2015 SCC Proceedings-Reduced PDF

SIMULIA COMMUNITY CONFERENCE
MAY 18–21, 2015 | BERLIN, GERMANY
PROCEEDINGS
3DS.COM/SIMULIA
TERMS AND CONDITIONS OF USE OF
THE USERS' CONFERENCE BOOK
The third-party research papers accessible to you in the Dassault Systèmes Simulia Corp. (“SIMULIA”)
Learning Community are made available to you for informational purposes only and are solely for
personal use. You may not use, reproduce, modify, sell or distribute copies of any third-party research
paper(s) in any form without the prior written permission of SIMULIA. Permission requests should be
sent to: simulia.conference@3ds.com.
The information presented in any third-party research paper is a product of the author(s) and does not
represent any view or opinion of SIMULIA. Further, SIMULIA does not endorse any views, opinions,
or methodologies contained in any third-party research paper or validate any results obtained by the
author(s) of any third-party research paper.
IN NO EVENT SHALL SIMULIA, ITS AFFILIATES, OR THEIR THIRD-PARTY PROVIDERS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, PUNITIVE, SPECIAL OR CONSEQUENTIAL DAMAGE
(INCLUDING WITHOUT LIMITATION, DAMAGES FOR LOSS OF BUSINESS PROFITS, BUSINESS
INTERRUPTION OR LOSS OF BUSINESS INFORMATION) AS A RESULT OF YOUR ACCESS TO ANY
RESEARCH PAPER EVEN IF SIMULIA OR ITS AFFILIATES HAS BEEN ADVISED OF THE POSSIBILITY
OF SUCH DAMAGES.
2015 SIMULIA Community Conference www.3ds.com/simulia 2

This volume contains the full manuscripts presented at the SIMULIA Community
Conference held in Berlin, Germany, May 18-21, 2015. The manuscripts are categorized
by industry, followed by company name in alphabetical order. We are grateful to the
authors for their efforts in preparing the papers that make up these proceedings. The full
manuscripts can be found on the SIMULIA Learning Community at www.3ds.com/slc.
CONTENTS
AEROSPACE & DEFENSE
12 Numerical Analysis to Optimize Peen-Forming Process Parameters
Y. Essa, M. Laspalas, E. García, A. Escolán, B. Hernández-Gascón, and F. Martín de la Escalera
(Aernnova Engineering Division S.A.)
22 Evaluation of Abaqus XFEM Capabilities for Crack Growth Analysis in

Aeronautical Structures
Ismael Rivero Arevalo and Javier Gomez-Escalonilla Martin (Airbus Defence and Space)
37 Bird Strike Analysis for Impact-resistant Design of Aircraft Wing Krueger Flap
Sebastian Heimbs, Wolfgang Machunze and Gerrit Brand (Airbus Group)
Bernhard Schlipf (Airbus Operations GmbH)
48 Use of Abaqus Explicit for Composite Sandwich Impact Damage Prediction

during Bird Impact
Marwan Al-Khalil (Airbus Operations Ltd.)
Richard Rigby and Erkan Kirtil (Airbus)
63 Time Accurate Simulation of Aeroelastic Flap Deployment with Free Play

Nicolas Reveles and George Antoun (ATA Engineering, Inc.)
James Fort (Dassault Systems SIMULIA Corp.)
75 Failure Analysis of CFRP Tubes with Integrated Rubber Layers Subjected to

Transverse Low-velocity Impact Loading
Enrico Stelldinger, Arnold Kühhorn and Markus Kober (BTU Cottbus–Senftenberg)
90 New Design of a Pressure Vessel Subjected to Blast Loads

Laurence Cousin and Pierre Evrard (CEA/DAM/DIF)
102 Predicting In- and Out-of-plane Damage Evolution in Fiber-reinforced Composites

Wouter Wilson (Fokker Landing Gear B.V.)
116 Nested Micromechanical and Structural Models for the Analysis of Discontinous
Long-fiber Thermoplastic Composite Materials and Structures
M. Hakan Kilic (Greene, Tweed & Co.)
129 An Approach Towards an Optimal Design of Composite Structures

Using Abaqus as FE-solver
Barbara Goller, Manuel Prackwieser and Manfred Gratt (INTALES GmbH)
144 Methodology for the Analysis of Tolerances in the Assembly Process of a Wing
Torsion Box Through FE Simulations
Agustín Chiminelli, Ruban Breto Pontaque and Miguel Angel Jimenez (ITAINNOVA)
Federico Martín de la Escalera and Luis Lizarduy (AERNNOVA)

160 High Velocity Perforation as a Benchmark Problem for Material Model Validation
Vladislav V. Vershinin (Moscow State University of Civil Engineering)
176 Simulation Lifecycle Management in the Real World: Practical Application

in PRIME Aerostructures
Andres Belles Meseguer (Prime Aerostructures)
188 Modeling of Heat Transfer across Bolted Joints in Abaqus/CAE

Volkan Coskun, Emir Ozkokdemir and Bulent Acar (Roketsan Missiles Inc.)
198 Modeling of Forming Process of Composite Materials Based on

Thermoplastic Matrix
Boris Fedulov, Mark Kantor, Alexander Safonov and Andrey Ushakov (Skolkovo Institute
of Science and Technology)
207 Finite Element Simulation of Thermal Barrier Coatings in Rocket Engines

Martin Bäker, Torben Fiedler and Joachim Rösler (Technische Universität Braunschweig)
219 Residual Strength of the Carbon Fiber Panel with Delamination

Mikhail Pavlov and Ivan Zharenov (TsAGI - Central Aerohydrodynamic Institute)
235 Shrinkage Modeling of Thermoplastic Wing Rib

Nikita Rozin and Ivan Zharenov (TsAGI - Central Aerohydrodynamic Institute)
Alexander Safomov and Andrey Ushakov (Skolkovo Institute of Science and Technology)
242 Analysis of the Carbon Fiber Wing Torsion Box After Low Velocity
Impact Damage
Ivan Zharenov and Mikhail Pavlov (TsAGI)
254 A Simulation Tool for Fatigue Analysis and Lifecycle Prediction

Daniel Kujawski, Phani C.R. Sree, Deepak Abburi and Joshua Kuok (Western Michigan University)
ARCHITECTURE, ENGINEERING & CONSTRUCTION

267 Elastoplastic Simulations with a Tangential Plasticity Constitutive Model for a Thin
Wall Bridge Pier Subjected to Various Non-proportional Cyclic Loading Conditions
Hideto Momii (ESTECH Corporation)
Seiichiro Tsutsumi and Riccardo Fincato (Joining and Welding Research Institute, Osaka University)
281 Study of Mechanical Compatibility at the Mortar-block Interface in a Heritage Building

and Numerical Modeling by Abaqus
Hicham Fihri Fassi and Y. Hammou (FSTS)
L. Van Parys and T. Descamps (Université de Mons)
M. Bideq (ENSA)
291 Improvement and Enhancement of Concrete Damage Plasticity Model

Yuji Nikaido, Yoshinori Mihara, Shohei Sawada and Yoshiyuki Takahashi (Kajima Corporation)
303 An Easy Procedure for Anisotropic Non-Linear Behavior of

Short-Fiber-Reinforced Plastics
Sascha Pazour, Wolfgang Korte and Marcus Stojek (PART Engineering GmbH)

313 Experience and Lessons Learnt from the SMART 2013 Benchmark
Maria J. Crespo, Javier Rodriguez, Luis Lacoma, Francisco Martinez and Joaquin Marti (Principia)
327 Advanced Finite Element Analysis of Deep Excavations using Abaqus

Yuepeng Dong (Singapore-MIT Alliance for Research and Technology)
Harvey J. Burd and Guy T. Houlsby (University of Oxford, UK)
Andrew J. Whittle (Massachusetts Institute of Technology, USA)
344 Modeling of Confined Inflatable Structures

Eduardo Sosa, Choo-Siang Wong and Ever J. Barbero (West Virginia University)
CONSUMER PACKAGED GOODS & RETAIL

360 Simplified Approaches for Modelling a Sealed and Partly Fluid Filled Container
Oswald Valtiner, Daniel Pielmeier, and Claudius Oblasser (ALPLA Werke Alwin Lehner GmbH & Co KG)
375 Modeling the Mechanical Behaviour of Some Complex Polymeric Materials by Relying
on the Nonlinear Capability in Abaqus
Colmar Wocke (Dow Chemical)
390 Analysis of Material-Related Instability Problems for Inelastic Thin Shells

Takaya Kobayashi and Yasuko Mihara (Mechanical Design & Analysis Corporation)
401 Application of Abaqus and Isight Simulation on Corrugated Board and Packaging
Arnoud Dekker (Smurfit Kappa)
405 Use of Abaqus/CAE and True-Load™ to Determine External and Internal Loading of
a Full Suspension Mountain Bicycle
Jay Maas (Trek Bicycle Corporation)
ENERGY, PROCESS & UTILITIES

417 Process integration between CAD and CAE applications for Isight based DOE analysis
Davide Pinna (Ansaldo Energia)
Luca Fattore (Exemplar)
431 Hydraulic Fracturing Simulation for Fracture Networks

Stephan Arndt, Tobias Hoeink and Wouter van der Zee (Baker Hughes)
Jianhu Nie (formerly of Baker Hughes)
449 Fracture Stability Assessment during SAGD operations using

4D Geomechanical Models
Wouter van der Zee and Marc Holland (Baker Hughes)
463 Finite Element Analysis for Post Cracking Design of Thermal Well Cements
Jueren Xie (C-FER Technologies)
478 Use of FEA and Radial Basis Functions for Reliability-based Design and Assessment
of Tubular Connection Sealability
Jueren Xie, Thomas Dessein and Feng Rong (C-FER Technologies)
495 Simulation of Hydraulic Fracturing of Unconsolidated Sands using Fully Coupled

Poro-Elastoplastic Models
Saurabh Puri, Juan Hurtado and Deepak Datye (Dassault Systèmes Simulia Corp)
Ganesh Dasari, Kevin Searles and Pablo Sanz (ExxonMobil Upstream Research Company)

509 Finite Element Verification of Non-Homogeneous Strain and Stress Fields during
Composite Material Testing
Lars Pilgaard Mikkelsen (DTU Wind Energy)
520 Advanced Fracture Modeling for Cuttings Re-injection

Jorge Garzon, Matias Zielonka, Kevin H. Searles and Pablo Sanz (ExxonMobil Upstream
Research Company)
536 Experimental Validation of Simulation Capabilities for Hydraulic Fractures Propagating

in a Porous Medium
Jing Ning, Gilbert Kao, Nikolay M. Kostov, Kevin H. Searles, Scott R. Buechler and Pablo F. Sanz
(ExxonMobil Upstream Research Company)
546 Dynamic Hydraulic Fracture Modeling for Wellbore Integrity Prediction in

a Porous Medium
Nikolay Kostov, Jing Ning and Pablo Sanz (ExxonMobil URC)
Shekhar Gosavi (ExxonMobil DC)
Piyush Gupta (ExxonMobil Upstream Research Company intern)
555 A Numerical Investigation on the Bedding Resistance of Laterally

Displaced Pipelines
Tim Gerlach, Martin Achmus and Mandy Narten (Institute for Geotechnical Engineering,
Leibniz University Hanover)
566 Pipe Flexibility Analysis and The Development of PCS - Pipe Calculation
System for Abaqus
Carlos Eduardo Alvares de Castro e Sousa (RMC Consultoria)
582 Design Optimization of a Curved Fir-Tree Root of Steam Turbine Using Computer Aided
Optimization and Finite Element Analysis
Shangjiong Tan, Weilin Shu, Gongyi Wang, Kai Cheng, Jiandao Yang (Shanghai Turbine Works
Company Ltd)
601 Increasing the efficiency of offshore rigid pipeline lateral buckling assessments
using a dedicated GUI and Isight
Carlos Charnaux, Graeme Roberts and Samuel Paul (Subsea 7)
621 Finite Element Simulation of Rotary Steerable Drilling Systems Advance Understanding
of Damaging Downhole Shock and Vibration
Nader Abedrabbo and Liam Lines (Weatherford)
636 Abaqus Analysis Methods on Highly Restrained Pipeline with

Soil Berm Formation
Shulong Liu, Emil Maschner, Teng Zhang, John Smyth and John Li (Wood Group Kenny)
HIGH-TECH
660 Achieving Higher Productivity on Abaqus Simulations with HPC and
Clustering Technologies
Pak Lui (HPC Advisory Council)

INDUSTRIAL EQUIPMENT
674 Robust and Optimal Strain Gauge Positions for Blade Vibration Testing
Piyawan Teufel (ABB Turbo Systems AG)
Jan Poland (ABB Corporate Research Center)
691 Prediction of Tin-Whiskers Generation Using Stress and Mass-Diffusion

Analysis
Takeshi Terasaki (Hitachi, LTd.)
702 Strength and Fatigue Life Analysis of Leaf Springs used in Heavy Duty Trucks
Mehmet Bakir, Murat Siktas, Serter Atamer (Mercedes-Benz Turk A.S.)
717 Coupled Analysis of Fracture Mechanics and Piezoelectricity in Active Layers in the
Abaqus Code Operated Through the Isight Tool
Eugenio Brusa (Politecnico Di Torino)
Mehdi Mohammadzadeh Sari (Politecnico Di Torino)
Cosima Fiaschi, Mauro Parodi (EXEMPLAR)
727 XFEM Simulation of Radial-Median Crack Evolution in Knoop Indentation of

Brittle Materials
Felix Rickhey (Sogang University)
Jin Haeng Lee (Korea Atomic Energy Research Institute)
Hyungyil Lee (Sogang University)
742 Evaluation of the Fatigue Strength of a Steel-Aluminum Tapped Thread Joint

with Local Concepts
Dominic Olveda (Technische Universität Darmstadt)
Ralf Waterkotte (Schaeffler Technologies GmbH & Co. KG)
Matthias Oechsner (TU Darmstadt)
756 Strip Shape Prediction using a Simplified Static Model of a 4-High

Cold Rolling Process
Mario Alberto Bello-Gomez, Martha P. Guerrero-Mata, Luis A. Leduc-Lezama,
Tania P. Berber-Solano (UANL)
Fernando Gonzalez (Ternium Mexico)
768 Validation of Flow Simulation on Abaqus/CEL™

Edgar Ramirez (Universidad Nacional Autónoma de México (UNAM))
Alan Saucedo, Osvaldo Ruiz, Victo Jacobo and Armando Ortiz (Engineering School, UNAM)
782 On the Numerical Implementation of a 3D Fractional Viscoelastic

Constitutive Model
Gioacchino Alotta (University of Palermo - University of Oxford)
Olga Barrera, Alan C. F. Cocks (University of Oxford)
Mario Di Paola (University of Palermo)
795 Hydrogen Embrittlement Mechanisms in Metals: a Modelling Approach

Olga Barrera, Alan C. F. Cocks and Edmund Tarleton (University of Oxford)

LIFE SCIENCES
808 Developing a Constitutive Model for a Highly Deformable Latex Rubber with
Implementations in Lagrangian Frameworks
Christopher Basciano, Marcus Rademacher, Patrick Downie and Anita Bestelmeyer (BD)
Brian Croop (DatapointLabs)
Jorgen Bergstrom (Veryst Engineering LLC)
820 Developing a Finite Element Head Model for Impact Simulation in Abaqus
P. G. Young and R.T. Cotton (Simpleware)
S.M. Qidwai, A. Bagchi and N. Kota (U.S. Naval Research Laboratory)
C. Pearce (Atkins Ltd.)
833 Impact Simulation of a Hearing Aid

Morten Birkmose (Soendergaard GN Resound A/S)
842 FE Modeling of a Flatfoot Deformity for Improving Surgical Planning

Zhongkui Wang and Shinichi Hirai (Ritsumeikan University)
Kan Imai, Masamitsu Kido and Kazuya Ikoma (Kyoto Prefectural University of Medicine)
TRANSPORTATION & MOBILITY

855 IDIADA's Virtual Proving Ground (VPG) for Durability Analysis
Jordi Arbiol, Xavier Armengol, Jose Antonio Muñoz, Inés Lama and Enric Aramburu
(APPLUS IDIADA)
869 Thermal Management of Electronic Devices used in Automotive Safety –

A DoE approach
Vinod Kumar, Vinay Somashekhar and Srivathsa Jagalur (Autoliv India Pvt. Ltd.)
880 Development of a Separator Plate for Valve Bodies in a Vehicle

Automatic Transmission
Harald Rebien, Tobias Tauscher and Rohit Ramkumar (DANA)
894 System Modeling and Simulation at Dana Holding

Frank Popielas (Dana Holding Corp)
904 New Constitutive Model for Woven Thermoplastic Composite Materials

Pieter Volgers (DuPont Performance Polymers)
Helga Kuhlmann and Zhenyu Zhang (DuPont Engineering Technologies)
920 Durability Assessment of Heavy Duty Diesel Engine Cylinder Head

A.H. Guzel, S.Erpolat, and O. Cicekdag (Ford Otosan A.Ş, Turkey)
931 Prediction of Compressive Characteristic of Hat Section High-strength Steel Frame

Taking Bauschinger Effect into Account
Kenji Takada, Masaki Matsuo, and Kenichiro Fukagawa (Honda R&D Co., Ltd.)
Kentaro Sato (JFE Steel Corporation)
Pierre Farahat and Anwer Yasin (Dassault Systèmes K.K.)
Lamine Diallo (Dassault Systèmes Canada Inc.)
Yangwook Choi (Dassault Systèmes Simulia Corp)
949 The Effect of Rubber in Strength Analysis for Strut Insulator Assembly
Hoo-Gwang Lee, Ji-Hoon Kim and Sung-Jin Yoon (Hyundai Motor Company)

956 Study on Body Rigidity, Strength and Fatigue Performance for Carbon Fiber
Reinforced Plastics Vehicles
Oh Se-Hee and Son Yu-Na (Hyundai Motor Company)
963 Simulation of Pipe Bending Process with Abaqus

Satoshi Ishikawa (IDAJ Co., Ltd.)
Yoshihiro Ishikawa (SANGO Co., Ltd.)
974 Automated Weight Optimization Process for Automotive Wheel Hub Bearing
Jungho An, Inha Lee, Seungpyo Lee (ILJIN Global Corporation)
Youngmin Cho and Sangkyu Park (Dassault Systèmes SIMULIA)
981 Automated Analysis in CATIA V5

Stephen Wallis (Jaguar Land Rover)
990 Simulation of Fully Coupled Thermo-Mechanical Effects in a Disc Brake Rotor

G.Karunakar, J. Oscroft , R. Taulbut and D. Steward (Jaguar Land Rover Limited)
997 Simulation of Combined Forming and Injection Molding Processes

Timo Mueller and Frank Henning (Karlsruhe Institute of Technology (KIT))
1011 Application of Predictive Engineering Tool (Abaqus) to Determine Optimized

Automotive Rubber Door Harness Grommet Design
Praveen Mishra and Dayananda Gowda (Mercedes Benz India R & D Pvt Ltd)
1018 Final Element Simulation of Blankholder’s Lift-off in a Deep Drawing Tool

Using Abaqus/Standard
Krzysztof Swidergal and Marcus Wagner (OTH Regensburg)
Christian Lubeseder, Ingo von Wurmb and Josef Meinhardt (BMW Group)
1024 Strength Assessment of Injection Molded Short-fiber-reinforced

Plastic Components
Wolfgang Korte, Marcus Stojek, Sascha Pazour (PART Engineering)
1034 Simulation Driven Design for Manufacture of a Pressed Composite

Automotive Tailgate
Andy Ngai and Mark Arnold (Penso Consulting)
1047 Brake System Model Reduction for Squeal Noise Study

Samuel Nacivet and Olivier Fazio (PSA Peugeot Citroën)
Jean-Jacques Sinou (Ecole Centrale de Lyon)
1062 Application of Coupled Eulerian Lagrangian Approach in Finite Element

Simulation of Friction Stir Welding
N. K. Sanjeev and B. P. Ravikiran (Robert Bosch Engineering and Business
Solutions Limited)
1080 Innovative Anisotropic Material Modeling Approach for

Fiber Reinforced Thermoplastics
Recep Yaldiz (SABIC)
Gertjan Kloosterman (Dassault Systèmes B.V. 3DS/SIMULIA)

1091 Topology Optimization of a Motorcycle Swing Arm Under Service Loads
using Abaqus and Tosca
Roberto Saponelli (Sacmi/Protesa)
Massimo Damasio (Exsemplar)
1102 Consideration of Manufacturing Pre-history Influence of Fatigue Life

of Sheet Metal Formed Automotive Parts
Dmitry Buzlaev (TESIS)
Viktor Kalaev and Dmitry Zheleznyakov (GAZ UEC)
Anders Winkler (Dassault Systèmes)
1113 Development of a Multifunctional Truck Cross Member Made of

Continuous Fiber Reinforced Plastics
Herwig Kirchberger (TEUFELBERGER Composite Ges.m.b.H.)
S. Fischmeister, H. Dannbauer and K. Puchner (Magna Powertrain, Engineering Center Steyr
GmbH & Co KG)
1126 Modeling the Draping of NCF Composite Preforms

Muhammad-Ali Siddiqui and Colmar Wocke (The Dow Chemical Company)
1141 Virtual Simulation of the Engine Hood Misuse Test, Development Practices
and Correlation Activities
Tamer Aydiner and Halil Bilal (Tofaş Türk Otomobil Fabrikası A.Ş.)
1154 INDEX

AEROSPACE & DEFENSE

Numerical Analysis to Optimize Peen-Forming
Process Parameters
Y. Essa1, M. Laspalas2, E. García3, A. Escolán2, B. Hernández-Gascón2, and
F. Martín de la Escalera1
1
R&T Department, Aernnova Engineering Division,
20 Manoteras Avenue. 4-5 floor, 28050 Madrid, Spain
2
Materials and Components Div. ITAINNOVA (Instituto Tecnológico de Aragón)
7 María de Luna Street, 50018 Zaragoza, Spain
3
NMF EUROPA (Aerometallic)
Andalucía Street. Tarazona Industrial Zone. 50500 Tarazona, Zaragoza, Spain
Abstract: Peen-forming is a practical technique for forming of metallic aeronautical components.

The method performance of this cold-working procedure mainly depends on the efficient selection
of process parameters. Currently, trial and error procedures, based on operators' experience are
used to determine peen-forming intensity patterns. This expensive and time consuming method
could be avoided by the accurate estimation of peen-forming process parameters. The absence of
a predictive tool to determine these parameters makes difficult to give a quick response to market
for the industrialization of new components. A virtual tool is developed to help designers to
evaluate the feasibility of applying the peen-forming technique and process parameters to
minimize drastically the number of rejected components in the process setup. Peen-forming
intensity is usually quantified by means of surface coverage, pre-stress condition, ball size, shot
velocity, component thickness, etc. The virtual software is called AERnnova Numerical
Optimization of Peen-forming Process (AERNOP). The scope of the present study is to determine
the process parameters optimization of peen-forming process using a 3D finite element method
(FEM) methodology.
Keywords: Virtual tool, flat pattern, residual stresses, optimization, AERNOP.
1. Introduction
Peen-forming process is characterized by striking the working surface of the panel with small
spherical steel shots at a high impact velocity, see Figure 1. Shot impact produces plastic
indentation surrounded by elastic region. After ceasing the shot impact, the plastic zone has
become permanently deformed and elastic stressed zone is likely to recover unloaded initial

conditions. Consequently, compressive residual stress distribution in the near-surface layer is
developed as a result of elasto-plastic deformation behavior. Residual stress distribution is
produced to maintain equilibrium in the peened layers. Compressive residual stress distribution
can extensively reduce the crack growth rate of surface-short-cracks thus increasing the fatigue
life of shot peened components.
Figure 1. Peen-forming process.
The efficiency of the peen-forming process depends on process intensity, which is the degree of
plastic deformation introduced by the impacting shots. The peen-forming intensity is mainly
quantified by surface coverage, shot characteristics, pre-stress condition, component
characteristics and shot velocity. Surface coverage is defined as the ratio of peened area to the
total exposed surface area.
Virtual numerical modeling of peen-forming has received increasing attention from research
community to analyse the compressive residual stress mechanism. This is due to the high
experimental measurements cost of direct residual stresses, which require the application of semi-
destructive tests, moreover time consuming. Multiple impact collisions during peening process
have been simulated using quasi-static analysis, Meguid et al. (1985), Li et al. (1991) and Frija et
al. (2006). However, dynamic analysis realized by Kobayashi et al. (1998) has demonstrated that
static residual stress profile is different from dynamic analysis.
Modeling of peen-forming is performed at increased level of complexity, starting by single shot or
by multiple shot impact events. Dynamic simulation of single shot was carried out by Al-Hassani
et al. (1999), Hong et al. (2008) and Meguid et al. (1999). Multiple shot impacts have been
conducted by Meguid et al. (2007) and, Miao et al. (2009) using finite element models. It is worth
mentioning that few references have addressed the final deformation process at component level
(Gardinier et al. (1999), Levers (1998), Miao et al. (2011) and Wang et al. (2006)).
In this study, a methodology that combines different scales to define peen-forming strategy and to
estimate the process parameters is presented, and applied for an aeronautical component.
As an antecedent of this work, it is worth to mention a previous contribution to Simulia
Community Conference, Laspalas et al. (2011), where a direct FEM methodology for the
simulation of peenforming process was presented. In this work the focus is to develop an inverse
methodology to optimize process parameters.

2. AERNOP simulation procedure
The numerical tool, namely AERNOP, is developed with the purpose of estimate the optimal
peen-forming process parameters. The tool is designed to help the industry to provide a quick
feasibility study of any kind of client peen-forming purchase order. The tool can evaluate the
viability to obtain the objective geometry by determining the required parameters, applying
inverse engineering. To do so, the objective geometry is taken as the starting configuration. Also,
the tool can check if obtained parameters are adequate using direct engineering, starting from the
flat pattern and predicting the deformation after peen-forming process. Finally, the reached
deformed shape can be correlated with the desired deformed shape.
Both approaches are based on the FEM. The inverse method consists on preparing a mesh model
of the objective geometry. Two parallel rigid analytical surfaces are modelized and frictionless
contacts are defined with the part geometry. The simulation is a static calculation where the
component is compressed between both rigid surfaces till it becomes flat (“sandwich method”).
The obtained stress distribution is afterwards used, first, for estimating the maximum pre-stress
that can be applied to the component, keeping the material in its elastic regime (see Figure 2).
(a) (b)
Figure 2: Sandwich method. (a) Objective geometry inside two rigid surfaces. (b)
Plane geometry showing the stress distribution.
Second, the stress distribution is used as a reference for the process parameter estimation. The
areas with compressive stresses are the areas that must be stretched by peenforming in the direct
process. This technique allows fixing the strategies to be applied in each zone of the component
(see Table 1).

Strategy Final shape
Peening on top surface
Peening on bottom surface
Peening on stiffening ribs
Table 1 – Peenforming strategies
The compressive stress value is related with the peenforming intensity required, but the solution is
not unique, different combination of peenforming parameters can result in the same overall
deformation.
Using a detailed FE simulation model, the peenforming process is analyzed at the material level. A
region of interest of the material is represented and a stream of randomly placed shots impacting
the surface is modelled using the discrete element modelling technique (DEM), and a proper
contact definition. The calculation is a transient dynamic simulation solved using Abaqus/Explicit
where the elasto-plastic response of the material under the surface is analysed as a function of the
different process parameters: shot size, velocity and coverage. The ranges of each variable are

selected to cover the full processing window. Figure 3 represents an in-plane stress distribution
together with the stress profiles through the thickness at different coverage ranges.
Stress [MPa]
0
Thickness [mm]
Figure 3: Multiple impact stress distribution profile evolution
This family of residual stress profiles is post-processed and a peenforming intensity function is
defined, which represent the deformation capability for each combination of peening parameters

(see Figure 3). Based on this intensity function and in the stress distribution obtained in the
component using the “sandwich method” a process estimator tool determines the optimal process
parameters, under predefined criteria. This is required because different combinations of
parameters (ball size, velocity, coverage, etc) might result in the same intensity, i.e. deformation
capability. These optimization criteria are based on the know-how of the manufacturing company
and additional requisites imposed by the client.
.
𝐼𝐼 = 𝑓𝑓(𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣, 𝑡𝑡ℎ𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖, % 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐, 𝑝𝑝𝑝𝑝𝑝𝑝 − 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠)

𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 = 𝑓𝑓(∅𝑏𝑏𝑎𝑎𝑎𝑎𝑎𝑎 , 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝)
Constant ball diameter
(ᶲball) and installation
pressure (p)
Intensity
function (Pa)
Figure 4: Intensity function (I).
3. Demonstration example
The developed methodology was applied on a complex aeronautical component having several
sizes of windows, different thicknesses along the panel, stringers and curvature representing an
example of objective demonstrator for a specific client order. Details of the objective work-piece
can be observed in Figure 4.

Figure 5. Details of the objective aeronautical demonstrator.
An appropriate finite element 3D quality mesh must be prepared and it has to take into
considerations the very small holes and the different thicknesses, moreover the stringers. The final
FEM mesh can be observed in Figure 5. The total mesh size has about 20000 elements.
Figure 6. Finite element mesh.
In Figure 6 is represented the coverage map derived from the developed tool. In this case, there is
no need to apply pre-stress boundary conditions.
The application of the previously described methodology result in a prediction of the peening
parameters, which include the shot size and velocity required, the pre-stress level and the coverage
distribution (Figure 6). In this case no pre-stress is required and only is necessary to apply shot on
the aeronautical side of the panel.

Figure 7: Coverage distribution on the demonstrator.
Finally, considering the obtained results, a study was done applying these acquired parameters on
a flat pattern using the calculation tool AERNOP, as a direct FEM method. This is for the sake of
comparison of the deformed shape obtained using the estimated parameters and the objective
geometry. The comparison between both geometries is performed by applying the methodology
known as “best fitting”, which applies a rotation and translation of the geometry to minimizing the
distances between the similar points of both geometries. The results can be shown in Figure 7,
where the product obtained after the direct method (gray) is superimposed to the target part
(green). The maximum error is 12mm, which can be observed in Figure 8. This is an acceptable
error over 2000mm demonstrator long.
Figure 8: Best fitting strategy between the geometry of the aircraft panel under
study (green) and that obtained by the direct method (grey).

Figure 9: Error distribution applying best fitting strategy.
4. Conclusions
The proved methodology, AERNOP, was conducted to simulate the peen-forming process,
estimate the process parameters and provide coverage contours to the operator in the industry. The
methodology was applied on a complex aeronautical demonstrator using sandwich method and
obtained results were used as a direct method. Best fitting process was applied to compare the
geometries. The results are very promising.
The developed simulation tool, AERNOP, allows predicting the peen-forming parameters. This
tool can be very helpful for designers to assess the process feasibility in early stages, allowing to
anticipate manufacturing problems and to take the necessary actions in advance. It can estimate, a
priori, the optimum shot peening parameters to be used by expert workers in the peen-forming
process setup, which reduce time and cost consumed during the process set up.
5. Acknowledgement
This project has been partially supported by the Centre for the Development of Industrial
Technology (CDTI), Spanish Ministry of Economy and Competitiveness. Their financial support
is greatly appreciated.
6. References
1. Al-Hassani, S.T.S., K. Kormi, K and D.C. Webb. “Numerical Simulation of Multiple Shot
Impact”, The 7th international conference on shot peening ICSP7, Warsaw 1999.
2. Frija, M., Hassine, T., Fathallah, R., Bouraoui, C., Dogui, A. “Finite Element Modelling of
Shotpeening Process: Prediction of the Compressive Residual Stresses, the Plastic
Deformations and the Surface Integrity”. Mat Sci Eng A, 426 173–180, 2006.

3. Gardinier, D.S. and Platts, M.J. “Towards Peen Forming Process Optimization”, The 7th
International Conference on Shotpeening ICSP7 Conference, Warsaw 1999.
4. Hong, T., Ooi, J.Y. and Shaw, B.A. “A Numerical Study of the Residual Stress Pattern from
Single Shot Impacting on a Metallic Component”, Advances in Engineering Software, vol.
39, pp. 743-756 2008.
5. Kobayashi, M., Matsui, T., Murakami, Y. “Mechanism of creation of compressive residual
stress by shot peening”. Int J Fatigue, 20 351–7, 1998.
6. Laspalas, M., Gómez, J., Martín de la Escalera, F., Sánchez, R., Jiménez, M.A. “Simulation of
Peen Forming Process of Aluminum Aeronautic Panel”. Id1242, Barcelona, Spain 2011.
7. Levers, A., and A. Prior, “Finite Element Analysis of Shot Peening”. Journal of Materials
Processing Technology, vol.80-81, pp. 304-308, 1998.
8. Li, A.K., Yao, M., Wang, D., Wang, R., “Mechanical Approach to the Residual Stress Field
Induced by Shotpeening”. Mater Sci Eng, A147 167–73, 1991.
9. Meguid, S.A., and Klair, M.S., “An Examination of the Relevance of Co-indentation Ctudies
to Incomplete Coverage in Shotpeening Using the Finite-Element Method,” J Mech Working
Technol, no. 11, 87–104, 1985.
10. Meguid, S.A., Shagal, G., Stranart, J.C. and Daly, J. “Three-Dimensional Dynamic Finite
Element Analysis of Shotpeening Induced Residual Stresses”, Finite Element in Analysis and
Design vol. 31, pp. 179-191, 1999.
11. Meguid, S.A., Shagal, G. and Stranart, J.C. “Development and Validation of Novel FE
Models for 3D Analysis of Peening of Strain-Rate Sensitive Materials”, Journal of
Engineering Materials and Technology, vol. 129, pp. 271-283, 2007.
12. Miao, H.Y., Larose, S., Perron, C. and Lévesque, M. “ On the Potential Applications of a 3D
Random Finite Element Model for the Simulation of Shot Peening”, Advances in Engineering
Software, vol. 40, pp. 1023-1038, 2009.
13. Miao H.Y, Larose, S., Perron, C., Lévesque, M. “Numerical Simulation of the Stress Peen
Forming Process and Experimental Validation,” Advances in Engineering Software 42 963–
975, 2011.
14. Wang, T., Platts, M.J. and Levers, A. “A process for Shot Peen Forming”, Journal of
Materials Processing Technology, vol. 172, pp. 159-162, 2006.

Evaluation of Abaqus XFEM Capabilities
for Crack Growth Analysis in
Aeronautical Structures
Ismael Rivero Arévalo and Javier Gómez-Escalonilla Martín
Fatigue and Damage Tolerance Department (TAETS11) – Military Aircraft
Airbus Defence and Space
Abstract: In the Fracture Mechanics field, standard industrial methods for crack growth analysis
are mainly based on analytic calculations as classical Finite Element approaches are not
practical to deal with discontinuities such as fatigue cracks due to the associated high
computational costs. eXtended Finite Element Method (XFEM) is one of the methodologies that
are being developed in the recent years in order to overcome the limitations associated to
classical approaches, especially for complex analysis. However limited industrial experience is
available to adopt this methodology as a standard practice in the aircraft industry.
XFEM capabilities available in Abaqus 6.14 have been evaluated by the authors to assess the
reliability and feasibility of the implementation of XFEM methodology to perform crack growth
analyses in aeronautical structures. A set of different cases, common in aeronautical structures,
have been selected for this purpose to check Abaqus capabilities on different configurations.
In the paper authors presents the results obtained for the evaluated cases and the validation of
these results using alternative methodologies for comparison. Current limitations in Abaqus
capabilities are also discussed and future developments are proposed for implementation in next
Abaqus versions.
Keywords: Aircraft, Fatigue, Crack, XFEM
1. Introduction
Reduction of structural weight and maintenance requirements are two main goals for aircraft
industry as these two facts directly penalize the aircraft operational costs. In the pursue of this
target, the development of more accurate, robust and reliable methodologies for structural analysis
has a key role as improved methods allows designing more optimized structural designs.
Focusing on the Fatigue and Damage Tolerance field, responsible of the sizing of many aircraft
components and the definition of structural maintenance requirements, the use of numerical
methodologies provides an important added value. For example the Finite Element Method (FEM)
allows the accurate calculation of Stress Concentration Factors for complex configurations where
analytic methodologies can only provide a conservative approach.
By the other hand, Crack Growth calculations are mainly based on analytic methodologies as no
numerical approaches are consolidated as standard industrial practices. The XFEM (eXtended

Finite Element Method) is one of the numerical methodologies that are appearing as an alternative
to classical analytic methods for crack growth calculations. It combines the benefits of FEM to
analyze complex structures with the capability of dealing with through-element discontinuities as
fatigue cracks with acceptable computational cost.
The authors are working on new methodologies based on XFEM to evaluate its capabilities for
crack growth calculations and its feasibility to become an industrial standard practice, the same
way as FEMs are used for stress calculations.
XFEM capabilities available in Abaqus 6.14 have been evaluated by the authors to assess the
reliability and feasibility of the implementation of XFEM methodology to perform crack growth
analyses in aeronautical structures. A set of different cases, common in aeronautical structures,
have been selected for this purpose to check Abaqus capabilities on different configurations.
In the paper authors presents the results obtained for the evaluated cases and the validation of
these results using alternative methodologies for comparison. Current limitations in Abaqus
capabilities are also discussed and future developments are proposed for implementation in next
Abaqus versions.
2. XFEM capabilities in Abaqus

The eXtended Finite Element Method is an extension of the conventional Finite Element Method
based on the concept of Partition of Unity (PUM). This methodology is able to deal with the
presence of discontinuities in a finite element by enriching degrees of freedom with special
displacement functions. Formulation and mathematical background of the XFEM approach
implemented in Abaqus can be found in the literature, (Du) and (Park, 2012).
The displacement vector of a node included in the enriched domain can be expressed
mathematically as follows:
𝑁 4
𝑢 = � 𝑁𝐼 (𝑥) �𝑢𝐼 + 𝐻(𝑥) 𝑎𝐼 + � 𝐹∝ (𝑥)𝑏𝐼∝ � (1)

𝐼=1 ∝=1
where NI(x) are the shape functions affecting the entire set of nodes in the domain, ui the nodal
displacement vectors of the entire set of nodes, aI the nodal enriched degrees of freedom for the set
of nodes whose shape functions supports are fully cut by the crack (set J), H(x) discontinuous or
‘jump’ functions, bIα the nodal enriched degrees of freedom for the set of nodes whose shape
functions supports are cut by the crack tip and Fα(x) adequate asymptotic functions for the
displacement field near the crack tip.
To mathematically describe the crack and track its growth the Levet Set Method is used. Two
level sets are required to model the crack:
Φ: describes the crack surface
Ψ: it is built so that its intersection with Φ is the crack front

The use of these two level sets allows the representation of the crack using only nodal data (no
explicit representation of the crack is required).
The combination of this methodology with the enriched shape of functions allows the modelling
of arbitrary crack growth without the necessity of remeshing.
Due to its high potential, this methodology has been implemented in several commercial codes
such as Abaqus (Du, Z.). In this work, authors explore the performance of the XFEM code
implemented in Abaqus 6.14 in fracture analysis of typical aeronautical structures.
Among the different XFEM approaches included in Abaqus, this paper will be focused on two of
them:
1) XFEM Stationary Crack
XFEM Stationary Crack approach implemented in Abaqus considers full element
enrichment (all the terms included in Equation 2 are taken into account). This
formulation allows dealing with the asymptotic stress fields that appears near the
crack tip.
This methodology can be used to calculate Stress Intensity Factors (SIFs) for
arbitrary cracks without adapting the mesh to the crack path (cracks can cross finite
elements and crack tip can be located inside a finite element).
The domain of the elements affected by the enriched formulation is defined by the
following Abaqus keyword:
*ENRICHMENT, ELSET=Solid-1, NAME=Crack-1, TYPE=STATIONARY CRACK,
Crack position is defined as an Initial Condition:

*INITIAL CONDITIONS, TYPE=ENRICHMENT
SIF evaluation is based on contour integration around crack tip. This analysis is
requested using the following keyword:
*CONTOUR INTEGRAL, CRACK NAME=Crack-1, CONTOURS=5, TYPE=K FACTORS, XFEM
The use of this formulation will result in obtaining a Stress Intensity Factor value for
each integration contour (the user can define the number of contours to be evaluated)
at each node located in the crack tip. In the examples included in the following
sections this point will be discussed with more detail.
For result post-processing, level sets Φ and Ψ (define the crack surface and tip) and
the XFEM status (intact, broken or partially broken) of the enrichment elements have
to be requested:
*NODE OUTPUT
PHILSM, PSILSM
*ELEMENT OUTPUT
STATUSXFEM

2) XFEM Crack Propagation using Low-cycle Fatigue Analysis based on Linear
Elastic Fracture Mechanics
The crack growth analyses included in this paper have been performed using the
Low-cycle Fatigue Analysis based on Linear Elastic Fracture Mechanics (LEFM)
implemented in Abaqus.
This methodology has been selected due to the following reasons:

- Low-cycle Fatigue Analysis is performed using Abaqus Direct-Cyclic
approach what significantly reduces the computational costs of the analysis,
as the damage growth can be extrapolated through several load cycles
avoiding the necessity of evaluating cycle-by-cycle the full model.
- Classical analytic methodologies for crack growth analysis are mainly based
on Linear Elastic Fracture Mechanics, so a great amount of material data is
available to be used with LEFM approach.
By the other hand the use of this methodology has certain limitations:
- A reduced element enrichment formulation is used in the analysis as the
term that takes into account the asymptotic stress field at the crack tip is not
taken into account. As a consequence crack tip cannot be located inside an
element.
- Linear elastic material behavior as the analysis is based on LEFM.
- Direct-Cyclic approach is not able to deal with problems in which changes

in the status of the contacts between the different components are produced.
This limitation also affects the contact between the crack surfaces (it can be
very important if compression load states are applied)
- The efficiency of the Direct-Cyclic approach is highly reduced if complex

loading spectra are applied as damage growth extrapolation through load
cycles is not possible.
The domain of the elements affected by the enriched formulation is defined by the
following Abaqus keyword:
*ENRICHMENT, NAME=Crack-1, ELSET=Solid-1.Solid, [TYPE=PROPAGATION

CRACK], INTERACTION=Crack
*SURFACE INTERACTION, NAME=Crack
Crack initial position is defined as an Initial Condition:

*INITIAL CONDITIONS, TYPE=ENRICHMENT

The LEFM criterion is defined by the following keyword. Paris’ law in terms of
strain energy release has to be defined by the user for the analyzed material in this
keyword (Equation 2).
*FRACTURE CRITERION, TYPE=FATIGUE, [NORMAL DIRECTION=1]
𝑑𝑎
= 𝑐3 ∆𝐺 𝑐4 (2)
𝑑𝑁
where da/dN is the increment of crack length in the applied load cycle, ΔG is the
incremental strain energy release and c3, c4 are material parameters.
The Low-cycle Fatigue Analysis (Direct-Cyclic) is defined by the following

keyword:
*STEP
*DIRECT CYCLIC, FATIGUE
For result post-processing, additionally to the outputs requested for Stationary Crack
approach, the incremental strain energy release is requested.
*ELEMENT OUTPUT
ENRRTXFEM
3. Practical applications
In this section the different cases that have been studied by the authors to evaluate Abaqus XFEM
capabilities will be presented. A set of representative cases in aeronautical structures is selected to
perform this evaluation. To evaluate the reliability of the method, the results obtained using
XFEM approach will be compared with other methodologies such as analytic calculations.
3.1 Edge cracks

Cracks growing from the edge of a plate are a very common scenario in the Fracture Mechanics
analyses performed for aeronautical structures. The configuration selected for the analysis is
shown in the next figure:

100 MPa
Crack length t=3.0mm

100mm
Edge crack
50.0mm
Figure 1: Edge crack scenario
For this configuration SIF calculation and crack growth analysis have been performed.
The model generated using Abaqus/CAE for the XFEM analysis of this configuration is shown in
the next figure. A max principal stress plot is shown for a certain crack length.
Figure 2: Edge crack. Abaqus model

Stress Intensity Factor calculation

SIF values have been calculated for several crack length / panel width ratios (from 0.1 to 0.6).
The next table shows the detailed results obtained for a crack length / panel width ratio of 0.2.
Results are shown for each through-thickness node and each integration contour.
Table 1: Detailed SIF results for crack length / panel width = 0.2
SIF – mode I Integration Contour

(MPa√mm) 1 2 3 4
1 1288 852 890 910
2 1261 755 807 824
Node 3 1270 769 816 834
4 1261 755 807 824
5 1288 852 890 910
As it can be seen, obtained values in each integration contour differ. So comparison with other
approaches is required to determine which contour provides the best results.
XFEM results will be compared with the analytic results obtained from Rooke and Cartwrigth
solution 1.1.20 (Rooke and Cartwright, 1974).
The analytic SIF result for the crack length / panel width ratio of 0.2 is 768 MPa√mm, which is
very similar to the results provided by contours 2 and 3. The next figure shows a comparison
between analytic values and XFEM results for contours 2 and 3 and the average between both
contours for the different crack lengths.
4500
Stress Intensity Factor (MPa√mm)
4000
3500
3000
2500
2000
1500
1000
500
0
0 0.2 0.4 0.6 0.8
Crack length / Panel width
Analytic XFEM_contour2 XFEM_contour3 XFEM_contour2+3
Figure 3: Edge crack. SIF results comparison

As it can be seen in the previous figure very accurate SIF results are obtained in contours 2 and 3,
so the average value between both contours can be taken as reference result.

Crack Growth analysis
The next figure shows the comparison of the crack growth results for XFEM and analytic Linear
Elastic Fracture Mechanics calculation.
60
50
Crack length (mm)
40
30
20
10
0
0 2000 4000 6000 8000 10000
Applied load cycles
Analytic XFEM
Figure 4: Edge crack. Crack Growth results comparison
A very good correlation is observed between both methodologies.
3.2 Hole cracks

Riveted joints are the most common configuration for structural attachments between aircraft
components. Due to this fact rivet holes are present in most aeronautical structures. The high stress
concentrations that appears at holes makes them susceptible to crack initiation what makes
necessary to perform analyses of cracks growing from holes.
In this case two different configurations will be analyzed: unloaded and loaded hole.
Crack growing from unloaded hole

An open hole that is not transmitting load will be analyzed. The selected configuration is shown in
the next figure:

100 MPa
t=3.0mm
D=4.8mm
A C=A 100mm
Symmetric crack
50mm
Figure 5: Unloaded hole crack scenario

Figure 6: Unloaded hole crack. Abaqus model

SIF values have been calculated for several crack length / hole diameter ratios (from 0.5 to 3.0).

The next table shows the detailed results obtained for a crack length / hole diameter ratio of 1.0.
Table 2: Detailed SIF results for crack length / hole diameter = 1.0

(MPa√mm) 1 2 3 4 5
1 786 518 552 561 564
2 790 506 537 546 550
Node 3 802 520 549 558 560
4 790 506 537 546 550
5 786 518 552 561 564
XFEM results will be compared with the analytic results obtained from NASA/FLAGRO Hole
Crack solution (JSC-22267, 1986) with the superposition of Rooke and Cartwright solution 1.1.5
(Rooke and Cartwright, 1974) to take into account the finite plate width effect.
The analytic SIF result for the crack length / hole diameter ratio of 1.0 is 518 MPa√mm, which is
again very similar to the results provided by contours 2 and 3.
The next figure shows a comparison between analytic values and XFEM results for contours 2 and
3 and the average between both contours for the different crack lengths.
1200
1000
SIF (MPa√mm)
800
600
400
200
0
0 0.5 1 1.5 2 2.5 3
Crack length / Hole diameter
Figure 7: Unloaded hole crack. SIF results comparison

Average value between contours 2 and 3 provides the most accurate solution.
Crack growing from loaded hole

A hole that is transmitting load from a rivet will be analyzed. The selected configuration is the
same that was analyzed for unload hole but adding the rivet load. Rivet load will be applied to the
hole surface through a pressure distribution.
100 MPa
D=4.8mm
t=3.0mm
1000N
A C=A 100mm
50mm
120 MPa
Figure 8: Loaded hole crack scenario


Figure 9: Loaded hole crack. Abaqus model
SIF values have been calculated for several crack length / hole diameter ratios (from 0.5 to 3.0).
The next table shows the detailed results obtained for a crack length / hole diameter ratio of 1.0.
Table 3: Detailed SIF results for crack length / hole diameter = 1.0

(MPa√mm) 1 2 3 4 5
1 806 679 717 697 712
2 825 687 728 711 729
Node 3 844 707 744 728 744
4 825 687 728 711 729
5 806 679 717 697 712
The analytic SIF result for the crack length / hole diameter ratio of 1.0 is 672 MPa√mm (obtained
using the same methodology that was used for unloaded hole), which is very similar to the results
provided by contour 2.
The next figure shows a comparison between analytic values and XFEM results for contours 2 and
3 and the average between both contours for the different crack lengths.

1600
1400
1200
SIF (MPa√mm)
1000
800
600
400
200
0
0 0.5 1 1.5 2 2.5 3
Crack length / Hole diameter
Figure 10: Loaded hole crack. SIF results comparison
Good correlation is observed, especially for small crack sizes. For large cracks XFEM results are
more conservative than analytic approach.
Multicrack scenario
Riveted structural attachments are done through a set of fasteners, so several holes are made at the
same structural component. To prevent widespread fatigue damage issues, scenarios with several
cracks growing from different holes have to be evaluated. An evaluation of XFEM capabilities to
deal with multicrack scenarios will be done in this section. A scenario consisting on two
embedded cracks on a plate will be analyzed (this is a simplification of a scenario with cracks
growing from two different holes when cracks are big enough to ignore the effect of the hole).
The next figure shows the analyzed configuration.

100 MPa
t=3.0mm
10 10
100
Tip Tip
50
Figure 11: Multicrack scenario

The next figure shows the comparison of the crack growth results for XFEM and analytic Linear
Elastic Fracture Mechanics calculation (superposition of Rooke and Cartwright solutions 1.1.5 and
1.2.3, (Rooke and Cartwright, 1974)).
14
12
10
Tip length (mm)
8
6
4
2
0
0 2000 4000 6000 8000 10000
Cycles
Tip1-XFEM Tip1-Analytic Ti1C-XFEM Tip1-Analytic
Figure 12: Multicrack scenario. Crack growth results comparison

A very good correlation between analytic and XFEM crack growth results is observed.

4. Conclusions
In this paper a set of practical cases from simple configurations to complex applications have been
evaluated using the XFEM capabilities implemented in Abaqus 6.14.
This methodology has demonstrated to be robust and reliable when obtained results obtained for
Stress Intensity Factor calculation and Crack Growth analysis have been compared with
alternative approaches. These results allows being confident about the potential of XFEM to
become in short period of time an alternative industrial standard approach in the Fracture
Mechanics field.
It would report clear benefits to the aeronautical industry as the improvement of the accuracy of
crack growth analysis directly reduce aircraft weight, as lighter structures can be designed by
reducing conservative approaches, and maintenance costs, as more accurate crack growth analysis
lead to lower structural inspection requirements.
However XFEM methodology currently available in Abaqus still requires development and
improvement of its capabilities. Limitations associated to the methodology when dealing with
contacts between components, material plasticity or complex loading spectra must be overcome to
provide an analysis tool able to deal with every aspect of a complex Fracture Mechanics analysis.
5. References
1. Du, Z. eXtended Finite Element Method (XFEM) in Abaqus. ©Dassault Systèmes.
2. JSC-22267. Fatigue and Crack Growth Computer Program NASA/FLAGRO. 1986
3. Park, Jun. Failure Prediction without Prescribing Crack Paths by using XFEM in Abaqus.
©Dassault Systèmes. 2012
4. Rooke, D.P., Cartwright, D.J. Compendium of Stress Intensity Factors. 1974
6. Acknowledgments
This paper is a result of the fructiferous collaboration between the fatigue departments of Airbus
Defence and Space and Safran Engineering Services. Going on with this collaboration will surely
come out in great results.
The authors also want to acknowledge the support provided by Ginés Boluda, Eduardo Oslé,
Ruben Tejerina and Fernando Sánchez from Airbus Defence and Space. Their contribution has
made possible to match the objectives proposed for this work.
The authors want to thank Principia for all their support with the use of XFEM capabilities
implemented in Abaqus.

Bird Strike Analysis for Impact-Resistant Design of
Aircraft Wing Krueger Flap
Sebastian Heimbs 1, Wolfgang Machunze 1, Gerrit Brand 1, Bernhard Schlipf 2
1
Airbus Group Innovations, 81663 Munich, Germany
2
Airbus Operations GmbH, 28199 Bremen, Germany
Abstract: Bird strike is a severe high velocity impact load case for all forward-facing aircraft
components and a major design driver due to the high energies and the strict safety requirements
involved. This paper summarises an experimental and numerical study to design a bird strike-
proof lightweight metallic Krueger flap as a high-lift device concept for a laminar wing leading
edge of a single aisle short range aircraft. The whole design process was based on numerical
optimisations for static load cases in combination with high velocity bird impact simulations, with
the focus on accurate modelling of the fluid-like bird projectile, the plasticity of the aluminium
material and the failure behaviour of the structural hinges and fastened joints. Finally, a full-scale
Krueger flap prototype was manufactured and tested under bird impact loading, validating the
numerical predictions and impact resistance.
Keywords: Bird strike, impact simulation, aircraft Krueger flap, gas gun test.
1. Introduction
Much research effort in aeronautics is currently dedicated to achieve a laminar flow wing for
transport aircraft, which significantly reduces air drag and hence fuel consumption. Laminar flow
requires the avoidance of any unevenness of the wing surface that could cause flow turbulences.
Since aircraft wings need high-lift devices to increase lift during low speeds of flight, extendible
slats are the most common leading edge high-lift devices, which involve a flow-disturbing step at
their trailing edge (Fig. 1a). Therefore, the more appropriate choice for a leading edge device
targeting at laminar flow is a Krueger flap, which is stowed on the lower side of the wing and
enables an undisturbed surface at least on the wing’s upper side (Fig. 1b) (Schlipf, 2011; Schlipf,
2013). The basic concept of the Krueger flap was invented by the German aerospace engineer
Werner Krueger in 1943 and was adopted in several commercial aircraft (e.g. Airbus A300, A310,
Boeing 707, 727, 737 and 747).
The design and sizing process of such a Krueger flap against structural failure with the target of
weight minimisation is based on typical flight loads and on particular risk load cases. As a
forward-facing component, the risk of bird strike impact is hence relevant both for structural
design and for part certification. Indeed, it turned out that bird strike is the design-driving load
case that requires structural reinforcements, which would not be necessary for conventional flight
loads.

step
(a) (b)
Figure 1. Illustration of (a) conventional slat and (b) Krueger flap as wing leading
edge devices.
Bird strike is the major high velocity foreign object impact load case for aircraft with more than
90,000 reported cases between 1990 and 2008 solely in the USA (Dolbeer, 2009). The collision
with a bird during flight can lead to serious damage to the aircraft. Consequently, the aviation
authorities require that all forward-facing components have to prove a certain level of bird strike
resistance before they are allowed for operational use. This can either be done by physical bird
strike tests with a gas gun (Bedrich, 1996) or by sufficiently validated finite element (FE)
simulations. Numerical methods are increasingly being used today for structural bird strike
analyses as an efficient and cost-effective alternative to full-scale tests (Heimbs, 2011a). Bird
strike simulation studies of conventional leading edge slats (Machunze, 2008; Heimbs, 2011b)
and of trailing edge flaps (Ritt, 2009; Smojver, 2010) can be found in the published literature, but
no such published studies seem to exist for Krueger flaps.
This paper presents a numerical and experimental study for the bird strike-proof design of a
metallic Krueger flap concept for a laminar wing of a transport aircraft. The focus is on the
accurate modelling of the fluid-like bird impactor, the plasticity of the aluminium material and the
failure behaviour of the kinematic hinges and fastened joints. Finally, a full-scale Krueger flap
prototype was manufactured and tested under bird impact loading for validation.
2. Krueger flap design principles

The Krueger flap in the focus of this study is supposed to be located on an outer position of the
wing and was designed as a metallic solution made of aluminium. It basically consists of outer
skins (top and bottom skin with leading edge) and an internal reinforcement structure containing
spars and ribs (Fig. 2). Metallic bolts and rivets were used to join different parts. The internal
structure needed to be designed, on the one hand, to include the necessary system components
(e.g. for de-icing) and, on the other hand, to carry the flight loads and bird impact loads. At the
same time, the design goal was to reduce the structural weight to a minimum. Hence, especially
the optimized selection of sheet thicknesses was the target of the numerical analysis. Since major
parts of the structure were supposed to be manufactured by milling, different thicknesses at
different positions of the flap can easily be implemented. The thicknesses should be as small as
possible to reduce the structural weight, but at the same time as large as necessary to withstand the
bird strike impact at various possible impact positions. The criteria for successful bird impact
resistance are twofold. Firstly, the kinematic parts and attachments of the flap may not fail or the
whole flap or parts of the flap may not break off. In such a case the detached parts of the flap

Krueger flap wing leading edge
bird
Figure 2. Krueger flap concept treated in this study.
might hit the empennage of the aircraft (vertical and horizontal tail plane) and lead to a critical,
uncontrollable flight status. Secondly, the damage tolerance requirements for the damaged
structure after impact necessitate carrying so-called ‘get-home loads’, which is a reduced load case
for safe landing at the next airport. The get-home load case is typically analyzed in a post-impact
simulation run, where – as a conservative approach – all damaged elements are initially removed
from the simulation model.
3. Model development for bird strike analysis

The simulations within this study were performed with the commercial FE software
Abaqus/Explicit 6.13. The following sections give some brief information on the most important
modelling issues, which are the nonlinear material modelling, the modelling of the kinematic
hinges and the modelling of the fastened joints, including failure for all those three cases. Finally,
the accurate modelling of the soft body projectile for the bird strike simulation is discussed.
3.1 Material modelling

All major parts of the Krueger flap are made of aluminium. The ductility and plastic deformation
of this material permits a high energy absorption capability. Material characterisation tests were
performed under both quasi-static and high-rate dynamic loading conditions on a servo-hydraulic
test machine to fully characterise the nonlinear material behaviour at various loading rates (Fig. 3).
The material plasticity and potential failure need to be covered adequately in the simulation
model. For this purpose, the yield stress vs. plastic strain was defined using tabular input to
represent the experimentally determined yield curve. Ductile damage initiation and damage
evolution were also implemented according to the test results. At the final point of failure the
element is eroded from the simulation. Strain rate effects were not implemented as the test results
and further literature data (Rodríguez-Martínez , 2011) prove that the material behaviour at the
velocities of interest is not strain rate-sensitive (Fig. 3).

Figure 3. Material characterization of aluminium AA 2024 for constitutive
modelling.
3.2 Kinematic hinges

Kinematic hinges exist at different positions of the Krueger flap and enable the rotation during
flap extension using either cylindrical or spherical hinges. The simulation model was not only
supposed to represent the degrees of freedom for rotation correctly, but also to cover potential
failure of the hinges. The most versatile approach to model such hinges in Abaqus is the use of
connector elements. They connect specific points and allow for the locking or unlocking of
individual translational and rotational degrees of freedom. Failure can either occur due to failure
of the bolt inside a hinge or due to failure of the lug of the kinematic part. Lug failure was
captured by detailed modelling of the parts with solid elements including material models with
failure (Fig. 4). Bolt failure by bending or shear loads was implemented into the connector
element by defining maximum loads, after which the element is removed. Even connector stops
could be defined easily for the connector elements, representing limits of rotational angles.
connector element lug
Figure 4. Modelling of kinematic hinges.

3.3 Fastened joints
Fastened joints based on bolts or rivets are mainly loaded in shear or normal tension and exhibit
elastic and failure behaviour depending on bolt and sheet material and thickness (Heimbs, 2013a;
Heimbs, 2013b). Again, connector elements are the most suitable approach in Abaqus to model
such joints connecting two or more parts at node-independent, discrete locations. Elastic, plastic
and damage behaviour can be ascribed to the six degrees of freedom in the connector section
definition. Elastic translational behaviour for tension and compression was defined based on the
equations of Gray and McCarthy (Gray, 2011; Gray 2012), which take into account the stiffness
and thickness of the bolt and the joining material. The definition of the connector strength,
plasticity and damage behaviour strictly depends on the failure mode, e.g. bearing, net tension,
pull through, fastener failure, etc. Therefore, for a general analysis, the prediction of failure mode
needs to be performed first based on respective equations for failure loads (e.g. given in the
literature (Schwarmann, 2003)) and the identification of the load case with the lowest failure load.
In the current study, bearing failure under shear loads and pull-through failure under normal loads
are the most relevant failure modes. The calculated failure loads indicate the beginning of
plasticity of the connector behaviour, see point A in Fig. 5. A simplified linear approach for the
description of the plastic deformation was used here by defining point B by a value of plastic
displacement and force. This point indicates the onset of damage (damage initiation). The
progression of damage (damage evolution) up to point C is represented by a linear softening
behaviour, characterised by a maximum displacement value when the connector element is
removed from the calculation. The values for the description of the plastic and damage behaviour
(points B and C) were derived from experimental force-displacement curves of joint failure tests in
aluminium AA 2024 plates (Langrand, 1999; Birch, 2005).
Figure 5. Illustration of elastic, plastic and damage behaviour of fastener element.
3.4 Bird projectile

When the bird hits the Krueger flap at the velocities of interest, it flows in a fluid-like manner over
the target structure with the large deformations of the spreading material being a major challenge
for FE simulations. Different numerical methods exist to model this fluid behaviour (Heimbs,
2011a). In this study, the smoothed particle hydrodynamics (SPH) approach was used, which is a
meshless Lagrangian technique with the fluid being represented by a set of discrete interacting
particles that can exhibit large deformations without the numerical problem of mesh distortion.

Since real birds are mostly composed of water, a water-like hydrodynamic response can be
considered as a valid approximation for the constitutive modelling of the bird projectile. A Mie-
Grüneisen-type equation of state (EOS) of the type us-up with parameters of water at room
temperature was used in this study.
For a separate validation of the projectile model independent of the target structure, impact test
results of birds being shot against rigid plates are typically used (Liu, 2014) , where the pressure or
force vs. time response is compared to ensure an accurate loading and stationary flow behaviour.
For this purpose, in-house test data from bird strike tests on rigid plates and rigid flap-like edges
were used to verify both the correct load-time history and the correct splitting of the SPH bird
model upon impact, which is of high relevance for the impact on the Krueger flap.
4. Bird strike simulation results

Based on these validated sub-models the final bird strike simulation model of the whole Krueger
flap including wing leading edge and bird impactor was generated and used for impact simulations
and various parameter studies. The impactor was a 4 lb bird with a velocity of 128 m/s. A total
time of 15 ms was simulated.
In a first study, the influence of the impact point in terms of horizontal and vertical variation was
assessed. Although the impact on the lateral positions close to the left and right edge are
challenging in terms of bending moment initiated in the Krueger flap, the impact directly in the
middle turned out to be the critical horizontal load case as the forces in the attachment struts and
the flap itself were the highest. During the variation of the vertical position, the impact onto the
Krueger flap’s front spar appeared to be the critical vertical load case, as the front spar is the major
load-carrying member assuring structural integrity and impact damage tolerance. It was decided to
use this combination of vertical and horizontal critical load case for the final real bird strike test. A
cross-sectional view of this simulation is shown in Fig. 6.
0 ms 0.75 ms 1.5 ms
2.25 ms 3 ms 7.5 ms
Figure 6. Cross-sectional view of bird strike on Krueger flap.

Nevertheless, the impact simulations on all other positions were used for the sizing of the metallic
part thicknesses targeting at compliance with the design goals of impact resistance and residual
strength. The thicknesses were optimized in a series of numerical simulations in order to assure
static strength (linear static FE analysis), bird impact resistance (explicit FE analysis) and post-
impact residual strength (linear static FE analysis with removal of damaged elements) with
minimum weight, see Fig. 7.
Although plastic deformation occurs at the flap’s leading edge, structural integrity is assured.
Neither the kinematic hinges nor the fastened joints exhibit critical failure loads. The final design
of the Krueger flap, derived from these simulations, was then used for the prototype
manufacturing for a real bird strike test.
1. Static strength analysis

for regular flight loads
(ultimate load)
2. Bird strike analysis for

investigation of structural
damage and fastener damage
Transfer of damage to static FE model Re-sizing (e.g. thickness increase)
3. Residual strength analysis

for get-home load case
considering damaged areas
removed elements
Residual strength ok? no

yes
Feasible design
Figure 7. Iteration loop for get-home load verification.

5. Full-scale bird strike testing
For validation of the numerical predictions and of the impact-proof design, a bird strike test was
performed at IABG with a full-scale prototype of the metallic Krueger flap. The flap was mounted
to generic metallic ribs, which were then attached to a rigid wall with a mass of 10 tons (Fig. 8).
The 4 lb bird projectile was accelerated in a gas cannon to an impact velocity of 128 m/s. The
target point was in the center of the Krueger flap being the worst case impact scenario.
Neither the Krueger flap nor any attachment parts failed under this impact load and no fastener
failure was observed. Moderate plastic deformations of the metallic flap occurred at the location of
impact, similar to the pre-test simulation results. The numerically predicted structural
deformations, reaction forces at the attachment points and strains on the upper and lower flap
surface are in close agreement to the test results, proving the bird strike resistance of the Krueger
flap and the accuracy of the numerical pre-test simulations (Fig. 9, Fig. 10).
Figure 8. Full-scale bird strike test set-up with gas cannon and rigid wall.
Figure 9. Maximum deflection of Krueger flap under bird impact (from high speed
video).

0 ms 3 ms
6 ms 15 ms
Figure 10. Simulation of bird strike on Krueger flap including realistic test set-up.
6. Conclusion
A procedure was presented to design and size a metallic Krueger flap for a laminar wing
application against bird strike loads using advanced numerical methods. The approach was based
on step-by-step validation of the modelling methods using experimental data. The major
modelling aspects turned out to be the nonlinear material modelling, the kinematic hinges
modelling, the fastened joints modelling and the hydrodynamic bird impactor modelling. The final
design that was derived from the impact simulations has proven bird strike-resistance in a real full-
scale gas cannon impact test.
The whole procedure using numerical methods for the design and sizing significantly increases the
efficiency of structural developments compared to conventional approaches that are mainly based
on tests. However, further improvements are possible in terms of multi-disciplinary optimization,
combining the three simulation steps (static strength analysis, impact analysis and residual
strength analysis) for material thickness optimizations in an automatic process to obtain more
quickly the final design with less manual interaction.
7. References
1. Schlipf, B., Payne, C., Gibbert, M., Dostmann, T., “Wing with a Krueger Flap and a
Gap Covering Device, and a Corresponding Method”, Patent EP 2356021 B1, 2013.
2. Schlipf, B., “Insect Shielding Krüger - Structural Design for a Laminar Flow Wing”,
DGLR Congress 2011, Bremen, 55-60, 2011.

3. Dolbeer, R.A., Wright, S.E., Weller, J., “Wildlife Strikes to Civil Aircraft in the
United States 1990-2008”, FAA National Wildlife Strike Database, Serial Report
Number 15, 2009.
4. Bedrich, L., Mertl, V., “Bird Impact Resistance of the Small Transport Aircraft”,
ICAS-96 - 20th Congress of the International Council of Aeronautical Sciences,
Sorrento, 1996, 175-182.
5. Heimbs, S., “Computational Methods for Bird Strike Simulations: a Review”,
Computers & Structures, 89, 2011(a), 2093-2112.
6. Heimbs, S., “Bird Strike Simulations on Composite Aircraft Structures”, SIMULIA
Customer Conference, Barcelona, 2011(b), 73-86.
7. Machunze, W., Middendorf, P., Keck, R., Roth, Y.C., “Design, Analysis and
Manufacturing of a Thermoplastic Slat”, EUCOMAS - European Conference on
Materials and Structures in Aerospace, Berlin, 2008, 121-129.
8. Smojver, I., Ivancevic, D., Numerical Simulation of Bird Strike Damage Prediction
in Airplane Flap Structure”, Composite Structures, 92, 2010, 2016-2026.
9. Ritt, S.A., Dobeneck, T., “High-velocity Impact Studies with Parametric High-lift
Devices”, ANSYS Conference & 27. CADFEM Users Meeting, Leipzig, 2009.
10. Rodríguez-Martínez, J.A., Rusinek, A., Arias, A., “Thermo-viscoplastic Behaviour of
2024-T3 Aluminium Sheets Subjected to Low Velocity Perforation at Different
Temperatures”, Thin-Walled Structures, 49, 2011, 819-832.
11. Heimbs, S., Schmeer, S., Blaurock, J., Steeger, S., “Static and Dynamic Failure
Behaviour of Bolted Joints in Carbon Fibre Composites”, Composites Part A, 47,
2013(a), 91-101.
12. Heimbs, S., Hoffmann, M., Waimer, M., Schmeer, S., Blaurock, J., „Dynamic
Testing and Modelling of Composite Fuselage Frames and Fasteners for Aircraft
Crash Simulations”, International Journal of Crashworthiness, 18, 2013(b), 406-422.
13. Gray, P.J., McCarthy, C.T., “An Analytical Model for the Prediction of Load
Distribution in Highly Torqued Multi-bolt Composite Joints”, Composite Structures,
93, 2011, 287-298.
14. Gray, P.J., McCarthy, C.T., “An Analytical Model for the Prediction of Through-
thickness Stiffness in Tension-loaded Composite Bolted Joints”, Composite
Structures, 94, 2012, 2450-2459.
15. Schwarmann, L., “HSB – Fundamentals and Methods for the Design and Evaluation
of Aeronautical Structures”, IASB, London, UK, 2003.
16. Birch, R.S., Vendrasco, L., White, M.D., Mahe, M., “Modelling the Dynamic Failure
of Riveted Joints in Aerospace Fuselages”, in: Impact Loading of Lightweight
Structures, M. Alves, N. Jones (eds.), WIT Transactions on Engineering Sciences, 49,
WIT Press, Southampton, UK, 2005.

17. Langrand, B., Deletombe, E., Markiewicz, E., Drazetic, P., “Numerical Approach for
Assessment of Dynamic Strength for Riveted Joints”, Aerospace Science and
Technology, 3, 1999, 431-446.
18. Liu, J., Li, Y., Gao, X., „Bird Strike on a Flat Plate: Experiments and Numerical
Simulations”, International Journal of Impact Engineering, 70, 2014, 21-37.
8. Acknowledgment
This work was conducted in the framework of the projects LuFo IV-3 HILIFT and LuFo IV-4
MOVE.ON focusing on laminar wing research, funded by the German Federal Ministry for
Economic Affairs and Energy (BMWi). The financial support is gratefully acknowledged.

Use of Abaqus/Explicit for Composite Sandwich
Damage Prediction during Bird Impact
M. Al-Khalil, E. Kirtil, R. Rigby
Airbus
Abstract: A method development program of testing and simulations was carried out to develop
bird impact NLFEA capabilities of composite sandwich damage prediction and bird dispersion
after penetration of the primary layer using Abaqus/Explicit. Curved composite honeycomb panels
(referred to as J-Nose) representing typical composite wing Fixed Leading Edge (FLE) structure
were subjected to bird strike to generate data for method validation. The test campaign was
tailored to produce various levels of damage and modes of failure; from minor localized core/skin
damage to panel perforation. In all tests high speed video were installed to capture the behavior
of the structure during the impact. In some tests, fully instrumented witness panel was installed to
enable measurement of deformation as a measure of bird residual energy and dispersion after
bird penetration of the sandwich panel. FE models were created of the tested structure and
analyzed using ABAQUS Explicit and the results were compared with tests. The analysis
performed used a standard Lagrangian bird model developed and validated at Airbus over many
years. In general, good correlation was obtained between tests and simulations results. It is known
that Lagrangian bird models suffer from numerical instability in case of impacting sharp edges
due to element distortion. To avoid such numerical difficulties, the SPH capability available in
Abaqus/Explicit was used to generate a representative SPH bird model. Test cases were re-
analyzed with the SPH bird model and results compared to the Lagrangian approach. A summary
of bird impact simulations and correlation against tests is presented in this paper.
Keywords: Aircraft, Wing, Leading, Edge, Bird, Damage, Impact, Composites, Sandwich
1. Introduction
Bird impact incidents on aircraft are not uncommon and they can cause structural and system
damage. Although measures have been implemented to prevent bird impacts from occurring, it is
impossible to prevent them totally. Further details regarding statistical reports related to bird
impact incidents can be found in [1-3]. All forward facing components of an aircraft are most
susceptible to bird strike such as leading edges of the wings and empennage, engine fan blades and
inlet, windshield, window frame, radome, landing gear and forward fuselage as illustrated in
Figure 1. The aviation authorities require that all aircraft structures in the vulnerable zones must
be designed to assure capability of continued safe flight and landing of the aircraft after bird
impact (EASA CS 25.631).

Figure 1 Illustration of aircraft components exposed to the risk of bird impact
Bird impact on aircraft structure is a complex phenomenon because of the high number of
variables involved. Such variables include material characteristics of the impacted structure and
bird behavior and its interaction with the aircraft structure i.e. accurately predicting the level of
pressure/forces transmitted to the structure. Traditionally, aerospace manufacturers rely heavily on
tests, supported by empirical equations, for design and certification of aircraft subjected to bird
impact. Such methods are not only time-consuming and expensive but they could lead to
conservative design, resulting in weight penalty. In addition, such tests are subject to a degree of
uncertainty due to the lack of consistency in bird impact test results for the same bird strike test
configuration.
With the increase of finite element software tools capabilities and hardware computing power, it is
now becoming more realistic to use such tools to develop finite element methods for predicting the
behavior of aircraft structures subjected to bird impact. However, it is important to emphasize that
the analysis/numerical techniques must be validated through correlation with tests on sufficiently
representative structure.
For almost two decades, Airbus has been involved in the development of bird strike
methodologies for aircraft structures, (mainly for metallic but more recently for composite) and
works closely with other Aerospace partners. A typical example of bird strike simulation and

correlation with test on generic fixed leading structure, shown in Figure 2, was performed in
collaboration with SONACA Aerospace manufacturer.
Figure 2 Simulation (Abaqus/Explicit) and test of bird impact on generic fixed leading
edge structure
Composite materials including sandwich construction are increasingly being used for aircraft
structures driven by high stiffness and strength to weight ratios. One of the main challenges of
using composite structure is their susceptibility to impact damage. Therefore, understanding their
behavior and in particular failure mechanisms is critical if the full benefit of using composites is to
be exploited.
Bird impact nonlinear finite element methodology development and validation for composite
sandwich J-Nose structure (situated in the wing leading structure) was performed using ABAQUS
Explicit. Tests were performed to generate data for correlation with simulations. The analysis
performed used a standard Lagrangian bird model developed and validated at Airbus. As part of
ongoing bird modelling enhancement, Smooth Particle Hydrodynamics (SPH) capability within
ABAQUS Explicit V613.3 was used and correlated with tests and Lagrangian approach. This
paper outline a summary of the work performed including bird impact simulations and correlation
with test results.
2. Method Validation
A bird strike test campaign was performed on a number of composite honeycomb panels,
representative of a typical composite Fixed Leading Edge structure referred to a J-Nose. The aim
of the test campaign was to provide test data for correlation with Abaqus/Explicit for method
validation in particular for a) material failure prediction and b) predictive capability for secondary
impact and dispersion after bird penetration of the J-Nose. It is important to note the impact
scenarios considered such as impact angle, bird impact velocities and impact locations do not
represent any specific aircraft flight configuration. The test campaign was tailored to produce
various levels of damage and modes of failure; from minor localized core/skin damage to panel
perforation. In all tests, high speed video cameras were installed at various positions to capture the
behavior of the structure and bird dispersion during the impact. Bird mass of 4 lb (1.81 kg) was
used for all the test campaign and this is in-line with the certification requirement of wing leading
edge structure. In some tests, a fully instrumented witness panel was installed to enable
measurement of deformation as a measure of bird residual energy and dispersion after bird

penetration of the J-Nose. Finite element models were created of the tested structure and analyzed
using a standard Lagrangian bird. However, SPH bird was also used for the test cases related to
secondary impact and the simulation results are presented in Section 3. It is not possible to present
all the results and therefore a sample of simulations and correlation with tests will be briefly
outlined in this paper.
ABAQUS /CAE 6.12-1 together with ALTAIR Hypermesh were used in preparation of the finite
element model and post-processing of the results. Abaqus/Explicit V6.12-1 was used for the
simulation. However, for the SPH bird, Abaqus/Explicit V6.14-1 was used.
2.1 Finite Element Model
Model and Structure Description: A typical composite sandwich wing fixed leading edge
structure test specimen is shown in Figure 3. It included the J-Nose, ribs, front spar, lower panel,
upper and lower covers. The front spar, ribs were made of aluminium alloys. The J-Nose sandwich
panel was constructed from carbon woven fabric with a honeycomb core. The upper and lower
covers were made of Carbon UD. The J-Nose is supported at the ribs and top & bottom covers and
butt-strap at the lower panel.
Top Cover
Ribs
Front
Spar
J-Nose Buttstrap
Bottom Cover Lower Panel Rib
Figure 3 FE model/geometry of tested specimen
The J-Nose and Ribs were meshed using elements with edge lengths of approximately 5mm and
further back the covers and spar were meshed with elements with edge lengths of approximately
10mm. The top and bottom covers were modelled as continuum shell element (SC8R) while the
metallic parts ribs and front spar were modelled as shell element S4R, shown in Figure 3. The
fasteners attaching various parts of the components were also included in the model using fastener
modelling capabilities within Abaqus. For the J-Nose, the skins were modelled with shell element
S4R and the honeycomb was modelled using solid element C3D8R, see Figure 4. It was necessary
to use triangular element and wedge element, mainly for mesh transition or corners but this was
kept to a minimum. The honeycomb solid elements should be roughly cubic within practical
limitations. It is assumed that the adhesive interface between skins and honeycomb is stronger

than the honeycomb material which has been borne out by coupon testing. Therefore, a surface-to-
surface tied interface is used to connect the skin shell elements to the solid honeycomb elements.
Also, Surface-to-surface ties are used to connect the top and bottom skins together at the panel
sides and at the edge of the honeycomb ramp.
Solid element
core
Conventional
shell skin
Figure 4 Modelling of the sandwich panel – J-Nose
The FE model included the full extent of the upper and lower covers, which were fully built-in in
the simulation at the points where they connected to the test rig as shown in Figure 5. It was
judged that these points were sufficiently far from the impact locations and the test rig is
considered to be rigid.
Figure 5 Specimen set up in the test rig

Material Definition: A simple orthotropic material model is used for the honeycomb which has
been found to work reasonably accurately for Non-Linear Finite Element Analyses. This material
model has an orthotropic elastic behavior by defining the engineering constants and also has a
plastic curve to represent the compression curve of the honeycomb. Element deletion is included.

The compression true stress and true plastic strain values are input as absolute values. It is
important to note that the compression curves should not include the initial peak which can occur
when compressing honeycomb, shown in Figure 6, as Abaqus cannot cope with the drop in stress
for orthotropic material model. This assumption is reasonable for bird strike where cracking and/or
penetration of the sandwich panel is the dominant phenomenon. However for hail or hard debris
impact a more refined methodology is required and method development work is ongoing within
Airbus to exploit the advanced honeycomb VUMAT capability available within Abaqus. This
would allow the peak stress in the compression curve to be retained as well as catering for non-
linear shear curves. This will enable a more accurate representation of honeycomb material
behavior which may be beneficial for hard debris impacts.
Figure 6 Compression curves with and without initial peak
The Fabric material model is used for woven (J-Nose skin) and is available as a built-in VUMAT
user subroutine in Abaqus Explicit. It can be accessed by naming the material such that it begins
with ABQ_PLY_FABRIC string. Non-linear in-plane shear stress-strain properties have been
taken into consideration in the fabric material model. The upper and lower covers were modelled
as a laminate of uni-directional plies with relevant properties in the 1 and 2 directions. True stress-
strain curves up to failure for the relevant aluminium alloy were used in the analysis.
Contact Definition: General contact was defined for the whole structure so that all components
can interact with each other and transfer load through the structure. A friction coefficient of 0.3
was used for the interaction between the structural components, though relative movement of these
components was not anticipated.

2.2 Bird Model
Observations from high-speed film recordings have shown that the bird disintegrates upon impact
and that thereafter its behavior approximates to that of a fluid. The bird model consists of a
cylinder with hemispherical ends and is shown in Figure 7 for a bird of 4 lb (1.81 kg). The length
of the Bird is approximately twice its diameter. The analysis performed used a standard
Lagrangian bird model developed and validated at Airbus over many years. The recommended
element type is the linear hexahedral, reduced integration C3D8R element.
Figure 7 Bird Model
As bird flow commences on impact, the behavior of the bird is governed by an Equation Of State
which relates the pressure to the density and internal energy. It is necessary to define the failure
criteria that will determine how the bird "flows". It is recommended that a small shear modulus, G,
be defined, along with tensile failure, damage initiation and damage evolution, in order that
elements, and therefore energy, can be removed from the bird material as it flows.
The general contact algorithm is used since it allows for a very simple definition of contact whilst
using sophisticated tracking algorithms to ensure that proper contact conditions are enforced
efficiently. In the analysis, it is allowed the interior of the bird to contact itself as elements erode.
The contact between the bird and the structure is considered as frictionless.
2.3 Simulation Results and Correlation with Tests – Lagrangian Bird

High speed video cameras were positioned at strategic locations to provide necessary test
parameters such as impact location, bird orientation on initial impact and bird dispersion which
can be used in the simulation and correlation with tests. It is important to note the actual measured
velocity of the bird in the test was used in the simulations. As mentioned earlier, a sample of
results will be presented in this section.

2.3.1 Damage to J-Nose but with no Bird Penetration
The main objective of this test is to produce damage to the J-Nose without penetration. This test
was performed using the configuration shown in Figure 8 with impact energy of 4.4kJ.
30°
Figure 8 Bird strike set up showing the impact location on J-Nose
The NLFEA results correlated very well with the test as shown in the Figure 9. It can be seen that
a through-thickness split which occurred in the test correlates well with that predicted by the
analysis. The analysis captured energy level at onset of failure in panels with good accuracy. No
penetration of bird debris through the panel was either predicted by the analysis or observed in the
test. This impact scenario demonstrates that the analytical methods are able to capture the energy
level at onset of failure in panel with good accuracy.
Core crushing / debonding Permanent deformation
Simulation: Split
in the panel
Test: split in the panel
Figure 9 Damage to J-Nose but no bird penetration in both NLFEA and test

2.3.2 Damage to J-Nose but with Bird Penetration
The objective of this test is to produce rupture in honeycomb and monolithic laminate (i.e. bird
penetration). This test was performed using the configuration shown in Figure 10 where the
impact location is between the two ribs, in the middle of the J-Nose. The impact energy of this test
was 12.7kJ.
Figure 10 Bird strike set up showing the impact location on J-Nose
NLFEA predicted bird penetration of the J-Nose as observed in the test, shown in Figure 11. The
simulation also predicted the level of damage severity as seen in the test.
NLFEA (rupture of J-Nose) Test (rupture of J-Nose)
Figure 11 NLFEA simulation and test results correlation

2.4 Secondary Bird Impact on Witness Panel after J-Nose Penetration
In order to provide data for the developed methodology in terms of bird residual energy and
dispersion of the bird after penetration, tests were performed where a metallic witness panel (2
mm thickness) was installed behind the leading J-Nose. The impact scenario was set up as shown
in Figure 12. To ensure the bird impacted the witness panel after it penetrates the J-Nose, the
tested specimen had a sweep angle of 15° and Angle of Attack (-6°) and bird impact energy of
8.8KJ. These parameters do not represent normal flight conditions.
Figure 12 Impact location and witness plate installation
In both test and simulation, most of the bird penetrated the panel and showed the same level of
damage of the J-Nose and bird dispersion, as shown in Figure 13 at similar time points.
1.8 ms 3 ms 4.8 ms
Figure 13 NLFEA results (above) and test (below) results at equivalent time of impact

The energy of the bird debris impacting onto the witness plate was also accurately captured by the
analysis. This is demonstrated by the close correlation of witness plate out of plane permanent
deformation as shown in Figure 14. Simulation predicted maximum out of plane displacement
within 0.7% of the test.
Simulation Test
Figure 14 Predicted and test permanent deformation of witness plate
The dispersion or spread of the bird debris as it penetrated the J-Nose and moved through towards
the witness plate was also captured accurately by the simulation. This is demonstrated in Figure
14, which shows the bird dispersion in test and simulation at the time the bird first impacts the
witness panel.
Simulation Test
Figure 14 NLFEA and test bird dispersion on impacting witness plate
The overall trend of strains measured at the rear face of the witness plate during the impact
correlated well with the simulations, as shown for a typical gauge in Figure 15. This supports that
the analysis captured accurately the energy and location of debris impacting the witness plate.

Figure 15 Strain correlations (simulation and test)
Figure 16 show the kinetic energies obtained in the simulation. The first drop in energy is caused
mostly by bird impact with the J-Nose panel while the second decrease in energy is mainly due to
the bird impact with the witness plate. From high-speed footage, the time interval between the bird
initial impact of J-Nose and witness plate was estimated to be 8.5ms compared with 8 ms in
simulation. Therefore, the average bird velocity after impact with the J-Nose was predicted well
by the analysis. This coupled with the good correlation of the amount and dispersion of bird
debris penetrated into the J-Nose (shown in Figures 14 and 15) indicates that the kinetic energy of
the bird debris passing through the J-Nose, and therefore impacting onto the witness plate, was
accurately captured.
~2kJ Model Kinetic Energy

Bird Kinetic Energy
Witness Plate Kinetic Energy
~8ms
~5kJ
Figure 16 Model, bird and witness panel KE energy

3. Secondary Bird Impact on Witness Panel after J-Nose
Penetration - SPH Bird
The new SPH capability in Abaqus/Explicit was used to perform some correlation based on the
available test results. For this purpose a SPH bird model was generated with a particle
arrangement of constant distances based on the Lagrangian bird geometry. Variations in particle
number were carried out to identify the influence on the results. A good compromise in accuracy
and run time on simple targets was found with the configuration shown in Figure 17.
Figure 17 4lb SPH bird model
The material model was not changed and taken from the Lagrangian bird model. The analysis in
section 2.4 was repeated with this SPH bird model.
In the SPH bird simulation, localized rupture of the J-Nose is shown in Figure 18 and this
compared well with test and simulations using Lagrangian bird (Figure 13).
Figure 18 Rupture of the J-Nose after impact
Some bird debris deflected at the outer surface of J-Nose as shown in Figure 19, in a similar
manner to Lagrangian bird simulation (Figure 13). The main bird mass travelled inside the J-Nose
towards the witness plate causing it to plastically deform.

1.8 ms 3 ms 4.8 ms
Figure 18 Perforation of J-Nose using SPH bird
Localized rupture of the J-Nose is shown in Figure 19 and this compared well with test (Figure 13)
and simulations using Lagrangian bird (Figure 13).
Using the SPH bird, the permanent out-of-plane displacement was predicted to be 26% less
(Figure 20) compared to the test result. A potential reason identified was the deformed status of
the SPH bird model just before the impact of the witness panel which showed some deviation
compared to the Lagrangian bird model. For the SPH bird simulation, it was not possible to extract
the energy curves as was done for Lagrangian bird (Figure 16) due to the current limitation in the
code. Such correlation would have shed more light on the differences between the SPH and
Lagranian results.
Figure 20 Deformation of witness panel in the simulation after impact

Additional test correlation activities to provide a more robust and accurate SPH bird model
concerning bird strike are ongoing. Providing advantages in analysis run time and material
separation the SPH approach can be a candidate to replace the Lagrangian bird model in case of
more complex impact cases.
4. Conclusions
A program of test and simulation campaign has been briefly outlined in this paper for method
validation of composite sandwich structure damage prediction due to bird strike. The methodology
was able not only to predict the onset of damage but also the penetration of the J-Nose and bird

dispersion. The numerical methods, mainly based on Lagrangian bird, can be used with confidence
for analytical work to support design on similar sandwich panel construction provided the relevant
material data. Simulations were also performed using SPH bird and the results were correlated
with test and Lagrangian simulation results.
5. References
[1] FAA Report, “Wildlife Strikes to Civil Aircraft in the United States 1990–2013”, National
Wildlife Strike Database, July 2014.
[2] http://wildlife.faa.gov
[3] EASA Report, “Bird Strike Damage and Windshield Bird Strike”, EASA.2008.C49, 2009.
6. Acknowledgement
The authors would like to thank the contribution of David Coldbeck of Spirit AeroSystems
(Europe) Ltd in providing the tests data, Richard Gibbon of Frazer-Nash for parts of the
simulations and Etienne Maillard of SONACA for the valuable discussions on the simulations and
providing bird impact tests on generic fixed leading structure for correlation.

Time-Accurate Simulation of Aeroelastic Flap
Deployment with Free Play
Nicolas D. Reveles,* George Antoun,* James Fort+
*
ATA Engineering, Inc.
+
Dassault Systems Simulia Corp.
Abstract: Today, numerous aeroelastic structures are deployed over a finite time period, such as
flaps, spoilers, control surfaces, and wheel bay and bomb bay doors. These devices must be both
divergence and flutter free. Current state-of-art simulations verify this by applying a quasi-steady
assumption that does not require the system to physically deploy within the computations. That is,
the structure is assumed elastic but does not possess the “large” motion associated with the
device’s path of travel. While these simulations have strong technical merit, especially in the case
where a structure is inadvertently locked in a partially deployed position, they are unable to
capture all of the relevant physics. When a component is deployed over a finite period, the flow
physics include additional unsteady aerodynamic effects that are lost without considering the
large motion of the device; in addition to the unsteady aerodynamics, the structure has inertial
terms that cannot be correctly accounted for with the quasi-steady assumption. It will be
demonstrated that by coupling Abaqus/Standard, which simulates the elastic response and flexible
multibody dynamic articulation, to the CFD solver Star-CCM+ via Simulia’s Co-Simulation
Engine (CSE), a time-accurate response to flap deployment can be achieved. This is compared to
results from a simulation utilizing the quasi-steady assumption.
Keywords: Aeroelasticity, CFD Coupling, Connectors, Coupled Analysis, Dynamics, Elasticity,

Flap, Free Play, Multi-Body Dynamics
1. Introduction
The determination of the aeroelastic response of deployable structures is a challenging but
necessary multidisciplinary task. Common structures such as flaps, spoilers, control surfaces, and
wheel bay and bomb bay doors all contain multiple configurations that require actuators to deploy.
These different configurations may have substantially different aerodynamic effects acting on
them between the fully deployed and withdrawn positions due to the possibilities of vortex
shedding, separation, and stall. These nonlinear aerodynamic phenomena are responsible for
determining the structure’s aeroelastic stability margin, the structural loads, and the actuator force
required for timely deployment.
Modern aeroelastic analyses accomplish this by examining the structural response of the part at a
fixed deployment angle to computed aerodynamic loading (Blades & Cornish, January, 2015). To
accurately model the nonlinear effects of the aerodynamic loading, the unsteady Reynolds-
averaged Navier–Stokes (URANS) equations are typically resolved on a computational mesh.
These equations allow for the efficient solution of the fluid dynamics by modeling the turbulence

rather than resolving it. The URANS equations may become inappropriate in situations where
large separated flow exists and the turbulent time scales begin to approach the unsteady flow time
scales. Large-eddy simulation (LES) enables resolution of these large turbulent scales while still
modeling the fine-scale turbulence; however, it does so at greatly increased computational costs
that may not be suitable for engineering-level analyses today. On the other hand, detached-eddy
simulation (DES) is a hybrid approach especially suitable for separated flows. With DES, the
URANS equations provide closure in regions where the mesh is too coarse to resolve the largest
turbulent eddies (e.g., boundary layers), and LES computations are performed in refined regions
where the turbulent scales may be captured (e.g., separated flow on a fine mesh).
The physical motion of the part itself may induce unsteady aerodynamic effects that modify
aeroelastic stability margins. With the addition of the structural inertial terms, it is possible that the
dynamic response of the deployment may be significantly different from the quasi-static response
to the elastic structure held at a fixed deployment angle.
One challenge in solving CFD problems with large relative motion is how to best represent the
moving domains. Today, the so-called “chimera” or overset approach (Mitcheltree, et al., 1988) is
most commonly employed in these situations. This technique allows multiple overlapping grids to
define the domain of interest. When the meshes are moving, at each time step several steps must
occur: holes are cut in any mesh that overlaps a body, and mesh fringe points are identified and
marked as “donors” and “receivers,” depending on whether they are used to interpolate to or from
another mesh, respectively. Finally, the fringe points are updated as an overset boundary
condition. For simulations with large relative motion, such as rotorcraft, the overset approach is an
enabling technology (Noack, June 2007).
Overset methods can be employed to account for the large motion of articulating bodies such as
flap deployments or bay door motions; however, these problems can be further complicated by the
mechanical wearing of parts over their lifetime that introduces an element of free play. This free
play allows some limited amount of unrestrained motion around the commanded position so that
the exact position of the motion body cannot be precisely controlled. Free play has been
recognized as an important parameter to consider in the aeroelastic analyses of components. The
FAA Advisory Circular 25.629 section 6.3.4 (Federal Aviation Administration, October 2014)
states, “Consideration of free play may be incorporated as a variation in stiffness to assure
adequate limits are established for wear of components such as control surface actuators, hinge
bearings, and engine mounts in order to maintain aeroelastic stability margins.” It also states in
section 7.1.4.1 that “Freeplay effects should be incorporated to account for any influence of in-
service wear on flutter margins,” as well as that “Once the analysis has been conducted with the
nominal, experimentally adjusted values of hinge moment coefficients, the analysis should be
conducted with parametric variations of the coefficients and other parameters subject to
variability.” Therefore, it is ideal to create a system model that enables incremental changes to the
free play to examine the effects and understand required tolerances.
When the aerodynamic response may be considered linear, it is possible to develop theoretical
mathematical models that describe the system (Tang & Dowell, November 2011). However, when
incorporating free play into aeroelastic analyses of deployable structures subject to nonlinear
aerodynamics, such models are not readily constructed. It is the intent of this paper to demonstrate
how these effects can be accounted for at reasonable computational expense through the use of
commercially available software packages. This will be demonstrated using a notional example of

an elastic flap deploying behind an elastic slat where the flap contains various amounts of free
play.
2. Demonstration Case
The demonstration case considered is depicted in Figure 1. The flap is designed to deploy by
sliding aft while simultaneously rotating into the flow at a positive angle of attack. Computations
were performed to approximately 18° (illustrated in Figure 2), past where stall and separation
begin. The actuation was analyzed with ±1° of free play as well as with no free play.
Figure 1. The geometry of the case considered consists of a leading slat and a trailing flap.
Figure 2. Three commanded deployment angles for the slat/flap system are depicted: 0°, 10°,
and 18°.
The notional flap geometry is made of a modified Wortmann FX-60-157 airfoil shape, frequently
used on sailplanes and occasionally in rotorcraft. Internally, the structure was assumed to consist

of a thick skin with a wing spar toward the quarter chord. The unmodified airfoil structure
contains a sharp and thin trailing edge that was difficult to make sufficiently stiff for the purpose
of a realistic deployable flap simulation, so the trailing edge was blunted to avoid overly
complicating the interpretation of the results.
Flight conditions are approximately that at sea level. The structure is flying at a Mach number of
0.35 and an ambient temperature of 300 K at atmospheric pressure. The chord length of the flap is
1.0 meter.
3. Numerical Model and Procedure

There are three important aspects to accurately modeling an aeroelastic deployable structure. First,
a validated and trusted structural model is obviously required to begin the analysis. The structure
for the considered demonstration case was modeled with the 6.14-1 release of Abaqus/Standard
(SIMULIA, 2014). Abaqus/Standard was responsible for computing the structural response and
the deployment schedule of the structure. An intentionally coarse mesh was constructed for the
purely notional geometry considered in these analyses, although it should be emphasized that, like
most simulations, application of this methodology to existing physical problems will only be as
accurate as the least-accurate model. The second important part of the aeroelastic model is the
simulation of the fluid dynamics. In these analyses, the commercial CFD code Star-CCM+ (CD-
Adapco, 2014) was used to model the fluid domain. Once again, an intentionally coarse mesh was
used to discretize the domain; however, application to aeroelastic problems where accuracy is
crucial would benefit from a finer mesh. The CFD mesh used in these demonstrations ran
efficiently on a single twelve-core desktop computer; finer meshes could be run on a small
computer cluster. The last component of the numerical model is linking the solid and fluid
domains. This was accomplished with SIMULIA’s Co-Simulation Engine (CSE). The CSE allows
for communication of aerodynamic loads and displacements between the two domains. Critically,
the two domains do not need to be point matched at the interface, as the CSE is capable of
performing the required interpolation between domains while conserving energy.
Figure 3. The FEM mesh was largely composed of continuum elements, with shell elements
on the flap sides.
The FEM mesh, depicted in Figure 3, consisted of 4,960 eight-node continuum elements with
reduced integration with material properties corresponding to aluminum. It is necessary to have a
watertight mesh geometry to couple with the fluid dynamic solver, so an additional 85 shell
elements were applied to the flap to seal off the hollow inside. Two connector elements (depicted

in Figure 4) attached the leading slat to the trailing flap, with the ends of the connectors attached
to the surrounding structural mesh with *COUPLING *KINEMATIC elements. Commanded motions in
these connectors actuated the deployment of the flap.
Figure 4. A schematic of the two connectors, with free play angle, θ.

Free play in the actuators was obtained by applying a simple nonlinear model to the connector
elasticity, as shown in Figure 5. As illustrated, there was no load over a one-degree region of free
play, after which the loads increased sharply. For studying an existing free-play problem, one
would want to measure the free play response in the device by applying deflections in both
directions and then measuring the loads to get the problem-specific response.
Figure 5. Free play was modeled as a region where a change in deflection could be obtained
without maintaining a supporting load.
The fluid dynamics were resolved with Star-CCM+ (version 9.06.009). The use of relative body
motion was accomplished using overset grids. The leading slat was made a part of the stationary
background grid, while the flap grid was allowed to move freely as an overset mesh. Star-CCM+
largely handles the overset process, computing fringe points for interpolation between the two

meshes automatically. Large body motion did require the use of the in-built “alternative” hole-
cutting scheme, which is more robust for close bodies, as well as resetting the mesh motion to zero
to avoid negative Jacobians in the refined boundary layer.
Figure 6. The CFD domain was composed of two overset meshes. The large background
mesh contains the leading slat while the flap is contained in the small overset mesh.
Both domains were meshed with hexahedral template cells, similar to an octree approach (Figure
6). The near-body meshes contained fifty prismatic cell layers to aid in capturing the boundary
layer with a stretching ratio of 1.2. As with any viscous CFD calculation, it is best to verify that
the wall y+ value at the first grid point is less than 1 in order to adequately resolve the boundary
layer. As observed in Figure 7, this condition was satisfied for the current conditions. If conditions
were varied, for example upstream dynamic pressure, it would be necessary to check wall y+
again.

Figure 7. Wall y+ values indicate appropriate initial wall spacing (y+ < 1).
The Menter k-omega SST (Menter, August 1994) detached-eddy turbulence model was employed
in these calculations. Increased refinement of the CFD meshes will further help resolve separated
flow physics as the computations incorporate increasingly more LES modeling.
Simulations were performed with one-millisecond time steps in Star-CCM+. Temporal
discretization was limited first-order, and five subiterations were applied to converge the flow
variables. Coupling between Star-CCM+ and Abaqus/Standard was performed explicitly at one-
millisecond intervals.
Before coupling began, the CFD calculations were performed with rigid bodies without motion to
allow nonphysical transients to leave the domain. After the solution became steady,
Abaqus/Standard was coupled with only elastic deformation included in the initial solution (i.e.,
without large body motion). The objective of this step is to avoid nonphysical excitement of the
structure at the start of the deployment. To improve computational efficiency of this step, the
structure was solved with a quasi-static approach that did not include inertial terms. This was done
until a steady response was obtained, indicating that the system was ready for the deployment
maneuver.
It should be noted that using the coarse meshes applied in this study, the computational cost was
dominated by the CFD calculations, and less than 1% of the run time was attributed to the
structural computations. Thus, while rigid, moving body simulations can be performed, the added
cost of coupling to Abaqus/Standard may be insignificant while providing insight into the
aeroelastic characteristics of a structure.
The deployment of the flap was computed with a dynamic structural analysis (therefore including
the inertial terms). As previously discussed, resolving the inertial response was an important
objective of this work. These computations were performed with and without free-play effects.
Comparisons to quasi-static analysis results were also made.

While the stability margins were not calculated in this effort, they are readily possible with this
methodology. To compute these modal parameters, one would vary the dynamic pressure for a
given free-play amount. An instability would be immediately apparent as a divergence of the
solution, enabling flutter or divergence speeds to be determined. Actual computation of the
frequency and damping requires additional analysis. As an example, see Bauchau and Wang’s
partial Floquet analysis for large multibody systems (Bauchau & Wang, October 2007), which is
well-suited to these types of computations (Zaki, et al., May 2010).
4. Demonstration Results
The described dynamic analysis procedure was compared to the quasi-static procedure where no
large body motion was included. Perhaps surprisingly, the dynamic response was, in fact, less
computationally expensive to obtain than the quasi-static response. This is because the goal of a
quasi-static analysis is an approximately steady result, which took more iterations to achieve. The
computational cost per iteration between the cases was approximately the same, as both used the
same meshes and the same codes. The dynamic analyses, however, only required that the
deployment maneuver be completed and were on average approximately four times faster to
produce than each quasi-static analysis. Furthermore, several quasi-static analyses were required at
each deployment angle while the dynamic analysis swept through all deployment angles in one
simulation. Thus, despite the additional complications associated with the moving geometry, the
dynamic analyses were considerably easier to obtain than the traditional quasi-static analyses.
The results of these analyses are depicted in Figure 8 and indicate that the dynamic loads tend to
oscillate about the mean of the quasi-static analysis. These amplified loads are imparted to the
actuator and surrounding structure and should be considered during a fatigue life assessment.
Figure 8. The results comparing dynamic (blue) and quasi-static (red) analyses. Left: one
degree of free play is included. Right: no free play is included.
It is useful to compare the effects of free play in Figure 8. With free play included, the moment
peaks and then decays, followed by another peak that repeats for the first ten degrees. This is
attributed to the flap impacting the stop at the end of the free play region, exciting the structure,
which then rings out. This is confirmed in Figure 9, which reveals that the actuator connector
comes to near-rest at the free-play stop toward the end of the simulation. It is thus unsurprising

that this impact behavior is entirely missing from the case where free play is not included, and that
instead a dynamic oscillation about the mean quasi-static loading occurs.
Figure 9. The actuator connector angle as a function of commanded angle.

Examination of the frequency response of the free play (Figure 10, left image) indicates an
excitation frequency of approximately 30 Hz. Performing a modal analysis within
Abaqus/Standard of the initial position indicates that the first mode is a flap-only rigid body mode,
followed by a mode at approximately 88 Hz. The inability to reproduce the dominant mode in the
response is due to the linear nature of the eigenvalue/eigenvector modal approach. Removing the
free play and repeating the analysis (modal response shown in Figure 11) indeed reveals the first
mode is 29.9 Hz, as expected.

Figure 10. Frequency response of the dynamic analysis with free play (on left) and without
free play (on right).
Figure 11. The first mode of the demonstration case when no free play is included.
Since the quasi-static analyses did not include inertial terms, it is possible to apply a single-pole
low-pass filter at the first modal frequency of 30 Hz to the dynamic results and compare the results
of the procedures again. Doing this (illustrated for results both with and without free play in
Figure 12) removes the modal excitation and allows additional comparisons to be drawn. It is clear
that the case with free play (Figure 12 left) does in fact contain the quasi-static response despite
this not being apparent in the unfiltered signal (Figure 8 left). While the case without free play
clearly follows the same trend before filtering, the application of the low-pass filter does aid in
reducing the magnitude of the excitation and follows the same trend as the case without free play.
This suggests that for certain classes of problems, a time-accurate dynamic analysis could supplant
multiple quasi-steady simulations.

Figure 12. Results of the dynamic analysis after a low-pass filter of 30 Hz is applied are
compared to quasi-static results both with free play (left) and without free play (right).
One key implication of performing the time-accurate deployment simulation is the discovery of
the magnitude of the oscillatory load on the actuator, which may be critical for fatigue life
assessments.
5. Conclusions
The following conclusions may be drawn:
• A practical methodology for performing free-play analyses on aeroelastic deployable
structures has been demonstrated.
• The effect of free play on dynamic loads may significantly exceed the loads predicted by
a quasi-static analysis. This is attributed to the dynamic impact of the flap on the actuator
as it transitions through the region of free play.
• The time-accurate deployment simulation enables a prediction of the magnitude of
oscillating loads, which is required when determining the fatigue life of the relevant
components, whereas quasi-static simulations could only predict the mean stress.
• Dynamic analyses may be performed at reduced computational costs compared to the
quasi-static analyses. This is because the quasi-static analyses attempt to produce a steady
solution where the transients have left the modeled system, which may require simulation
times that exceed the time of deployment.
• A single dynamic analysis may be used with a low-pass filter below the first modal
frequency to obtain results similar to quasi-static simulations.
• The described methodology is well suited for aeroelastic stability margin calculations
through standard time-accurate stability procedures. Doing so requires varying the
problem’s relevant parameters such as the freestream dynamic pressure and the amount
of free play to determine stability.
6. Acknowledgements
The authors would like to thank Dr. Eric Blades of ATA Engineering, Inc., for suggesting this
type of analysis, as well as for discussions regarding the computation of aeroelastic stability
margins.
7. References
Bauchau, O. & Wang, J., October 2007. Efficient and Robust Approaches to Stability Analysis of
Large Multibody Systems. Journal of Computational and Nonlinear Dynamics, 3(1).
Blades, E. & Cornish, A., January, 2015. Aeroelastic Stability Predictions of a Business Jet
Landing Gear Door Using High Fidelity Fluid-Structure Interaction Tools. Kissimmee, FL.
CD-Adapco, 2014. Star-CCM+ 9.06.009 User's Manual.

Federal Aviation Administration, October 2014. Aeroelastic Stability Substantiation of Transport
Category Airplanes, No. 25.629-1B.
Menter, August 1994. Two-Equation Eddy-Viscosity Turbulence Models for Engineering
Applications. AIAA Journal, 32(8), pp. 1598-1605.
Mitcheltree, R., Salas, M. & Hassan, H., 1988. Grid embedding technique using Cartesian grids
for Euler solutions. AIAA Journal, 26(6), pp. 754-756.
Noack, R., June 2007. A Direct Cut Approach for Overset Hole Cutting. Miami, FL.
SIMULIA, 2014. Abaqus 6.14 User's Manual.
Tang, D. & Dowell, E., November 2011. Aeroelastic Response Induced by Free Play, Part I:
Theory. AIAA Journal, 49(11), pp. 2532-2542.
Zaki, A., Reveles, N., Smith, M. & Bauchau, O., May 2010. Using Tightly-Coupled CFD/CSD
Simulations for Rotorcraft Stability Analysis. Phoenix, AZ.
8. Appendix
Simulations were performed on a Hewlett-Packard Z420 desktop computer containing the Intel®
Xeon® CPU E5-1650 with a clock speed of 3.20 GHz and 32.0 GB of RAM. Eight cores were
used in each simulation. Dynamic simulations required 208 CPU-hours while quasi-static
simulations each required between 168 and 376 CPU-hours. Abaqus/Standard alone required
approximately just one CPU-hour.

Failure Analysis of CFRP Tubes with Integrated
Rubber Layers Subjected to Transverse Low-
Velocity Impact Loading
E. Stelldinger, A. Kühhorn, M. Kober
BTU Cottbus–Senftenberg, Chair of Structural Mechanics and Vehicle Vibration Technology,

Siemens-Halske-Ring 14, 03046 Cottbus, Germany
Abstract: Among the numerous advantages of CFRP laminated structures there are some
drawbacks, such as the high sensitivity in terms of impact loading. Particularly low-velocity
impacts result in non-visible and barely-visible impact damage respectively. The three-
dimensional state of stress in the impact zone leads to matrix cracking and delamination inside the
composite laminate and possibly to fibre failure for higher impact energies. It can be very difficult
to detect such damage by visual inspection, and additionally, in most cases a significant reduction
of strength and stiffness can be expected. To improve the damage tolerance a rubber layer, named
KRAIBON®, is integrated in the composite layup of cylindrical carbon/epoxy tubular specimens.
Numerous low-velocity impact tests, using different rubber compounds and layups, are carried
out. The degree of damage is examined using microsectioning. A reliable prediction of the
initiation and propagation of failure, caused by impact loading, requires suitable failure criteria
such as the theories of Puck, Cuntze or the LaRC05 criterion. Within this work a theory for initial
and progressive failure, based on the Puck failure criterion, is implemented in Abaqus/Explicit
using a user material subroutine (VUMAT). The initiation and propagation of interface
delamination is modelled using cohesive layer elements. A constitutive hyperelastic material law
provided by Abaqus is used to model the constitutive behaviour of the rubber by fitting the
hyperelastic constants to experimental test data. The numerical predictions of the local pattern of
damage and the impactor reaction force histories, computed by explicit finite element
calculations, have been compared with the experimental results.
Keywords: Composites, Delamination, Failure, Impact
1. Introduction
The continuously increasing use of CFRP structures in many industrial branches promotes the
development of new failure criteria and also the enhancement of existing theories. The aim is to be
able to describe the material behaviour under all possible load conditions. In view of the fast
growing computational power, it is possible to analyze more and more complex structures, but
also to implement more extensive failure criteria into finite element analysis software. The driving

force for the continuous development and improvement of the simulation tools is the possibility to
reduce time-consuming and costly component tests. However, in order to verify the simulation
tools, some component tests are indispensable. In particular, the complex damage behaviour of
CFRP structures makes the calculation of damage due to impact very difficult. An impact causes
concentrated out-of-plane loads, which leads to inter-fibre failure, fibre failure and delaminated
areas between the plies. Particularly insidious are damages inside the laminate, which cannot be
detected by visual inspections, the so-called barely visible impact damages. Such impact loads can
occur in a variety of situations, for example tool drops during maintenance and foreign object
impacts like hail or stone chip.
In cycle sport CFRP frames are now used in a large number. Unfortunately, the bicycle frames are
exposed to numerous impact loads, such as stone chips. In the worst case this can lead to a non-
visible damage within the laminate and a subsequent sudden total failure of the bicycle frame.
There are already some efforts to increase the impact resistance of the bicycle frames (Kaiser,
2007). Own experimental studies have shown, that the integration of rubber layers into a CFRP
laminate, results in a significant increased impact energy damage threshold. However, an efficient
optimization of the impact resistance-enhancing measures only is possible by reliable simulations
of the damage behaviour of CFRP laminates. In the present work, a combination of the failure
theory of Puck and the cohesive layer technology is used.
2. Rubber-layer integration
In order to improve the impact resistance of a CFRP laminate, a rubber layer (KRAIBON®), with a
thickness of 0.5 mm, was integrated. Two different rubber compounds were used and placed at
different positions within the laminate. However, for the first numerical simulations presented here
only the rubber mixture with the internal name SAA9509/21 was used.
2.1 Constitutive law
For the simulation of the material behaviour of the rubber, a hyperelastic model, using the Ogden
strain energy potential
N
2 µi N
1
U =∑ (λ1 i + λ2 i + λ3 i − 3) i + ∑
α α α
( J el − 1) 2i ,
i =1 αi i =1 Di
with an order of N = 3 was applied. Here, λi are the principal deviatoric stretches and Jel is the
elastic volume strain. The material coefficients µi and α i have been calibrated by Abaqus on the
basis of tensile and compressive test data provided by Gummiwerk KRAIBURG. Due to the lack

of volumetric compression test data, the Poisson´s ratio has been assumed to be ν = 0.4995.
Taking into account the relationship
3K 0 / µ 0 − 2
ν= ,
6 K 0 / µ0 + 2
the initial shear modulus
N
µ 0 = ∑ µi
i =1
and the initial bulk modulus
2
K0 =
D1
the value for D1 is computed as approximately 0.002, by using the parameters listed in Table 1. D2
and D3 are set to zero. The verification of the Drucker stability showed a stable behavior of the
model for all strains. More details can be found in the Abaqus Users Manual.
Table 1. Ogden material parameters for the rubber compound SAA9509/21.
i µi αi Di
1 -55.498 4.698 0.002
2 23.930 5.324 0
3 32.575 4.037 0
3. Impact tests
The damage response of laminated composites due to low-velocity impact has been studied
experimentally by many authors. Some surveys are published by (Agrawal, 2014; Cantwell, 1991).
For the implementation of low-velocity impact tests two common procedures exist: Testing by
drop-weight impact towers, which is more commonly used and the use of a pendulum impact
tester (charpy pendulum). In the present investigation, a charpy pendulum, equipped with a
spherical impactor (D = 15 mm) and a force transducer, was used. The impact energy can be
varied by the initial excursion of the pendulum. As test objects tubular specimens with a length of
200 mm and a diameter of 60 mm, made of CFRP and rubber, were used. In most cases, the focus
was on the investigation of the damage profiles due to impacts with an impact energy range from
Eimp = 10 J upwards. The test objects were mostly plates, made of fibre-reinforced plastics
(Agrawal, 2014). However, in practical application cases, real components usually have curved
surfaces. On the basis of extensive experimental studies (Ehrlich, 2004) showed, that the curvature

of the impacted surface, cannot be neglected with respect to the impact behaviour. Another finding
was, that structures with convex curved impact surfaces have a significantly greater damage than
planar structures. Experimental tests of bicycle frame tubes, made of CFRP, have shown that even
impact energies of Eimp = 0.5 J cause significant damage within the laminate (Kaiser, 2007). For
this reason, three samples were tested for each layup with a relatively small impact energy starting
at 0.15 J. Subsequently, the impact energy was increased in 0.15 J steps until a first drop in the
load response was evident (see Figure 1).
threshold load Fthr

0,6 J 0,75 J
0,45 J
impact energy Eimp
0,30 J
0,15 J
Figure 1. Exemplary force-time histories.
In addition, numerous tests were carried out at higher impact energies and with impact energies
just below the determined delamination threshold values. For the investigation of the extent of
damage, the specimens were examined by microsectioning. The cross sectional cut runs directly
through the point of the initial contact between impactor and specimen. Some of these
microsectioning examinations have shown that there are a few inter-fibre failures without any
indication in the force-time history. This leads to the conclusion that each drop in the load
response is caused by delaminations or fibre failures. For the implementation of the numerical
simulations only the reference sample layup and the layup with the best impact performance were
selected: Reference: [±45/02/±45/02/±45]; I1O: [±45/RUBBER/02/±45/02/±45].

4. Failure models used for numerical investigations
Thanks to the efforts within the framework of the three world wide failure exercises (Hinton,
2004; Kaddour, 2012; Kaddour, 2013), great progress could be achieved in the field of failure
analysis of fibre reinforced plastics. In the present work Puck´s model (Puck, 1996; Puck, 1998)
has been selected because it was able to achieve very good results in the first and second exercise.
This action plane-related failure criterion is physically based on the hypothesis of Coulomb and
Mohr and suitable for brittle fracture. In addition, a degradation model has been chosen, which
enables a reduction of the stiffness of damaged regions, depending on the result variables of
Puck´s approach and the stress conditions in this region. In order to make these models available
for Abaqus/Explicit, these theories have been implemented in a user defined material model
(VUMAT). As the propagation of delaminations cannot be described with this model, a cohesive
layer approach provided by Abaqus/Explicit has been used.
4.1. Fibre failure (FF)
Damage initiation
One of the basic characteristics of Puck´s failure criterion is the distinction between FF and IFF.
To predict the fiber breakage an extended fiber fracture criterion has now been established. This
formulation involves the fibre perpendicular stresses, which, due to the Poisson´s effect, cause
additional stresses in the fibres (Puck, 1998; Puck, 2002; Deuschle; 2012):
1  

E|| 

  R||t for[...] ≥ 0
f E FF = σ 1 − ν ⊥|| − ν ⊥|| f ⋅ mσf (σ 2 + σ 3 ) with  c
± R||t ,c   E|| f   − R|| for[...] < 0
If the stress exposure increases to fE FF = 1, fibre breakage occurs. The magnification factor mσf
takes into account the inhomogeneous stress field in the matrix. The transverse stresses are slightly
larger near the fibre, so the factor is set to the value mσf = 1.1 for CFRP (Deuschle; 2012).
Damage propagation
As soon as fibre breakage occurs, all stiffnesses are degraded to 1% of their initial value.

4.2. Inter - fibre failure (IFF)
Damage initiation
The IFF model of Puck is a very well-proven model and has been used by many authors. For this
reason, the explanations have been kept quite short. The basic idea is, that the fracture occurs on a
fiber-parallel plane, on which the stress exposure reaches the value of fE(θ) = 1 at first. For this
reason, all acting stresses must be transformed to potential fracture planes in a range
of θ = [-90°; 90°]:
σ n (θ ) = σ 2 ⋅ cos2 θ + σ 3 ⋅ sin 2 θ + 2τ 23 ⋅ sin θ ⋅ cos θ

τ nt (θ ) = −σ 2 ⋅ sin θ ⋅ cos θ + σ 3 ⋅ sin θ ⋅ cos θ + τ 23 ⋅ (cos2 θ − sin 2 θ )
τ n1 (θ ) = τ 31 ⋅ sin θ + τ 21 ⋅ cos θ
An angle of θ = 0° means that the fracture surface runs perpendicular to the 2-direction. The
following two expressions of the stress exposure only depend on these three action plane-related
stresses and the fracture resistances related to the action plane:
2 2
 1 pt    τ (θ )  2  τ (θ )  pt
f E IFF (θ ) =  At − ⊥Aψ  ⋅ σ n (θ ) +  nt A  +  n1 A  + ⊥Aψ ⋅ σ n (θ )
     
 R⊥ R⊥ψ    R⊥⊥   R⊥||  R⊥ψ
for σn(θ) ≥ 0 and
2 2 2
 p⊥cψ   τ nt (θ )   τ n1 (θ )  p⊥cψ
f E IFF (θ ) =  
⋅ σ n (θ ) +  A  +   + A ⋅ σ n (θ )
 RA   A 
 ⊥ψ   R⊥⊥   R⊥||  R⊥ψ
for σn(θ) < 0. Here, the following relationships are valid:
R⊥c
R⊥At = R⊥t ; R⊥A|| = R⊥|| ; R⊥⊥
A
= c
;
2(1 + p⊥⊥ )
p⊥t ,ψc t ,c
p⊥⊥ p⊥t ,||c τ nt2 τ n21
= cos 2
ψ + sin 2 ψ with cos 2 ψ = ; sin 2
ψ = .
R⊥Aψ A
R⊥⊥ A
R⊥|| τ nt2 + τ n21 τ nt2 + τ n21

In reality, not only the three action plane-related stresses (σn, τnt and τn1) are relevant for IFF. For
this reason, Puck supplemented his theory with two extensions. The first one takes into account
the influence of high fibre parallel stresses (σ1). Due to high stresses in 1-direction some fibres
already fail before reaching the strength limit. The results are local micro-fractures and fibre-
matrix debonding, which reduces the resistance against IFF. The second extension concerns the
influence of non-fracture plane stresses and probabilistic effects in an analytical manner. Both
extensions have been integrated into the VUMAT, but should not be discussed in detail here. The
definition of all required variables is made according to the recommendations of the authors.
Detailed explanations can be found for example in (Deuschle, 2012).
Damage propagation
Numerous studies have shown, that prior to the formation of first macroscopic cracks, a variety of
micro cracks leads to a nonlinear material behaviour (Kopp, 1999). However, the consideration of
these nonlinearities should be a part of future work. Within the meaning of pucks failure
condition, a stress exposure of fE IFF = 1 leads to first macroscopic cracks. There are already
different ways to consider the post-failure behaviour of composite materials. For IFF, continuum
damage models, based on the smeared crack approach, are still by far the most widely used.
Smeared crack means, that the effects of a discrete crack are evenly distributed over the whole
continuum. In context of FEM, the material stiffness must be degraded in the material point where
fE IFF has reached the value of one. A very simple procedure is the “ply-discount method”, which
suddenly degrades the stiffnesses of the ply to the residual stiffnesses. Other possibilities would
be, for example, to degrade on the basis of an empirical function (depending on fE IFF) or the
“constant IFF stress exposure approach” (Puck, 2002; Deuschle, 2010). The latter approach was
chosen, because this is very simple and yet reasonable. Here, the stress exposure is kept constant
at fE IFF = 1 by the ply stiffness values. The idea behind this is, that when the stress exposure of a
layer exceeds the value of 1, immediately another crack occurs and thus reduces the stiffness and
the stress exposure, respectively. Micro mechanical studies have shown that the degradation of the
individual material parameters (E2, E3, G12, G13, G23) should be carried out in dependence of the
fracture angle θfp. In (Deuschle, 2010) the development of an approach for fracture angle-
dependent degradation is presented. In (Deuschle, 2012) a simplified version of this approach is
used. However, using these approaches for the present problem leads to an unstable damage
evolution. The reason is that due to the loads in thickness direction no load redistribution in
adjacent layers is possible. Therefore, only the in-plane material parameters (E2, G12,) were
degraded, which leads to the following modified degradation rule, following (Deuschle, 2012):
 E ⋅ (1 − δ ⋅ cos(θ fp )) for σ n > 0

E2deg =  2
 E2 for σ n ≤ 0
G12deg = G12 ⋅ (1 − δ ⋅ 0,6 ⋅ cos(θ fp )) .

Where δ = [0;0.97] is the damage variable. The upper limit of δ was set to δmax=0.97 to ensure a
residual E-moduli of 3% (Knops, 2003). Following (Deuschle, 2012) it is assumed, that the G-
moduli experience about 0.6 times the reduction of the E-moduli. When IFF occurs, the damage
variable δ is gradually increased to maintain a constant fE IFF = 1, until all stiffnesses have reached
their residual stiffness values.
4.3. Delamination
Damage initiation
A delamination is caused by the three stresses, σ33, σ13 and τ23, acting on the interface between two
layers. Consequently, a delamination initiation criterion should be based on these three stresses. A
very good overview of existing delamination failure criteria can be found in (Abrate, 2011).
Puck´s action plane-related failure criterion includes the delamination as special case. Thus, a
delamination occurs if the stress exposure fE reaches its maximum value of fE =1 on an action-
plane of θ = ±90°.
However, with regard to the damage propagation due to delamination initiation, a suitable theory
must be used, which accounts for the stress singularity at the delamination front. Abaqus provides,
by the use of cohesive elements, some very powerful displacement and energy based degradation
approaches. For this reason, an Abaqus-integrated delamination initiation approach was used.
Besides the maximum nominal stress criterion another stress based criterion, the quadratic
nominal stress criterion
2 2 2
 t n   t s   tt 
 0  +  0  +  0  =1,
 t n   t s   tt 
is provided by Abaqus and used in the present work. For the application of this approach the
constitutive response of the cohesive layer has to be defined in terms of a traction-separation law.
Where tn is the nominal traction stress in 3-direction, ts and tt in 1- and 2-directions. tn0, ts0, and tt0
are the maximum bearable nominal stresses in the three directions, respectively. Following the
recommendations of Puck, the interlaminar strengths must be reduced with an weakening factor of
0.8 ÷ 0,9. The reasons are increased fibre waviness and imperfections in the form of air-
entrapping, which on this interfaces are more likely to occur due to fibre crossings (Puck, 1996).
The strength values reduced by a factor of 0.8 can be found in Table 3. Under the assumption that
the constitutive behaviour is uncoupled, the nominal traction stresses can be calculated as follows
t n   Enn 0 0  ε n 
    
t s  =  0 Ess 0  = ε s 
t   0
 t  0 Ett  ε t 

Where εn, εs and εt are the nominal strains, which are equal to the separations if the nominal layer
thickness is set to T=1. More information can be found in the Abaqus analysis user´s guide.
Damage propagation
The damage propagation is defined in the form of an evolution law based on fracture-energy. To
specify the dependence of the fracture energy on the mode mix, the power law form is used:
α α α
 Gn   Gs   Gt 
 C  +  C  +  C  =1.
 Gn   Gs   Gt 
Where GnC, GsC and GtC are the values for the fracture toughness, which refer to the critical
fracture energies required to cause failure in the respective direction. The corresponding values are
given in Table 3.
4.4 Material properties
For the manufacturing of the tubular specimens the unidirectional prepreg KUBD1507 was used.
The corresponding material data, which are used for the numerical simulations, are listed in
Table 2 and Table 3.
Table 2. Material data KUBD1507.
Density ρ = 1556 kg/m³

Elastic properties E1 = 121400 MPa*1; E2 = E3 = 7000 MPa; E1f = 234000 MPa
G12 = G13 = 4000 MPa*2; G23 =3200 MPa*2
ν12 = ν13 = 0.3*2; ν23 = 0.4*2; ν||┴f = 0.2*2
Strengths R||t = 1680 MPa*1; R||c = 1200 MPa*2
R┴t = 135 MPa; R┴c = 200 MPa*2
R┴|| = 85 MPa
Inclination parameter pt┴|| = 0.35*3; pc┴|| = 0.30*3; pt┴┴ = pc┴┴ = 0.275*3
*1 Test data; *2 Assumed; *3 recommendation of the author
Table 3. Material data cohesive layer.
Density ρ = 1556 kg/m³

Elastic properties Knn = 7000 MPa; Kss = Ktt = 4000 MPa*2
Strengths t0n =108 MPa; t0s = t0t = 65 MPa
Fracture toughness GnC = 0.8 N/mm*2; GsC = GtC = 0.3 N/mm*2; α = 1*2
*2 Assumed

5. Numerical investigations
The finite element model was created using Abaqus/CAE (see Figure 2). The CFRP tubular
specimens have a length of 200 mm and an outer diameter of 60 mm. The tube is supported about
half the circumference on the back. The space between the two support elements is 40 mm. The
impactor is modelled as a rigid body with a diameter of 15 mm.
specimen supports rubber

(rigid body)
z
y
x
impactor
impact-zone
(rigid body) cfrp
Figure 2. Finite element model
The impactor revolves around an axis in a distance of 382 mm. The inertia about this axis was set
to 0,384 kg/m², which is exactly the inertia of the real charpy pendulum. Consequently, the impact
energy can be varied by changes in the initial angular velocity. The contact between the different
contact surfaces is modelled using the general contact algorithm provided by Abaqus/Explicit. In
normal direction a “Hard” contact was defined, while the tangential behaviour is simulated by a
penalty formulation using a friction coefficient with a value of µ = 0.3. To reduce the effort due to
mesh modifications and associated allocations of the material orientations, the impact-zone (see
Figure 2) is modelled as a separate region and tied to the surrounding mesh-region. Some studies
for the estimation of errors which are induced by using this approach can be found in (Keskin,
2015). Inside the impact-zone the lateral dimension of the elements is about 0.2 mm up to 1 mm in
the edge region. Through the thickness one C3D8R element per laminate ply is used. Each layer
has a thickness of 0.14 mm, while the interface has a thickness of 0.01 mm and is discretized with
COH3D8 elements. The rubber layer is modelled using four fully integrated elements (C3D8)
through the thickness. To prevent hourglass modes of the C3D8R elements the stiffness relaxation
hourglass control has been used.

6. Results
6.1. Reference test series
The test results are listed in Table 4, showing a fairly good repeatability of the IR – test series. All
initial delaminations have been detected at the same impact energy, within a very narrow range of
threshold loads. The accordance of the force-time histories is quite good, but in the simulation, the
first delamination already occurs at a contact force of about 580 N (see Figure 5). Accordingly,
also the extent of the delaminations is significantly greater in the simulation (see Figure 3 (top)
and 4 (top)). Figure 3 (top) shows an example of a cross-section cut of a sample after a 0,3 J –
impact. The detected damages are IFF (Mode A) in ply 10 and a delamination at the interface 9-
10. In addition a cross-section cut due to a 0.45 J – impact is shown (see Figure 4 (below)).
Figure 3. Section cut views: Experiment
Figure 4. Section cut views: Simulation (IR, 0.30 J)

Table 4. Test results IR* - series.
Specimen name Impact energy Eimp [J] Threshold load Fthr [N]
IR-1 0.30 719
IR-2 0.30 774
IR-3 0.30 707
* [±45 /02/±45/02/±45]
Figure 5. Force vs. time histories of IR-1.
6.2. I1O test series
The impact test series I1O shows with respect to the delamination threshold load a certain
scattering (see Table 5). As a result, the respective first delaminations occurred at different impact
energies. At a high impact energy, the correspondingly high internal energy at the time of the
damage initiation leads to greater damage. The damage in consequence of the impact energy of
0.6 J is only a small delamination at the interface 5-6, while the damage as a result of the 1.35 J –
impact has significantly greater proportions (see Figure 6).
Table 5. Test results I1O* - series.
Specimen name Impact energy Eimp [J] Threshold load Fthr [N]
I1O-1 0.75 951
I1O-2 1.35 1090
I1O-3 0.60 836
* [±45/RUBBER/02/±45/02/±45]

Numerical investigations were carried out at all three impact energies (0.60 J; 0.75 J; 1.35 J). An
impact energy of 0.60 J leads to no damage in the numerical simulations. The 0.75 J - impact
caused IFF (mode A) located in layer 2 and layer 10, but no delaminations occurred. Finally the
1.35 J - impact simulation led to large delaminations and numerous IFF (mode A and mode C).
Figure 8 shows a comparison of the contact force histories due to the 1.35 J – impact. The good
agreement of the initial slopes of both curves shows, that the total stiffness can be reproduced very
well by the use of the constitutive model, as described in Section 2.1. However, the damage
process begins at a slightly higher contact force than predicted in the experiment. In addition, the
major delaminations were predicted at the interfaces 7-8 and 8-9, in contrast to the interfaces 8-9
and 9-10 in the experiment. These gaps can be explained by assuming, that the matrix crack tips
act as delamination initiation trigger. These stress singularities due to IFF are not considered by
the simulation. In the numerical model, the effects of the IFF are taken into account only by some
load redistribution to adjacent layers.
Figure 6. Section cut views (Experiment)
Figure 7. Section cut views: Simulation (I1O, 1.35 J)

Figure 8. Force vs. time histories of I1O-2
7. Conclusions
The action plane-related failure criterion of Puck, including a suitable degradation model, was
implemented using a user defined material subroutine (VUMAT). In addition cohesive layer
elements were used for the prediction of delamination initiation and propagation. The results of
the numerical simulations show a good accordance to the experimental investigations. The
experimentally determined improvement of impact tolerance by integrating a rubber layer
(KRAIBON®), could be reproduced by the numerical simulations. The good agreement of the
numerical and experimental force-time histories of the I1O - test series show, that the material
behaviour of the rubber was very well captured by the hyperelastic material law. This opens up the
possibility to optimize parameters such as rubber compound, rubber layer thickness and stacking
sequence in terms of an impact resistance improvement.
8. References
1. Abrate, S., „Impact Engineering of Composite Structures“, SpringerWienNewYork, 2011.
2. Agrawal, S., Singh, K. K., Sarkar, P. K., “Impact Damage on Fibre-Reinforced Polymer
Matrix Composite – A Review”, no. 48, pp. 317-332, 2014.
3. Cantwell W., Morton J., “The Impact Resistance of Composite Materials – A Review”,
Composites, no. 22, pp. 347–362, 1991.

4. Deuschle, H. M., „ 3D Failure Analysis of UD Fibre Reinforced Composites: Puck´s Theory
within FEA”, Thesis, Universität Stuttgart, 2010.
5. Deuschle, H. M., Kröplin, B. H., „ Finite Element Implementation of Puck´s Failure Theory
for Fibre-Reinforced Composites under Three-Dimensional Stress”, Journal of Composite
Materials, no. 46, pp. 2485-2513, 2012.
6. Ehrlich, I., “Impactverhalten schwach gekrümmter Strukturen aus faserverstärkten

Kunststoffen”, Dissertation, Universität der Bundeswehr München, 2004.
7. Hinton, M.J., Kaddour, A.S., and Soden, P.D. “Failure Criteria in Fibre Reinforced Polymer
Composites: The World-Wide Failure Exercise, Elsevier, 2004.
8. Kaddour, A. S., and Hinton, M. J., “Benchmarking of Triaxial Failure Criteria for Composite
Laminates: Comparison between Models of ‘Part (A)’ of ‘WWFE-II’”, Journal of Composite
Materials, no. 46, pp. 2595–2634, 2012.
9. Kaddour, A. S., Hinton, M. J, Smith, P. A., and Li, S., “The Background to the Third World-
Wide Failure Exercise”, Journal of Composite Materials, pp. 2417-2426, 2013.
10. Kaiser, M., “Zur Anwendung von kohlenstofffaserverstärktem Konststoff im Hochleistungs-

Rahmenbau von Sporträdern”, Dissertation, Technische Universität Kaiserslautern, 2007.
11. Keskin, A., et al. “On The Quantification of Errors of a Pre-Processing Effort Reducing
Contact Meshing Approach” , AIAA SciTech 2015, Kissimmee, Florida, 5 - 9 January 2015.
12. Knops, M., “Sukzessives Bruchgeschehen in Faserverbundlaminaten”, Dissertation, RWTH-

Aachen, 2003.
13. Kopp, J. W., “Zur Spannungs- und Festigkeitsanalyse von unidirektionalen

Faserverbundkuststoffen”, Dissertation, RWTH-Aachen, 1999.
14. Puck, A.,”Festigkeitsanalyse von Faser-Matrix-Laminaten, Modelle für die Praxis“, Carl-
Hanser-Verlag, München, 1996.
15. Puck, A., and Schürmann, H., „Failure Analysis of FRP Laminates by Means of Physically
Based Phenomenological Models“, Composites Science and Technology, pp. 1045-1067,
1998.
16. Abaqus Users Manual, Version 6.13, Dassault Systémes Simulia Corp., Providence, RI.

New Design of a Pressure Vessel Subjected
to Blast Loads
L.Cousin, P.Evrard
CEA, DAM, DIF, F91297 Arpajon, France
Abstract: A new design of containment vessel has been proposed to conduct confined detonation
experiments. In order to guarantee the confinement of the detonation products, the risk expressed
as a probability of failure has to be quantified. This quantification is done using probabilistic
analyses which require data from experiments and simulations in order to be sustained. When a
blast stresses a spherical vessel, many different mechanical phenomena appear and have to be
studied using adapted models. A blast leads to reflected waves in the structure. A numerical
chaining is used to access the vessel dynamic structural response. This chaining consists in
linking CATIA V5 with Abaqus/CAE to realize the analysis models and in linking our
hydrodynamic code to Abaqus /Explicit in a weak coupling: high speed hydrodynamic
simulations provide transient pressures which are used to act on the inner shell of the vessel. Two
mechanical phenomena and their associated finite-elements models are of interest. First, we
focus on the nominal model used to study the dynamic response of the vessel itself, underlining
the most important components (internal furniture, material laws, type of mesh...) to be taken into
account. The spherical vessel has five ports. Each of them has a cover bolted to the vessel. Then,
we review a model the aim of which is to understand the bolts behavior when the vessel is
subjected to high dynamic stresses. This model consists in a first implicit analysis followed by an
explicit analysis which continually takes into account the bolts pre-tension loads. Experiments
were performed in order to validate this weak coupling. We end up showing the good agreement
between experimental and numerical results.
Keywords: Bolt Loading, Connectors, Constitutive Model, Coupled Analysis, Dynamics,
Experimental Verification, Explosive, Impact, Safety, Vibration.
1. Introduction
The CEA has developed a new design of containment vessel for carrying out confined detonation
experiments. This paper briefly presents the confinement vessel used for experiments and its
associated instruments which allows understanding its behavior when submitted to blast loads.
This behavior of the vessel has to be simulated because of the need existing in safety
demonstration: to guarantee the confinement of the detonation products, the risk expressed as a
probability of failure has to be quantified. This quantification is done using probabilistic analyses
which require data from experiments and simulations in order to be sustained. The second point of
the paper focuses on the numerical chaining used to master the blast effects on the vessel: this
chaining goes from the Design Office, working with CATIA V5, to the Simulations Office, in
which Abaqus models are prepared. Abaqus/Explicit is also linked in a weak coupling with our
hydrodynamic code which provides information to load the inner shell of the vessel. The third
point then describes the nominal Abaqus/Explicit model used to study the response of the vessel
when submitted to a transient impulse loading. Results from the chaining are compared to
experimental data. The fourth and last part consists in a description of another Abaqus model used

to understand the behavior of the bolts which link five covers to the vessel body. The model
consists in a first Abaqus/Standard analysis followed by an Abaqus/Explicit analysis using the
*IMPORT functionality. A different bolts modeling is proposed because of the limitations of the
*Pre-tension Section functionality only available in Abaqus/Standard, allowing for the bolts
pre-tension loads to be taken into account in the dynamic simulation.
1. Containment vessel configuration

The containment vessel is presented in Figure 1. It consists of a steel spherical body with five
ports each equipped with a cover:
- a steel cover at the top, used for general access,
- four aluminum side covers, used for diagnostics.
Each cover is connected to the vessel body by two concentric circles of bolts and three
independent seals. The vessel is connected to a steel frame by its four side ports.
Figure 1 : Vessel for confined detonation experiments
The vessel is monitored during each experiment in the same way:

- several strain gauges are attached to the outer skin vessel, at strategic locations and at the
center of the five covers,
- for each cover, a gauge is attached to a bolt belonging to the inner circle of bolts and
another gauge is attached to a bolt belonging to the outer circle of bolts.
2. Numerical chaining
We use a numerical chaining (Figure 2) which links:
- the Design Office to the Simulations Office which uses Abaqus (Figure 2, path (1))
- and also the Simulations Office which uses our hydrodynamic code to our Simulations
Office which uses Abaqus (Figure 2, path (2)).

Design Office Simulations Office
CATIA V5 Home-made
hydrodynamic code
(1) Simulations Office

(2)
Abaqus/CAE
Abaqus/Standard
Abaqus/Explicit
Figure 2 : Numerical chaining

2.1. From Design Office to Simulations Office
The first link enables to go from the geometric product assembly done with CATIA V5 in the
Design Office to the finite elements model done in the Simulations Office using Abaqus/CAE
(Figure 2, path (1)). A first VB script tool is used to put in the user’s folder all native CATIA
V5 files (CATParts and CATProducts) concerning the model as well as materials data
associated with the parts. The materials data consist in a ‘.lib’ file, an input file and a file
giving the links between materials and their associated parts in the CATIA model. Materials
are defined in a complete manner in the CATIA library: mechanical characteristics defining
thermal behavior as well as dynamic behavior are known for most materials. Once the user’s
folder contains all these data, Abaqus/CAE is opened and the user accesses a plug-in allowed
for the CATIA V5 and the materials data to be loaded in. Once the file to import is chosen:
- CATIA V5 geometries are imported in Abaqus/CAE,
- Imported parts are renamed following definitions names existing in the CATIA model,
- Instances are created automatically in Abaqus/CAE according to the imported parts,
- Materials data existing in the folder are imported and materials are assigned to the
suitable parts,
- A table ‘material/behavior’ is available for the user: in this table one can choose the most
appropriated behavior for the analysis he wants to lead (characteristics of a thermo-
mechanic behavior, characteristics of a dynamic behavior type Johnson-Cook plasticity
law...)
This gateway is useful to guarantee the quality of our Abaqus models: users are sure to work with
the appropriated geometries and materials properties.
2.2 Weak coupling between a home-made hydrodynamic code and

Abaqus/Explicit
The second link (Figure 2, path (2)) consists in a weak computational coupling between our
hydrodynamic code and Abaqus/Explicit. In a mechanical way, the vessel is submitted to a blast

load. This blast load is computed using a specific hydrodynamic code developed by the CEA. The
code simulates explosive detonation, internal gas expansion and shock waves propagation in 2-D
Eulerian analysis models. A characteristic load function of an explosive detonation inside the
vessel is presented in Figure 3 (a): a transient impulse load can be observed, due to the shock
wave, followed by a long term quasi-static load caused by the evolution and the heating of gaseous
detonation products. Because of furniture inside the vessel, the pressure field versus time is not
homogeneous on the inner shell. The pressure field must be computed at different points onto the
inner shell in order to take into account the heterogeneous dynamic load and must be computed for
a sufficient duration to cover the several reverberations inside the vessel. The way these pressure
data are used in Abaqus is explained in paragraph 3.1.
Figure 3 : (a) Typical pressure-time evolution; (b) Example of an 2-D Eulerian

hydrodynamic model
3. Abaqus/Explicit analysis model in order to study the vessel

response
3.1. Constitutive Abaqus/Explicit analysis model

Working assumptions
The modeling of the vessel presented in Figure 1 is accomplished using a half-structure shown in
Figure 4. Furniture exists rounding the explosive charge which is not represented here but taken
into account in the modeling.

Top cover
Vessel body
Lateral
Shields
covers
Furniture
Axial
connectors
SymetrySymmetry
plan plane
Figure 4 : Half-analysis model for the vessel’s dynamic response
Boundary conditions
This half-vessel is fixed in the Y-direction by two AXIAL connectors type which have an
elasticity modulus of one. Symmetry plane conditions are introduced for the nodes belonging to
the cross-section plane (Figure 4).
Mesh
A 3D continuum elements mesh is applied. The analysis model contains 1,5 million 7 millimeter
linear hexahedral elements C3D8R (Figure 5).

Figure 5 : Mesh of the half-model
Loads
Furniture is driven using speeds computed from the 2D-Eulerian hydrodynamic analysis. The
speed data are used through a predefined field defined in the Abaqus/CAE Load Module. The
resulting keyword is:
*Initial Conditions, type=VELOCITY
In the same time, all the pressure-time fields are used to load the inner shell of the vessel. We
define one amplitude curve per pressure-time field. These fields are applied by slices which are
defined creating:
- a surface: the inner shell,
- analytical fields to define the altitude of each slice in the Y-direction. This is done in the
Abaqus/CAE Load Module through a table presented in Figure 6 (a), accessible from the
Tools/Analytical Field menu,
- a load, type ‘pressure’ which allows to link the appropriated analytical field, surface and
amplitude curve. The good definition of the slices can be monitored in the Visualization
Module (Figure 6 (b)), loading the model of interest.
This functionality allows us to load the vessel without introducing any partitions of the structure
which are a problem to keep a regular mesh.

Figure 6 : (a) Definition of an analytical field describing a slice named
‘AF_carto_pression_S10’; (b) Verification of the definition of the slice in the
Visualization Module
Materials
The high strength steel vessel behavior, the steel top cover behavior and the aluminum side
covers behavior are supposed to be elastic perfectly plastic. The shields behavior is supposed to
follow a Johnson-Cook law.
The whole steel furniture behaviors are considered to be elastic perfectly plastic, coupled with a
shear failure criterion to take into account their failure when impacting the inner shell of the
vessel.
Interactions
A general frictionless contact is used in the model between the furniture and the vessel inner shell.
In order to take into account impacts between the furniture, failed or not, and the vessel, we create
different surfaces:
- an exterior surface on the inner shell of the vessel, named ‘exterior_vessel_inner_shell’
- an exterior surface on the whole furniture, named ‘exterior_furniture’
- interior surfaces in the whole furniture in order to take into account contacts existing
between the future failed furniture and the inner shell vessel . These surfaces are named
‘interior_furniture’

Exterior surfaces are created on geometric entities, in a classical way in Abaqus/CAE, using the
default selected tool. In order to create interior surfaces, the user has to define surfaces type
‘Mesh’ and change the default selected tool with ‘Select From interior Entities’ (Figure 7).
Select From Interior Entities
Figure 7 : Modification of the default selected tool to define interior surfaces

In the *Contact keyword, the following relations between surfaces must appear:
*Contact, op=NEW
*Contact Inclusions
Interior_furniture ,
Exterior_furniture,
Interior_furniture , Exterior_furniture
Interior_furniture, Exterior_vessel_inner_shell
Exterior_furniture ,Exterior_vessel_inner_shell
Outputs
Some of the elements belong to nodes sets according to the gauges attached to the outer shell of
the vessel during experiments. The strain tensor is the output of interest. We assign a specific
spherical material orientation (R, T, Z) on the vessel body so that the strain tensor is computed in
the same orientation as the gauges orientation. This is done to be able to compare experimental
and numerical data.
3.2. Analysis model validation
Several experiments were carried out using different explosive charges. The vessel is always
monitored with gauges which locations are the same as the ones presented with the red dots in
Figure 8. Experimental and numerical maximum Von Mises stresses values are compared for each
gauge. One of the comparisons made is presented in Figure 8 . Good agreement can be observed
excepted at the south pole and the north pole (top cover) of the vessel: the finite elements model
underestimates the experimental data at the poles. These phenomena can be observed for each
experiment comparison. Some more work is led at the moment in order to improve this fact.

Maximum Von Mises stresses (MPa)
Simulation
Experiment
J12
South North
Side Covers
Figure 8 : Maximum Von Mises stresses- Experimental measures and numerical

results comparison
4. An analysis model for the bolts behavior study

4.1. Constitutive Abaqus analysis model
Working assumptions
The Abaqus model used in order to understand the behavior of the bolts when the vessel is
submitted to a blast load is presented in Figure 9. It consists in the same half-model as before, with
the same symmetry plane and boundary conditions. No furniture is taken into account. Bolts used
to assembly the covers to the vessel body are modeled. The top cover is assembled by two
concentric circles of M36 bolts. The side covers are assembled by two concentric circles of M24
bolts.
Cylindrical Two concentric

local circles of
orientation M36 bolts
Two rows
of M24 bolts
Figure 9 : Analysis model configuration

Interactions
The two shields are linked to the vessel using the *TIE functionality. Contacts defined between the
covers and the vessel body take into account a friction of 0,1.
Mesh
All the parts are meshed with C3D8R linear hexahedral elements. The analysis model contains 1,5
million 7 millimeter linear hexahedral elements C3D8R.
Bolts modeling
General concept
The M36 bolts (top cover) and the M24 bolts (side covers) are modeled using beam elements (in
red on Figure 10) associated with connectors in order to firstly impose the pre-tension load in the
bolt and secondly monitor the existing axial force in the bolt when the vessel is submitted to a
dynamic solicitation due to the blast.
Zoom Diameter under

Kinematic
the head bolt
Cover Coupling
2
Connector
Vessel Thread grip height 1
Figure 10 : Bolts modeling- Beam element associated with connector

The connector defines a kinetic relation between two points. One of these nodes is the end of the
head bolt side beam. The second node is named differently from the first one but has the same
location: the connector element length is null. Figure 10 shows a zoom of the modeling: the
connector, built between node 1 and node 2, is connected in series with the beam (in red). The
second node of the connector is the kinematic coupling node reference. This coupling is
representative of the head bolt leaning on the covers: the slave nodes underlying surface is defined
according to the diameter under the head bolt. A kinematic coupling is also defined at the other
end of the beam in order to take the bolt thread grip height into account: the slave nodes
underlying surface is defined according to the real bolt thread grip height.
Beam elements characteristics

Linear beam elements, B31, are used to model the bolts. The behavior of these steel bolts is
supposed to be elastic. The diameter, d, and the length, L, of the beam are computed in order to
obtain the suppleness in tension and flexion of the M24 bolts and the M36 bolts respectively.
These assembly parameters are computed using the COBRA software which follows the VDI2230
[ 1].

Connector elements characteristics
‘Cartesian + Cartan’ connectors type are used. They are associated with local cylindrical
orientation (R, T, Z) such as defined in Figure 9. All translation and rotation degrees of freedom
are activated (U1, U2, U3 + UR1, UR2, UR3). The designation ‘1’ is equivalent to the local R-
direction, the designation ‘2’ is equivalent to the local T-direction, and the designation ‘3’ is
equivalent to the local Z-direction. The connectors are used to impose the pre-tension load in the
bolt and secondly to know the axial force history when the vessel is submitted to a dynamic
solicitation due to the blast so that their stiffness is infinite.
Sequence of analysis
The simulation is conducted in two times:
• A first implicit simulation is accomplished using Abaqus/Standard in which the pre-
tension load is imposed. This time is a two steps analysis:
o the first step is a step for contact initiation. The degrees of freedom of the
connectors linked with beams on the top cover are all locked. The degrees of
freedom of connectors linked with the beams of the side covers are all locked
excepted the translation degree of freedom in the beams axis direction. A very
small displacement is given to the side covers in this beam axis direction,
o in the second step, the imposed displacements are off and the connectors linked
with the beams of the top cover are relaxed in the beams axis direction. The pre-
tension load is then passed through the whole bolts defining a load type
‘Connector Force’ in the third local direction (Z).
• The final state of this first Abaqus/Standard analysis becomes the initial state of a
dynamic explicit simulation accomplished using Abaqus/Explicit using the *Import
functionality. The vessel is submitted to the blast load following the same methodology
as described in the paragraph 3.1, the bolts pre-tension load being taken into account.
This pre-tension load state is maintained locking all the degrees of freedom of the
connectors. Finally, we access the axial force history in the connectors which is
equivalent to the axial force seen by the bolts.
4.2. Analysis model validation

Uniaxial gauges are attached to two M36 bolts and to two M24 bolts. In both cases, one of these
bolts is located on the inner circle of bolts and the other on the outer circle of bolts. These gauges
allow us to know the initial force level obtained as a result of the tightening torque and to follow
the time-force evolution during the experiment.
The measures accuracy is +/-13%. The finite elements model is evaluated on its capacity to
estimate the greatest force passing through the bolts. Figure 11 shows the experimental and
numerical maximum levels of force for a given experiment in:
• The M36 bolts of the top cover located in the inner circle (TC INT),
• The M36 bolts of the top cover located in the outer circle (TC EXT),
• The M24 bolts of the side cover located in the inner circle (SC INT),
• The M24 bolts of the side cover located in the outer circle (SC EXT).
Very good agreement is found. This notification is valid for others experiments.

Figure 11 : Maximum forces passing through the bolts. Comparisons between
experimental and numerical data.
5. Summary
This paper describes the tools of a numerical chaining used to design a confinement vessel
submitted to blast load. This chaining links first CATIA V5 to Abaqus/CAE and secondly the
CEA hydrodynamic code to Abaqus/Explicit in a weak coupling. The Abaqus /Explicit model
dealing with the dynamic response of the vessel is presented and a comparison between
experimental and numerical results is shown. The paper describes also an Abaqus/Standard-
Abaqus/Explicit model the aim of which is to be able to estimate the higher axis force existing in
the bolts of the covers when the vessel is solicited by a blast load. Comparisons with experimental
results validate the two proposed models.
6. References
[ 1] VDI 2230 “Systematische Berechnung hochbeanspruchter Schraubenverbindungen

Zylindrische Einschraubenverbindungen” Verein deutscher Ingenieure

Predicting in- and out-of-plane Damage Evolution in
Fiber-reinforced Composites
W. Wilson
Fokker Landing Gear B.V., The Netherlands
Abstract: Fokker Landing Gear has a history in development of composite technology for landing
gear applications. In order to successfully design and qualify composite landing gear parts it is
essential to be able to reliably predict the mechanical and failure behavior of the composite
material. Therefore the goal of this study was to develop, calibrate and validate a material model
which can be used to reliably predict the failure behavior of fiber-reinforced composites. In the
developed material model the fibers and resin are modelled as separate materials with their own
specific material and failure behavior. The interaction between the fibers and resin is accounted
for using a (proprietary) modified Mori-Tanaka approach. For the resin both the plasticity and
damage behavior is included. For fiber failure new damage laws have been developed. Cohesive
surfaces are used to model the delamination behavior. The implementation of this material model
and cohesive damage laws is done in Abaqus/Standard. Comparison with test data showed that
with the material model the failure behavior in many different tests could be predicted very well.
Keywords: Composites, Constitutive Model, Damage, Delamination, Fabrics, Failure, Landing

Gear, User-Defined Material,
1. Introduction
Fokker Landing Gear has a history in the development of composite technology for aircraft
landing gear applications. Due to the highly safety critical nature of landing gear components, in
order to successfully design and qualify these components it is essential to be able to reliably
predict the mechanical and failure behavior of the composite material.
Traditional models used for modelling composites are in most cases linear elastic and use failure
criteria (in most cases only valid for uni-directional materials) to predict the moment of failure.
Hence, they do not take into account the nonlinear behavior of the resin and the actual damage
initiation and damage evolution. With these traditional models the mechanical behavior and failure
moment of thin undisturbed composite sections can be predicted relatively well. However, for
thick composites and especially around stress concentration (e.g. open holes, pin loaded holes) the
traditional models under predict the failure loads by a large margin (e.g. Chang et al., 1984;
Whitworth et al., 2003). Over the last years more detailed material models have been developed
that include the effect of resin damage and plasticity (e.g. Johnson, 2001; Chen et al., 2014).
Although big steps forward have been made, with the models currently available it was not
possible to reliably predict the failure behavior of the composite materials using textile (woven)
fiber reinforcements. In order to address this issue Fokker Landing Gear has developed a new
material model for modeling composites (Wilson, 2011, 2013) with the capability of modelling
textiles composites. In this material model the fibers and resin are modelled as separate materials
with their own specific thermo-mechanical and failure behavior. The interaction between the
fibers and resin is accounted for using a (proprietary) modified Mori-Tanaka approach. The
implementation of this material model and cohesive damage laws is done in Abaqus/Standard.

Previously it has been shown that composites with different textile architecture (2x2 twill fabric
layers and 2D braided layers with varying angles) can be described with the same set of material
data (Wilson, 2011). It was also shown that this material model is capable of predicting failure
around stress concentrations (Wilson, 2013).
The goal of the current study was to further improve this model with new failure laws and to
incorporate the plasticity behavior of the resin, such that it can more reliably predict the failure
behavior of fiber-reinforced composites.
2. Method
2.1 Material model
2.1.1 Total stress
The total stress in the material is given by
 totf i  totf
σtot = 1 − ∑ ρf σr + ∑ρ σ i
f
i
f , Equation 1
 i =1  i =1
where σ r and σ f are the total resin and fiber stress tensors, respectively, totf the number of fiber

directions and ρfi the fiber volume fraction in the i-th direction efi , respectively. When the strains
in a composite are relatively low, it can be assumed that the relative volume fractions of the resin
and fiber stay constant during deformation. Hence, ρfi is taken as a constant.
2.1.2 Microstructure model

Following the rules of mixtures the total strain in the material is given by
 totf i  totf
ε = 1 − ∑ ρ f ε r + ∑ρ ε i i
f f , Equation 2
 i =1  i =1
where ε r and ε fi are the resin and fiber strain tensors, respectively. When we express ( ε fi ) as a
function of the resin strain ( ε r )
εfi = 4 A fi : ε r , Equation 3
this Equation 2 becomes

 totf 
( )
totf
ε = 1 − ∑ ρfi ε r + ∑ ρfi 4 A fi : ε r , Equation 4
 
 i =1  i =1
from which follows that
−1
 totf  
( )
totf
ε r = 1 − ∑ ρfi  4 I + ∑ ρfi 4 A fi  : ε . Equation 5
 i =1  i =1 

Here 4 A fi is the strain concentration tensor that relates the (local) fiber strain to the resin strain,
which is computed using a proprietary modified Mori-Tanaka approach (based on Mori & Tanaka,
1973).
2.1.3 Fiber properties

The fibers are assumed to behave linear elastic and anisotropic, with a different stiffness in
compression and tension. Fiber damage is assumed to be a brittle fracture. Although it is assumed
that a fiber fails immediately once damaged, the total damage in a fiber bundle is assumed to
evolve more gradually, mainly due to the differences in fiber orientations and resin distribution
around the fibers (which influences the local fiber stresses). The following function for the
evolution of the damage parameter Df is used
 κ  κ − κ i ,f 
Df =  c,f  f  . Equation 6
 
 κ f  κ c,f − κ i ,f 
κi,f and κc,f are the values of history parameter κf at which damage initiation starts and at which the
fibers have completely failed. For tensile failure the history parameter κf is set equal to the
maximum fiber strain over time. The function for κ used for compressive failure is proprietary to
Fokker Landing gear, but is based on the assumption that fiber micro-buckling is dependent on the
compressive fiber strain, effective resin stiffness and the out-of-plane strains (compressive out-of-
plane strains are assumed to inhibit micro-buckling and tensile out-of-plane strains are assumed to
promote micro-buckling).
2.1.4 Resin properties

The resin is assumed to have an elasto-plastic behavior, with linear elastic behavior up to the point
where the material starts to plastically deform. For the plastic behavior an exponential Nadai
Ludwik hardening law is chosen.
ε r ,p (t ) 
n

σ y (t ) = σ y ,0 1 +  Equation 7
 ε 0 
where ε r ,p is the plastic equivalent strain, σy,0 the initial yield stress and ε0 and n material
parameters. Assuming von Mises plasticity the following flow rule is used
3 ε r ,p dip
ε r ,p = σr Equation 8
2 σ rip
with σ r the in-plane equivalent von Mises stress, and σ r the in-plane deviatoric part of the
ip dip
resin stress tensor. For the resin damage the following function is used (Johnson, 2001)
 σ ip 
Dr = α log r  Equation 9
 Fr 

where Fr is the in-plane von Mises stress at which damage initiates and α a positive constant that
determines the rate of damage evolution.
2.1.5 Mesh dependency

When a material gets damaged, part of the material can no longer carry any loads. Due to this the
effective stiffness of the material decreases. Analyses with such softening behavior are normally
very sensitive to mesh-dependency. In Abaqus the Hillerborg's fracture energy proposal
(Hillerborg et al., 1976) is available which for most cases can (partly) alleviate the mesh-
dependency. A more general, but computationally less efficient, method for alleviating mesh-
dependency is a nonlocal damage theory (e.g. Pijaudier-Cabot and Bazant, 1987; Bazant and
Pijaudier-Cobot, 1988; Comi, 2001). In the non-local damage theory the local strain ε that
determines the damage initiation is replaced by a nonlocal strain ε . This nonlocal strain is taken
as a weighted average of the strain within a volume V.

n
  ξ i2  
 1    ∫
V (ξ i ) exp − 2 ε i (ξ i )

 2l 
ε (x) =
VV ∫
g (ξ )ε ( x + ξ )dV = i =1 n


 ξ i2 
, Equation 10
∫V (ξ i ) exp − 2 
i =1  2l 

where l is the length scale of the material, V is the volume of punt i with a distance ξ i from the

current location x.
For modelling failure in composites, next to alleviating mesh-dependency there is another reason
to use a non-local method. As mentioned in the previous paragraph compressive fiber failure is
generally assumed to occur via micro-buckling of the fibers. For micro-buckling to occur, high
strains or stresses should be present over a certain length of the fiber. Hence, very local stress or
strain peaks will probably not lead to micro-buckling. As the nonlocal method uses an average
strain over a certain area, it will filter out these very local strain peaks. For the tensile failure of
fiber the Hillerborg's fracture energy proposal is used.
2.2 Delaminations
2.2.1 Initiation laws
Mode 1 delaminations are assumed to occur due to excessive tensile out-of-plane stresses alone.
σ 33
≥ 1 , Equation 11
F33
The initiation and growth of mode 2 delaminations in composites is mainly governed by the
transverse shear stresses (τ13, τ23). Compressive out-of-plane stresses are known to inhibit the
initiation and growth of mode 2 delaminations while tensile out-of-plane stresses are known to
promote the initiation and growth of mode 2 delaminations (Lecuyer and Engrand, 1992). A law
for damage initiation based on the above assumptions is given by

max (τ 1,3 ,τ 2,3 )
+ βσ 33 ≥ 1 , Equation 12
Ftvsh
where Ftvsh is the transverse shear strength and β a positive constant. Mode 3 failure is assumed to
occur by the same rules as Mode 2.
2.2.2 Implementation
To implement Equation 11 and 12 in Abaqus/Standard the maximum stress criterion is used.
τ 13 τ 23 σ 33
, , ≥ 1 , Equation 13
F13 F23 F33
To implement Equation 12 into Equation 13 F13 and F23 have to be made dependent on the out-of-
plane stress (σ33). To do this Equation 13 has to be rewritten to
max (τ 13 , τ 23 ), σ 33 ≥ 1, Equation 14
Ftvsh − β *σ 33 F33
where β* is given by β/Ftvsh. The implementation of this damage initiation law in Abaqus/Standard
is described in detail in reference (Wilson, 2013).
2.3 Implementation in Abaqus and general model description

The material model as discussed in the section 2.1 (including the nonlocal damage theory) has
been implemented in Abaqus/Standard using the user-subroutines UMAT and USDFLD. For more
details about the implementation see (Wilson, 2013). The element deletion option is used to delete
elements in which the fibers have completely failed (this option is available in combination with
UMAT since Abaqus v6.14-1). All finite element models used in this paper consisted of C3D8R
elements with enhanced hourglass control. When possible due to symmetry, only half of the total
test sample was modeled. Each ply was modeled as a separate element layer. The layers were tied
together using cohesive surfaces. Initiation and growth of delaminations was included in the
material description of the cohesive surfaces (see paragraph 2.2). Surface-to-surface small-sliding
contact was used for the cohesive surfaces. Contact between samples and the test fixtures was
modelled using surface-to-surface finite sliding contact. Boundary conditions were chosen such
that they mimic the test conditions most accurately.
2.4 Determination of unknown material parameters

For the current study a composite material consisting of a combination of a quasi-UD 2x2 twill
fabric and braids of carbon fibers embedded in an epoxy resin was used. As the actual material
data is proprietary to Fokker Landing Gear only relative differences between test and numerical
data will be given. Calibration of the unknown material properties is done in the following
sequence:
1. Linear stiffness properties (Young’s moduli and Poisson’s ratios of the fiber and resin
and 2 shape parameters) (number of parameters = 7)

2. Non-linear resin properties (Plastic and damage behavior of resin) (number of parameters
= 7)
3. Delamination properties (Mode 1 and 2 failure) (number of parameters = 3)
4. Tensile failure properties of the fibers (number of parameters = 2)
5. Compressive failure properties of the fibers (number of parameters = 3)
For step 1 to 4 only samples consisting of the 2x2 twill fabric were used. For Step 5 a combination
of the 2x2 twill fabric and braids was used. In the next paragraphs the used calibration methods are
briefly described.
2.4.1 Linear resin and fiber properties

Isight 5.9 was used to calibrate the linear material properties (Young’s moduli and Poisson’s ratios
of the fiber and resin). As optimization method the “pointer method” is used. The simflow that
was used is depicted in Figure 1. To make the simflow as efficient as possible several coupon
tests were modeled within one Abaqus simulation. Furthermore, three Abaqus/Standard
simulations were run in parallel in the Isight simflow (Figure 1A). In total 7 coupon test were
simulated:
• Compression in 0º and 90 º direction (ASTM D695)
• Tension in 0º and 90 º direction (ASTM D3039)
• In-plane shear test (ASTM D3518)
• 2 different 3 point bending tests (ASTM D2344)
Displacements and reaction forces were extracted from the Abaqus simulations using History
output. Using the Calculator component (“Compute stiffnesses” component in Figure 1A) the
sample stiffnesses were computed from the history data, after which the Excel component was
used to read in the stiffnesses as computed from test data. The Calculator component was then
used to compute the relative errors between the test data. As objective function the sum of all
relative errors is used.
2.4.2 Non-linear resin properties

Isight 5.9 was used to calibrate the nonlinear material resin properties (Plastic and damage
behavior of the resin). As optimization method the “pointer method” is used. To be able to make a
distinction between the plasticity and damage behavior an in plane shear test with cyclic loading
was used (Paepegem et al., 2006). The Isight simflow (Figure 1B) consisted of an Abaqus
component in which the in-plane shear test is simulated and a Matlab component in which the
stiffness, plasticity and damage as function of the shear strain are determined and compared to the
test data. As an objective function the sum of the relative errors of these curves is used.

A B
Figure 1. Isight simflow for the optimization of A) linear stiffness properties, B)

non-linear resin properties, and C) compressive fiber failure properties.
2.4.3 Delamination properties

To determine the delamination properties first the mode 1 properties were determined using a peel
test (ASTM D5528). The two unknown material properties are the damage initiation stress and the
energy release rate parameter (G1c). The latter can be directly determined from test data. The
damage initiation stress was obtained by manual optimization.
For the optimization of the Mode 2 parameters 3 different 3-point bending tests (ILS1 [0f/90f]3s,
ILS2 [90f/0f]3s, ILS3 [0f6] 1) were used (ASTM D2344). The unknown parameters for the Mode 2
failure were determined by a manual optimization. To minimize the number of unknown material
parameters, the following relation between the τ13 and τ33 was used (Turon et al., 2010 ).
GIIc
τ 13 = τ 33 , Equation 15
GIc
2.4.4 Tensile failure of fibers
For the determination of the tensile fiber failure properties 2 tensile tests were used (ASTM
D3039), one in warp (0º) and one in weft (90º) direction. Tensile failure strains (κi,f from Equation
7) were directly taken from the test data. The damage evolution parameters (κc,f from Equation 7)
were determined by manual optimization.
2.4.5 Compressive failure of fibers

For the determination of the compressive 0º-fiber failure properties 3 compression tests (COMP1
[45f/-45f/0f4]s, COMP2 [45f/-45f/0f3/45f/-45f/0f3/-45f/45f] and COMP3 [0±b453,0f4]s)1 were used. The
1
Here f means that the ply consist of the 2x2 twill fabric, and ±b45 mean the ply consists of a ±45º
braid.

90º -fiber failure properties were manually derived from the 0º-fiber failure properties. The
calibration (optimization) was done using Isight 5.9 (Figure 1C). As optimization method the
“pointer method” is used. As an objective function the sum of the relative errors in failure load
was used.
2.5 Validation
For validation of the model the following tests were used.
Tensile test
The tensile test was performed according to ASTM D3039 with a tapered specimen with a layup
of [0±45b/04f]s (Figure 2). The FEA model consisted 3324 C3D8R elements with enhanced hourglass
control. Due to symmetry, in the FEA models only a half of the total sample was modeled.
Figure 2. Top-Left) Photograph of tensile test setup; Top-Right) Dimensions;

Bottom) FEA mesh.
Compression test
The compression test was performed according to ASTM D3410-03 with a modified sample
(layup of [0±45b/04f]s) and a NASA short block compression fixture (Figure 3). The FEA model
consisted 4534 C3D8R elements with enhanced hourglass control. Due to symmetry, in the FEA
models only a half of the total sample was modeled.
Figure 3. Top-Left) Photograph of compression test setup; Middle) Dimensions;

Right) FEA mesh.

Bearing compression & tension
The bearing tests (compression & tension) were performed according to ASTM D5961-09 (Figure
4). The tested samples had a layup of [0±45b/04f/+45f/-45f]s. The loading pin had a diameter of 19.03
mm. The FEA models consisted of 6508 and 8131 C3D8R elements with enhanced hourglass
control, for the compression and tension test respectively. The pins were modelled as an analytical
rigid. The parts inside the lower clamps was not included in the models, instead the bottom planes
were fixed in all directions. Due to symmetry, in the FEA models only half of the total sample
was modeled.
Figure 4. Left) Photograph of bearing compression test setup; Middle) Dimensions;

Right) FEA mesh (left: compression; right: tension).
Open hole compression

The open hole compression tests were performed according to ASTM D6484 (Figure 5). The
tested samples had a layup of [0±45b/04f/0±45b/0±45b/0±45b/04f/0±45b]. The FEA model consisted of 7712
C3D8R elements with enhanced hourglass control. Due to symmetry, in the FEA models only half
of the total sample was modeled.

Figure 5. Left) Photograph of open hole compression test setup; Middle)
Dimensions; Right) FEA mesh.
3. Results
3.1 Calibration of material properties
3.1.1 Linear resin and fiber properties
The resulting errors between the FEA predictions and test data are given in Table 1.
Table 1: Resulting error [%] between FEA model and mean test data
ECE0 ECE90 ETE0 ETE90 υTE0 Eips υips EILS1 EILS2
0 0 2 0 1 3 4 3 1
3.1.2 Non-linear resin properties

In Figure 6 the shear stress, resin damage and plastic shear strain are plotted as a function of the
axial or shear strain for both the average test data and FEA data. The FEA results corresponded
very well with the test data.
0.6 0.03
Plastic shear strain [-]
Axial stress [MPa]
150 FEA
0.4 0.02 test
100
D [-]
0.2 0.01
50 FEA FEA
test test
0 0 0
0 0.02 0.04 0 0.02 0.04 0 0.02 0.04
Axial strain [-] Shear strain [-] Shear strain [-]
Figure 6. Resulting nonlinear resin response as a function of the shear strain. Left)
Shear stress, Right) Resin damage, Bottom) Plastic shear strain.

3.1.3 Delamination properties
In Figure 7 the resulting force-displacement graph of the FEA model and test data is given for the
G1c peel test and one 3-point bending test (ILS3). For the other 2 3-point bending tests the relative
difference between the FEA and test data was approximately 5%.
FEA
Test
Force
Force
0 Vertical displacement Displacement
Figure 7. Left) Force-displacement graph of peel test, Right) Force displacement

graphs of 3 point bending test ILS3 (blue=no damage; red=failed).
3.1.4 Tensile failure of fibers
The resulting error between the FEA predictions and test data for both the 0º and 90º failure was
<1%. The failure modes of both simulations corresponded very well to the test data.
3.1.5 Compressive failure of fibers

The computed failure loads corresponded very well with the test results (see Table 2).
Table 2: Resulting Errors in compression strength
COMP1 COMP2 COMP3
ERROR [%] +3 -3 +5
3.2 Validation
For the validation tests the relative errors between the predicted failure loads and the test data are
given in table 3.

Table 3: Errors between FEA and testdata
Error [%]
Full model No fiber No fiber No resin No
compression tension plasticity & delaminations
failure failure damage
Tension +1
Compression +6
Open hole -2 >100 -2 +6 -3
compression
Bearing tension -7 +2 -6 +36 +1
Bearing +1 +94 +1 +69 +5
compression
In figure 8 to 10 photographs of the failed open hole compression and bearing tests samples are
given together with the deformed meshes of the failed FEA simulations. The predicted failure
modes corresponded very well with failure modes seen in the tests. For all tests a combination of
resin damage/plasticity, delaminations and fiber damage was seen. To study the effect of these
different failure mechanisms on the final failure of the samples additional simulations were
performed in which each failure mechanism was selectively turned off. The results are given in
Table 3. The open hole compression test seemed to be mainly dependent on the compressive
failure. Although the resin plasticity and damage also slightly influenced the results (considering
that the compressive failure is dependent on resin plasticity and damage). For the bearing tension
and compression test the influence of the resin damage and plasticity was much larger. In the
bearing tension test both compressive and tensile fibers failure occurred almost simultaneously,
which can be the reason that failure can be relatively well be predicted by including just only one
of these failure mechanism. All tests were slightly dependent on the inclusion of delamination
initiation and growth.
Figure 8. Left) photograph of failed open hole compression test specimen, Right)
Deformed FEA model after failure

Figure 9. Left) photograph of failed bearing compression test specimen, Right)
Figure 10. Left) photograph of failed bearing tension test specimen, Right)
4. Discussion
In order to successfully design and qualify safety critical composite parts, it is essential to be able
to reliable predict the mechanical and failure behavior of composite materials in “disturbed” areas.
Therefore the goal of this study was to develop, calibrate and validate a material model which can
be used to reliably predict the failure behavior of fiber-reinforced composites.
For the current study, the model of Wilson (2013) was extended to include the plasticity of the
resin, updated failure laws and an improved microstructure model. In this material model the
fibers and resin are modelled as separate materials with their own specific material and failure
behavior. The interaction between the fibers and resin is accounted for using a (proprietary)
modified Mori-Tanaka approach. For the resin both the plasticity and damage behavior is
included. For fiber failure new damage laws have been developed. Cohesive surfaces are used to
model the delamination behavior. The implementation of this material model and cohesive
damage laws is done in Abaqus/Standard.
The nonlinear stiffness behaviour of the model was previously validated (Wilson et al. 2011) by
predicting the mechanical behaviour of braids (with different configuration) using material
properties that were fitted on the 2x2 twill fabric data used in this paper. In the current study it is
shown that also the in- and out of plane failure behaviour can be predicted very well for a large
variety of loading conditions and many different layups.
During testing a failure often occurs very rapidly. After the failure it is very difficult to see how
that failure was initiated and what the sequence of (possible) different failure mechanisms was. In
the current study we have shown that a model as developed in the current study can be used to

explain the sequence (and importance) of the different failure mechanisms in the total failure
process.
In the current study the material model was calibrated using coupon test data, and validated using
element test data (including stress concentration). For a full validation the model should be
validated for both additional element test and full thickness components or subcomponent.
5. References
1. Abaqus Users Manual, Version 6.14-1, Dassault Systems Simulia Corp. Providence, RI
2. Bazant, Z.P., Pijaudier-Cobot, T.G.P., Nonlocal continuum damage, localization instability
and convergence, J. Appl. Mech., Vol. 55, pp.287–293, 1988
3. Chang, F. et al., Failure of Composite Laminates Containing Pin Loaded Holes, Journal of
Composite Materials. vol. 18, 255-278, 1984
4. Jing-Fen Chen, Evgeny V. Morozov, Krishnakumar Shankar, Simulating progressive failure
of composite laminates including in-ply and delamination damage effects, Composites Part
A,, Volume 61,185-200 (2014)
5. Comi, C., A non-local model with tension and compression damage mechanisms, Eur. J.
Mech. A/Solids, Vol. 20, pp.1–22. 2001
6. Hillerborg, A., M. Modeer, and P. E. Petersson, Analysis of Crack Formation and Crack
Growth in Concrete by Means of Fracture Mechanics and Finite Elements, Cement and
Concrete Research, vol. 6, pp. 773–782, 1976.
7. A.F. Johnson, Modelling fabric reinforced composites under impact loads, Composites Part
A, Volume 32, Issue 9, Pages 1197–1206 (2001)
8. Lecuyer, F., Engrand, D., A methodology for the identification of a criterion for delamination
initiation, JNC8:751-762, 1992
9. W. Van Paepegem, I. De Baere, J. Degrieck., I. Modelling the nonlinear shear stress-strain
response of glass fibre-reinforced composites. part I: Experimental results. Composites
Science and Technology, 66(10):1455-1464, 2006.
10. Pijaudier-Cabot, T.G.P., Bazant, Z.P. Nonlocal damage theory, J. Engng. Mech. (ASCE),
Vol. 113, pp.1512–1533, 1987
11. Mori, T., Tanaka, K., 1973. Average stress in matrix and average elastic energy of materials
with misfitting inclusions. Acta Metall. 21, 571–574.
12. A Turon, PP Camanho, J Costa, J Renart, Accurate simulation of delamination growth under
mixed-mode loading using cohesive elements: definition of interlaminar strengths and elastic
stiffness, Composite Structures 92 (8), 1857-1864, 2010
13. Whitworth, H. A., et al., Failure analysis of composite pin loaded joints., Composite
Structures, vol. 59, 261–266, 2003
14. Wilson.,W., Virtual testing of composites using Abaqus., 2011 SIMULIA Customer
Conference, Conference Proceedings, May 2011
15. Wilson W, Prediction of damage evolution in composites, 2013 SIMULIA Customer
Conference, Conference Proceedings, May 2013

Nested Micromechanical and Structural
Models for the Analysis of Discontinuous
Long-fiber Thermoplastic Composite Materials
and Structures
M. Hakan Kilic, Ph.D.
Greene, Tweed & Co., U.S.A.
Abstract: This study presents a three-dimensional (3D) micromechanics-based nonlinear

framework for the analysis of discontinuous long-fiber (DLF) thermoplastic composite materials
and structures. The proposed material modeling framework is a nested micromechanical
approach that explicitly recognizes the composite system within the cross-section of a DLF
composite member. The overall modeling approach is able to predict both the elastic and
nonlinear response of the composite material based on the in-situ properties and response of the
fiber and matrix constituents. Coupon tests are used to calibrate the in-situ linear and nonlinear
properties of the fiber and matrix. The nonlinear calibration of the matrix is done by using the
overall axial shear stress-strain response generated from V-notch shear tests. The multi-scale
constitutive framework is used with Abaqus for the analysis of DLF composite structures. The
material subroutine (UMAT) is used to implement the DLF nonlinear micromechanical model. A
damage modeling approach is also integrated with the micromechanical model to form a
framework for the progressive damage analysis of DLF composite structures. The new “element
deletion control by state variables” functionality in Abaqus/Standard version 6.14 is also
employed as another approach for the progressive damage analysis. Several examples are
presented to demonstrate the predictive capabilities of the coupled multi-scale material and
structural analysis approach. The overall structural modeling framework is numerically efficient
and accurate when compared with experimental response of DLF structural components.
Keywords: Composites, Constitutive Model, Micromechanics, Experimental Verification,
Progressive Damage, Thermoplastic, Failure, Discontinuous Long-fiber, Multi-scale
1. Introduction
One of the challenges remaining for composites in commercial aerospace applications is the
ability to produce highly-complex three-dimensional (3D) components with an acceptable value
proposition versus metals. Discontinuous long-fiber (DLF) thermoplastic composites are therefore
introduced to provide complex-shaped composite parts with both part count and weight reduction
advantages over simple loaded metal parts. DLF composites also bridge the gap between complex
machined metal components and cost-intensive continuous fiber composite lay-up technologies.
Most DLF thermoplastic products are produced by chopping carbon fiber reinforced prepreg
unidirectional tape into “flakes” or “chips” with predefined dimensions. This process enables
cost-effective complex-shape metal replacement parts as seen in Figure 1. High thermoplastic
resin viscosity results in very uniform fiber/matrix distribution throughout the component,
including complex part features. Molded-in inserts and “net molding” capabilities result in cost-
effective manufacturing (reduced trimming or assembly labor content, reduced material waste).
Flexibility in DLF composite part design and manufacturing also allows consolidation of multiple

individual components into a single net-molded component which results in cost and weight
savings. Thermoplastic matrix polymers for DLF materials in Aerospace applications are
typically PEEK or PEKK [1].
Carbon fiber reinforced Molded-in holes

prepreg unidirectional Molded-in inserts Varied thickness
tape
Molded reinforcements
Chopped
“flakes” or
“chips” (bulk
form)
Molded-in hole,
helicoil insert
Molded-in holes & counter sinks Molded-in labeling
Figure 1. Greene Tweed’s compression molded Xycomp® DLF parts
Certification of DLF parts currently is achieved by point design which requires testing large
number of individual parts; hence, it is time consuming and expensive. As the use of compression
molded DLF composites in the aircraft industry increases, there is a need to develop certification-
by-analysis methods verified by limited experimental testing. However, there are currently no
standard or established design and analysis methods for DLF composites due to the random fiber
orientations, effects of material flow on fiber orientation, limited material test data, and
nonlinearity in material behavior. Accurate modeling of DLF material behavior is therefore
needed to establish proper stress analysis methodologies to predict DLF part performance.
To that end, this study introduces effective integration of 3D nonlinear micromechanical and
structural models for the progressive damage and nonlinear analysis of DLF composite materials
and structures. The DLF composite system is made of AS4/PEEK material. The proposed
framework consists of linearized micromodel homogenization schemes and stress correction
algorithms, which result in enhancing computational efficiency and accuracy. The effective
material response is generated using a 3D micromechanical model consisting of a rectangular unit-
cell with four fiber and matrix sub-cells. Approximate traction continuity and strain compatibility
relations are expressed in terms of the average stresses and strains of the sub-cells. The fiber and
matrix responses are explicitly recognized in this micromodel. The nonlinear elastic behavior is
attributed to the matrix. The micromechanical modeling framework is integrated into a general
3D finite element (FE) code, Abaqus. The result is a new global-local nonlinear modeling
approach for the analysis of DLF composite structures. A simple progressive damage approach is
also utilized within the framework along with 3D Tsai-Wu failure criterion. The new “element
deletion control by state variables” functionality in Abaqus/Standard version 6.14 is also

employed as another approach for the progressive damage analysis. Compression molded DLF
coupons and parts are tested and analyzed using the proposed multi-scale modeling framework.
The framework shows very good predictive capabilities for the overall effective properties,
nonlinear response, and ultimate load of DLF composites.
2. Nested micromechanical and structural framework for DLF

composites
A combined micromechanical and structural framework is developed for the nonlinear and
progressive damage analysis of DLF composites. A nonlinear 3D micromechanical model for the
DLF medium is used to generate the effective in-plane and out-of-plane responses of the material.
The proposed nested structural and material micromodeling approach is schematically illustrated
in Figure 2. The upper level depicts an FE structural model using 3D continuum elements. At the
lower level, a representative volume element (RVE) in the form of a unit-cell (UC) is used at each
Gaussian material point in order to generate an effective homogenized nonlinear response for DLF
medium.
2
Structural Level
3 1
Z
1
2
3
Element Level
material point
3D-FE
3
3D Micromechanics-based Material Model for DLF
Composites 2 Homogenized material:
DLF
X3
3D equivalent response with
s in-plane isotropy
h 1-h
matrix
1-b (1) f iber (2) matrix e

s
s 1
(3) matrix f iber
b
(4) f iber ξb e
X2
Figure 2. Nested micromechanical and structural framework.
The structural framework is generated by applying the developed micromechanical model within a
3D FE formulation in order to form a global-local nonlinear analysis of DLF structures. The
material subroutine (UMAT) of the ABAQUS FE code is used to implement the nonlinear

micromechanical model. This subroutine operates at each integration point and its task is to
update the stress, stiffness, and history variables, for a given strain increment. For finite strain
increments and nonlinear material response, the tangential stresses and strains typically violate the
nonlinear constitutive relations of the sub-cell materials. In order to satisfy the actual stress-strain
relationships as well as the traction and compatibility constraints, a correction scheme is needed.
Therefore, a stress update algorithm is developed which consists of two steps: (1) a predictor step,
which may be a tangential or a trial elastic predictor; followed by (2) a correction step, in which
the trial elastic state is adjusted in order to arrive at accurate stress values. Using a predictor step
alone, without subsequent correction, may lead to unacceptable accumulation of error. This makes
the task of efficient numerical implementation of the proposed micromodel very important in
order to perform nonlinear analysis of realistic DLF composite structures.
3. Micromechanical model for the DLF composite system

The DLF material is considered as a medium where resin is reinforced with flakes randomly
distributed in the plane of the layer. Therefore, the DLF effective medium can be represented with
an in-plane isotropic model. The DLF UC model generates 3D equivalent response with in-plane
isotropy for the DLF medium by a weighted average of two UC models (Figure 3). The first is a
matrix-mode layer (transverse), which is modeled using a UC where the fibers are completely
surrounded by a matrix phase. The second is a model for a fiber-mode layer, in which the fibers
are not shielded by the matrix, and both constituents have the same average strains.
X3
h 1-h
A (1) f iber (2) matrix Matrix mode

1-b
(Part-A)
1
(3) matrix
b s
B
(4) f iber ξb
X2 Fiber mode
(Part-B)
Figure 3. DLF unit cell.
The DLF UC model is shown in Figure 3 as a collection of four sub-cells. It is also constructive to
note that the matrix-mode layer (part A) is composed of sub-cells (1) and (2), while the fiber-mode
layer (part B) is composed of sub-cells (3) and (4). The out-of-plane direction is represented by
axis 3. The spatial variation of the displacement field in each sub-cell is assumed such that the
stresses and deformations are spatially uniform. Traction continuity at an interface between sub-
cells can therefore be satisfied only in an average sense. The 3D stress-strain constitutive
characterization for the fiber and matrix constituents is performed at the lowest level of the nested
modeling framework. In the current study, the fiber is modeled as a linear material and the matrix
constituent is modeled as a nonlinear material.

The DLF micromodel is driven by the in-situ response of its fiber and matrix constituents (sub-
cells). In this study, those are the properties of the AS4 carbon fiber and the PEEK resin. V-notch
in-plane shear test results are used to calibrate the micromodel in the linear stress-strain range.
The nonlinear material response of the composite is attributed to the matrix and it is modeled
using the J2- deformation theory along with the Ramberg-Osgood (R-O) nonlinear uniaxial
representation. The in-plane shear test results are also used to calibrate the nonlinear matrix
behavior. The R-O stress-strain for the matrix is calibrated by varying its parameters until the
overall nonlinear response for the composite is matched with the V-notch test results. The
calibration of the model is concluded with finding the three R-O parameters.
4. Prediction of the DLF elastic properties and stress-strain

behavior
The prediction capability of the micromodel for DLF composites is examined for coupons that are
subject to in-plane tension and compression and out-of-plane shear loading. These tests were
performed in displacement control mode in order to reach the ultimate load and a complete
breakage of the coupon. Strain gages were used to monitor the strains at the center of the coupons.
The ratios of computed effective moduli and Poisson’s ratios to the experimentally obtained ones
for the DLF material are reported in Table 1. Experimental values are the average of the test
results performed at room temperature. The predicted values from the present micromodel
compare favorably with those obtained experimentally.
Table 1. Ratio of predicted to experimentally obtained effective elastic properties.
Property E11 E22 G13 G23 ν12
Predicted/Test 1.00 1.00 0.99 0.99 0.90
Figure 4 shows the experimental stress-strain curves (with repeated tests) and the prediction of the
model (solid line) for out-plane shear behavior. The stress-strain curves are plotted as the stress,
monitored from the load cell of the machine, against the strain gage readings. The predicted result
from the micromodel is in excellent agreement with the test results. These results demonstrate the
ability of the proposed nonlinear micromodeling formulation to capture the nonlinear behavior of
DLF material.

Out-of-Plane Shear
Micromodel prediction
Test results
Figure 4. Predicted and experimental nonlinear out-of-plane shear stress-strain

behavior of DLF material.
5. Progressive damage analysis

Progressive damage modeling is very important because it allows for the simulation of degraded
structural responses, especially when the structure can continue resisting additional applied loads
despite failure at limited locations. The first step in the damage analysis is to establish a suitable
failure criterion for damage initiation in the material. Then, the effective elastic properties for
composite material have to be degraded gradually or are set to a small number for the composite
material. In the case of micromechanics-based modeling, progressive damage methodology can
be applied at the constituent level.
To that end, 3D Tsai-Wu failure criterion is used to detect damage initiation in the DLF material.
The criterion represents a general quadratic failure surface in the stress space. Both the in-plane
and out-of-plane modes of failure are considered in this study. Ultimate stress values required to
establish the 3D Tsai-Wu failure criterion are obtained from the coupon tests.
A damage index, F, is introduced as the maximum value of the Tsai-Wu failure criterion that an
integration point has experienced during all loading states. Damage index contours are used to
show the progression of damage in the structure. When F is larger than 1 at an integration point, it
means that the Tsai-Wu failure criterion has been exceeded and the progressive damage
formulation is active. Once damage is detected, the contribution of fiber sub-cells to the effective
properties of DLF composite material is set very close to zero. The reason to apply this
methodology is because damage initiates and propagates between the boundaries of DLF flakes in
the material where the fibers are not continuous. The elastic and nonlinear properties of matrix
sub-cells are degraded gradually at the damaged areas.

Another progressive damage analysis approach is also considered in this study. Instead of
degrading the constituent properties at the micromechanics-level, stiffness of a damaged element
is not taken into account in the global stiffness matrix. The new “element deletion control by state
variables” functionality in Abaqus/Standard version 6.14 is employed to perform this type of
progressive damage analysis approach. An additional state variable is assigned in the user
material subroutine to deactivate the material point when the damage index F is larger than 1 for
the 3D Tsai-Wu failure criterion.
6. Nonlinear and progressive damage analysis of DLF composite

plate under four-point bending load
The micromechanical and structural framework is used to predict the performance of a DLF plate
under four-point bending load as seen in Figure 5. Both of the proposed progressive damage
approaches, micromechanics-level and element deletion, are also used in the FE analysis. An FE
model with quarter symmetric boundary conditions is developed to represent the actual conditions
of the testing. Interactions are used at the loading and support pin locations. Kinematic coupling
is used to apply the displacement to the loading pin.
The predicted part performances using the 3D nonlinear micromodel along with the
micromechanics-level and element deletion progressive damage approaches are compared with the
experimental results as shown in Figure 6. Load and displacement values in Figure 6 are obtained
at the center of the loading fixture. As seen in Figure 6, good agreement is shown when
comparing the experimental results and the FE model that incorporates the combined
micromechanical and progressive damage approaches. In the case of using the 3D nonlinear
micromodel along with the micromechanics-level damage approach, the predicted ultimate load is
the last point before the failure point (larger red circle) on the predicted load-displacement curve
in Figure 6. The predicted ultimate load, from the FE model, is very close to the average of the
experimental results.
Quarter-Symmetric FE model
Loading Pin
Support Pin
Planes of Symmetry
Kinematic
Interaction coupling
at the Interaction at
support the loading ∆
location location
Figure 5. Quarter-symmetric FE model for four-point bending DLF plate.

1.0
0.9
Predicted
Ultimate Load
0.8
Average Ultimate Test Load

0.7
0.6
Normalized Load
0.5
∆
0.4
0.3
0.2 Prediction with using 3D nonlinear DLF

micromodel and progressive damage
Prediction with using 3D nonlinear DLF
0.1 micromodel and element deletion
Test Results
0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Normalized Cross-head Displacement
Figure 6. Predicted and experimental response of four-point bending plate.
First failure load is predicted to occur at the normalized load of 0.507 at which the damage index
value for the 3D Tsai-Wu failure criterion exceeds 1for the first time. First failure occurs in the
outermost layers in tension because the uniaxial tensile strength of the DLF material is lower than
the uniaxial compression strength.
7. Nonlinear and progressive damage analysis of DLF composite

T-beam under four-point bending load
The micromechanical and structural framework is also used to predict the performance of a DLF
T-beam part under four-point bending load, as shown in Figure 7. An FE model with quarter-
symmetric boundary conditions is developed to represent the actual conditions of the part testing.
The proposed progressive damage approach at micromechanics-level is also employed in the FE
analysis. Material directions are selected such that the out-of-plane direction is always normal to
the surfaces including the fillet area. Flow of the material can also be seen in the CT-scan image
of a T-beam cross-section. Loading and support pins are modeled using hexagonal and tetrahedral
elements, respectively. Hexagonal elements are used to model the DLF T-beam. Kinematic
coupling is used to apply the displacement to the loading pin.

Test Fixture
Specimen
2
Loading Pin
CT-Scan
Image
Support Pin
Plane of flakes are in 1-2

plane. 3-direction (out-
of-plane) is normal to
the exterior surfaces.
Test Fixture
2
Planes of
Symmetry 1
Support Pin
Loading Pin
Quarter-Symmetric
FE model
Figure 7. Quarter-symmetric FE model for DLF T-beam.
Two different geometries are considered for the nonlinear and progressive damage analysis of
DLF T-beams. First geometry has a larger fillet radius and thinner web thickness than the second
one as shown in Figures 8 and 9.
The predicted and experimentally obtained part performances are compared with each other in
Figures 8 and 9. Good agreement is shown when comparing the experimental load-displacement
results and the predicted ones from the FE model that incorporates the combined micromechanical
and progressive damage models. The predicted ultimate loads are the last points on the predicted
load-displacement curves. The predicted ultimate load values from the FE model are very close to
the average values of the experimental results.

400
1.000
Predicted
350
0.875 Average Ultimate Test Load Ultimate Load
A
300
0.750
T
250
0.625
Normalized Load
3R
L
200
0.500
150
0.375 tw
100
0.250
50
0.125 Prediction with using 3D nonlinear DLF
micromodel with progressive damage
Test Results
0.0000
0.00
0.000 0.05
0.125 0.10
0.250 0.15
0.375 0.20
0.500 0.25
0.625 0.30
0.750 0.35
0.875 0.40
1.000
Figure 8. Predicted response of T-beam using the DLF micromodel and

progressive damage approach – configuration 1.
450
1.125
Predicted
Ultimate Load
400
1.000 Average Ultimate Test Load
A
350
0.875
T
300
0.750
Normalized Load
R
0.625
250 L
0.500
200
0.375
150 2tw
0.250
100
Prediction with using 3D nonlinear DLF
0.125
50 micromodel with progressive damage
Test Results
0.000
0
0.000
0.00 0.125
0.05 0.250
0.10 0.375
0.15 0.500
0.20 0.625
0.25 0.750
0.30 0.875
0.35
Figure 9. Predicted response of T-beam using the DLF micromodel and

progressive damage approach – configuration 2.

8. Comparison of predicted and experimentally obtained damage
locations for DLF angle bracket
The micromechanical and structural framework along with the micromechanics-level progressive
damage approach is also used to predict the performance of a DLF angle bracket part. Twenty-
nine parts were tested under the loading condition shown in Figure 10. Each part was attached to
the steel loading arm and base plate using steel bolts. An FE model with symmetric boundary
conditions is developed to represent the actual conditions of the part testing.
∆ Symmetrical
FE model
Contact b/w
bottom surf ace of
Contact b/w
angle bracket &
surf ace of bolts
steel f ixtures Contact b/w
& surf ace of
analytical rigid pin Plane of flakes are in 1-2
angle bracket
& notch surf ace plane. 3-direction (out-
of-plane) is normal to
the exterior surfaces.
Fasteners: C3D10
C3D8R C3D10I
C3D4
Tie constraints
b/w surf ace of
bolts & steel Displacement
f ixtures
loading
Figure 10. FE model for the DLF angle bracket.
Figure 11 shows the comparison of the predicted damage locations in the DLF part with the
damage seen at the end of the part testing. Gray areas in the FE contour plots are the predicted
damaged areas where the damage index value for the 3D Tsai-Wu failure criterion exceeds 1.
Good agreement is shown when comparing the experimental results and the FE model predictions
for the damage locations in the DLF part.

Test
FEA predictions
Figure 11. Predicted and experimentally obtained damage locations for the DLF
angle bracket.
9. Conclusions
An effective integration of micromechanical and structural framework is formulated for the
nonlinear analysis of DLF composite materials and structures. The DLF micromodel explicitly
recognizes the response of the fiber and matrix constituents at the lower level of hierarchy in the
framework. The developed model predicts the overall nonlinear behavior of DLF composite
material under a general multi-axial state of loading. Coupon tests are used to experimentally
verify the predicted stress-strain behavior using the micromodel. The DLF micromodel
demonstrates good prediction capabilities for the effective properties and multi-axial nonlinear
behavior of DLF composites.
The framework is easily incorporated within a classical displacement-based FE structural
modeling using 3D continuum elements and additional stress correction algorithm. A damage
modeling approach is also integrated with the micromechanical model to form a framework for the
progressive damage analysis of DLF composite structures. The new “element deletion control by
state variables” functionality in Abaqus/Standard version 6.14 is also employed to perform another
type of progressive damage analysis approach. The modeling approaches are applied to DLF
composite parts made up from an AS4/PEEK material system. Ultimate (failure) loads are
predicted by the framework for different DLF composite parts. Contours of the attained maximum
value of the 3D Tsai-Wu criterion are plotted for the DLF parts to indicate the damage
progression. The framework shows very good predictive capabilities for the overall effective
properties, nonlinear response, and ultimate load of DLF composites.

10. References
1. Greene, T., “Discontinuous Long Fiber Composites for Complex Shape Metal
Replacement,” CAMX 2014 Conference Proceedings, Orlando, FL, June 2-5, CAMX –
The Composites and Advanced Materials Expo, CD-ROM-13 pp, 2014.
11. Acknowledgements
The author would like to thank the colleagues in the Quality Department of Greene, Tweed and
Adnan Khawaja for creating the part scan images. He would also like to acknowledge and thank
the colleagues in the Advanced Technology Group of Greene, Tweed-Switzerland for performing
the mechanical tests.

An Approach Towards an Optimal Design of
Composite Structures Using Abaqus as FE-solver
M. Prackwiesera, M. Gratta and B. Gollera*
a
INTALES GmbH Engineering Solutions, Innsbrucker Str. 1, 6161 Natters, Austria
*goller@intales.com
Abstract: The present paper shows the development of an efficient, fast and reliable optimizer
for composite and metallic parts of lightweight structures. The algorithm aims at identifying the
optimal configuration of different structural parts concerning thickness, fibre orientation, number
of plies, etc. This leads to mass savings and also a decrease of the development time in the
structural dimensioning phase. Due to the limited applicability of classical optimization
algorithms like gradient based or evolutionary methods in case of large Finite Element models
with a high number of design variables, a novel approach is presented where the optimization
problem is tackled by a heuristic adaption procedure on element level. This approach will be
illustrated on small numerical examples to show the functionality of our optimizer. Abaqus user
subroutines provide accessible interfaces which are embedded in our optimization workflow in
order to apply the optimization routine to large industrial problems. An example for such a
problem is given at the end of this paper to demonstrate the smooth interaction between
optimization routine and Abaqus.
Keywords: Aircraft, Composites, Design Optimization, Minimum-Weight Structures, Optimization,

Output Database, Post-processing, Shell Structures.
1. Introduction
One main aspect for the competitiveness in the design of lightweight structures is the
identification of weight saving opportunities while fulfilling multiple constraints in terms of
allowable values for stresses, strains, displacements etc. The variables to be optimized are the
number of plies of different parts, the orientation of the fibres and also their stacking sequence in
case of composite parts. For metallic parts, the thicknesses are to be optimized, while the diameter
represents the design variable related to fasteners. For real structures, the number of variables to
be optimized might soon reach a limit where classical optimization algorithms, like e.g. gradient-
based methods or also evolutionary algorithms, might not be applicable due to the curse of
dimensionality. An infeasible high number of model evaluations would be required in order to
identify a possible solution.
The topic of the here described project is the development of a novel approach where the
optimization problem is tackled by a heuristic adaption process on the element level. More
specifically, the starting point of the algorithm is the element stiffness matrix, which is adjusted
according to the temporary state in terms of failure criteria at a certain load level. The external
forces are applied in several load increments until the full load level is active and during each of
these increments an adaptation process is performed. The process can be compared with the

mechanical adjustment of biological materials, like wood and bones, where the stiffness formation
process is determined by the acting forces.
The key aspect in this process is the access of the element routines and the development of an
adaptation strategy for each element type in order to obtain optimal convergence. A number of
state-of-the-art element formulations have therefore been implemented and linked with the
commercial Finite-Element (FE) code Abaqus. In order to manage the independently adjusted
properties of each element, an efficient data-management is required which has been solved by
connecting an SQLITE-database to the element routines, where the values of each iteration of all
variables to be adjusted are written and accessed in the next iteration. Strategies for the type and
magnitude of change of the distinct design variables have been investigated where especially the
robustness of the solution, i.e. its sensitivity with respect to small deviations of input parameters,
is in the focus of the studies.
The main advantage of this approach is its applicability to large and complex FE-models, where
also several million degrees of freedom do not lead to algorithmic limitations. In addition, the
strategy can be used for models involving different kinds of elements, like shell elements
including layered composite, sandwich or isotropic (metallic) shell sections, and also fastener
elements.
The structure of the present paper is as follows: after a brief discussion of the theoretical
background in Sec. 2, the implementation of the element routines in the framework of Abaqus
with the link to the SQLITE-database is addressed in Sec. 3. The augmentation of the element
routines by failure criteria and the respective strategies for adaptation are topic of Sec. 4. The
quality of the employed adaptation strategies is judged by means of some small test examples.
2. Theoretical background
The goal of the present project is the optimal design of lightweight structures such that the weight
is minimized and all design criteria are fulfilled. This can be expressed as a constrained
optimization problem, i.e.
min f (x ) (1)
subject to g i (x ) ≤ ci for i = 1 m , (2)
where x are the design variables, like thickness, fibre angle, fastener type, etc., f (x ) is the
objective function, i.e. the total mass, and g i (x ) are the constraints which are defined by failure
modes, design rules and feasibility of construction. Classical optimization algorithms might soon
reach their feasibility limits since the number of design variables is in the range of hundreds or
even thousands for large FE-models.
Hence, an alternative approach based on an iterative adaption theory on element level is followed
in this paper, which is inspired by ideas like "fibre steering" (aligning the fibres along the most
effective direction), "computer aided internal optimization" (Kriechbaum, 1992), and "multi-
domain topology optimization" (Ma, 2006). More specifically, during the structural analysis the
status of each element in terms of fulfilling the investigated criteria is checked and - if necessary -
the values of the design variables of the element are changed. This change may imply an increase

of the thickness in case of a metal, the change of the draping angle or of the stacking sequence in
case of composite elements or also the increase of the diameter in case of a fastener element. In
order to consider a possible load distribution when changing the stiffness of the structure due to
this local adjustment, the external load is applied iteratively. In this way, the structure can "grow"
according to the acting load, which is a concept inspired by mechanical adaptation of biological
materials. As it also applies to wood, the growth is only additive and hence, the material is never
reduced. In this way, it can be ensured that in case of multiple load cases only the critical load
cases are driving for each element, while a lower external force does not change the element
properties.
The crucial point is the definition of the adaptation strategy, which has to be defined for each
failure criterion. A measure of the degree to which the limit factor is exceeded for each of the
criteria is the so-called reserve factor (RF), which defines the factor by which the load is increased
or decreased, respectively, in order to reach the limit state given by an RF equal to 1.0. Hence,
this value is a measure by which factor e.g. the thickness of a metallic element or the diameter of a
fastener is to be increased. In case of multiple design variables belonging to one element or to one
criterion, the strategy becomes more involved and has to be specific to each case. The approaches
which have been developed will be described in detail in the following section.
3. Development of the optimization procedure

The idea behind the optimization process is the adaptation of the properties of each element
according to its state with respect to the failure criteria. Hence, for each material/element type a
strategy for adjusting the element has been developed and implemented in the element routine (see
Sec. 5.1). Based on the current values of the element forces, moments and/or stresses, the design
variables of the analysed element are adjusted.
In the following, the strategies applied to the different element types are discussed and an example
will illustrate the functionality of the algorithm.
3.1 Composite elements
The first kind of failure criteria which is investigated are composite failure modes. To quantify an
eventual failure, we calculate for each criterion the quotient of actual occurring value of strain or
stress and the corresponding threshold marking failure. Thus we obtain a degree of utilization. A
value above 1 means failure.
One frequently applied approach is the maximum strain criterion. The requirement is formulated
such that the maximum strain ε1 occurring in an element must not exceed a user-defined
threshold, which is usually in the range of 0.3% to 0.4%:
 ε1
f = for ε1 ≥ 0
 εt
Maximum strain criterion :  (3)
ε
 f = 1 for ε < 0
 εc 1


εt  allowable tension strain
εc  allowable compression strain
In addition, two other criteria are applied, which are based on comparison stresses. In case of fibre
failure the Yamada-Sun criterion is relevant because it compares the acting and allowable stresses
in fibre direction:
 2 2
 f =  σ 1  +  σ 12 
 for σ 1 ≥ 0
 X   S
  t    (4)
Yamada − Sun criterion : 
2 2
  σ1  σ 
 f =   +  12  for σ 1 < 0
X 

  c  S 
Xt  allowable tension stress in fibre direction
Xc allowable compression stress in fibre direction
S allowable shear stress

If the Hashin criterion is the decisive criterion we can assume that the matrix fails, because the
stresses perpendicular to the fibre direction have been exceeded:
 2 2
 f =  σ 2  +  σ 12  for σ ≥ 0
 X  2
 t  S 
Hashin criterion :  (5)
2 2
  σ 2   σ 12 
 
 f =  X   S  for σ 2 < 0
+
  c  
Xt  allowable tension stress in fibre direction
Xc allowable compression stress in fibre direction
S allowable shear stress

Thus we obtain three degrees of exploitation in which the highest value is corresponding to the
maximum material utilization.
The following strategy adapts the layup to meet all criteria. It can be applied to both four-node
quadrilateral and three-node triangular shell element. In addition, also for the dimensioning of the
face sheets and core thickness of sandwich elements and of the curved element this approach is
employed.
3.1.1 Strategy to adapt the composite structure
In case the requirement is not fulfilled, the following adaptation strategy is applied: the section
forces for initially isotropic material properties (smeared stiffness of an initial stacking) are
calculated at each element centre. The load of all load cases is applied in several increments and

thus we obtain the first section forces for an incremental load of e.g. 10%. Then the optimization
routine starts to search the dominating direction for each element using a heuristic method. The
result is the draping angle of the stacking which suits the most with the applied loads and
generates therefore the lowest degree of utilization for the element. Further, a grouping parameter
g is calculated, which depends on the degree of anisotropy of the stress state. The proposed
stacking then follows a strict sequence, 45/90/-45/0g, which is repeated until the required
membrane and bending stiffness is obtained. The parameter g describes how many plies with a
direction of the draping angle follow each other. If these adjustments are not sufficient to fulfil all
criteria, an additional ply is added to the stacking of the element and the entire structure is
realigned in the following step. This is repeated until all load increments of all load cases are
applied and a final stacking is obtained for all elements.
The approach will be illustrated using quadrilateral monolithic elements. For triangular elements
or all other previous mentioned elements, the same strategy can be applied.
3.1.2 Numerical example: plate with hole
The first example used for illustrating the method is a plate with hole under tension load as it can
be seen in Figure 1. The structure is clamped at the left end and loaded with constant point loads
of 2000 N in positive x-direction at the right edge. The plate is made out of composite (fabric),
where the initial configuration is given with 4 plies (45/0/0/45) with a ply thickness of 0.28 mm.
The Young's modulus in both directions is given with 62000 N/mm2.
Figure 1. FE-model of the plate with hole under tension

The design variables are the number of plies and the draping angle, where the values after the
optimization process are shown in Figures 2 and 3. The direction of the fibres is optimal if it is
aligned with the direction of the main load path. Since a plate with hole under tension is a well-
known benchmark example, the direction of the principal stresses is known from analytical
solutions. This direction can be recognized also in Figure 2, where however, the curvature of the
directions around the hole is not as pronounced as in the analytical solutions. This can be
explained by the fact that the material is not isotropic due to the fibres and hence the optimal
direction is not equal to the draping angle. The draping angles of the single elements as indicated
in Figure 2 represent therefore a plausible solution.
The number of plies is shown in Figure 3 for each element of the model. As it is known from
analytical solutions, the highest stresses are located at the top and bottom of the hole. These are

also the two regions where the highest thickness increases are performed during the optimization
process. The highest value for the number of plies has been set to 50 (user input prior to the
analysis) and this is also the value which has been reached by the elements in these areas. The
definition of a maximum values for the plies is motivated by the situation that an increase of the
number of plies provokes a higher attraction of loads to this area (due to the higher stiffness) and
hence an unrealistic thickening would occur around stress peaks. By limiting the number of plies
to a user defined value, a certain re-distribution of the stresses around the location of the peak
values can be obtained. The increase of the number of plies in the two regions from the hole to the
support reflects the direction of the forces in the structure. Also at the support a certain increase
can be seen due to the higher stresses.
Figure 2. Draping angle after optimization
Figure 3. Number of plies after optimization
3.2 Metallic elements

In addition, a simultaneous optimization of the thickness of metallic parts is envisioned. Thus, the
optimizer allows the structural dimensioning of hybrid parts, since the stiffnesses of metal and
composite components are interdependent in terms of the global stiffness of the whole structure.
The strategy of optimising is analogous to the composite one and is also based on a heuristic
approach. The thickness of each shell-element is therefore adapted according to its state with
respect to the von Mises yield criterion (calculated for a general plane stress load scenario)
σv (6)
σ v = σ 12 − σ 1σ 2 + σ 22 + 3σ 122 ≤1
σ v ,allowed

σ 1 / 2  normal stresses
σ 12  shear stresses
σ v  von Mises stresses
3.3 Fastener elements
To meet all requirements of an optimizer which should be able to support an engineer in the
dimensioning of complex structures, also fasteners are considered in the optimising process. Three
criteria assess an eventual failure of a fastener and thus steer its dimensioning. We take into
account a combined failure criterion
2 2
 σn   σs 
  +  ≤ 1, (7)
σ  σ 
 n ,allowed   s ,allowed 
σ n  acting and allowable normal stresses

σ s  acting and allowable shear stresses,
a pull-trough failure criterion
P and σ PT
σ PT = ≤ 1, (8)
1 σ
(D − d ) π
2 PT ,allowed
4
P acting force
d diameter of the collar
D diameter of the head
and a bearing failure criterion, that compares the stresses which are transferred from the collar of
the fastener to the plate material with the allowable bearing stresses
Pshear σ BEA (9)
σ BEA = ≤1
dπt 0 σ BEA,allowed
Pshear  acting shear force
d diameter of the collar
t0  thickness of thinnest connecting plate.
4. Smoothing to provide manufacturability

In the optimization procedure, manufacturing issues have to be considered meaning that the term
"optimal configuration" of structural parts implies the condition of manufacturability. In terms of

manufacturing issues especially a smooth surface is desired, meaning that there is no abrupt
change in thickness or stacking of a part. Since in our approach each element can theoretically
have a different stacking or thickness than its neighbour, it is necessary to smooth the element
properties over the structure to get connected homogenous areas.
The idea is not to smooth the final stacking, but to pass the section forces already smoothed to the
optimizing routine in every load increment. Since the structure is consequently adapted for the
smoothed section forces, additional optimizing loops without increasing the load after the full load
are performed in order to adapt the elements to the actually occurring section forces.
The smoothing has always to be understood as a compromise. We want to somehow maintain the
information calculated from the optimization routine, but consider also the manufacturing process.
Hence it seems reasonable to weight elements which are nearer to the considered element higher
than others whose distance is larger. Thus we are going to use a weighting function ϕ which is
dependent on the distance r between the centroids of two elements to determine the weights for
the smoothing routine:
 1  r  2 
ϕ : r → exp−    , (10)
 2  λ  
where λ has to be chosen by the user (we set it to 1). The weighting ϕ is now strongly scale
dependent. Therefore we have to transform the distances between neighbours from the interval
[0, R] to an interval [0, b] . The variable R is the radius within which the elements are considered as
neighbours. We transform the intervals by the affine transformation:
x−c
γ [ a ,b ],[ c , d ] : x → a + (b − a) , (11)
d −c
where we usually set a = 0, b = 0, d = R. Now the weight vector is given by:
wikl = wik = ϕ (γ [ a ,b ],[ c , d ] (rik ) ), where rik = xi − xk . (12)
After all the weights are calculated the whole weight vector needs to be normalized:
∑
nk
i =1
wik = 1 for all k = 1,, N . (13)
Then, the smoothed section forces are given for every load increment l by
~
sf kl ( j ) = ∑ w sf i k
l
( j ), (14)
i∈N k
~ l ( j) =
sm k ∑ w sm ( j ),
i∈N k
i
l
k for k = 1,  , N and j = 1,2,3, (15)
where N k is a set containing the indices of the elements belonging to the neighbourhood of the

k-th element and N is the total number of elements of the model. The actually occurring section
forces are expressed by sf k and smk .
4.1 Numerical example for smoothing

We demonstrate the effects of smoothing on a cylindrical roof, which is loaded with a distributed
load on the left half of the structure. Figure 4 shows the obtained number of plies after optimizing
the roof with the strategy presented above. The blue surfaces are marking composite elements with
few plies whereas the more the elements tend to appear red the thicker the stacking is.

As first step we have to choose a suitable radius for the neighbour search. The radius R has a great
influence to what extent the properties of the elements are smoothed. In Figure 5 we can see that
all red elements affect the green in the middle. A smaller radius means to stay closer to the
‘optimal’ solution and a larger radius creates a more continuous stacking and thickness.
We set the radius for this example to R=50 and the angle for the allowed variation of the element
normals to α=30° (Figure 5).
The result of the smoothing process is a roof with more homogenous areas (Figure 6). This is also
evident in the maximum number of layers which are decreasing from 40 to 35. The total number
of plies of the whole roof construction is increasing. Thus, the structure is less optimized with
regard to the input data, i.e. it has a higher weight. However the advantage is that the gained result
after smoothing is less sensitive to slight changes in the input parameters and is therefore more
robust.

Figure 5. R-α-neighbourhood (red) of an element (green)
Figure 6. Number of plies after smoothing
5. Embedment of ABAQUS in the optimization process

In order to implement the proposed optimization algorithm at element level, an access to the
element routine has to be provided. This is necessary since the layup of the shell elements and the
diameter of the fasteners are changing between different steps of one Abaqus job. This means that
the mechanical and geometrical properties, which are required for the definition of the element
stiffness matrix, are changed according to the demands of the optimization process. The data to
exchange the properties of the individual elements are handled over an external database. Every
single element has to access this database before calculating its local stiffness for the global
assembly. Therefore full access to the element routine is crucial to render the interventions of the
optimizer possible. Abaqus provides interfaces for the user-defined interference in the solution
process (Abaqus User Subroutines Reference Manual). The user subroutine employed for the
implementation of the element routines is called UEL. Beside the mandatory committal of the
element stiffness matrix to the Abaqus solver, several parameters, as for example the current step

number, can be read in and out over this interface. This allows us to steer the optimizer via the
information provided by the subroutine. Thus, in order to make the optimizer applicable to a
certain region of a model, the type of element used for that component must be implemented as a
user subroutine. Hence, the most frequently applied element types have been coded in FORTRAN
in terms of the UEL for Abaqus, which are three- and four-node shell elements, a sandwich
element, an element which can model the actual geometry of a curved part and a fastener element.
All element formulations are state-of-the-art and were compared with success to the Abaqus
elements to which they correspond.
Figure 7. Solution sequence with interfaces to user-subroutines
Figure 7 shows the sequence of a linear static analysis in Abaqus and those stages of the analysis
where the user subroutines needed by the optimization algorithm are invoked.
As already mentioned, we need to manage the results of the optimization process over a database
in order to provide them for the consecutive calculation. The user subroutine UEXTERNALDB
provides an interface between Abaqus and an external database. For complex Finite Element
models, a large amount of data has to be handled since the current properties of each element have
to be stored in every iteration and for each load case and retrieved in the next iteration. An
SQLITE-database provides a suitable means for this problem since data can be stored and
accessed efficiently. The optimization algorithm is based on adaption of the element properties
during an iteratively increased external load. In order to handle this iterative increase of the load
while performing a full analysis as shown in Figure 7, a step in Abaqus is understood as an
iteration of the optimizer for one load case. For this purpose multiple steps are defined in the
Abaqus input file. For a pre-defined number of n iterations, n steps have to be defined for each
load case k, where in the i-th step the total external load is multiplied with the factor i/n. All

together are resulting n times k steps for an optimization run. The optimizer creates a separate
database for every load iteration and load case which contains the current layup and the
exploitation of each element according to failure criteria from Sec. 3. Thus, the full history of the
optimization process is available. After all load cases were calculated for a load increment, a
database is created which summarizes all results for this load increment and in which the different
stacking per element are merged to a new basis for the next load increment. Thereby, the results
of the optimization run can be traced completely.
Figure 8. Screenshot of the results database (each row for an element)
Although the results in the SQLITE-database format (Figure 8) are easy to handle, a graphical
illustration of the results becomes necessary for a quick overview or to place the results at the
disposal for customers or other potential users. User elements cannot be displayed in Abaqus
/CAE and the results for section forces and stresses are only available in the database. Therefore,
we developed our own viewer (called ALETHIA) to provide the visualization of the results.
Figure 9. Aletheia viewer to illustrate results by accessing the SQLITE database

6. Industrial example
Finally, the optimizer is applied to a large industrial example. The level of detail is decisive here,
since even small components can cause several thousand parameters to optimize in case of a
configuration with multiple different composite layup areas consisting of several materials with
several failure criteria and consequently with different constraints for each element. This creates
an optimization task with a vast multi-dimensional parameter space, several constraints depending
on the properties of the element and the objective to find a configuration of the structure which is
able to carry the applied load cases and whose parts are not oversized at the same time, meaning
that the result is as light as possible.
The structure we consider in the following consists of metal parts and composite components
which in turn can be divided according to their setup into tape, fabric and sandwich areas. Several
different bolt types serve as fastener to connect the different parts. Thus, all components are
included which can be part of a lightweight structure.
The FE-model we consider consists of over 800.000 elements, thereof 460.000 shell elements.
These in turn can be assigned to 2000 different stacking and setups in the start configuration.
Summing up, the whole model is defined by over four million degrees of freedom.
Depending on the configuration of the FE-simulation, the time which is necessary for one step in
Abaqus (linear computation to converge for a certain load level) varies from less than an hour to
several hours when contact between different parts of the model is fully considered (over contact
pairs). The functioning of the optimizer causes that this time has to be multiplied by about 11: 10
times to apply the load with an increment of 10% and a last time to calculate the exploitation of
the final structure (see Section 4). Since the 45 load cases are independent, they have to be applied
successively, and as such the total amount of steps which has to be evaluated rises to 495. The
high amount of steps in one Abaqus job and the consideration of contact in the calculation lead to
two problems for which workarounds were necessary.
The first problem is to define two user defined elements as contact pair which is not possible by
default. The simplest solution we found is to cover all user elements with “dummy” elements. In
doing so, we assign to Abaqus built-in elements (for example S4 or S3) a negligible stiffness and
no weight so that they are not affecting the calculation. We use these elements then to define
master and slave of the contact pair.
The second problem is the number of CLOADS per step. The load is defined over an include file for
each load case which consists of about two million lines with node, degree of freedom and value.
By starting a job with more than 35 steps the pre.exe of Abaqus terminates with an ERROR
message because unit 25 exceeds the limit of 16 GB. Since this limit cannot be increased currently
the only solution was to divide the job into several jobs with restarts and adapt our FORTRAN
tools to this new circumstance. Python tools are generating the input-files for the start and the
restarts and a bash file guarantees that the whole optimization process is still running
automatically.
Once the workarounds are implemented, the optimizer is applicable to the presented and any other
lightweight structure regardless of the size. Since the 45 load cases are acting independent, each

can be seen as individual optimization task. In Figure 10 and Figure 11 the resulting number of
plies and thicknesses of two components out of the full structure are shown. It shall be noted that
for this analysis no smoothing procedure has been applied, the results hence represent for each
element the optimal configuration for the 45 load cases.
Figure 11. Thickness of metallic component after optimization

The optimizer can hence consider all 45 different scenarios, is satisfying multiple constraints for
each material and provides as result a structure which is able to carry the applied load cases and
whose parts are not oversized at the same time, meaning that the result is as light as possible. In
total 495 evaluations of the full model are necessary to achieve this result which compared to other
optimization algorithms is more than competitive.
7. References
1. R. Kriechbaum, J. Schafer and M. Mattheck. CAIO (computer aided internal optimisation): A
powerful method to optimise fiber arrangement in composites. In First European conference
on smart structures and materials, 281-284, 1992
2. Z.D. Ma, N. Kikuchi, C. Pierre and B. Raju. Multidomain topology optimization for structural
and material design. ASME Journal for Applied Mechanics 73(4): 565–573, 2006
3. Dassault Systèmes. Abaqus 6.12 Analysis User's Manual.
4. E. Verginer. New methods for lay-up optimization of composites, Master Thesis, Innsbruck
2014

5. A. Kofler. Local and global smoothing of data for lay-up optimization of composites, Master
Thesis, Innsbruck 2014
8. Acknowledgements
This financial support of the Austrian Space Applications Programme of the Austrian Ministry for
Transport, Innovation and Technology (BMVIT) is deeply appreciated by the authors.
The authors also thank A. Kofler and E. Verginer, diploma students at the University of Innsbruck,
for their contributions to the paper.

Methodology for the Analysis of Tolerances in the
Assembly Process of a Wing Torsion Box
Through FE Simulations
A. Chiminelli1, F. Martín de la Escalera2, R. Breto1, E. Duvivier1, J.L. Núnez1, L.
Lizarduy3, M.A.Jimenez1.
1
Instituto Tecnológico de Aragón – ITAINNOVA, María de Luna 7-8, 50014, Zaragoza,
Spain.
2
Aernnova Engineering Solutions Ibérica, Avda. Manoteras 20, 28050, Madrid, Spain.
3
Aernnova Manufacturing Engineering, Portal De Gamarra 40, 01013, Vitoria-Gasteiz,
Spain.
achiminelli@itainnova.es
Abstract: This work presents a methodology for the analysis of deviations that might be
produced during the assembly process of aircraft components using the finite element
method as calculation tool. The methodology allows determining the effect that different
deviations in the constituent parts may have on the final tolerances of an assembly
considering the influence of the joining techniques and the different operations that are
usually carried out during such processes. This permits to evaluate if certain deviations
in some parts would maintain the final assembly within the specified tolerances or,
conversely, may accumulate/propagate negatively generating a non conformity. In
addition, the simulations performed allow analysing the possible effect of deviations
coming from the assembly tooling or different fastening sequences. In this sense, the work
performed demonstrates that the finite element method can be used effectively to improve
the processes commonly used in the manufacturing of components constituted by several
parts and with tight tolerances.
This work is part of the activities performed within the project “284961 CLEAN SKY,
Simulation of the assembly tolerances for composite aircraft structures – SATCAS”, led
by ITAINNOVA and performed under the supervision of Aernnova Engineering Division
as Topic Manager. SATCAS is a sub-project associated to the activity of Assembly
Simulation defined in the work-package WP 3.2.1.2 of BLADE – SFWA - Clean Sky
programme, led by Airbus.
Keywords: Simulation Methodology, Assembly Processes, Tolerance Analysis,

Deviations, Aircraft Structures.

1. Introduction
Independently of the manufacturing processes and their sophistication, parts always
present deviations and imperfections. The wear of the tools and moulds, misalignments in
the fixation systems and jigs, errors that the operators may introduce, deformations
caused by the own fabrication processes and many other factors; will inevitably generate
defects and variations in the components. In systems constituted by many parts, the
assembly processes acquire a special relevance, affecting the way on which the
deviations might accumulate/propagate.
For the aeronautic and automotive industries, the study of accumulative tolerances has
gained a significant relevance since its influence in the manufacturing costs is generally
important [1]. These studies allow predicting deviations in assemblies, understand how
they are produced, identify the main sources and, finally, obtaining recommendations to
reduce the manufacturing non-conformities.
The conventional tolerances analysis techniques focus on the evaluation of the impact
that geometric and/or dimensional deviations in the constituent parts may have in the
dimensions of the resultant ensembles. In most of these methodologies the parts are
considered as rigid bodies (i.e. the constituent components do not deform during the
assembly processes).In many applications this hypothesis is inadequate because:
-The assemblies are strongly affected by the part compliances (the hypothesis of parts as
rigid bodies is not acceptable because they are significantly deformed during the
assembly).
-The assemblies have such tight tolerance requirements that, despite the fact that the parts
behave almost rigidly, the deformations generated during the assembly processes cannot
be dismissed.
In those cases, issues come into play such as the interaction between parts as a
consequence of the deviations generated by the tools or the assembly jigs, the distortions
generated in the joints, or the spring-back effects in the parts. From a methodological
point of view, the incorporation of the part compliances in the tolerance analysis of
assemblies is usually carried out through the integration of mechanical/structural
calculation techniques with the methods of tolerances accumulation previously described
[2-4].
The methodology presented in this work is based on the Finite Element Method (FEM) as
calculation tool to analyse the accumulation and propagation of deviations considering

the parts compliance, necessary taking into account the tight tolerances that apply for the
NLF (Natural Laminar Flow) Wings object of study of SATCAS and the flexibility and
size of these assemblies.
The numerical methodology developed covers, among other, the following aspects:
- Definition of FEM strategies for the simulation of the assembly process.
- Development of an application for the treatment of results.
The method has been applied to the analysis of a NLF Wing Box defined within the
framework of the BLADE project previously mentioned.
Taking into account the complexity of the processes used in aeronautics, their simulation
becomes a major challenge. In this sense, it must be considered that the deformations in
the final ensembles can be caused by the deviations in the constituent parts, deviations in
the assembly jigs or by the assembly operations itself (for example deformations
introduced by the temporary fasteners or the bolts). All these possible sources of
deviations should be taken into account in the numerical simulations. The FEM strategy
proposed in SATCAS aims to cover all these aspects using different techniques to
simulate the operations and/or steps that may have an effect in the final deviations of the
resultant component.
Finally, the tolerances analysis performed in this work focuses on the waviness that might
appear in the Upper Cover surface, which can produce the lost of the laminar flow. This
analysis of waviness requires a relatively complex treatment of the FE results that, as it
will be explained later, has been implemented in a computer code and constitute part of
the methodology proposed.
This work has been developed as part of the project SATCAS - Simulation of the
assembly tolerances for composite aircraft structures. The main objective of the project
was the development of a methodology for the analysis of assembly tolerances
considering the propagation/accumulation of deviations coming from multiple sources
(parts, assembly jigs and deviations associated to the assembly process) and the parts
compliance. From the point of view of its application, the method is useful to: determine
the key driving parameters in the processes, analyse the possible influence of the joining
techniques (type of bolts, nuts, washers...), evaluate different assembly strategies, support
the decision regarding the fixation sequences and determine how the components
deviations can affect to the final assembly tolerances.
Within the methodology developed, this paper focuses in the simulation strategies
proposed for the assembly processes and in the application developed for the treatment of

results. In order to describe it and to show its operation, the results obtained from one of
the cases analysed is presented: a worst case for a deviation in the cradles of the assembly
tool.
2. Numerical simulation of the assembly process
The FE simulation strategies have been defined taking into account the assembly process
proposed for the NLF Wings. Among others, the following aspects have been identified
as critical in these processes:
- Simulation of initial fixations through temporary fasteners.
- Simulation of the bolts/rivets (including in-plane strains produced locally that,
accumulated, might generate global deformations).
In assemblies where liquid and/or solid shims are used to compensate the clearances that
usually appear between the constituent parts, the initial fixation through temporary
fasteners becomes an important assembly phase in terms of tolerances, since the
deformations that are produced by them are not avoided by the shims. Within the
methodology proposed, the fasteners are represented through two simple connector
elements and an additional set of ‘shell’ elements that represent the fastener head (surface
for contact, Figure 1). The first connector introduces the load (with a low stiffness to
guarantee that the entire load goes to fix the parts together) and the second one is used to
finally introduce the axial stiffness of the main rod of the fastener. The force can be
controlled through a load that operates in the main axis of the first connector. This
simulation strategy allows considering that the position of the bolts may change during
the assembly process (since the parts are being deformed and the drilling of the holes
could be executed in the course of the process).
Figure 1. Representation of the temporary fasteners in the FE models.

The forces that are introduced by the temporary fasteners were determined
experimentally, using a load cell as it is shown in figure 2.

Since hundreds of temporary fasteners might be required in a large aeronautical
ensemble, a script has been developed by ITA that allows to introduce the virtual
fasteners (connectors, shells, etc) automatically defining the positions of the pilot holes.
Figure 2. Measurement of the forces introduced by the temporary fasteners.

In relation with the bolts, they have been introduced in the simulations through:
- Connector elements, which allow to adequately reproducing the joints stiffness.
- A local field of in-plane strains, that is introduced to take into account the deformations
that are generated locally in each bolt and that may accumulate to generate further
‘stretching’ on the parts.
The definition of the connector elements in the models vary depending on the type of
bolt/nut to be considered. The adjustment of their stiffness has been performed through a
correlation based on a detailed numerical-experimental study carried out considering
different type of joints (Figure 3 and 4).

Grip length (50mm)
3 mm
Sensor LVDT 10 mm
L=242 mm
L=38 mm
Figure 3. Detailed numerical-experimental study of bolted joints.
Figure 4. Detailed FE models vs. simplified FE models.

The approach proposed for the introduction of the in-plane strain fields that are generated
locally in the bolted joints is based in the introduction of fictitious thermal expansions in
the materials (Figure 5). Through them, and imposing adequate changes of temperature, a
thermal stretching can be generated to simulate the mentioned phenomenon (depending
again on the type of bolt, diameter, nut, etc).

Figure 5. Strategy proposed for the introduction of the in-plane strains. Temperatures
(left) and strains (right).
The thermal expansion coefficients and the jumps of temperature are adjusted depending
on the substrates thicknesses, the materials involved and the type of bolts used. The radial
strains has been obtained from simple 3D-solid simulations performed for the different
materials/thicknesses and considering the preload introduced by the bolts (Figure 6.a.).
The values of the preloads were obtained experimentally in a previous study performed
by ITA. Then, the jump of temperature to obtain an equivalent radial strain (adjusted with
the average value of the strains through the thickness obtained from the 3D simulations)
has been determined by 3D-shell simulations considering the mesh shape (although with
a higher refinement) defined for the FE models of the assemblies in the zones of the
overlaps (Figure 6.b).
Figure 6. FE models used to adjust the thermal expansion coefficients and the jumps of
temperature for the introduction of the in-plane strains.

It is worthy of mention that the strategies presented are focused on FE models of large
ensembles, for simulations to determine/predict the deformations and the ‘waviness’ that
might be produced during the assembly processes in a global sense. A detailed analysis of
the deformations that might be generated locally near the bolts requires FE models as the
ones shown in figure 3, in a different scale than the ones covered by the approach
presented. However, the global deformations that can be produced by an accumulation of
the in-plane local strains that are introduced by the bolts should be covered by the global
FE analyses in some way. The strategy proposed is presented as a feasible approach to
cover this particular source of deviations. In terms of meshes, the FE strategies should be
suitable to be applied to models based mainly on ‘shell’ elements of reasonable size (in
order to maintain the computational cost of the simulations within practical limits). This
constitutes an additional challenge for the FE strategies proposed.
As for the temporary fasteners, the introduction of the connectors for the bolts and the
definition of the temperatures and materials properties required for the simulation of the
in-plane strains (through a proper definition of different sets of nodes and elements) have
been simplified through a script developed by ITA specifically with these purposes.
The type of FE calculations proposed for the assembly simulations are lineal-elastic and
without geometric or material nonlinearities. This assumption is based on two
considerations:
- The deviations that will be analysed are in all cases small enough to guarantee that
“small - displacement” hypothesis applies.
- The assembly process will never produce plastic deformations or damage in the parts.
Taking into account this, the main source of non-linearities in the simulations are the
contacts/interactions between the constituent parts/components (including the heads of
the temporary fasteners).
3. Treatment of results
In addition to the FE models itself, since as it has been mentioned the study is focused in
the analysis of the deviations and ‘waviness’ that might be produced in one of the
surfaces of the wings, a set of scripts/tools were needed to perform the analysis of results.
These tools were integrated in a code and constitutes part of the methodology developed.
The waviness is evaluated mainly through the ‘b/a’ ratio, with ‘b’ and ‘a’ as defined in
the following figure (measured against the nominal surface).

Figure 7. Parameters for waviness calculation.
Basically, the code/program developed performs the following tasks:
1.- Reading the FE results, obtaining the displacements of each node for the deformed
configuration (after each FE simulation, with a certain scenario of deviations).
2.- Execution of a best-fit between the nodes of the un-deformed configuration and those
of the deformed one.
3.- Reading of the original CAD surfaces (in order to be able to represent and manipulate
them).
4.- Projection of all the deformed nodes on to the CAD surfaces.
5.- Sectioning algorithm. The zone of interest is divided in several chord-wise or span-
wise sections where the profile and the b/a ratio are going to be calculated.
6.- Waviness calculation in the different sections.
The interface of the program developed is shown below.

Figure 8. Interface of the code/program developed.
An example of the application of this procedure for the treatment of results is presented
in the next section.
4. Application to an aircraft torsion box
As it has been previously introduced, the methodology has been applied to the analysis of
the assembly of the Torsion Box of a NLF Wing. Since the assembly jigs can be an
additional source of deviations, the FEM strategy defined also contemplates the

simulation of these tools and the systems used for the fixation/positioning of the parts
(cradles, vacuum vents/cups, etc).
Then, the FE model developed is composed by the following main parts:
- The assembly jig cradles for the Upper Cover and the Leading Edge, with a set of
connector elements representing the vacuum vents/cups.
- The assembly jigs supports for the positioning/fixation of the Front Spar and the Rear
Spar, with connectors to represent the fixation elements.
- Connectors for the positioning and guiding of the rest of parts.
- The constituent parts: Upper Cover, Leading Edge, Front Spar, Rear Spar, Ribs and the
Lower Covers (Front, Central and Rear).
- The connectors and the rest of elements used to represent the bolts, the temporary
fasteners and other fittings.
- The elements used to simulate the shim and sealant.
The parts mentioned are shown in the next figures (Figure 9):

Figure 9. Main parts/elements constituting the FE model for the RH Wing.
The full model developed is composed by 409243 elements and 341034 nodes. The
materials involved in the assembly have been defined as linear-elastic, isotropic for the
aluminium and the steel components and orthotropic for the Upper Cover that has been
designed with a carbon fibre – epoxy resin composite. The total amount of bolts in the
model is 3113 and the amount of temporary fasteners is 433. Their distribution is shown
in the next figure.
Figure 10. Elements used for the temporary fasteners & bolts in the Wing Torsion Box
model.

The exact assembly sequence defined for the ensemble is introduced in the simulations
through different calculation steps. For the base assembly process, more than 600 steps
are required.
The methodology has been applied to diverse cases: different deviations scenarios in the
constituent parts (for the cumulative tolerances analysis), deviations in the assembly tool,
alternative fastening sequences, etc. In this paper, only one of these cases is presented.
The FE model has been prepared using Abaqus/Standard v6.12. The simulations have
been performed using the same software.
4.1 Analysis of a deviation in the cradles of the tooling
The case presented is a deviation of +0.1 mm and -0.1 mm in the profiles of two
subsequent cradles of the assembly tool. These deviations are shown schematically in the
next figure. The figure also shows the results obtained from the FEM simulation of the
process and the field of deviations after the treatment of results with a detail of one
section (the one located in the zone where the deviations were introduced).

Deviations considered FEM results (displacements)
Treated results - final deviations after the assembly process and ‘b/a’ ratio calculated
Figure 11. Case analysed. Deviations considered and results.
The main conclusions that have been obtained from this specific analysis are:
- The deformation obtained in the section in the zone affected by the deviations
considered do not generate a waviness out of the acceptability range. Then, these
deviations in the tooling are acceptable.
- Considering the maximum vacuum level for the suction cups used to fix/support the
Upper Cover and the Leading Edge, a significant initial deformation might be introduced
in both parts. This is produced by the forces done by the vents/cups and by the fact that
there is a zone (corresponding to the zone in red in figure 11) where there is a nominal
gap between the Leading Edge and the cradles, allowing the parts to deform freely.

Moreover, these distortions are maintained throughout the entire process and their effects
do not disappear after the assembly (once the suction cups fixations are released), since
once the parts are assembled, they cannot recover freely (elastic recovery). In this sense,
a redesign of the fixation strategy in these zones is recommended in order to reduce these
deformations.
5. Conclusions
A methodology for the study of aeronautic assembly processes in terms of deviations and
tolerances has been introduced. The methodology, based on the finite element method as
calculation tool to take into account the parts compliances, has been applied to the
assembly of an aircraft Wing Box. Within the SATCAS project, the FE strategies
proposed for the simulation of these processes has demonstrated its feasibility, adequacy
and potential for this type of studies. An example of this type of simulations has been
presented in this paper.
The methodology developed has demonstrated to be useful for the analysis of deviations
that may arise during assembly processes caused by deviations in the constituent parts,
deviations in the assembly jigs and/or effects associated to the assembly operations itself.
Moreover, it has proven that it might be helpful to improve the processes (through
recommendations for the fixation systems, the assembly/fastening sequences, etc) or even
the design of the assembly jigs.
6. References
[1] “Variation simulation of fixtured assembly processes for compliant structures
using piecewise-linear analysis”, Michael Leon Stewart Master Thesis, Brigham
Young University, 2004.
[2] “An integrated methodology for statistical tolerance analysis of flexible

assemblies”, Alan J. Mortensen Master Thesis, Brigham Young University, 2002.
[3] “Deformation analysis of large aerospace components during assembly”, M.

Saadat, L. Cretin, R. Sim, F. Najafi, The International Journal of Advanced
Manufacturing Technology, March 2009, Volume 41, Issue 1-2, pp 145-155.
[4] “Predicting Deformation of Compliant Assemblies Using Covariant Statistical

Tolerance Analysis”, M. R. Tonks, K. W. Chase, C. C. Smith, Models for
Computer Aided Tolerancing in Design and Manufacturing, 2007, pp 321-330.
7. Acknowledgements
The research leading to these results has received funding from the European Union's
Seventh Framework Programme (FP7/2007-2013) for the Clean Sky Joint Technology
Initiative under grant agreement n° 284961.

High Velocity Perforation as a Benchmark Problem
for Material Model Validation
Vladislav V. Vershinin
Department of Structural Mechanics, Moscow State University of Civil Engineering, 26
Yaroslavskoe shosse, Moscow 129337, Russia
Abstract: High velocity perforation problem is taken as a benchmark to validate material

plasticity and fracture models. 2024-T3(51) aluminum alloy is chosen as a reference material.
Numerical simulations of a plate perforation using Johnson-Cook plasticity model with two sets of
parameters, Johnson-Cook fracture criterion with one set of fitting parameters and pressure and
Lode dependent ductile fracture criterion with three sets of material constants are performed into
Abaqus/Explicit. Calibration of all the material constants is accomplished using results of quasi-
static and high strain rate experiments. Problem formulation, plasticity and fracture models and
their parameters, heating effects and friction are estimated independently in the context of their
effect on the accuracy of the numerical solution. Ballistic limit velocity and residual velocity tests
are employed for validation procedure. For one of the considered problem-models combination
quantitative agreement with maximum discrepancy of less than 10% and good and physically
sound qualitative correlation with the experimental results were obtained.
Keywords: Plasticity model, fracture criterion, plate perforation, numerical simulation.
1. Introduction
Utilization of appropriate plasticity and fracture models in numerical simulation of various

processes followed by large deformation and fracture of a material is of paramount significance
for an accurate prediction of major qualitative and quantitative characteristics. High velocity
perforation of metal plates is among such problems. The perforation process is characterized by
high rate deformation and fracture. Stress states of the projectile and target materials are time-
variant and complex, while failure mechanisms of the interacting bodies are dependent on their
geometry and mechanical properties. That’s why ballistic behavior of plates made of a particular
material can be successfully used as a benchmark problem for validation of this material plasticity
and fracture models.
Metal behavior has been extensively investigated for more than a century. Dozens of metal
plasticity and fracture models have been developed, however open questions are still present that
is proved by ongoing theoretical research. Plastic behavior of particular alloys is anisotropic
(mainly in thin sheets) as well as dependent on pressure and the third invariant of the stress
deviator (e.g., Barlat et al., 1991; Karafillis and Boyce, 1993; Brünig, 1999; Kroon and Faleskog,
2013; Cazacu and Barlat, 2004; Bai and Wierzbicki, 2008; Seidt and Gilat, 2013), so that it is
necessary to employ newly developed elaborate plasticity models for a careful prediction of
yielding of such materials. Fracture behavior of some metals is also rather complex, sometimes
exerting dependence on stress tensor invariants (see Bai and Wierzbicki, 2008, 2010; Khan and

Liu, 2012; Oyane et al., 1980; Johnson and Cook, 1985; Lou et al., 2013, 2014) and coupling with
plastic behavior in the sense of its influence on strength of metals (Gurson, 1977; Tvergaard and
Needleman, 1984; Kachanov, 1958; Lemaitre, 1985; Chaboche, 1988a, 1988b; Brünig and Gerke,
2011). However, due to a comparatively large number of independent experimental tests necessary
to calibrate elaborate models simple plasticity models and fracture criteria are still popular.
In the present paper plate perforation by a rigid spherical projectile is considered as a benchmark
problem. As a calibrated and validated material 2024-T3(51) aluminum alloy is chosen. In the first
part of the article results of the experiments conducted by the author are combined with the data
available in literature and presented as a reference solution for the considered problem. In the
second part of the work a description of the employed plasticity and fracture models, their
parameters calibration procedure and results of numerical simulation utilizing these models are
presented. During the numerical simulation problem formulation, heating effects, friction,
elaboration and parameters of plasticity and fracture models are assessed independently in the
sense of their influence on the qualitative and quantitative accuracy of modeling.
2. Benchmark problem description
As a benchmark problem for validation material plasticity and fracture models it was decided to
consider plate perforation test with residual and ballistic limit velocities taken as reference
parameters. One of the best alternatives for being considered as a benchmark material was 2024-
T3(51) aluminum alloy, since large amount of experimental data and calibrated models of this
material are available in literature. Evidently, plate perforation by a rigid projectile should be
considered, so the simplest variant of this problem was chosen - circular 2024-T3(51) aluminum
alloy plates normally impacted at the center by a rigid spherical projectile.
It was shown in (Vershinin, 2015) that for circular 2024-T3(51) aluminum alloy plates with
normalized thicknesses H D ≥ 0.19 , where H is the plate thickness and D is the projectile
diameter, ballistic limit velocity Vbl (in m s ) follows an empirical relation of the form
0.673
622 ⋅ ( H D )
Vbl = − 33 (2.1)
This relation was obtained from combination of the published results (Senf and Weimann, 1973)
and new experimental data. In the new set of experiments free circular 2024-T3(51) aluminum
alloy plates with a diameter d = 81.4 mm and thicknesses in the range of = H 0.9 ÷ 12.2 mm were
used. Spherical projectiles with a diameter D = 10.0 mm were made of AISI 52100 alloy steel.
The Vbl values were determined as a half-sum of the lowest projectile velocity at which the plate
perforation occurred and the highest projectile velocity at which a hard steel sphere did not
perforate the plate. Additional details on the experiments can be found in (Vershinin, 2015).
It should be mentioned, that 2024-T3(51) aluminum alloy exerts no scale effects (Weimann, 1974)
and boundary conditions have almost negligible influence on the Vbl values for d D > 8 (Bivin,
2008, 2011). At the same time Eq. (2.1) is limited from above by cavitation phenomenon and
should be used very carefully for velocities greater than 1000 m s .

3. Plasticity and fracture models. Implementation into
Abaqus/Explicit
During numerical simulation the steel spherical projectile was modeled as elastic solid and the
aluminum target plate – as destructible elastic-plastic solid. Plasticity and fracture models utilized
for modeling of 2024-T3(51) aluminum alloy and implemented into Abaqus/Explicit are described
below. They are Johnson-Cook plasticity model (Johnson and Cook, 1983) and two
phenomenological fracture models – Johnson-Cook fracture criterion (Johnson and Cook, 1985)
and ductile fracture criterion with pressure and Lode dependence (Lou et al., 2014), which is
implemented into Abaqus/Explicit in a tabular form.
3.1. Johnson-Cook plasticity model
It was reported by many authors (see, for example, Barlat et al., 1991; Bai and Wierzbicki, 2008;
Seidt and Gilat, 2013), that 2024-T3(51) aluminum alloy is not isotropic and for careful modeling
of its behavior anisotropic and/or pressure and Lode angle parameter dependent yield functions
must be utilized. However, some authors (e.g., Lesuer, 2000) claimed, that stress-strain behavior
of the 2024-T3(51) alloy is isotropic and insensitive to mean stress. This statement was supported
(Lesuer, 2000) by compression and tension tests on a split Hopkinson pressure bar using samples
made from 4-mm thick plates. Stress-strain curves obtained in compression for the in-plane 0°
(rolling direction), in-plane 90° and normal (through the thickness) orientations almost coincided
(maximum deviation was 20 MPa). Material response in tension and compression under large-
scale plastic flow was also indistinguishable. So, it was decided to try Johnson-Cook plasticity
model (Johnson and Cook, 1983) at first. The choice was motivated by relative simplicity of the
model, its widespread use and because it was already implemented into Abaqus/Explicit. As it will
be shown below, the choice justified itself.
The Johnson-Cook plasticity model is based on the von Mises yield surface with the associated
flow rule. According to this model, isotropic hardening law depends on the strain rate and
temperature rise and has the following form:
  T − T q 

σ= n 
(
  )
  0
 A + Bε pl  1 + C ln ε pl ε 0  1 −  T − T  
 (3.1)
  m 0  
where σ is the von Mises stress, ε pl is the equivalent plastic strain, ε pl and ε0 are the current
and reference strain rates, respectively, Tm is the melting temperature, T0 is the transition
temperature below which the material strength has no temperature dependence, and
A, B, C , n, q are five material constants, which need to be calibrated from experiments. As it is
seen from Eq. (3.1) effects of strain hardening, strain rate hardening and temperature softening are
uncoupled.
Due to high strain rates the plate perforation process was assumed to be adiabatic, so that
numerical analysis was performed including effects of adiabatic heating. The temperature increase
∆T owing to plastic work is calculated on each increment through solving at each integration
point the following heat equation:

χ
T = σ : ε pl (3.2)
ρ cV
where ρ is the density, cV is the specific heat at constant volume, χ is the ratio of plastic energy
converted to heat, σ is the stress tensor and ε pl is the plastic strain rate tensor. Target material
physical, elastic and thermal properties assigned for the numerical simulation are taken from
(Lesuer, 2000; Teng and Wierzbicki, 2004) and presented in Table 1, and constants of the
plasticity model (5.1) are taken from (Lesuer, 2000; Johnson and Cook, 1983) and given in Table
2. Stress-strain curves obtained for the two material constant sets, which are presented in Tables 1,
2 and indexed as “P-I” and “P-II”, on a single element subjected to uniaxial tension under
isothermal and adiabatic conditions with ε pl = 1 are shown in Fig. 1.
Table 1. Physical, elastic and thermal properties of 2024-T3(51) aluminum alloy

Index ρ , kg m 3 E , GPa ν cV , J ( kg ⋅ K ) χ Tm , K T0 , K
P-I 2770 74.66 0.3 875 0.9 775 293
P-II 2770 74.66 0.3 875 0.9 775 293
Table 2. Constants of the Johnson-Cook plasticity model for AA2024-T3(51)

Index A, MPa B, MPa n C q ε0 , s −1
P-I 369 684 0.73 0.0083 1.7 1.0
P-II 265 426 0.34 0.0083 1.7 1.0
1,200
1,000
Von Mises stress, MPa
800
600
400
P-I, isothermal
P-I, adiabatic
200 P-II, isothermal
P-II, adiabatic
0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Equivalent plastic strain
Fig. 1. Stress-strain curves obtained in Abaqus/Explicit for the two material

constant sets on a single element subjected to uniaxial tension under isothermal
and adiabatic conditions with ε pl = 1

3.2. Johnson-Cook fracture model
The Johnson-Cook failure model is based on the value of the equivalent plastic strain at element
integration points. The damage parameter is defined as
ω
= ∑ ( ∆ε pl ε plf ) (3.3)
where ∆ε pl is an increment of the equivalent plastic strain, ε plf is the equivalent plastic strain at
failure, and the summation is performed over all increments in the analysis. When ω ≥ 1 failure is
assumed to occur. An expression for the equivalent plastic strain at failure ε plf which is used in
Abaqus/Explicit slightly differs from the original formula (Johnson and Cook, 1985) and has the
following form:
  p    ε pl    T − T0  
ε plf =
 D1 + D2 exp  D3   1 + D4 ln     1 + D5   (3.4)
 
  q     ε0    Tm − T0  
1 3
Here p = −σ m = − trace ( σ ) is the pressure stress, q= σ= ( S : S ) , where S= σ + pI is the
3 2
deviatoric stress, D1 − D5 are five material constants, which need to be calibrated from
experiments, and the other variables were defined earlier in the text (see Subsection 3.1). Eq. (3.4)
differs from the original formula in the sign of the parameter D3 . Values of the failure parameters
D1 − D5 are taken from (Kay, 2003) and presented in Table 3. Thermal properties are identical to
those used in the Johnson-Cook plasticity model (see Table 1).
Table 3. Parameters of the Johnson-Cook failure model for AA2024-T3(51)
Index D1 D2 D3 D4 D5 ε0 , s −1
JC 0.112 0.123 1.5 0.007 0.0 1.0
3.3. Ductile fracture criterion with pressure and Lode dependence

In Abaqus/Explicit the ductile fracture criterion may be defined in a tabular form as a function of
equivalent plastic strain rate ε pl , stress triaxiality η = − p q , and the normalized third deviatoric
3 13
stress invariant
= ξ (=
r q) cos ( 3θ ) , where r =  27 det ( S ) 2  is the third invariant of
deviatoric stress and θ is the Lode angle. The function ξ (θ ) can take values from ξ = −1 , for
stress states on the compressive meridian, to ξ = 1 , for stress states on the tensile meridian.
Damage accumulation and failure occur in the same manner, as in the Johnson-Cook fracture
model (see Subsection 3.2).
To obtain values of equivalent plastic strain at failure ε plf for different stress states a newly
developed fracture model (Lou et al., 2014) was used. The proposed ductile fracture criterion has
the form

C1 C2
 2   1  3− L 
   η + + C  ε plf =
C3 (3.5)
 2  1+ C  2 
 L +3    3 L +3  
Here C1 , C2 , C3 are three material constants to be calibrated from experiments,
L= (
3 tan (θ ) − 3 ) ( tan (θ ) + 3 ) is the Lode parameter which is related to the normalized third
deviatoric stress invariant ξ in the following way:
( ) ( L + 3)
32
− L 9 − L2
ξ= 2
(3.6)
Parameter C = − σ 1 σ represents the sensitivity of the cut-off value for the stress triaxiality to
microscopic structure of a material. By the cut-off value the stress triaxiality below which fracture
will never occur is meant. The Lode dependent cut-off value for the stress triaxiality is proposed to
take a form of
3− L
η+ +C =
0 (3.7)
3 L2 + 3
To calibrate the parameters C1 , C2 , C3 experimental data (Bao, 2003; Seidt, 2010) was used.
Values of stress state variables and equivalent plastic strain at failure ε plf from these two
experiment sets are presented in Table 4. It should be emphasized, that because of the fact that
stress state parameters (η , ξ , L ) are varying in the loading process, average values of these
parameters were used. That’s why three additional tests available (Khan and Liu, 2012) were not
included in the data set – the authors reported only values of the stress state parameters at failure.
The subscript “av” will be dropped and (η , ξ , L ) will be understood as the average values, unless
otherwise stated. Stress state parameters from the experiments (Bao, 2003; Seidt, 2010) along with
the cut-off surfaces for the stress triaxiality (see Eq. (3.7)) at three different values of the
parameter C are shown in Fig. 2.
Table 4. Equivalent plastic strains at failure and corresponding stress state
parameters for 2024-T3(51) aluminum alloy
Test # Ref. ε plf η ξ L
1 [93] 0.170 0.9300 1.000 -1.0000
2 [93] 0.210 0.6100 0.097 -0.0561
3 [93] 0.210 0.0124 0.055 -0.0318
4 [93] 0.260 0.1170 0.500 -0.3054
5 [93] 0.280 0.6300 1.000 -1.0000
6 [93] 0.310 0.3430 1.000 -1.0000
7 [93] 0.330 0.3560 0.984 -0.8666
8 [93] 0.341 -0.2240 -0.800 0.5531
9 [93] 0.356 -0.2330 -0.820 0.5749
10 [93] 0.360 0.3690 1.000 -1.0000
11 [93] 0.380 -0.2340 -0.810 0.5638
12 [93] 0.450 -0.2780 -0.910 0.6940
13 [93] 0.460 0.4000 1.000 -1.0000
14 [93] 0.480 0.3570 0.979 -0.8477

15 [93] 0.620 -0.2480 -0.840 0.5980
16 [94] 0.1036 0.5802 -0.0383 0.0221
17 [94] 0.1675 0.0006 0.0012 -0.0007
18 [94] 0.1951 0.5648 0.1474 -0.0855
19 [94] 0.1951 0.6052 0.2072 -0.1207
20 [94] 0.2017 0.7304 0.1085 -0.0628
21 [94] 0.2022 0.8556 1.0000 -1.0000
22 [94] 0.2160 0.2663 0.9336 -0.7352
23 [94] 0.2204 0.4974 0.6478 -0.4145
24 [94] 0.2419 0.4781 0.1896 -0.1103
25 [94] 0.2441 0.4319 0.8881 -0.6607
26 [94] 0.2447 0.7631 1.0000 -1.0000
27 [94] 0.2562 0.3376 0.9993 -0.9710
28 [94] 0.2656 0.1585 0.6581 -0.4228
29 [94] 0.2727 0.6765 1.0000 -1.0000
30 [94] 0.2832 0.5994 -0.4417 0.2662
31 [94] 0.2909 0.6072 1.0000 -1.0000
32 [94] 0.3080 0.5243 1.0000 -1.0000
33 [94] 0.3190 0.5994 -0.5180 0.3179
34 [94] 0.3224 0.3491 1.0000 -1.0000
35 [94] 0.5152 -0.1574 -0.6538 0.4193
ξ
1.0
0.8
0.6
0.4
0.2
0.0 η
-0.2
-0.4
-0.6 Experimental data [93]

Experimental data [94]
Cut-off surface I (C = 0)
-0.8
Cut-off surface II (C = 1/3)
Cut-off surface III (C = 1/2)
-1.0
-1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Fig. 2. Stress state parameters at failure and cut-off surfaces for the stress
triaxiality (see Eq. (3.7)) for 2024-T3(51) aluminum alloy
Calibration of the parameters C1 , C2 , C3 at three different values of the parameter C was

performed using MATLAB Curve Fitting tool. The results of the calibration procedure are
presented in Table 5.
Table 5. Calibrated material constants of the ductile fracture criterion for 2024-
T3(51) aluminum alloy
Index C C1 C2 C3
L-I 0 2.413 0.2762 0.2991
L-II 1/3 2.628 0.4879 0.3062
L-III 1/2 2.694 0.5842 0.3089
4. Results of numerical simulation
The numerical simulation was performed in both axisymmetric and 3-dimensional formulations
(indexed “AX” and “3D”, respectively). The projectile impacted the target at its center;
dimensions of the target and the projectile were equal to corresponding dimensions of the
projectile and the target plates used in the experiments performed (see Section 2). In all cases,
except those in which influence of friction was estimated, frictional coefficient was assumed to be
equal to 0.2. Steel properties defined were the following: Young’s modulus E = 210 GPa ,
Poisson’s ratio ν = 0.3 , density ρ = 7830 kg m3 .
As it is known processes followed by material fracture are irregular. However, analysis of the tests
conducted as well as other available experimental results (e.g., Wingrove, 1973; Borvik et al.,
2002) reveals that perforation of plates by rigid spherical nosed axisymmetric projectiles, which
occurs through both shear plugging and ductile hole enlargement, can be considered as
axisymmetric. So, it was decided, that axisymmetric problem formulation could also be employed
(at least for targets with H/D ≥ 0.4, whose mechanisms of failure correspond to the mentioned
above). It is realized by the author, that perforation of thin plates followed by petalling is strictly
unsymmetrical process, but to complete the set axisymmetric formulation was applied to the
whole range of the normalized thicknesses considered.
In the first set of the numerical tests Vbl values for different combinations of problem formulation
and plasticity and fracture models were determined. The Vbl values obtained are presented in
Table 6, while a disagreement of the numerical results with the experimental data (Vershinin,
2015) is shown in Fig. 3.
Table 6. Comparison of Vbl values obtained numerically with the experimental data
H D 0.1 0.2 0.3 0.6 1.2 2.0

Experiment 138.3 177.6 243.6 408.0 670.2 958.7
AX/P-I/JC 109.5±0.5 175.5±0.5 235.5±0.5 361.5±0.5 571.5±0.5 924.5±0.5
3D/P-I/JC 91.5±0.5 154.5±0.5 236.5±0.5 382.5±0.5 542.5±0.5 934.5±0.5
3D/P-I/L-I 97.0±2.0 167.5±2.5 239.5±2.5 345.0±1.0 670.0±1.0 995.0±5.0
3D/P-I/L-II 96.5±0.5 164.0±1.0 226.5±2.5 361.0±1.0 653.0±3.0 982.0±2.0
3D/P-I/L-III 96.5±0.5 169.0±1.0 223.0±1.0 369.5±0.5 639.0±1.0 977.0±1.0
AX/P-I/L-III 114.0±1.0 187.0±1.0 241.0±1.0 349.0±1.0 637.0±1.0 977.5±2.5
3D/P-II/L-III 93.0±1.0 157.0±2.5 213.0±3.0 336.0±1.0 597.0±3.0 922.5±2.5

10
-5
Error, %
-10
-15
AX/P-I/JC
-20 3D/P-I/JC
3D/P-I/L-I
-25 3D/P-I/L-II
3D/P-I/L-III
-30 AX/P-I/L-III
3D/P-II/L-III
-35
0 0.5 1 1.5 2
H/D
Fig. 3. Discrepancy between the numerical and experimental results (see Table 6)
It is clearly seen, that neither of the formulation-models combinations tested catches performance
of thin plates with H D < 0.15 . Since two different yield surfaces were tested (cases “3D/P-I/L-
III” and “3D/P-II/L-III”) and stress state of the target in the contact area changes with H D
increase, the most likely cause of such a discrepancy is that the fracture locus is poorly defined in
some area in the η − ξ plane. And there is a ring of truth about it, because, despite their large
amount, experimental data on fracture of 2024-T3(51) aluminum alloy, which were used for the
fracture locus calibration, are mostly concentrated in the first quadrant (see Fig. 2).
Meanwhile, compared with the experimental results, which follow Eq. (2.1), i.e. plates with
H D ≥ 0.19 , the error of the numerical results obtained using the “3D/P-I/L-III” formulation-
models combination does not exceed 10%. And not only the quantitative agreement, but also good
qualitative correlation of this formulation-models combination with the experiment was achieved.
Views of plates of various thicknesses H D after their perforation by a rigid spherical body with
velocity nearly Vbl are presented in Fig. 4, and in Table 7 dimensions of the ejected plugs for
H D in the range 0.1–2.0 are compared with the values obtained experimentally. In Fig. 5
temperature field in the contact area of the H D = 1.2 plate during its perforation is shown along
with view of the corresponding test specimen after impact. Molten metal, which leads to the
“snow cap” appearance on the projectile, and development of the “dead zone” are clearly seen. It
is also remarkable, that for plates with H/D ≥ 0.19 all of the problem-models combinations tested
were within 20% of the experimental results.

Of course, there is certain dissimilarity in the numerical results and experimental data. The most
pronounced take place for the cases, in which relative errors in Vbl determination are the largest.
That is plates with H D = 0.3 and H D = 0.6 . In both cases it is lack of ductility and fracture
locus poorly defined in some area of the η – ξ plane, which causes a discrepancy with the
experiment.
In the target with H D = 0.3 petals formed eventually break off. It should be mentioned, that
petals break off actually takes place at higher projectile velocity, so that this phenomenon in the
numerical simulation is just an early development of a really existing failure mechanism. As for
the plate with H/D = 0.6, projectile passage through the hole, formed in the target after the plug
ejection, varies in the numerical simulation and in the tests. As it is shown in Table 7, the plug
diameter is smaller than the projectile diameter. So, in the experiment the projectile pushes the
target material to the side (through ductile hole enlargement mechanism), when passing through
the target. In the numerical simulation, in contrast, target material on the whole inner surface is
just eroded. This yields lower resistance to perforation and underestimation of the ballistic limit.
Also it should be noticed, that no developed spalling on the front side of thick plates, as it takes
place in the experiment (Vershinin, 2015), was obtained in the numerical simulation.
Table 7. Dimensions of the ejected plugs. Comparison with the experiment
H D 0.1 0.2 0.3 0.6 1.2 2.0
h plug , Experiment No plug 0.91H 0.87H 0.68H 0.49H –
d plug , Experiment No plug 0.53D 0.80D 0.89D 1.13D –
h plug , 3D/P-I/L-III 0.87H 0.90H 0.86H 0.73H 0.49H 0.39H
d plug , 3D/P-I/L-III 0.30D 0.54D 0.59D 0.73D 1.11D 1.11D
a) b)

c) d)
e) f)
Fig. 4. Views of plates of various thicknesses H D after their perforation by a rigid
spherical body with velocity nearly Vbl : a) H/D = 0.1, V0 = 97 m/s; b) H/D = 0.2, V0 =
170 m/s; c) H/D = 0.3, V0 = 224 m/s; d) H/D = 0.6, V0 = 370 m/s; e) H/D = 1.2, V0 = 640
m/s; f) H/D = 2.0, V0 = 978 m/s

Fig. 5. Temperature field in H D = 1.2 plate during its perforation with V0 = 640 m s
and view of the corresponding test specimen after impact
In the second set of the numerical tests residual velocities Vr for plates of different thicknesses
were determined using the best formulation-models combination “3D/P-I/L-III”. Normalized Vr
versus V0 data are compared with the experimental ones (Senf and Weimann, 1973) and shown in
Fig. 6. Both numerical and experimental results scatter around a single curve, again confirming the
validity of the utilized models.

6
Experiment, H = 3 mm, D = 12.7 mm
3D/P-I/L-III, H = 2 mm, D = 10 mm
5
3D/P-I/L-III, H = 3 mm, D = 10 mm
3D/P-I/L-III, H = 6 mm, D = 10 mm
4
3D/P-I/L-III, H = 12 mm, D = 10 mm
Vr /Vbl

3 3D/P-I/L-III, H = 20 mm, D = 10 mm
0
0 1 2 3 4 5 6
V0/Vbl
Fig. 6. Normalized residual projectile velocities. Comparison with the experiment

Finally, an effort to estimate influence of friction and heating effects was made. To accomplish
this friction coefficient was, firstly, set equal to zero (no friction forces) and then to 0.47 (this
value is usually considered as coefficient of kinematic friction for contact pair aluminum-steel).
Results of numerical simulation of these cases are presented in Table 8 and indexed as “NF” (no
friction) and “MF” (maximal friction), respectively. Also presented in Table 8 are results obtained
numerically without any adiabatic heating effects and coefficient of friction set equal to 0.2. This
data set is indexed as “NH” (no heating). All other material constants and problem formulation for
these three extra cases were similar to those used in the “3D/P-I/L-III” formulation-models
combination.
Table 8. Comparison of Vbl values obtained numerically for three extra cases with
the experimental data
H D 0.1 0.2 0.3 0.6 1.2 2.0
Experiment 138.3 177.6 243.6 408.0 670.2 958.7
3D/P-I/L-III 96.5±0.5 169.0±1.0 223.0±1.0 369.5±0.5 639.0±1.0 977.0±1.0
3D/P-I/L-III/NF 91.5±1.5 – – 313.0±1.0 – –
3D/P-I/L-III/MF 102.0±1.0 – – 367.5±0.5 – 982.5±2.5
3D/P-I/L-III/NH 96.0±1.0 – – 375.0±1.0 – 967.5±2.5
As is clearly seen form Table 8, influence of friction and heating effects on Vbl values is
insignificant for the whole range of the target thicknesses considered. It should be only

emphasized that setting friction coefficient equal to zero results in significant drop in target
resistance to penetration, that is not consistent with the reality. This result may be considered as
evidence of non-zero friction coefficient between a projectile and a target during high-velocity
perforation despite high strain rates and melting.
5. Discussion and conclusions
Numerical simulations of circular 2024-T3(51) aluminum alloy plate perforation by a rigid

spherical projectile were conducted in both axisymmetric and 3-dimensional formulations using
Johnson-Cook plasticity model with two sets of fitting parameters, Johnson-Cook fracture
criterion with one set of fitting parameters and pressure and Lode dependent ductile fracture
criterion with three sets of material constants, which were implemented into Abaqus/Explicit code.
It was found that valid plasticity and fracture models are of equal importance for an accurate
modeling of high-rate, non-linear problems with material fracture, and disregarding of this fact can
lead to a typical mistake (Seidt et al., 2013), that one can make. That is utilizing elaborate
plasticity model and simple fracture criterion, or vice versa, that may result in a better agreement
with the experimental results, but be physically wrong.
With all the formulation-models combinations employed the simulations were within 20% of the
experimental results for the targets with H/D ≥ 0.19. Moreover, with the Johnson-Cook plasticity
model and the ductile fracture criterion with pressure and Lode dependence, which were calibrated
using available experimental data, implemented into Abaqus/Explicit code and employed for 3-
dimensional finite element analysis, quantitative agreement with maximum discrepancy less than
10% and good and physically sound qualitative correlation with the experimental results for
circular 2024-T3(51) aluminum alloy plates with the normalized thicknesses H D ≥ 0.19 were
obtained. These results confirmed the validity of the formulation-models combination employed
and allow one to consider numerical simulation of high velocity perforation as an alternate way for
validation of material plasticity and fracture models.
Influence of friction and heating effects appeared to be insignificant for projectile velocity up to
1000 m s , except the case when friction coefficient was set equal to zero, that resulted in
significant drop in target resistance to penetration not consistent with the reality and revealed non-
zero friction coefficient between a projectile and a target despite high strain rates and melting.
Acknowledgements
This work is a part of the research project financially supported by the Russian Foundation for
Basic Research (project No. 14-08-00016).
References
1. Bai Y, Wierzbicki T. A New Model of Metal Plasticity and Fracture with Pressure and
Lode Dependence. Int J Plast 2008;24(6):1071-96.
2. Bai Y, Wierzbicki T. Application of Extended Mohr-Coulomb Criterion to Ductile
Fracture. Int J Fract 2010;161(1):1-20.

3. Bao Y. Prediction of Ductile Crack Formation in Uncracked Bodies. PhD Thesis.
Cambridge, MA: Department of Ocean Engineering, Massachusetts Institute of
Technology; 2003.
4. Barlat F, Lege DJ, Brem JC. A Six-Component Yield Function for Anisotropic Materials.
Int J Plast 1991;7(7):693-712.
5. Bivin YK. Strain and Fracture of Circular Plates under Static and Dynamical Loading by
a Spherical Body. Mech Solids 2008; 43(5): 798-807.
6. Bivin YK. Fracture of Circular Plates on Normal Impact by a Rigid Spherical Body.
Mech Solids 2011; 46(4): 597-609.
7. Borvik T, Langseth M, Hopperstad OP, Malo KA. Perforation of 12 mm thick steel plates
by 20 mm diameter projectiles with blunt, hemispherical and conical noses: Part I:
Experimental study. Int J Imp Eng 2002;27(1):19-35.
8. Brünig M. Numerical Simulation of the Large Elastic-Plastic Deformation Behavior of
Hydrostatic Stress-Sensitive Solids. Int J Plast 1999;15(11):1237-64.
9. Brünig M, Gerke S. Simulation of Damage Evolution in Ductile Metals Undergoing
Dynamic Loading Conditions. Int J Plast 2011;27(10):1597-617.
10. Cazacu O, Barlat F. A Criterion for Description of Anisotropy and Yield Differential
Effects in Pressure-Insensitive Metals. Int J Plast 2004;20(11):2027-45.
11. Chaboche JL. Continuum Damage Mechanics: Part I – General Concepts. J Appl Mech
1988;55(1):59-64.
12. Chaboche JL. Continuum Damage Mechanics: Part II – Damage Growth, Crack
Initiation, and Crack Growth. J Appl Mech 1988;55(1):65-72.
13. Gurson AL. Continuum Theory of Ductile Rupture by Void Nucleation and Growth:
Part I – Yield Criteria and Flow Rules for Porous Ductile Materials. J Eng Mater Technol
1977;99:2–15.
14. Johnson GR, Cook WH. A Constitutive Model and Data for Metals Subjected to Llarge
Strains, High Stain Rates and Temperatures. In: Proceedings of the 7 th International
Symposium on Ballistics, Hague; 1983, p. 541-47.
15. Johnson GR, Cook WH. Fracture Characteristics of Three Metals Subjected to Various
Strains, Strain Rates, Temperatures and Pressures. Eng Fract Mech 1985;21(1):31-48.
16. Kachanov LM. Time of the Rupture Process under Creep Conditions. Izvestiya Akademii
Nauk SSSR. Otdelenie Technicheskikh Nauk 1958;8:26-31 (in Russian).
17. Karafillis AP, Boyce MC. A General Anisotropic Yield Criterion Using Bounds and a
Transformation Weighting Tensor. J Mech Phys Solids 1993;41(12):1859-86.
18. Kay G. Failure Modeling of Titanium Ti-6Al-4V and Aluminum 2024-T3 with the
Johnson-Cook Material Model. Technical Report DOT/FAA/AR-03/57. Livermore:
Lawrence Livermore National Laboratory; 2003.
19. Khan AS, Liu H. A New Approach for Ductile Fracture Prediction on Al 2024-T351
Alloy. Int J Plast 2012;35:1-12.
20. Kroon M, Faleskog J. Numerical Implementation of a J2- and J3-Dependent Plasticity
Model Based on a Spectral Decomposition of the Stress Deviator. Comput Mech
2013;52(5):1059-70.
21. Lemaitre J. A Continuous Damage Mechanics Model for Ductile Fracture. J Eng Mater
Technol 1985;107(1):83-9.

22. Lesuer DR. Experimental Investigations of Material Model for Ti-6Al-4V Titanium and
2024-T3 Aluminum. Technical Report DOT/FAA/AR-00/25. Livermore: Lawrence
Livermore National Laboratory; 2000.
23. Lou Y, Huh N, Lim S, Pack K. New Ductile Fracture Criterion for Prediction of Fracture
Forming Limit Diagrams of Sheet Metals. Int J Sol Str 2012;49(25):3605-15.
24. Lou Y, Yoon JW, Huh H. Modeling of Shear Ductile Fracture Considering a Changeable
Cut-Off Value for Stress Triaxiality. Int J Plast 2014;54:56-80.
25. Oyane M, Sato T, Okimoto K, Shima S. Criteria for Ductile Fracture and Their
Applications. J Mech Work Tech 1980;4(1):65-81.
26. Seidt JD. Plastic Deformation and Ductile Fracture of 2024-T351 Aluminum under
Various Loading Conditions. PhD Thesis. Columbus, OH: Department of Mechanical
Engineering, The Ohio State University; 2010.
27. Seidt JD, Gilat A. Plastic Deformation of 2024-T351 Aluminum Plate under a Wide
Range of Loading Conditions. Int J Sol Str 2013;50(10):1781-90.
28. Seidt JD, Pereira JM, Gilat A, Revilock DM, Nandwana K. Ballistic Impact of
Anisotropic 2024 Aluminum Sheet and Plate. Int J Imp Eng 2013;62:27-34.
29. Senf H, Weimann K. Die Wirkung von Stahlkugeln auf Dural-Einfach-und
Mehrplattenziele. EMI Report No. V6-73 (in German), 1973.
30. Teng X, Wierzbicki T. Effects of Fracture Criteria on High Velocity Perforation of Thin
Beams. Int J Comput Meth 2004;1(1):171-200.
31. Tvergaard V, Needleman A. Analysis of the Cup-Cone Fracture in a Round Tensile Bar.
Acta Metall 1984;32(1):157-69.
32. Vershinin VV. Validation of Metal Plasticity and Fracture Models through Numerical
Simulation of High Velocity Perforation. Int J Sol Str 2015 (submitted).
33. Weimann K. Penetration of Steel Spheres in Aluminum Targets. EMI Report, Е3/74 (in
German), 1974.
34. Wingrove AL. The influence of projectile geometry on adiabatic shear and target failure.
Metall Trans 1973;4(8)1829-33.

Simulation Lifecycle Management in the
Real world: Practical Application in PRIME
Aerostructures
A. Belles Meseguer
PRIME aerostructures GmbH, Inkustraße 1-7/4 A-3400 Klosterneuburg
a.belles-meseguer@primeaero.at
Abstract: Simulation has become a key factor in the development of competitive and innovative
products. However, the implementation of simulation in an efficient way goes further than
choosing the most powerful analysis tools or working with the best engineers: Managing
simulation data, the usage of non-standard methods or communication problems, are some of the
issues that need to be solved.
In Prime Aerostructures, the implementation of the Simulation Lifecycle Management System from
Simulia has given numerous benefits. Sometimes it is difficult to see the necessity of SLM if not
from a practical application. Therefore the objective of this discussion is to show through some of
our projects from the aeronautical and automotive industries how SLM is being used and how it
has improved our simulation processes.
During the development of airplane components with composite materials collaborative work
between design and simulation is needed. This is accomplished by the composites modeling tools
from Catia and Abaqus, which offer bi-directional data transfer links between design and
analysis. SLM guarantees an easy up-to-date management of these data, including geometry
revisions, laminate definitions or testing results among others, with total traceability.
We also present our success analyzing sports car exhaust systems, where different analyses such
as a heat transfer, a thermo-mechanical or a dynamic simulation are performed for every new
project. Scenario definition tools help us standardizing the methods, giving the engineer the steps
to follow. In this way it is easier to follow the state of a project and accessing to its data.
Keywords: Simulation Lifecycle Management, SLM, FEM, Composites…
1. Introduction
1.1. The need for Simulation Lifecycle Management (SLM)
Importance of simulation tools in the development of products has increased in the last years and
continuous growing is expected (CIMdata, 2011). A change from verification to design purposes
can be observed in the use of FEM or CFD analyses. Due to the improvement in computing
technology, nowadays more integrated tools exist that help in the product development phase,

whereas in earlier times simulations were mainly done with verification purposes. It is one of the
goals of PRIME aerostructures GmbH (PRIME hereafter) to serve as reliable partner in those
design and development activities, delivering analysis and simulation solutions, as well as in
certification tasks.
A problem of the role of simulation in the product development, where virtual prototyping has
gained great importance, is the management of the data. Large amounts of data are created during
the simulations, including geometry or specifications needed to perform the analyses, calculation
files themselves, results or reports, etc.
In Figure 1 can be seen an example that depicts it. For the analysis of a sportscar exhaust system
several inputs are needed, as material or geometry data. In order to study the mechanical behavior
on the car several calculations need to be performed. Among them there is a G-Shock static
analysis, a thermo-mechanical analysis, or a dynamic analysis, all requiring its FEM model. For
the definition of the loads another inputs would be necessary, as for example vibrational data from
the motor. At the right side of Figure 1 some possible result outputs are shown.
Figure 1 – Sketch of the multiple data sources in the analysis of a sportscar exhaust system.
This input/output data example can be regarded to a single case. If new revisions are involved
during the design process, the amount of files to control scales and their management becomes a
complex task. In addition, when different teams work together, communication problems occur
frequently. Therefore, the management of all that intellectual property becomes a crucial point.
If typical ways for data storage are used, as was the case of PRIME before implementing SLM,
following problems are implied (CIMdata, 2011):
− loss of data,
− use of non-standard methods,

− communication problems.
Companies identify insufficient focus on knowledge capture, being data sharing and reuse major
constraints on the value available from simulation and analysis. Data needed by collateral and
downstream project functions is too often unavailable, outdated, or captive to error-prone manual
methods of dissemination and re-entry.
Figure 2 – Loss of data and communication problems occur frequently in absence of a management
system.
If we analyze the efficiency of simulation and analysis activities, it is dependent on four factors:
simulation technology, people, integration in the process, and data management. Regarding the
first two points, companies are aware of the importance of using the latest simulation technology,
both software and hardware, and also of working with the analysis engineers. These factors can be
considered straightforward to solve.
However, a good integration of the simulation activities in a company is not so simple. By that can
be understood the function of the simulation activities with verification or development purposes,
and how the simulation team works and communicate with the rest of the company. For PRIME,
mainly focused on the simulation side, it means to have an efficient and reliable communication
with the clients with no problems or misunderstandings in sharing data.
The last efficiency factor is the data management itself. Finding the right information in less time
is a challenge given the large amount of data, but still worse problems can appear with an
inadequate data management. Cases of simulations done with wrong design revisions or using
wrong input data are known by anyone, being the cost of such “mistakes” critical.

Efficiency of
Simulations
Simulation Data
Technology management
Integration in
People the process
Figure 3 - Efficiency factors of simulation.
1.2. Hierarchical structure

Systems like PDM or SLM have been massively used since its integration several decades ago,
and they have become necessary for an efficient product development. But in the case of
simulation environments similar systems are only partially used. Many companies still do not rely
on those technologies for the management of CAE (Computer Aided Engineering) tasks.
One of the reasons of this lack of implementation is given by the higher complexity of simulation
processes. Contrary to CAD data that is organized according to a hierarchical physical assembly,
simulation data is more complex and have different sources as shown before. The structure of a
management system should not be based on physical hierarchy but on the simulation process,
creating a product structure from a CAE point of view.
1.3. Case of small enterprises

When speaking about SLM the picture of a large company arises: innumerable employees working
in team from different departments on the same enterprise and accessing to the same data and
information. That is the focus that Dassault Systèmes seems to be targeting, as in so a large
environment a management system is essential.
But what is the case of small enterprises? Even with a small number of users, as in the case of
PRIME, the same problems exist. Likewise the amount of data and its management its critical.
Besides, the risk factor in simulation engagements of small/medium companies is higher since the
loss of IP or time are critical factors that put the profitability of projects in danger. It cannot be
afford to not have the data under control and spend too much time looking for information or
understanding where the things come from.
As this was the case of our company, the chosen solution was the implementation of SLM for the
simulation environment management. In this manuscript we report on our experience in the use of
SLM as solution for the above commented problems. Advantages and disadvantages of SLM are
depicted with practical examples from simulations done in PRIME.

2. Implementation of Dassault Systèmes SLM Solution
SLM solution from Simulia has the traceability and possibilities of a typical data management
system together with special features for simulation works. In PRIME the SLM version
V6R2013X was implemented, which is based on ENOVIA.
ENOVIA is the system already established in other Dassault Systèmes brands. It serves as base for
capturing, managing, and securing the intellectual property. Figure 4 shows an example of how the
data is accessed. Revisions of the documents are secured, and information as who made a change,
when, or why, is available. Other possibilities can be seen at the upper side of Figure 4 that are
common to each document. For example, under “Lifecycle” the state of a document can be
changed or followed. In “History” a list of changes is presented. Or it is possible to create a
“Route”, which essentially is a series of tasks that can be assigned to different persons. These are
some of the capabilities.
Figure 4 – Example of document view in ENOVIA: Data management capabilities as revision control,
traceability, or data security are assured.
The benefit of SLM for simulation activities reside in its specialized tools:
− Live Simulation Review: serves as navigation tool for search and access to simulation
data in context of 3D design data. It is an extension of 3DLive, allowing the users to
access to 3D representation of simulation data.
− Model Editor: is an environment in which the preferred modeling tools are available to
directly work with, even with 3rd party simulation models and representations. It enables
a more cohesive editor for collaborative work.
− Scenario definition: Processes can be defined as templates and simulations with different
steps can be independently run, creating the so called scenarios. In addition, it creates a
representative view for organizing the data involved in the analyses. Automated tasks can
also be defined and controlled.
A certain number of companies are already are using SLM, but compared with the amount of CAE
users, its impact is relatively small. Some examples of companies that are already using SLM are
DANA (DANA, 2013), Borg Warner Turbo Systems, or Kimberly-Clark Corp (Pieper, 2011),
among others.

Figure 5 – View of Live Simulation Review used in DANA, extracted from (DANA, 2013).
3. SLM success in PRIME

The main goal of implementing SLM in PRIME was an improvement in the management of
analysis and simulation data. To have all data under control and to know where the things come
from, while assuring a quick access to it, were some of the challenges to be addressed. For that
purpose SLM seemed to be appropriate. Especially of importance is the Scenario Definition
module, which has been the principal tool for our purpose.
One of the situations most subjected to possible errors is when plenty of revisions are made during
the development of a product. Misunderstandings can easily happen and wrong data used, or false
results can be supplied. This was exactly the case of the design of a lightweight arm for an
automated robot of the company ENGEL, developed together with PRIME (see Figure 6).
In a first sight the arm does not look much complex. However, the robot operates a high speeds
and the movements need to be very precise. Thus, the technical requirements were a very light and
stiff component at the same time. Several proposals were studied and analyzed, ending up with a
composite/aluminum design as the best solution.

Figure 6 – Composite arm in ENGEL automated robot. Left: 3D image of the robot, courtesy of
ENGEL Gmbh. Right: CAD and FEM views of the composite arm.
For that project the organization of data was done with SLM using the Scenario Definition
Features. In Scenario Definition a “Process” can be defined, which in this case corresponds to the
project itself. The “Process” shows a is made as a structured tree in order to organize information
and analysis tasks. In Figure 7 the structure is shown using the graphical view, called “Impact
Graph”. It can be seen how a tree composed of different categories with links between them
creates an input/output hierarchical structure. The inputs of the process box, in dark blue, are
called “Activities”. In the ENGEL project they represent the design revision, but the system leaves
the user to choose what they specifically mean. An “Activity” can represent a single task or an
automated task as well. Here it was used to represent a revision cycle, each cycle gathering its
CAD data and analysis results.

A total of eight revisions were performed, as shown in Figure 7. For each revision (Activity) the
inputs can be expanded and the information followed, being organized under different
predetermined categories as Product, Specifications, or Internal Data, among others. An important
point is that the CAD geometry under Product is based on a link. It is not a copy, but a connection
to the original assembly made by the designer. In this way it is always know from which geometry
a simulation is, even from which revision. At the right side of the outputs can be seen, as for
example the FEM solutions or the reports.
There are situations in which a person not directly working on the project would need to look for
some information in it. An example would be a manager that wants to see the final results. Or it
could be a new analyst who is going to work in future developments. In those cases the graphic
representation from Figure 6 is an easy and understandable way to access to the needed data.
Figure 7 – Impact Graph of ENGEL lightweight arm simulation process.
Figure 8 – Impact Graph of ENGEL lightweight arm simulation process (detail).

The strategy of using Scenario Definition has been followed for other projects in PRIME. The
analysis of sportscar exhaust systems represents one of them. Since several simulations need to be
performed for each different exhaust in a more or less similar way, it was an adequate case for
creating a template. For each project similar simulations with the same type of inputs/outputs
exists, as sketched in Figure 2. The graphical representation of the template (Figure 9) depicts how
the single simulations are defined as “Activities”, and how the inputs/outputs are organized in
them. The heat transfer the inputs are expanded, showing the two categories used in it: Product
and Specifications. Under Specifications information like the material database or testing results
are to be found/linked.
Similar information is to be found in the other three single simulations: G-Shock static analysis,
Thermo-Mechanical analysis, and Dynamic analysis. Again, necessary information to perform a
simulation or used to evaluate results is available and organized. The whole structure is then saved
in form of a template that will be used for each new exhaust to be analyzed, having the same
structure. We have noted that in this way the knowledge is no longer so heavily dependent on the
person who initially performed the analyses, as the use of templates makes easier to follow the
same methodology and to separate tasks among the team.
Figure 9 – Impact Graph view of an exhaust system analysis.
One option that in PRIME has considered important is the types of documents for the storage of
revisions in SLM. It is desired to have revisions of the data, but when the files are large, as the
FEM results can be, storage capacity can be a problem. SLM gives the possibility of choosing if
every revision is saved or only the latest, which is suitable for very large data.

Figure 10 – Location of FEM database in Scenario Definition structure.
Above the graphical view has been shown, but in order to have a better understanding of how
SLM works, some explanation is necessary: The highest unit of organization in the system is in
“Workspaces”. In our company they represent projects, although it could be differently arranged.
Once the workspaces exist, users can be assigned to have access to them and the rights they have
can be selected. In them folders and documents are created, as well as processes or activities of the
Scenario Definition tool. Each of them needs to be understood as a metadata object which has
many other possibilities and properties than a typical document.
Figure 11 shows some options that a Process document type has. The Impact Graph would give as
similar views as in Figure 7 or Figure 9. Other options like the creation of a task assignment, the
management of a lifecycle, or looking for a revision history, are easy to access.
Figure 11 – Properties of a Process in SLM.

Coming back to the exhaust project, Figure 12 shows a situation where a new analysis is done and
a Route is created. That project was created using a template but an extra analysis was done. The
steps necessary to perform that extra analysis were controlled by a Route, which is very
straightforward to define.
Figure 12 – Creation of Route (set of tasks) related to an Activity.
A key option of Scenario Definition is the creation of automated tasks. Figure 13 depicts a case
where an automated task is used to extract plots directly from the FEM results. The Activity uses a
“connector” in order to call an application, which executes the given script. Inputs and outputs of
that script lie on the system, and the task is performed with automatically. This particular script
calls Abaqus/Viewer and uses the FEM results from the dynamic analysis to create the desired
graphs. Once the graphs are created, they are automatically saved in the system. In case new
results are calculated, with a single click the updated plots can be created again.

Figure 13 – Automated task for extraction result plots using Abaqus/Viewer.
4. Conclusion and outlook

It has been some more than a year since SLM in its version V6R2013X was adapted in PRIME.
After this time a clear beneficial effect of its implementation has been observed. Especially the
simulation data, including inputs of results of the analyses, are secured and located. In contrast,
some negative points related with the degree of maturity of the software can be found.
Projects and simulations managed with SLM are no longer so dependent on the persons who work
on them. The reason is a much more structured organization, having located all information
involved in the calculations. After a period of adaptation, working in the system has been efficient.
Less time is used looking for data and confusions about design revisions have disappeared.
5. References
EBARA (2014). Hidenobu Okamoto, Hiroyoshi Watanabe, and Michiko Sugiyama. On-going
Project on Simulation Lifecycle Management at EbaraProceedings of 2014 SIMULIA Community
Conference.
DANA (2013). Dana’s Engineers Team Up to Accelerate Product Design with Simulation
Lifecycle Management. SIMULIA Community News, November 2013.
CIMdata. (2011). Simulation Lifecycle Management "More than data management for
simulation". http://www.3ds.com/products-services/simulia/products/simulation-lifecycle-
management/.
Pieper (2011). Chris Pieper. Simulation Lifecycle Management Solves the Hard Challenges of
“Soft” Products. SIMULIA Realistic Simulation News, January/February 2011.

Modelling of Heat Transfer across Bolted Joints in
Abaqus/CAE
Volkan Coskun, Emir Ozkokdemir, Bulent Acar
ROKETSAN Missile Industries
Abstract: Bolted joints are frequently used connections in many engineering systems, and crucial
parts of the heat transfer path in structures. During the design phase, heat transfer across the
joints must be modelled properly in order to determine the accurate temperature distribution on
components. Usually in finite element models, perfect thermal contact between mating surfaces is
assumed. However; this approach is inadequate and may bring on misleading results owing to the
equality of temperature assumption between surfaces. Complex configuration of the joints and
non-uniform contact pressure distribution due to bolt preload make the prediction of transient
heat transfer characteristics across the joints arduous. In this study, transient conduction heat
transfer through the bolted joints was modelled in Abaqus/CAE by defining gap conductance
between the abutting surfaces. Two correlations were utilized to estimate the thermal contact
resistance at joints. Impact of bolt preload and resulting pressure distribution on heat transfer and
temperature distribution were investigated. Experiments were performed to verify the analyses
results by conducting various temperature measurements. Finally, coupled temperature
displacement analysis of a missile guidance section under aerodynamic heating was conducted
and results were compared with the results of an uncoupled heat transfer analysis in which perfect
thermal contact was assumed at all interfaces.
Keywords: Bolted Joints, Gap Conductance, Thermal Contact Resistance
1. Introduction
Customer requirements lead to more complex missile systems and using components susceptible
to temperature. One of the most challenging duties to ensure mission success of missiles is to keep
some components such as avionics within their narrow operating temperature range during flight.
To determine whether employing thermal management systems is necessary or not, temperature
distribution on the components shall be accurately estimated. Accurate estimation of the
temperature distribution also enables us to decide which and to where thermal protection systems,
e.g. ablative shields and phase change materials, shall be implemented.
Major thermal load a missile is subjected to during its flight at supersonic and hypersonic regimes
is the aerodynamic heating which can be briefly described as conversion of the kinetic energy of
air into thermal energy due to viscosity (see Figure 1). Aerodynamic heating brings on very high
heat flux rates q”, up to several millions W/m2 which makes thermal contact resistance’s effect at
interfaces significant. The higher the heat flux is, the higher temperature drop across an interface
takes place due to the following relationship,
𝑞 " = 𝑘𝑔𝑎𝑝 (𝑇1 − 𝑇2 ) = 𝑘𝑔𝑎𝑝 Δ𝑇
where T1 and T2 denote the temperatures of the mating surfaces and kgap is the thermal contact
conductance which is also called as gap conductance (see Figure 2).

Figure 1. Aerodynamic heating during flight (courtesy of Roketsan).
Figure 2. Temperature drop across a joint, (Gilmore, 2002).

In this study, heat transfer through bolted joints was modelled in Abaqus\CAE by defining gap
conductance between mating surfaces. Two models were utilized to estimate the thermal contact
conductance at joints. Impact of bolt preload and resulting pressure distribution on heat transfer
and temperature distribution were investigated. Experiments were performed to verify the analyses
results by conducting various temperature measurements. Finally, coupled temperature
displacement analysis of a missile guidance section under aerodynamic heating was conducted and
results were compared with the results of uncoupled heat transfer analysis in which perfect thermal
contact was assumed at all interfaces.
2. Heat transfer in bolted joints

Bolted joints are crucial parts of the heat transfer path in missile structure. In general, perfect
thermal contact between mating surfaces is assumed. However, this approach may bring on

misleading results especially when heat fluxes are very high. Complex configuration of the joints
and non-uniform pressure distribution due to bolt preload make the prediction of heat transfer
characteristics across the joints arduous.
Figure 3 depicts a typical bolted joint construction (a), pressure distribution at the interface (b),
and heat flow through the joint (c). Pressure is at its peak near the bolt hole and assumed to be
zero outside of the pressure cone. Since gap conductance is a function of pressure, heat flow lines
constrict in the region where the pressure is high.
Figure 3. Heat transfer in a bolted joint ((a) a bolted joint, (b) pressure distribution
at the interface, (c) heat flow), (Madhusudana, 2014).
There exist several correlations for thermal contact conductance in the literature. They can be
broadly classified as ones with elastic deformation assumption and ones with plastic deformation
assumption (Hasselström, 2012). They relate gap conductance to the interface pressure, surface
quality parameters and material properties.
3. Test sample and FE model

Test sample used in experiments is similar to a pipe flange construction (see Figure 4). While one
end of the sample is closed, the other end is open. The test sample was heated at the closed side for
a certain time with a hot air gun whose heating power can be adjusted up to several kW, while all
other surfaces were insulated. Temperature distribution on the test sample was measured with K
type thermocouples. A series of thermocouples placed on the sample were aligned with the bolt
centerline, and another series of thermocouples were aligned with an artificial line lying along the
mid of the two neighboring bolts.

Figure 4. Test sample.
Before tests were conducted, the arithmetic average roughness and flatness of the mating surfaces
of the flange were measured. Flatness of the surfaces were 0.013 and 0.005 mm. The RMS
roughness of the surfaces were calculated as 0.65 and 0.78 μm by assuming a Gaussian
distribution of asperities.
Tests were carried out at three different torque levels, i.e. Γ, 2Γ, and 4Γ Nm. After each test, the
sample was let cool down for a time to reach ambient temperature and have uniform temperature
distribution. Afterwards, torque level was adjusted with a torque meter and a new test was started.
After tests were completed, FE model of the test sample was created in Abaqus/CAE. Since
pressure is zero at interfaces in uncoupled heat transfer analysis, coupled temperature-
displacement analysis must be conducted to model the pressure dependent thermal contact
conductance. Therefore, linear solid coupled temperature-displacement elements (C3D8T) were
used to build the model. Since the model is symmetric, only 1/12 of the test sample was built (see
Figure 5). Surface-to-surface contacts were defined for all mating surfaces except tie contact was
used for the interface between the nut and bolt.

heat flux
hot side of
flange
bolt force
symmetry
restraint
cold side
of flange
Figure 5. FE model of the test sample.

Two correlations taken from the literature were utilized to define the thermal contact conductance
between the mating surfaces. The first correlation, i.e. Mikic’s correlation (Gilmore, 2002), is
based on the elastic deformation assumption, and the second one, i.e. Tien’s correlation
(Hasselström, 2012), is based on the plastic deformation assumption. GAPCON subroutine was
used to define the gap conductance in Abaqus/Standard, because GAPCON offers flexibility in
specifying dependencies of gap conductance on the interface pressure, surface quality parameters
and temperature dependent material properties.
4. Results
Three different torque levels and two different thermal contact conductance correlations were used
in the simulations. In total, six simulations were conducted and results were compared with the
test data. Temperature drop across the two different sections of the flange is chosen for
comparison since temperature drop is the major variable related to the gap conductance and
interface pressure (see Figure 6). Figure 7, 8 and 9 show the temperature drop across the flange for
different torque levels of Γ, 2Γ and 4Γ Nm, respectively.

point 1 point 3
point 4
point 2
Figure 6. Locations where temperature drops were compared.
Figure 7. Temperature drop across the flange (Γ Nm) ((a) between point 1 and point
2, (b) between point 3 and point 4).
Figure 8. Temperature drop across the flange (2Γ Nm) ((a) between point 1 and
point 2, (b) between point 3 and point 4).

Figure 9. Temperature drop across the flange (4Γ Nm) ((a) between point 1 and
As seen from the figures, temperature drop between the points 1 and 2 is lower than the
temperature drop between the points 3 and 4. This is due to the higher interface pressure and gap
conductance in the proximity of the bolt hole. Figure 10 shows the temperature distribution and
contact pressure at the flange interface of the cold side when the bolt torque is 4Γ Nm and time is
Θ. Higher temperature and contact pressure values are observed near the bolt hole.
o
Figure 10. Temperature (in C) and pressure distribution (in MPa) at the interface.
Fully coupled heat transfer analysis with Abaqus/Standard results in acceptable results compared
to the test data especially when time is less than 0.8Θ. The Mikic’s correlation yields more
satisfactory results than the Tien’s correlation when torque level is lower than 4Γ Nm. Difference
between the two correlations becomes negligible as torque level is increased.
Figure 11 demonstrates the effect of bolt torque on temperature drop across the flange. As torque
level is increased, temperature drop across the flange reduces. It is an expected result since the gap
conductance is almost proportional to the interface pressure which is determined by the bolt
torque.

Figure 11. Effect of the bolt torque on temperature drop ((a) between point 1 and
5. Aeroheating analysis of a missile guidance section

As stated above, temperature of the avionics shall not exceed the maximum operating temperature
during the flight. Therefore, in order to decide the necessity of implementing thermal protection,
accurate thermal analysis shall be made. Figure 12 shows the quarter model of a missile guidance
section built in Abaqus/CAE. The model consists of two casings fastened with bolts, plate and
avionics. Bolted joint connections are also used to fasten the plate to the casing2 and the avionics
to the plate (which is not shown in the figure).
bolts casing2
casing1
bolt
avionics
plate
Figure 12. Quarter model of the guidance section.

Aerodynamic heating environment the missile is subjected to was determined via CFD analysis in
the form of recovery temperature and heat transfer coefficient versus time. This thermal load was
defined in Abaqus/Standard by introducing surface film condition on the outer surfaces of the
casings. All other surfaces were assumed to be adiabatic.

Linear solid coupled temperature-displacement elements (C3D8T) were used to build the model.
Surface to surface contacts were defined between all mating surfaces except tie contact was used
between the threaded regions of the bolts and holes. GAPCON subroutine was utilized to describe
the thermal contact conductance according to Mikic’s correlation, because Mikic’s correlation
yields more satisfactory results at lower torque levels and the differences between the correlations
becomes negligible at higher torque levels. So it covers broader range of torque levels.
Figure 13 shows the temperature distribution on the guidance section at the end of the flight. Heat
flows to the avionics through the bolted joint connection. Therefore, heat transfer in the bolted
joints and the thermal contact conductance have a direct impact on the temperature of the avionics.
o
Figure 13. Temperature distribution (in C) on the guidance section.
The same aeroheating analysis was also conducted with heat transfer elements (DC3D8) after
removing bolt loads and converting all surface to surface contacts to the tie contacts. This is the
case in which perfect thermal contact is assumed between all mating surfaces. Figure 14 and
Figure 15 compare the results obtained from the fully coupled temperature-displacement analysis
and uncoupled heat transfer analysis.
o
Figure 14. Temperature distribution (in C) on the plate and avionics ((a) coupled
temperature-displacement analysis, (b) uncoupled heat transfer analysis).

error ~8 oC
error ~5 oC
Figure 15. Volume average temperature variation of the plate and avionics.
Figure 14 and Figure 15 show the temperature distributions on the plate and avionics, and volume
average temperature variation of the plate and avionics, respectively. As seen clearly, there exists
a significant difference between the results of the analyses when taking into account the narrow
temperature margin of the avionics. Perfect thermal contact assumption results in higher
temperature estimation on the avionics which may lead to the unnecessary use of thermal
protection.
6. Conclusion
This paper presents the modelling of heat transfer in bolted joints in Abaqus/CAE by defining
pressure dependent thermal contact conductance between mating surfaces in a fully coupled heat
transfer analysis. Several tests were conducted by taking numerous temperature measurements on
the test sample and measurements were compared with the results of the fully coupled heat
transfer analysis. Fully coupled temperature displacement analysis in Abaqus/Standard gave
satisfactory results compared to the test data which emphasizes the flexibility and trustworthiness
of Abaqus. Furthermore, fully coupled and uncoupled heat transfer analyses of a missile guidance
section were conducted. Results indicate that there is a significant difference between the
estimated results, as perfect thermal contact assumption overestimates the temperature on the
components.
7. References
1. Gilmore, D.G., “Spacecraft Thermal Control Handbook Volume I: Fundamental
Technologies,” The Aerospace Press, El Segundo, California, 2002.
2. Hasselström, A. J., and Nilsson, U. E., “Thermal Contact Conductance in Bolted Joints,”
Diploma Work, Chalmers University of Technology, Sweden, 2012.
3. Madhusudana, C. V., “Thermal Contact Conductance,” Springer International Publishing,
Switzerland, 2014.
4. Abaqus Users Manuel, Version 6.13-1, Dassault Systemes Simulia Corp., Providence, RI.

Modeling of forming process of composite
materials based on thermoplastic matrix
Fedulov B.N., Kantor M.M., Safonov A.A. and Ushakov A.E.
Skolkovo Institute of Science and Technology
Abstract: Composite materials based on thermoplastic matrix became a popular choice as a

material for modern structures. Nevertheless, the manufacturing process of this type of materials
have many technology parameters, which have to be determined before the first composite part is
produced. The study of influence of all parameters on final quality of composite part by means of
technological experiments are time and cost consuming. This makes engineers to study the way of
the modeling of thermoplastic composite forming process. This research is dedicated to the
modeling of thermoplastic material under different conditions. Several approaches to capture
specific for this material features are performed. The method to model the influence of
crystallinity on mechanical properties of composite material and on final residual stresses is
analyzed. An approach to model shear nonlinearity in composite prepregs is performed. The
analysis of defects initiation in the thermoplastic composites under technological temperature
cycle is also performed. All numerical procedures and special subroutines based on Abaqus
software are presented. Eventually a complete set of engineering tools using Abaqus software
needed to model the forming process of thermoplastic composite details is realized.
Keywords: Composite material, Thermoplastic composite material, Thermoplastic material

forming, Thermoplastic composite residual stress, Technology process thermoplastic material,
Failure due to technology residual stress. Nonlinear shear behavior of thermoplastic composites.
1. Introduction
Thermoplastic composites took a significant part in the market of new materials. This type of
materials gives essential and cost effective results in automotive and aerospace industries. Fast
fabrication, easy storing and capability of welding make them attractive as a choice for the modern
constructions. Composites with thermoplastic matrix have special characteristics based on their
thermoplastic features. The main feature of this kind of material is the presence of crystallinity in
its internal structure. The degree of crystallinity determines significantly the mechanical properties
of the material. Low ratio of crystallinity leads to the low stiffness of the material, and on the
contrary, maximum value of the crystallinity ratio gives highest values of the material stiffness.
Another special aspect is additional volumetric shrinkage due to increase of the ratio of
crystallinity. This volumetric shrinkage caused by chemical material phase transition can be the
reason of the essential residual stress, which cannot be ignored. Manufacturing experiments with
variation of all technological parameters, in the way to get good quality product, takes a lot of time
and eventually are expensive. This fact makes all attempts of modeling a complete manufacturing
cycle of the product based on thermoplastic matrix essentially important from the view of the
practice.

2. Forming process
The first modeling step of the thermoplastic composite product manufacturing is its forming. In
order to analyze models and validate the technology process it was proposed a special double-
dome surface represented in work [1] (Figure 1):
Figure 1: Double-dome surfaces: (a) female die, (b) male punch. [2]
The corresponded virtual model based on Abaqus software was developed (Figure 2):
Figure 2: Double dome surfaces and prepreg Finite Element Model

Thermoplastic preform during forming process has about 200°C-400°C temperature and
demonstrates weak properties in the direction with no reinforced fibers. The idea realized in this
work is that usage of material model based on standard Abaqus lamina properties with low shear
modulus. After few attempts of modeling and analyzing experimental data from picture frame tests
(Figure 3) the ideal plasticity in case of shear loading was added.

Figure 3: Different specimens shear test curves [2]
Large majority of researchers uses explicit solvers in order to perform forming simulation and this
is justified on grounds of equilibrium and contact convergence problems. However, explicit solver
makes everybody uses some special technics to reduce calculation time especially in case of the
usage of stiffness dependent material damping. Our research is successful attempt of implicit
solver application for thermoplastic forming problems. We face with contact convergence
problems when the default contact stabilization was used. On the one hand, the default contact
stabilization implies the reduction of stabilization from the beginning of the step to the end of the
step due to reduction factor, such behavior causes the contact convergence problems at the end of
the step, on the other hand, it implies that the magnitude of stabilization is additionally multiplied
by the fraction of the step remaining, correspondingly such behavior cause problems in the start
increments. We recommend using the reduction factor for contact stabilization equals 1 and
constant amplitude through the step. These settings will ensure constant contact stabilization
through the modeling step. On the Figure 4 one can see the contact stabilization settings used in
the research. Figure 5 shows the final result of the thermoplastic forming. Shear angles in radians
are presented in the contour plot.
Figure 4: Contact stabilization settings used in the research

Figure 5: Finale stage of the thermoplastic forming, shear angles, radians
More over our forming modeling analysis shows that the usage of plasticity model presented in the
work [3] has a real perspective. Figure 6 shows that during the thermoplastic forming one can find
that there are prepreg areas subjected to biaxial compression and biaxial tension. Stress state
parameter triaxiality is shown on the Figure 6 contour plot.
Figure 6: Stress triaxiality

Plasticity criteria from [3]:
�𝐴𝑖𝑗𝑘𝑙 (𝜉)𝜎𝑖𝑗 𝜎𝑘𝑙 = 𝐶𝑜𝑛𝑠𝑡, (1)
where = 𝜎/𝜎0 , 𝜎 = 𝜎𝑖𝑖 /3, 𝜎0 = �3/2𝑆𝑖𝑗 𝑆𝑖𝑗 , 𝑆𝑖𝑗 = 𝜎𝑖𝑗 − 𝜎𝛿𝑖𝑗 .
Parameter ξ helps to identify the type of stress and chose particular constants 𝐴𝑖𝑗𝑘𝑙 for example in
cases of tension, compression or pure shear. Plasticity flow with criterion (1) can be realized on
the base of Abaqus USDFLD subroutine. Strain rate dependency can be added by means of
standard Abaqus tools into the yield condition also.

3. Crystallinity model
Crystallinity ratio is a temperature history depended value, thus it has integral of temperature
value in its definition. The most popular approach to obtain value of the crystallinity is described
in [4]. There are two mechanisms to grow crystals: first one is the nucleation and second is the
growing of present ones. The total value of crystallinity is the sum of results of these two
mechanisms with special weights:
𝑋𝑣𝑐 = 𝑋𝑣𝑐∞ (𝑤1 𝐹𝑣𝑐1 + 𝑤2 𝐹𝑣𝑐2 )
where
𝑋𝑣𝑐 – Degree of crystallization
𝑋𝑣𝑐∞ - Equilibrium degree of crystallinity
𝑤1 - Weight factor for first mechanism (grow of crystals)
𝑤2 - Weight factor for second mechanism (nucleation of crystals)
𝑤1 + 𝑤2 =1
Each mechanism contribution is summarizing by next equation:
𝑡
𝐶2i 𝐶3i
𝐹𝑣𝑐i = 1 − 𝑒𝑥𝑝 �−𝐶1i � 𝑇𝑒𝑥𝑝 �− � + �� 𝑛i 𝑡 𝑛i−1 𝑑𝑡�
�𝑇 − 𝑇𝑔 + 51.6� 𝑇(𝑇𝑚i − 𝑇)2
0
where
𝐶11 , 𝐶21 , 𝐶31 , 𝐶12 , 𝐶22 , 𝐶32 , − Experimental constants
𝑇𝑔 - Glass transition temperature
Tm1, Tm2 – Temperatures for melting of crystals
n1, n2 – Avrami constants for corresponding mechanisms
The calculation of crystallinity parameter 𝑋𝑣𝑐 was realized by special subroutine and used in
UMAT program to influence on mechanical properties of thermoplastic composite.

For the case of 35°C cool rate and for 40 plies specimen, based on PEEK matrix, typical
distribution of crystallinity shown on Figure 7.
0.3
0.2
0.1
0
0 0.001 0.002 0.003 0.004 0.005 0.006
Figure 7: crystallinity distribution through the thickness of 40 plies specimen
4. Residual stress
In order to approximate effective properties of composite material and calculate additional
volumetric shrinkage there is an idea to use crystallinity distribution at the next step. The
reasonable approach is to use Chamis [5] and Bogetti [6] micromechanics equations in
combination with work [4], where average property of thermoplastic matrix (PEEK) are obtained.
Transversal residual stresses values through the thickness are shown on Figure 8 for the 40 ply
unidirectional specimen. One can see that taking into account crystallinity influence gives a result,
which is more correct.
Figure 8: Residual transversal stress distribution in 40 ply unidirectional specimen

5. Thermoplastic matrix failure
We can estimate failure of thermoplastic product during cool down process by application of the
obtained residual stress to the meso-model, (Figure 9).
Figure 9 Applying forces to the meso-model

Briefly, a complete algorithm of failure estimation is shown on Figure 10.
Material model used for matrix failure prediction is taken from the work [8] and based on
Drucker-Prager plasticity [9]. Failure accumulation is based on work [10] and has integral form
with capability to take into account the history of loading of the material:
𝑡
𝑑𝜀 𝑝𝑙
𝜔𝐷 = � 𝑝𝑙
0 𝜀𝐷 (𝜉, 𝜀̇)
𝜔𝐷 = 1 –matrix failure
Obtain effective properties for temperature

analysis
(Specific heat, conductivity, ... )
temperature
analysis
Volume deformation of matrix material due to

crystallization
Crystallization
Effective properties for composite based on

Residual stress
Temperature and Crystallization results
Defect nucleation
Figure 10: Algorithm scheme for matrix failure estimation

Figure 11 shows the fringe of parameter 𝜔𝐷 at the initiation of failure, in case of plane strain
transversal tension condition of meso-model.
Figure 11: Distribution of parameter 𝜔𝐷 at initiation of failure

Figure 12 shows the areas of active yielding during loading.
Figure 12: Active yielding areas
6. Conclusions
The approach to estimate manufacturing failure of thermoplastic composite products is presented.
It was shown that in the case of thermoplastic matrix the ratio of crystallinity plays important role
in predicting of the residual stress. It is possible to conclude that Abaqus software has all
necessary tools to realize a complete cycle of design development of thermoplastic products, from
forming process to meso-level failure prediction.

7. Acknowledgments
This work was supported by RFBR grant 14-01-31322 and the part of this work was carried out in
the Perm National Research Polytechnic University with support of the Government of Russian
Federation (The decree № 220 on April 9, 2010) under the Contract № 14.В25.310006, on June
24, 2013.
8. References
1. Composites Benchmark Forum; 2008. http://www.wovencomposites.org/index.php [accessed
07.12.12].
2. Harrison, P., Gomes, P., Correia, R., Abdiwi, F., and Yu, W.R. (2012) Press forming the
double-dome benchmark geometry using a 0/90 uniaxial cross-ply advanced thermoplastic
composite. In: ECCM 15 - 15th European Conference on Composite Materials, Venice, Italy,
24-28 Jun 2012.
3. Lomakin E.V., Melnikov A.M., Fedulov B.N. Constitutive models for anisotropic materials
susceptible to loading conditions // Mechanics and Model-Based Control of Advanced
Engineering Systems, Springer Verlag GmbH, Berlin, 2014
4. Crystallization kinetics of polyetheretherketone (peek) matrices Chris N. Velisaris and James
C. Seferis. Polymer Engineering & Science. Volume 26, Issue 22, pages 1574–1581,
December 1986 DOI: 10.1002/pen.760262208
5. Chamis C. C. Simplified composite micromechanics equations for hygral, thermal and
mechanical properties. 38th Ann. Conf. of the Society of the Plastics INdustry (SPI)
Reinforced Plastics/Composites Inst., Houston, Tex., 7-11 Feb. 1983
6. Travis A. Bogetti, John W. Gillespie Jr, Process-Induced Stress and Deformation in Thick-
Section Thermoset Composite Laminates. Journal of Composite Materials May 1992 vol. 26
no. 5 626-660.
7. Wesley E. Lawrence1, James C. Seferis1, and John W. Gillespie Jr.2. Material response of a
semicrystalline thermoplastic polymer and composite in relation to process cooling history.
Polymer Composites Volume 13, Issue 2, pages 86–96, February 1992
8. Lei Yang,Ying Yan, Jian Ma, Bo Liu. Effects of inter-fiber spacing and thermal residual
stress on transverse failure of fiber-reinforced polymer–matrix composites. Computational
Materials Science. Volume 68, February 2013, Pages 255–262
9. Abaqus 6.14 manual
10. Hooputra, H., H. Gese, H. Dell, and H. Werner, “A Comprehensive Failure Model for
Crashworthiness Simulation of Aluminium Extrusions,” International Journal of
Crashworthiness, vol. 9, no.5, pp. 449–464, 2004.

Finite Element Simulation of Thermal Barrier
Coatings in Rocket Engines
Torben Fiedler, Martin Bäker, Joachim Rösler
Institut für Werkstoffe, Langer Kamp 8, 38106 Braunschweig, Germany
Abstract: Rocket engines need to withstand extreme gas temperatures. To achieve this, the engine
is lined with copper and cooled with liquid hydrogen. Nevertheless, creep processes can lead to
damage and failure of the component. To avoid this, rocket engines can be protected with thermal
barrier coatings. Standard coating systems as used in gas turbines are problematic because of
larger thermal stresses between the copper substrate and the coating due to a large thermal
mismatch. In this work, we use finite element simulations to study the stress evolution in a coating
material tailored for application in a rocket engine. The influence of the thermal conductivity on
the resulting stress state is discussed and general conclusions for the design of thermal barrier
coatings in rocket engines are drawn.
Keywords: Coatings, Thermal Stress, Aerospace, Rocket Engine
1. Introduction
The combustion chamber in rocket engines is exposed to high thermal and thermomechanical
loads. The process gas can reach temperatures of 3200°C (Greuel, 2002). Since no industrial
material can withstand these temperatures, the rocket chamber is lined with copper and cooled on
the inside using liquid hydrogen in cooling channels with temperatures of approximately -240°C.
The surface temperature of the copper liner can reach temperatures of up to 600°C (Raj, 2007)
with large heat fluxes of more 100 MW/m² (Popp, 1996, and Quentmeyer, 1977). The copper liner
above the cooling channels has a thickness of 1mm so that the thermal gradient and the large
pressure difference between combustion chamber and cooling channel cause a high thermo-
mechanical load. This load and thermo-mechanical fatigue cause failing due to the so called dog-
house effect (Figure 1) where cooling channels deform visco-plastically, leading to fracture of the
copper liner after a few thermal cycles (Riccius, 2004).

Figure 1. Dog-house effect
In gas turbines, where the process gas is also hotter than the service temperature of the turbine
blade material, thermal barrier coatings are used to reduce the thermal load and thus the
mechanical stresses. These coatings comprise two layers: a so-called bond coat (usually a
NiCrAlY alloy) serves as oxidation protection and improves the adhesion of the top coat material
(yttria-stabilized zirconia) that serves as thermal protection (Bürgel, 2011).
A similar protection method might also be used in rocket engines, coating the copper liner with a
material with lower thermal conductivity and higher service temperature. However, directly
transferring the thermal barrier coating systems used in gas turbines to rocket engines is not
feasible (Schloesser, 2011) because the very low thermal conductivity of zirconia would lead to
extremely high surface temperatures. Furthermore, the coefficient of thermal expansion differs
widely between a standard NiCrAlY coating and the copper substrate (see also Figure 5 below), so
that high thermal stresses occur at the interface between the copper substrate and the coating.
Therefore, a new coating system needs to be developed that is suitable for the application in rocket
engines. Adding copper to a standard NiCrAlY alloy increases the coefficient of thermal
expansion and thus reduces the thermal stresses between substrate and coatings. The thermal
conductivity of NiCrAlY is already roughly one order of magnitude smaller than that of copper.
Furthermore, the service temperature of standard NiCrAlY in gas turbines is approximately
1100°C, much higher than the allowed temperatures in copper alloys. Therefore, a NiCuCrAl
coating can possibly serve both as thermal barrier coating and as bond coat.
To evaluate the feasibility of this coating system, it is important to understand the conditions
inside the coating and especially the thermal stresses at the interface and inside the coating.
Experimentally, this can be done using a laser testing bay described in section 2. A finite element
model is used to determine temperatures and stresses inside the coating and to estimate the
conditions inside a rocket engine, where component tests are costly to perform.
2. Experimental conditions
Due to the extreme conditions inside a rocket engine and to the high costs of full-scale tests, it is
not feasible to perform a large number of experiments using rocket engine experiments. To test
thermal barrier coatings under realistic conditions, a laser test bed was constructed, consisting of a
3 kW diode laser with a special optics to produce a broad focal point with 20 mm diameter. To
ensure optical coupling between the specimen and the laser light at a wave length of 808 nm, the
specimen can be coated using Fe3O4. In this way, the surface of plate shaped specimens can be

heated up to 1500°C in less than 0.8 s. Surface temperatures are measured using a two-color
pyrometer. To study cyclic loading of the coatings, the procedure is repeated up to 50 times.
Specimens can then be metallographically prepared to study the interface between coating and
substrate and to understand failure mechanisms. The experimental setup and results for standard
thermal barrier coating systems are described in detail in (Schloesser, 2011).
To understand the failure mechanisms of the coatings, it is necessary to know the conditions at the
interface during the cycling experiments. Stresses are generated due to the thermal mismatch of
the expansion coefficients of the materials and are exacerbated by the roughness of the interface.
Since the interface region is not accessible for measurements during the experiments, finite
element simulations are used to calculate temperatures and stresses at the interface.
To transfer the results to the condition inside a rocket engine, a fully coupled fluid-structure model
of a rocket engine is used (Kowollik, 2011 and Kowollik, 2013). The calculation shows that a
steady state is reached after 0.3 s if a copper substrate with a 100 µm NiCrAlY coating is used.
This calculation can then be used to determine the heat transfer between the coating and the hot
gas and between the copper substrate and the cooling channel. For this, a simple one-dimensional
model is used (Bürgel, 2011). The calculated mean film coefficient on the hot-gas side is
αh=63.5 kWm-2K-1 , the cooling-surface film coefficient is estimated as 455.7 kWm-2K-1.
Using these numbers, the temperature of the coating’s surface and of the coating-substrate
interface can be estimated for different coating thickness as a function of the thermal conductivity
of the coating, see Figure 2.
Figure 2 Temperatures and heat fluxes calculated for different values of thermal
conductivity and different coating thickness d.
Although this is only an estimate because the film coefficient is assumed not to depend on the
coating, the plot shows that a coating with low thermal conductivity will lead to prohibitively high
surface temperatures, since even ceramic thermal barrier coatings cannot withstand surface
temperatures of more than 1200°C (Bürgel, 2011). If a coating thickness of 100 µm is assumed,

the thermal conductivity should not fall below ~20 W/mK to keep the surface temperature
sufficiently small. This leads to a reduction of the maximum substrate temperature of ~200 K.
3. Finite element model

3.1 Model Description
The two-dimensional geometry of the finite element model was created using the Python scripting
interface of Abaqus/CAE. An input deck was written and was directly modified for the variation
of boundary conditions and material properties. All calculations were performed using
Abaqus/Standard.
Figure 3. Two-dimensional model geometry
The two-dimensional model geometry is shown in Figure 3. The model consists of a copper
substrate with a height of 2 mm (laser testing) or 1 mm (rocket combustion chamber) and a
coating with a thickness of 113 µm. The interface between the materials is assumed to be
sinusoidal in shape to model the typical interface roughness due to a thermal spraying process
(Freborg, 1998). The model is assumed to represent a small strip of material taken out of a large
specimen. To ensure correct boundary conditions, the lower edge of the model is restricted in y-
direction, whereas the upper edge of the model is tied in normal direction to straight a rigid surface
that can only move in the y-direction. These boundary conditions allow for thermal expansion in
the y-direction with distorting the shape of the strip of material. In the z-direction, a generalized
plane strain condition is assumed.

Figure 4. Mesh at the interface substrate/coating
In the laser experiment, the copper specimen is placed on a nickel block that serves as a heat sink.
This heat sink was included in the same part. The Young’s modulus of the heat sink material was
set to a small value (1 MPa) and its coefficient of thermal expansion was set to that of copper so
that the thermal stresses in y- and z-direction generated by the heat sink material are negligible.
The specimen was meshed using CPEG4RHT elements. 22000 elements were used in the copper
substrate and the coating, using mesh refinement near the interface to resolve the stresses, see
Figure 4. The model including the heat sink contained an additional 130000 elements in the heat
sink and the extended copper-substrate. Linear elements were chosen to allow for a very fine mesh
at the interface with reasonable computational costs. Reduced integration helps to avoid problems
due to shear locking since elements at the sinusoidal interface experience some bending. For the
purely elastic simulation done here, hybrid elements ae not necessary; these elements were chosen
because in future simulations creep will be considered and large plastic deformations can be better
represented by hybrid elements.
3.2 Material properties
Since the laser testing times are extremely short, no creep of the materials was considered. A
purely thermo-elastic simulation was sufficient to estimate the stresses at the interface. Material
properties for the copper substrate are taken from (Kupferdatenblatt, 2005, and Dies, 1967, and
Fassin, 2015). Material properties of the NiCrAlY coatings are taken from (Pawlowski, 2008, and
Altun, 2008, and Taylor, 2004, and Rösler, 2004) . Since NiCuCrAl is a new material, the
properties have to be estimated. The CTE was measured at the bulk material (Fiedler, 2014), but it
is estimated that it will not differ much from the CTE of the NiCuCrAl coating, since the coatings
are dense (porosity < 1%). Figure 5 shows the values of CTE for the copper-substrate, the
NiCrAlY alloy and the NiCuCrAl alloy. It can be seen that adding 30% copper increases the CTE
and will thus reduce thermal stresses at the interface. It is also apparent that the largest thermal
stresses will occur at intermediate temperatures of about 900-1000 K.

Figure 5. Thermal expansion coefficients related to 273 K
For the heat transfer analysis and the determination of the surface-temperatures, the thermal
conductivity is of importance: The thermal conductivity of the Cu-substrate is approximately 320
W/mK. For NiCrAlY it is 12 W/mK at 800 K and 17 W/mK at 1300 K. The thermal conductivity
of the NiCuCrAl coating has not been measured so far. Therefore, a parametric study with
different thermal conductivities is carried out.
3.3 Thermal boundary conditions

For the simulation of the laser experiment, the left edge of the heat sink was set to a fixed
temperature of 293.14 K. The surface of the specimen was heated within 1 second to the final
temperature of 1287.84 K, using the temperature profile that was determined pyrometrically in the
laser experiments.

To determine the temperature profiles and stresses during exposure for longer times in rocket
applications, film coefficients and sink temperatures for heat transfer were provided on the
coating-surface and the downside of the copper-substrate based on the calculations in section 2.
The hot-gas temperature at the coating surface was set to 3502 K with a film coefficient of
63.5 kWm-2K-1. At the cooling site of the copper substrate, the sink temperature of the liquid
hydrogen was set to 40 K, using a film coefficient of 455.7 kWm-2K-1.
4. Results
4.1 Laser cycling experiments
During laser cycling, thermal stresses build up at the interface between the copper substrate and
the coating. Delamination on the micro-scale usually occurs by crack formation in a direction
parallel to the interface, so that the 11-component of the stress is most relevant. Figure 6 shows
Figure 6. Stress S11 in MPa at the interface substrate (left) and coating (right) for
the NiCrAlY coating at 1273 K

this stress component for the case of a standard NiCrAlY coating on copper. Since the CTE of
copper is larger than that of the coating, tensile stresses build up in the so-called “valley” region of
the interface roughness (upper side in Figure 6). These stresses may cause delamination and
failure. Experimentally, it was found that a NiCrAlY coating fails after several thermal cycles due
to delamination (Schloesser, 2011).
Figure 7. Simulation of the laser experiment with a NiCuCrAl coating: Temperature

(solid line) and Stress S11 (segmented line) in the valley of the interface
roughness-profile for different values of the thermal conductivity λ; the reference
thermal-conductivity of NiCrAlY is named λ0.
The thermal stresses are due to the CTE mismatch between copper and NiCrAlY. If copper is
added to the coating material, the CTE significantly increases, leading to a reduction of the
thermal stresses. Since the thermal conductivity of NiCuCrAl is not known, simulations with
different values of the thermal conductivity were performed, setting the conductivity to 150%,
100%, 50%, and 25% of the value in NiCrAlY.
Figure 7 shows the development of the stress and the temperature at the valley position of the
interface region for a NiCuCrAl coating. Although the temperature history depends on the thermal
conductivity, the maximum stresses are very close and occur at an interface temperature of about
950K. The reason for this is that the difference of the CTE between copper and NiCuCrAl
becomes smaller at higher temperatures, see Figure 5.
Figure 8 shows the stress history at the valley region for a NiCrAlY and a NiCuCrAl coating, both
with a thermal conductivity of NiCrAlY. It is apparent that the maximum stresses are strongly
reduced when copper is added to the coating material. If creep is considered, the reduced stresses
will also lead to a smaller amount of visco-plastic deformation so that residual stresses after
cooling to room temperature would be reduced as well. From this point of view, adding copper to
the coating material should serve to improve the lifetime of the coating.

Figure 8. Simulation of the laser experiment: Stress S11 for a NiCrAlY and a
NiCuCrAl coating
4.2 Simulation of a rocket engine

The laser testing experiments are temperature-controlled because the laser power is directly
controlled using the measured surface temperature. Inside a rocket engine, changing the coating
material’s thermal conductivity will also affect the temperature profile and the maximum
temperatures, see also Figure 2. The second model (see section 3.1) has been used to study the
effect of the thermal conductivity when the sink and gas temperature and the film coefficient are
fixed (see section 3.3).
Figure 9, left shows the surface temperature of the coating and the interface temperature for a
NiCuCrAl material with different heat conductivities. If the thermal conductivity is too small, the
surface temperature of the coating is far beyond the service temperature of nickel-based materials.
Even a large thermal conductivity still leads to rather high surface temperatures of 1400 K.
However, this does not imply that NiCuCrAl cannot be used as a coating system because the
thickness of the coating was kept fixed in the simulation. A coating thickness of 100 µm, as
assumed here, reduced the temperature of the interface region to values below 700 K, which is
below the values reached in the laser cycling experiments. Therefore, a thinner coating might be
used.

Figure 9. Temperature, maximum Stress S11 at the substrate/coating interface and
elastic strain energy in the coating for different values of the thermal conductivity
λ, the reference thermal-conductivity of NiCrAlY is named λ0.
Figure 9, right shows the maximum stress S11 at the interface region. As expected, the maximum
stress occurs for the highest thermal conductivity of the coating, but absolute values are still below
150 MPa and will probably not be sufficient to cause significant delamination.
However, failure of a coating system may also be affected by buckling of the coating due to large
compressive stresses in the in-plane directions. Buckling is driven by the stored elastic energy.
Since the stresses in 22- and 33-direction are considerably larger than the 11 stresses except
directly at the interface, the total stored energy inside the coating can be used as a rough indicator
of the tendency of the coating to buckle. (Note that this is only an estimate because the stresses in
22- and 33-direction are tensile near the interface and will thus not contribute to buckling, but may
instead cause segmentation cracks. Since these cracks might also reduce the lifetime and increase
the probability of delamination, a large value of the total stored elastic energy can be considered
unfavorable.)
Figure 9, right shows the total stored elastic energy inside the coating. As expected, the energy
becomes smaller with increasing heat conductivity because the copper temperature increases so
that the thermal stresses inside the coating are reduced.
5. Discussion and Conclusions

The results presented here show that adding copper to a NiCrAlY coating system is very likely to
improve the lifetime of the coating on a copper substrate. Thermal stresses at the interface are
strongly reduced compared to the case of a standard NiCrAlY alloy. Furthermore, it can be
expected that the adhesion will be improved if copper is present in the coating.
Simulations using the rocket engine model show that the – so far unmeasured – thermal
conductivity of the coating material must not be too small to avoid extreme surface temperatures
and to increase the copper temperature because otherwise large thermal stresses will develop
inside the coating that may lead to buckling or segmentation cracks.

Depending on the actual thermal conductivity of NiCuCrAl coatings, a coating consisting of only
this material may not be feasible because the thermal gradients in the coating would be large and
surface temperatures might be extreme. However, NiCuCrAl would still be highly useful as a
bond coat material on which a second top coat is applied. The simulations shown here already
allow to draw some conclusions on the desired properties of the top coat material: (i) its service
temperature should exceed that of NiCuCrAl; (ii) its thermal conductivity should be sufficiently
large (at least as large as that of NiCrAlY) to avoid extreme surface temperatures; (iii) the
coefficient of thermal expansion should lie below that of NiCuCrAl so that thermal mismatch
stresses in the 22- and 33-direction that might case buckling are reduced (ideally the CTE value
should be chosen so that the mean thermal strain at the high coating temperature is comparable to
the mean thermal strain inside the copper material). Further finite element simulations including a
top coat will be performed in the future to facilitate the material selection of the top coat.
6. References
1. Altun, O., and Boke, Y. E., and Kalemtas, A., “Problems for determining the thermal
conductivity of TBCs by laser-flash method”, Journal of Achievements in Materials
and Manufacturing Engineering, no. 30, 2008
2. Bürgel, R., and Maier, H. J., and Niendorf, T., “Handbuch Hochtemperatur-
Werkstofftechnik“, Vieweg, 2011
3. Dies, K., ”Kupfer und Kupferlegierungen in der Technik“, Springer, 1967
4. Fassin, M., Institute of Applied Mechanics RWTH Aachen university, private
communication, 2015
5. Fiedler, T., and Fedorova, T., and Rösler, J., and Bäker, M., “Design of a Nickel-
Based Bond-Coat Alloy for Thermal Barrier Coatings on Copper Substrates”,
Metals, no. 4, 2014
6. Freborg, A. M., and Ferguson, B. L., and Brindley, W. J., and Petrus, G. J.,
"Modeling oxidation induced stresses in thermal barrier coatings", Materials Science
and Engineering A, no. 245, 1998
7. Greuel, D., and Suslov, D., and Haidn, O., and Fritscher, K. “Thermal Barrier
Coatings for Cryogenic Rocket Engines”, AIAA journal, no. 1, 2002
8. Kowollik, D., and Tini, V., and Reese, S., and Haupt, M., “Application of a
Viscoplastic Damage Model in a 3D FSI Analysis of a Rocket Nozzle”,
Sonderforschungsbereich/Transregio 40 - Annual Report 2011, 2011
9. Kowollik, D. S. C., and Horst., P., and Haupt, M. C., “Fluid-Structure Interaction
Analysis Applied to Ther Barrier Coated Cooled Rocket Thrust CChamber with
Subsequent Local Investigation of Delamination Phenomena”, Progress in
Propulsion Physics, no. 4, 2013
10. “Kupferdatenblatt CuCr1Zr“, Deutsches Kupferinstitut, 2005
11. Pawlowski, L., “Science and Engineering of Thermal Spray Coatings”, John Wiley &
Sons Ltd., 2008

12. Popp, M, and Schmidt, G, „Rocket engine combustion chamber design concepts for
enhanced life“, AIAA, ASME, SAE, and ASEE, Joint Propulsion Conference and
Exhibit, no. 32, 1996
13. Quentmeyer, R. J., “Experimental Fatigue Life Investigation of Cylindrical Thrust
Chambers, AIAA/SAE Propulsion Conference, no. 13, 1977
14. Raj, S. V., and Ghosn, L. J., and Robinson, C., and Humphrey, D., “High heat flux
exposures of coated GRCop-84 substrates”, Materials Science and Engineering A,
no. 457, 2007
15. Riccius, J., and Haidn, O., and Zametaev, E., “Influence of Time Dependent Effects
on the Estimated Life Time of Liquid Rocket Combustion Chamber Walls”, 40th
AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibit, 2004
16. Rösler, J., and Bäker, M., and Aufzug, K., “A Parametric study of the stress state of
thermal barrier coatings Part I: creep relaxation”, Acta Materialia, 2004
17. Schloesser, J., and Bäker, M., and Rösler, J., “Laser cycling and thermal cycling
exposure of thermal barrier coatings on copper substrates”, Surface and Coatings
Technology, no. 206, 2011
18. Schloesser, J., and Kowollik, D., and Bäker, M., and; Rösler, J., and Horst, P.,
“Thermal Barrier Coatings in Rocket Engines - A Multiscale Simulation and
Development Approach”, Sonderforschungsbereich/Transregio 40 - Annual Report
2011, 2011a
19. Taylor, T., and Walsh, P., “Thermal Expansion of MCrAlY Alloys”, Surface and
Coatings Technology, no. 177-178, 2004
7. Acknowledgement
Financial support has been provided by the German Research Foundation (Deutsche
Forschungsgemeinschaft – DFG) in the framework of the Sonderforschungsbereich Transregio 40,
Teilprojekt D2

Residual Strength of the Carbon
Fiber Panel with Delamination
*M.V. Pavlov, *I.A. Zharenov
* TsAGI - Central Aerohydrodynamic Institute or, Moscow distr., Russia
Abstract: Carbon fiber reinforced materials are getting widely spread. But at present time
method of the calculation of impact residual strength for carbon fiber parts is poorly developed.
Especially this method is underdeveloped in case of the delaminated composite parts. Residual
strength is very essential for any project where the composite material parts are used. The
possibility to estimate the value of residual strength of the carbon fiber reinforced parts is
highlighted in terms of safety for Aerospace structures.
In this work we propose a method to calculate the carbon fiber panel residual strength of
the civil plane wing prototype with respect to delamination progress simultaneously with
degradation law of the material properties. This approach has been developed and realized based
on ABAQUS software.
All numerical models are presented and the dependence of panel residual strength on the
delamination size was shown based on these models.
Keywords: Composite Materials, residual strength, delamination, Hashin’s criteria.
1. Introduction
FE analysis is presented in this paper. The carbon fiber panel of the civil plane wing
prototype is a research object. The complete description of this object and material properties is
shown. Obtained results are presented.
The Objective: To determine residual strength of the specimens without/with delamination
due to 90, 140 and 370 Joules energy impact into trim.
2. Finite element model

The object of research is a part carbon fiber panel of the civil plane wing prototype. Figure
1 shows the sketch of the object and it’s material properties. (all the dimensions are in mm)

Material Properties
E11 E22=E33 G12=G13=G23 ν12 ν13=ν23
16800 821 446 0.322 0.015
Figure 1 The object of research
Mesh for the part was built using 8-node quadrilateral in-plane general-purpose continuum
shell elements, with hourglass control of elements. Meshed part is shown in the Figure 2. Total
number of elements: 53 370.
Figure 2 - Mesh
3. Boundary conditions
To get the RF (reaction forces and moments) from the point all BCs were made through
center points with the coupling to the surfaces. The compression load was applied to the front edge
of the panel from the coupling point. Opposite edge was encased in all degrees of freedom, Figure
3.

Figure 3 – Boundary conditions
4. Hashin criteria & VCCT modeling in Abaqus/Standard

In Abaqus the damage initiation criteria for fiber-reinforced composites are based on
Hashin’s theory. These criteria consider four different damage initiation mechanisms: fiber
tension, fiber compression, matrix tension, and matrix compression [1, 2].
In order to make the criteria work and the material properties degradation, damage
parameters were entered:
Tensile Strength Compressive Strength Shear Strength
Longtitudinal 268 107 11.3
Transverse 8.77 21.6 11.3
In the case to model delamination progress in Abaqus/Standard the VCCT (Virtual Crack
Closure Technique) was used. VCCT is based on the assumption that the strain energy released
when a crack is extended by a certain amount is the same as the energy required to close the crack
by the same amount [1, 3].
VCCT Properties:
Mode I Mode II Mode III
0.056 0.32 0.32
From the global model of the panel a local one was built. It has the size 20x20 mm and
located in the panel center, Figure 4.

Figure 4 – Delamination area (20x20 mm, delamination- the node not marked red)
Each layer has VCCT interaction. After loading this model we got EFENRRTR (Effective
energy release rate ratio) of every layer, Figure 5, Figure 6.
2.50E-08
Effective energy release rate
2.00E-08
ratio, kgs/mm
1.50E-08
1.00E-08
5.00E-09
0.00E+00
1 2 3 4 5 6 7 8 9101112131415161718192021222324252627282930313233
Layer
Figure 5 – Plot of ENERRTR

Figure 6 – Maximum of EFENRRTR
Maximum of EFENRRTR is 2.241e-08, this energy is between layers 17 and 18. After we
got the value of max of EFENRRTR and the layers from local model we located the declamation
in this place in full-sized model.
5. Results
A series of numerical panel using Abaqus were used to demonstrate differences in
delamination damage, but generally it can be divided into two parts:
1. Determination of residual strength of the undamaged panel
2. Determination of residual strength of the panel with delaminations.
5.1. Determination of residual strength of the undamaged panel

Initiation of Hashin criteria starts at values U (Displacement) 1.16 mm and RF (Reaction
Force) 258 586 kg. Damage initiation begins from the fiber compression mechanism. Stress
distribution at this time is shown in Figure 7, 8.

Figure 7 – Stress distribution
Figure 8 – Stress distribution
Figure 9 – Hashin criteria initiation

Fiber destruction initiates in layers 2 and 3 (counting from stringer flange) Figure 9 and 10.

Figure 10 – Ply 2 (00)
Figure 11 – Ply 3 (-450)

The plot shows loading conditions (red line) and the place (blue dot) where Hashin criteria
initiate.

300000
275000
250000
225000
Reaction force, kgs
200000
175000
150000
NOVCCT
125000
HFC
100000
75000
50000
25000
0
0 1 2 3 4 5
Displacment, mm
Figure 12 – Panel without delaminations
5.2. Determination of residual strength of the specimens with
delamination
Initiation of Hashin criteria start at values U (Displacement) 1.21 mm and RF (Reaction
Force) 258 586 kg. Damage initiation begins from the fiber compression mechanism. Stress
distribution at this time is shown in Figure 9 and 10.

Figure 13 - Stress distribution

Fiber destruction initiates from layers 2, 3, 5, 8, 10, 11, 13, 14, 16 (counting from stringer
flange). The place of the damage is shown in the figures below.

Figure 17 – Ply 3 (-450)

Figure 20 – Ply 10 (00)
Figure 21 – Ply 11(+450)
Figure 22 – Ply 13 (00)

Figure 23 – Ply 14 (-450)
Figure 24 – Ply 16 (00)

The strain energy release at value U=1.91 and RF= 265 627, Figure 25
Figure 25 – Initiation of EFENRRTR, max

After energy starts to release the delamination goes on to accrue until the stress is removed,
the delamination shape is shown below.
(a) (b)
Figure 26 – Delamination shape (a) – before, (b) - after
300000
275000
250000
Reaction force, kgs
225000
200000
175000
150000 VCCT
125000
HFC
100000
75000 EFENRRTR
50000
25000
0
0 2 4 6 8 10
Displacment, mm
Figure 27 – Panel with defect, 90 Joules
The plot above shows: green point – Hashin criteria initiation, blue point – the strain
energy release.
In the same way two different delaminations were tested after been subjected to impacts of
140 and 370 Joules, the analysis results are shown below.

300000
250000
Reaction force, kgs
200000
150000 VCCT 140

EFENRTR
100000
FC
50000
0
0 0.5 1 1.5 2
Displacment, mm
Figure 28 – Panel with defect 140 Joules
300000
250000
Reaction force, kgs
200000
150000 VCCT 370

EFENRTR
100000
FC
50000
0
0 0.5 1 1.5 2
Displacment, mm
Figure 29 – Panel with defect 370 Joules
Finally, we’ve got the generalized diagram of results for panel with/without delaminations.

300000
250000
Reaction force, kgs
200000
VCCT 90
150000
VCCT 140
VCCT 370
100000
noVCCT
50000
0
0 0.5 1 1.5 2
Displacment, mm
Figure 30 – Generalized diagram
Strength for panel without defect 281 738 kgs.
Strength for panel with defect 90 Joules 275 594 kgs.
Residual strength for panel with defect 90 Joules 97,82%
6. Conclusion
The paper is devoted to strength analysis of carbon fiber specimen consisting of skin
supported by two stringers. The analysis was divided into two parts:
1. Strength determination of the panel without delamination;
2. Determination of residual strength of the specimens with damage (delaminated area after
impact of 90, 140, 370 Joules energy).

The main result of this paper is a comparative chart curve of loading the specimen with and
without delamination. The residual strength of the specimen: after strike 90 was 97.82%, after
strike 140 was 79.2%, after strike 370 was 62,64%. It can be considered that the presence of
delamination after impact with of 90 Joules energy does not affect greatly the panel residual
strength.
7. References
2. Hashin Z. Failure Criteria for Unidirectional Fiber Composites. Journal of Applied
Mechanics 1980;47:329-34.
3. Wu, E. M., and R. C. Reuter Jr., “Crack Extension in Fiberglass Reinforced Plastics,”
T and M Report, University of Illinois, vol. 275, 1965.

Shrinkage Modeling of Thermoplastic Wing Rib
*A. E. Ushakov , *A.A. Safonov, **I.A. Zharenov, **N.V. Rozin
*Skolkovo Institute of Science and Technology, Moscow distr., Russia;
** Central Aerohydrodynamic Institute or TsAGI, Moscow distr., Russia
Abstract: During manufacture of parts made of thermoplastic composites in the process of

thermoforming liquid of resin transfers to solid-state with occurrence of residual stresses in part
which may lead to distortion of the shape of the product. Transition from liquid to solid is called
the crystallization process, the reverse process – melting. In order to predict products distortion
mathematical modeling techniques are developed. Methods are allowed to evaluate the degree of
warpage of the selected design solutions and on the basis of these data to change the process
parameters or product design.
Behavior model of thermoplastic material has been developed by Skolkovo Institute of Science and
Technology and then realized in Abaqus user subroutine UMAT.
The paper presents an example of the application of this material model for thermoforming
process of the wing rib on the last states of the process – without draping process. The material of
the wing rib is Tenax®-E TPCL PEEK-HTA40. Every ply was modeled by own solid element. The
feature of modeling of plies is that one ply may differs in thickness in 2,3 times in construction that
why they were modeled in CATIA.
Geometry distortion in modeling has good agreement with the experimental data.
Keywords: Composite Materials, thermoplastics, thermoforming
1. Introduction
Thermoplastic composite materials in modern aircraft manufacturing gradually become more and
more common. One of the design solutions of thermoplastics at the moment are such the elements
if wing as ribs, produced by hot pressing.
In hot pressing heated plate of composite material is laid in the desired shape, and then stamping.
During the process the composite material subjected to high temperature and pressure from tool.

Figure 1. Typical process of thermoforming
In the pressing process the press plates are heated and melted composite material. Individual
layers of material compacted to form good contact and durable bond between the adjacent layers.
Then composite material is cooled at a rate to obtain a desired degree of crystallinity of the matrix.
During all these processes is critical to maintain the proper temperature of the composite structure
and the pressure exerted press plates. Otherwise, there will be significant residual stresses,
unsatisfactory degree of crystallinity of the matrix, insufficient adhesion between layers and other
defects.
In this paper, to solve the problem of estimating the distortion of the spoiler rib, the following
analytical models of behavior of thermoplastic material:
• The model of thermal analysis
• The model of crystallization.
These models were implemented in Abaqus by user subroutine UMAT.
2. The initial data for the calculation of distortion

The material of the spoiler rib is Tenax®-E TPCL PEEK-HTA40 [1] - is a ThermoPlastic
Consolidated Laminate (TPCL) made of several layers of thermoplastic powdered woven fabric.
The powdered woven fabric is composed of the high tenacity carbon fibre product Tenax®-E
HTA40 E13 3K 200tex and a PEEK matrix. The consolidated laminate can be simply heated
above its melting temperature and then stamped in a metallic mould within a press in a few
minutes. Properties of material are in Table 1.

Figure 2. Typical process of thermoforming
Table 1. Physical and mechanical properties Tenax®-E TPCL PEEK-HTA40

Conditioning / Typical value
Properties (test direction)
Test temperature
Melting point Peak DSC 343°C 649°F
Glass transition Onset 23°C, 50% r.h./ 143°C 289°F

temperature 23°C, 50% r.h
Tensile* (warp, 0°) Modulus 23°C, 50% r.h./ 60 GPa 8.7 Msi
ISO 527-4 23°C, 50% r.h
Strength 963 MPa 139.7 ksi
Compression* Modulus 23°C, 50% r.h./ 59 GPa 8.6 Msi

(warp, 0°) Strength 23°C, 50% r.h 725 MPa 105.2 ksi
EN 2850 Type B
Flexural (warp, 0°) Modulus 23°C, 50% r.h./ 64 GPa 9.3 Msi
EN 2562 Type A Strength 23°C, 50% r.h 1166 MPa 169.1 ksi
*normalized to nominal thickness (0,31mm) (0,31mm)
Stacking sequence: (0,90)3/(0,90)r3
2.1 Technological process parameters
Simulated process defined by the following parameters:

• billet heating temperature in the module for IR heating: 390 ± 5 ° C.
• Duration of exposure shaped body in the closed mold: 90 seconds.
• Length of the work piece conveying unit for pressing after heating: <5 seconds.
• Cooling rate: <120 K / min (393 ° C).
2.2 Geometrical parameters

Geometry of spoiler rib and the tool are illustrated in Fig.3-4.

Figure 3. Spoiler rib geometry
Figure 4. Tool geometry (modeled as rigid body)
Parameters to calculate the stresses in the user procedure UMAT, defined in the User Material in
Abaqus are shown in Table 2.
Table 2. User material properties
1 E1f Elastic fiber modulus
2 E2f Elastic fiber modulus
3 ν12f Poisson's ratio of fiber
4 ν23f Poisson's ratio of fiber
5 G12f Fiber shear modulus

6 Vf Fiber volume content
7 Arm_dir Warp direction
8 Initial cristallization Cristallization parameter
Parameters of the other equations of the model material (taking into account the loss of weight,
melting point, glass and others) were given in the relevant procedures and functions FORTRAN-
module.
3. FEA-modeling
Every ply was modeled by own solid element. The feature of modeling of plies is that one ply may
differs in thickness in 2,3 times in construction that why they were modeled in CATIA, see Fig. 5.
Figure 5. Spoiler rib stacking in CATIA

Figure 6. Plies orientation in Abaqus
3.1 Contact interactions
To implement interaction between rib and tooling in the crystallization process of the binder in the
computational model were set conditions of mechanical contact between the surface of the rib and
tooling at the first step of the calculation. In the second step of the calculation (warpage)
mechanical contact was turned off. At this point, the rib is located in the matrix and the calculation
of warpage is performed.
3.2 Temperature boundary conditions
To ensure the technological conditions of melting and crystallization of the preform given time
variable temperature. In step between dwell punch and die temperature of the billet is 390 ° C and
held for 90 seconds. Then, the contact with the matrix is removed and the preform is cooled at a
speed of 393 ° C / min. up to room temperature.
Temperature was set for all nodes ribs.
4. Warpage FEA-modeling results
Scalar fields of stresses and displacements were obtained for spoiler rib on the step of dwell and
release from the mold until equilibrium state of preform, see Fig. 7.

Figure 7. Displacement field of spoiler rib at the last step of calculating
Deviations rib belts was about 1 °, in which the original slope angles snap 91 °, see. Distortion
values in the production of belts also was about 1 °.
The calculation estimates the distortion of geometry based on thermoforming implemented in

Abaqus user model of thermoplastic material behavior with the layers of solid modeling elements
and the corresponding orientation, showed the possibility of using this material model for such
complex parts like spoiler rib.
Verification of the model requires the specification of parameters of the experimental user
material model. In the next phase of work is scheduled to enter into account the degree of
interlayer coupling through the calculation of the density of contact and autohesion (interdiffusion
of thermoplastic polymers).
5. References
1. Nam, J.-D. and J. C. Seferis, 1992, “Generalized composite degradation kinetics for polymeric
systems under isothermal and nonisothermal conditions,” Journal of Polymer Science Part B:
Polymer Physics, 30(5), 455-463.
2. Ozawa, T., 1971, “Kinetics of non-isothermal crystallization,” Polymer, 12(3), 150-158.
3. Blundell, D. J. and B. N. Osborn, 1983, “The morphology of poly(aryl-ether-ether-ketone),”
Polymer, 24(8), 953-958.
4. Talbott, M. F., G. S. Springer and L. A. Berglund, 1987, “The Effects of Crystallinity on the
Mechanical Properties of PEEK Polymer and Graphite Fiber Reinforced PEEK,” Journal of
Composite Materials, 21(11), 1056-1081.

Analysis of the Carbon Fiber Wing Torsion Box after Low
Velocity Impact Damage
*I.A. Zharenov, *M.V. Pavlov
* TsAGI - Central Aerohydrodynamic Institute, Moscow distr., Russia
Abstract: Composite products are gaining wide popularity in industrial production. But
after getting impact damage composite parts do not behave as metals due to absence of plasticity.
It is very important to analyze BVID damage and residual strength of the composite structure after
impact damage.
This problem is particularly critical for designing primary structures in aerospace

industry. One of the most important elements of the airplane construction: carbon fiber wing
torsion box was analyzed in this research.
The analysis was carried out on the damage after low velocity impact with use of Virtual
Crack Closure Technique and Hashin criteria with corresponding degradation model. For
detailed representation of stress fields submodeling technique was used as well.
Based on ABAQUS software all necessary engineering tools and complete approach to
analyze composite structures were developed.
Keywords: Composite Materials, residual strength, VCCT
1. Introduction
In Aerospace industry is very important to deliver best safety of the primary structure as
you can. For now you have an exact answer only in metal structures. So what happens if you are
using composite primary structures? You have to take into account small defects after impact
damage, because they are so dangerous in composite structures. In this paper presented method to
solve a problem - residual strength and influence on the structure of the BVID after low velocity
impact.

Figure 1 Prototype of the airplane wing
2. Theoretical part
Impact damage analysis was presented for the upper skin panel.
All parts in the analysis produced from carbon fiber reinforced composite material
For impact damage analysis Hashins criteria was used [2]. Dimensions of the
delamination areas were determined with use of the Virtual Crack Closure Technique (VCCT) [3].
Margins of safety are defined from ply strengths and delamination strength of the
structure components.
Ply strenght, КH, defined based on the Hashins criteria. Hashin criteria have a good correlation
with experimental data, and it distinguishes different modes of the destruction.
Residual strength coefficients are defined for four modes of ply damage:
, где - damage initiation index under fiber
tension ;
, где damage initiation index under fiber

compression ;
, где damage initiation index under matrix
tension ;
, где damage initiation
index under matrix compression .

Where
XT – Longitudinal tension strength;
XC - Longitudinal compression strength;
YT - Transverse tension strength;
YC - Transverse compression strength;
SL - Longitudinal shear strength;
ST- Transverse shear strength;
α - Shear stress contribution coefficient into fiber tension.
Damage initiation indexes showing material state under loading condition. How close
material to fail under four different modes.

Fracture strength coefficients, KG, are defined from:
1. If initiation of the delamination is under load lower than design load РР, so
,
where - load of the initiation of the delamination.
2. If initiation of the delamination is under load more than design load РР, so
,
where - the equivalent strain energy release rate, - the critical

equivalent strain energy release rate.
3. FE model Upper skin panel

For the analysis was developed large scale global model of the prototype of the torsion
wing box section (Figure 2). Local FE model of the upper skin panel was developed as well
(Figure 3). Global model is loaded by design load. Sub model boundary conditions were used for
the local model (Figure 4).
Figure 2 Global and local model
Figure 3 Local model of the upper skin panel

Figure 4 Sub model location with respect to global model
Stress components are calculated and shown in Figure 5. Damage initiation indexes are
presented in Figure 6, Figure 7 as well.
Figure 5 Max principal in-plane stress (left), min principal in-plane stress (center), Tresca shear stress
(right)
Figure 6 Damage initiation indexes of the skin without impact damage under loading

Figure 7 Damage initiation indexes of the stringers without impact damage under loading
Effective energy release rate ratio was obtained by VCCT method and is shown in Figure 8.
Figure 8 Effective energy release rate ratio (left), Gequiv/Gc vs design load (right)
4. Submodel impact damage. Impact energy 140 Joules

In this chapter FE model of the impact damage is presented. Rigid spherical impactor is
pushed to the panel with different rate of impact energy. This analysis was done in
Abaqus/Explicit[1]. In this work presented workflow for the impact energy 140 Joules, but the
entire work was carried out for the 50, 90, 140, 200 and 240 Joules. And final results are shown
for all energy values.
FE submodel of the reinforced upper skin panel is presented in Figure 9. In Figure 10 you
can see maximum defect depth (maximum displacement, mm).

Figure 9 Global model and local model location (left), local FE model (right)
Figure 10 Maximum displacement after impact damage, impact energy 140J
Fields of properties degradation coefficients are presented in Figure 11, Figure 12, Figure
13. Delamination area is shown in Figure 14. And all data are tabulated in Table 1.
Figure 11 Fiber damage under compression (left), tension (right).

Figure 12 Matrix damage under compression (left), tension (right).
Figure 13 Matrix damage under shear
Figure 14 Delamination area
Table 1. Damage parameters

dcf dtf d cm d tm
ds
(fiber (fiber (matrix (matrix
(shear)
compression) tension) comp) tension)
Maximum 0.655 0.818 0.541 0.877 0.917
Average 0.328 0.409 0.271 0.439 0.459
Area of the damaged
129 71 63 715 722
plies,mm2
Delamination area, mm2 15338
5. Damaged material diagrams

Damage parameters were calculated for all energy rates and presented in Table 2. Graph
representation is shown in Figure 15, Figure 16 and Figure 17.
Sdam – area of the material with degraded properties after impact,

Sdelam – delamination area.
Table 2. Average values of the damage indexes and damaged material areas
Impact energy, J dcf dtf dcm dtm ds Sdam, mm2 Sdelam, mm2
50 0.201 0.252 0.064 0.385 0.426 226 6274
90 0.288 0.370 0.204 0.420 0.439 437 11161
140 0.328 0.409 0.271 0.439 0.459 722 15338
200 0.339 0.429 0.282 0.444 0.483 852 19079
240 0.352 0.438 0.298 0.452 0.487 946 20321
Figure 15 Delamination area versus impact energy
Figure 16 Damage indexes versus impact energy

Figure 17 Area of the damaged ply material versus impact energy
6. Residual strength
In this part FE model for calculating the residual strength is presented in the Figure 18.
Red area is the area of the material with degraded properties, 25mm*10mm.
Damaged material properties are presented in Table 3.

Table 3. Average properties of the damaged ply material
Damaged material properties.
Impact energy, J
Parameters Raw monoply
50 90 140 200 240
E1, kgs/mm2 15200 11764.80 10203.00 9602.60 9367.00 9199.80
E2, kgs/mm2 846 656.07 582.26 546.09 538.90 528.75
G12, kgs/mm2 446 256.23 250.21 241.51 230.81 229.02
XC, kgs/mm2 156 124.72 111.07 104.91 103.12 101.17
XT, kgs/mm2 268 200.60 168.97 158.39 153.16 150.62
YC, kgs/mm2 21.7 20.31 17.28 15.83 15.58 15.23
YT, kgs/mm2 8.8 5.41 5.10 4.94 4.89 4.82
S, kgs/mm2 8.4 4.83 4.71 4.55 4.35 4.31

Figure 18 FE model for the residual strength calculation
7. Damageability energy is 140 Joules

Damage initiation indexes are presented in Figure 19, Figure 20. The values of the
indexes are shown in Table 4.
Figure 19 Damage initiation indexes of the skin with impact damage under loading, HXТ, HXС, HYТ, HYC
Figure 20 Damage initiation indexes of the stringer with impact damage under loading, HXТ, HXС, HYТ, HYC
Figure 21 Effective Energy release rate ratio.

Effective energy release rate ratio field is shown in Figure 21. Two diagrams of the
Effective Energy release rate ratio are shown in Figure 22.
Figure 22 Gequiv/Gc versus design load for raw and damaged skin panel
Table 4. Damage initiation indexes and residual strength coefficients. Impact energy, 140 J
HXT HXC HYT HYC KXT KXC KYT KYC
Skin 0.328 0.382 0.299 0.413 3.049 2.618 3.344 2.421
Stringer 0.274 0.472 0.182 0.486 3.650 2.119 5.495 2.051
Minimal value, K 2.05
Table 5 shows the residual strength value.
Table 5. Residual strength. Impact energy, 140 J
Delamination area Residual strength
Skin 0.505
8. Residual strength diagrams

In this chapter residual strength diagram is presented. As described above, the same method
was applied with different values of impact energy. Coefficients Kply and Kdelam are presented in
Table 6. Graphical representation is shown in Figure 23.
Kply – Strength coefficient of the ply,
Kdelam – Fracture strength of the skin.

Table 6. Residual strength

Impact damage, Joules Kply Kdelam
0 3.53 1.76
50 3.29 0.94
90 2.16 0.71
140 2.05 0.505
200 1.72 0.33
240 1.50 0.30
Figure 23 Residual strength versus impact energy diagram
9. Conclusions
The Abaqus software contributed greatly to solve this challenge. The Hashin criteria and
VCCT allow developing this residual strength calculation method after low velocity impact– one
of the most important tasks in aerospace industry.
10. References
1. Abaqus Users Manual, Version 6.12, Dassault Systémes Simulia Corp., Providence,
RI.
2. Hashin Z. Failure Criteria for Unidirectional Fiber Composites. Journal of Applied
Mechanics 1980;47:329-34.
3. Wu, E. M., and R. C. Reuter Jr., “Crack Extension in Fiberglass Reinforced Plastics,”
T and M Report, University of Illinois, vol. 275, 1965.

A Simulation Tool for Fatigue Analysis and
Lifecycle Prediction
Daniel Kujawski*, Phani C.R. Sree, Deepak Abburi and Joshua T.L. Kuok
Western Michigan University
Mechanical and Aerospace Engineering, Kalamazoo, MI 49008-5343, U.S.A.
Abstract: This paper presents an interactive web-based simulation tool for a modern fatigue
analysis and lifecycle prediction methodology of smooth and notched components. Users around
the globe with diverse degree of fatigue familiarity may access it via Internet by means of multiple
platforms such as desk- and lap-top computers, tablets and/or smart-phones. In particular, the
users with a limited fatigue analysis background would benefit from “on-the-fly” fatigue learning
experience. This is accomplished by means of proper guidance through a step-by-step process and
providing specific details and explanations without the need of a tutorial handbook. It makes a
self-explanatory and frustration free web-based simulation software, which allows users to learn
the fatigue fundamentals while expanding their knowledge on modern fatigue analysis methods.
For variable amplitude loading a dedicated spectrum analysis tool is provided for a potential
clean-up and desire modifications of a raw spectrum data. Subsequently, a rainflow method is
used and the corresponding hysteresis loops at the notch-root are determined. Then, the relevant
interactive graphs, calculated values and tables are displayed. The web-based learning tool can
be accessed at www.fatiguenet.com.
Keywords: Web-based software, fatigue analysis, notch stresses, Neuber’s rule, spectrum loading.
1. Introduction
In general, fatigue damage occurs in materials when subjected to variable amplitude stresses that
are below the ultimate tensile strength, or even the yield strength of the material (Dowling, 2013).
To avoid fatigue cracking, components are designed by allowing a sufficient safety factor. Too
high of a safety factor would result in bulky and expensive components. Hence, R&D departments
employ highly trained engineers to perform fatigue analysis on components before and during
prototype development. This involves extensive time and resources to create new designs. Due to
cut offs in R&D budgets many new designs are based on old and proven designs. In the global and
highly competitive market, the R&D departments are mostly afforded by relatively large
organizations. Routinely the R&D departments acquire dedicated fatigue software. On the other
hand, smaller organizations seldom have R&D departments but they may have a periodic need for
fatigue analysis. Therefore, there is an overall demand for a straightforward web-based fatigue
analysis tool accessible via Internet, which can be used occasionally. Specifically, in an age of
technological advancement, which is growing exponentially, there is an increasing trend that
students in colleges shift more towards an ‘e-learning’ platform to grasp knowledge outside of the
classroom environment.
In order to carry out a meaningful fatigue analysis, requires some fundamental fatigue life analysis
knowledge, which most students would not have learned during their undergraduate studies. Also,
most practices in small companies are limited when it comes to fatigue analysis and life prediction

know-how experiences. For more than a decade, fatigue analyses websites are available on-line
(e.g. References 4-6). At present, these websites provide material database and elaborated fatigue
analyses that also include fatigue crack propagation and multi axial fatigue analyses. Hence, this
paper is focused on a simple, self-explanatory and interactive web-based fatigue analyses learning
tool dedicated to the first time or unexperienced users.
Nomenclature
E modulus of elasticity ε’f fatigue ductility coefficient

b fatigue strength exponent ∆γ/2 shear strain amplitude on critical plane
c fatigue ductility exponent σa normal stress amplitude
FS Fatemi-Socie parameter σar fully reversed stress amplitude, applied
K coefficient in Fatemi-Socie parameter σar G fully reversed stress amplitude, Goodman
kt elastic stress concentration factor σar M fully reversed stress amplitude, Morrow
kf fatigue notch factor σar SWT fully reversed stress amplitude, SWT
2Nf number of reversals to failure σ’f fatigue strength coefficient
R stress ratio, σmin/σmax σm mean normal stress
S nominal normal stress σmax maximum normal stress
SWT Smith-Watson-Topper parameter σ min minimum normal stress
ε a normal strain amplitude σN max maximum normal stress on critical plane
ε a e elastic strain amplitude, σa/E σu ultimate tensile strength
ε a p plastic strain amplitude, ∆ε p/2 σy yield strength
The aim of this paper is to present an interactive web-based tool for learning a modern fatigue
analysis and life prediction methodology of smooth and notched components. The web-based tool
may be easily accessed by means of multiple platforms such as
desktop and laptop computers, tablets and/or smart-phones, as
illustrated in Fig. 1. Users with limited fatigue background
would benefit from the proposed interactive “on-the-fly”
learning experience. At each step, concise fundamental fatigue
information is provided without the need of a tutorial handbook.
It constitutes interactive, self-explanatory web-based
educational tool, which allows the users to learn the fatigue
fundamentals as well as to expand and master their knowledge
on modern fatigue analysis methods. It can be used alone or in
tandem with commercial FEA software.
Figure 1 Illustration of the
web-based access.

For variable amplitude loading histories a rainflow method is used to count the cycles and to
determine the corresponding hysteresis loops at the critical locations. In addition, a dedicated
spectrum analysis tool is provided, which enables clean-up and desired modifications of a raw
spectrum data such as to eliminate midpoints, truncate or clip a given spectrum. During analysis
the relevant interactive graphs and calculated values are displayed. The following sections provide
a summary of the proposed web-based interactive fatigue analysis learning tool.
2. Structure/flowchart of the web-based software
The core structure of the proposed web-based software shown in Fig. 2 provides multiple options
for users to select. Unexperienced or first time users could be confused given so many options. In
order to assist such users, brief explanations and tips for each step are provided as shown in
Appendix. The objective of these explanations together with provided equations regarding fatigue
life predictions approaches is to map-up common similarities and differences among them.
Figure 2 General structure of the educational version of the web-based fatigue

analysis tool.

3. Web-based software
The website is structured to be simple and self-explanatory, flowing from Step 1 to Step 7, as
shown in Fig. 3. The current website features an educational version of the software along with
video examples of using the software. The website also provides additional spectrum analysis
software to prime the raw spectrum data before fatigue life prediction analysis can be carried out
using the educational version. In this section, the analysis procedure is discussed with different
options available on the website. As the user enters the website (www.fatiguenet.com), the home
page is presented. Under ‘Educational Version’, the ‘Software’ tab is selected, which presents the
life predication software. Below are the steps involved with explanation and screenshot of the steps
depicted in Fig. 3. In this educational version, the user would be presented with helpful tips and
information at each step.
Step 1: Select Material

In this step, the desired material, units (SI or English), and language are selected. The educational
version provides five different materials with an additional option to input user defined material
properties. Six languages: English, Chinese, French, German, Japanese, Russian, and Spanish are
available to select with English being default language.
Figure 3 Screenshot of Step-1 to Step-7 from website.

Step 2: Specimen Type
In this step, the users are provided with an option to choose a smooth or a notched
specimen/component. The ‘smooth’ is used in generic terms, i.e. it applies to notches when the
local elastic/plastic strains and stresses are known, e.g. from FE analysis. Selecting a smooth
specimen would result in a kt value of 1 to be used in the analysis, which the users cannot change.
On the other hand, for a notched specimen the users would be able to input the desired kt (or kf)
value (kt = 2 is default) in the proceeding step. A plasticity corections for notch-root stresses and
strains are obtained using Neuber’s rule (Neuber, 1961).
Step 3 : Input type

If smooth specimen is selected in Step 2, users would have the option to select either stress or
strain as the input data type. If notched specimen is selected the nominal stress, Sn , is the input
data type (where Sn < σy).
Step 4: Loading Type

The user can select three loading types to be analyzed. The software is designed to analyze the
following loading types:
1. Constant Amplitude
2. Block Loading
3. Spectrum Loading
.
Step 5 : Input Data
In this step kt (or kf) value is specified. If smooth specimen is selected in step 2, the value is
displayed as 1 and cannot be altered. In case of a notched specimen, the default value is 2 but may
be changed according to the users need. For a constant amplitude loading the user shall specify the
loading levels as S level 1 and Slevel 2. Loading always starts from zero and continues towards Level 1,
which can be the maximum or the minimum stress. Proper information is displayed if the user
chooses Block Loading or Spectrum Loading options.
Step 6 : Neuber’s Analysis/Hysteresis Loop(s)

Selecting “Analyze” in Step 5 displayes Neuber’s curve (the upper curve in Fig. 4a) which is
interacting with the Ramberg-Osgood stress-strain curve (the lower curve in Fig. 4a). For a given
material the interactive Neuber’s curve is valid for any value of the product of Sn and kt (or kf).
Also, instead of kt*Sn (or kf*Sn ) value, the elastic notch-root stress can be used and the Neuber’s
curve will provide the corresponding plasticity correction for estimated elastic-plastic stress and
strain. It can be noted that a deviatoric formulation of Neuber’s rule (Kujawski, 2014) is not active
in this educational version.
Then, selecting the ‘Hysteresis’ button displays the hysteresis loop shown in Fig. 4b or loops for
the Block Loading or Spectrum Loading options.

(a) (b)
Figure 4 Screenshot of (a) the interactive Neuber’s graph and (b) the hysteresis
loop.
Step 7 : Life Prediction Methods

In this step the life prediction method based on stress, strain or stress-strain is selected. The stress-
based approach includes four well known approaches for mean stress effects, i.e. (Goodman,
1930), (Morrow, 1968), (Smith, 1970) and (Walker, 1970). Strain-based approaches consist of
Morrow (Morrow, 1968) and the strain versions of Goodman and Kujawski-Ellyin (Kujawski,
1995) models. The SWT parameter (Smith, 1970) is available as the stress/strain-based method. A
critical plane method of Fatemi-Socie (Fatemi, 1988) and the deviatoric formulation of the SWT
parameter (Kujawski, 2014) are not actve in this educational version.
The output window provides a consolidated numerical values for maximum, minimum, mean and
amplitude of stresses and strains at the notch tip. The life predictions graph is depicted in Fig. 5
and the corresponding fatigue lives are provided in Fig. 6. After completion of the analysis, three
printing options are avalilable such as: print this page, print this table, and print explanations (Fig.
6).

Figure 5 Screenshot of graphical representation of life prediction for various
approaches.
Figure 6 Screenshot of comparison table showing numerical values of life

prediction.
4. Spectrum Analysis Tool
Besides the life analysis software, the website also provides a spectrum analysis tool to
prime/optimize the raw spectrum data before fatigue life analysis can be carried out. The software
is divided into 5 main steps:
1. Cleaning
2. Rainflow method
3. Removing small cycles
4. Clipping
5. Truncation

The software also consists of short explanation videos for the 5 steps listed above to assist user in
further understanding each process.
4.1 Cleaning
Spectrum analysis should always only contain peaks and valleys in its data. In this step, a raw
spectrum data would be cleaned by removing its intermediate points, and thus leaving only peaks
and valleys in the spectrum data so that it can be used for spectrum analysis.
Figure 7 Intermediate points removal - before and after cleaning.
As depicted in Fig. 7, the raw spectrum data had 6 points in total, which consists of 2 intermediate
points. Once the cleaning process is done, the 2 intermediate points were removed, the graph on
the right in Fig. 7. Hence, now leaving only 4 points when compared to the initial 6 data points.
4.2 Rainflow Method

Rainflow method is used to count the number of cycles for the life prediction of the component as
shown in Fig 8. A cycle will not be counted if the previous range, ∆σXY, is larger than the
following range, ∆σYZ (Fig. 8a). But if the opposite occur a cycle X-Y-X’ is counted (Fig. 8b).
(a) (b)
Figure 8 Rainflow method (a) No cycles counted, (b) A cycle of X-Y-X’ is counted.
4.3 Removing Small Cycles

If the loading histories containe ranges smaller than the user defined threshold value, ∆th, that does
not affect the life, therefore the cycle(s) associated with those ranges could be removed. This
shortens the analysis, since the number of loading cycles to be considered is reduced. In this step,
a threshold range, Δth, is specified (e.g. ∆th=15), and any range smaller than Δth will be removed,
and turning points will be updated. This method of removing the small cycles is illustrated in Fig.
9.

Figure 9 A spectrum before and after removing small cycles with ranges ∆ ≤ 15.
4.4 Clipping
Clipping allows users to modify the raw spectrum data of the component, by eliminating some of
the highest loading peaks. This is done when the user specifies its Mm ax value, and any point(s) in
the loading histories that is larger than Mm ax will be clipped down to the Mm ax value, turning
points will be updated subsequently. This method is illustrated in Fig. 10 shown below. Points 2
and 4 from the graph on the left of Fig. 10 are clipped down to the Mm ax red dotted line. If
intermediate points are created it will be removed, the final result is shown on the right graph in
Fig. 10.
Figure 10 A spectrum before and after clipping.

4.5 Truncation
Truncation is very similar to clipping. Instead of eliminating high overload peaks, truncation
eliminates underload peaks in loading histories. This is done in the similar way as clipping, by
specifying its Vtruncation level, and any points in the loading histories that are below the specified
Vtruncation level will be removed, and turning points will be updated. This is illustrated in Fig. 11.
Figure 11 A spectrum before and after truncation.
Points 1,5,6 and 7 on the left side of the graph in Fig. 11 are truncated up to the Vtruncation level ,
turning points are then subsequently updated.
5. Conclusions
An interactive web-based learning tool for a modern fatigue analysis and life prediction
methodology of smooth and notched components has been developed. Users from around the
globe may access it via Internet by means of multiple platforms such as desktop and laptop
computers, tablets and/or smart-phones. In particular, the users with a limited fatigue background
would benefit from possibility to expand their fatigue learning experience. The tool can be used to
analyze constant amplitude, block loading and spectrum loading histories. Dedicated spectrum
analysis software is also included. Life predictive capabilities include stress-based, strain-based
and stress-strain based approaches.
Acknowledgements
This research is supported by Western Michigan University Technology Development Fund.
6. References
1. Dowling, N. E, “Mechanical Behavior of Materials,” fourth ed., Prentice Hall, Upper
Saddle River, New Jersey, 2013.

2. Fatemi, A., Socie, D. F, “A Critical Plane Approach to Multiaxial Fatigue Damage
Including Out-of-phase Loading,” Fatigue Fract Eng Mater Struct., Vol. 11, 1988,
149–165.
3. Goodman, J., “Mechanics Applied to Engineering,” London, Longmans Green, 1899 [see
also, vol. 1, 9th ed., 1930].
4. https://www.efatigue.com
5. http://fde.uwaterloo.ca
6. http://fadoff.cz/page
7. Kujawski, D, and Ellyin, F, “A Unified Approach to Mean Stress Effect on Fatigue
Threshold Conditions, International Journal of Fatigue, Vol.17, 1995, 101-106.
8. Kujawski, D, “On Energy Interpretations of the Neuber’s Rule,” Theoretical and Applied
Fracture Mechanics, Vol. 73, 2014, 91-96.
9. Kujawski, D, “A Deviatoric Version of the SWT Parameter,” International Journal of
Fatigue, Vol.67, 2014, 95-102.
10. Morrow, J. D, “Fatigue Properties of Metals,” In: Fatigue Design Handbook Sec. 3.2.
SAE Advances in Engineering, 1968, 21–30.
11. Neuber, H, “Theory of Stress Concentration for Shear Strained Prismatic Bodies with
Arbitrary Non Linear Stress Strain Law,” Trans. ASME, Journal of Applied Mechanics,
1961, 544–555.
12. H. Smith, K. N., Watson, P., Topper, T. H, “A Stress–Strain Functions for the Fatigue of
Metals,” Journal of Materials, Vol. 5, 1970, 767–778.
13. Walker,K, “The Effect of Stress Ratio During Crack Propagation and fatigue for 2024-T3
and 7075-T6 Aluminum,” Effects of Environment and Complex Load History on Fatigue
Life, ASTM STP462,West Conshohocken, PA, 1970, 1-14.
Appendix: Explanations provided in each step
Life Analysis Tool Educational version supports up to 100 points in spectrum/block loading
(uniaxial only). Other loading conditions such as torsion, bending and multiaxial are supported in
the full version.
Step1: Select Material

Selection from five different materials with custom material option is available in the educational
version (more materials are available in the full version)
Select SI unit or English unit
Note: More detail explanations and descriptions for each step are available under i icon.
Stress Strain
SI units MPa m/m or mm/mm
English units ksi in/in

Step 2: Specimen/Component
Smooth
Select input P
Notched
Step 3: Input Type

Nominal Stress Sn = P/An An
(Sn < σ0 where σ0 is the yield strength)
Step 4: Loading Type

Three loading types are available for selection:
1. Constant Amplitude Loading
2. Block Loading P
3. Spectrum Loading
Step 5: Input Data

For smooth specimen/component
kt (or kf) = 1, is set as default and cannot be changed
For notched specimen/component
kt (or kf) = 2, is set as default and can be changed accordingly
Constant Amplitude Loading
Input stress Slevel1 and Slevel2 or strain ε level1 and ε level2
Loading starts from zero and proceeds to Slevel1 (or ε level1) then is cycling between Slevel1 (or ε level1)
and Slevel2 (or ε level2).
Slevel1 (or level1) can be positive or negative.
Block & Spectrum Loadings
User have the option to download sample file or choose their own data file for analysis. Users can
select a normalized factor if desired.
Analyze (by clicking on)
An Interactive Neuber’s chart is displayed corresponding to Slevel1 or ε level1, respectively.
Step 6: Hysteresis
An initial loading and hysteresis loop (s) curves are displayed.
Step 7: Life Prediction

Three live prediction approaches are available for selection:
1. Stress based (Goodman, Morrow, SWT, Walker),
2. Strain based (Morrow and strain versions of Goodman & Kujawski-Ellyin),
3. Stress-Strain based (SWT, Fatemi-Socie, SWT-deviatoric)
Life predictions are determined for all models within the selected approach (stress, strain or
stress-strain based).
Note: For σm < 0 Goodman’s relation is typically set to σarG = σa.
Output
Relevant local and nominal stresses and strains, endurance limits, calculated lives and fatigue
graph(s) are displayed.
An entire screen or table only can be printed by selecting a suitable printing button.

ARCHITECTURE, ENGINEERING
& CONSTRUCTION

Elastoplastic Simulations with a Tangential
Plasticity Constitutive Model for a Thin Wall Bridge
Pier Subjected to Various Non-proportional Cyclic
Loading Conditions
Hideto Momii1, Seiichiro Tsutsumi2, Riccardo Fincato2
1)
ESTECH Corporation
2)
Joining and Welding Research Institute, Osaka University
Abstract: Various elastoplastic constitutive models have been proposed to predict the structural
stiffness; most of them use an associated flow rule, which defines a plastic flow direction normal
to the yield surface. These models, based on this assumption, have contributed to the predictions
of elastoplastic deformations of solid structures in several applications, however, in some cases,
they tend to overestimate the structural stiffness. In order to overcome this defect, we have
adopted an unconventional elastoplastic model capable of taking into account the generation of
the inelastic strain rate not only along the direction normal to the yield surface but also along the
tangential one. In this paper the aforementioned model has been studied by applying a series of
non-proportional loading paths to a thin wall pier and comparing the results obtained with the
ones derived by neglecting the tangential contribution.
Keywords: Non-proportional Cyclic Loading, Elasto-plastic FE Analysis, Tangential Plasticity,

Low Cycle Fatigue Life, Constitutive Equation.
1. Introduction
Thin wall steel bridge are widely used in highway bridges, buildings, and offshore structures. In
Japan, on 17th January 1995, the Hyogoken-Nanbu earthquake hit huge region around Kobe and
Osaka and its magnitude, 7.3 degree, was higher than that assumed level. Many bridge piers of the
highway were severely damaged due to these cyclic excitations. Seismic solicitations are naturally
quite complex, the amplitude changes during earthquake as well as its direction, which usually is
not uni-directional but rather bi-directional and non-proportional in lateral direction. Therefore, in
order to correctly design the structural stiffness for those conditions, several cyclic non-
proportional loadings have been investigated in this work, trying to understand the reliability of
the model to catch a realistic behavior for the bridge pier. Moreover, bridges structural parts,
buildings and offshore piers may not experience seismic actions only one time, therefore, in
addition to the previous considerations, the design method must be able to take account also the
cumulative damages due to loading history.

On the other hand, numerous experimental studies have shown that fatigue life depends not only
on the strain amplitude but also on the strain path as well, in fact, it changes if proportional or non-
proportional cyclic loading conditions are applied (c.f. Itoh, 2013; Hoshide, 2011; Reis, 2011).
The base of cantilever structures including bridge piers, for instance, is one of the typical site
where the principal stress/strain directions changes under the macroscopic unidirectional loading
conditions, as well as for more complex operational loading conditions. Therefore, it is important
to predict accurately the plastic deformations under general loading conditions and to reflect them
in a damage counting law for fatigue life prediction.
The conventional elastoplasticity models (Drucker, 1988), based upon the premise that the interior
of yield surface is a pure elastic domain, have contributed to the predictions of elastoplastic
deformation of solid materials such as metals, geomaterials and concrete. They, however, cannot
describe cyclic loading behavior below the yield stress of materials due to the neat distinction into
two separate domains: elastic and plastic ones. An additional drawback is the unrealistically stiff
response that might be generated in multiaxial loading problems, especially for those where a non-
proportional loading path is considered. The main reason for this is the assumption that the
inelastic strain is not induced by the stress rate component tangential to the yield surface but it is
uniquely directed outward along the normal to the plastic potential surface, underestimating the
real development of irreversible contributions.
In order to overcome these defects various elastoplasticity models have been proposed up to the
present (Hashiguchi, 2009). Many of them however, except the one adopted in this study
(Hashiguchi, 2001, 2011, 2014; Tsutsumi, 2005), may not be applicable for a general
deformational behavior including cyclic loading conditions.
In this paper an unconventional elastoplastic constitutive model based on the Extended Subloading
Surface model with additional modifications to catch the so-called “Tangential Plasticity”
(Tsutsumi, 2005), has been adopted for the numerical analyses. This theory allows to take into
account the contributions of the plastic deformations, even within the elastic domain of
conventional plasticity theories, and the tangential inelastic strain induced by a stress rate
component tangential to the plastic potential surface. Experimental results have been compared
with the numerical ones carried out including or not the effect of tangential plasticity.
2. Elastoplastic constitutive equation

2.1 Basic description of extended subloading surface model
In the present work the extended subloading surface model (Hashiguchi, 2009), in the form of the
cutting-plane return mapping method (Simo, 1985; Hashiguchi, 2011, 2014), has been adopted for
a fast and accurate computation.
For sake of brevity, the extended subloading surface constitutive equations and its return mapping
formulation will not presented in the present paper; the reader is referred to the lecture notes
(Hashiguchi, 2009) for a full and detailed explanation of the theory. As a general description of the
model it can be mentioned here that the smooth transition from elastic and plastic domains is
achieved by adding a new loading surface (i.e. subloading surface), which is created by means of a
similarity transformation from the conventional yield one. The insertion of a mobile similarity

center enriches further the formulation, allowing to a more realistic description of the irreversible
strain accumulation during cycling loading.
2.2 Extension to the tangential plasticity

Most of the conventional plasticity models, and unconventional one as well, adopt an associate
flow rule for the definition of the plastic strain rate/increment. Under this assumption it is possible
to catch a realistic response in case of proportional loading paths, where the ratio among the
principal stresses and their directions are kept constant. However, whenever a non-proportional
loading condition is imposed, they tend to overestimate the material stiffness.
In order to take into account the tangential stress rate component some preliminary hypotheses
(Hashiguchi, 2001; Tsutsumi, 2005) are needed:
1. the stress rate is linearly related with the tangential stretch (where A is a stress function):

Dt = A σ*t (1)
2. the additive decomposition of the stretching holds:

D = D e + D p + Dt (2)
3. the tangential stress component, and the inelastic stretch associated, are purely deviatoric
(Rudnicki, 1975);
4. no hardening can be generated by a stress rate component tangential to the yield surface.
The main idea of this new numerical algorithm is to adapt the return mapping, formulated with an
associate flow rule, for taking into account the tangential relaxation in order to keep the
advantages of a fast but accurate computation.
In particular, the last of the assumptions (i.e. number 4), allows us to split the normal and
tangential stress rate component effects, evaluating the first with the cutting-plane method and the
second with a sort of ‘radial return mapping’.
After the stress is brought back to lie on the correct plastic surface (in Figure 1) through a series of
local convergences on cutting-planes, the tangential stress rate component can be subtracted from
the total stress state. The definition of the rate involves some mathematical passages starting from
the substitution of Equation 1 into Equation 2 and inverting the relationship to express the stress
rate as function of the strain one.
In order to split the aforementioned contributions, the generic stress function A is given as:
 T 
=A = ; T ξ R
b
(3)
 2G − 2GT 
Where T is an exponential function that depends on two material constants and b, and on the
similarity ratio R. Finally it is possible to write the stress rate as in Equation 4, where the
tangential one is expressed in the lower expression:

(4)
However, once the subtraction is performed, the new stress state will not satisfy anymore the local
equilibrium, and then a correction must be introduced as shown in the following system of
equations:
σ (nk++11),2 correct= σ (nk++11),1 − ( σ*t + correction )
o
Ν correct ,2 = σ *(n +k1+1),2 correct σ *(n +k1+1),2 correct )

(5)
=λntangential
+1 3 2 σ *(n +k1+1),2 correct − 3 2 σ *(n +k1+1),1
σ (nk++11),final= σ (nk++11),2 correct − 2 3 λntangential

+1 Ν correct ,2
It is worth mentioning that this strategy does not require an iterative procedure, since the distance
( k +1),2 correct
between σ n +1 and the final stress state point can be easily evaluated without slowing
down excessively the computational time.
2.3 Consideration of stress plateau response

As an additional feature, for a more realistic description of the material behavior, a special
numerical procedure has been added in order to freeze the material hardening after the yielding,
obtaining the strain plateau visible in Figure 2.
Figure 1. Tangential stress correction procedure.

Table 1. Material constants
Elastic moduli [GPa],
Initial value of yield surface [MPa]
Isotropic hardening
Stress plateau threshold
Kinematic hardening
Evolution of normal yield ratio
Translation of similarity center
Tangential plasticity
Outlined squared: Experimental results (Nakajima, 2000)

Lined: Simulation results
Figure 2. Stress-strain responses of SS400 for the strain ranges =0.01, 0.02,
0.03, 0.04, 0.05 (the responses of the models with (T=0) and (T≠0) exhibit the
same results under uniaxial stress cycles.)

Briefly, the user can set the value of the cumulative isotropic hardening variable (i.e. H in the
subloading surface model) as a threshold beyond which an expansion of the yield surface is
possible. Below that value the plastic deformations can occur without contributing to the material
hardening, as it happens in a perfect plasticity case.
3. Comparison with test result

3.1 Description of benchmark experiments
The experimental results, compared against the numerical simulations, were carried out by
Nishikawa, 1998. Figure 3a shows the sketch of the specimen, fixed at the base and subjected to a
constant axial load and quasi-static cyclic lateral displacements at the top of the column.
Geometrically the sample consists of a thin wall bridge pier, with a circular cross section of 9 mm
thickness, which is assumed to be made of structural steel SS400 (JIS) with a stress plateau
domain just after the yielding. As shown in Table1 and Figure 2, the material constants are fitted
using the test results performed by Nakajima, 2000. The geometric properties of the specimen are
shown in Table 2, where h is the height from the base of the column, 2r is the outside diameter of
(a) (b) (c)
Figure 3. Schematic representation of experimental system and its finite

element model.
Table 2. Geometric properties of the specimen.

h 2r t A0 Pz Pz/ Grade
(mm) (mm) (mm) (mm2) (kN) ( ) (JIS)
3403 891.0 8.9 24663 880.1 0.121 SS400

the cross section, t is the wall thickness, A0 is the cross section area and Pz is the constant axial
load.
3.2 Description of finite element model and boundary conditions

The finite element analyses were conducted by means of the commercial finite element code
Abaqus/Standard ver.6.13. Figure 3b displays the finite element model and boundary conditions,
which reproduce the ones experimentally realized by Nishikawa, 1998. In order to reduce the
calculation costs, the upper half of the column is modeled by beam elements (B31), the remaining
part by brick elements (C3D8) and by a rigid body, which is used to connect the solid with the
beam elements at the interface. Six elements have been used for the thickness discretization and
the mesh density is higher in the lower part of the column, where local buckling is expected to
happen.
The subloading surface model with tangential stress rate effect has been implemented into Abaqus
via user subroutine UMAT. In order to identify the material constants, we have used the
measurement results under the monotonic and cyclic stress-strain curves (Nakajima, 2000). The
results of the calibration, and the SS400 steel material constants, are shown in Figure 2 and Table
1 respectively. On the other hand the material constants for tangential plasticity, reported in Table
1, cannot be obtained by fitting the uniaxial stress-strain curve, therefore they were defined
through a trial-and-error approach, comparing the outcome of the FE simulations against the
Black thin line: Experimental result (Nishikawa, 1998)

Colored bold line: Simulation results
(a) w/o tangential plasticity (b) w/t tangential plasticity
Figure 4. Relationship between horizontal displacement and horizontal load at

loading point on top: (a) elastoplasticity without tangential stress rate effect and
(b) elastoplasticity with tangential stress rate effect.

As shown in Figure 3b, the base of specimen is constrained, whereas a double boundary condition
is applied on top: the first one is a compressive axial load Pz, constant through all the analysis, and
the second one is a series of cyclic lateral displacements dx along the x-direction, whose amplitude
is increased during the simulation. The axial load Pz is set at 12.1% of the full-section yield load,
which is calculated from the nominal yield stress at cross section area. The horizontal
displacements dx are calculated by:
d x = nd 0
H y h3
d0 = (6)
3EI
 Pz  z
H
= y σ y − 
 A0 h
where n is the number of loading cycles, σ y is the yield stress of material, z is the section
modulus of the specimen, I is the second moment of area. The d0 is the loading point displacement
when the lower part of the specimen yields. The history of the lateral displacement dx is given in
Figure 3c.
3.3 Results and Discussions

The comparison between the hysteresis responses of the column obtained in the FE simulation,
with (T≠0) and without tangential stress rate effect (T=0), and the experimental ones are
w/o tangential plasticity
w/t tangential plasticity
(a) +4d0 (b) +5d0 (c) +6d0 (d) +7d0 (e) +8d0 (f) +9d0
Figure 5. Evolution of the local buckling at the bottom part of the specimen and
maximum principal strain distributions: the upper figures are calculated by
elastoplasticity without tangential plasticity and the lower figures are calculated
by elastoplasticity with tangential plasticity.
presented in Figure 4. The curves in Figure 4b are in better agreement with the experiment than

Figure 6. Comparison of the local deformations, from +1d0 to +9d0, at the bottom
of column between (a) w/o tangential plasticity and (b) w/t tangential plasticity.
the ones obtained with the conventional plasticity (T=0). The tangential plasticity model can, in
fact, capture both: the ultimate load and the decreasing of the strength for the each of the post
cycle peaks during the cycles.
Figure 5 depicts the deformations carried out by using the conventional and the tangential
plasticity models in the FE simulations, whereas Figure 6 compares the horizontal profiles of the
sample obtained by the nodal displacements around the bottom of the column. Both of them
indicate that the localization carried out by means of the tangential plasticity algorithm tends to
occur at an earlier stage, moreover it is enhanced with respect to the one resulting from the
conventional plasticity model.
4. Various bi-directional loading conditions in the cross section

4.1 Loading paths
In this section seven loading paths (Ucak, 2014) are applied to the model in order to investigate
differences in response under bi-directional loading conditions. The schematic representations of
these seven loading processes are shown in Figure 7(a)~(g). Paths BS and BA trace a butterfly
shape, symmetrically BS and asymmetrically BA, in the X-Y displacement space. Both XU and
XP trace a cross shape path, however the number of changes of the loading direction in each cycle
is different. Paths DM and SQ have the same shape, but the direction at the maximum
displacement in each cycle is different. Case CR traces a simple circular path.

(a) BS (b) BA (c) XU (d) XP
(e) DM (f) SQ (g) CR (h) Ex. CR Cyc. 1 to 9
Figure 7. Schematic representation of the loading conditions.
Figure 7(h) shows CR displacement history from cycle one to nine as an example. All the paths
consist of nine cycles and their displacement amplitudes are increased through the cycles.
4.2 Numerical results and comparisons

Numerical calculations have been carried out using the same material constants for both the return
mapping and the tangential plasticity algorithm. Figure 8,9 and 10 show the hysteresis responses
at the top of the specimen through the cycles. As for the case shown in Figure 4, the
elastoplasticity model with tangential plasticity better estimates the minimum ultimate loads
compared to the ones obtained with the conventional plasticity (T=0). In addition, the numbers of
cycles necessary to reach the critical conditions are different in the two models: the results
obtained by considering the tangential plasticity tend to achieve the ultimate load at an earlier
stage. This means that, under operational loading conditions, the predictions simulated by the
conventional elastoplasticity models may overestimate the structural resistance, with more or less
serious implications for the safety factors design.

(a) BS
(b) BA
(c) XU
Figure 8. Comparisons of the hysteresis loops under different loading

conditions (from (a) BS to (c) XU). Left figures show the responses in the x-
direction, right figures show the ones along the y-direction.

(d) XP
(e) DM
(f) SQ

conditions (from (d) XU to (f) SQ). Left figures show the responses in the x-
direction, right figures show the ones along in the y-direction.

(g) CR

condition (g) CR. Left figure shows the responses in the x-direction, right figure
shows the responses along the y-direction.
5. Conclusions
The paper presented the numerical results based on the extended subloading surface model
incorporating the tangential plasticity effect in order to capture the material response under a non-
proportional loading condition. The algorithm has been used to simulate the behavior of a thin
wall bridge pier subjected to biaxial cyclic loadings. As a result of the comparison between the
conventional plasticity model and the tangential plasticity model against experimental data, it can
be concluded that the latter has higher predictive capability. The conventional plasticity, in fact,
overestimated the ultimate load, whereas the tangential plasticity has been proved to be able to
catch the experimental peak load. Moreover a decrease of the peak loads and an anticipation of the
ultimate loads at earlier stage have been observed whenever the tangential relaxation has been
taken into account for all of the seven bi-directional loading paths. This indicates the importance
of considering the tangential plasticity effect in order to achieve more accurate and reliable
predictions under operational loading conditions. The further validation of material constants for
tangential plasticity is left as a future work.
6. References
1. Abaqus Users Manual, Version 6.13-1, Dassault Systémes Simulia Corp., Providence, RI.
2. Drucker, D. C., “Conventional and unconventional plastic response and representation,” Appl.
Mech. Rev. ASME 41, page 151-167, 1988.
3. Hashiguchi, K., and Tsutsumi, S., “Elastoplastic constitutive equation with tangential stress
rate effect,” Int. J. Plasticity, vol. 17, page 117-145, 2001.

4. Hashiguchi, K., Pfeiffer, F., Wriggers, P., “Elastoplasticity theory,” Lecture notes in applied
and computational mechanics, Springer, Berlin, 2009.
5. Hashiguchi, K., and Mase, T., “Subloading surface Model and its Return-mapping
Formulation,” ACTA 60th Nat. Congr. Of Theoretical & Applied Mechanics, Japan, 2011.
6. Hashiguchi, K., Suzuki, N., and Ueno, M., “Elastoplastic deformation analysis by return-
mapping and consistent tangent modulus tensor based on subloading surface model (1st
report, formulation of return-mapping),” Transaction of the JSME, vol. 80, No. 812, 2014. (in
Japanese)
7. Hoshide, T., “Biaxial Fatigue Life Predicted by Crack Growth Analysis in Various Material
Microstructures Modeled by Voronoi-Polygons,” J. Mater. Eng. Perform., vol. 20, page 1497-
1504, 2011.
8. Itoh, T., Sakane, M., and Ohsuga, K., “Multiaxial low cycle fatigue life under non-
proportional loading,” Int. J. Press. Vess. Pipi., vol. 110, page 50-56, 2013.
9. Nakajima, N., and Yamada, M., “Hysteresis Loop Characteristics of Structural Carbon Steel
SS400 in Very Large Plastic Zones,” J. Struct. Constr. Eng. AIJ, No. 536, page 17-22, 2000.
(in Japanese)
10. Nishikawa, K., Yamamoto, S., Natori, T., Terao, K., Yasunami, H., and Terada, M.,
“Retrofitting for seismic upgrading of steel bridge columns,” Eng. Struct., vol. 20, page 540-
551, 1998.
11. Reis, L., Li, B., and Freitas, M., “Biaxial fatigue for proportional and non-proportional
loading paths,” Fatigue & Fracture of Eng. Mater. Struct., vol 27, page 775-784, 2004.
12. Rudnicki, J. W., and Rice, J. R., “Conditions for localization in pressure-sensitive dilatant
materials,“ J. Mech. Phys. Solids, vol. 23, 371-394, 1975.
13. Simo, J. C., and Oritz, M., “A unified approach to finite deformation elastoplasticity based on
the use of hyperelastic constitutive equations,” Compt. Meth. Appl. Mech. Eng., vol. 49, page
221-245, 1985.
14. Tsutsumi, S., and Hashiguchi, K., “General non-proportional loading behavior of soils,” Int. J.
Plasticity, vol. 21, page, 1941-1969, 2005.
15. Ucak, A., Tsopelas, P., “Load Path Effects in Circular Steel Columns under Bidirectional
Lateral Cyclic Loading”, J. Struct. Eng, ISSN (online): 1943-541X, 2014

Study of mechanical compatibility at the mortar-
block interface in a heritage building and numerical
modeling by Abaqus
H. FIHRI FASSI1 , Y. HAMMOU1, L. VAN PARYS2, M. Bideq3 , T. DESCAMPS2
1. Faculté Des Sciences et Techniques (FST), Route de Casablanca, SETTAT ,Morocco

hichame.fihri-fassi@uhp.ac.ma
2. Université de Mons(UMONS) - Faculté Polytechnique, 20 Place du Parc, MONS ,Belgium
3. ENSA, Safi, Morocco
Abstract :This research work aims at modeling the mortar-block assembly for analyzing the
potential occurrence of mechanical problems at interfaces as the consequence of an eventual
incompatibility. The study considers a heritage building located in Morocco. The authors focus
their attention on a representative part of the building and perform numerical simulations,
studying the stress repartition inside the masonry. For carrying out efficient analysis, the
mechanical characteristics (Modulus of Resistance, Modulus of Elasticity) are collected through
non-conventional testing techniques dedicated to the field of Heritage Conservation that are
briefly described. The scenario considered in the proposed study concerns a bearing wall restored
with a non-adequate mortar. The goal of the modeling works is to compare the repartition of
internal solicitations between the original situation and the restored one in order to quantify the
risk, for materials that should be conserved, associated with the “stress shielding” phenomenon
and to justify the particular attention to be paid for the choice of a specific mortar for
interventions to be carried out on the concerned building.
Key words: masonry, numerical simulation, compatibility, stress shielding, Abaqus
1. Introduction
Masonry has been used for centuries in many places around the world. Most of the master pieces
of our heritage are then erected with this material. For several reasons, these heritage structures
should be preserved in order to be transmitted to following generations. Although they may appear
trivial, restoration works are easily impacted by parasite problems associated with (essentially
mechanical) incompatibilities between conserved and brought materials. These problems affect,
economically and technically, the durability of the operations and constitute a point of interest for
people engaged in structural mechanics.

2. Heritage masonry: specificities, weaknesses and challenge
Erectable with many types of bricks or stones with mortar joints according to varied patterns, the
masonry concept concerns a wide range of building materials that are characterized by
heterogeneous mechanical behavior [1]. Based on old constituents and ancient techniques (even if
the know-how could be of advanced level at their construction period), heritage masonry may
suffer of low mechanical properties still restricted by degradations induced by aggressive
environments or exceptional conditions to which they are exposed [2,3]. In order to preserve
ancient structures, restoration campaigns will generally opt for the replacement of altered parts by
more reliable ones. Such kind of works carried out on masonries may sometimes concern bricks or
stone but, in almost any case, will lead to partial mortar replacement (at least pointing operations).
A specific attention should then be paid in the prescription of the works: contemporary elements
should achieve a sufficient compatibility (mechanical, aesthetical, …) with the ones that are kept.
3. Non-conventional data acquisition on materials in heritage buildings

In order to better document the problems of incompatibility between contemporary mortars and
old masonries, the characterization of mechanical properties is of crucial interest in the approach
of the subject. Nevertheless, extracting great quantities of materials within a structure whose value
justifies the conservation in order to perform conventional characterization tests is a delicate and
discussed spot. For the needs of the present research, the authors rely on experimental devices
which make it possible to collect mechanical data on materials (compressive strength through the
Modulus of Resistance and rigidity through the Modulus of Elasticity) through very few
destructive approaches, namely:
1- “Translational Cutting Test” [3], to obtain the properties of resistance of materials. Not usual in
classical material mechanics, it is widely used in the petroleum industry. It consists in tracing at
the surface of a sample by removing successively thin layers of material with a rectangular
Polymer Diamond Coated cutter of width w, a groove at constant shallow depth d. The cutter is
inclined by an angle and is displaced at constant horizontal velocity v. The magnitude and
inclination of the force F acting on the cutter are accurately measured and recorded during the test.
Thanks to a dedicated phenomenological model, the Modulus of Resistance of the tested material
may be computed from the relationship between the recorded force and the destroyed section. This
technique is characterized by its simplicity, its time efficiency, the measuring accuracy and the
main characters of the test which is relatively not very destructive due to the limited quantity of
the necessary sample for the realization of the test as well as not dependent on the experimental
procedures and analysis compared to the type of studied materials.

Figure 1. Schematic action of the PDC cutter (left) and experimental device at
UMONS (right)
2- “Pulse Sonic Test” [4], to obtain the properties of rigidity of materials. Usually encountered in
classical material mechanics where it is used in many fields, it relies on the well-known
relationship linking the sound velocity inside a material with the dynamic version of the Modulus
of Elasticity. Many experimental devices are commercially available: an emitter sends a sonic
pulse on a given location at the surface of the material, a receiver detects it at another place and
embedded computer-aided facilities allow determining the sound velocity based on the distance
between emitter and receiver.
4. Masonry typology for present study
Typical of systems that may be encountered in Morocco, the masonry pattern constituting the
minaret walls of AL MASJID AL AADAM Mosque in the old medina of Safi town (Figure 2) is
chosen as starting point for the present study, focusing on potential restorations that would affect
the front wall. In its preliminary shape, the proposed parametric study will rely on numerical
simulations carried out on an elementary cell that is representative of the masonry encountered in
the minaret wall (Figure 3).

Figure 2. Al Masjid Al Aadam rebuilt in 1807
Figure 3. elementary cell: front view (left), section (right)
The proposed method is general: once developed, it is directly transposable to any other masonry
pattern concerned by restoration operations where contemporary mortar is brought. The
elementary cell is composed of the conserved elements (block and mortar) and of the new mortar
that is used for pointing both faces of the considered wall after weathered mortar material has been

removed on a 2 cm depth. The preserved blocks are parallelepiped elements of 9 x 6 x 19 cm3 and
all the joints (head joints and bed joints) present a thickness of 1 cm. Due to its position inside the
minaret wall, the cell is assumed to be impacted by a static load pressure of 10 daN/cm2, taking the
weight of above elements (wall and flagstone) into account. For replicating the natural
confinement acting on the cell while circumventing edge effects, rigid plates are placed on the
superior and under the inferior face of the elementary cell with hard contact and friction without
sliding conditions and continuity conditions are imposed on lateral facets of the cell. Numerical
simulations are performed with Abaqus software [5], coupled with Matlab which manages the
parametric study (see below) and controls the Abaqus calculation files. A tridimensional model is
considered. In this preliminary approach, homogeneous and isotropic linear materials are
considered for the simulations although more complicated studies (even with damage for blocks
and/or mortar) are in progress.
5. Simulated scenario
In classical conservation approaches, repointing is a restoration operation that will concern joints
(usual weak point of the structure) once they will have reach a given degree of alteration. The
intervention consists of the digging of the joints manually or with a mechanical tool. According to
their cohesion, the adherence and the state of degradation, the joints are dismantled on 1 to 3 cm of
depth (in our scenario 2 cm) in order to guaranty a satisfying embedding for the new mortar that
will be brought. Before repointing, masonry will be washed by brushing and often vacuum-
cleaned with compressed air or high-pressure water flows depending on the friability of materials.
Once the surface of the joint humidified, the filling must be made following distinct steps to
maximize adherence and avoid any possible shrinking of the new mortar.
Mechanical incompatibility problems associated with repointing operations intervenes in
relationship with the recourse to great resistance and/or great stiffness contemporary mortars
compared to those of old block and mortar. The usage of inappropriate binders (nervous cement,
industrial limes, …) in mortar mixes is the crucial problem of the 20th and 21st centuries. It may
give birth to mortars that are not adapted to the “work” of the old buildings. In the short-term, the
stress shielding phenomena has to be feared as it creates zones with stress increase and can
potentially promote damages in sensitive blocks (Figure 4). In the long-term, a mortar that is too
rigid will not be able to follow the displacement of the structures over the years; tensions will
appear between the blocks and the mortar, mechanically affecting the structural behavior of
masonry.
For helping practitioners to orientate their day-to-day practice, the authors undertook to determine
range of parameters associated with limit state configurations. Based on such boundary
knowledge, it becomes possible to quantify practical risks associated with situations that are faced.
In the proposed parametric study, the fixed morphology of the masonry cell will be considered
under fixed loading but the Modulus of Elasticity of the new mortar will be modified (between 1
and 30 times the MOE of conserved mortar) for several values of MOE for the conserved block
(between 40 and 120 times the MOE of conserved mortar). Such an approach is likely to highlight
the occurrence of the stress shielding phenomenon.

Table. 1 – Mechanical properties of various materials used in the parametric
modeling
Designation of Modulus of Elasticity

Poisson's ratio[-] Density [kg/cm3]
Material [daN/cm2]
Conserved mortar 3 000 0.25 0.0020
Conserved block 40 x 3000  120 x 3000 0.25 0.0020
New mortar 1 x 3000  30 x 3000 0.25 0.0020
Figure 4. Stress shielding scheme: masonry repointed with identical mortar (left)
or too rigid mortar (right)
6. Preliminary Results
With regard to field observations, the present study presents evolution of results around edges of
blocks concerned by the restoration process. The treatment of the results of the simulation are
involved for a local interest volume on the corner of the block, see (Figure 5)

Figure 5. schematic view of the elementary cell and local interest volume for
results visualization
For each pertinent face of the local interest volume, the evolution of stresses associated with
specific damage patterns are plotted in Figure 6, Figure 7 and Figure 8. The maximal tensile stress
on the Stretcher face may be responsible for crack initiations while compressive stresses for Head
and Bed faces could eventually lead to crushing if blocks are composed with weak or weathered
material. The presented results of these preliminary investigations outline the stress shielding
phenomenon may take place under specific range of properties and, further than some boundaries,
no critical evolution should be feared any more. Such considerations may now be used as
guidelines for the setup of advanced simulations that are now in progress, namely taking damage
into account with the Barcelona Model implemented in Abaqus.

Figure 6. Evolution of ratio “max tensile stress / max tensile stress (ref before
restoration)” for Stretcher face (S) of local volume, with the ratio ENM / ECM for
several ECU/ECM values and under fixed loading

Figure 7. Evolution of the ratio “compress stress max / compress stress max (ref
before restoration)” for Header face (H) of local volume, with the ratio ENM/ ECM for
Figure 8. Evolution of the ratio “compress stress max / compress stress max (ref
before restoration)” for Bed face (B) of local volume, with the ratio ENM/ ECM for
7. Conclusions and further works
Although mechanical incompatibility plays a great role in the durability associated to restoration
operations carried out on heritage masonry, the current recourse to trial & error approaches is not a
fatality. The present work proposes preliminary results showing that numerical simulation may be
used for documenting the situation faced by practitioners. Since these results are still partial, no
strong conclusion may be edited. Nevertheless, encouraged by the presented observations, further
investigations are now in progress. They rely on damageable material models for the blocks and
will help establishing practical tools, usable on the building sites for risk estimation.

8. Acknowledgement
The authors would like to thank the Agence Universitaire de la Francophonie as well as Wallonie
Bruxellles International for their financial support that allowed both the authors from Morocco to
join the Civil Engineering Department of UMONS for initiating this research work.
9. References
1. R.PEALES, Modélisation du comportement mécanique par éléments discrets des ouvrages
maçonnés tridimensionnels : contribution à la définition d'éléments de contacts surfaciques,
Thèse de doctorat de l'université de Montpellier II, 23-36, 2007.
2. A.THOIKONOMOPOULOU, Approches numériques pour l’étude du comportement des
structures maçonnées anciennes : un outil base sur le calcul a la rupture et la visualisation
graphique, Thèse de doctorat de Ecole Nationale Supérieure D’architecture De Paris La
Villette Ecole Doctorale «Ville, Transports Et Territoire», Université Paris-Est, 6-9, 2009.
3. F.FOUCHAL, Contribution à la modélisation numérique des interfaces dans les structures
maçonnées, Thèse de doctorat de l’université de Reims Champagne-Ardenne , 43-51, 2006.
4. L. VAN PARYS, F. DAGRAIN, L. LEOSKOOL , The recourse to cutting-based tomography
for quantifying the surface strengthening of altered stones, ICDS12-International conference
durable structures: from construction to rehabilitation Lisbon, Portugal.
5. ABAQUS Version 6-10-1, User documentation. Dassault system Simulia.

Improvement and Enhancement of Concrete
Damage Plasticity Model
Yuji. Nikaido, Yoshinori Mihara, Shohei Sawada and Yoshiyuki
Takahashi
Kajima
Corporation
Abstract: The purpose of this paper is to make an improved concrete damage plasticity
model that focus on a post cracking behavior. First, we constitute the material model
that simulates a process of closing the large tensile cracks and recovering the
compression stiffness. Second, the model is enhanced by adding another feature that
simulates a transition from compression to tension again. Finally, experimental
simulation analyses are conducted for validation of efficiency.
Keywords: Civil Engineering, Concrete, Constitutive Model, Plasticity, Damage, Cyclic

Behavior
1. Introduction
In order to properly assess the safety margins of the reinforced concrete structures
under a cyclic loading, it is necessary to properly simulate damage process of the
concrete; however, concrete damage models in commercial analysis codes are rarely
practicable in terms of a transition behavior from tension to compression, and vice versa.
In Abaqus, elasto-plastic concrete models have been introduced based on plastic
damage theory from ver.6.3. Though they have enabled elasto-plastic evaluations under
a monotonic loading in practical use, it is hard to say that they can simulate an
appropriate behavior under a cyclic loading.
2. Constitutive Law of Concrete

2.1 Isotropic Damage Plastic Theory
In the elasto-plastic analysis of reinforced concrete structures, it is necessary to

appropriately configure material properties of concrete, rebar and bond
relationship of them.
In particular, material properties of concrete are complex, because compression,
tensile and shear behavior is respectively different.
Since Abaqus6.3, isotropic plastic damage model has been introduced. It is
characterized by putting damage into a scalar value that does not depend on the

direction. According to the model, damage can be captured by the entire state
including tensile, shear, and compressive failure.
Figure 1. Image of reinforced

concrete structures
Uniaxial characteristics in tension and compression in the material models are

shown in Figure 2. On tension side, a stress-strain curve follows the linear elastic
relationship E0 until it reaches the value of the fracture stress σt0 corresponding to
the generation of micro-cracks in the concrete. Once it exceeds the fracture stress,
the formation of micro-cracks is expressed by the tension softening response. On
the other side, under uniaxial compression stress strain response, a characteristic
is in accordance with the linear elastic relationship E0 until the stress reaches the
initial yield stress σc0. Response in the plastic zone is generally characterized by
softening from the ultimate stress σcu.
According to the test results, it has been found that the slope of the unloading
response is reduced, when it is unloaded from any point on strain softening
portion of the strain curves. In the isotropic plastic damage theory, its reduction of
the elastic stiffness of the material is treated as damage variable. Damages are
represented by two damage variables dt and dc as the reduction factor (zero: no
damage to one: fully damage) of the elastic stiffness tension and compression
side. These variables can be a function of plastic strain, temperature, and other
field variables.

tensile stress compressive stress
σt σc
(a) (b)
σt0
σcu
σc0
E0
E0
(1-dc)E0 compressive strain
tensile strain εc
(1-dt)E0
εt
pl el pl el
εt εt εt εt
Figure 2. Uniaxial behavior of

concrete in compression and tension
side
2.2 Behavior under cyclic load
It is known that reinforced concrete recover stiffness even if tensile cracking and stiffness
reduction occurs in the tension side. It is result from closing crack again in the process of
stress is shifted to the compression side. In order to simulate the crack closing process, the
stiffness recovering coefficient of wt and wc are set for Abaqus.
Figure 3 is a schematic diagram of a process of a stiffness recovery shown in the Abaqus

theory guide [8]. Thus, user can set a stiffness recovering behavior by using a coefficient wc
(0 ~ 1) how the tension damage involve the stiffness at the transition to compression side.
And it is the same for a transition back to the tension side.
Figure 3 shows default settings of Abaqus that represents stiffness recovering at the transition
from tension to compression side and damage cumulating at the transition from compression
to tension side.

σt
σt0
E0
wt=1 (1-dt)E0
wc=0
(1-dc)E0 (1-dc)(1-dt)E0 wc=0 ε

wc=1
E0
Figure 3. Uniaxial behavior of

concrete under cyclic loading
2.3 Problem of current concrete model and proposal of new model
In certain situations, such as shown in Figure 3, a constitutive law of concrete in

Abaqus properly captures the macroscopic properties of the concrete. However, in
case of a large tensile strain shown in Figure 4, stiffness recovery occurs earlier
than expected. It can be attributed to the current material model that judges a
transition from tension to compression stiffness recovery by zero stress point.
tensile strength
Figure 4. Transition behavior with

large tensile strain in current material
model

Figure 5 shows an example of behavior of concrete by actual cyclic loading test
in [19]. After receiving large tensile stress as shown in the figure, the concrete
stiffness reduces as expected in accordance with the degree of damage. Thereafter,
through a process of closing the cracks, the concrete gradually recovers the
stiffness by interlocking on crack surfaces.
In order to solve the problems shown in Figure 4, we propose a new material
model in which compressive stiffness recovery criteria is modified from a zero
stress point to a zero strain point. This change enables to simulate the crack
closing process that is quite similar to the actual behavior of the concrete such as
Figure 5. It is intended that the reduced tensile stiffness is mainly dominant
during the tensile strain state and the compressive stiffness become dominant
from transition to the compressive strain.
Tensile s tress
Tensile
di splacement
(μm)
(Actual behavior in an experiment) (Proposed material model)

Figure 5. Actual Behavior of concrete and
proposed model

3. Verification Analysis
In this section, verification analyses are performed to check the operation of proposed
material model.
3.1 Material Properties

Concrete material characteristics for the study are shown in Table 1. It is the
same properties as verification of the concrete material constitutive law in [2].
Table 1. Concrete Material Characteristics
Young’s Modulus 33.8 GPa
Compressive Strength 30.9 MPa
Strain at -3
2.2×10
Compressive Strength
Tensile strength 1.57 MPa
Strain at -5
4.6×10
Tensile Strength
Poisson’s Ratio 0.2
Concrete tension stiffening and tension damage characteristics are set according
to Eq.(1) and Eq.(2). It was proposed by Izumo et al [5] and has been verified
through the dynamic simulation analysis of shear walls [4].
c
ε 
σ t = σ cr  cr  (1)
 εt 
d
ε 
dt = 1 −  cr  (2)
 εt 
Here,
σ t : Tensile Stress σ cr : Tensile Strength ε t : Tensile Strain ε cr : Strain at Tensile Strength c = 0.4 d = 1.0

Similarly, Concrete compression hardening and compression damage
characteristics are set according to Eq.(3) and Eq.(4). Concrete compression
hardening and compression damage characteristics follow the previous report [2],
which is based on [21] and [7].
E ε ε 
2
E ε 
σ c =  0 c −  c  1 +  0 − 2  c  f cm (3)
 E cl ε cl  ε cl   E cl  ε cl 

σc
dc = 1 − (4)
E0 (ε c − ε p )
Here,
σ c : Compressive Stress f cm : Compressive Strength
ε c : Compressive Strain ε cl : Strain at Compressive Strength
E0 : Young ' s Modulous Ecl : Secant Modulous at f cm ε p : Plastic Strain
40
Compr essi ve St r ess( N/mm2)
30
20
10
0
0 0. 002 0. 004 0. 006 0. 008 0. 01
Compr essi ve St r ai n εt
1
0. 75
Damage f act or dc
0. 5
0. 25
0
0 0. 002 0. 004 0. 006 0. 008 0. 01
Pl ast i c St r ai n εpl
- Compression -

2
Tensi l e St r ess( N/mm2)

1. 5
0. 5
0
0 0. 002 0. 004 0. 006 0. 008 0. 01
Tensi l e St r ai nεt
1
Tensi l e damage f act or dt
0. 8
0. 6
0. 4
0. 2
0
0 0. 002 0. 004 0. 006 0. 008 0. 01
Pl ast i c St r ai nεcr
- Tension -
Figure 6. Concrete properties for
analysis
3.2 Conditions of Analysis
Analysis for verification is carried out in in the way shown in Figure 7.
An element for the analysis is the reduced integration solid elements (C3D8R).
STEP1 STEP2 STEP3

Tension Unload Compression
Figure 7. Procedure of analysis (half

cycle)
3.3 Results of Analysis
The results of the verification analysis by material conditions of the previous
section are reported.
3.3.1 Tension to Compression
For the basic verification, Analysis starts from a half cycle study.
Figure 8 shows the stress-strain relationship in the analysis by proposed and
current material model. It appears that the left shows a rigid recovery
behavior at zero strain proposed in 2.3. The right shows the analysis results
from a current model of the same material and analysis conditions for
comparison, but it result in unrealistic transition to the compression side
rigidity stress zero point.
2 2
1 1
0 0
-1 -1
-2 -2
0 0. 002 0. 004 0. 006 0. 008 0. 01 0 0. 002 0. 004 0. 006 0. 008 0. 01
Tensi l e St r ai nεt Tensi l e St r ai nεt
(Proposed model) (Current model)

Figure 8. Results of analysis
(half cycle)

3.3.2 One cycle
Figure 9 shows the stress-strain relationship in the one cycle analysis by

new material model. It appears that proposed model properly work on the
compressive damage area after recovering stiffness. Moreover, it also
appears that stress-strain relationship transmits tensile damaged stiffness at
the zero-stress point such as a transition of current material model shown in
Figure 3.
-10
St r ess( N/mm2)
-20
-30
-0. 006 -0. 004 -0. 002 0 0. 002 0. 004 0. 006 0. 008
enlarged St r ai nε enlarged
2
0
-10
1
St r ess( N/mm2)
St r ess( N/mm2)
-20
0
-30
-1
-0. 006 -0. 005 -0. 004 -0. 003 -0. 002 -0. 001 0 -0. 004 -0. 002 0 0. 002 0. 004 0. 006 0. 008
St r ai nε St r ai nε
Figure 9. Results of analysis (one cycle)

3.3.3 Two cycles
Figure 10 shows the stress-strain relationship in the two cycle analysis by

new material model. It is lucid that proposed model follows current material
model under partitioning tensile strain. There are not sufficient data in such
regions, however, the current material model seems realistic and the impact
on an overall behavior would be limited.
-10
St r ess( N/mm2)
-20
-30
-0. 006 -0. 004 -0. 002 0 0. 002 0. 004 0. 006

enlarged St r ai nε enlarged
2
0
1. 5
-10
St r ess( N/mm2)
St r ess( N/mm2)
1
-20
0. 5
-30
0
-0. 006 -0. 005 -0. 004 -0. 003 -0. 002 -0. 001 0 -0. 004 -0. 003 -0. 002 -0. 001 0 0. 001 0. 002
St r ai nε St r ai nε
Figure 10. Results of analysis (two cycles)

4. Summary
In order to properly assess the safety margins of the reinforced concrete structures
under a cyclic loading, we propose a new material model in which compressive stiffness
recovery criteria is modified from a zero stress point to a zero strain point. This change
enables to simulate the crack closing process that is close to the actual behavior of the
concrete. Though much still remains to be unsettled about a behavior of damaged
concrete, I believe we made some contributions to simulate it. Further consideration will
be needed to yield any findings toward the practical use.
5. Acknowledgments
The authors wish to acknowledge deeply grateful to Dr. Shashwat Sinha, Senior
Application Engineer of Dassault Systems Simulia Corp. who form a model and give us
carefully considered feedback and valuable comments.

An Easy Procedure for Anisotropic Non-Linear
Behavior of Short-Fiber-Reinforced Plastics
Sascha Pazour, Wolfgang Korte, Marcus Stojek
PART Engineering GmbH, Germany
For fiber reinforced parts the consideration of anisotropic material behavior is required to
receive reliable results. In the scope of this fact a procedure is described how to consider these
effects in terms of process-structure interaction and how to achieve possible benefits such as
weight reduction and shorter development cycles shown for examples of the industry .
The developed procedures show how to consider the anisotropic mechanical behavior of injection
molded short-fiber-reinforced plastics parts in FE analysis. It is shown how the existing
information, which is provided by injection molding simulation software, can be processed and
transferred into mechanical simulation models. The procedure is outlined with practical
applications.
Keywords: Composites, Fiber-Reinforced, Injection molding, Process-Structure-Interaction,

Anisotropic, Hill´s Potential
1. Introduction
It is evident that due to the anisotropic behavior of short-fiber-reinforced parts the structural
analyst is interested in considering these effects. Such a consideration may result in a more
accurate prediction of the mechanical behavior of the investigated component or part.
In the following article, a strategy will be described to predict the elastic-plastic anisotropic
behavior of injection-molded short-fiber-reinforced plastics based on the theory of Hill.
2. General
The injection molding process causes a local anisotropic material behavior within the part due to
fiber orientations in the short-fiber-reinforced plastic. These orientations vary not only at different
locations in the part but also across the wall thickness. The laminar flow of the melt in
combination with the velocity profile in the flow channel lead to characteristic fiber orientations
across the wall thickness with fibers mainly aligned parallel to the flow direction in the outer so-
called shear layers and fibers mainly aligned transverse to the flow direction in the mid layer.

Figure 1. Fiber Orientation Across the Wall Thickness of an Injection Molded Part
2.1 Micromechanics of Short-Fiber-Reinforced Plastics

In the past a lot of researches dealt with micromechanical modeling of fiber-reinforced plastics. In
this paper we consider the theory of Tandon and Weng. Tandon and Weng developed an analytical
transversal isotropic approach. They derived closed analytical equations from Eshelby's theory of
ellipsoidal inclusions. Their model can be extended also for the orthotropic case but remains
purely elastic. The model provide the elastic response of the composite in different directions by
assuming that the composite behaves as a continuum. The individual behavior of the different
phases of the material - fiber and matrix - cannot be distinguished anymore.
However for a lot of cases in practical applications these assumptions do not restrict the usability
of this model in order to make reasonable statements with regard to the part stiffness and design
limit.
2.2 Consideration of Fiber Orientation Distributions
In real injection molded short-fiber-reinforced parts there are fiber orientation distributions instead
of ideally aligned fibers and because of the fact that in a finite element model the element size is
never small enough to include only one fiber the material model has to have properties referring to
that distribution. It is evident that for nearly ideal aligned fibers different mechanical properties
result than for nearly un-oriented fibers. This has to be considered in the definition of the
anisotropic material properties.

Figure 2. Fiber Orientation and Degree of Orientation in an Injection Molded Plastic Example Part
For this a so-called orientation averaging can be applied (Advani and Tucker). The mechanical
properties determined for the idealized unidirectional volume element are weighted with their
local degree of orientation. For a narrow distribution the degree of orientation is close to unity and
for a very broad distribution close to zero (quasi-isotropic). The information about the local
degrees of orientation can be extracted from the orientation tensor which is provided for every
element as standard output by most injection molding solvers.
Figure 3. Orientation Tensor Example

3. Fundamentals
The analysis and interpretation of results obtained by means of an anisotropic analysis is generally
carried out otherwise than it is done in an isotropic analysis. In particular, the evaluation of all
stress and strain invariants such as principal stresses and strains or the von Mises stress is not
meaningful in an anisotropic analysis. Because all components of the stress and strain tensor here
refer to local element coordinate systems that are generated by the software CONVERSE.
Accordingly, only the components of the stress or strain tensor should be evaluated. Since by
definition in CONVERSE, the 11-direction is always aligned with the main fiber orientation
direction, the individual tensor components e.g. σ11, σ22, σ33 therefore also always relate to the
fiber orientation directions within each element. For example, for fiber orientations transferred
from surface or midplane injection molding models, the stress component σ11 of the composite is
the stress in fiber direction. The stress component σ22 is the stress component of the composite
transverse to the fiber direction in the plane of melt flow. The stress component σ33 is the stress
component of the composite transverse to the fiber direction transverse to the plane of melt flow
(i.e. pointing in direction of the wall thickness). In a surface or midplane injection molding
simulation fiber orientations in the out-of-flow plane do not exist by definition. The melt flow
plane is assumed parallel to the local part surface respectively cavity wall.
In the case of fiber orientations transferred from 3D injection molding models, the stress
components σ22 and σ33 cannot further distinguished with regard to their alignment to the cavity
wall. Here, the fibers can be oriented in all spatial directions (melt flow respectively fiber
orientation in wall thickness direction is possible).
3.1 Stiffness Assessment

The stiffness of a part calculated with the linear-elastic anisotropic material model applied in
CONVERSE has to be considered as base stiffness. At which base stiffness is being understood as
the stiffness of the part in the approximately linear-elastic range. If the load on the part is such that
the stress in the supporting cross-sections of the structure does not exceed the nearly linear-elastic
range the calculated base stiffness should reflect the real stiffness of the part within the general
model accuracy. This should be valid even if local stress concentrations (e.g. in notched areas)
exceed the linear elastic limit significantly and would undergo plastic deformations in reality as far
as this yielding does not affect the global deformation of the part.
Also for dynamic oscillating loads and in modal analyses the calculated base stiffness should
match the real behavior within the overall model accuracy. On the one hand all mechanical solvers
linearize at the operating point the response of the structure anyway for dynamic analyses in the
frequency domain. And on the other hand the oscillation amplitudes are comparatively small, so
that the assumption of a linear material response is justified anyway as far as the structure is not
subjected to an initial load, so that the operating point is shifted into the nonlinear range of the
stress-strain characteristic. The limitation of the linear-elastic material model that is applied in
CONVERSE is for most practical cases acceptable. Since, fiber-reinforced plastics behave as
opposed to unreinforced grades over a relatively large range approximately linear. Furthermore

plastics parts undergoing long term loads or dynamic loads should anyway not stressed to the
nonlinear range. Stresses should be kept well below the fracture or yield point and should remain
in the nearly linear-elastic range. Hence, the tabulated ultimate failure values of the material, in the
context of a conservative part design, have to be reduced accordingly and can then be compared
with the calculated linear stresses and strains.
3.2 Strength Assessment

If a strength assessment has to be done based on an anisotropic analysis the calculated stresses
respectively strains have to be compared with appropriate failure indicators. The assessment of
failure has to be done by distinguishing the local fiber orientations. Accordingly, different failure
indicators parallel and transverse to the fiber orientation have to be determined. As an estimate for
the strength of the composite parallel to the fiber direction (σ11) the strength of the composite (σB
or σS) from material Databases such as CAMPUS Plastics may be used. As an estimate for the
strength of the composite transverse to the fiber direction (σ22) the strength of the matrix material
of a corresponding unreinforced grade may be used. These values are of course only rough
estimates. More reliable failure limits can be obtained by using injection molded test plaques with
a film gate in order to achieve aligned fibers along the side of the plaque. From these plaques
specimen can be taken for tensile tests parallel and transverse to the fiber orientation.
4. Elasto-Plastic Behavior of Short-Fiber-Reinforced Plastics
Even though in most cases linear elastic material behavior is enough for several analysis tasks in
some cases an elastic-plastic formulation is required. Fortunately with a new approach this
material behavior can be given by a built in material model in structural simulation software.
Therefore a material formulation in a user subroutine (UMAT) is not necessary anymore.
4.1 The Hill´s Potential

The *POTENTIAL command is able to define stress ratios for anisotropic yield and creep
behavior. It can be used in conjunction with the *CREEP or *PLASTIC option. Also rate
dependency can be considered. Here only the Plastic option is used and explained.
Anisotropic yield behavior is created through the use of yield stress ratios Rij In the case of
anisotropic yield the yield ratios are defined with respect to a reference yield stress σ0.
The Hill's potential function is an extension of the von Mises function. In a rectangular Cartesian
system it can be expressed as:
The components F, G, H, L, M, N are defined as:

where each σij is the degree of orientation dependent yield stress value. σ0 is the user-defined
reference yield stress.
The Rij values are the anisotropic yield stress ratios and the τ0 can be evaluated with σ0/0,577.
Therefore the six yield stress ratios are defined as follows.
These values have to be defined for each degree of orientation in the material card.
Obviously the *POTENTIAL command is not able to consider different behavior between tensile
and pressure loads.
In the following picture the broad distribution of the nonlinear behavior of short-fiber-reinforced
materials is shown. 1 is the highest and 0 the lowest stiffness and strength possible.
With considering this effects such materials can be simulated a lot closer to reality then by using
isotropic material behavior (only one curve).

Figure 4. Plastic Behavior for Different Fiber Distribution
5. Workflow
To assign fiber orientation, the fiber distribution and non-linear anisotropic material behavior to a
FE-model the use of a software tool seems to be the best choice.
There are several software tools available. Some are direct interfaces from the injection molding
solver and therefore only support this specific solver and others are third-party products like the
one used for this example (CONVERSE). The workflow can be seen as a general workflow for
this type of simulation, no matter what software is used.
Figure 5. Workflow

The only material data required to create such a material is a stress-strain curve of your composite
material (e.g. from Campus Database).
6. Example Results
Unfortunately the created values for different degrees of orientation cannot be validated on Test
results since it is not possible to produce test specimen with only one certain fiber direction and
distribution. Nevertheless it is possible to simulate test specimen with all kinds of orientation and
distribution in it and the physical max. and min. values can be checked by comparing the result
non-linear behavior of the degree of orientation 1 and 0 with the behavior of a two phase
representative volume element. The RVE shows the behavior of a 100% orientation of fibers in
and perpendicular to the fiber direction.
Figure 6. Tensile Test Example (Material PBT GF20)

Figure 7. Representative Volume Element (Material PBT GF20)
7. Abaqus Syntax commands

Orientation information has to be given by local orientation systems at each element by using a
*DISTRIBUTION card and assigned through an orientation card *ORIENTATION,
DEFINITION=COORDINATES
The fiber distribution is given by orientation tensor values as field variables, where there are only
the values in and across the main fiber direction required because of the fact that the sum of all
three is always 1 *INITIAL CONDITIONS, TYPE=FIELD, VARIABLE=1.
The material properties have to be provided different in the linear-elastic case and the elasto-
plastic case. In the linear elastic case the material properties are defined by using the
*ELASTIC,TYPE=ENGINEERING CONSTANTS,DEPENDENCIES=number command.
The elastic-plastic anisotropic material behavior is possible through the *POTENTIAL command.
All these formulations are supported by the Software CONVERSE.
8. Summary and Prospect

With the outlined procedure it is possible to obtain reliable results for structural simulation of
short-fiber-reinforced plastic components.
The anisotropic nonlinearity can be considered with existing Abaqus Syntax commands without
any user subroutine. That makes the simulation a lot faster and robust.
The simulation speed is even the same as for isotropic material definition.

With existing methods the opportunity is given to simulate nonlinear anisotropic temperature
dependent material behavior with a minimum of effort.
Even strain rate dependency, creep and damage behavior can be added on top.
But this would have been leaded too far for this documentation.
9. References
1. Abaqus Analysis User’s Manual, Version 6.12-2, Dassault Systémes Simulia Corp.,
Providence, RI.
2. Korte, W., Stojek, M., “Entwicklung einer Software zur orientierungsgradabhängigen
Kalibrierung eines anisotrop elasto-plastischen Standard-Materialmodells zur FEM-Simulation
von kurzfaserverstärkten Kunststoffbauteilen” unpublished report ZIM Koop research and
development project No. KF3056701RR2, Bergisch Gladbach, Germany, 2015.
3. Stommel, M., “Entwicklung und Umsetzung einer hybriden Simulationsmethodik zur
effizienten Berechnung und Auslegung von kurzfaserverstärkten Kunststoffbauteilen”
unpublished report ZIM Koop research and development project No. KF2214510RR2,
Saarbrücken, Germany, 2015.
4. CAMPUS website, www.campusplastics.com
5. PART Engineering GmbH website, www.part-engineering.de
6. M. Stommel, M. Stojek, W. Korte, „FEM zur Berechnung von Kunststoff- und
Elastomerbauteilen“, Carl Hanser Verlag, 2011
7. CONVERSE Documentation, CONVERSE 3.6 Manual, Feb. 2015
8. Tandon, G.P. and Weng, G.J, "The Effect of Aspect Ratio of Inclusions on the Elastic
Properties of Unidirectionally Aligned Composites", Polymer Composites, Vol. 5, Nr. 4, John
Wiley & Sons, 1984
9. Eshelby, J.D. "The Determination of the Elastic Field of an Ellipsoidal Inclusion and Related
Problems", Proceedings of the Royal Society of London 241, No. 1226, 376-396, Bir-
mingham, 1957
10. Advani, S.G. and Tucker, C.L. "The use of tensors to describe and predict fiber orientation in
short fiber composites", J. of Rheology Vol. 31 (1987) No. 8, p. 751-784
10. Acknowledgment
Part of the investigations in this paper are sponsored by Federal Ministry of Economics and
Technology due to a resolution of German Federal Parliament in the scope of the funding program
„ZIM - Zentrales Innovationsprogramm Mittelstand “
Teile der Untersuchungen in dieser Veröffentlichung sind gefördert durch das Bundesministerium
für Wirtschaft und Technologie aufgrund eines Beschlusses des Deutschen Bundestages im
Rahmen des Förderprogramms „ZIM - Zentrales Innovationsprogramm Mittelstand“

Experience and Lessons Learnt from the
SMART 2013 Benchmark
M.J. Crespo, J. Rodríguez, L. Lacoma, F.
Martínez and J. Martí
PRINCIPIA Ingenieros Consultores
Abstract: The French CEA, together with EDF and the IAEA, recently organised an
international benchmark to evaluate the ability to model the mechanical behaviour of
a typical nuclear reinforced concrete structure subjected to seismic demands.
The participants were provided with descriptions of the structure and the testing
campaign; they had to propose the numerical model and the material laws for the
concrete (stage #1). A mesh of beam and shell elements was generated; for modelling
the concrete a damaged plasticity model was used, but a smeared crack model was
also investigated.
Some of the initial experimental results, with the mock-up remaining in the elastic
range, were provided to the participants for calibrating their models (stage #2).
Predictions had to be produced in terms of eigen-frequencies and motion time
histories. The calculated frequencies reproduced reasonably the experimental ones;
the time histories, calculated by modal response analysis, also reproduced adequately
the observed amplifications.
The participants were then expected to predict the structural response under strong
ground motions (stage #3), which increased progressively up to a history recorded
during the 1994 Northridge earthquake, followed by an aftershock. These results were
produced using an explicit solver and a damaged plasticity model for the concrete,
although an implicit solver with a smeared crack model was also investigated.
The paper presents the conclusions of the pre-test exercise, as well as some
observations from additional simulations conducted after the experimental results
were made available.
Keywords: Concrete, damage, time history.
1 Introduction
There are considerable uncertainties involved in the prediction of the response of reinforced
concrete structures to seismic events. To advance the knowledge and capabilities in that regard,
the French Commissariat á l’Énergie Atomique (CEA), Électricité de France (EDF) and the
International Atomique Energy Agency organised the international benchmark named the SMART
2013 project.

The exercise involves a number of shaking table tests carried out in a reinforced concrete mock-
up, as well as predictive calculations regarding the outcome of those tests. Principia participated in
the project by conducting the necessary calculations for three of the four stages and the present
paper summarises the findings.
2 Description of the model (Stage #1)

Both the geometry and material properties assigned to the model were described in Stage #1.
However some changes were introduced in the course of the project, as suggested by the results
already obtained or by requirements of the analyses.
2.1 Geometry
The structure is a 1/4th scaled mock-up, with a trapezoidal shape, three stories, of reinforced
concrete. It is representative of a typical, simplified half part of an electrical nuclear building. The
geometry is with the aim of studying the torsional behavior.
The mock up has a total height of 3.1 m, a width of 3 m and the two lateral walls have a width of
2.55 m and 1.05 m respectively. All the lateral walls and the three floors have a thickness of 10
cm. There is also a central pillar with a square section of 10 cm x 10 cm. Figure 1 shows a plan
view and a lateral view of the mock-up. This two views together with the 3D ones that will be
presented from the finite element mesh provide a complete idea of the mock-up geometry.

Figure 1 Plan and lateral view of the mock-up.
2.2 Materials
The mock-up involves two materials: concrete and reinforcing steel. In stage #1, the material tests
were not yet available and the material properties used in the analyses were conventional design
values.
There are several material models in Abaqus [SIM13] that are suitable for modelling concrete.
That selected was the “smeared crack concrete model” available in the implicit code
Abaqus/Standard. It provides a general capability for modelling concrete in all types of structures,
including beams, trusses and solids. The model consists of an isotropically hardening yield surface
that is active when the stress is dominantly compressive and an independent “crack detection
surface” that determines if a point fails by cracking. It uses oriented damaged elasticity concepts
(smeared cracking) to describe the reversible part of the material’s response after cracking failure.
A parabolic stress-strain law was used, with an initial modulus of 32 GPa, a peak stress of 30 MPa
at 0.22% strain, and an ultimate strain of 0.35%, as presented in Figure 2. The tensile strength is
2.4 MPa with an assumed fracture energy of 180 J/m2. The resulting biaxial behaviour is shown in
Figure 3.

35
30
25
Stress (MPa)
20
15
10
0
0.00% 0.10% 0.20% 0.30% 0.40%
Strain
Figure 2 Uniaxial behaviour of concrete
Due to convergence problems experienced in some of the runs in Stage #3, a decision was made to
change from implicit to explicit integration, using Abaqus/Explicit. Since this module does not
support the former material model, the “concrete damaged plasticity” model was used instead. It
combines non-associated multi-hardening plasticity and scalar (isotropic) damaged elasticity to
describe the irreversible damage that occurs during the fracturing process. It is meant to provide an
accurate approach for modelling cyclic events, allowing the user control over the stiffness
recovery effects.

0.2
0.0
-0.2
-0.4
Normalised stress (-)
-0.6 Elastic
Failure
-0.8
-1.0
-1.2
-1.4
-1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2
Normalised stress (-)
Figure 3 Biaxial behaviour of concrete
The reinforcing steel was modelled as an elastoplastic material with isotropic (von Mises)
plasticity. Some experimental information had been provided, namely the Young’s modulus, the
elastic and shear strengths, and the ultimate strain. Although the simulations of the material tests
were performed with reinforcing bars embedded in the concrete elements, those of the mock-up
used the “rebar layer” technique, which does not consider individual bars but the averaged
stiffness effects on the concrete elements.
2.3 Finite element mesh
The mesh used to represent the mock-up consists mainly of shell elements (8.559); a few (230)
linear beam elements were also used for the pillar and the three beams on the floors.
The shaking table has 10,253 elements, as in the mesh provided by the organising committee

(OC). The total number of nodes is 20,247. Figure 4 presents two views of the mesh.
Figure 4 General views of the finite element mesh

3 Linear Dynamic Analysis (Stage #2)
In Stage #2 linear dynamic analyses were performed. The first activity consisted in conducting
modal analyses under three different hypotheses:
• Mock-up fixed at its base with no additional masses.
• Mock-up fixed at its base and loaded with additional masses.
• Mock-up fixed to the shaking table and loaded with additional masses.
The frequencies corresponding to the first three natural modes calculated for each of the three
above hypotheses appear in Table 1.
The first three modal shapes are similar under the three hypotheses. The first mode is the main one
in the X-direction (see axes in Figure 4), the second one is the main one in the Y-direction, and the
third one is a torsional mode around the vertical Z-axis. Figure 5 presents the three modal shapes
for the third hypothesis.
The frequency differences observed between the first two hypotheses are consistent with the
increase in the global mass of the model. The differences between the second and third hypothesis
arise from the flexibility of the shaking table.
OC provided the participants with experimentally determined frequencies, which are also
indicated in Table 1. The differences between the calculated and experimental frequencies are
probably due to the material properties which, at this stage, had to be given standard values;
experimentally determined properties only became available later.
Table 1 Calculated and measured frequencies (Hz) under three hypotheses
Mode 1 Mode 2 Mode 3
Fixed base, no masses 19.9 35.4 65.4
Fixed base, with masses 8.40 15.4 29.7

Fixed to shaking table, with masses 5.65 8.76 19.0
Experimental 6.28 7.86 16.5

Figure 5 Modes of the mock-up fixed to shaking table and loaded with masses
This stage also included a linear time domain analysis for two time histories (identified as Runs 6
and 7), both with a PGA of 0.10g. Run 6 is a synthetic white noise, while Run 7 is a scaled design
signal.
The input motion was introduced in the calculation by imposing displacements at the actuators,
i.e.: creating an additional node linked with a distributing coupling to the four nodes of the
shaking table surrounding the actuator. This type of connection distributes forces but does not
introduce any stiffness.
Modal dynamic analyses were conducted for the different base motions and associated input
motions at the actuators. The damping was assigned on a mode-by-mode basis, using the
information provided in the experimental report by OC: 2.6% for the first mode, 4.2% for the
second, and 5.5% for the third one. A damping ratio of 6% was assigned to all the higher modes.
The results were presented in terms of absolute displacements and accelerations.
In general, for this stage the results obtained were of the same order as those provided by OC.
Space limitations do not allow an exhaustive comparison, but a sample result has been selected for
display, which is the horizontal X-displacement of point A in floor 3. Figure 6 provides the
comparison of calculated and measured results for Run 6. The rest of the comparisons presented in
the paper correspond to that same location.
4 Nonlinear Dynamic Analysis (Stage #3)

Convergence problems were experienced with Abaqus/Standard for some of the runs
contemplated in this stage. The concrete cracking caused a sudden release of the internal energy
inducing a reduction of the time increment to the order of 10 µs. The convergence could be
improved by changing the concrete fracture law or by adding numerical damping but, since such
strategies would alter the structural behaviour, a decision was made to change to the explicit solver

in Abaqus/Explicit [SIM13]. The time increment is now around 2 µs, smaller than in
Abaqus/Standard but much faster per increment since no matrix needs to be inverted or factorised.
Besides, the increment is almost not reduced during the simulation.
As mentioned earlier, this required changing the material model. The main problems posed by the
new formulation of the concrete behaviour are that the reversals are somewhat stiffer than
expected and that, with increasing numbers of cycles, the equivalent plastic strain may ratchet up
with little physical basis.
Some results are presented below for all the mandatory runs, which are no. 9, 13, 17, 19, and 23.
All the results included here correspond to the same X-displacement of point A in floor 3. For
each of the five runs mentioned, the comparisons appear in Figure 7 through Figure 11.
The large displacement part of the response is considered to be generally satisfactory in all cases,
perhaps particularly in Runs no. 13, 17 and 19. However, there are important differences in the
amplitudes of the oscillations during the second half of the history which, although their
significance may not be very important, they are still not easy to explain.
Experimental Calculated
0.003
0.002
0.001
displacement (m)
-0.001
-0.002
-0.003
0 10 20 30 40 50
Time (s)
Figure 6 Calculated and measured X-displacements for point A, floor 3, Run 6

0.003
0.002
0.001
displacement (m)
-0.001
-0.002
-0.003
0 1 2 3 4 5 6 7 8 9
Time (s)

0.015
0.01
0.005
displacement (m)
-0.005
-0.01
-0.015
0 2 4 6 8 10 12 14 16
Time (s)

0.03
0.02
0.01
displacement (m)
-0.01
-0.02
-0.03
0 2 4 6 8 10 12 14 16
Time (s)
0.05
Experimental
Calculated
0.03
displacement (m)
0.01
-0.01
-0.03
-0.05
0 2 4 6 8 10 12 14 16
Time (s)

0.01
Experimental
Calculated
0.005
displacement (m)
-0.005
-0.01
0 1 2 3 4 5 6 7 8
Time (s)
5 Conclusions
Calculations were conducted for the first three stages of the International Benchmark SMART
2013, which involve predictions of the response of a mock-up in a series of shaking table tests. As
a result of the work conducted, the following conclusions can be offered:
1. The calculated modal frequencies reproduce only moderately those determined
experimentally. The differences are partly attributed to the fact that, at the time,
standard rather than experimentally determined material properties had to be used.
2. The low-amplitude time history analysis did provide a reasonable match of the
experimentally determined response.
3. The strong non-linearities and softening caused by the large-amplitude motions in
stage #3 gave rise to convergence problems in Abaqus/Standard which could not be
satisfactorily solved while maintaining the material description. This caused a
migration to Abaqus/Explicit as solver, which also entailed moving from a smeared
crack model to a damage plasticity model for the concrete.
4. Finally, the calculations performed in stage #3 agreed reasonably well with the
experimental results at least for the large displacement part of the response, though
the amplitude of the oscillations in the second half of the history is somewhat less
satisfactory.

6 References
[SIM13] SIMULIA (2013) “Abaqus Analysis User’s Guide”, Version 6.13,

Providence, Rhode Island.
[RIC13] Richard, B. and Chaudat, T. “Presentation of the SMART 2013
International Benchmark”, Document no.
DEN/DANS/DM2S/SEMT/EMSI/ST/12-017/E, 9 September.

Advanced Finite Element Analysis of
Deep Excavations using Abaqus
Yuepeng Dong1, Harvey J. Burd2, Guy T. Houlsby2, and Andrew J. Whittle3
1, Singapore-MIT Alliance for Research and Technology, previously University of Oxford
2. University of Oxford, UK
3, Massachusetts Institute of Technology, US
Abstract: Deep excavation, widely used for underground constructions, is a very complex soil-
structure interaction problem in geotechnical engineering. Its performance is influenced by the
sophisticated soil behavior, complex retaining structures, and variable construction sequences. As
the excavation goes deeper and larger in scale, challenges arise for both practical applications
and research. Finite element analysis is an effective tool to study its mechanism, which can
consider both geotechnical and structural aspects. It provides necessary information on the
performance of deep excavations for design purpose, and can also be used to predict the behavior
of deep excavation and provide guidance for the construction. The main issues involved with the
finite element analysis of deep excavations include selecting appropriate constitutive models for
soils and structures, the simulation of the construction procedure, and the modelling of the
soil/structure interface. These issues are addressed in detail based on the use of Abaqus. The
application of this procedure is demonstrated through a detailed analysis of a deep excavation
case history. Results showed that the finite element analysis can capture the performance of the
deep excavation satisfactorily.
Keywords: Deep excavations, finite element analysis, case histories
1. Introduction
Deep excavations are widely used in urban areas for the development of underground space, e.g.
subway stations, basements for high-rise buildings, underground car parks and shopping centres.
However, the excavation process inevitably alters the stress states in the ground and may cause
significant wall deformations and ground movements. Especially when the excavation is close to
adjacent infrastructure, e.g. buildings, tunnels, buried pipelines, piled foundations, the excavation
induced ground movements must be carefully monitored and controlled within an acceptable
amount, to avoid any potential damage to these classes of infrastructure.
As a very complex soil-structure interaction problem, the performance of deep excavations is
influenced by the sophisticated soil behavior, complex retaining structures, and variable
construction sequences. Finite element analysis is an effective tool to study its mechanism, which
can consider both geotechnical and structural aspects such as soil behavior, details of structures,
and construction sequences. It provides necessary information on the performance of deep

excavations for design purpose, and can also be used to predict the behavior of deep excavation
and provide guidance for the construction.
Advanced finite element analysis of deep excavations should adopt accurate modelling procedures
for geotechnical and structural behavior, e.g. (i) the geometry of the excavation, (ii) detailed
retaining structures, (iii) correct construction sequence, and (iv) reliable material models and input
parameters.
The main issues involved with finite element analysis of deep excavations include selecting
appropriate constitutive models for soils and structures, the simulation of the construction
procedure, and the modelling of the soil/structure interface (Potts & Zdravkovic 2001). These
issues are addressed in detail in this paper based on the use of Abaqus version 6.11. The
application of this procedure is demonstrated through a detailed analysis of a deep excavation case
history. Results showed that the finite element analysis can capture the performance of the deep
excavation satisfactorily.
This paper is mainly based on relevant research conducted at University of Oxford and
Massachusetts Institute of Technology (MIT).
2. Components in braced deep excavations
2.1 The soil

The soil is a critical part to deal with in the modelling procedure. This is particularly related to the
partition and mesh of the model according to the excavation geometry, and the selection of
element types and constitutive models.
The soil can be modelled with tetrahedral or hexahedral continuum elements in 3D analysis. For
each type of elements, there are normally linear or quadratic elements (although higher order
elements are also possible) with full or reduced integration. In general, quadratic elements are
more accurate than linear elements in theory, but they are also expensive in calculations if the
same mesh is used. Hexahedral elements are suitable for excavations with regular geometry, e.g.
rectangular ones. Linear hexahedral elements with reduced integration (C3D8R) in Abaqus can be
used in large scale and complex analyses, on the balance of accuracy and efficiency. However,
higher order elements are recommended for more accurate purposes when computational resource
is not a problem.
The performance of deep excavations is strongly influenced by the soil behavior. In finite element
analysis, it is vitally important to use realistic soil models to obtain more reliable results. The
selection of constitutive models largely depends on the software that is used. Abaqus includes
several basic models for soils (e.g. Mohr-Coulomb, Drucker-Prager, and Modified Cam clay), and
also provides the interface for advanced users to implement their specific constitutive models
through the subroutine UMAT or VUMAT. A multiple-yield surface soil model (Houlsby 1999)
has been implemented into Abaqus by the first author to consider the small-strain stiffness
nonlinearity of the soil, and this model is used extensively at University of Oxford. Advanced soil
models, MIT-E3 soil model (Whittle 1987; Whittle & Kavvadas 1994), and MIT-S1 soil model
(Pestana 1994; Pestana & Whittle 1999), developed at Massachusetts Institute of Technology
(MIT), have also been implemented into Abaqus and successfully used for deep excavations as
well as other geotechnical problems.

2.2 The retaining wall
The retaining wall is generally modelled using either shell elements or solid elements in finite
element analysis, if the wall installation process is modelled using the Wished-In-Place (WIP)
method. Shell elements have no geometric thickness and are easier to generate in the mesh. In
addition, the internal forces and bending moments can be obtained directly. However, it is found
that the computed wall deflection and ground movement from shell element wall are larger than
those from solid element wall, and this is because the shear stress on the back of soil-wall interface
provides beneficial bending moment about the wall centreline when the diaphragm wall is
modelled with solid elements, whereas shell element wall does not have this effect (Potts &
Zdravkovic 2001; Zdravkovic et al. 2005). Moreover, when the wall installation process is
modelled explicitly in the model, e.g. the Wall-Installation-Modelled method, including the trench
excavation, bentonite slurry, and concrete injection, only solid elements are suitable to model the
wall because the geometry of the wall needs to be considered. Therefore, solid elements are
preferred to model the retaining wall in this paper. In terms of the output of internal forces and
bending moments, this can be achieved through embedding the shell element wall at the centreline
of solid element wall and assigning shell element wall with the same thickness of the retaining
wall but a small Young’s modulus if the wall is assumed as a linear elastic material. The low
stiffness of the embedded shell element wall will not affect the deformation of soil element wall,
but the internal forces and bending moments can be computed from the output of shell element
wall.
Another issue related to the retaining wall is modelling the construction joints in the wall, as
shown in Fig. 1. For example, the diaphragm wall is discontinuous in the horizontal direction
along the sides of the excavation because it is constructed by sections of wall panels.
Consequently, it cannot sustain any significant out-of-plane bending, and the horizontal stiffness
of the wall is also much smaller than the stiffness of the solid concrete. The influence of joints can
be considered by using the anisotropic wall approach (Zdravkovic et al. 2005) in which the
stiffness in the direction along the joints is reduced.
Fig. 1 Schematic view of different wall types (Zdravkovic et al. 2005)

The retaining wall is adequate to be represented by a linear elastic material for simplicity. Cracks
may exist in the concrete structures and propagate during the excavation, which can be considered
by reducing the nominal stiffness of the concrete components or using a more detailed concrete

crack model, such as the concrete damaged plasticity model in Abaqus. However, using a more
sophisticated concrete model will increase the complexity of the analysis.
2.3 The supporting system
The support system usually consists of horizontal components (e.g. reinforced concrete beams and
floor slabs, and temporary steel struts), and vertical components (e.g. piles and columns). In finite
element analysis, beam elements can be used to model the beams, piles, columns and steel struts,
whereas shell elements are suitable for floor slabs. Opening accesses are usually designed in the
floor slabs during the top-down excavation, and they can be modelled explicitly in the model.
In a similar way to the material model used for the retaining wall, the reinforced concrete
components in the support system can also be assumed to behave linear elastically for simplicity
or represented by a detailed concrete crack model. The concrete beams and floor slabs will shrink
or expand during the curing process and due to the variation of ambient temperature, which may
have large influence on the excavation behaviour. The thermal effects can be considered in the
analysis by introducing thermal strains. Moreover, gaps may develop at the connection between
the retaining wall and horizontal bracing structures, and concrete floor slabs may creep under
service condition, which has the tendency to induce larger wall deformations. These effects can
also be approximated by using the thermal shrinkage method.
2.4 Adjacent infrastructure
The adjacent infrastructure close to the excavations, e.g. buildings, utility pipelines, tunnels, and
pile foundations, may be included in the model if they are major concerns in the project, although
this would increase the complexity of the analysis. An example of this approach is given in Dong
(2014) in which the settlements of buildings and buried pipelines are investigated.
2.5 Constraints and contacts
The connections between different structural components (e.g. retaining walls, concrete beams,
floor slabs, and vertical piles) in deep excavations are usually assumed to be rigid (e.g. cast or
welded) in numerical analyses, but they may be more complicated in reality (e.g. pined
connections, and gaps) and they may change as the excavation proceeds. For simplicity, tie
constraints are usually applied at the connections of these components in the analyses.
The interface behavior between the soil and structures (mainly the soil/wall interface and the
soil/pile interface) can be modelled using contact pairs at the interface associated with a contact
model. However, it should be noted that contact is a highly complex mechanism and may cause
numerical difficulties. An extended Coulomb frictional contact model is available in Abaqus.
Modification of this model is described later in this paper, and the influence of interface properties
on the excavation behavior is investigated through parametric studies in Dong (2014).
2.6 Boundary conditions
The region to be analysed in deep excavations is of relatively small dimensions compared with
those of the surrounding medium. In finite element analysis, the normal practice is to extend the
mesh to some distance away from the zone of interest, and apply fixed displacement boundary
conditions there.

3. Material models
3.1 Constitutive models for the soil

The built-in soil models in Abaqus are conventional and do not contain modern concepts of soil
modelling, e.g. small-strain stiffness nonlinearity, recent stress history, and anisotropy.
Fortunately, Abaqus provides interfaces for advanced users to implement their own constitutive
models through subroutines UMAT and VUMAT. Advanced soil models are crucial to capture the
sophisticated soil behavior and improve the accuracy of the analyses. Three advanced soil models
are described here.
3.1.1 Multiple-yield surface soil model
A multiple-yield surface soil model (Houlsby 1999) has been implemented into Abaqus by the
first author using UMAT to consider the small-strain stiffness nonlinearity of the soil. This soil
model takes into account the non-linear behavior of soil at small strains, and also includes effects
such as hysteresis and dependence of stiffness on recent stress history. It uses multiple yield
surfaces within the framework of work-hardening plasticity theory. Non-linearity of the small-
strain response is achieved using a number of nested yield surfaces of the same shape as the outer
fixed failure surface. Those inner yield surfaces translate as the stress state changes, while the
bounding von Mises failure surface models perfect plastic behavior with no work hardening. As a
stress point moves in the stress space and encounters a yield surface as shown in Fig.2, the
stiffness reduces as shown in Fig.3, and the yield surface moves with the stress point. It thus
models the non-linearity of response at small strains, and the effect of recent stress history.
σ1
A
σ2 σ3
Fig.2 Yield surface after stress path from O to A and back to O (Houlsby 1999)

Shear stress τ
G2 = g 2G0
C2 = c2 su
G1 = g1G0
C1 = c1su
G0
O Shear strain γ
Fig.3 Shear stress-strain cure

The model is specified by the small strain shear modulus, G0 , the bulk modulus K and a set of
non-dimensional parameters ci and gi ( i = 1, n ) that are used to specify the size and work
hardening characteristics of each of the inner surfaces. The size of each inner surface is ci C and
the tangent shear modulus when the i th surface is active is gi G0 . A total of nine inner yield
surfaces are used in the analyses described in this paper, a balance between accuracy and
computational efficiency. Calibration of this soil model for Shanghai Clay is described in Dong
(2014). This soil model has been applied to tunneling (Burd et al. 2000) and deep excavations
(Dong 2014) with success.
3.1.2 MIT-E3 soil model
The MIT-E3 soil model (Whittle 1987; Whittle & Kavvadas 1994) is an advanced constitutive
model which captures the nonlinear and inelastic behavior of normally consolidated to lightly
consolidated clay (OCR<8) such as (i) small-strain stiffness nonlinearity, (ii) anisotropic stress-
strain-strength, and (iii) hysteretic and inelastic behavior due to cyclic loading. The model is
formulated based on the theory of incrementally linearized elasto-plasticity and consists of three
distinct components: (i) an elasto-plastic model for normally consolidated clays, (ii) a perfectly
hysteretic formulation, and (iii) bounding surface plasticity. Fig. 4 shows the shape of the yield,
failure and loading surfaces used in the MIT-E3 model.

Fig. 4 Yield, failure and load surface of MIT-E3 soil model (Whittle & Kavvadas
1994)
The model has 15 material constants and a series of state variables describing (i) the effective
stresses, (ii) the size and orientation of the bounding surface, (iii) the effective stresses at the most
recent reversal state, (iv) the strain accumulated since the last reversal state, (v) the size of the
bounding surface at the last reversal state, and (vi) the size of the load surface at first yield. This
soil model has been implemented to Abaqus by Hashash (1992), and extensively used in deep
excavations (Whittle et al. 1993; Hashash & Whittle 1996) as well as other geotechnical analyses.
3.1.3 MIT-S1 soil model
MIT-S1 model (Pestana 1994; Pestana & Whittle 1999) is a generalized effective stress soil
model, for describing the rate independent behavior of freshly deposited and over-consolidated
soils. It is a third generation of soil model, developed at MIT, after MIT-E1, and MIT-E3. It
inherits the basic structure of its predecessors (MIT-E3), but also introduces significant
improvements in the form of the bounding surface and hardening laws. A new framework of
compression behavior unifies the modelling of clays and sands. The model introduces new
expressions for describing small-strain non-linearity, therefore enhancing predictions of shear
modulus degradation for both clays and sands. The model also incorporates a variable Poisson’s
ratio to describe more realistically the variation of K 0 measured during one-dimensional swelling
and subsequently reloading. Fig. 5 shows the bounding surface of MIT-S1 soil model and
compares with isotropic yield surface of Modified Cam Clay model and anisotropic bounding
surface of MIT-E3 model.

Fig. 5 Comparison of yield surfaces of MIT-S1 with MCC and MIT-E3 (Pestana &
Whittle 1999)
The model requires, in general, 16 input parameters to characterize a general soil specimen.
Typical predictions for sands or clays, on the other hand, requires only 14 or 13 parameters,
respectively. Evaluation of this soil model on sands and clays are shown in (Pestana, Whittle &
Salvati 2002) and (Pestana, Whittle & Gens 2002) respectively. The MIT-S1 soil model has also
been implemented to Abaqus, and application of this model to deep excavations in Berlin Sands
can be referred to Nikolinakou et al. (2011).
3.2 Constitutive models for the structure
The structural components in deep excavations (e.g. the diaphragm wall, horizontal beams and
floor slabs, and vertical piles) are mainly reinforced concrete structures, and it is adequate to
model them as a linear elastic material for simplicity. The effect of cracks is normally considered
by reducing the nominal value of the Young’s modulus of the intact concrete.
Abaqus provides several more realistic models for reinforced concrete structures, e.g. the concrete
smeared cracking model, and concrete damaged plasticity model, for advanced analyses. The
concrete damaged plasticity model in Abaqus provides a general capability for modelling
reinforced concrete structures. It uses concepts of isotropic damaged elasticity in combination with
isotropic tensile and compressive plasticity to represent the inelastic behavior of concrete. It
consists of the combination of non-associated multi-hardening plasticity and scalar (isotropic)
damaged elasticity to describe the irreversible damage that occurs during the fracturing process.
However, the disadvantage of using these models is associated with the difficulties in determining
the large number of material parameters and the numerical instabilities during the calculation.
As discussed earlier in this paper, the retaining wall is discontinuous in the horizontal direction
and has construction joints in the wall. The anisotropic wall approach (Zdravkovic et al. 2005) is
an adequate method to consider the effect of construction joints. The retaining wall is modelled as
a cross anisotropic linear elastic material to investigate its capability in capturing the observed
excavation behavior (Dong 2014). For the local coordinate system in Fig.6, the Young’s modulus
𝐸 along 2-direction is reduced by a factor of β ( β < 1.0) , while the material is isotropic in 1-3
plane.

E1=E3
Eout/Ein=E2/E1 =β
3
2
1
1 – wall thickness
2 – wall length
3 – wall depth
Fig.6 Local directions for cross anisotropic wall

Thermal effects of concrete beams and floor slabs, as discussed earlier in this paper, may affect the
wall deformation and ground movement. This can be considered in the concrete material
properties by specifying a coefficient of thermal expansion α , and a temperature change ∆T .
3.3 Constitutive models for the soil/structure interface
An extended version of the classical isotropic Coulomb friction model is available in Abaqus for
use with general contact analysis capabilities. The extensions include an additional limit on the
allowable shear stress, anisotropy and the definition of a secant friction coefficient. The standard
Coulomb friction model assumes that no relative motion occurs if the equivalent frictional stress
τ eq
= τ 12 + τ 22 (where τ 1 and τ 2 are shear stresses on the interface), is less than the critical stress
τ crit which is proportional to the contact pressure p in the form τ crit = µ p (where µ is the
friction coefficient). It is possible to put a limit on the critical shear stress, τ crit = min ( µ p,τ max ) ,
where τ max is a user specified constant. If the equivalent stress reaches the critical stress (
τ eq = τ crit ), slip will occur.
In geotechnical engineering, it is reasonable to link τ crit with the undrained shear strength of the
soil su , e.g., τ crit= α ⋅ su , where α is an coefficient. Further modification can be made through a
subroutine FRIC to make τ max as a variable. This modified version of the extended Coulomb
friction model is illustrated in Fig.7. The influence of the value of α on the excavation behaviour
is investigated through a number of parametric studies in Dong (2014).

τ
,
µ
1
O p
Fig.7 Extended Coulomb friction model
4. Modelling procedures
4.1 Geostatic analysis

Geostatic stress-displacement analysis is the first step for equilibrium iteration to set up the initial
stress state in the ground prior to the construction. The vertical stress is governed by the gravity,
while the horizontal stresses are related to the vertical stress through the coefficient of lateral earth
pressure at rest K 0 . There are well established procedures to define the value of K 0 in both
normally consolidated and overly consolidated soil.
4.2 Wall and pile installation
The diaphragm wall is a cast-in-situ reinforced concrete retaining wall constructed using a slurry
supported trench method. The installation process includes deep mixing, slurry supported
excavation, placing the reinforced cage, concrete casting and curing. The bored pile installation
follows the similar process. The installation process modifies the in-situ stress state in the ground
close to the trench and influences the starting point for analysis of the post-construction behaviour.
In addition, it results in substantial ground movements and settlements of adjacent infrastructure,
which may be significant compared to those induced by the excavation itself. The influence
depends on the groundwater level, panel width, and changes in pore water pressure during
diaphragm wall installation. Therefore, although simple, the widely used Wished-In-Place (WIP)
method for the wall and pile installation in the finite element analyses of deep excavations
neglects the installation effect on the ground movement and the subsequent excavation behaviour.
The effects of diaphragm wall installation have been investigated through finite element analyses
by several researchers, e.g. Ng & Lings (1995), Ng & Yan (1999), Gourvenec & Powrie (1999),
and Burlon et al. (2013). The construction of a single panel is modelled by the following
procedure:
1. Excavate the trench and apply the hydrostatic bentonite pressure on the trench surfaces
simultaneously;
2. Cast the panel with concrete by increasing the lateral pressure inside the trench to a
bilinear wet concrete pressure envelope as proposed by Lings et al. (1994);

3. Cure the concrete panel by replacing the trench with concrete material and remove the
applied bilinear concrete pressures simultaneously.
4.3 Soil removal and structure installation
The soil excavation and structure installation can be modelled by deactivating or reactivating the
corresponding finite elements in the model. When the elements are deactivated in the analysis, the
contribution of mass and stiffness of that element is removed from the global Mass and Stiffness
matrix.
4.4 Dewatering and consolidation
A dewatering process is normally used inside the excavation when the ground water table is high,
e.g. in Shanghai. The dewatering and consolidation processes can be conducted using a
displacement-pore pressure coupled analysis with an effective stress soil model. The general
process in the finite element analysis is described in this section, although they are not modelled in
the analyses in this paper, due to the limitation of the soil model used. The pore water pressure at
the ground water table level and the excavated soil surface is set to zero in the analysis. Changes
of pore water pressure with time are calculated with the drainage boundary conditions and
specification of permeability and consolidation characteristics of the soil. Detailed analysis can be
seen in Zheng et al. (2014)
5. Analysis of deep excavation case histories
Application of the previous procedures will be demonstrated through a detailed study of a well-
documented deep excavation case history.
5.1 General description of the case history
Shanghai Xingye Bank building is a high-rise building (82.5m high) with a three-level basement.
The structure is constructed from a reinforced concrete frame and the building is founded on deep
piles (Xu 2007). Construction of the basement began in February 2002, and finished in December
2003. The excavation depth, as shown in Fig. 8, is 14.2m on the west side, and 12.2m on the east
side. The excavation is surrounded by a number of densely packed buildings and several old
service pipes. Ground improvement (e.g. root piles, and spiral injection piles), therefore, was
conducted outside the excavation to mitigate the detrimental effects on the buildings. A
comprehensive field measurement was conducted during the excavation process. This site is
situated on typical Shanghai Clay which is composed of Quaternary sediments and has high water
content, low shear strength, and high compressibility.

Fig. 8 Plane view of adjacent infrastructure (Xu 2007)
The A-A sectional view of the excavation in Fig. 9 shows briefly the structure of the retaining
system which is mainly composed of the diaphragm wall, horizontal beams and floor slabs, and
vertical piles and columns.
Fig. 9 Layout of instrumentation at the construction site (Xu 2007)

5.2 Finite element model
A detailed 3D finite element model was created to represent at an appropriate level of detail of
structure and various construction sequences.
The soil and diaphragm wall are modelled with 8-noded solid hexahedral elements with reduced
integration (C3D8R). The support system includes vertical piles, horizontal beams, and floor
slabs. The piles and beams are modelled with 3D 2-noded beam elements (B31), while the floor

slabs are modelled with 4-noded quadrilateral shell elements with reduced integration (S4R).
Three adjacent buildings (i.e., ECADI building, CB building, and SJB building, as shown in Fig.
8) are included in the model. The external and internal walls, as well as the floor slabs, are
modelled with linear quadrilateral shell elements with reduced integration (S4R). The bottom of
the building is embedded into the piled-raft foundation which is modelled with solid and beam
elements. The ground improvement around the excavation, as shown in Fig. 8, is represented by
shell elements for simplicity. Two buried pipelines (see Fig. 8) on two sides of the exuviation are
modeled using beam elements.
Fig. 10 Mesh and boundary conditions of the model

The soil is represented by the multiple surface plasticity model, as described earlier in the paper.
The stiffness G0 and shear strength su parameters increase linearly with depth. The diaphragm
wall and the surrounding ground improvement are modelled as an anisotropic linear elastic
material to consider the joints in the wall. The horizontal beams and floor slabs are modelled as a
linear elastic material with thermal effects. The buildings are assumed to behave linear elastically
because no obvious cracks or damage were observed in the buildings during and after the
excavation. The piles and pipelines are also assumed to be a linear elastic material.
The analysis is conducted in undrained conditions considering the short period of construction and
low permeability of the clay. The real construction sequence is followed in the analysis.
5.3 Computed results
Correlation between the numerical results and field measurement, presented in Dong (2014),
indicates that the advanced finite element analysis can reproduce the observed excavation
behaviour satisfactorily. This section only shows some displacement contours of ground and
structures at the final stage of the excavation.
The vertical displacement of the whole model is shown in Fig. 11. As can be seen, the soil inside
the excavation moves upwards due to the stress relief caused by the soil removal, while the soil
outside the excavation settles down, and the settlement is larger behind the wall centre but smaller

around the wall corner, indicating the 3D effect or corner effect. The adjacent buildings settle with
the ground, and the settlement is larger close to the diaphragm wall and becomes smaller far away
from the excavation, which means the building inclines towards the excavation. It also shows that
the ECADI building settles more than the other two buildings.
Fig. 11 Vertical displacement of whole model (unit:m)

As shown in Fig. 12, the largest wall deflection is concentrated at the wall centre area close to the
excavation formation level, while the deflection at the corner is smaller due to the corner effect.
The shape of the diaphragm wall also affects the wall deflection. For example, the wall deflection
is larger in the longer piece of wall, while smaller in the shorter piece of wall.
Fig. 12 Wall deflection (unit:m)

The displacement contour of the support system is shown Fig. 13. The piles move upwards due to
basal heave. The displacement of beams and slabs are constrained by diaphragm wall and piles,
and the differential settlement is clearly seen which may cause cracks inside the concrete.

Fig. 13 Displacement distribution of the supporting system
As shown in Fig. 14, the ground improvement deforms with the ground and the pattern is similar
to the diaphragm wall deformation. The deformation is larger behind the centre of the diaphragm
wall at the final excavation formation level, and smaller at the corner.
Fig. 14 Deformation of the ground improvement (unit:m)

It can be seen in Fig. 15 that the settlement of the buried pipelines is mainly concentrated in the
region behind the excavation, which confirms that the excavation has a large impact on the
adjacent buried pipelines.
Fig. 15 Settlement of the buried pipeline (unit:m)

6. Conclusions
This paper discussed the procedures in advanced finite element analysis of deep excavations based
on the use of Abaqus. The key components in the analysis include the soil, the retaining wall, the
support system, adjacent infrastructure, constrains and contacts, and the boundary conditions. The
material constitutive models are critical to capture the main excavation behaviour. The models for
the soil, structures, and contact are described in details. Advanced soil models are implemented
into Abaqus through UMAT, which considers more sophisticated soil behaviors such as small-
strain stiffness nonlinearity, and anisotropy. An anisotropic wall approach is adopted to consider
the construction joints in the retaining wall, and thermal effects are considered in the horizontal
beams and floor slabs. The extended Coulomb friction model in Abaqus is modified through a
subroutine FRIC, in which the shear resistance at the soil-structure interface is related to the
undrained shear strength of the soil. The main modelling procedures in the analysis are also
described in this paper, including the geostatic analysis, wall and pile installation, soil removal and
structure installation, dewatering, and consolidation. The proposed procedure is demonstrated
through a detailed study of a deep excavation case history. Results showed that the advanced finite
element analysis can reproduce the observed excavation behaviour satisfactorily.
7. Acknowledgement
The main work described in this paper is based on the research conducted at University of Oxford
during the DPhil study of the first author supervised by Prof Harvey J. Burd and Prof Guy T.
Houlsby. The case history is kindly provided by Dr Zhonghua Xu who collected and analysed the
field measurement in his PhD thesis. Extension of this work is currently undergoing with Prof
Andrew J. Whittle at MIT, based in Singapore-MIT Alliance for Research and Technology.
8. References
1. Burd, H.J. et al., 2000. Modelling tunnelling-induced settlement of masonry buildings.

Proceedings of the Institution of Civil Engineers: Geotechnical Engineering, 143(1), pp.17–
29.
2. Burlon, S., Mroueh, H. & Shahrour, I., 2013. Influence of diaphragm wall installation on the
numerical analysis of deep excavation. International Journal for Numerical and Analytical
Methods in Geomechanics, 37(11), pp.1670–1684.
3. Dong, Y., 2014. Advanced Finite Element Analysis of Deep Excavation Case Histories. DPhil
thesis, University of Oxford, Oxford, UK.
4. Gourvenec, S.M. & Powrie, W., 1999. Three-dimensional finite-element analysis of
diaphragm wall installation. Geotechnique, 49(6), pp.801–823.
5. Hashash, Y.M.A., 1992. Analysis of deep excavations in clay. Sc.D thesis, Massachusetts
Institute of Technology, Cambridge, Boston, U.S.
6. Hashash, Y.M.A. & Whittle, A.J., 1996. Ground movement prediction for deep excavations in
soft clay. Journal of Geotechnical and Geoenvironmental Engineering, 122(6), pp.474–486.
7. Lings, M.L., Ng, C.W.W. & Nash, D.F.T., 1994. The lateral pressure of wet concrete in
diaphragm wall panels cast under bentonite. Proceedings - ICE: Geotechnical Engineering,
107(3), pp.163–172.

8. Ng, C.W.W. & Lings, M.L., 1995. Effects of Modeling Soil Nonlinearity and Wall
Installation on Back-Analysis of Deep Excavation in Stiff Clay. Journal of Geotechnical
Engineering, 121(10), pp.687–695.
9. Ng, C.W.W. & Yan, R.W.M., 1999. Three-dimensional modelling of a diaphragm wall
construction sequence. Geotechnique, 49(6), pp.825–834.
10. Nikolinakou, M.A. et al., 2011. Prediction and interpretation of the performance of a deep
excavation in berlin sand. Journal of Geotechnical and Geoenvironmental Engineering,
137(11), pp.1047–1061.
11. Pestana, J.M., 1994. A unified constitutive model for clays and sands. ScD thesis,
Massachusetts Institute of Technology, Cambridge, Boston, U.S.
12. Pestana, J.M. & Whittle, A.J., 1999. Formulation of a unified constitutive model for clays and
sands. International Journal for Numerical and Analytical Methods in Geomechanics, 23(12),
pp.1215–1243.
13. Pestana, J.M., Whittle, A.J. & Gens, A., 2002. Evaluation of a constitutive model for clays
and sands: Part II – clay behaviour. International Journal for Numerical and Analytical
14. Pestana, J.M., Whittle, A.J. & Salvati, L.A., 2002. Evaluation of a constitutive model for
clays and sands: Part I - sand behaviour. International Journal for Numerical and Analytical
15. Potts, D. & Zdravkovic, L., 2001. Finite Element Analysis in Geotechnical Engineering: Vol.
2 - Application, London: Thomas Telford Limited.
16. Whittle, A.J., 1987. A constitutive model for overconsolidated clays with application to cyclic
loading of friction piles. ScD Thesis, Massachusetts Institute of Technology, Cambridge,
Boston, U.S.
17. Whittle, A.J., Hashash, Y.M.A. & Whitman, R. V, 1993. Analysis of deep excavation in
Boston. Journal of Geotechnical Engineering - ASCE, 119(1), pp.69–90.
18. Whittle, A.J. & Kavvadas, M.J., 1994. Formulation of MIT-E3 constitutive model for
overconsolidated clays. Journal of Geotechnical Engineering - ASCE, 120(1), pp.173–198.
19. Xu, Z.H., 2007. Deformation Behaviour of Deep Excavations supported by Permanent
Structure in Shanghai Soft Deposit. PhD thesis, Shanghai Jiao Tong University, Shanghai,
China.
20. Zdravkovic, L., Potts, D.M. & St. John, H.D., 2005. Modelling of a 3D excavation in finite
element analysis. Geotechnique, 55(7), pp.497–513.
21. Zheng, G. et al., 2014. Test and numerical research on wall deflections induced by pre-
excavation dewatering. Computers and Geotechnics, 62, pp.244–256.

Modeling of Confined Inflatable Structures
Eduardo M. Sosa*, Choo-Siang Wong† and Ever J. Barbero†
Department of Civil and Environmental Engineering
*
West Virginia University, Morgantown, WV, USA

†
Department of Mechanical and Aerospace Engineering
West Virginia University, Morgantown, WV, USA
Abstract: This work presents the development of Finite Element (FE) models to simulate different
stages of the design and operation of constrained inflatable structures. Such structures can be used,
for example, for sealing segments of underground civil infrastructure such as large pipes or tunnels.
In such applications, the confined inflatable structure is prepared for placement, either permanently
or temporally, and left ready for deployment, inflation, and pressurization when needed. Once
positioned and in operation, the inflatable may be subject to a number of possible loading scenarios
and interaction with the surrounding confining environment, which must be evaluated beforehand.
FE models developed for this work include models for folding, positioning, deployment,
pressurization, and in-service loading of constrained inflatable structures. Models were created and
analyzed with the Simulia/Abaqus simulation suite. Membrane materials, relaxation techniques,
mass scaling, as well as a multi-chamber approach were evaluated and implemented in order to
better correlate predicted results to available experimental data. The result is a modeling procedure
that can be used to predict the performance of an inflatable structure for a given confining
environment. Thus, the models can be used as a predicting tool for optimization of different aspects
of the design and field operation of the inflatable structure.
Keywords: Abaqus/CAE, Abaqus/Explicit, Confined Inflatable, Deployment, Folding, Membrane.
1. Introduction
Large inflatable structures for protection of underground infrastructure, such as railway tunnels or
large pipes, has been under development at West Virginia University (WVU) since 2007 (Barrie,
2008, Martinez et al, 2012, Fountain, 2012, Barbero et al, 2013-a/b and Sosa et al, 2014-a/b). The
implementation of large inflatable structures (also called inflatable plugs) inside civil infrastructure
such as transportation tunnels is intended to prevent or reduce the damage induced by hazardous
events by creating a compartment to contain the threat. Potential threats include flooding, smoke or
noxious gases that can propagate along a tunnel system and compromise its functionality and
structural integrity. The inflatable structures are typically installed at specific locations of the tunnel
in order to create a compartment that can isolate the compromised region. As part of this
development, WVU has designed several prototypes (Barbero et al, 2013-b) and conducted multiple
validation tests of confined inflatable structures at full and reduced scale (Martinez et al, 2012,
Barbero et al, 2013-a, Sosa et al, 2014-a/b). Some of these validation tests were performed using
specially built testing facilities designed to simulate a segment of an actual rail transportation tunnel
(Barbero et al, 2013-a, Sosa et al, 2014-a/b).

However, performing tests, especially at full-scale in either an actual tunnel segment, or in the
specially built testing facilities, is a very complex task. Validation tests performed at full-scale
demonstrated that it can take several iterations to achieve acceptable results which cannot be
predicted in advance. Therefore, the development of numerical models using Finite Element
Analysis (FEA) becomes imperative as a predicting tool that can anticipate the change in
performance of the system under design and operational changes. Calibrated models are also used
for parametric studies and design optimization that otherwise would be very costly to implement at
experimental level. Examples of tests performed at full-scale are illustrated in Figure 1. Additional
details on testing configurations and experimental results are reported by Martinez et al, 2012,
Barbero et al, 2013-a, Peil et al, 2012 and Sosa et al, 2014-a/b.
Figure 1: Full-Scale Inflatable Plugs: a) Test performed in an actual tunnel in 2008

(Barrie, 2008, Martinez et al, 2012); b) and c) Tests performed at WVU (Barbero et
al, 2013-a, Sosa et al, 2014-a/b).
This paper aims to introduce some of the work carried out for the development of FE models of
confined inflatable structures used for protection of underground tunnels. These models were
created and analyzed based on information available from experimental results of tests performed
at full-scale (Barbero et al, 2013-a; Sosa et al, 2014-a/b). The two main objectives of this study are:
1. Development of FE models of confined inflatable plugs to simulate:
a. The structural components of the inflatable
b. The folding process of the inflatable
c. The deployment of the inflatable under confined conditions
d. The interaction of the inflatable with different elements of the confining surface
and the global and local conformity to those elements
2. Comparison of numerical results with available experimental data in order to adjust
modeling parameters and improve the predicting capability of the models.

2. Finite Element Model Setup
2.1 Modeling Tools
Abaqus/CAE and Altair’s HyperMesh were chosen as finite element pre-processors for creating
geometries, meshing and renumbering of nodes and elements into a scripting-friendly form. For
solving the models, Abaqus/Standard and Abaqus/Explicit were selected to solve the simulations.
In the post-processing phase, Python, Tool Command Language (TCL) Script, and Abaqus Scripting
were implemented to generate numerical outputs, and Abaqus/CAE was used to visualize and
present the numerical results graphically.
2.2 Geometries
The development of a comprehensive model able to reproduce the different stages of the work
performed experimentally required the creation of several components that constituted the whole
model. The two main components are the inflatable plug and the confining environment
representative of a typical tunnel segment in which the inflatable will be installed. Additional
components used in the modeling process included auxiliary planes representative of sitting base
and tools and procedures that were used during the folding process of an actual full-scale prototype.
The features of the different components of the FE model were created considering the different
stages of implementation of an inflatable structure in a tunnel section. These stages included folding,
placement in the storage area, relaxation, and deployment of the inflatable plug.
The first main component of the FE model is the inflatable plug. It consists of a cylinder with two
hemispherical end-caps. Each hemispherical end-cap includes three partitions on its surface created
for manufacturing purposes as illustrated in Figure 2. The perimeter of the cylindrical portion of the
inflatable is designed to be larger than tunnel’s perimeter. Over sizing is added to achieve
conformance of the plug to the enclosure taking into account wrinkles and sag that may occur during
deployment. Better fitting, or conformance, result in better sealing, which is the ultimate objective
of the system.
The structural membrane of the inflatable consist of a multiple layer system. The multi-layer system
is modeled with an equivalent single membrane built from M3D4 membrane elements with
equivalent thickness, mass and stiffness representative of the macro-mechanical behavior of the
structural membrane. The inflatable plug also includes two metal fittings used as air fill and air
release ports. These two fitting are located on one of the hemispherical end-caps. Since they are
much more rigid than the structural membrane, they are modeled by R3D4 rigid elements. The total
mass of the inflatable plug including metallic fittings is approximately 907 kg.
Figure 2(a) shows how the geometry of the inflatable plug was initially partitioned into several
auxiliary surfaces and edges. The purpose of creating these auxiliary surfaces and edges is to delimit
the position of fitting areas, internal chamber surfaces, folding surfaces and folding lines at the
cylindrical region of the plug. Figure 2(b) shows the meshed configuration of the inflatable plug.

Figure 2: Inflatable plug: An actual full-scale prototype (left); (a) Initial geometry
with auxiliary partitions; (b) Final meshed configuration.
The second main component of the FE model is the model of the tunnel section. In this study, the
tunnel section in which the inflatable plug will be placed is assumed to be non-deformable. A typical
tunnel profile is illustrated in Figure 3. This cross section was used for the analyses performed in
this work. Figure 3 also shows the 3D geometry and the meshed configuration of the tunnel used
for all the analyses. Rigid elements R3D4 were used to represent the tunnel section.
During the folding process, the inflatable plug will interact with a flat base and auxiliary rotational
plates created to simulate the folding process. Similarly to the tunnel section, these surfaces are
considered non-deformable and were meshed with R3D4 rigid elements. Figure 4 displays all the
components involved in the development of the FE model including the inflatable plug, tunnel,
auxiliary base, and auxiliary rotational plates.
Figure 3: Tunnel cross section: Typical profile prepared for receiving an inflatable
plug (left); Meshed configuration used in the FE models (right).

Figure 4: All components of the FE model.
3. Folding Process
The folding process implemented in this work consisted of a combination of rigid body rotations
and translations applied as boundary conditions to selected nodes and elements previously marked
on the surface of the inflatable. Folding lines were created from sets of nodes positioned on the
surface of the cylindrical portion of the inflatable plug which guided the folding process. The FE
model of the folding process is developed as a set of geometric transformations that simulates an
actual folding sequence (Barbero et al, 2013-a; Sosa et al, 2014-a).
The model created for simulation of the full folding process included the following general steps:
1) Unconstrained, unstressed inflation; 2) flattening and grounding; 3) and folding by rolling. In the
first step, the starting position of the inflatable was in unconstrained and unstressed conditions and
using the ideal design geometry. During the second step, the flattening was achieved by
simultaneous use of horizontal displacements and vertical gravity force. An illustration of the
implementation of these two steps is shown in Figure 5. Once the inflatable is flattened and laying
on the ground (represented by the base plate shown in Figure 4), the folding process began by
forming a longitudinal wrinkle designed to hold and sequentially release membrane material during
the deployment process. This membrane material was held by ties that were modeled by connectors
placed discretely along one of the longitudinal edges of the artificial wrinkle. The flattened inflatable
at the beginning of the folding sequence is illustrated in Figure 6. The third step of the folding
consisted of gradual rolling of the flattened inflatable by successive lifting and partial rotations of
the partial folds by using the rotational rigid plates illustrated in Figure 4. Three sets of lifting and
rotation maneuvers, combined with gravity force, were applied to complete the folding sequence.
Figure 7 illustrates the sequence of the last set of folding maneuvers and the final folded
configuration.

Figure 5: Sequence of deflation, flattening and grounding in preparation for
folding.
Figure 6: Flattened inflatable structure at the start of the folding process.
Figure 7: Folding sequence: (a) Vertical lifting; (b) Rotation; (c) Action of vertical
gravity force. This sequence was repeated three times.

4. Positioning of the folded inflatable
The simulation of the placement process consists of a combination of a rigid body rotations and
translations of the folded plug in order to place it in the storage area. Once in place, the folded plug
is connected to the tunnel along a predefined horizontal line located inside the storage area that
provides alignment to the inflatable during the deployment and inflation process. The placement is
completed by further translation originated by a horizontal gravity force. Figure 8 shows different
views of the placement sequence of the folded plug within the tunnel section. Once the folded plug
is positioned in the storage area, a vertical rigid plane that represents an actual enclosure holds the
folded plug within the volume available for storage in a typical tunnel section.
(a)
(b)
Figure 8: (a) Initial condition for placement; (b) Orientation of folded plug after
rotation and horizontal translation.

The rigid plane not only simulates the enclosure but also maintains the folded inflatable plug within
the storage area during the implementation of the dynamic relaxation process. The main purpose of
implementing the dynamic relaxation process into the simulation is to restore distorted elements
back to their original condition before doing the deployment simulation. Most of available
commercial software execute the relaxation process through passing either node’s or element’s
coordinates from an initial configuration to a reference configuration. In the simulation, the rigid
plane pushes and closes the storage area when relaxation simulation commences as illustrated in
Figure 9. Figure 9(d) shows the final result of the implementation of dynamic relaxation and gravity
force. The result of both effects defined the starting position of the folded plug for the simulation of
deployment. The relaxation process is also added to the deployment simulation in order to minimize
the kinetic energy, stabilize the simulation process, and avoid distortions on the membrane surface
when the inflatable is fully positioned inside the tunnel section.
Figure 9: (a) Rigid plane before closing the storage area; (b) closing the storage
area; (c) Folded plug after implementation of dynamic relaxation; (d) Action of
vertical gravity force and initial state of the folded plug before deployment.

5. Simulation of Deployment
5.1 Inflation Algorithm and Model Setup
The Uniform Pressure Method (UPM) was implemented in this work. This method has been widely
used in airbag simulations for more than two decades and demonstrated to predict reasonably good
results. The UPM method was found to be simple, computationally efficient and adequate for
modeling relatively slow inflation since the inertia of the inflation gas can be neglected. The
capability of reproducing the actual deployment behavior with relatively low computational cost
and high accuracy were two of the key factors that led to the adoption of the UPM as inflation
method. The implementation of more sophisticated algorithms, such as the Coupled Eulerian
Lagrangian (CEL) method and the Arbitrary Lagrangian Eulerian (ALE), or mesh-free methods,
such as the Smoothed Particle Hydrodynamics (SPH) or the Finite Pointset Method (FPM), was
ruled out at this stage primarily because of their elevated computational cost (in the order of about
4 to 15 times higher than UPM depending on the mesh density) and the difficulty of obtaining all
the necessary parameters to create an accurate model.
The main assumptions of UPM are that the pressure in the airbag is spatially uniform during the
inflation; it also assumes no heat transfer and that the inflating gas behaves as an ideal gas with
constant specific heats. A limitation of this simplified algorithm is the inability to simulate local
fluid effects because the formulation does not involve the equations of fluid dynamics that describe
the movement of fluid. The multi-chamber and the fluid exchange approaches were needed to extend
the capability of the UPM in order to replicate the actual deployment behavior obtained from
experiments at full-scale.
The UPM is implemented in Abaqus/Explicit based on the formulation proposed by Wang and
Nefske (1988). It requires the definition of surface-based cavities to model the fluid-structure
interaction during the deployment and inflation process. This capability allows to use standard finite
elements to model the membrane of the inflatable; it requires a surface definition on the cavity
boundary for coupling the deformation of the membrane and the pressure exerted by the inflation
fluid (air in this case); it also requires definition of the fluid behavior, fluid exchange to model the
flow of the fluid, and the definition of the inflator properties for modeling the deployment of the
cavities. Some of the advantages of implementing a deployment model based on surface-based
cavities include the possibility of straightforward modeling of the adiabatic process, a more precise
modeling of the inflation process including properties of the multiple cavities (or chambers), as well
as relatively rapid convergence of the analysis which shortens the computational time.
Most of the pre-simulation conditions for modeling deployment of a confined inflatable are similar
to the ones that are typically applied to simulations of airbags. However, additional assumptions
have been made in order to define some of the particular conditions of the simulation. These
additional assumptions were based on the behavior observed during actual deployments of full-scale
experimental prototypes reported by Barbero et al, 2013-a, Sosa et al, 2014-a); they included:
 Vertical gravity force is applied on the entire inflatable plug from the beginning to the
end of the simulations.
 Air at ambient temperature of 80°F and standard ambient pressure is assumed to exist
during the inflation of the plug.
 The volumetric values of air flow and inflation time are based on experimental results.

 Inflow air behaves as an ideal gas without heat transfer into the fluid cavity (adiabatic
process).
 The air is assumed to be transferred into and within plug through an orifice which has a
discharge coefficient of 1.0.
 The incoming air enters into the plug through the inflation fittings.
 The dynamic relaxation process applies to the model during entire deployment
simulation.
 Internal chamber walls were used to direct the airflow inside the inflatable and to
calculate the volume of fluid within the plug.
 The equivalent external fabric is assumed to be impermeable.
The model setup for simulation of deployment comprised the inclusion of a passive restraining
mechanism consisting of multiple uniaxial ties installed in the surface of the inflatable during the
folding process. The purpose of the ties is to produce a sequential release of membrane material in
order to improve the level of local conformity on the walls and roof of the tunnel. These ties were
represented by connector elements and are assumed to have translational degrees of freedom during
the deployment simulation.
The model setup also included the creation of virtual internal chambers used to guide the airflow.
The position of the internal chambers delimited by chamber walls placed inside the inflatable plug
is illustrated in Figure 10. In this model, the inflation port is located in Chamber 2.
(a) (b)
Figure 10: Position of chamber walls and flow direction within the inflatable plug;
(a) Unguided airflow; (b) Guided airflow.

5.2 Initial Configuration before Deployment
The folded plug placed in the storage area illustrated in Figure 9(d) was used as starting point of the
deployment simulation. In this configuration, the folded plug sits on the base of the storage area
under the effect of its own weight. A vertical virtual gate holds it up until the deployment sequence
is activated. The folded plug is connected to the tunnel section at line B as shown in Figure 9(d),
and the nodes of this line are fixed in all directions. This boundary condition represents the ties that
fasten and restrain the inflatable plug to the tunnel section and are assumed to be unbreakable during
the deployment simulation.
An unfolded reference configuration (initial metric) illustrated in Figure 10 was defined to specify
the unfolded stress-free configuration of the inflatable plug. The folded plug illustrated in Figure
9(d) was used as initial configuration and the reference configuration were used to account for the
wrinkles that arise from the folding process. Under this procedure, since a reference configuration
was specified for all elements, any initial stress conditions specified for the same element are
ignored, therefore no initial stresses were included in the membrane of the folded bag. Typically,
during the modeling of folded membranes initial errors and mesh distortions are introduced
inevitably during the folding process. This distortions in the folded mesh may lead to initial stresses
that can affect the final shape of the deployed inflatable structure by creating fictitious wrinkles or
bogus stress concentrations that do not exist in the real structure.
Gravity force and dynamic relaxation are applied on the folded plug from the very beginning of the
simulation. Upon removal of the virtual vertical gate, the gravity force induces the fall of the plug,
which starts to unroll and therefore initiating the deployment process. The tunnel section is assumed
to be a rigid body fixed in X, Y, and Z direction. A friction coefficient 0.19 was used between plug
and tunnel; and a self-friction coefficient of 0.21 was used for the fabric-to-fabric friction. These
two values define how the surface of the membrane interacts with itself and the tunnel surface as
the deployment develops in the simulation. The hard contact option was selected to be the contact
method between parts.
The simulation plan showing various deployment conditions is summarized in Table 1. The folded
plug in Deployment #1 did not include connector elements to control the release of membrane
material; moreover, the air flow was not directed within the plug. When the air flow is not directed,
the air fills up the plug evenly, spontaneously, and immediately once it enters into Chamber 2, as
illustrated in Figure 10(a). The uniform pressure is calculated and applied on the entire internal
surface of the plug when the airflow is not directed within the plug. Deployment #2 included
connector elements installed during the folding process. For Deployment #3, a stiffness scale factor
was included in the connector elements in order to compensate for the dynamic response associated
to the mass scaling factor. Deployment #4 is similar to Deployment #3, but with a specific fluid
exchange designed to guide the airflow within the plug. From Figure 10(b), the air fills up the first
half (from Chamber 1 to Chamber 3) immediately after entering into the plug. After this initial step,
the second half (Chamber 4) of the plug starts inflating at the 20th second.

Table 1: Test Plan of Deployment Simulations.
Deployment # Connector Elements Air Flow Directed
1 No No
2 Yes, without mass scaling No
3 Yes, with mass scaling No
4 Same as Deployment 3 Yes
5.3 Results
Figure 11 shows results corresponding to Deployments #1 and #4 as extreme cases compared to
available experimental results. The airflow in Deployment #1, #2, and #3 was not directed, so the
plug was inflated uniformly and instantaneously once the inflation part has been activated. The fluid
exchange timing in Deployment #4 contributed to replicate the actual deployment performance.
The FE simulation can be used to evaluate the conformance, or fit, of the inflatable to the tunnel
geometry. Lack of conformity reveals itself in the form of gaps between the inflatable and the tunnel
wall, mainly observed at geometrical transitions such as corners and around obstructions such as
pipes to which the inflatable has to adapt in order to seal. Lack of conformity is directly responsible
for the leakage rate; that is, the flow rate that the plug is incapable of stopping, which obviously
needs to be minimized. The simulation is also capable of predicting wrinkles in the plug fabric. The
position and extension of wrinkles can be predicted accurately and they compare very well with
experimental observations available in the literature. These wrinkles are the result of suboptimal
deployment. Both wrinkles and gaps can be minimized by optimization of folding sequence;
deployment sequence including air flow rate; position, number, and strength of passive restraining
elements, and friction coefficient, among other factors. The effect on conformance of any and all of
these factors, isolated or in combination, can be predicted by the simulation, thus minimizing the
need for costly and time consuming experimentation. The excess of membrane material forms
wrinkles on the tunnel wall instead of on the corner of the floor. This result is similar to what it was
observed experimentally. The occurrence of wrinkles is in part related to how the membrane
material is distributed and driven during the deployment, but it is also function of the percentage of
manufacturing over sizing left in the perimeter of the cylindrical portion of the plug. The extra
membrane material is added to account for unforeseen elements in the tunnel that can add to the
perimeter to be sealed. It also accounts for the non-uniform nature of the deployment process that
can lead to an irregular distribution of membrane material.
The level of global conformity of the plug to the tunnel section was evaluated by computing the
contact area obtained for each deployment. Abaqus has the capability of computing the contact area
between two objects, which in this case corresponded to the inflated plug and the tunnel walls. Table
2 summarizes the contact area for each deployment case. The cylindrical portion of the plug has a
theoretical contact area of 71.98 m². This area should provide sufficient slippage resistance to handle
an external force that will try to push the plug along the tunnel section that the plug itself is intending
to seal.

Experimental result FEM: Deployment #1 FEM: Deployment #4
Figure 11: Deployment sequences: Experimental (from Barbero et al, 2013-a, Sosa
et al, 2014-a) vs. Deployments #1 and #4.

Figure 12 shows a close-up view of experimental and simulation results illustrating the level of
agreement reached by the finite element model particularly at the corners and change of angles in
the perimeter of the tunnel.
Figure 12: Comparison of: (a) simulation results (Deployment #3) and (b)
experimental results (Barbero et al, 2013-a, Sosa et al, 2014-a).
Table 2 shows that for Deployment #1 the contact area increased 14% with respect to the original
design contact area, even though the model did not include any connector elements to drive up
membrane material, which created at least three bridging spots in the upper part of the tunnel, as
observed in Figure 13. The extra contact area is generated by the confining effect of the tunnel
section at the intersection of the hemispherical end-caps and the cylindrical part of the inflatable.
The connector elements implemented in Deployment #2 failed earlier than expected. Still, the
contact area increased 27% and reduced visible bridging to two places only. From the calculated
contact area, the addition of connector elements clearly improves the global conformity by nearly
duplicating the percentage of increase of contact area with respect to Deployment #1. Deployment
#3 included connectors with scaled up elastic stiffness and failure force which delayed the breakage
and left no visible bridging spots. Deployment #3 showed a contact enhancement of 32% with
respect to the theoretical contact area. Deployment #4 predicted a contact enhancement slightly
higher than Deployment #3 (+2%). Deployments #3 and #4 demonstrated to be the closest to the
experimental results. The experimental contact area is not easily measurable, and it is not available
for comparison with the simulation results. The simulation results demonstrate the advantage of
using FE modeling to predict or estimate quantities, such as contact area, that cannot be obtained
directly from physical experiments.
Table 2: Contact Area after Completion of Deployment.

Deployment #
Parameter
1 2 3 4
Unconfined Inflation Cylindrical Contact Area (m²) 71.98 71.98 71.98 71.98
Confined Inflation Simulation Contact Area (m²) 81.51 91.49 95.27 96.27
Contact Enhancement % +14% +27% +32% +34%
Visible Bridging Spots 3 2 0 0

6. Conclusions
This study presents the development of finite element models to simulate the installation and
deployment of large confined inflatable structures. Models for folding, placement and deployment
of an inflatable structure used to seal a closed conduct representative of a tunnel segment were
presented. The strategy used for developing the models presented in this work can be used for further
evaluations including design optimization of the inflatable, alternative folding sequences, and
alternative ways to guide the air flow within the inflatable plug, as well as the influence of using
different airflow rates during the deployment and inflation process. Alternative confining perimeters
and additional features placed in the interior of the profile to maximize the sealing capacity of the
plug can be modeled as well. The deployment simulation results as well as the level of conformity
of the simulated inflatable structure to the confining perimeter are in excellent agreement with
7. References
1. Abaqus User’s Manual, Documentation Collection Version 6.11, Dassault Systémes Simulia Corp.,
Providence, RI.
2. Altair HyperWorks User’s Manual. Altair Engineering Inc., Troy. MI.
3. Barbero, E. J., E. M. Sosa and G. J. Thompson. “Testing of Full-Scale Confined Inflatable for the
Protection of Tunnels” Proceedings of the VI International Conference on Textile Composites and
Inflatable Structures, Structural Membranes 2013, Munich, Germany, October 9-11, 2013 (a).
4. Barbero, E. J., E. M. Sosa, X. Martinez and J.M. Gutierrez, “Reliability Design Methodology for
Confined High Pressure Inflatable Structures.” Engineering Structures. Volume 51, pp. 1–9, June,
2013 (b).
5. Barrie A., “Going Underground: Homeland Security Works on Tool to Prevent Tunnel Disasters”.
FoxNews.com, September 05, 2008. http://www.foxnews.com/story/0,2933,417461,00.html
6. Fountain, H., “Holding Back Floodwaters with a Balloon”, The New York Times/Science
Supplement, Nov. 19, 2012. http://www.nytimes.com/2012/11/20/science/creating-a-balloonlike-
plug-to-hold-back-floodwaters.html?pagewanted=all&_r=0
7. Martinez X., J. F. Davalos, E. J. Barbero, E. M. Sosa, W. Huebsch, K. Means, L. Banta, G.
Thompson, “Inflatable Plug for Threat Mitigation in Transportation Tunnels”. Proceedings of the
Society for the Advancement of Material and Process Engineering Conference (SAMPE 2012),
Baltimore, MD, May 21-24, 2012.
8. Peil K. L., E. J. Barbero, E. M. Sosa. “Experimental Evaluation of Shear Strength of Woven
Webbings”. Proceedings of the Society for the Advancement of Material and Process Engineering
Conference (SAMPE 2012), May 21-24, Baltimore, MD, May 21-24, 2012.
9. Sosa, E. M., G. J. Thompson and E. J. Barbero, “Testing of Full-Scale Inflatable Plug for Flood
Mitigation in Tunnels”. Transportation Research Record: Journal of the Transportation Research
Board, Volume 2407, Structures 2014, Vol. 2, pp. 59-67, 2014 (a).
10. Sosa, E. M., G. J. Thompson, E. J. Barbero, S. Ghosh, K. L. Peil. “Friction Characteristics of
Confined Inflatable Structures”, Friction, 2(4), pp. 365–390, November, 2014 (b).
11. Wang, J. T., and Nefske, O. J., “A new CAL3D Airbag Inflation Model,” SAE paper 880654, 1988.

CONSUMER PACKAGED
GOODS

Simplified Approaches for Modelling a Sealed and
Partly Fluid Filled Container
O. Valtiner, D. Pielmeier and C. Oblasser
ALPLA Werke Alwin Lehner GmbH & Co KG
Abstract: Containers (bottles) made of polymers are exposed to manifold loads during their life
cycle. They are open or sealed as well as empty or filled. Depending on the mentioned conditions
a container behaves different under loading. This effect can be observed in everyday life. Imagine
opening a shaken plastic bottle filled with a carbonated beverage. Abaqus provides simulation
methods for completely filled containers with either a pneumatic or a hydraulic fluid. Real
containers barely are totally filled with a hydraulic fluid. In general there is always a so called
head space filled with gas (pneumatic fluid). This paper presents two methods for modelling the
two-phase system (liquid and gas) inside a partly filled container using a fluid cavity instead of
meshing the volume inside the container. Therefore a one-phase system equivalent to the two-
phase system is described and applied to the model using two different methods – an explicit and
an implicit one. These methods are implemented by using subroutines which allow extending the
functionality of Abaqus. Applying these approaches allows cost efficient simulations of various
loads like excess pressure, vacuum, temperature change, top load, squeezing etc. for sealed, partly
filled containers. The paper describes the background and assumptions necessary to define the
one-phase system, verifies the model by comparing the FEA results with analytical solutions and
shows worked examples.
Keywords: Container, Bottle, Partly Filled Container, Consumer Packaged Goods, Hot Fill,
Fluid-Structure Interaction, Fluid Cavity, Multi-Phase Simulation, Ideal Gas and Bulk Modulus.
1. Introduction
To accelerate and improve the development of new containers (bottles) made of polymers the use
of numerical simulations nowadays is state of the art. As short as the lifetime of containers for
consumer packaged goods (CPG) seems to be, as manifold are the conditions and loads they are
exposed to. Each condition and loading type requires different simulation techniques to ensure
efficient development. Typically process simulations of extrusion blow moulding process (EBM)
or stretch blow moulding process (SBM) are performed to predict the wall thickness distribution
off a container. Product simulations predict the behaviour of the bottle exposed to different loads
and different conditions. Some product simulations correspond directly with tests usually
performed during customer approval procedure or for quality assurance. Classically these are:
• Top load test / simulation
• Squeezing test / simulation
• Bulging test / simulation

Figure 1 shows a typical top load test configuration where a bottle is located between a bearing
plate on the bottom and a bearing plate on the top. The top plate is moved downward and the load-
displacement curve is recorded.
Figure 1. Top load test (left), simulation (right).
Other product simulations are not related to physical tests or are replacing expensive and time
consuming tests. A question of interest for such a simulation can be the behaviour of containers
filled with hot liquid and sealed at high altitude, afterwards while cooling they are transported to
see level where they are palletized. Simulations of this kind are required in order to ensure a
proper performance of the container during its entire life cycle. Looking at a containers life cycle
there are phases during which the container is either open or sealed. In both cases the container is
empty or filled with liquid. Even in a filled container there is always a so called head space
containing gas. Figure 2 illustrates these four possible conditions.
Simulations dealing with condition (a) in Figure 2 just have to consider the mechanical properties
of the material the container is made of. In case of condition (b) for many simulations fluid
sloshing and gravity effects of the fluid mass can be neglected and the fluid can be modelled by
applying hydrostatic pressure on the bottle. Condition (c) can be well modelled in
Abaqus/Standard and Abaqus/Explicit using surface-based fluid cavity technique. It provides the
coupling between the deformation of the fluid filled structure and the pressure exerted by the
contained fluid on the cavity boundary of the structure (Dassault Systémes Simulia Corp., 2014).
This technique supports two different types of fluids:
• Pneumatic fluids
• Hydraulic fluids
Pneumatic fluid can be used to model condition (c) in Figure 2.
Condition (d) in Figure 2 is characterized by the presence of two different fluids – a pneumatic
fluid (gas) and a hydraulic one (liquid).

Gas Gas
Gas Gas
Liquid Liquid
(a) (b) (c) (d)
Figure 2. Conditions of a container during its life cycle –

(a) empty/open, (b) partly filled/open, (c) empty/sealed, (d) partly filled/sealed.
For cases where fluid sloshing and gravity effects of the fluid can be neglected a surface-based
fluid cavity filled with a single artificial bulk (see Figure 3) can be used to model this condition. In
the following two approaches achieving this goal are presented.
2. Modelling approaches for partly filled/sealed containers

In order to avoid expensive meshing of the volume inside the container and describing liquid and
gas by constitutive models an approach using surface-based fluid cavity is chosen.
Surface-based
Gas fluid cavity
Artificial bulk
Liquid Liquid
(a) (b)
Figure 3. Partly filled/sealed container –

(a) two-phase system and (b) equivalent one-phase system.
To describe the properties of the artificial bulk inside the fluid cavity two different approaches are
presented.

The first approach is an implicit one using the user subroutine UFLUID of Abaqus/Standard to
define the constitutive model of the artificial bulk.
The second approach – an explicit one – can be applied to Abaqus/Standard as well as to
Abaqus/Explicit. Therefore the artificial bulk is considered as hydraulic fluid but with a fluid
cavity pressure (gauge pressure) dependent bulk modulus 𝐾(𝑝).
2.1 Implicit approach using subroutine UFLUID
User subroutine UFLUID serves to define a fluid constitutive model being used for hydrostatic
fluid elements used in cavities. The subroutine requires the definition of the fluid density 𝜌(𝑝, 𝜃),
where 𝑝 is the fluid cavity pressure and 𝜃 is the fluid cavity temperature, its fluid pressure
compliance
dρ
C p = − ρ −2 (1)
dp
and the fluid temperature compliance 𝐶𝜃 defined as (Dassault Systémes Simulia Corp., 2014):
dρ
Cθ = − ρ −2 (2)
dθ
2.1.1 Fluid density 𝜌(𝑝, 𝜃)
Let’s consider a container as shown in Figure 4 (a) exposed to an ambient pressure 𝑝𝑎𝑚𝑏 . Inside
the container an initial fluid cavity pressure (initial gauge pressure) 𝑝𝑖𝑛𝑖 is acting. The gas with the
initial volume 𝑉𝐺,𝑖𝑛𝑖 and initial temperature 𝜃𝑖𝑛𝑖 is considered to be an ideal gas with the specific
gas constant 𝑅𝑠 . The absolute initial pressure 𝑝�𝑖𝑛𝑖 results from the ambient pressure 𝑝𝑎𝑚𝑏 and the
gauge pressure 𝑝𝑖𝑛𝑖 to:
p ini = pamb + pini > 0 (3)
According to the equation of state (EOS) of an ideal gas its density is (Cerbe & Wilhelms, 2005)
p ini
ρG ,ini = . (4)
θini * Rs
The gas mass 𝑚𝐺 inside the container remains constant and can be calculated as follows:
mG m=
= G ,ini ρG ,ini *VG ,ini (5)
The liquid is considered to be incompressible and the volume 𝑉𝐿 independent of the temperature.
Hence its density 𝜌𝐿 and mass 𝑚𝐺 is constant:
mL ρ=
= L *VL const. (6)
The total container volume 𝑉𝑖𝑛𝑖 of the initial configuration (a) in Figure 4 is the sum of the initial
gas volume 𝑉𝐺,𝑖𝑛𝑖 and the constant liquid volume 𝑉𝐿 . In addition the initial fraction of liquid
𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 in the cavity is defined by:

VL VL
L frac=
,ini = (7)
Vini VG ,ini + VL
Looking at the deformed configuration (b) in Figure 4 the state of the gas is described by the fluid
cavity pressure 𝑝 leading to the absolute pressure
p= pamb + p > 0, (8)
the temperature 𝜃 and the volume

p ini *VG ,ini *θ
VG = (9)
θini *p
which results from the EOS of an ideal gas. The process leading from the initial configuration to
the deformed one is considered to be isotherm.
Ambient pressure
Gas: Gas:
, , , , , ,
, , Gas Gas , ,
Container volume
Container volume
Liquid: Liquid Liquid:

isotherm Liquid
, ,
(a) (b)
Figure 4. Partly filled/sealed container –

(a) initial configuration and (b) deformed configuration.
The total mass 𝑚 inside the deformed configuration (b) in Figure 4 is calculated from the mass of
the gas 𝑚𝐺 and the mass of the liquid 𝑚𝐿 . Taking equations (5) and (6) leads to:
=m ρG ,ini *VG ,ini + ρ L *VL (10)
The total volume 𝑉 is the sum of the gas volume 𝑉𝐺 and the constant liquid volume 𝑉𝐿 .
Considering equation (9) leads to:
=V + VL (11)
θini *p

Since the density 𝜌(𝑝, 𝜃) of the deformed configuration (b) in Figure 4 is required one now can
take equations (10) and (11) to obtain:
m ( ρG ,ini *VG ,ini + ρ L *VL ) *θini * p

ρ ( p ,θ=
) = (12)
V p ini *VG ,ini *θ
+ VL *θini * p
Taking advantage of equation (8) and of 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 defined in equation (7) reforming equation (12)
leads to the following expression:
( ρG ,ini *(1 − L frac ,ini ) + ρ L * L frac ,ini ) *θini *( pamb + p )

ρ ( p, θ ) = (13)
p ini *(1 − L frac ,ini ) *θ
+ L frac ,ini *θini *( pamb + p )
Equation (13) is the required expression for the density of the artificial bulk (see Figure 3) to be
used in the user subroutine UFLUID.
2.1.2 Fluid pressure compliance 𝐶𝑝 and fluid temperature compliance 𝐶𝜃
As described in section 2.1 the user subroutine UFLUID requires the definition of the fluid
pressure compliance 𝐶𝑝 and the fluid temperature compliance 𝐶𝜃 as defined in equations (1) and
(2). Applying equation (13) to the definition of 𝐶𝑝 and 𝐶𝜃 one obtains following two expressions
needed for the user subroutine UFLUID:
p ini *(1 − L frac ,ini )*θ
Cp = −
( ρG ,ini *(1 − L frac ,ini ) + ρ L * L frac ,ini ) *θini *( pamb + p ) 2
(14)
p ini *(1 − L frac ,ini )

Cθ = (15)
( ρG ,ini *(1 − L frac ,ini ) + ρ L * L frac ,ini ) *θini *( pamb + p )
Equations (3), (4), (7), (13), (14) and (15) describe the constitutive model of the artificial bulk in
Figure 3 consisting of the liquid and the gas phase.
2.1.3 Application of fluid properties and initial conditions to the model
Physical properties of each single phase (gas and liquid) have to be hard-coded into the user
subroutine UFLUID. These are the specific gas constant 𝑅𝑆 , the ambient pressure 𝑝𝑎𝑚𝑏 , the fluid
density 𝜌𝐿 and the initial liquid fraction 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 of the initial container volume.
As Abaqus/Standard hands over the cavities reference node number to the subroutine UFLUID the
node number could be used to hand over one or even more of the previous described hard-coded
parameters to the subroutine UFLUID. This workaround can make the subroutine more flexible
but requires attention to the node numbering of the model.
The initial fluid cavity pressure 𝑝𝑖𝑛𝑖 and the initial temperature 𝜃𝑖𝑛𝑖 are applied to the cavities
reference node.

2.1.4 Verification of the user subroutine UFLUID
In order to evaluate the accuracy of the previous described approach and the functionality of the
subroutine UFLUID the simple model shown in Figure 5 is used. Therefore the fluid cavity
pressure 𝑝 obtained from the simulation is compared with the analytical solution.
Top plate
Gas
(20%)
100mm
Liquid
(80%)
Bottom
plate 50mm
Figure 5. Rigid cylindrical tube, filled with 20% gas (air) and 80% liquid (water).
The model consists of a tube which is rigid in radial direction (achieved by constraining
translational DOFs in 𝑥 − and 𝑦 −direction) and sealed by rigid top- and bottom plates (achieved
by constraining all DOFs). Tube and plates are modelled using S4R structural shell elements with
a linear elastic material and no thermal expansion. The volume inside the tube is filled with 20%
air and 80% water. I.e. the initial fraction of liquid 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 = 0.80. The following properties are
used:
• Ambient pressure 𝑝𝑎𝑚𝑏 = 0.101325 MPa
• Specific gas constant 𝑅𝑆 = 288.027 J/kgK
• Fluid density 𝜌𝐿 = 1.0𝑒 − 9 t/mm³

Top plate
Gas
Gas
(20%)
Gas
120mm
100mm
90mm
Liquid
Liquid Liquid
(80%)
(a) (b) (c)
Figure 6. Cylindrical tube – (a) initial condition, (b) elongated 𝑽𝑮,𝒊𝒏𝒊⁄𝑽𝑮 = 𝟎. 𝟓 and
(c) compressed 𝑽𝑮,𝒊𝒏𝒊 ⁄𝑽𝑮 = 𝟐. 𝟎.
The initial temperature is 𝜃𝑖𝑛𝑖 = 350 K. Two different values for initial gauge pressure are
considered: 𝑝𝑖𝑛𝑖 = 0.0 MPa and 𝑝𝑖𝑛𝑖 = 0.101325 MPa. The inside of the tube is modelled by a
surface based fluid cavity.
The following types of loads and combinations thereof are imposed to the tube:
• Change of temperature from 𝜃𝑖𝑛𝑖 = 350 K to 𝜃 = 280 K
• Elongation or compression of the tube by moving the rigid top plate up- or
downward (see Figure 6)
To obtain the analytical solution for the fluid cavity pressure 𝑝 the gas filled head space is
described by the equation of state (EOS) of an ideal gas (Cerbe & Wilhelms, 2005) and leads to
p ini *VG ,ini ( pamb + pini )*VG ,ini p *VG
= = . (16)
θini θini θ
Considering equations (3) and equation (8) leads to the analytical expression for the fluid cavity
pressure:
( pamb + pini )*θ VG ,ini

=p * − pamb (17)
θini VG

Table 1 summarises the comparison between the fluid cavity pressure 𝑝 calculated by simulation
using subroutine UFLUID and the analytical solution according equation (17) for several load
cases. Values marked with a star (*) indicate that these are the imposed loads to the model. E.g. in
case 1 the fluid cavity temperature 𝜃 is decreased from the initial temperature 𝜃𝑖𝑛𝑖 = 350 K to
𝜃 = 280 K while the volume of the gas is kept constant.
Table 1. Comparison of simulation and analytical solutions for gauge pressure 𝒑.

Initial configuration Deformed configuration
𝑽𝑮,𝒊𝒏𝒊 𝒑 [𝐌𝐏𝐚] 𝒑 [𝐌𝐏𝐚]
Load case 𝒑𝒊𝒏𝒊 [𝐌𝐏𝐚] 𝜽 [𝐊] [−]
𝑽𝑮 Simulation Analytical
Case 1 0.00 280* 1.0 -0.020265 -0.020265
Case 2 0.00 280* 2.0* 0.060795 0.060795
Case 3 0.00 280* 0.5* -0.060795 -0.060795
Case 4 0.101325 350 0.5* 0.00 -4.2e-9
Case 5 0.101325 280* 1.0 0.060795 0.060795
Case 6 0.101325 280* 2.0* 0.222915 0.222915
Case 7 0.101325 280* 0.5* -0.020265 -0.020265
Looking to the last two columns of Table 1 it can be seen that simulations and analytical solutions
are identical.
2.1.5 Application of the method to a real containers top load simulation
The modelling approach is applied to a real bottles top load simulation as shown in Figure 7. In
this type of simulation the top plate is moved downward with constant velocity and the force vs.
displacement curve is observed.
The bottle is discretised by means of S4R structural shell elements. Friction between top plate and
bottle as well as between bottle and bottom plate is considered.
Three simulations are performed:
a. Bottle empty/open
b. Bottle partly filled/open with initial liquid fraction 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 = 0.92
c. Bottle partly filled/sealed with initial liquid fraction 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 = 0.92
Simulations (a) and (b) don’t require the described modelling approach as the bottle is open and
the gas phase doesn’t activate any pressure when its volume changes. The simulations (a) and (b)
just serve as references for simulation (c). For all three simulations the force vs. displacement
curves are plotted to be compared with each other.
The simulation (a) of the empty bottle consists of a single step during which the top plate is moved
downward (in negative 𝑧 − direction) with constant velocity (top load simulation).

Top plate
Bottom
plate
Figure 7. Model for top load simulation.
The simulations (b) and (c) of the partly filled bottle are divided into two steps (see Figure 8):
• Bulging simulation:
The bottle is open and exposed to hydrostatic pressure leading to a deformation of
the bottle (bulging). The initial liquid fraction 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 is determined from the
deformed bottle shape resulting from this step.
• Top load simulation: As described for simulation (a).
Step 1 Step 2
(Bulging simulation) (Top load simulation)
Figure 8. Partly filled/sealed bottle – not deformed (left),

deformed after bulging simulation (centre) and after top load simulation (right).
Figure 9 shows the force vs. displacement curves of all three simulations.

Figure 9. Force vs. displacement curves of top load simulations, deformed bottle.
The simulations (a) and (b) of the open bottle show almost the same top load peak. The partly
filled/open bottle shows at the first part of the curve a little stiffer behaviour as the empty one. The
partly filled/sealed bottle (c) shows an approx. 12% higher top load peak as the other simulations.
2.2 Explicit approach using pressure dependent bulk modulus 𝑲(𝒑)
For surface based fluid cavities the hydraulic fluid model is used to model nearly incompressible
fluid behaviour and fully incompressible fluid behaviour. Compressibility is introduced by
assuming a linear pressure-volume relationship. The required parameters for compressible
behaviour are the bulk modulus 𝐾 and the reference density 𝜌𝑅 (Dassault Systémes Simulia Corp.,
2014). As shown in in Figure 3 the two-phase system should be modelled as an equivalent one-
phase system. The one phase system (artificial bulk) cannot be considered as incompressible as the
presence of the gas brings compressibility into the system. But with increasing compression of the
gas the entire system gets stiffer. Considering a pressure dependent bulk modulus 𝐾(𝑝) instead of
a constant one meets this fact.
2.2.1 Pressure dependent bulk modulus 𝐾(𝑝)
A fluids bulk modulus 𝐾 is described by (Dassault Systémes Simulia Corp., 2014):
 V ( p, θ ) − V ( p = 0, θ ) 
p = −K *  (18)
 V ( p = 0, θini )
 

For the following considerations the liquid phase is assumed to be incompressible and its thermal
expansion is ignored. The gas phase is treated as ideal gas and the process is supposed to be
isotherm. Using these assumptions the volume 𝑉(𝑝, 𝜃) is the same as described in equation (11).
Using equation (8) here leads to:
V ( p, θ )
= + VL (19)
θini * ( pamb + p )
Converting equation (18) the pressure and temperature dependent bulk modulus equals:
 V ( p = 0, θini ) 
K ( p, θ ) = − p*   (20)
V 0, θ ) 
 ( p, θ ) − V ( p =
For a constant fluid cavity temperature 𝜃 equation (20) represents a linear relationship between the
fluid cavity pressure 𝑝 and the bulk modulus 𝐾(𝑝).
35
30
25
20 273.15
Bulk modulus
15 313.15
353.15
10
393.15
5
0
0 0.5 1 1.5 2
Fluid cavity pressure
Figure 10. Bulk modulus 𝑲(𝒑) for 𝒑�𝒊𝒏𝒊 = 𝒑𝒂𝒎𝒃 = 𝟎. 𝟏𝟎𝟏𝟑𝟐𝟓 𝐌𝐏𝐚,
𝜽𝒊𝒏𝒊 = 𝟐𝟗𝟑. 𝟏𝟓 𝐊 and 𝑳𝒇𝒓𝒂𝒄,𝒊𝒏𝒊 = 𝟎. 𝟗𝟐 .
Figure 10 shows exemplarily the linear relation of equation (20) for a container initially filled with
92% liquid (water) and 8% gas (air).
2.2.2 Thermal expansion
Thermal expansion can be expressed in terms of the fluid density

ρ ( p = 0, θini )
( p 0,=
ρ= θ) , (21)
1 + 3*α (θ ) * (θ − θ 0 ) − 3*α (θini ) * (θini − θ 0 )
where 𝜃0 is the reference temperature for the coefficient of thermal expansion and 𝛼(𝜃) is the
mean (secant) coefficient of thermal expansion (Dassault Systémes Simulia Corp., 2014).
Defining
θ 0 ≡ θini (22)
and reforming equation (21) leads to:
 ρ ( p = 0, θini )  1
=α (θ )  − 1 *

(23)
=  ρ ( p 0, θ )  3* (θ − θini )
Equation (13) describing 𝜌(𝑝, 𝜃) can be used in equation (23) to calculate 𝛼(𝜃). It becomes
obvious that 𝛼(𝜃) is independent of the temperature 𝜃. It depends on the ambient pressure 𝑝𝑎𝑚𝑏 ,
the initial conditions 𝜃ini , 𝑝ini and 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 . Figure 11 shows exemplarily 𝛼(𝜃) for different initial
conditions 𝑝𝑖𝑛𝑖 and 𝐿𝑓𝑟𝑎𝑐,𝑖𝑛𝑖 . Ambient pressure 𝑝𝑎𝑚𝑏 = 0.101325 MPa and initial temperature
𝜃ini = 293.15 K are kept constant.
1.2E-03 1.2E-03
1.0E-03 1.0E-03
Coefficient of thermal expansion
Lfracini=0.99
8.0E-04 8.0E-04
Lfracini=0.8
6.0E-04 6.0E-04
4.0E-04 4.0E-04 Lfracini=0.5
2.0E-04 2.0E-04 Lfracini=0.01
0.0E+00 0.0E+00
273 323 373 423 273 323 373 423
Fluid cavity temperature
Figure 11: Coefficient of thermal expansion for different initial conditions

𝒑𝒊𝒏𝒊 = −𝟎. 𝟎𝟓 𝐌𝐏𝐚 (left) and 𝒑𝒊𝒏𝒊 = 𝟏. 𝟎𝟎 𝐌𝐏𝐚 (right).
2.2.3 Application of fluid properties and initial conditions to the model
The idea of the explicit approach is to take the fluid cavity pressure 𝑝𝑖 from a converged increment
𝑖, calculate the corresponding bulk modulus 𝐾𝑖+1 (𝑝𝑖 ) to be used for the following increment 𝑖 + 1.
This approach takes advantage of the possibility to include field variables into the definition of the
bulk modulus in Abaqus/Standard as well as in Abaqus/Explicit. This way it is feasible to define

the bulk modulus depending on a field variable representing the fluid cavity pressure. As shown in
Figure 10 the bulk modulus is a linear function of the fluid cavity pressure. Hence two points of
the curve are sufficient to describe the bulk modulus for a constant temperature. This definition
can be completed by adding definitions for different discrete temperatures.
One can use the subroutine UAMP to define a customized amplitude representing the fluid cavity
pressure as a function of time. Therefore a sensor measuring the fluid cavity pressure 𝑝 is defined
in the history output. The so defined amplitude is used to specify the time history of the
additionally defined field variable corresponding to the field variable used in the definition of the
bulk modulus as described above.
2.2.4 Verification of the method
In order to evaluate the method same model as shown in Figure 5 is used. The results from the
simulation are compared with the results from simulations using the implicit approach (UFLUID)
described in section 2.1. The initial conditions and the material properties are the same as in
section 2.1.4.
Gas (20%)
Gas,
Liquid
Liquid
(80%)
Time = 0 sec Time = 1 sec
Figure 12. Fluid cavity pressure for different maximal time increments.
The explicit approach keeps the pressure dependent bulk modulus 𝐾(𝑝) constant within a time
increment. Hence the choice of the maximal allowed time increment has a strong influence of the
accuracy of the simulation. In contrast the implicit approach (UFLUID) considers the change of
constitutive relations within the time increment.
Figure 12 shows fluid cavity pressure vs. time plots for different maximal time increments used
for the explicit definition of 𝐾(𝑝). It also compares the result with the implicit approach
(UFLUID). Figure 12 corresponds to case 2 as described in Table 1.

It is quite clear that for very small maximal time increments both approaches lead to the same
result. As the maximal time increment increases the two approaches differ from each other.
3. Conclusions and discussion

Two approaches for modelling the two-phase system (liquid and gas) inside a partly filled and
sealed container using a surface defined fluid cavity are presented – an implicit and an explicit
one. The implicit approach takes advantage of user subroutine UFLUID describing the constitutive
model of the two-phase system. The explicit approach uses the hydraulic fluid model of
Abaqus/Standard and defines a pressure dependent bulk modulus 𝐾(𝑝) using a field variable
representing the fluid cavity pressure from the last increment. To achieve this user subroutine
UAMP in combination with a sensor output is needed. The background of both approaches is
described and comparisons with analytical solutions for a simple model prove the accuracy of both
methods.
The presented approaches consider following assumptions:
• The process is isotherm.
• The liquid phase is incompressible and its thermal expansion is neglected.
• The gas phase is considered to be an ideal gas.
• Sloshing and gravity effects of the fluid mass can be neglected.
• The initial temperature and total initial pressure (sum of ambient and initial gauge
pressure) of the gas describe the reference state for density and thermal expansion.
4. References
Cerbe, G., & Wilhelms, G. (2005). Technische Thermodynamik. München: Hanser.
Dassault Systémes Simulia Corp. (2014). Abaqus Analysis User's Guide, Version 6.14.
Providence, RI, USA: Dassault Systémes Simulia Corp.

Modeling the Mechanical Behaviour of Some
Complex Polymeric Materials by Relying on the
Nonlinear Capability in Abaqus
Colmar Wocke
Dow Europe GmbH, Bachtobelstrasse 3, CH-8810 Horgen, Switzerland
Abstract: Polymeric materials are complex in their mechanical behaviour; inter alia, they exhibit
large strain, anisotropic and irreversible response which is often accompanied by stable localized
necking behaviour. When developing suitable computational models for such materials, one
cannot do this without the underlying support of a FEM program that offers inherently non-linear
solution capabilities.
Within Dow there has been some internal activity, concerned with the development of such a
polymeric material computational model. Excellent agreement has been obtained in modeling
simple tension tests of linear low density polyethylene (LLDPE) and polyethylene terephthalate
(PET) films. In addition to showing the results of these model calibrating efforts, in the case of the
LLDPE film, some simulation results from application areas for the developed model are
presented. These application areas typically come from within advanced, flexible packaging
applications where minimum usage of material has an environmental benefit. The severe
deformation and generally complex behaviour able to be captured by the model, highlights how
indispensible the core capabilities of Abaqus were in developing the Dow modeling capability.
Keywords: Abaqus, Polymers, Consumer Packaging, Anisotropic, Necking, Non Linear, Fracture,
Environmental, Wastage, Plastic

1. Introduction
Modern polymers offer many advantages; this is especially true in consumer packaging
applications. We, as consumers, have become accustomed to purchasing products that are well-
protected from damage and/or appear attractive in see-through packaging. For products that may
spoil, such as fruit and vegetables, remaining fresh on the shelves for a very long time is an
advantage; less has to be discarded by the retailer and this in itself represents an environmental
benefit. In both the developed and developing world ca. one-third of all plastics consumption goes
into packaging. So, any reduction in wastage of packaging material would have a direct and
beneficial impact on the environment. Much of packaging design and even material specification
contains a significant trial-and-error component, relying heavily on the accumulated experience of
those involved along the value-chain. Finite Element Method simulation can clearly enable new
and economical possibilities in packaging design to be studied in advance, coupled to material
specifications that are perhaps better-suited to resisting the externally-imposed environmental
loads. Correctly applied and meaningful simulation can help reduce wastage and have a positive
economic benefit for and on the environment. A generic packaging FEM example, with some
important details for subsequent Abaqus analysis highlighted, is shown in Figure 1.
For the Finite Element Method simulation to be useful as a design aid, the material properties
themselves must be meaningfully integrated into a computational model to the level of detail
required of the analysis prediction; the computational model must enable the real materials to be
represented. This is no easy task, since the challenges and complications are tough and many;
some of these are noted in the following non-exhaustive list:
1. The computational model must deal with severely non-linear behaviour (thin films, very
flexible, large strains of ca. 800%, with a large permanent plastic set and oftentimes
exhibiting local material instability).
2. Failure and fracture needs to be represented in order to investigate misuse scenarios or
intentional separation.
3. Individual films are laminated together, thus requiring a shell-type model; there is
internal interaction between film layers via internal adhesive layers; all of these
constituents have differing individual responses.
4. Film orthotropic behaviour needs to be sensibly modeled – not only in the film’s small
strain response.
5. A real laminated structure has imperfections, together with some mostly-unknown levels
of internal stress.
6. There are theoretical shortcomings that have an impact on what can and cannot be
modeled.
7. Often one is missing test data – this is usually why a simulation is required, else the test
would suffice.
8. The analyst is usually severely time/resource constrained and cannot easily perform
material model development and real problem analysis at the same time, unless the
materials are very simple to describe.

Within Dow there has been an internal initiative to develop suitable material computational
models for solving thin film, laminated structures using polymeric materials. Our initial candidate
materials have been linear low density polyethylene (LLDPE) and polyethylene terephthalate
(PET). The intent of the paper is not to discuss the detailed computational model development –
this is proprietary to Dow, but rather to illustrate some of the analysis possibilities that are enabled
by such a model. The paper shows some results from tensile test calibration of the developed
model in the case of the far more-challenging LLDPE film and also some detailed applications
thereof; we solve a pouch-dropping problem to demonstrate an intended future use, as well as a
real-life bag handle application.
Doubled
thickness
S3R, S4R elements
R3D4 elements
Figure 1. A FEM example of a modern, flexible package made from laminated

polymeric materials.
2. Overview of pertinent capabilities offered by Abaqus

Abaqus has, in its suite of capabilities and procedures, much of what is required to solve our
problem; our analysis requirements are listed in the Introduction. At its core, Abaqus is a non-
linear code; it is not a linear code that has some added-on features of non-linearity. In order to deal
with failure and fracture during the rapid event of a bag impacting the floor, Abaqus/Explicit is
used as the analysis development code.

Polymeric films are usually designated as having MD-orientation or CD-orientation; these
mnemonics refer to the machine direction and cross-direction, respectively. The mechanical
response is generally different along these directions and the nomenclature stems from the
processes used to manufacture these films – they’re typically manufactured on a blown film line in
the case of LLDPE and cast film line in the case of PET. It is relatively-straightforward to specify
such a composite shell, as illustrated in Figure 2, using the keyword-form:
**
** Define the LAMINATED Shell wall
**
*SHELL SECTION, ELSET = ALL_ELEMENTS, COMPOSITE,
ORIENTATION = REFERENCE_ALIGNED
<LLDPE_Shell_Thickness>,5,Material_DOWLEX_VUMAT,0.0,SHELL_LLDPE
<PET_Shell_Thickness>,5,Material_PET_VUMAT,0.0,SHELL_PET
**
REFERENCE ALIGNED
Projection
Local
2-axis
Local 1-axis
PET layer (strain to ca. 30%)
MD
LLDPE layer
(strain to ca.
800%, with
severe local CD
instability)
Figure 2. An example of using the projection of axes when specifying a laminated
orthotropic film structure making up a pillow-pouch wall.
The alignment of the material is generally along some “axis” of the bag; in the case of a non-
structured mesh it is not easy to specify these material orientations across the complicated mesh;
Abaqus offers help here, in that the projections onto the shell surface from the
REFERENCE_ALIGNED system, as illustrated in Figure 2, are used. Care must be exercised here – the

projection swops around when the angle between the local element normal to the surface and the
newly defined system with the axis labeled REFERENCE_ALIGNED is ≤ cos0.1º - see for
example, in the case of small-strain shells, Sec. 3.6.3 in the Abaqus Theory Manual (Simulia,
2013).
Abaqus offers some considerable extensibility in its providing an interface, enabling the
FORTRAN (or C) coding of material behavior falling outside that already represented in the
material library. The keyword-form for invoking this is:
*MATERIAL, NAME = MATERIAL_DOWLEX_VUMAT

*DEPVAR
11
**
** Optional data lines to ease output interpretation
**
*USER MATERIAL, CONSTANTS = 7
<Material_DOWLEX_1>,<Material_DOWLEX_2>,<Material_DOWLEX_3>,
<Material_DOWLEX_4>,<Material_DOWLEX_5>,<Material_DOWLEX_6>,
<Material_DOWLEX_7>
The User Subroutine is called at every material point during the analysis and updates the stress
tensor and the material state variables (11, in this case), from their initial values and from the
kinematic increments passed into the routine. The User Subroutine then defines the generally
complex behavior of the material that needs to be adequately represented.
Content-filled packages can be relatively-easily analyzed using the Eulerian or SPH analysis
capability. The “Impact of a water-filled bottle” analysis example, documented as Problem 2.3.2
in the Abaqus Example Problems Manual (Simulia, 2013) illustrates the procedure very well for
an analyst unaccustomed to dealing with these problems.
Geometric imperfections can be included by perturbing the perfect mesh with some fraction of the
eigenmodes of the structure/package; these are first extracted in a frequency step in
Abaqus/Standard. The general workflow, for an imperfection-seeded dropping of a
package/structure onto the floor, in keyword-form is as presented below. Initial stresses can be
included as initial model data; however the specification of this is less than trivial for a real bag or
pouch structure.
2.1 Extracting frequencies
*HEADING
Extracting Layered Frequencies

*INCLUDE, INPUT = BAG_MODEL.INP
*STEP, NAME = EXTRACT_FREQUENCIES, PERTURBATION
*FREQUENCY, EIGENSOLVER=LANCZOS, NORMALIZATION=DISPLACEMENT
, <begin_frequency>, <last_frequency>, , ,
*NODE FILE, MODE = <BEGIN_MODE>, LAST MODE = <LAST_MODE>
U
*OUTPUT, FIELD, VARIABLE=PRESELECT
*END STEP
2.2 Adding imperfections and dropping
*HEADING
Dropping after adding imperfection
*INCLUDE, INPUT = BAG_MODEL.INP
**
*INITIAL CONDITIONS, TYPE = STRESS (DIFFICULT, IN PRACTICE)
**
*IMPERFECTION, FILE = EXTRACT_FREQUENCIES, STEP = 1
<mode_1>,<scale_factor_mode_1>
<mode_2>,<scale_factor_mode_2>
.
.
<mode_n>,<scale_factor_mode_n>
*STEP, NAME = DROP
*DYNAMIC, EXPLICIT
, .1
*END STEP
3. Model evaluation and partial calibration

3.1 Strip tension test
We present the results from our model in predicting the very large strain, slow-speed tensile
stretching of a thin strip of DOWLEX™ LLDPE film; the strip is 100mm long by 15mm wide and
the film thickness in this case is 130 microns (thinner films have been analyzed quite
successfully). The analysis has been conducted in Abaqus/Explicit 6.13-3 using S4R elements and
treating/modeling the problem as quasi-static, through suitable monitoring of the internal energy

and kinetic energy after mass scaling. The test data collected from this test, inter alia,
supplemented other input data for the model calibration and we use the average of 5 such tests as a
representative curve for the experimental results; the agreement between all 5 curves was
excellent, only deviating in the ultimate strain that each sample was able to sustain.
When pulled in the CD-orientation, the sample experiences significant localized necking; in the
case of MD-orientation the neck is far more diffuse and tends to spread across the whole length of
the sample; this is in agreement with what is observed in practice. In both cases, the neck is
stable, but the degree of localization is significantly different, as shown in Figure 3.
CD MD
Figure 3. Differences in tensile strip necking, aligned in CD- and MD-orientation.
The influence of imperfections for the two orientations is also different. There are not only
geometric imperfections in such a film, but also some unknown distribution of internal stress from
the manufacturing operation. Both of these effects have been accounted for by including only
geometric imperfection; a reasonable scaling of the higher-frequency eigenmodes was applied.
The MD-orientation tends to smooth out imperfections with increased stretch of the tensile strip,
whereas the CD-orientation retains them and draws them into the stable necking region, as shown
in Figure 4. This drawing-in of the imperfections, causes a marked wrinkling of the film in the
neighbourhood of that zone, which can be quite clearly seen with the naked eye in the
experimental tests. The computational model development and tuning was performed on the basis
of assuming no imperfections in the film; any inclusions of imperfections in the analysis
represents degradation that can be expected in its response, should the film deviate more severely
from an accepted manufacturing norm.

Figure 4. Local wrinkling of tensile strip when pulled in CD-orientation, after
imperfections (eigenmode perturbations) have been included in the mesh.
Characteristic „blip“
Figure 5. Analysis vs. Experiment when pulled in CD-orientation.

The predicted engineering stress vs. engineering strain for the tensile strip in comparison to the
measured results is excellent; this is especially true in the early part of the curve up to an
engineering strain of ca. 4.0 in the case of the CD-orientation and for virtually the whole extent in
the case of MD-orientation. The introduced geometric imperfections have very little effect in the

MD-orientation while for the CD-orientation, the extent of the stable necking zone in the curve in
Figure 5 is influenced. The introduction of imperfections shortens the marked characteristic “blip”,
which is so typical of LLDPE materials. The engineering values were obtained from the analysis
results of the “crosshead” displacement and force and suitably converting these, using the initial
dimensions of the sample. In the model, the “crosshead” was a single node, tied to the free end of
the tensile tab using a *MPC type BEAM. The results are compared to experiment in Figures 5 and 6.
Figure 6. Analysis vs. Experiment when pulled in MD-orientation.
The same computational model, only suitably oriented is thus able to predict two different
responses for such a DOWLEX™ LLDPE thin film; this is an important analysis result, since in a
real bag or pouch with a complex stress field the internal load apportioning is moderated by the
stiffness of the film. The degree to which external loading is absorbed along the two directions in
turn affects the propensity for failure along that direction. In addition, the appearance of the neck
and its marked difference along the two directions influences the internal stress distribution of any
laminate made from these (and other) materials. The necking is clearly a complex, non-linear
phenomenon that develops and extends as a function of strain and material orientation.
3.2 Elmendorf tearing test
The Elmendorf tearing test (ASTM D1922 : Propagation Tear Resistance of Plastic Film and Thin
Sheeting) is a standard test used in industry for assessing the toughness of polymeric films or
laminates having an initial crack; a sample is torn at high rate in a dynamic testing machine,

similar to a Charpy test for toughness testing of metals. The test is used in a comparative sense,
when ranking the performance of different candidates. The sample is torn, subjecting the crack
area to mainly antiplane shearing/tearing; thus Mode III in the language of crack opening
displacement measurements. The Abaqus model, used the “seam crack” feature in CAE with
which to generate the initial crack; the outline of the model is shown in Figure 7.
Seam crack
Figure 7. Elmendorf tearing test specimen, showing seam crack and reduced
density plot of the material orientation vectors.
This test is a challenging numerical test; it is highly-dynamic, the films are thin and flimsy with
significant failure. The problem is run with no form of mass scaling; thus the true dynamics in the
problem apply. PET-film is usually co-laminated with LLDPE in practice, since the PET-film fails
at a significantly lower strain than does the LLDPE. We can certainly predict improvements in
resistance of the PET through lamination with LLDPE, as shown in Figure 8 but it is very difficult
to get these models to run to completion. In this figure, the crack that originally ran for
considerably longer was halted by the addition of the LLDPE-layer in the laminate. Remedies that
have been tried, with varying degrees of success, is to treat the model as a membrane through
permanent removal of degrees of freedom 4,5 and 6 and adjusting the hourglass stiffness to the
maximum value recommended by Abaqus, using the keyword:
*SECTION CONTROLS, NAME = STABILIZER, HOURGLASS = RELAX STIFFNESS

1. , 3. , 3.
The error message encountered is either “Excessive rotation increment at one node” or perhaps
“Excessive distortion of element”, even in the case of using “standard” material descriptions. The
standard diagnostic that is printed out by Abaqus, does suggest that the model may have failed
completely. When the laminate is included into a bag or pouch containing a fluid, then it usually
runs to completion; this is attributed to the far better energy absorption, say damping, of the fluid
in that kind of problem and/or the lower degree of Mode III tearing in the laminate.

Reduction
in extent of
advanced
crack
Figure 8. Improved resistance to tearing in PET-layer by addition of an LLDPE-layer

in the Elmendorf tearing test; plotted on the original, undeformed mesh.
4. Application problems
As an initial test of the suitability of our general procedure and computational models, two
industrially-representative problems have been analyzed.
4.1 Bag handle tearing
The bag-handle tearing test that has been modeled, mimics the overload that a packaging bag may
sometimes be subjected to. The analysis has been conducted in Abaqus/Explicit 6.13-3 using S3R
& S4R elements for the laminated bag wall and analytical rigid surface ARSC-elements to represent
the load application device; geometrically simple rigid contact surfaces can be very easily
represented using such ARSC-elements. General contact has been specified in the problem, with all
surfaces included in the definition through usage of the parameter ALL EXTERIOR. All surfaces
interactions were regarded as frictionless. One can view (post process) the results for any
particular film within the laminate by selecting it from the RESULTS tab in Abaqus/Viewer and then
choosing the relevant layer in the list that appears. The output of such a selection for the LLDPE
layer and the PET layer is shown in Figure 9, with the ARSC elements having no contoured result.

ARSC elements
LLDPE PET
Figure 9. Different Mises stress state in LLDPE layer and PET layer in laminate.
The bag-tearing analysis results show very good agreement with experiment; the location of
severe tears and their development with crosshead displacement, as shown in Figure 10, mirrors
the visual evidence from the test. The test has introduced an asymmetry into the loading of the bag
which is not modeled; in practice there is some considerable geometric and (maybe) stress
imperfection in the bag which may cause or exacerbate this asymmetry; the model is for an
assumed-perfect bag handle.
Figure 10. Failed PET-layer in laminate highlighted by the remaining central red
region that represents the LLDPE-layer in the contour plot.

4.2 Pouch dropping problem
The pouch measures 112.5mm by 50mm by 245mm and while no specific product is modeled, it is
perhaps typical of a small “coffee-bean” pouch as illustrated in Figure 1. The thickness of plies in
the laminate have been chosen to suit the analysis and for easy reference to the already developed
data and model for the tensile test. The laminate, as modeled, consists of a 12micron PET-film
and a 130micron DOWLEX™ LLDPE film; all the overlapping bonding edges of the pouch have
had their thicknesses doubled, to represent a perfect bond. As a severe numerical test, it is fully-
filled with an essentially incompressible fluid (water) and dropped onto a frictionless rigid floor;
one expects significant strain energy to be absorbed by stretching of the pouch walls, since there is
no free space in the pouch. The water model is the same as in Problem 2.3.2 in the Abaqus
Example Problems Manual (Simulia, 2013), only converted to SI-units. The water is initially
modelled/meshed as C3D4 elements in CAE and converted to PC3D elements within the initial
step of the analysis at time = 0.0, as has been illustrated in Example Problem Problem 2.3.2, by
including the 4th line on the data of the *SECTION CONTROLS keyword:
*SECTION CONTROLS, NAME=EC-1, ELEMENT CONVERSION=YES, CONVERSION CRITERION=TIME,

KERNEL=CUBIC
1., 1., 1.
[blank line]
[blank line]
1, 0.
While, in practice the pouch would exhibit imperfections, these have been neglected in the
analysis presented in the paper; they can be easily added using the eigenmode perturbation
technique as has been discussed and illustrated in the case of the tensile strip test.
The pouch has been given initial linear and rotational velocities, as it is on the point of striking the
floor.
*INITIAL CONDITIONS, TYPE= ROTATING VELOCITY

FOUR_CORNER_BAG_SI.BAG_NODES, -0.1, -0.125,-0.125,-2.0
0.0,0.0,0.0,0.0,1.0,0.0
*INITIAL CONDITIONS, TYPE= ROTATING VELOCITY
WATER_SI.WATER_NODES, -0.1, -0.125,-0.125,-2.0
0.0,0.0,0.0,0.0,1.0,0.0
**
The acceleration due to gravity is also included, acting in the negative-z direction.

*DLOAD
FOUR_CORNER_BAG_SI.BAG_ELEMENT,GRAV, 9.8, 0., 0., -1.
WATER_SI.WATER_ELEMENT, GRAV, 9.8, 0., 0., -1.
At the end of the analysis, the pouch is still intact. The four graphics in Figure 11 show the
magnitude U of the deformed shape, LE, MISES and a computed indication SEVERE of the severity of
potential failure of the laminate. The plot is for the envelope of the maxima through the shell wall.
Interestingly, the peak contour values related to material performance - LE, MISES and SEVERE - do
not occur at the same point on the bag. The points on the pouch that are the most SEVERE-ly close
to failure are at the flopped-over neck region and the base that has contacted the rigid surface.
U LE
MISES SEVERE
Figure 11. Pouch after impact with floor – envelope of contours of U, LE, MISES
and SEVERE.
The differing degrees of orthotropy and thickness differences between the various film layers in
the laminate is the cause of the shift in the peak values for the LE,MISES and SEVERE variables on
the pouch wall-surface; hence the need for computing a failure SEVERE-ity. The necking
differences along the CD- and MD-directions for the LLDPE, as well as the milder orthotropy of
the PET-film all contribute to how the complex internal load-sharing in the laminate occurs.

5. Conclusions
The major effort in our work was the establishment of a computational framework for a thin film
of LLDPE-type material, using Abaqus. This class of material can sustain very large strain of ca.
800% before failure, with a large permanent plastic set and exhibiting local and differentiated
material instability along two directions. Our developed model shows excellent agreement with
experiment for the low-speed tensile test cases for the particular grade of DOWLEX™ LLDPE
film that has served as the basis for this paper. We have also investigated the usage of our
approach to other grades of LLDPE and film thicknesses, with success.
By comparison, the PET film is not as extensible and does not exhibit the necking instability,
although it also shows differentiated response along its MD- and CD-axes. The computational
model has been used successfully within a laminate of these two film-types and the initial
performance is very encouraging. We are able to compute the SEVERE-ity of failure envelope of the
laminate and indeed for any film layer.
The continued development and extension of the capability is an on-going effort within Dow. We
have identified improvements in our parameterization and improvements in our ability to extract
information from test data; we are currently working on implementing these and correlating
computations with experiments on dropping of pouches. We hope to be able to give feedback on
these activities and additional packaging applications for the model at future SIMULIA Customer
Conferences.
6. Acknowledgements
Acknowledgements go to Drs. M-A. Siddiqui and P. Shembekar from Dow for helping with the
CAE creation of some of the analysis models for computer runs associated with this paper. Thanks
to Dr. S. Bensason, also from Dow, for supplying the required material test data during the
computational model development.
7. References
1. Abaqus 6.13 Documentation Collection, Abaqus Example Problems, Keywords & Theory
Manuals, Simulia.
® ™Trademark of The Dow Chemical Company

Analysis of Material-Related Instability Problems for
Inelastic Thin Shells
Takaya Kobayashi and Yasuko Mihara
Mechanical Design & Analysis Corporation
Abstract: The buckling of axially compressed cylindrical shells is known to be sensitive to

imperfections, as is the buckling of spherical shells under external pressure. Regarding these
problems, a very slight imperfection can be the starting point from which local deformation is
initiated and developed and, consequently, overall structural stability is lost. Therefore, it has
been very difficult to pursue an analysis using the conventional approach, which is typically
represented by the arc-length method. The methodology developed in our previous study
demonstrates the applicability of the artificial damping method of Abaqus from which we can
overcome the difficulty in performing the analysis due to local instability. This paper describes the
application of the technique for the generation of singularities in large-deformed viscoelastic and
elastoplastic cylindrical shells, which is expected to be utilized in the packaging of food, daily
articles, and pharmaceuticals.
Keywords: Shell, Buckling, Riks method, Artificial damping, Viscoelasticity
1. Introduction
Geometrically nonlinear static problems sometimes involve buckling or collapse behavior, in
which the load-displacement response shows a negative stiffness and the structure must release
strain energy to remain in equilibrium. Several approaches can be used to model such behavior.
Among them, path tracing, based on the arc-length method, is the most fundamental procedure.
This method is used for cases in which the loading is proportional, that is, in which the load
magnitudes are governed by a single scalar parameter. If the analysis process traces unstable paths
under the global load-displacement response with negative stiffness, the arc-length method is
effectively usable. However, if the instability is localized (e.g., surface wrinkling, material
instability, or local buckling), there will be a local transfer of strain energy from one part of the
model to the neighboring parts, and global solution methods may not work. This class of problems
must be solved either dynamically or with the aid of artificial damping. The buckling of a real,
thin-walled shell is typically a local phenomenon, and it may be triggered by a small, local
disturbance. In terms of the buckling problem related to elastic cylindrical shells under axial
compression, experimental results have shown that after primary buckling, secondary buckling
occurs, accompanied by successive reductions in the number of circumferential waves at every
path jumping. In a previous study, the authors traced this successive buckling of the elastic
cylindrical shells using the latest general-purpose finite element technology with a static artificial
damping method (Kobayashi, 2012). The study accomplished a fully automatic and seamless
simulation of successive path jumpings in the deep post-buckling region and showed good
agreement with Yamaki’s experimental results (1984) and Esslinger’s high-speed photography
(1970). This paper describes the examples of the application of the artificial damping technique:
the generation of singularities in large-deformed viscoelastic cylindrical shells under axial
compression, as well as wrinkling during elastoplastic cup drawing.

2. Artificial damping method
The artificial damping method is a technique used to achieve total energy balance in static analysis,
compensating for strain energy changes due to local elastic instability. In the present study,
Abaqus (Dassault, 2013), the general-purpose finite element code, was used due to its powerful
nonlinear capability to overcome the analytical instabilities encountered in thin-shell problems,
while the artificial damping method or its comparable quasistatic capabilities are available in other
codes, such as ADINA, ANSYS, and Marc. Abaqus provides an automatic mechanism for
stabilizing unstable quasi-static problems through the automatic addition of volume-proportional
damping to the model. Via Equation (1), viscous forces are added to the global equilibrium
equations, as shown in Equation (2):
𝐅v = 𝑐𝐌𝐌 ∗ 𝐯𝐯 (1)
𝐏𝐏 − 𝐐𝐐 − 𝐅v = 𝟎 (2)
where 𝐌𝐌 ∗ is an artificial mass matrix calculated with unity density, c is a damping factor, 𝐯𝐯 =
∆𝐮𝐮/∆𝑡𝑡 is the vector of nodal velocity, ∆𝐮𝐮 is the vector of incremental displacement, ∆𝑡𝑡 is the
increment of time (which may or may not have a physical meaning in the context of the problem
being solved), 𝐏𝐏 is the total applied load, and 𝐐𝐐 is the internal force. When local instability occurs,
the deformation rate of that portion begins to increase, and locally released strain energy is
dissipated due to the appended damping effect. Consequently, quasi-static analysis makes it
possible to treat the problems involved in local instability.
Empirically, as is the case in general, moderate stabilizing can be achieved by adjusting the
damping factor c in Equation (1) so that the total dissipated energy does not exceed 1–10% of the
total strain energy. In terms of Abaqus use, the damping factor c is specified by an indirect input
parameter called “the dissipated energy fraction,” which is the ratio of the allowable incremental
damping energy (the dissipated energy for a given increment) to the total strain energy. The
fraction has a default value of 2.E-4. The damping factor for each individual element is adjusted to
meet the requirements at both the global and the local element level. However, since the artificial
damping energy is classified as non-conservative, the resultant energy always increases, while the
strain energy may decrease. Therefore, the above-mentioned treatment is just a target and can be
corrected in some situations, such as when there is rigid body motion or when significant non-
local instability occurs. In such cases, the damping factor is chosen such that the average element
damping matrix component, divided by the step time, is equal to the average element stiffness
matrix component multiplied by the dissipated energy fraction.
Based on our practical experience, lowering the dissipated energy fraction to 1/100–1/1,000 of the
default value (2.E-4) provides highly sensitive analytical performance (Kobayashi, 2012). This
technique is useful, for instance, to achieve precise expressions of the mode jumping along with
the sharp dropping of the load for thin shell problems.

3. Viscoelastic cylindrical shells under axial compression
The time-dependency of the material properties can affect the buckling behavior. Taking the
viscoelasticity of resin materials as an example, this kind of material-related instability may occur
in the formation process of thin-walled structures, such as the blow-forming of plastic containers
or the formation of polymer thin films. In this study, we performed experimental and analytical
investigations of viscoelastic cylindrical shells under axial compression. The generation of
singularities was studied in comparison with the elastic cylindrical shells.
For The test shells were constructed from Mylar (polyethylene terephthalate) film. Because the
Mylar film has both a high elastic limit and also a glass transition temperature lower than 100 °C,
the buckling experiment can be easily conducted in the glass transition temperature region of the
material. Figure 1 shows the storage modulus E', the loss modulus E”, and the loss tangent tanδ as
functions of temperature. These dynamic viscoelastic properties of the Mylar film were measured
for angular velocities 3.16, 10, 31.6, and 100 rad/sec with an ascending temperature rate of
2 °C/min in the temperature range of 30 to 200 °C. A TA Instruments RSA III dynamic
mechanical analyzer was used to measure the dynamic viscoelastic properties. This measurement
device is equipped with a temperature-controlled oven with a solid structure and a high volume
flow rate, providing excellent performance in temperature control. The glass transition
temperature Tg was estimated to be 70 °C. To create a master curve, the WLF formula, as shown
below, was applied for the time-temperature reduction.
𝐶1 (𝑇−𝑇R )
log�𝛼(𝑇)� = − (3)
𝐶2 +(𝑇−𝑇𝑅 )
where
𝑇R = 𝑇g + 50, 𝑇g = 70 ℃, 𝐶1 = 8.86, 𝐶2 = 101.6
The generalized Maxwell model is the most general form of the linear model of viscoelasticity. It
takes into account that the relaxation does not occur at a single time but at a distribution of times.
The coefficients for the generalized Maxwell model were obtained from the master curve via the
optimization approach so that the relaxation curve from Equation (4) could be calculated
numerically. The relationship between time-domain constants and frequency-domain constants is
expressed as shown in Equation (5) and Equation (6). Using these formula, the numerical
constants for the generalized Maxwell model (Equation (4)), which provides a close
approximation of the actual measurement data, can be obtained.
𝐸r (𝑡𝑡′) = 𝐸e + ∑𝑁 ′
𝑛−1 𝐸n exp(−𝑡𝑡 /𝜏𝑛 ) (4)
2 𝜔′2
𝜏n
𝐸′ = 𝐸e + ∑𝑁
𝑛=1 2 𝜔′2 𝐸n (5)
1+𝜏n
𝜏n 𝜔′
𝐸′′ = ∑𝑁
𝑛=1 2 𝜔′2 𝐸n (6)
1+𝜏n

where
Er: relaxation modulus,
Ee: long-term modulus once the material is totally relaxed,
En: elastic modulus of the nth Maxwell element,
τ n: relaxation time of the nth Maxwell element,
N: number of Maxwell elements in parallel.
1.E+10 1.E+03
60℃ E'
1.E+09 70℃ E'' 1.E+02
E' tanδ
1.E+08 80℃ 1.E+01
E',E'' [Pa]
tanδ [-]
1.E+07 1.E+00
1.E+06 1.E-01
1.E+05 1.E-02
1.E+04 1.E-03
0 50 100 150 200
Temperature [deg.C]
Figure 1. Dynamic viscoelastic properties for Mylar film.

Material Buckling test temperatures of 60, 70 (= Tg), and 80 °C were assigned so that each
temperature corresponded to the glassy condition, the glass transition condition, or the rubbery
condition, respectively. Figure 2(a) to Figure 2(c) show the relaxation curves for each temperature.
According to the time–temperature superposition principle, the curves, as a function of time, do
not change shape as the temperature is changed but appear to only shift left or right based on
Equation (5). With the planned time duration of the buckling test set at 300 sec, the estimated
relaxation zones (1–300 sec) are indicated in the figures. As is clear from the figures, the zones are
located across the glassy condition, the glass transition condition, or the rubbery condition.
1.E+10 1.E+10
Relaxation Modulus [Pa] a
1.E+09 1.E+09
1.E+08 1.E+08
Test time duration
1.E+07 1.E+07
1.E+06 1.E+06 Test time duration

T = 60 deg.C T = 70 deg.C
1.E+05 1.E+05
1E-01 1E+02 1E+05 1E+08 1E+11 1E-05 1E-02 1E+01 1E+04 1E+07
Time [sec]
Time [sec]
(a) 60 °C; Glassy condition (b) 70 °C; Glass transition condition

1.E+10
1.E+09
1.E+08
1.E+07
1.E+06
Test time duration
T = 80 deg.C
1.E+05
1E-09 1E-06 1E-03 1E+00 1E+03
Time [sec]
(c) 80 °C; Rubbery condition
Figure 2. Relaxation curves for Mylar film.
We have confirmed that the uniaxial FE model provides results that are identical to the
experimental data shown in Figure 2. The viscoelastic property of the Mylar film, especially its
temperature dependency, was precisely measured so that it could serve as the background for
achieving this agreement (Kobayashi, 2012-2). Figure 3 shows the compression test results for the
cylindrical shells under each test temperature. The test shells, with a diameter of 80 mm, a height
of 100 mm, and a wall thickness of 0.30 mm, were created with a simple longitudinal lap joint.
The shells were fully fixed at both ends into a thick aluminum plate and were compressed at a
constant temperature bath up to an axial shortening of 5 mm in 300 sec. The test temperatures
were 60, 70, and 80 °C, as mentioned above. The buckling behaviors varied depending on these
temperature conditions.
Figure 4 shows the load-shortening curves. The analytical results showed close agreement with the
experimental results. In the glassy condition (60 °C), the test shell showed typical elastic buckling
behavior accompanied by successive reductions in the number of circumferential waves (from n =
8 to n = 6) at every path jumping. Likewise, the test shell buckled with an elastic deformation of n
= 8 within the glass transition region (70 °C). However, in this case, the strain energy was relieved
immediately due to stress relaxation, and consequently, the initial buckling pattern remained as a
permanent deformation as shown in Figure 3(b). In the rubbery condition (80 °C), although the
elastic buckling still provided the specific deformation mode of n = 8 in the very early stage of the
buckling process, the initial buckling shape was replaced immediately by an irregularly deformed
shape with a round edge due to rapid stress relaxation as shown in Figure 3(c).

(a) 60 °C; Glassy condition (b) 70 °C; Glass transition condition
(c) 80 °C; Rubbery condition
Figure 3. Viscoelastic buckling of Mylar cylindrical shells under

axial compression: Experiment (left); FEM (right).
In general, a change in elastic modulus due to viscoelastic relaxation has a range of ten to the third
power or more, as shown in Figure 2. Viscoelastic material properties actually provide stabilizing
effects regarding dynamic aspects. However, in a static scheme, the rapid decrease in elastic
modulus may cause the loss of structural stability. The artificial damping method also overcomes
analytical difficulties due to material-related instability because the instantaneous stiffness change
can be reflected in the global equilibrium equations with appropriately adjusting the damping
force so as to correspond to the states of the local and individual instabilities in the model.
Incidentally, the explicit dynamic method is poorly suited to lengthy events (empirically more
than one second).

500 50
Exp. 60 deg.C FEM 60 deg.C
Load ( 60 and 70 deg.C ) [N] a

400 40
Load ( 80 deg.C ) [N] a

(a) 60 deg.C (Glassy condition)
300 30
200 20
n=8 n=7
n=6
(b) 70 deg.C
100 10
(Glass transition condition)
(c) 80 deg.C (Rubbery condition)

0 0
0 1 2 3 4 5
Displacement [mm]
Figure 4. Load-shortening curves for each temperature condition.
4. Wrinkling during elastoplastic cup drawing

This problem is presented as a benchmark at the 9th International Conference and Workshop on
Numerical Simulation of 3D Sheet Metal Forming Processes (NUMISHEET 2014) (Dick, 2014).
Advanced can manufacturing technology and cost control efforts have resulted in a consistent
reduction of the net metal weight used, which has led to beverage cans with thinner sidewalls,
reduced neck diameters, and smaller base diameters. However, small-diameter cans increase the
likelihood of wrinkling during production, as show in Figure 5. A schematic view of tools used for
the drawing process, including their dimensions, is shown in Figure 6. The drawing occurs
continuously during one punch stroke, and the punch speed is 140 mm/sec. The recommended
friction coefficient is 0.03, and the blank holding force is given as 8900 N. AA 5042 aluminum
alloy (t = 0.2083mm) and aluminum-killed draw-quality (AKDQ) steel (t = 0.2235mm) are
considered as sheet materials. These materials are assumed to have the YLD2000 anisotropic yield
criteria (Barlat, 2003), as shown in Figure 7, in which s11 and s22 represent the stress in the rolling
direction and the stress in the transverse direction, respectively. The anisotropic elasto-plastic
material behaviors were implemented in Abaqus by using the UMAT user-defined material
subroutine.
Experiment FEM
Figure 5. Draw cup with wrinkles.
Figure 6. Schematic view of tools for the drawing process (Dick 2014).
400 500
400
300
300
200
200
100 100
0
s22
0
s22
-100
-100
-200
-200
-300
-300 -400
-400 -500
-400 -300 -200 -100 0 100 200 300 400 -500 -400 -300 -200 -100 0 100 200 300 400 500
s11 s11
(a) AA5042 aluminum alloy (b) AKDQ steel
Figure 7. Anisotropic yield surfaces.
In this study, we traced the local buckling behavior of anisotropic elasto-plastic thin shells using
the artificial damping method. The automatic seamless simulation provided good agreement with
the experimentally observed buckling behavior, which covered the entire drawing process well, as
shown in Figure 8. While explicit dynamic codes have been conventionally used in the field of
sheet metal forming analysis, one of the practical advantages of using implicit codes is that they
make it possible to calculate eigenmodes at any midpoint in the analysis. Figure 9(a) shows the
results from the eigenvalue analysis for AA5042. The eigenvalue analysis was performed at a very
early stage of the punching process (2 mm of stroke). The magnitude of this punch stroke is small
enough that one can assume a nearly uniform circumferential stress distribution on the shell. The
plastic strain generated at this point was around 1–2%. One noteworthy fact is that all the

extracted eigenvalues were negative. This means that the buckling is triggered by a relaxation of
the punching stress of the sheet, which may be due to the development of the plastic strain of the
sheet and/or the sheet sliding out from the draw die.
The curves in Figure 9(a) represent the relationship between the normalized eigenvalue λ and the
number of circumferential full-waves n, where m is the number of radial half-waves. In the
vicinity of the lowest critical value, multiple eigenmodes with different numbers of waves (n = 7–
12 for m = 1) exist. The proximity of the eigenmodes must have impaired the reproducibility of the
experiments; our history analysis of the punching process (Figure 8(a)) provides a unique buckling
mode with n = 12. On the other hand, the curves for AKDQ do not indicate any obvious minimum
point, as shown in Figure 9(b). This suggests that the deformation of the sheet tends to remain
axisymmetric (n = 0) and that the punching process does not show local buckling behavior. Our
history analysis for AKDQ (Figure 8(b)) supports this estimation.
m=1 m=1
n = 12 n=0
10 10
AA5042, H1=5.207mm,
AA5042, Abaqus
H1 = 5.207 mm, Abaqus AKDQ, HH1=3.429mm,
AKDQ, Abaqus
1 = 3.429 mm, Abaqus
AA5042, H1=5.207mm, Experiment AKDQ, H1=3.429mm, Experiment
8 8 AKDQ, H1 = 3.429 mm, Experiment
AA5042, H1 = 5.207 mm, Experiment
6
Load (kN)
6
Load (kN)
4 4
2 2
0 0
0 5 10 15 20 0 5 10 15 20
Stroke (mm)
Stroke (mm)
Figure 8. Load-stroke diagram during the drawing process.
m = 1, n = 12 m = 2, n = 12 m = 1, n = 0
5 5
4 4
λ/λ1st1st
3 3
λ/λ1st1st
λ/λ
λ/λ
2 m=3
m =3 2 m =3
m=3
m =2
m=2 m =2
m=2
1 1
m =1
m=1 m =1
m=1
The lowest critical values
0 0
0 5 10 15 20 25 0 5 10 15 20 25
nn nn

Figure 9. A Relationship between the normalized eigenvalue
and the number of circumferential full-waves n.
The buckled area on the shell can be regarded as a circular plate with a hole at the center and the
inner and outer boundaries clamped. Timoshenko (1961) suggested that such a circular plate,
when subjected to radial loading, buckles in multiple waves along the circumference. When the
ratio of the inner and outer radii approaches unity, the buckling conditions for a compressed ring
are analogous to those of a long, compressed rectangular plate. In the case of the rectangular plate,
a typical deformation mode indicates several half-waves in the direction of compression but only
one half-wave in the perpendicular direction, as shown in Figure 10. The lengths of the half-waves
are supposed to approach the width of the plate because a buckled plate is subdivided
approximately into squares. A single eigenvalue will be extracted for this deformation mode.
However, manufacturing experience indicates that the real process conditions, such as the
mechanical properties of the metal sheet, the tooling geometry, and the contact conditions
(including lubrication), cause multiple buckling modes with different numbers of circumferential
waves in close proximity. This suggests that the effective width of the buckled area is influenced
by the above-mentioned factors. In fact, the conference organizer reported that the number of
circumferential full-waves was found to be n = 13 (AA5042) and n = 11 (AKDQ) in their
benchmark experiment, whereas our analyses provides n = 12 (AA5042) and n = 0 (AKDQ). The
relationship between the eigenmodes obtained from the linear eigenvalue analysis and the
deformation modes actually observed in the buckling process requires further investigation.
Axial stress distribution Before buckling
After buckling
Figure 10. Analogy of buckling behavior between a compressed ring (left)

and compressed rectangular plate (right).
5. Conclusion
In designing a modern lightweight structure, it is of technical importance to ensure its safety
against buckling under the applied loading conditions. This study presents the application of the
artificial damping method to practical material-related instability problems. Replacing the kinetic
energy with the artificial damping energy makes it possible to analyze instability problems within

the static analysis scheme. The proposed technique is useful not only for elastic local buckling
problems but also for material-related instability problems.
6. References
1. Abaqus Users Manual, Dassault Systems Simulia Corp., 2013, Version 6.13, USA.
2. Barlat, F., Brem, J.C., Yoon, J.W., Chung, K., Dick, R.E., Lege, D.J., Pourboghrat, F., Choi,
S.H. and Chu, E., 2003, Plane stress yield function for aluminum alloy sheets—part 1: theory,
International Journal of Plasticity, 19, 1297–1319.
3. Dick, R., Cardoso, R., Yoon, J.W. and Paulino, M., 2014, Benchmark 4, The 9th International
Conference and Workshop on Numerical Simulation of 3D Sheet Metal Forming Processes
(NUMISHEET2014), Melbourne, http://www.numisheet2014.org/
4. Esslinger, M., 1970, Hochgeschwindigkeitsaufnahmen von Beulvorgang dunnwandiger,
axialbelasteter Zylinder, Der Stahlbau, 39 (3), 73-76.
5. Kobayashi, T., Mihara, Y. and Fujii, F., 2012, Path-Tracing Analysis for Post-buckling Process
of Elastic Cylindrical Shells under Axial Compression, Thin-Walled Structures, 61, 180–187.
6. Kobayashi, T., Sato, M. and Mihara Y., 2012, Application of Thermo-Viscoelastic Laminated
Plate Theory to Predict Warpage of Printed Circuit Boards, Viscoelasticity, Juan de Vicente
(Editor), InTech, ISBN 980-953-307-335-9.
7. Timoshenko, S.T. and Gere, J.M., 1961, Theory of Elastic Stability, 2nd ed., McGraw-Hill,
New York, pp.389–392.
8. Yamaki, N., 1984, Elastic Stability of Circular Cylindrical Shells, North-Holland,
Netherlands, p.222.

Application of Abaqus and Isight Simulation on
Corrugated Board and Packaging
Author: Arnoud Dekker
Smurfit Kappa Development Centre
Abstract: The presentation is about the application and impact of Abaqus and iSight simulations
for corrugated board and packaging within Smurfit Kappa. Abaqus and Isightare used for the
development of the prediction formulas, used in our prediction tool PaperToBox. PaperToBox is
considered as industry leading. This tool is internally used by 1800 colleagues, resulting in >500
000 calculations per year. Also this tool provides the target values for the newly rolled-out quality
measurement system, the “Board Referee”.
Smurfit Kappa is one of the leading providers of paper-based packaging solutions in the world,
with around 42,000 employees in approximately 350 production sites across 32 countries and with
revenue of €8.1 billion in 2014. We are located in 21 countries in Europe, and 11 in the Americas.
We are the only large-scale pan-regional player in Latin America. With our pro-active team we
relentlessly use our extensive experience and expertise, supported by our scale, to open up
opportunities for our customers. We collaborate with forward thinking customers by sharing
superior product knowledge, market understanding and insights in packaging trends to ensure
business success in their markets. We have an unrivalled portfolio of paper-packaging solutions,
which is constantly updated with our market-leading innovations. This is enhanced through the
benefits of our integration, with optimal paper design, logistics, timeliness of service, and our
packaging plants sourcing most of their raw materials from our own paper mills. Our products,
which are 100% renewable and produced sustainably, improve the environmental footprint of our
customers. Our headquarters are in Dublin with regional headquarters in Paris (Europe) and
Miami (the Americas).
Keywords: Corrugated Board, Bending strength, Design Of Experiments, Response Surface

Creation
1. Introduction
The presentation is about the application and impact of Abaqus and iSight simulations for
corrugated board and packaging within Smurfit Kappa.
The shown example is the use of Abaqus and Isight for the development of the prediction
formulas, used in our prediction tool PaperToBox. PaperToBox is considered as industry leading.
This tool is internally used by 1800 colleagues, resulting in >500 000 calculations per year. Also

this tool provides the target values for the newly rolled-out quality measurement system, the
“Board Referee”.
2. Formula development for PaperToBox

The first example is the use of Abaqus and Isight for the development of the prediction formulas,
used in our prediction tool PaperToBox. PaperToBox is considered as industry leading. This tool
is internally used by >1800 colleagues, resulting in >500 000 calculations per year. Also this tool
provides the target values for the newly rolled-out quality measurement system, the “Board
Referee”, which measures the bending strength of corrugated board.
2.1 Work flow

The complete work flow is starting with the FEA model and ends with business application, see
picture 1.
Figure 1. Work flow from FEA to the business.

2.2 Pre-processing and post-processing with Python
The use of Isight requires fully parametric FEA models. Using Python scripts for pre-processing
and post-processing is needed, because the whole geometry of the corrugated board is dependent
on complex formulas. Abaqus/CAE is used as the pre- and post-processor.
Figure 2. Bending strength model.
Figure 3. Detail of bending strength model.
2.3 Using Isight for generating response surface formulas.

With Isight a DOE is created based on the properties and value range of the PaperToBox input.
For the different flute types, a full factorial DOE with selected parameters was run..
In Isight the DOE component in combination with simcode component is used. In the simcode
component each FEA model is generated with a pre-processing Python script using Abaqus/CAE,
run with Abaqus/Standard and post-processed with a post-processing Python script, again using
Abaqus/CAE.
After the DOE has run, the approximation function with quartic polynomals is used to generate the
response surface. The error using the approximation function was less than 1% for any sample
point. These formulas can be directly copied into Excel with minor editing.

2.4 Implementation in PaperToBox and subsequent usage
The Excel file with the response surface formulas is embedded in PaperToBox, an in-house
developed web-based application. This allows all the users (>1800) in the corrugated board and
packaging plants (>125) to use PaperToBox and have instantaneous results for their calculations.
This allows the Smurfit Kappa plants to optimize the paper usage (> 3000 M€) in the corrugated
board and packaging.
Also PaperToBox provides the target values for the newly rolled-out quality measurement system,
the “Board Referee”, which measures the bending strength of corrugated board. the corrugated
board and packaging plants (>125) will measure more than 600 000 samples per year.
Figure 4. Board Referee, developed by Smurfit Kappa.
3. References

Use of Abaqus/CAE and True-LoadTM to Determine
External and Internal Loading of a Full Suspension
Mountain Bicycle
J. Maas, Analysis Engineer, Trek Bicycle Company
Abstract: Trek Bicycle Corporation has long been at the leading edge in the bicycle industry.
Treks bicycles are subjected to the most rigorous testing in the industry and their frames are
covered by a lifetime warranty. To maintain the highest level of safety and quality, real world
loads need to be properly understood. Trek Factory Racing professional athletes are pushing
that understanding; most recently attempting a front flip across a 72 foot canyon. The loads
generated for such an event are certainly beyond current understanding and testing protocols.
A process using Abaqus and the True-LoadTM plug-in (Wolf Star Technologies) will be presented
that effectively converts a full-suspension mountain bicycle into a load transducer. A fully
instrumented Trek Aluminum Session 29er mountain bike (strain gauges, accelerometer, shock
sensors, GPS, etc…) is used with high speed video to record a high load event. These data, in
conjunction with the Abaqus / True-LoadTM process, are used to determine the external loads
on the bicycle. A comparison of laboratory testing loads to the external loads is made. The
implementation of the Abaqus / True-LoadTM process increases Treks knowledge of loading
environments and aligns with Treks history of innovation and reliability.
Keywords: Abaqus CAE, True-LoadTM, Bicycle, External Loads, Internal Loads, Experimental
Verification
1. Introduction
Trek Bicycle Corporation has long been at the leading edge in the bicycle industry. Treks’
bicycles are subjected to the most rigorous testing in the industry and their frames are covered by a
lifetime warranty. To maintain the highest level of safety and quality, real world loads need to be
properly understood. Trek Factory Racing professional athletes are pushing that understanding;
most recently attempting a front flip across a 72 foot canyon, see Figure 1. The loads generated
for such an event are certainly beyond current understanding and testing protocols.
The closet industry test requirements (Ref. ISO 4210-6, Annex C) to this event load the rear of the
bicycle to 2800 N to ensure safety. Trek performs a similar laboratory test that goes above and
beyond this test (minimum required load before failure is 5338 N (1200 lb)) and additionally tests
the bicycle to failure, see Figure 2. We believe these tests neither load the frame to the level of
professional riders nor in the same manner.

Figure 1. Trek professional mountain bike rider, Tom van Steenbergen, tries to
front flip the 72-foot canyon gap at Red Bull Rampage (October 2014).
Figure 2. SS-EN ISO 4210-6 Tire clearance test and the Trek DOP test
configurations.
The goal of this work is to develop a process using Abaqus/CAE and the True-LoadT M plug-in
(Wolf Star Technologies) that can quantify the loads created in the field during extreme use cases.
These data can then be compared to current laboratory tests, used to develop new laboratory tests,
or modify existing ones.
2. Finite Element Model

The test platform is a production Session 29er Aluminum bicycle and is modeled in Abaqus/CAE
using solid (C3D10) and shell (S8R) elements. All shell elements are constant thickness even
though the tubing of the bicycle is butted. The tapered sections of these tubes have a thickness
equal to the average thickness of the tubing through that region. Several portions of the bicycle
are modeled as aluminum solids and are indicated by the green elements in Figure 3 below. The
rocker link is model using solid elements as well but is made from Magnesium AZ90 alloy.

Note that during the event time of interest, see discussion in Section 4, the suspension system is
under a relatively constant state equal to 140 mm of front fork compression and a fully
compressed rear shock. The front fork is idealized using beam (B31) elements and the rear shock
using connector elements. The rear end of the bicycle is oriented such that the rear shock is in the
fully compressed configuration.
Tie constraints are used to model welds, load transfer from the fork to frame, and at frame pivots.
The connections of the parts, i.e. rocker link to the main frame, are modeled using connector
elements. Beam, link, and hinge connectors are used throughout the finite element model (FEM)
depending upon the application.
Figure 3. Aluminum Session 29er finite element model.

FEMs of the wheel and tire system have not yet been developed and, therefore, have not been
included. As such, boundary conditions are applied to the center of the rear and front dropouts. At
the rear all degrees of freedom (DOF) are held fixed except rotation about the axle. At the front all
DOF are held fixed except rotation about the axle and movement in the FWD/AFT direction.
A set of assumed unit load cases were applied at the center of the bottom bracket (BB) and at the
left hand side handle bar locations. The objective of the unit load cases are to allow the FEM to be
placed in a multitude of independent strain states that True-LoadT M can amplify and combine in
such a way as to match the strains produced in the field. The initial set of unit loads at the BB
included all force directions (X, Y, and Z) and moments Mx and My. The initial set of unit loads
at the handle bar (HB) included all force directions and moments Mx and Mz. The set of unit loads
were subsequently reduced during the FEM refinement stage of the analysis where the model and
its boundary conditions and unit loads are iterated until an acceptable level of strain correlation is
reached.

2.1 True-LoadTM
True-Load™ is a set of Abaqus/CAE plug-ins that essentially turns any structure into its own load
transducer. The Wolf Star True-Load™ construct creates a mathematical relationship between
user supplied unit loads and strain gage strain measurements. As a result measured strain data can
be used to back calculate real-world loading on the structure.
True-Load™ consists of three major components: Pre-Test; Post-Test and True-QSE™. Pre-Test
will be used to locate strain gauges on the structure. Pre-Test will create a strain proportionality
matrix that will be used to relate strain values to load cases. Post-Test uses the strain
proportionality matrix defined in the Pre-Test module and measured strain data to create loading
histories. Post-Test will create a True-QSE™ event that will be used to interrogate the structure.
The True-QSE™ module will be used to open the event created by True-Load™ to allow the user
to perform advanced post processing functions. The overall flow of using True-Load™ is
illustrated in Figure 4. See the True-LoadT M documentation for further discussion regarding the
theory used to determine optimal strain gage locations and determination of loading histories for
each unit load case.
Figure 4. True-LoadTM process flow.

Using Pre-test to determine optimal strain gage placement requires the generation of an element
set in Abaqus/CAE during the preprocessing phase. This element set is then used in Pre-Test as a
set of candidate elements for gage placement. Within Pre-Test the element set and unit loads cases
are selected and then the strain proportionality matrix is calculated. True-LoadT M provides a set of
tools for rotating, orienting, and moving strain gages for both fine-tuning the quality of the strain
proportionality matrix and to ease placement of the gages on the structure. Additionally, tools are
available for generating dimensioned figures for specifying the strain gage placement.
3. Instrumentation and Testing

23 data channels (12 strain gages, 3 tri-axial accelerometers, 1 linear potentiometer, and 1 Hall
Effect sensor) were used on board a DTS (Diversified Technical Systems, Inc.), SLICE data
acquisition system (DAS), see Figure 5. Data was collected at 5,000 Hz each (7 million data
points per minute). The use of this very small and completely on-board DAS allowed for the first
time data collection during extreme mountain bike events with no hindrance on the rider what-so-

ever. Previously used DASs required the use of a backpack to carry the data collection and battery
hardware and, therefore, also required a set of wires from the bicycle to a backpack which greatly
restricted the riders’ movement. The bottom image of Figure 5 shows the old DAS as was used
for a road vibration measurement test where the backpack and required wire harness can be seen.
Twelve strain gages (Vishay Precision Group) were placed throughout the frame according to the
True-LoadT M Pre-Test analysis. Three tri-axial accelerometers (Dytran Instruments, Inc.) were
mounted to the frame and fork (non-drive side AFT chain stay, Center of Head Tube, rear of seat
tube) and were used for visualization of the frame movements during the event of interest.
Additionally, a KA Sensors linear potentiometer, commonly used in the motorsports and
automotive industry, was used to measure front fork travel and a small and inexpensive Hall Effect
sensor (obtained from DigiKey), was incorporated into the rear linkage and calibrated to indicate
the rear shock length, see Figure 6. Sensor cables and wiring were custom designed and installed
to reduced bulk and weight. A small 2600 mAh battery was used to power the system, lasts for
many hours of testing, and was mounted directly to the main frame.
Figure 5. Session 29er mountain bike showing the DTS SLICE DAS (top). Old DAS
requiring backpack and wire harness (bottom).

Figure 6. Linear potentiometer (left) used for front fork displacement
measurements and Hall Effect sensor (right) used for rear shock displacement
measurements.
For this initial research two load cases were post-processed fully through the True-LoadT M
process. The first load case was a fairly large straight drop off of a wooden structure (Deer
Hunter, see Figure 7). The second was a jump landing with substantial sideways movement of the
bicycle during landing. The purposes of using these two load cases were:
1) Determine external and internal loads for a load case capable of fully compressing
the rear suspension. Compare these loads to the laboratory ‘equivalent’ test (Drop
Out Push).
2) Compare lateral forces generated in the sideways landing case (Mojo, data segment
10) to both Deer Hunter and Drop Out Push.
Figure 7. Deer Hunter straight drop landing load case (left) and Mojo “sideways”
landing load case (right), Trek Trails.

4. Post Processing and Analysis
Data from one event on the Deer Hunter trail is shown in Figure 8. The trail used for this testing
contains several turns, berms, rocks and roots etc… Data was collected from the beginning of the
trail to shortly after landing the jump of concern. Data shown in Figure 8 is for the portion of the
event while the rider was airborne to just after landing the jump. Note that all data channels are
“quite” while the rider is airborne, e.g. all strain data shows zero strain and the front fork,
accelerometers show zero acceleration, and rear shock show zero change in length.
Upon impact, strain channels become active again and suspension begins to compress. Note the
portion of the data indicated by the dashed red bars. There is a 12 msec portion of time where the
suspension is again in a somewhat steady state but compressed. For this period of time the rear
suspension is fully compressed and the front fork is in a steady 140 mm of compression. This is
the configuration for which the FEM was generated.
Strain data for this 12 msec period of time is shown in Figure 9 and is the strain data used within
True-LoadT M for the strain correlation analysis. Note the sharp spike in strain shortly after the
beginning of this 12 msec period of time. The rear suspension is fully compressed but not all of
the energy from the impact has been absorbed. The bicycle frame itself needs to absorb this energy
as strain energy. A rapid increase in strain shows up strongly in the chain stays and seats stays.
Figure 8. Sensor data from Deer Hunter landing case, data segment 23. Data
starts with bicycle airborne. Note that shock velocity data are calculated data.

Figure 9. Strain gage data from Deer Hunter load case, 12 msec of data during
impact where rear shock and front fork are in a near steady state configuration.
The strain proportionality matrix created in True-LoadT M Pre-Test was imported into Post-Test
along with the strain gauge data discussed above. True-LoadT M Post-Test generated a load
amplification curve for each of the unit load cases that minimize the error between the field strains
and the finite element strains. An example amplification curve for the forward and downward load
on the bottom bracket is shown in Figure 10.
The peak amplification for the X direction load on the bottom bracket occurs at time equal to 2.2
msec and corresponds to a load factor of 17.25, or a total BB-Fx load of 17,250N (3,880 lb). The
BB-Fy maximum load factor occurs at time equal to 2.6 msec and corresponds to a load factor of
4.90, or a total load of 4,900N (1,100 lb). Note how the peak in the load amplification curves
corresponds to the peak the in strain signature in the chain stays and seat stays shown in Figure 9.
Figure 10. Load amplification curves for BB-Fx (left) and BB-Fy (right); Deer Hunter
load case.

Additional post-processing tools provided by True-loadT M include the strain data cross plot, and
strain gauge / FEM strain vs time for each strain channel. As can be seen in Figure 11 the load
amplifications have provided a reasonably close strain state in the bicycle frame as compared to
the field measured strains. Only one of nine strain gage channels used in the analysis has an error
greater than 8% and in that case (gage 2 – Drive Side Forward Chain Stay) the error is
conservative and off by approximately 90 µε.
Figure 11. Strain data cross plot (left) and history plot (right) for all strain gages;
Deer Hunter load case.
True-LoadT M QSE generates an output database in which all loading histories are combined and is
useful to visualize the loaded bicycle and its deflected shape. Additionally, reaction forces can
easily be extracted from the QSE ODB. Reaction forces at the center of the rear drop out are
compared to forces applied during the DOP test case in Table 1. The reaction forces shown are for
the maximum loading condition during each event.
Note that the loads generated in the field cases both far exceed the DOP lab case. Also note the Z
(lateral) load for both field cases is non-zero but is not large. Additionally, the Mojo field case,
which was assumed would generate a much larger Z load than the Deer Hunter load case, only
generates about 330 N lateral force.
Table 1. Field Cases versus Drop Out Push rear end loads.
Load Case Y Load (N) Z Load (N)
DOP Lab Case 5338 0
Deer Hunter Field Case 8663 -129
Mojo Field Case 6386 -331

An additional step in the analysis process is necessary to enable the extraction of internal loads.
For ease of processing, the load amplification factors at a single point in time are extracted and
used as inputs to a single time step static finite element model. This model is created to enable the
request for nodal forces due to elemental stresses (NFORC) which are needed to extract loads
within the Free Body Diagram (FBD) cut feature within Abaqus/CAE. Table 2 shows the load
distribution through the seat stays and chain stays as extracted from Abaqus /CAE using the FBD
cut tool.
Table 2. Load distribution through Seat Stays and Chain Stays.
Load Case Seat Stay Loads Chain Stay Loads
Fx Fy Fz Fx Fy Fz
DOP Drive Side -2997 -1316 410 2975 -1328 -441
Non-drive Side -3036 -1333 -379 3058 -1361 411
Deer Hunter Drive Side -1962 -865 317 8245 -3645 -1113
Non-drive Side -1956 -851 -196 7516 -3302 1120
Mojo Drive Side -1425 -615 141 5499 -2413 -909

Non-drive Side -1592 -707 -254 6016 -2651 692
As can be seen by scrutinizing the load distributions, the DOP case has a significantly larger
proportion of the rear end load passing through the seat stays. It appears that the fixity provided at
the seat collar and top of the head tube during the DOP lab case contribute to drawing load
through the seat stays. Also note that the chain stays are much more highly loaded in the field
cases than they are in the DOP lab case. This is also seen by looking at strain plot comparisons for
Deer Hunter load case versus DOP lab case, see Figure 12.
Figure 12. Principal strain distribution in the chain stays for the Deer Hunter (left)
and DOP (right) load cases.

5. Conclusions / Summary / Future Work
A full suspension mountain bike was successfully instrumented with strain gages, accelerometers,
and suspension travel sensors and data was collected for post processing within Abaqus/CAE and
True-LoadT M. With the use of True-LoadT M we were able to generate load amplification curves
and subsequently compile a quasi-static event that allows for the extraction of external loads and
internal loads. These loads were compared to a laboratory case (DOP) and it is found that loads
generated in the field far exceed those currently being tested. Additionally, it is found that as a
result of the boundary conditions in the lab the strain state of the seat stays and chain stays is
different in the laboratory than it is in the field.
The FEM accurately idealized the bicycle frame with shell and solid elements but the fork was
inadequately idealized with beam elements. Additionally, wheels were not modeled and
assumptions were made regarding boundary conditions at the center of the rear and front drop
outs. Further refinement of the FEM to include these details may result in even better strain
correlation.
It was seen that lateral loading of the rear end of the bicycle during field testing was small.
Additional field load cases will be tested in future testing that should generate larger lateral
loading, i.e. over-rotated 360 degree rotated jump. Also, additional testing is planned with a
professional athlete with plans to perform even larger stunts to measure the amount of load
increase for these even more extreme events.
We are also investigating the use of multi-body dynamics software to allow scaling of load cases
in the box. This removes the need to field test for every case where loads are needed to be
determined.
6. References
1. Cycles – Safety requirements for bicycles – Part 6: Frame and fork test methods, ISO
4210-6:2014, Swedish Standards Institute, Stockholm, Sweden.
2. User Manual / Documentation for True-LoadT M, Wolf Star Technologies,
Milwaukee, WI.

ENERGY, PROCESS
& UTILITIES

Process Integration between CAD and CAE
Applications for Isight-based DOE Analysis
Ing. Davide Pinna Ing. Luca Fattore

Ansaldo Sviluppo Energia Exemplar
Via Lorenzi 8 Corso Vittorio Emanuele II 161
16152, Genova. ITALY 10139, Torino. ITALY
+39 010 6553249 +39 011 435051
davide.pinna@aen.ansaldo.it luca.fattore@exemplar.com
Abstract: Product development engineers often use simulation to evaluate the performance of a
single design - a part or a system - characterized by a specific CAD representation, specific input
data, and materials properties.
Simple parametric analyses and "what-if?" studies are frequently performed in order to try to
improve the design. However, the use of automated optimization and Design-Of-Experiments
techniques that allow efficient scientific exploration of the whole Design Space to establish the
correlation between the various parameters and to identify optimal solutions is instead less
frequent, especially when complex CAD models and various CAE tools are involved.
This paper is focused on CAD and CAE workflow integration for the design of a Gas Turbine
vane. The model of the complete process, implemented in the SEE-Fiper environment and
involving all the calculation tools, is described.
The goal of the project is to explore the Design Space using a DOE analysis to identify an
optimized vane design with geometry that leads to simpler machining and that satisfies both
minimum creep life and minimum fatigue life criteria.
Keywords: CAD, CAE, Collaborative Engineering, Gas Turbine Blade, Gas Turbine Vane, Isight,
Multidisciplinary Process, Data and Process Management, Process Automation, Turbomachinery.
1. Introduction
Ansaldo Energia (AEN) is Italy's leading producer of thermoelectric power plants including
process, mechanical, civil, installation and start-up engineering. The production centre is split into
three product lines: gas turbines, steam turbines and generators.

Among the components designed in AEN, one of those with higher-tech is certainly the gas
turbine vane. It includes various types of challenges: strength, durability, flexibility and
maintainability.
The project described in this paper concerns the design and implementation of an automatic
multidisciplinary process for the thermo-mechanical analysis of a cooled gas turbine vane, capable
of handling both standalone activities and optimization activities. The process starts from a
parametric cold CAD model of the internal channel integrated with the CAE model so that
changes in the CAD model are automatically reflected in the CAE reference model as multiple
analysis runs are launched during the optimization process.
This process has been applied to the exploration of the Design Space for the identification of a
new configuration optimized in order to obtain a vane with greater creep and fatigue life and with
a lower cooling air consumption and, not less important, to increase the know-how of the experts
in 3D modeling and thermo-structural analysis on the use of analytical techniques with parametric
CAD models.
As shown in Figure 1, the internal geometry of the vane airfoil examined in this paper is
composed of a single channel along its length.
Figure 1. Vane views with internal channel

The disciplines involved in the analysis are 3D modelling, heat transfer, mechanical integrity and
lifing (low cycle, creep and high cycle) evaluation. With this automated procedure, all of the tasks
concerning the 2D thermal calculation, the 3D thermal calculation, the 3D thermo-structural
calculation and the post processing of the results can be launched in a single run, and studies can
be performed by varying in a simple way selected parameters at the beginning of the process.
This project provides for the integration of many processes in Isight which can be distributed to
many machines in order to optimize all the IT resources (time/license/timeframe) involved in the
company.
The activities carried out during this project are:

 The creation of an automated and repeatable procedure:
o Process for parametric CAD model modification
o Process for automatic mesh on CAE model
o Process for thermo-mechanical analysis and verification
 The process validation.
 Design of Experiments analysis (DOE).
 Results analysis and identification of the optimized solutions.
2. Process Description
This paper describes a multi-disciplinary process for multi-objective optimization of a gas turbine
vane. Figure 2 shows the main workflows of the overall process captured in SIMULIA’s Isight
process integration and design optimization software. This is a multidisciplinary process, very
complicated; we use different software (commercial and homemade) and different methodologies
and, thanks to Isight, we have very useful management of all of the results with visibility on the
state of the vane run to run.
Figure 2. Master workflow
The process is divided into the following workflows:

- CAD Modification
The CAD modification is based on the UG-NX CAD system and the aim is to update
the internal channel section geometry of the vane airfoil in order to obtain a new vane
with a complete new internal channel. The internal channel is a body (see Figure 4)
composed of 17 section profiles that describe the shape of the cooling air channel; this
body is finally Boolean subtracted from the vane solid body. The 17 channel sections
defining the shape of the channel body are parameterized as shown in Figure 3 using 4
geometry dimensions, introduced below:
[mm]RAGGIO_LEADING_EDGE=nn
[mm]RAGGIO_TRAILING_EDGE=nn
[mm]OFFSET_CURVA_PRESSURE_SIDE=nn
[mm]OFFSET_CURVA_SUCTION_SIDE=nn5
OFFSET
SUCTION SIDE

R LEADING_EDGE
R TRAILING_EDGE
OFFSET
PRESSURE SIDE
Figure 3. Example of one of the profiles that is updated by the parameters

In Figure 4, one example of the internal channel body operation based on 17 section
shapes is shown, in which all of the parametric shapes have been modified
independently. In order to avoid sudden curvature changes in the channel shapes in
the radial direction, as shown in the right side of Figure 4, parabolic variation rules
along the vane span have been introduced as MAIN DOE geometry parameters.
Figure 4. Example of the internal channel body operation based on 17 section

shapes when all of the parameters of the shapes have been modified
independently; the dashed curves show the parabolic rules used to drive the
parameters of all 17 sections in order to avoid an unrealistic channel shape.
The main parabolic channel thickness and radius variation equation along the vane
span is based on two geometry characteristics: the maximum position and maximum
value of the parabola.
Finally, eight global geometry parameters for the parabolic variation rules along
the vane span have been specified as candidates for the DOE analysis to use to
define the vane cooling channel shape: max_pos and max_value for trailing and
leading radii, and max_pos and max_value of suction side and pressure side

thicknesses. Figure 5 shows a number of different geometry variations that can be
obtained with these parameters.
Figure 5. Examples of internal channel geometry variations studied with the

DOE process
The Isight process task related to the geometry updates is shown in Figure 6: the first
part is a loop on all the channel sections in order to update their geometry parameters
based on the parametric parabola shape. Finally the CAD vane assembly is updated
with the new section shapes.

Figure 6. CAD modification process and automatic mesh generation inside UG-
NX
- Automatic Mesh
The second important step in the automation is the FEM model creation. This task is
performed automatically inside the NX-CAE environment, and the FEM model
creation procedure is based on CAD surfaces tagging in order to control the mesh
phase and the ANSYS component creation.
The final result is an ANSYS cdb file with the mesh and the component definition that
are used in the next step to apply boundary conditions through some ANSYS macros.
The total number of nodes in the automatic process has been limited to 0.5M with
respect to the original FEM model used. Figure 7 shows the details of the two FEM
models (automatic and manual) compared side by side.
Figure 7. Detail of the original vane FEM model on the left (structural and
thermal) and the automatic FEM model on the right created in the Isight
process: the number of nodes is 1/3 of the original but the quality satisfies the
analysis criteria.

- 3D Thermal Analysis
In this task, a 3D thermal analysis is carried out to obtain the vane’s 3D thermal map
distribution. The boundary conditions for the 3D thermal analysis are passed directly
from a 2D thermal analysis (2D temperature maps, flow network result, 2D HTC
distribution). In this implementation, the 2D thermal results are not updated within the
new vane channel geometry: instead, a preliminary sensitivity analysis has been
performed to check if the introduction of this 2D thermal analysis into the loop is
needed, but currently the channel geometry effect on the temperature in the flow has
been considered negligible.
Figure 8. 3D Thermal analysis task

In Figure 9 below, a temperature map comparison between the manual vane analysis
versus the automatic one is shown, with the same channel geometry in both cases. The
temperature difference is quite small (less than 1°C).
Figure 9. Final temperature map from the thermal analysis: comparison of the
manual results (on the left) with the automatic procedure (on the right)

- 3D Mechanical Analysis
In this process, a 3D structural analysis is carried out on the vane, applying structural
loads (pressure) and the map of the temperatures obtained with the 3D thermal
analysis, to get the final vane’s 3D stress map. Figure 10 shows the automated Isight
process, and Figure 11 shows the results of the analysis on pressure and suctions
sides. Note that a VCOLLAB visualization file is automatically generated from the
ANSYS RST results file for easy and efficient evaluation of the results.
Figure 10. 3D thermal-mechanical analysis process
Figure 11. Stresses in the vane, pressure and suction side
- Thermo-structural post-processing
The post-processing task is divided into three main parts, as shown from Figure 12 to
Figure 16. The aim of this task is to evaluate the overall performance of the channel
configuration based on maximum Von Mises stress, the creep plastic strain, the LCF
(low cycle fatigue) safety factor for the turbine pre-defined working mission, and the
final envelope combined (LFC+creep) damage factor.

 Calculation of the margin of safety
Figure 12. Post processing for margin of safety

The margin of safety is based on Ansaldo’s specific requirement. The vane is sliced
along 6 planes, then the max stress point is identified at the suction and pressure sides
and the stress path is extracted from the max stress point to the shortest distance to the
channel. Along the stress path some stress evaluations are performed in order to get
the margin of safety comparing the membrane, bending and maximum stress within
the material allowable at the specific temperature in the slice. This computation
process is automatically performed combining post-processing ANSYS macros and an
Excel file (see Figure 13).
Figure 13. Example of the Excel worksheet used to verify the stress limit in the
vane
 Creep analysis (to determine the stress due to creep)
Figure 14. Post processing for creep analysis

In Figure 14, after the ANSYS creep simulation the max creep strain is evaluated in
the vane and a VCOLLAB model is created.

 Low cycle fatigue (LCF) analysis (to determine the time of fatigue failure)
Figure 15. Post processing for LCF analysis

A Matlab in-house code is used to evaluate the LCF fatigue margin of safety
combining cyclic stress (computed by ANSYS in the Thermo-mechanical task)
simulating a required total amount of startup-shutdown cycles.
 Cumulative damage evaluation
Figure 16. Final integrity margin of safety envelope evaluation

A Matlab in-house code is used to evaluate the cumulative damage, combining LCF
fatigue margin of safety with creep damage in order to get the final envelope margin
of safety of the vane. This evaluation is performed at different mission scenarios
(combining specific hours of startup and number of shutdown-startup cycles).
3. Process Validation
A preliminary validation of the whole process has been performed using the nominal channel
design parameters.
In Figure 17, the “official” vane is compared to the one automatically created by the Isight task.
The slices created are mainly used by the engineer to see the visual positions of the planes used for
the design validation and results extrapolation, but in the automatic Isight process all of the
positions are codified in the procedure and no more visual help for the human engineers is needed,
so a unique CAD solid is created.

Figure 17. The original CAD model used for the manual vane simulation is sliced (on the
left); the automatic procedure creates a unique solid (on the right, the nominal DOE
configuration)
Figure 18. The stress map from the mechanical-thermal analysis: comparison of the manual
results (on the left) with the automatic procedure (on the right)
In Figure 18, the Von Mises stress contour from the official manual vane stress analysis is
compared to the analysis produced with the automatic Isight process. Some differences are
apparent, but the critical location and the global behavior are matched.

4. Results
In Figure 19, a relevant subset of geometry configurations obtained from the DOE study is shown.
The max stress, max temperature and max creep strain results for each configuration have been
evaluated; the table of results shows how increasing the mean vane thickness of the airfoil profile
leads to a double effect of increasing the creep strain at the training edge (TE) while decreasing
the strain at the leading edge (LE). However, the leading edge is more critical, so a solution with a
thick airfoil is preferable. In any case, the vane has significant margin before the critical lifing
condition is reached.
As shown in Figures 20 and 21, the stresses and the temperatures are not affected by variations in
the vane airfoil thickness. Also, the leading and trailing edge radii have only a second order effect
on the creep of the vane.
stress blade
temp blade Strain CREEP blade [%]
[Mpa]
thickness OFFSET trailing radius leading radius
caso, note TE LE max TE LE
[mm] % %
sottile -1,5 100 100 N+2 N-9 N N -0,013 N+0,008
nominale_DOE 0 100 100 N N N N N
2,5 100 100 N+2 N+4 N+0,7 N+0,0096 N-0,004
3 100 100 N+3 N+4 N+0,9 N+0,011 N+0,004
spesso 3,5 100 100 N+3 N+3 N+1 N+0,011 N+0,001
spesso rag racc + 3,5 110 110 N+0,010 N+0,0032
spesso rag racc + 3,5 110 120 N+0,0097 N-0,0043
Figure 19. Some relevant configurations and the overall results obtained from the DOE
campaign: N is the Nominal value
Figure 20. The temperature distribution with 3 different relevant thickness configurations

Figure 21. The stress distribution along the vane for 3 different airfoil mean thicknesses
5. Conclusion
The DOE results on the turbine vane object studied in this paper help to illustrate some interesting
considerations regarding the internal channel design.
• No critical lifing condition is violated as a result of the channel thickness and radius
distribution along the vane span; therefore an increase in the thickness may lead to an
overall lower creep strain and it will not significantly affect the life of the part.
• Innovative considerations in the internal channel design coming from manufacturing
and/or quality approach can be evaluated without compromising the structural
requirements of the vane.
This Isight project has shown the potential benefit of an automatic procedure that involves a
parametric CAD model in all the steps of checking the design requirements for the final design
validation. Some lessons learned can be summarized:
• Creating an automatic and repeatable procedure for design studies provides the ability to
track all the information/files/methods from the CAD definition to the FEM validation so
that they can be re-used in future studies.
• The increase in knowledge of the multidisciplinary CAE tools achieved by the engineer
involved in the development of the Isight process has enabled a new approach, leading to
a more powerful and thorough automatic design exploration procedure compared to the
classic trial-and-error manual experience.
• By enabling the quick exploration of new and innovative designs, the engineer’s
knowledge in the specific field of interest is increased, thus encouraging him to explore

beyond heritage configurations and drive forward with new innovative approaches and
designs.
6. Bibliography
The following papers have provided inspiration for this project and have been the starting
point for the development of our innovative contribution to the design of the gas turbine.
1. Cofer J.I., and Hahn Y., “Design Study of Dovetail Geometries of Turbine Blades using
Abaqus and Isight,” ASME paper GT2012-68566, ASME Turbo Expo, Copenhagen,
Denmark, 11-15 June 2012.
2. Di Sisto P., Parodi M., “New Simulation Process for the Life Management of the Gas
Turbine Rotors,” ASME paper GT2013-96033, ASME Turbo Expo, San Antonio, Texas,
03-07 June 2013.
3. Karl A., “Simulation Driven Design Enabling Robust Design,” presented at the 2010
SIMULIA Community Conference, 25-27 May 2010.
4. Mattone M., Maggiore M., and Di Sciuva M., “Integrated Environment for Virtual Product
Development,” Greek-Italian Aeronautics Days: Growing by Sharing, Turin, Italy, 27
October 2006.
5. Panchenko Y., Moustapha H., Mah S., Patel K., Dowhan M.J., and Hall D., “Preliminary
Multi-Disciplinary Optimization in Turbomachinery Design,” presented at the NATO
RTO AVS Symposium, Paris, France, 22-25 April 2002, and published in RTO-MP-089.
6. Pinna D., and Cittadella F., “DOE Analysis for Internal Cooling Configuration of Gas
Turbine Blade,” presented at the 2013 SIMULIA Community Conference, Wien, Austria,
22-24 May 2013.
7. Pinna, D., and Macciò, M., “Collaborative Engineering Process for Multidisciplinary
Optimization of a Gas Turbine Component,” presented at the 2011 SIMULIA
Community Conference, Barcelona, Spain, 17-19 May 2011.
8. Pinna, D., “L’ingegneria Collaborativa: l’esperienza in Ansaldo Energia,” Master in
“Innovazione tecnologica e sviluppo prodotti per la generazione di energia,” Università
degli Studi di Genova, Ansaldo Energia, 2010.
7. Acknowledgments
The authors wish to thank Dr. Alberto Federici and Dr. Francesco Cittadella for his contributions
during the process implementation.

Hydraulic Fracturing Simulation for Fracture
Networks
Stephan Arndt, Wouter van der Zee, Tobias Hoeink
Baker Hughes
Jianhu Nie
formerly of Baker Hughes
Abstract: The term ‘Unconventional Reservoirs’ is used in the oil and gas industry for
hydrocarbon reservoirs that have very low permeability in the magnitude of microdarcy (μd) and
therefore rely on artificially introducing pathways for fluids and gases, commonly by using
hydraulic fracturing techniques, to enable economic production. This approach of using
fracturing stimulation campaigns is also increasingly used to improve production in mature oil
and gas fields. Industry estimates for North America indicate that more than half of all stimulation
treatments have no impact on production, despite this technique having been in use in numerous
onshore wells for more than a decade. Creating fractures connecting the source rocks to the well
whilst allowing fluids to flow, either with new hydraulic fractures and/or activating existing
natural fracture sets can have a large impact on production. Optimising the current approach
based on determining its many parameters and adapting experience based knowledge to other
regions in the world poses a complex challenge with the goal to understand and predict the
effectiveness of the stimulation campaigns. Significant progress in the simulation of fluid driven
crack propagation for a single fracture occurred with recent updates of Abaqus capabilities
including coupled fracture and pore fluid flows with porous medium deformation. This paper
looks at the challenges using different modelling approaches using Abaqus such as Cohesive
Elements (COH), Extended Finite Element Method (XFEM), Smooth Particle Hydrodynamics
(SPH) and coupled Eulerian Lagrangian methods (CEL) with the goal to extend applications to
more complex geometries with interaction of multiple fractures and stress shadowing effects.
Keywords: Cohesive Elements, Coupled Analysis, Coupled Lagrange Euler (CEL), Crack
Propagation, Damage, DFN, Drilling, Fracture Initiation, Fracture Propagation, Geomechanics,
Horizontal Wells, Pore Pressure, Poro-Elasticity, Reservoir, Smooth Particle Hydrodynamics
(SPH), Wellbore, XFEM.
1. Introduction
A large number of publications on the topic of ‘Unconventional Reservoirs’ are available and only
a few key points are summarized here with the focus on Hydraulic Fracturing Simulation using
Abaqus’ coupled pressure-deformation capabilities, including cohesive elements, extended finite
element method and other techniques.
For production from unconventional reservoirs, it is important to overcome the limitations in flow
rates caused by the low permeability. One aspect is increasing the reservoir contact, i.e. the length

of the wellbore positioned inside the reservoir. Given the restricted vertical height typically
available compared to the large horizontal distances, directional drilling has become a key
technology as horizontal wells can significantly increase the length of the wellbore in contact with
the reservoir. The second key technology is increasing the flow into the well using pathways
created by hydraulic fracturing. Injection fluids are pumped into the well to open fractures at
specific points, then particles, called proppant, are added to keep the fractures open. Proppant
selection (one example is fine-grained sand) includes a choice of particle sizes to fit into the
fractures and volume and stiffness will influence the extent to which the fractures remain open and
permeable.
This leads to three major challenges:
1. Lateral placement of the wells
2. Placing of fracturing stimulation stages along the well
3. Fracture design and operational parameters
Figure 1. Schematic of a horizontal well and hydraulic fracturing stage.

Knowledge of the in-situ stress is required in the design process. With the assumption that the
major principal stresses in the earth’s crust are usually well aligned with the vertical and
horizontal directions (note these are always compressive unlike in engineering applications), one
can divide the possible stress states into:
• Normal Faulting – the vertical stress is the major principal stress, Sv > SHmax > Shmin
• Strike-Slip – the vertical stress is the intermediate principal stress, SHmax > Sv > Shmin
• Reverse Faulting – the vertical stress is the minor principal stress, SHmax > Shmin > Sv

Geomechanical modelling allows us to further constrain major and minor principal stress, using
information such as borehole breakouts in image or caliper data, tensile fractures, leak-off tests,
etc. For a complete textbook on these topics, the authors refer to (Zoback, 2008). In complex
environments, i.e. in the presence of viscous materials such as salt (Van der Zee, 2011) or thermal
stress due to injection (Holland, 2015), a 4D Geomechanics approach using the Finite Element
Method is required to determine the stress state.
The importance of the stress state becomes apparent with the knowledge that hydraulic fractures
propagate normal to the smallest principal stress. Therefore, in a reverse faulting environment, a
fracture would be in a horizontal plane and a horizontal well cannot achieve multiple independent
fractures, whereas in normal faulting and strike-slip environments the horizontal well can be
aligned with the least principal stress and multiple hydraulic fractures can be staged. The stress
anisotropy, the difference between major and minor principal stress, can also have a major impact
on hydraulic fracturing, as low stress anisotropy can lead to complex fracture networks due to
interaction and stress shadowing effects.
A number of additional aspects are important:
• Reservoir and rock mass properties are heterogeneous. This aspect requires a 3D volume
modelling approach and statistical methods for representation in the simulations.
• Operational considerations, including the pressure pumping equipment selection,
fluid/proppant properties and energy requirements will determine the allowable ranges
for select simulation parameters.
• Fracture initiation in strong formations can introduce a shift in the importance of
parameters from those known and designed, such as fluid and proppant properties, to
those unknown or having high uncertainty, such as stress and fracture toughness of rock.
Finally, it should be noted that an important application area is found in geothermal energy. The
same processes described above are involved, but additional complexity arises from the fluid and
rock temperatures having a significant influence on the scenario.
Industry benefits of a full 3D hydraulic fracturing simulation capability are found in:
• Optimizing the use of hydraulic fracturing technology, including rig and drilling time,
fluid and proppant quantities, energy requirements and other cost factors.
• Reducing the number of fracturing campaigns that do not add to production. Currently,
many designs rely on experience and an expensive trial-and-error approach.
2. Hydraulic fracturing overview

Many practical approaches are used in the industry to create calibrated models, typically using a
number of assumptions to reduce problem complexity to the level where analytical solutions are
available or fast numerical methods can capture certain aspects of the problem. Common model
categories for the different software solutions include:
1. Single planar fracture
2. Discrete Fracture Networks
3. Complex fractures

As the vocabulary used around the general topic of hydraulic fracturing simulations can be
ambiguous, examples for each category are shown below. The aim of this paper is to contribute to
the understanding and development of simulation capabilities for complex fractures.
2.1 Single planar fracture
One of the advantages of fast analysis of a single fracture using numerical methods is to provide
real-time advice on operational parameters and react to changes in observed versus predicted
behavior. Fractures performed inside the reservoir, bounded by cap rock and basement, can
comprise hundreds of meters of length. This type of analysis, including the use of log data from
the well, can be used to adjust fluid volumes, pressures, proppant concentration, etc., and has been
used in the industry for decades. The inclusion of proppant transport allows predictions of fracture
conductivity (Figure 2), one of the direct measures of the success of the treatment.
Assumptions regarding fracture geometry and analytical forms with a list of references are
summarized in Economides (2013).
Figure 2. Example composite plot of fracture width contours (Meyer, 2014)

2.2 Discrete Fracture Networks
For fractured reservoirs, the connectivity of conductive fractures is important for well trajectory
planning. Assuming critically stressed fractures (ratio of shear and normal stress exceeding the
friction coefficient) are conductive, mapping current and future stress states, following depletion
or injection, onto the fracture planes, a network of connected fractures can be used to assess
production scenarios. This is done by creating representative planar features where the
distributions for spacing and orientation can come from wellbore image analysis or seismic
attributes, referred to as Discrete Fracture Networks (DFN), and resolving the geometric
intersections. Introducing new fractures is typically not part of the workflow in this category. An
example for a DFN analysis of reservoir fractures connected to a wellbore performed using
JewelSuite™ is shown in Figure 3.

Figure 3. DFN connectivity analysis with JewelSuite™ (Holland, 2014).
2.3 Complex Fractures
The assumption of a single planar fracture extending from the wellbore (simple fracture) is often
not justified and more complex fractures can evolve. In cases of low-stress anisotropy, reservoir
depletion can lead to a change in the stress regime and new fractures in a hydraulic fracturing
stimulation campaign at a later stage can open perpendicular to the initial set of fractures. This can
create a complex fracture network, allowing a much larger total fracture volume as shown in the
schematic in Figure 4, after (Fisher, 2004) and (Warpinski, 2008). Additionally, three-dimensional
effects can occur on a small scale in the near wellbore volume, as the stress concentration and flux
around the well can distort the fracture geometry and reorientation can occur. These effects
influence fracture aperture and can become problematic for fluid transport of the proppant around
the curvature of the fracture.
Figure 4. Complex Fracture Network (Fisher, 2004; Warpinski, 2008).

3. Fluid-driven fracture propagation
Independent of the question of whether fractures are pre-existing features of the rock mass or
introduced by fluid-driven fracture propagation, any simulation of rock mass behavior under
deformation quickly converges on the difficulty of introducing discontinuities into the modeling
approach.
3.1 Discontinuous modelling of rock
This is at first a geometric challenge, which for finite element analysis translates into a meshing
challenge, as both the lateral dimension (thickness) of the fracture in the surrounding rock mass is
very small compared to both the typical joint spacing and the required discretization along the
fracture path. This impacts computation of the stress intensity factor and the crack tip opening
deformation for realistic fracture propagation simulation. The ratio becomes extremely
unfavorable when considering reservoir dimensions (reservoir thickness in tens of meters, fracture
length hundreds of meters in some cases) compared to fracture openings in the range of
millimeters or below.
In the last ten years, significant progress has been made in modeling the mechanical behavior of
jointed rock, with the three-dimensional volumes in the simulation often referred to as Synthetic
Rock Mass (SRM). Those simulations performed with Abaqus/Standard and Abaqus/Explicit,
both under static (including explicit quasi-static simulation) and dynamic loading (rock bursts,
blasting and seismicity) significantly improved with the introduction of cohesive elements in
Abaqus version 6.5-1. Examples of geomechanical modeling of faults in Abaqus are shown in
(Arndt, 2007a) using a continuum approach with fine mesh discretization and an implementation
showing discontinuous behavior for frictional fault sliding using the general contact capabilities in
Abaqus/Explicit (Arndt, 2007b) for problems where the high confinement stress justifies the lack
of cohesion for this modeling approach. The use of the explicit solver allowed efficient use of the
parallel solver on a cluster of 32 CPUs, addressing the model size requirements in excess of 10
million degrees of freedom. The first application of Abaqus for SRM samples using cohesive
elements and a Discrete Fracture Network (DFN) approach is shown in (Arndt, 2009). Here, the
discontinuous nature of rock mass is represented with samples containing representative joint
spacing and orientation using both the cohesive element approach, but underlying contact allows
complete separation at the joints on mobility of the newly formed particles. The question of
scalability of rock strength parameters is investigated and applied to problems of fragmentation in
mass mining methods. This leads to further developments for a fully coupled pore-pressure
deformation examples introducing flow in conductive faults in (Beck, 2010), with the limitation
that no change in permeability or any fracture opening and propagation was considered. An
approach to calibrate the strength of joints is shown in (Arndt 2013), where the different behaviors
observed on different length scales are used to derive properties for rock mass consisting of
parallel bedding planes explicitly contained in very detailed models. A detailed review of
numerical methods used in rock mechanics and rock engineering including modelling of fracture
processes and of fluid flow and heat transfer in rock masses can be found in (Brown, 2012).
Crack opening and coupled pressure deformation both for cohesive elements and XFEM were
introduced in Abaqus version 6.14-1. A detailed overview of both methods together with
validation using analytically solvable cases is provided by (Zielonka, 2014). It is shown that for all
regimes, mapped by first fluid storage dominated by either crack opening or leakage and second

by energy dissipation by either fluid viscosity or fracture toughness, Abaqus shows good
agreement with analytical solutions and convergence of results quality for mesh refinement.
3.2 Choice of modeling approach
For the purpose of discussing advantages and limitations of the various modeling approaches
possible for fluid-filled fractures using Abaqus, it seems worthwhile, in the author’s opinion, to
differentiate those methods that integrate the multi-physics challenge into a single analysis
technique (or element formulation approach) and methods that require defining an interface
between two very different simulation methods (i.e. a region where results quantities such as
pressure or deformation are exchanged).
The latter category is independent of the question of having an internal coupling interface defined
in Abaqus or requiring external coupling with an independent simulation code (co-simulation).
Theoretically, a coupled multi-physics approach defining an interface could take advantage of the
different domains (a coupled pore-pressure deformation continuum/discontinuum domain
combined with a purely fluid filled domain) using different discretization. Whilst most of the
examples mentioned in the above discussion regarding the modeling of faults and SRM use
explicit time integration (Abaqus/Explicit), therefore providing the advantage of overcoming the
numerical instabilities that are commonly introduced by strain-softening materials, no pore-
pressure coupled modeling capabilities are currently available in Abaqus/Explicit. Still, multi-
physics simulations using Abaqus/Explicit might be of interest for approximating fluid fracture
propagation in low-permeability cases. Therefore, using this differentiation of methods in this
second category only co-simulation could be considered capable of fully coupled simulation.
The explicit method solver solution time, scales approximately linearly with element number, if all
other aspects including element size (determining stable time increment) remain similar. This
allows very large models to be analyzed, in particular in combination with the explicit parallel
solver, allowing efficient scalability for several hundred CPUs. Information on the scalability of
the implicit solver for this problem type was not available to the authors.
The main options for hydraulic fracturing simulation using Abaqus/Standard are:
• Cohesive Elements
• eXtended Finite Element Method (XFEM)
Additional options include using Abaqus/Explicit (currently limited to fluid fracture propagation
without coupled pore pressure) or third-party software:
• Coupled Lagrange Euler (CEL)
• Smooth Particle Hydrodynamics (SPH)
• Co-simulation
As this work focuses on the analysis capabilities provided by the suite of Abaqus software
products, the very general case of co-simulation with other analysis products has not been
investigated. Emphasis should be given to the fact that from the entries in the list above, only
XFEM allows a simulation without the requirement to introduce a possible fluid fracture path
predefined in the model. Fracture surfaces cannot intersect because each element can only contain
a single enhancement. For all analysis cases, a pre-existing or initial fracture has to be introduced.

A coupled CEL-continuum approach has been briefly investigated and shown to open closed
fractures with a fine discretization of the fluid domain, but insufficient progress diverted the work
to the other methods.
Many of these modeling approaches require further development effort to increase coverage of the
physical effects the simulation aims to replicate, but as complexity increases, the validation of
simulation results becomes a challenge. Analytical solutions are not available to capture the full
range of desired effects and validation would either require a meaningful laboratory test setup or a
full field data approach. The latter case is desirable, but quantities that can be measured in-situ are
limited and costs can be prohibitive. Alternatively, simulation results from other codes can be used
and the work presented in this paper is also used to validate simulation results performed with the
GEOS framework at the Lawrence Livermore National Laboratory (Settgast, 2014).
3.3 Coupled pressure-deformation cohesive elements
Abaqus/Standard provides cohesive elements with constitutive behaviours for:
• Coupled pressure-deformation (Poroelasticity)
• Normal and tangential pore fluid flow
• Element degradation and crack opening
• Crack fluid flow (Poiseuille flow)
An example problem in (Abaqus, 2014a), “10.1.5 Hydraulically induced fracture in a wellbore”
demonstrates the application of coupled pressure-deformation cohesive elements to this type of
problem. A pumping stage propagates a fracture, the high pressure fracturing fluids then bleed-off
into the reservoir, boundary conditions substitute the effect of proppant to preserve the fracture
width and subsequent reservoir drawdown occurs.
This method faces the limitation that the crack path must be defined in advance in the mesh.
3.4 XFEM
With the validation results from (Zielonka, 2014) it is shown that XFEM can used as an alternative
solution to the cohesive approach above. (Abaqus, 2014b), “1.19.5 Propagation of hydraulically
driven fracture using XFEM” also demonstrates good agreement between both methods. The
question of how XFEM fractures behave in complex stress fields seems to remain open, as
rotation of the principal stresses in the above example does not generate a curved fracture.
3.5 SPH
To investigate the ability to model fluid flow inside the fracture using Abaqus with a view on
further applications to rough fracture surfaces and proppant transport, the Smooth Particle
Hydrodynamics (SPH) capability in Abaqus/Explicit has been used in a common benchmark
problem: the Poiseuille flow between stationary parallel plates with parameters listed in Table 1.
This simulation shows very good agreement with the analytical solution as shown in Figure 5 and
Figure 6.

Table 1. Parameters for plane Poiseuille flow (Morris, 1997; Holmes, 2011)
Variable Value
Width L 0.001 m
Viscosity μ 0.001 Pa s
Density ρ 1000 kg/m3
Pressure Gradient F 0.0001 m/s2
Figure 5. Abaqus model with solid element mesh parallel plates and SPH particles
seeded in the fluid domain.
Figure 6. Comparison of the predicted SPH flow velocity with analytical solution
(height profile evolution over time, left, center velocity, right).

Investigations confirming mesh refinement (or particle density) convergence of this method and
on the influence of artificial viscosity to control stable time increment size on shear viscosity were
performed as well. These findings and continuation questions are outside the scope of this paper
and will be published separately in the future.
4. Application of cohesive element hydraulic fracture simulation

Three examples were investigated that used the cohesive element approach, based on the
validation for both cohesive elements and XFEM to correctly describe fluid-filled fracture
propagation and success in modelling intersecting fractures.
4.1 2D Simulation with intersecting fractures
To test the simulation capabilities for intersecting fractures, a quasi 2D-model with square
geometry is constructed in CAE. Abaqus/CAE 6.14-1 does not support the 12-node displacement
and pore pressure three-dimensional cohesive element in the “Mesh” module. Meshing can be
performed by using the 8-node, three-dimensional cohesive element and subsequently editing the
element type. The optional offset parameter allows the automatic generation of the mid-surface
nodes (nodes 9-12) that contain only pore pressure degrees of freedom. This can be performed in
Abaqus/CAE via Model, Edit Keywords or separately editing the input file to modify the entry for
the element definition for type COH3D8 by changing the element type and specifying an offset
value:
*ELEMENT, TYPE=COH3D8P, OFFSET=offset
The mesh stack orientation should be defined in the “Mesh” module to ensure the cohesive
elements are inserted correctly.
This approach does not allow generating intersecting fractures. In this case, all connected elements
need to have common pore pressure nodes, which is not compatible with the underlying geometry
and requires either further editing or mesh generation via scripting. The independent constitutive
and geometric thicknesses of the elements would allow inserting zero thickness cohesive elements
at all element boundaries without distortion of the mesh.
For this example, a model is created by defining a single layer of C3D8P elements (regular mesh,
100m x 100m, 1m size) intersected by a narrow geometry domain of 0.01m to contain the
cohesive elements (COH3D8P) in the mid-planes. The center piece solid element is deleted and
one node for each side of the layer is created in the center to be used as common pore pressure
nodes on the four attached cohesive elements.
The parameters from validation cases in (Zielonka, 2014) are used in the simulations (Young’s
modulus E = 17 GPa, Poisson Ratio v = 0.2, Fluid viscosity μ = 1 cPoise). An initial open fracture
is defined on the left side, with the initial length chosen as 25% of the model dimensions, using
the keyword:
*INITIAL CONDITIONS, TYPE=INITIAL GAP
ElementSet

Both pressure and inflow volume-controlled boundary conditions yield the same results, either
using:
*BOUNDARY
NodeSet, 8, 8, InjectionPressure
or
*CFLOW, AMPLITUDE=RampUp
NodeSet, , InjectionVolumeRate
Following an initial *GEOSTATIC step to ensure equilibrium of the initial conditions for stress and
pore pressure, a *SOILS, CONSOLIDATION analysis step is performed using the unsymmetric matrix
solver as required for this problem type. Changes in pore pressure can be large, so an adequate
choice of the convergence tolerances for pore pressure, defined using the UTOL parameter, is
necessary. The initial in plane stress is set to S11 = 15MPa, S11 = 10MPa and S33 = 10MPa. For the
second model, S11 and S33 are swapped, rotating the stress 90 degrees.
The fracture propagation for the two models is shown in the time sequence in Figure 7 using the
fracture aperture variable (PFOPEN) and a 500x displacement magnification.
The simulation demonstrates the expected behavior. Initial tests were performed with the injection
in the center and initial gaps defined for all four cohesive elements connected to the central pore
pressure nodes to confirm fracture propagation occurs only in the plane normal to the least
principal stress. In this example the possible fracture paths are limited to the cohesive element
regions, so in the scenario where the least principal stress is horizontal, a redirection of the fracture
propagation direction can only occur at the center of the model. At this stage the fracture does not
propagate further laterally.
An important observation in the sequence of pictures is the reduction of aperture of the horizontal
fracture as the fluid escapes into the vertical fracture and stress shadowing occurs. This very
realistic behavior has significant implications for real-life stimulation treatments, as proppant
transport inside the fluid is affected in rather complex ways and the proppant might choke the
fracture tip.

Time SH = 15.0 MPa / SV = 10.0 MPa SH = 10.0 MPa / SV = 15.0 MPa
300 s
600 s
900 s
1200 s
Figure 7. Evolution of fluid injection showing fracture aperture (PFOPEN) for

cohesive elements with displacement magnification of 500x.

4.2 Validation against experimental data
Results in (Blanton, 1982) allow the validation against experimental data. Here, samples of
dimensions 30cm x 30cm have been fabricated using synthetic rock blocks (“hydrostone”) with
existing fractures embedded at different angles (90°, 45°,60°) and subjected to different tri-axial
compressive stress states before fluid injection propagates a hydraulic fracture. Three of these
experiments, showing all three different observed interaction modes in the same model geometry,
were chosen for validation (Table 2).
Table 2. Validation experiments chosen from (Blanton, 1982)
Fracture Interaction
Test SXX SZZ
Angle θ Mode
CT-4 60° 12.0 MPa 10.0 MPa Opening
CT-8 60° 20.0 MPa 5.0 MPa Crossing
CT-21 60° 14.0 MPa 5.0 MPa Arrest
As several parameters from the experiments, such as injection fluid pressure and hydrostone
strength (fracture toughness), are not provided in the paper because we follow the argumentation
in (Fu, 2012) to narrow the possible range according to observations from the tests. This also
allows for comparison with these and other numerical simulations using the same reference.
Contrary to the approach in (Fu, 2012), where crack propagation crossing the natural fracture is
not allowed and the stress intensity factor KI is evaluated instead, the Abaqus simulation allows
crossing through the natural fracture by extending the cohesive elements to the specimen
boundary.
An initial gap opening has to be specified in the model for the cohesive elements at which fluid
injection occurs as in the previous example. It would seem obvious to use the same approach to
create the natural fracture, but the element degradation of the shear stiffness changes the stress
equilibrium in the geostatic step significantly, especially in the case of high differential stress (CT-
8). The fracture toughness has no major influence on the fracture propagation; therefore, it is
sufficient to have the natural fracture closed both in the initial conditions and in equilibrium by
choosing a shear strength larger than the shear stress on the inclined plane. The presence of
cohesive elements subsequently allows the fracture to open.
A coarser regular mesh and a finer advancing front mesh with ~4000 and ~20,000 elements,
respectively, have been tested with similar results. A time sequence of the three cases is shown in
Figure 8, where the fine mesh is suppressed for better clarity. Displacements are magnified 100x.

CT-4 (Opening) CT-8 (Crossing) CT-21 (Arrest)
Time
12.0 MPa 10.0 MPa 20.0 MPa 5.0 MPa 14.0 MPa 5.0 MPa
0.30 s
0.45 s
0.60 s
0.75 s
Figure 8. Evolution of fluid injection showing gap flow volume rate (GFVR).

As shown in the evolution of fracture gap flow volume rate, all three tests show the correct
fracture interaction mode. CT-21 does show minor fluid propagation along the natural fracture, but
this is small compared to the fluid volume and fracture opening on the other side of the injection
point. Again, higher confining stress shows a smaller crack opening in CT-4 as in the previous
example.
4.3 3D Simulation with parallel fractures
Current research into the interaction of different fracture stages, stress shadowing between parallel
fractures and alternative sequencing between adjacent parallel wells (‘Zipper Frac’) demonstrates
the significance of 3D simulation capabilities. (Bunger, 2014) and (Izadi, 2015) provide analyses
of the parameters for closely spaced fractures and a detailed analysis of the zipper frac concept
using a continuum damage approach with Abaqus is presented in (Shen, 2014).
The chosen test case for 3D simulation of parallel fractures with cohesive elements is the
simultaneous propagation of equally spaced fractures under the assumption of equal injection
pressure, neglecting wellbore hydraulics and perforation friction effects. A 3D model was built
using Abaqus/CAE, including the wellbore geometry (75/8 in diameter), a reservoir volume of 40m
vertical extent with cap rock above and basement below for a total height of 60m, five parallel
planes meshed with cohesive elements with 5m separation, and partitioning to allow for mesh
transition from the well. Lateral symmetry is used to reduce element numbers. No vertical
symmetry was used while building this model to allow for gravity and stress gradient effects
(although the case presented here does not include these effects). The model geometry and mesh
with ~100,000 elements are shown in Figure 9.
Figure 9. 3D model with horizontal well, reservoir volume, five planes containing
cohesive elements for hydraulic fractures (left) and mesh discretization (right).
Using pressure boundary conditions as in the previous examples, volume flow and fracture
propagation occurs. Higher lateral stress and rock stiffness in the cap rock and basement provide
fracture containment inside the reservoir.
Unlike a simulation with a constant volume flow rate for each fracture, the constant pressure
generates flow rates affected by the stress changes. The inner fractures become restricted by the
presence of neighbors on both sides and fracture aperture and total volume are reduced
significantly as seen in Figure 10. This quantitative simulation experiment is consistent with

published simulation results. Further work is required to quantify the effects above and validate
results against other available sources.
Figure 10. Evolution of fracture aperture in simultaneous parallel fractures.

Analysis times for this 3D problem are in the range of several hours using a single computer with
16 CPU cores, indicating that there is no immediate practical limitation for this simulation
approach.
5. Conclusions
The stated challenge for this paper is to extend applications to more complex geometries with
interaction of multiple fractures and stress shadowing effects. By choosing the cohesive element
approach in Abaqus/Standard it becomes possible to perform 3D hydraulic fracturing simulations
that satisfy the following criteria:
• The analysis capability includes essential physical behaviors such as coupled pressure-
deformation (Poroelasticity), fracture propagation and fracture fluid flow with leak-off.
• Validation of propagation of a single fluid-filled fracture against analytical solutions,
such as KGD geometry and Penny-shaped cracks, and mesh convergence have been
shown (Zielonka, 2014).
• Successful validation of 2D fracture interaction observed in experiments. Different
interactions modes such as fracture diversion (or opening), crossing a natural fracture,
and arrest are predicted correctly.
• The implementation of intersecting cohesive elements used in the 2D examples can be
extended to include all element boundaries in the model in 3D, providing a general 3D
simulation capability.
• Interaction of parallel fractures in 3D simulations and qualitatively correct behaviors as
stress-shadowing-limiting fracture apertures, as shown in simulations using other codes,
is observed.
• Acceptable performance and time incrementation for large 3D problems is demonstrated,
combined with scalability using the parallel solver.

A major limitation lies in the necessary choice of either using cohesive elements or XFEM.
Cohesive elements placed at solid element boundaries limit the possible fracture path to element
boundaries and XFEM does not support intersecting fractures or nucleating new fractures.
These are common challenges in the focus on coupled 3D hydraulic fracturing simulation
capabilities for the oil and gas industry in recent years and the desire to include more effects will
drive further developments. With the approach shown in this paper Abaqus in its current release
can be used to increase understanding of interaction of fractures in stimulation campaigns and to
help reduce expensive trial-and-error approaches.
6. References
1. Abaqus Benchmarks Guide, version 6.14-1, Providence, RI, USA, 2014.
2. Abaqus Example Problems Guide, version 6.14-1, Providence, RI, USA, 2014.
3. Arndt, S., Beck, D., Reusch, F., Thin, I., Stone, C., Heap, M. & Tyler, D. “Deep and High
Stress Mining – Deformation and Seismicity”, In Proceedings of Abaqus International Users’
Conference. Paris, France. May 22-24, 2007.
4. Arndt, S., Beck, D. & Reusch, F., “Modelling of Large Open Pit Stability Using ABAQUS”,
In Proceedings of Abaqus International Users’ Conference. Paris, France. May 22-24, 2007.
5. Arndt, S., “Advances in Mining Simulations”, In Proceedings of Simulia Customer
Conference. London, England. May 18-21, 2009.
6. Arndt, S., Fillery, B., “The Importance of Simulation in Geotechnical and Mining
Applications”, In Proceedings of Simulia Customer Conference. Barcelona, Spain. May 17-
19, 2011.
7. Beck, D., Fillery, B., & Reusch, F., “3D Hydro-mechanical Simulation of Faulted Open Pit
Slopes”, 44th U.S. Rock Mechanics Symposium and 5th U.S.-Canada Rock Mechanics
Symposium, 27-30 June, Salt Lake City, Utah, 2010.
8. Blanton TL. An experimental study of interaction between hydraulically induced and pre-
existing fractures. Proceedings of SPE Unconventional Gas Recovery Symposium. Pittsburgh,
Pennsylvania. Society of Petroleum Engineers, 1982; 559–571. DOI: 10.2118/10847-MS.
9. Brown, T., “Deep Mining 2012 - Y. Potvin (ed)”, Australian Centre for Geomechanics, Perth,
ISBN 978‐0‐9806154‐8‐7, Perth, 2012.
10. Bunger, A.P., Peirce, A.. Numerical simulation of simultaneous growth of multiple interacting
hydraulic fractures from horizontal wells. In ASCE Shale Energy Engineering Conference,
Pittsburgh, PA, July 21–23 2014. ASCE.
11. Economides, M.J, Hill, A.D., Ehlig-Economides, C., Zhu, D., “Petroleum Production
Systems”, Prentice Hall, 2012.
12. Fisher, M.K., Heinze, J.R., Harris, C.D., Davidson, B.M., Wright, C.A., Dunn, K.P.,
“Optimizing Horizontal Completion Techniques in the Barnett Shale Using Microseismic
Fracture Mapping”, 90051-MS SPE Conference Paper, 2004.
13. Fu, P., Johnson, S.M., Carrigan, C.R., “An explicitly coupled hydro-geomechanical model for
simulating hydraulic fracturing in arbitrary discrete fracture networks”, Int. J. Numer. Anal.
Meth. Geomech. 2012.
14. Holland, M., “JewelSuite™ DFN Plugin”, Baker Hughes, 2014.

15. Holland, M., “Directionless fracture stability assessment based on a 4D geomechanical model
of a SAGD scenario”, First Break, Vol 33, No 3, March 2015, pp. 47 – 52, EAGE, 2015.
16. Holmes, Williams & Tilke, Int. J. Numer. Anal. Meth. Geomech., 35, 419-437 (2011).
17. Izadi, G., Settgast, R., Moos, D., Baba, C., Jo, H., “Fully 3D Hydraulic Fracture Growth
Within Multi-Stage Horizontal Wells”, ISRM, 2015 – to be published.
18. Meyer, B. R., “MFRAC User’s Guide, Tenth Edition”, Meyer Fracturing Software Group,
Baker Hughes Incorporated, USA, 2014.
19. Morris, Fox & Zhu, J. Comp. Phys., 136, 214-226 (1997).
20. R. R. Settgast, S. M. Johnson, P. Fu, S. D.C. Walsh, C. Annavarapu, Y. Hao, and J.A. White,
F.J. Ryerson, “GEOS: A Framework for Massively Parallel, Multi-Physics Simulations.
Theory and Implementation.” LLNL Technical Report 654611. (LLNL-TR-654611).
21. R R. Settgast, S. M. Johnson, P. Fu, S. D.C. Walsh, Lawrence Livermore National
Laboratory, “Simulation of hydraulic fracture networks in three dimensions utilizing
massively parallel computing resources”, Unconventional Resources Technology Conference,
August 2014.
22. Shen, X., “Numerical Analysis on the Interaction between Two Zipper Frac Wells Using the
Continuum Damage Method”, OTC-24975-MS, Offshore Technology Conference-Asia, 25-
28 March, Kuala Lumpur, Malaysia (2014).
23. Warpinski, N.R., Mayerhofer, M.J., Vincent, M.C., Cipolla, C.L. and Lolon, E.P., 2008,
“Stimulating Unconventional Reservoirs: Maximizing Network Growth While Optimizing
Fracture Conductivity”, SPE 114173 presented at the SPE Unconventional Reservoirs
Conference, Keystone, Colorado, U.S.A., 10-12 February 2008.
24. van der Zee, W., Ozan, C., Brudy, M., Holland, M., “3D Geomechanical Modeling of
Complex Salt Structures”, In Proceedings of Simulia Customer Conference. Barcelona,
Spain. May 17-19, 2011.
25. Zielonka, M. G., Searles, K. H., Ning, J., Buechler, S. R., “Development and Validation of
Fully-Coupled Hydraulic Fracturing Simulation Capabilities”, In Proceedings of Simulia
Customer Conference. Providence, RI, USA. May 19-22, 2014.
26. Zoback, M. D., “Reservoir Geomechanics”. Cambridge University Press, 2008.
7. Acknowledgment
The authors would like to thank Baker Hughes management for permission to publish this article,
Pengcheng Fu, Randy Settgast (Lawrence Livermore National Laboratory) and Brent Fillery
(Rolls Royce Aero Engines) for stimulating discussions about numerical simulation methods and
David Martin (DS SIMULIA South), Xiaoliang Qin (DS SIMULIA Headquarters), Xu Huang
(Baker Hughes Diamond Research) and Huaning Zhu (DS SIMULIA Headquarters) for valuable
discussions regarding SPH modelling.

Fracture Stability Assessment during SAGD
operations using 4D Geomechanical Models
Marc Holland, Wouter van der Zee
Baker Hughes, Reservoir Technology and Software
Abstract: In the oil and gas industry, steam assisted gravity drainage (SAGD) is an operation
where steam is injected into a reservoir to lower the viscosity of heavy oil to allow its production.
The injection of steam leads to thermally induced stresses and pore pressure changes in the
reservoir and its proximity. Although the pressure and temperature evolution can be modelled
with reservoir simulators, they neglect the change of the stress field in relation to these properties,
which can impact the stability of faults and fractures. These active faults and fractures, in turn,
affect the fluid transport properties. In order to assess the potential for fault and fracture (re-
)activation we created a model of a SAGD operation by coupling the outcome of a reservoir
simulation to a 4D geomechanical finite-element model using Abaqus. The stresses calculated in
the Abaqus model are used as the input for our fracture stability calculations. First, a
conservative analysis uses a directionless assessment without the need for discrete fracture and
fault orientations. It shows extensive critically-stressed regions above and below the injection
well, which extend into the cap rock. A second analysis using the available fault models reveals
that the bounding faults have critically stressed patches toward the end of the stimulation.
Quantitative risk assessment is performed to investigate the influence of a range of input
parameters on the fault stability and to quantify the risk of fault reactivation. The outcome of the
study stresses the need for a detailed discussion of the cap rock thickness and the impact of
potential of fractures.
Keywords: Oil and Gas, Geomechanics, Steam Assisted Gravity Drainage, SAGD, Reservoir,
Thermal Expansion, , Abaqus, JewelSuite, , Stress, Displacement, Permeability, Cap Rock, Seal
integrity, Faults, Fractures, Stability, Critically Stressed, Reactivation
1. Background
The choice of how to produce hydrocarbons from the subsurface depends on a variety of
circumstances. For shallow reservoirs holding hydrocarbons which are too viscous to be produced
economically with conventional means, a technique can be applied which is called “steam assisted
gravity drainage” or SAGD (Butler, 1991). Here the viscosity of the hydrocarbons is reduced by
locally increasing the rock temperature with the injection of hot steam. For this operation typically
two sub-horizontal wells, parallel in the vertical plane, are drilled. In the upper well steam is
injected over a prolonged time. Through convection and conduction the rock surrounding the well
is heated and the elevated temperatures mobilize the hydrocarbons which are then produced from
the lower well. The operational parameters for SAGD need to be constrained to honor both
economic as well as physical limits. The economic limits are primarily based on technical

expenses such as the cost of surface equipment and the cost of prolonged steam injection. Physical
limits on the other hand are primarily linked to the safety of the operation, where the containment
of the steam inside the reservoir is of prime importance. The injected steam is confined in the
reservoir by the same sealing elements which are responsible for the accumulation and
containment of hydrocarbons over geological time scales. Seals may be created by bounding faults
which may act as lateral flow barriers, as well as “cap-rock” which is impermeable and acts as a
vertical flow barrier. The integrity of these sealing elements needs to be maintained during the
injection of steam to avoid leakage of steam and hydrocarbons into shallower layers, the aquifer,
or even to the surface (ERCB, 2010; ERCB, 2013). A compromised seal may thus impose serious
health, safety and environmental threats.
Figure 1 – The SAGD model represents a shallow reservoir at

approximately 280 m depth. The reservoir layer is approximately 70 m thick
as defined by the two horizons and is laterally bound by the two major fault
surfaces. A third small fault is presumed to be present inside the reservoir.
These model elements are represented by triangulated surfaces. Other
objects include well bores for the injection and production as well as two
reservoir grids (blue) in which the injection of the steam and production of
the hydrocarbons are modelled. The contoured horizon at the top
describes the surface topography.
In order to assess the operational parameters and minimize the risk of losing the seal integrity we
combine the outcome of a reservoir simulation with a geomechanical simulation carried out in
Abaqus. Here, the temperature and pore pressure evolution of the reservoir section are fed into a
mechanical simulation of the subsurface system to solve for the evolution of the effective stresses.

These results are then used to assess the likelihood of causing fracture slip, which could
compromise the seal integrity. This workflow is demonstrated in a shallow reservoir undergoing
SAGD as illustrated in Figure 1. Here, a reservoir compartment of approximately 70 m thickness
is located at a shallow depth of approximately 280 m. The reservoir layer is laterally bound by two
normal, low-offset faults. In this reservoir two SAGD operations are carried out where steam is
injected into two sites simultaneously over a prolonged period. The increase in the rock
temperature at these sites is locally more than 250°Celsius.
2. Pre-processing and geomechanical simulation

To assess the risk of fracture activation or seal breach, an integrated approach is necessary in
which different simulation techniques and grids need to be combined. As previously mentioned,
the input for the geomechanical simulation is partly taken from the reservoir simulations. The
reservoir model is a purpose-made fluid flow simulation, typically carried out to constrain the
economic limits. Using a finite-difference approach, the thermal and pressure evolution of the
reservoir layer is modelled using both injection and production. The reservoir simulation considers
the chemical composition of the hydrocarbons, the associated phase changes and hydrocarbon
saturations. The reservoir grids typically are layer conforming. Grid cells are commonly IJK
organized hexahedral cells with high length to height ratios so they better replicate the flow
properties imposed by geological layering.
To integrate the data we use the sub-surface modelling platform JewelSuite ™ in which the
reservoir models are loaded along with supporting structural elements (faults, horizons,
topography) and well data. Two reservoir simulation grids are present in which the temperature
and pressure evolution of the SAGD operation were modelled. These reservoir grids each have
about 40 thousand grid cells (33×17×73 elements in IJK), holding in total 80 reported time steps.
These highly-specialized simulations cover only the reservoir section and are laterally confined as
the changes in temperature and pressure are believed to act locally. The geomechanical simulation
of the stress evolution is carried out with Abaqus. In order to link these simulations, several pre-
and post-processing steps are necessary, which are described below in more detail.
The sub-surface stress state in the reservoir is dependent not only on the pressure and temperature
evolution as calculated by the reservoir simulation but it is a function of a much larger system.
The contribution of the vertical load of the overburden for example requires the geomechanical
model to be extended upwards to the topographic surface. Effects related to the stiffness of the
rock layers require not only the knowledge of the elastic parameters throughout the model, but
also require the model to be extended laterally (“sideburden”) and downwards (“underburden”). In
order to capture these influences and to avoid artifacts from the model boundary, the mesh for the
geomechanical model covers a vertical column from the surface to about twice the depth of
interest (>560 m) and has lateral dimensions of at least three times the size of the area of interest
in which the steam injection is conducted. The geomechanical simulation therefore covers a much
larger volume compared to the reservoir simulation. As property data and the supporting structural
elements are typically missing outside the volume of interest, user interaction and user decisions

are required to construct a large enough property volume for the geomechanical simulation. The
triangulated surfaces available for topography, faults and horizons are thus extended outwards and
the geological grid is constrained by these extended surfaces. This geological grid honors the
structural constraints and incorporates available formation top data from well-based
interpretations. The purpose of using a geological, layer-conforming grid is to replicate the
geological layering to assist in the population of geomechanical properties. The required
properties for the simulation are density, the elastic properties Young’s modulus and Poisson’s
ratio as well as the ambient stresses, pore pressure and information on the thermal expansion
coefficient. These properties are typically modelled based on the well log data to form a verified
well-centric 1D geomechanical model. The 1D data is then extrapolated from the wells into the
3D geological grid to form a 3D property container. Details on these two workflow steps are
beyond the scope of this paper, but see e.g.: Holland et al. (2010) or Zoback (2007).
The constraints for the mesh for the geomechanical simulation are prepared in the JewelSuite™
subsurface modeling software. The triangulated surfaces of the structural elements are re-meshed
in such a way that a “water-tight” structural model is created. As the triangulated surfaces are used
as mesh seeding points in Abaqus, we use the triangle size on the surfaces to enforce a meshing
gradient. The Abaqus mesh is constructed from tetrahedrons. As the replication of thin but
laterally extensive layers with tetrahedrons is computationally expensive, we only use high
resolution tetrahedrons in the central volume of interest and gradually increase the element size
outwards (van der Zee, et al., 2011). The “coarsening step” of enforcing a gradual increase of
triangle size is performed in the subsurface modeling software to ensure quality control and ease
of interaction with the original objects. The meshing of the geomechanical simulation grid is
performed in Abaqus/CAE using a data exchange through the Python™ interface. Details of the
meshing procedure can be found in van der Zee et al. (2011). Upon successful creation, the mesh
is subsequently imported and visualized in the subsurface modeling software. If necessary, poor
elements can be identified in the mesh and the triangulated surfaces can be modified accordingly
for another meshing run. In the presented scenario the geomechanical model is built with three
reservoir faults and a few structural horizons. Grid or mesh refinement is done closer to the well
trajectories for a high resolution in the volume of interest. The created finite element mesh
consists of 809,000 C3D10MP elements (Figure 2).
We did not implement the faults as sliding surfaces (e.g. contact modeling) in the analysis because
the slip of such a contact would alter the stress field. Although modeling such behavior may be
more accurate, the purpose of the analysis is not to model the consequences of fault slip but to
assess the risk of slip initiation. The faults are included as geometric features in the mesh where
the absence of slip provides stress continuity, which allows us to conservatively assess the risk of
slip. This is required as the faults’ properties and the faults’ orientations are not precisely known.
We assess the risks and uncertainties behind these assumptions below, using dedicated analytical
geomechanical techniques (see Post-processing and results).

Figure 2 – A k-slice from the two reservoir simulation grids (left) and the
same slice through the 3D mesh used for the geomechanical simulation
(right). Both images use temperature for the color coding.
The Abaqus input file for the geomechanical analysis is created fully using the JewelSuite™
software. Using the mapping capabilities and the unit management system of the software, some
material properties are mapped (e.g., temperature, Figure 2) from the 3D geological grid onto the
3D mesh and provided as field variables where the Abaqus material properties in the mesh are
made dependent on these field variables. Other material properties are defined per element set
(Table 1). The mesh for the finite element simulation consists of C3D10MP elements and a linear
poro-elastic material model. The necessary boundary conditions for the simulation are derived
from the regional geological stress field as present in the 1D geomechanical models.
After an initialization step to obtain a stress state which is in equilibrium with the input parameters
(e.g. density and gravitational load, see van der Zee et al., 2011), the changing pore pressure and
temperatures are applied to the model. The pore pressure field is applied as changing boundary
conditions (d.o.f. 8) at every node in the model. The temperature field of the reservoir models (as
present in the 3D geological grid) is applied using the *Temperature option on the nodes with
changing temperatures.

The created input file is then submitted to Abaqus/Standard for calculation. From the 80 reported
time steps of the reservoir simulation, every 2nd step was used in the geomechanical simulation.
Upon completion the results are read back into the JewelSuite™ subsurface modeling software for
post-processing in the geological context.
3. Post-processing and results

During the post-processing step we assess fault stability using a dedicated geomechanical model in
the JewelSuite™ subsurface modeling software. The full stress tensor and pore pressure obtained
from the finite element simulation are used for the geomechanical analysis, and a Mohr-Coulomb
failure criterion is used as the basis for determining whether faults may have slipped. In this step,
we focus on the re-activation of preexisting fault and fracture planes. The effective normal stress
and shear stress component on each plane is calculated based on the fracture orientation and the
surrounding stress field (Chou & Pagano, 1967). Assuming no cohesion (because the faults and
fractures already exist) and a sliding friction coefficient of 0.6 (Byerlee, 1978) we calculate the
ratio of the critical shear stress to the actual shear stress. This measure is called Tau Ratio (van der
Zee, et al., 2014). Values of Tau Ratio close to 1.0 or above are considered to be critically
stressed, where the fracture is at risk of slipping. Due to asperities on the fracture planes, slip
would create openings along the fracture plane and thus increase the fracture’s permeability
(Barton & Zoback, 1995). Critically stressed fractures or faults therefore may pose a risk of
compromising seal integrity.
There are significant uncertainties in determining fault and fracture orientations, locations, and
densities. First, the faults which are to be tested in the analysis are commonly interpreted from
subsurface seismic data. These data sets may have a considerable spatial uncertainty and a coarse
resolution making it difficult to provide accurate interpretations. Small-scale undulations of the
fault plane may be unresolvable, leaving the analysis biased toward the average orientation of the
faults. Second, smaller scale fractures are not visible at all on sub-surface seismic as they fall
below the resolution of the measurements. However, since fractures are extremely common in all
rock, their potential contribution to the risk should be assessed.

Figure 3 – Top view of two of the reservoir faults. The faults are color-
coded for Tau Ratio (P50) at the final step of the analysis. Red colors
indicate critically stressed sections of the fault. The four marked points
indicate locations on the fault on which the time series for the P10, P50 and
P90 values of Tau ratio are sampled, shown in Figure 4. The blue and red
spheres are showing critically stressed sections on the fault. The green
and yellow spheres show non-critical sections. The temperature body of
the steam injection is rendered semi-transparent for reference.
We map the stress field from the 3D mesh onto the nodes of the triangulated fault surfaces to
calculate the fault slip risk of the bounding faults. Rather than calculate only one pair of shear and
effective normal stress values for every single node (Holland, 2015), we perform 1,000 Monte
Carlo simulations at each node to incorporate the model uncertainties. We believe that such an
uncertainty assessment would be too time consuming in Abaqus, and therefore decided to run
multiple analyses on the results in JewelSuite™. At each node, some the original input parameters
are altered by random offsets using Gaussian distributions. The pore pressure is allowed to be
changed with a standard deviation of 0.1 MPa, the dip angle by a standard deviation of 10.0
degrees, the azimuth of the dip direction by 5.0 degrees and the sliding friction coefficient by
0.05. From these 1,000 realizations for every node, the results are sorted and only the P10, P50
and P90 results stored on the triangulated surfaces (Figure 3). Figure 4 shows a plot in which the
time series of Tau Ratio is presented at the four locations indicated in Figure 3.

1.6
Point 1
Point 1
1.4
Point 1
1.2 Point 2
Point 2
1 Point 2
Tau Ratio
Point 3
0.8
Point 3
Point 3
0.6
Point 4
0.4 Point 4
Point 4
0.2
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
Time Step
Figure 4 – Time series for the Monte Carlo simulation of Tau Ratio at four
selected locations shown on the fault in Figure 3. Both the blue and red
curves exceed the critical limit of 1.0. The green point lies outside the
influence distance of the steam injection and shows little to no change. The
P50 (solid, bold) result of the yellow point moves to more stable conditions.
Note the different ranges between the P10- P90 pairs (dashed lines), where,
e.g., a much larger uncertainty is present in the red curves compared to the
blue curves.
The four probed locations in Figure 4 show significantly different results. Two of the sampled
points show a trend of Tau Ratio (P50) that exceeds the critical limit of 1.0. Point 1 shows the
most critical values. The uncertainty range of point 1 is however noticeably smaller compared the
range from point 4. The time series of point 3 - located at some distance from the stream injection
- remains constant through the time series whereas point 2 becomes more stable over time.

Figure 5 – Volume rendering of the directionless Tau Ratio in which the
critical values are shown as opaque red clouds. The critical sections are
located directly above and below the temperature bodies (Last time step in
analysis, temperature body is rendered as reference).
In order to understand the results in more detail and also assess the sub-seismic fractures which
may be present, we perform a volume operation on the geomechanical 3D mesh directly, which is
explained in detail in Ansari et al (2012) and Holland (2015). We use the effective stress tensor
from the Abaqus simulation to calculate Tau Ratio for the most critical fault or fracture orientation
possible. This conservative approach reveals whether the worst-case fault orientation would yield
critical conditions. As no observed fault orientations are needed for this calculation, the analysis
can be applied to the entire stress volume and can be rendered accordingly (Figure 5).
The volume rendering of the results shows that potentially critically stressed conditions are present
below and above the steam injection, rendered in Figure 5 as opaque red clouds. The growth of the
clouds coincides primarily with rising temperatures in the direct vicinity of the steam injection and
less with the pore pressure changes. The critical volume extends upwards into the impermeable
cap rock. We further investigate the cap rock integrity by extracting the maximum tau ratio over
the entire time series from the mesh at the interface between the cap rock and the reservoir (Figure
6). The critical areas are located directly above the steam injection bodies. The two lower
hemisphere plots give an indication of the fracture orientations which would yield critical values
at these locations, where red sections in the lower hemisphere plots reveal the fracture orientations
which would be critically stressed. The two locations probed show that a wide range of fracture or

fault orientations are potentially critically stressed. Despite their close proximity (30 m), the
critical orientations of fracture poles differ significantly between the two stereo plots.
Figure 6 – Map view of the most critical Tau Ratio values of the time series
extracted at boundary between the cap rock and reservoir. Areas that could
have critically stressed fractures are rendered in red. For two locations,
lower hemisphere plots are drawn which highlight the critical fracture or
fault orientations (Lower hemisphere stereo plots show critical orientations
of fracture poles).

4. Discussion
Figure 7 – Sketch of the changes in the stress conditions above and at the
sides of the radially expanding temperature body. At the sides, the
expansion increases the minimum horizontal stress component, making
the Mohr Circle smaller. Above and below, the thermally induced stresses
increase the vertical stress component, making the Mohr Circle larger.
Stress perturbations in the presented SAGD scenario are a result of the thermal expansion of the
rock in response to the steam injection. The displacement caused by the injection happens radially
around the sub-horizontal wellbores and reaches displacements of approximately 10 cm in the
sub-surface, and 2 cm at the surface (Holland, 2015). The stress field perturbations are confined to
the proximity of the steam injection. Fault surfaces in close proximity to the injection sites
become critically stressed and potentially shearing early in the time series, presenting a risk for
loss of the lateral seal integrity (Figure 3, Figure 4). As these faults partly extend into the
impermeable cap-rock, a slip event in this scenario may also lead to vertical fluid migration
(Figure 1). An assessment of the worst-case fracture stability reveals the critical volumes in which
a potential for fracture activation exists (Figure 5). The critical volumes are located above and
beneath the steam injection. This pattern is related to the stress perturbation of thermal expansion
of the rock within the present stress field.
In this scenario, the well trajectories are drilled in the direction of the maximum horizontal stress
SHmax (Table 2). The thermal expansion causes the thermally induced stresses to act radially
around the well bore. Above and below the well bore, the changes primarily increase the vertical

stress, whereas laterally the minimum horizontal stress is increased. As the minimum horizontal
stress is approximately the minimum principal stress and the vertical stress the maximum principal
stress, a Mohr Circle (Jung, 1946/1947) for a location left or right of the steam injection would
shrink in diameter. A Mohr Circle on top or below would increase in diameter (Figure 7).
The stress perturbations are acting locally, but the volume of the potentially critically stressed
fractures extends vertically into the cap-rock (Figure 5), and the critical area can be mapped onto
the interface between the reservoir and cap-rock (Figure 6). Here a detailed fracture assessment
should be carried out to rule out the possibility of the existence of critically oriented fractures. Due
to the radial expansion of the rock around the steam injection well bores and the interaction of the
two injection sites, the critical fracture orientations change with location (Figure 6).
5. Conclusion
Subsurface operations where significant changes in temperature or pressure are expected are likely
to experience local perturbations of the stress conditions. This was demonstrated on a SAGD
scenario in which the thermal expansion caused by the injection of steam leads to the changes. For
a risk assessment, the results of a reservoir simulation can be used to feed into a geomechanical
simulation of the subsurface system. Due to the fundamentally different simulation meshes and
different properties and requirements of the simulations and meshes, an integration platform is
required which enables the construction and population of the anticipated 3D mesh for the
geomechanical simulation. The required extension in size as well as the population of the model
(with properties which are typically available as a well-centric 1D geomechanical model) are
carried out in JewelSuite™. Here also a structural model is constructed based on triangulated
surfaces.
Enforced gradients in the triangle size of these surfaces are used to constrain the tetrahedron sizes
of the Abaqus mesh. Using a linear-elastic material model, a one-way coupling is used where we
provide the temperature and pore pressure evolution as boundary conditions to the
Abaqus/Standard solver. The post-processing of the results is done in the JewelSuite™ subsurface
modeling software to provide for the application of specialized geomechanical probing and
analysis tools.
Upon the identification of critically stressed fractures, locations and time steps, purpose made
tools help the user to identify fracture orientations with high risk.
6. Acknowledgement
The authors like to thank Baker Hughes Inc. management for permission to publish this article.

7. References
Ansari, S. et al., 2012. Caprock Integrity Case Study for Non-thermal Polymer Flooding Project
Using 4D Reservoir Coupled Geomechanial Simulation. SPE Heavy Oil Conference,
Calgary/Canada, SPE (157912).
Barton, C. A. & Zoback, M. D., 1995. Fluid flow along potentially active faults in crystalline rock.
Geology, 15(5), pp. 467-470.
Butler, R. M., 1991. Thermal recovery of oil and bitumen. Englewood Cliffs, N.J.: Prentice Hall.
Byerlee, J. D., 1978. Friction of rock. Pure & Applied Geophysics, Volume 116, pp. 615-626.
Chou, P. C. & Pagano, N. J., 1967. Elasticity - Tensor, Dyadic, and Engineering Approaches.
Princeton, N.J.: Van Nostrand.
ERCB, 2010. Total E&P Canada Ltd.Surface Steam Release of May 18, 2006, Joslyn Creek
SAGD Thermal operation, Alberta: Energy Resources Conservation Board.
ERCB, 2013. Primrose East Bitumen Emulsion Release, January 3, 2009, Alberta: Energy
Resources Conservation Board.
Gu, F., Chan, M. Y. & Fryk, R., 2011. Geomechanical-Data Acquisition, Monitoring and
Applications in SAGD. Journal of Canadian Petroleum Technology, 50(6), pp. 9 - 21.
Holland, M., 2015. Directionless Fracture Stability Assessment based on a 4-D Geomechanical
Model of a SAGD scenario. First Break, 33(3), pp. 47-52.
Holland, M. et al., 2010. Value of 3D Geomechanical Modeling in Field Development - A new
Approach using Geostatistics. Al-Khobar, Society of Petroleum Engineers.
Jung, F., 1946/1947. Der Culmannsche und der Mohrsche Kreis. Österreichisches Ingenieur-
Archiv, 1(4-5), pp. 408-410.
van der Zee, W., Ozan, C., Brudy, M. & Holland, M., 2011. 3D Geomechanical Modeling of
Complex Salt Structures. Barcelona, SIMULIA Customer Conference.
van der Zee, W., Plug, W.-J., Niewland, D. & Holland, M., 2014. Geomechanical Optimization of
Underground Gas Storage Operation. Providence, SIMULIA Community Conference.
Zoback, M. D., 2007. Reservoir Geomechanics. Cambridge: Cambridge University Press.

8. Appendix
The SAGD scenario used in this publication is synthetic. Although closely aligned to a realistic
scenario, the structural model, the positioning of the reservoir grids, the external stress field as
well as the geomechanical properties are fictional.
Table 1 – Material properties used in the analysis per element set.

Element sets Top horizon Density (kg/m3) Young’s Poisson’s Thermal
modulus ratio expansion
(GPa) coefficient (K-1)
Top Topography 2,200 3.0 0.3 -
Layer 1 Horizon 1 2,250 1.5 0.25 3·10-5
Layer 2 Horizon 2 2,270 2.7 0.27 3·10-5
Layer 3 Horizon 3 2,310 1.2 0.26 3·10-5
Layer 4 Horizon 4 2,350 1.0 0.27 3·10-5
Base Horizon 5 2,390 1.5 0.3 3·10-5
Table 2 – Parameters for initial stress field.
Property Value
Effective Stress Ratio Shmin at surface 0.7
Effective Stress Ratio Shmin at 280 m TVD 0.5
Effective Stress Ratio SHmax at surface 0.9
Effective Stress Ratio SHmax at 280 m TVD 0.7
Azimuth of SHmax 35 DEG

Finite Element Analysis for Post-Cracking Design of
Thermal Well Cements
Jueren Xie
C-FER Technologies, Canada
Abstract: Steam Assisted Gravity Drainage (SAGD) and Cyclic Steam Stimulation (CSS) are
typical thermal well technologies used for the recovery of heavy oil and oil sands resources.
Cyclic thermal loading, resulting from temperature variation with the peak operating
temperatures in the range of 220°C to 350°C, often subjects the well cement sheath to significant
longitudinal and circumferential tensile or compressive stresses, leading to potential failure
modes, including cracking, crushing and de-bonding. Failure of cement sheath would compromise
the well and caprock integrities. Therefore, ensuring satisfactory performance of cement sheath
over the well life has been one of the key challenges in thermal well designs.
Traditional well design uses the stress-based design concept which limits the longitudinal and
circumferential cement stresses to the tensile strength value. The stress-based design has been
proven to be overly conservative and often impossible to meet for thermal well cements. For
developing a more suitable design criterion, advanced finite element analysis (FEA) modeling
techniques are required to capture the complex cement behavior in thermal wells.
This paper presents an FEA approach by using a damaged plasticity model to capture the complex
thermal and mechanical behavior of cement material. The damaged plasticity model considers the
thermal conductivity, tri-axial stress effect, strain-hardening due to plasticity, strain-softening due
to cracking and cyclic loading responses. This paper also presents a post-cracking design concept
as a new design basis to allow limited cement cracking, providing that the permeability of the
cement remains within an acceptable limit. An illustrative example is presented to demonstrate
the use of FEA approach and post-cracking design concept for the design of thermal well cements.
Keywords: Casing, Cement, Compressive Strength, Cooling, Cooling Rate, Damage, Cracking,
Finite Element Analysis (FEA), Heating, Heating Rate, Monte Carlo Simulation, Plasticity,
Tensile Strength, Thermal Cement, Thermal Cycle, Thermal Well, Young’s modulus
1. Introduction
Operation of thermal wells presents cyclic thermal loading resulting from temperature changes,
with the peak steam temperature in the ranges from 220°C to 275°C for SAGD wells and from
330°C to 350°C for CSS wells. When the steam is first injected into the wells, steep thermal
gradients across the cement sheath may develop due to the large temperature difference between
the steam and the surrounding formations. Casing material, with the high thermal conductivity and
a high coefficient of thermal expansion of steel, tends to expand quickly, subjecting the relatively
cold cement sheath to tensile circumferential stresses. As the cement material typically has a low

tensile strength, the thermal expansion of casing during heating can potentially cause the cement
sheath to crack. In the cooling phase, as a result of well shut-in, the difference in thermal
contraction of casing, cement and formation material may also cause loss of the cement bond to
the casing string. Cement cracking and de-bonding may allow for fluid migration along the well,
which may compromise the hydraulic isolation role of the cement sheath. This is often of special
concern in the caprock region, as axial cracking through this zone could allow steam and
hydrocarbons to migrate to overlying formations, or even to surface. Cement cracking and
de-bonding also present concerns for potential Stress Corrosion Cracking (SCC) for casing string.
To consider the impact of thermal load, thermal wells require the use of thermal cement, which
according to the Alberta Energy Regulator (AER) is a blend which does not exhibit a significant
reduction in strength when subjected to temperatures greater than 360°C (AER, 1990).
Traditionally, a stress-based design concept is used in designing well cement. This approach limits
the tensile stress in the cement sheath to the tensile strength of the material to prevent the
occurrence of cement cracking. To determine the stress conditions in the cement sheath, several
FEA studies have been carried out by various investigators (e.g. Thiercelin, 1997; Bosma, 1999;
Ravi, 2008). Some early FEA studies used thermo-elastic material and one- or two-dimensional
structural models. Use of such models may disregard the tri-axial stress conditions and Poisson’s
ratio effects in the cement. Since stress-based design has been typically used, little effort has made
in modeling the non-linear material behavior, such as yielding of the casing and yielding, crushing
and cracking of the cement sheath. The Mohr-Coulomb material model used by some may
simulate cement failure by the application of a shear failure criterion; however, this approach
appears to overlook the effects of cement cracking and crack propagation.
Xie et al. (2011) used a smeared cracking model implemented by Abaqus to investigate the impact
of heating rate on the cracking potential of the cement sheath during the heating phase for SAGD
wells. The smeared cracking constitutive model simulated the cracking and crushing behavior of
the cement material under monotonic loading (e.g. heating operation). One of the shortcomings of
this study was that the analyses were limited to the heating phase of thermal operations and the
cement response to cyclic thermal loading was not analyzed.
This paper presents a review of cement mechanical properties and proposes a new design basis:
post-cracking design concept. To facilitate the use of the post-cracking design approach, a
damaged plasticity model is recommended to capture the thermal-mechanical behavior of cement
in thermal wells. The damaged plasticity model uses the plasticity theory to consider the tri-axial
stress effect, strain-hardening due to plasticity, strain-softening due to cracking and cyclic loading.
An analysis example is presented in the paper to show the impact of heating and cooling loading
on cement performance.
2. CEMENT MECHANICAL PROPERTIES

Comprehensive FEA models often require the use of the stress-strain material responses under
uniaxial tensile and compressive loading. Schematic representations of such responses are shown

in Figure 1. Mechanical properties for cement materials include elastic modulus, Poisson’s ratio,
tensile strength, compressive strength, fracture energy and thermal expansion coefficient
(MacGregor, 1992; Xie et al., 1995). These properties often vary with temperature. The key
thermal properties include the thermal conductivity and specific heat.
(a) Uniaxial Tension (b) Uniaxial Compression
Figure 1 Schematic Representation of Stress-Strain Relationships in Cement

under Uniaxial Tension and Uniaxial Compression
2.1 Key Mechanical Parameters
Table 1 presents a list of key mechanical parameters such as elastic modulus, Poisson’s ratio,
compressive strength and tensile strength.
Dean et al. (2002) showed a type of API Class G cement with Young’s modulus of 3.1 GPa and
compressive strength of 6.9 MPa.
Stiles (2006) studied the mechanical properties of five cement formulations after exposure to a
temperature of 340°C. The five formulations were: 1) conventional Class G; 2) thixotropic system
(Class A); 3) foamed Class G; 4) high strength and low density system; and 5) flexible and
expanding low density system. Stiles determined that the mechanical properties change
significantly after 24 months of curing at temperature and the degree of changes vary among
cement formulations. The key mechanical properties determined by Stiles, corresponding to the
five formations were: elastic modulus of 7.3, 2.3, 4.0, 6.5 and 4.2 GPa before curing; 12.1, 6.8,
6.1, 1.4 and 1.4 GPa after 24 months curing; effective compressive strengths of 32.3, 15.5, 23.1,
48.9 and 37.2 MPa before curing; ;56.2, 40.7, 28.3, 11.9 and 13.8 MPa after curing; tensile
strengths of 2.6, 2.3, 2.6, 3.2 and 2.2 MPa before curing; and 7.3, 6.4, 4.5, 1.6 and 1.5 MPa after
curing.
DeBruijn et al. (2009) performed a study of SAGD well integrity using three cement formulations:
flexible cement, thixotropic cement and conventional thermal cement. The key mechanical
properties considered which correspond to these three cements were: elastic modulus values of

2.1, 5.5, 8.5 GPa; compressive strengths of 9, 28.8 and 48 MPa; and tensile strengths of 1.2, 4.4
and 4.8 MPa.
Table 1. Description of Key Mechanical Parameters for Cement.
Parameter Description
Elastic modulus (Ec) It is a measure of stiffness during the initial elastic deformation under
tensile or compressive loading. The elastic modulus in tension is often
assumed to be the same under compression.
Poisson’s ratio (ν) It is the ratio of transverse to axial strain
Compressive strength (fc’) It refers to the uniaxial compressive strength as measured by a
compressive test of a standard test cylinder of cement.
The effect of confinement on cement compressive stress-strain
response should also be considered. Generally, the radial confining
stress increases the compressive strength and post-peak stress level.
Tri-axial stress tests can be performed using a tri-axial Hoek cell (Xie et
al., 1995).
Tensile strength, (ft’,) It is taken to refer to the cement strength under uniaxial tensile loading.
It is, however, often determined through a flexural test of a plain cement
beam or a split cylinder test, rather than the direct uniaxial tension test.
2.2 Cement Cracking and Fracture Energy

Fracture mechanics based FEA was first presented by Hillerborg et al. (1976) for concrete and
cement materials. The key assumption was that the stresses act across a crack as long as it is
narrowly opened. The mechanical response during the narrowly opened cracking phase is called
strain-softening. In order to capture the cracking and strain-softening behavior,
Hillerborg et al. (1976) proposed the use of fracture energy value in the FEA model.
There are two alternative test approaches to quantify the strain softening and fracture energy of
cement in tension: the direct tension test and the flexural test on notched beams (Xie et al., 1995).
Both types of tests must be conducted with the testing machine in stroke control to capture the
descending part of the load-deflection response. Figure 2 presents schematics of tensile stress
versus cracking-displacement relationship. The dashed curve shows the typical variation from
coupon tests. In FEA modeling, such a relationship is often simplified by a linear curve, shown in
Figure 2 as the solid line. The area underneath the tensile stress-displacement curve is called the
fracture energy,(Gf) as defined by Hillerborg et al. (1976).
Macharia et al. (2011) determined that the fracture energy values of Portland blast furnace cement
were in the range of 0.138 to 0.211 N/mm for dry samples and 0.117 to 0.150 N/mm for sealed
samples. Padevet and Zobal (2011) found significantly higher values of fracture energy for cement
paste with fly ash.

Figure 2 Schematic Representation of Stress vs. Cracking Displacement
Response and Fracture Energy of Cement
2.3 Creep
Creep is the permanent, time-dependent deformation to relieve stresses within the cement material.
Cement that is subjected to sustained forces over long-durations is prone to creep. Creep may
reduce the effects of cracking that occurs in cement; however, this effect requires further
validation through testing.
3. POST-CRACKING DESIGN CONCEPT

Traditionally, the stress-based design approach was used which limited the tensile stress in the
cement sheath to the tensile strength of the material to prevent the occurrence of cement cracking.
However, the tensile strength limit requirement appears to be stinging, as the typical heating
operation inevitably imposes initial cracking in the cement sheath in thermal wells (Xie et
al, 2011). In an effort to develop a more reasonable design criterion, a post-cracking design
concept was recently proposed (Xie, 2014a, 2014b and 2015). The post-cracking design concept
was based on the fracture mechanics by Hillerborg et al. (1976) who considered that the stresses
act across a crack for cement type materials, as long as the crack opening is narrow. Therefore,
one can postulate that the permeability of cement materials may be affected by the initial cracking;
however, this effect would not be significant until the cracks open widely. The post-cracking
design concept allows the use of cement beyond the tensile strength limit, as long as the crack
width (i.e. crack displacement) in the cement is limited.
The post-cracking design criterion can be described as:

ucr ≤ (ucr)allowable

where ucr is the crack displacement and (ucr allowable) is the allowable crack displacement. The
crack displacement can be determined from FEA, while the allowable crack displacement should
be determined from physical tests.
The crack displacement threshold can be defined by controlling the permeability to an acceptable
level, while maintaining sufficient residual strength of the material for potential further loading.
The permeability of cracked cement and concrete materials has been studied by several
investigators (Wang, Jansen and Shah, 1997, Picandet, Khelidj and Bellegou, 2009, Akhavan,
Shafaatian and Ajabipour, 2013, Walsh, et al., 2012; Abdoulghafour, et al., 2013). Figure 3 shows
the schematic representation of the relationship between permeability (in logarithm scale) and
crack width, based on the review of published work on cement and concrete materials.
The work, completed by Wang, Jansen and Shah (1997) suggested that for a cracked concrete
sample, after unloading, the water permeability coefficient increased from 1×10-9 cm/s to
1×10-6 cm/s (as the crack opening displacement increased from 0 to 50 microns); however, the
water permeability coefficient increased from 1×10-6 cm/s to 1×10-3 cm/s as the crack opening
displacement increased from 50 to 200 microns. The key finding of the study was that the concrete
crack opening had little effect on permeability if the cracking opening displacement was less than
50 microns.
Figure 3 Schematic Representation of Permeability and Crack Width Relationship

In the light of the work completed by Wang, Jansen and Shah (1997), this paper uses 25.4 microns
as the crack width limit for the post-cracking design for cements, allowing for the design of
cements beyond tensile strength limit while controlling the permeability to a reasonable low level.
The threshold is expressed as:
(ucr)allowable = 25.4 Microns
Macharia et al. (2011) determined that the fracture energy values of Portland blast furnace cement
were in the range of 0.138 to 0.211 N/mm for dry samples, and 0.117 to 0.150 N/mm for sealed

samples. The noted fracture energy values suggest that the ultimate crack opening displacement
could be significantly larger than 25.4 microns. Therefore, using 25.4 microns as a design limit
would retain sufficient residual strength to sustain further monotonic or cyclic loading on the
cement material.
It is important to point out that the post-cracking design concept acknowledges that the limited
cement cracking displacement is acceptable, providing the design forces are
displacement-controlled and bounded. Since the well thermal strains and formation movements are
controlled and bound by the magnitude of the well temperature and production conditions, it is
suitable to use of post-cracking design concept for thermal well cements.
4. CEMENT CONSTITUTIVE MODELS

Use of the post-cracking design concept requires an advanced FEA model which is capable of
capturing the three-dimensional stress/strain conditions, initial elastic response described by
Young’s modulus and Poisson’s ratio, thermal expansion of the material, non-linear stress-strain
relationships under compressive loading, cracking and crack propagation under a tensile stress
condition and cyclic loading responses.
The fracture mechanics based FEA approaches proposed by Hillerborg et al. (1976) have been
implemented in some FEA programs. For example, the commercial FEA program
ABAQUS (2013) has implemented the smeared cracking model and damaged plasticity model,
based on Hillerborg’s theory. As mentioned earlier, Xie et al. (2011) used a smeared cracking
model to investigate the impact of heating rate on the cracking potential of the cement sheath
during the heating phase for SAGD wells. The smeared cracking model consists of an
isotropically-hardened yield surface that is active when the stress is dominantly compressive and
an independent “crack detection surface” that determines if a point fails by cracking. It is designed
for applications in which the cement or concrete is subjected to essentially monotonic straining at
low confining pressures.
Furthering the smeared cracking model, a damaged plasticity model is recommended in order to
better capture the thermal-mechanical behavior of cement in thermal wells, especially under cyclic
thermal-mechanical loading (Xie 2014a). The damaged plasticity model uses concepts of isotropic
damage in the elastic behavior in combination with isotropic tensile and compressive plasticity to
represent the inelastic behavior of cement. This plasticity-based damaged model assumes that the
main two failure mechanisms are tensile cracking and compressive crushing. It consists of the
combination of non-associated multi-hardening plasticity and scalar damage in the elastic behavior
to describe the irreversible damage that occurs during the fracturing process. The model is
designed for applications in which cement or concrete is subjected to monotonic, cyclic and/or
dynamic loading under low confinement. A similar damaged plasticity model for concrete was
developed by Xie et al. (1996), with the consideration of high confining stress effects.
As shown in Figure 1a), under uniaxial tension, the stress-strain response follows a linear elastic
relationship until the value of the failure stress(𝜎𝑡0 ) is reached. The failure stress corresponds to
the onset of micro-cracking in the concrete material. Beyond the failure stress, the formation of

micro-cracks is represented with a softening stress-strain response. Under uniaxial compression, as
shown in Figure 1b), the response is linear until the value of initial yield. In the plastic regime, the
response is typically characterized by stress hardening, followed by strain softening (beyond the
ultimate stress (𝜎𝑐0 )).This representation, although somewhat simplified, captures the main
features of the response of cement.
When the cement material is unloaded from any point on the strain softening branch of the
stress-strain curves, the unloading response is weakened, represented by the degradation of the
elastic modulus. Two damage variables (dt and dc) are used to describe the degradation of the
elastic modulus in tension and compression. These two damage variables are assumed to be
functions of the plastic strains’ temperature and field variables. For the cyclic loading analysis, the
stiffness recovery effect associated with stress reversal (i.e. crack closing) is considered by
introducing stiffness recovery factors.
The same concepts for the damage variables and stiffness recovery factors are used for the
multi-axial mechanics modeling.
5. ANALYSIS EXAMPLE
An illustrative scenario for a SAGD well was analyzed using the damaged plasticity model for a
cement sheath. The case was analyzed to study the impact of steam heating rate on cement
cracking and the impact of cooling phase on potential de-bonding.
5.1 Structural and Material Models

Figure 4 shows the FEA model, using axisymmetric solid elements. The model included
intermediate and surface casing strings, cement sheaths and surrounding formations.
To accurately capture the temperature distribution field around the well, the model used a mesh
density of one element per meter along the axial direction. In the radial direction, ten elements
were used for each of the casing and cement layers and more than 30 elements of increasing radial
thickness for the formations.
The example considers a well with an overburden thickness of 360 m and caprock thickness of
10 m. The overburden is assumed to have properties that change linearly with depth from surface
to the top of the caprock with the elastic modulus of 100 MPa to 1.35 GPa, thermal expansion
coefficient of 2.0×10-6/°C to 1.6×10-6/°C, initial temperature of 4°C to 9°C and thermal
conductivity of 2.0 W/m·K to 1.7 W/m·K.
The caprock has an elastic modulus of 1.19 GPa, thermal expansion coefficient of 5.0×10-6/°C,
initial temperature of 10°C and thermal conductivity of 1.7 W/m·K.
The injection well is completed with a surface casing of 406.4 mm (16 in), 96.7 kg/m (65 lb/ft)
H40 casing material landed at 150 m MD in a 508.0 mm (20 in) hole; an intermediate casing of
298.5 mm (11.75 in), 69.9 kg/m (47.0 lb/ft) L80 casing material landed at 750 m MD in a

374.5 mm (14.75 in) hole and a horizontal liner of 219.1 mm (8.625 in), 47.6 kg/m (32 lb/ft)
820 m in length with the liner hanger at 740 m MD in a 269.9 mm (10.625 in) hole. The steel
materials are assumed to have an elastic modulus of 200 GPa, thermal expansion coefficient of
12×10-6/°C and thermal conductivity of 45 W/m·K.
The surface and intermediate cement sheaths are modeled using the damaged plasticity model. The
cement is assumed to have the elastic modulus of 8.5 GPa, compressive strength of 48 MPa,
tensile strength of 4.8 MPa, thermal expansion coefficient of 9.0×10-6/°C and thermal conductivity
of 1.0 W/m·K. In defining the damaged plasticity model, the fracture energy is assumed to be
0.15 N/mm. Assuming a linear tensile stiffening curve, the critical crack displacement at zero
tensile stress is determined to be 0.0625 mm. The damage variables are assumed to be 0.9 at the
critical cracking displacement for tension and at the critical crushing strain for compression. The
stiffness recovery ratios are assumed to be 0.2.
Formation
Cement
Intermediate Casing
Surface Casing
Figure 4 FEA Model showing Various Formation Layers (left) and Wellbore (right).
5.2 Modeling of Thermal Loading in Heating Phase

The analysis focused on the impact of the heating rate during the heating phase. Seven heating
rates from 0.15°C/min to 5°C/min were analyzed. The peak steam temperature was assumed to be
220°C and the initial temperature was 5°C. The relationships between the casing ID temperature
and time are shown in Figure 5.

Figure 5 Various Heating Rates Considered in Analysis
The heating phase was analyzed in two steps. First, an equilibrium stress state condition was
applied using geostatic analysis to consider the effect of the weight of the overburden and caprock.
Secondly, a coupled thermal-displacement analysis was performed to determine, as a function of
time, the temperature distributions around the wellbore and the stresses and deformations in the
casing strings and cement sheaths that result from material thermal expansion and formation
confinement.
Figure 6 presents the contour plots of temperature and tensile damage variable in the casing,
cement and formation. This is shown in the caprock interval after 5 hours, 2 days and 1 year of
heating for the well subjected to the base case heating rate (0.6°C/min.). The intermediate casing
reaches a relatively uniform temperature along the full length of the wellbore; however, a
significant temperature gradient from the casing to the outer surface of the cement sheath and out
from the wellbore into the formation was found even after one year of heating.
The cement cracking is caused by the various thermal expansion amounts of the casing, cement
and formation materials. The immediately-heated intermediate casing would expand to cause high
circumferential stress in cement sheath. As shown in Figure 6, the temperature decreases through
the cement thickness. As a result, the cement portion adjacent to the casing would expand further
and the initial cracking would develop in the cement portion adjacent to the formation.
Figure 7 presents the history of the crack displacement over the caprock interval for the seven
heating rates from 0.15°C/min to 5°C/min studied. It shows that cracks develop in the cement in
the early stages of heating when the temperature gradient is high. Heating for several days tends to
warm up more cement and formation volumes and slows down the further cracking. The
magnitude of the cement crack displacement increases with quicker heating rates.

After five days of heating, the crack displacement values are 8.5, 13, 19.5, 25.1, 30.5, 38 and 42.5
microns for the heating rates of 0.15°C/min, 0.3°C/min, 0.6°C/min, 1.0°C/min, 1.5°C/min,
3.0°C/min and 5.0°C/min, respectively. Assuming that a threshold value of crack displacement is
set as 25.4 microns, the highest allowable heating rate would be approximately 1.0°C/min.
Cement
5 Hours
Overburden
Heating
Well Center
Caprock
Casing
2 Days
Heating
1 Year
Heating
Figure 6 Temperature (°C) Profile (Left) and Tensile Damage Variable (Right) for
the Base Case during Heating
Figure 7 Maximum Cement Crack Displacement

5.3 Modeling of Cooling
For the cooling phase, a steady state analysis was performed to determine the impact of heat
dissipation to the surrounding formations on the cement mechanical behavior. Since the steady
state analysis was performed, cooling time was not specifically defined. Instead, analysis results
are presented in terms of casing ID temperatures. Figure 8 presents the temperature distribution
and radial stress in the cement sheath corresponding to the casing ID temperatures (over caprock
interval) of 220°C, 109°C and 38°C during the cooling phase. The non-uniform temperature
distribution in the casing and cement causes a decrease in the compressive radial stress in the
cement sheath next to the casing OD. The radial compressive stresses in the cement next to the
casing are 5.35, 3.59 and 2.37 MPa, corresponding to the casing temperatures of 220°C, 109°C
and 38°C. If the radial stress in the cement changes from compression to tension, de-bonding
between the cement and casing occurs. The results suggest that the cooling phase increases the
potential for de-bonding between the casing and cement; however, in this case, the de-bonding
does not occur even at 38°C casing ID temperature during cooling.
Cement
Wellbore
Overburden
Temp = 220 °C
Well Center
Caprock
Casing
Wellbore
Temp = 109°C
Wellbore
Temp = 38°C
Figure 8 Temperature (°C) Profile (Left) and Radial Stress (MPa, Right) for the Base
Case during Cooling
6. CONCLUSIONS
This paper presents a review of cement mechanical properties and proposes a new design basis for
thermal well cements: post-cracking design concept. A damaged plasticity model is recommended
for analyzing the thermal-mechanical behavior of cement in thermal wells. The proposed analysis
and design approach is demonstrated by analyzing an illustrative scenario of a SAGD well.
The following conclusions are made:

• Post-cracking design concept is proposed as a new design basis for cement in thermal
wells. The post-cracking design allows for limited cement cracking, providing that the
permeability of the cement does not increase beyond acceptable values. A threshold of
25.4 micron is recommended as the crack displacement limit for cement design.
• The damaged plasticity model considers the tri-axial stress effect, strain-hardening due to
plasticity, strain-softening due to cracking and cyclic loading. It is a robust numerical tool
for analyzing cement behavior in thermal wells.
• An illustrative scenario case was analyzed under the thermal mechanical loading
resulting from the heating and cooling operations of a SAGD well. The analysis
demonstrated that quicker heating rates tend to increase the potential of cement cracking.
For the cooling phase, the potential for de-bonding between the casing and the cement
increases as the wellbore temperature decreases with cooling time.
7. NOMENCLATURE
dt = Degradation of the elastic modulus in tension
dc = Degradation of the elastic modulus in compression
Ec = Elastic modulus
ft’ = Tensile strength
fc’ = Compressive strength
Gf = Fracture energy
ucr = Crack displacement
ν = Poisson’s ratio
8. REFERENCES
1. ABAQUS. (2013). User’s Manual. (Version 6.13).
2. Abdoulghafour, H., Luquot, L. and Gouze P. (2013). Characterization of the Mechanisms
Controlling the Permeability Changes of Fractured Cements Flowed Through by CO2-Rich
Brine. Environment Science & Technology, 47 (18), 10332-10338.
3. Akhavan, A., Shafaatian, S.M.H., Ajabipour, F. (2012), Quantifying the effects of crack
width, tortuosity, and roughness on water permeability of cracked mortars, Cement and
Concrete Research, 42, 313-320.
4. Alberta Energy Regulator. (1990, July). Directive 009: Casing Cementing Minimum
Requirements.
5. Bosma, m., Ravi, K., Van Driel, W. and Schreppers, G.J. (1999, Oct. 3-6). Design approach to
Sealant Selection for the Life of the Well. SPE Annual Technical Conference and Exhibition,
Houston, TX, USA. SPE 56536.
6. Dean, G. D., Torres, R. S. (2002). Novel Cement System for Improved Zonal Isolation in
Steam Injection Wells. SPE/Petroleum Society of CIM/CHOA 78995.

7. Debruijin, G., Garnier, A., Brignoli, R., Bexte, d., and Reinheimer, d. (2009, March 17-19).
Flexible Cement Improves Wellbore Integrity in SAGD Wells. SPE/IADC Drilling
Conference and Exhibition, Amsterdam, The Netherlands, SPE/IADC 119960.
8. Hillerborg, A., Modeer, M. and Peterson, P-E. (1976). Analysis of Crack Formation and
Crack Growth in Concrete by Means of Fracture Mechanics and Finite Elements. Cement and
Concrete Research, Vol. 6, 773-781.
9. Macharia, m. m., Sato, R., Shigematsu, A. and Onishi, H. (2011, Aug. 14-16). Study of
Fracture Energy Properties of Portland Blast Furnace Cement Type-B Concrete with Partial
Replacement of Aggregate Porous Ceramic Course Aggregate. 36th Conference on Our
World in Concrete & Structures, Singapore.
10. MacGregor, J. G. (1992). Reinforced Concrete – Mechanics & Design. (2nd ed.). Prentice
Hall.
11. Padevet, P. and Zobal, O. (2011, Sept. 20-22). Fracture Energy of Cement Paste with
Addition of the Fly Ash. Modeling of Mechanical and Mechatronic Systems, 4th International
Conference, Herlany, Slovak Republic.
12. Ravi, K., Hunter, B. and Kulakofsky, d. (2008, March 10-12). Job Design and Cement System
to Contain Steam and Improve Heavy Oil Recovery. World Heavy Oil Congress 2008,
Edmonton, AB, Canada. WHOC 2008-381.
13. Stiles, D. (2006, Feb. 21-23). Effects of Long-Term Exposure to Ultrahigh Temperature on
the Mechanical Parameters of Cement. Miami, Florida. IADC/SPE 98896.
14. Thiercelin, M.J., Darguad, B., Baret, J.F. and Rodriguez, W.J. (1997, Oct. 5-8). Cement
Design Based on Cement Mechanical Response. SPE Annual Technical Conference and
Exhibition, San Antonio, TX, USA. SPE 38598.
15. Walsh, S. d. C, Du Frane, W. L., Mason, H. E. and Carroll, S. A. (2013). Permeability of
Wellbore-Cement Fractures Following Degradation by Carbonated Brine. Rock Mech Rock
Eng 46, 455-464.
16. Wang, K., Jansen, D.C., Shah, S.P. (1997), Permeability study of cracked concrete, Cement
and Concrete Research, 27, 381-393.
17. Xie, J., Elwi, A.E. and MacGregor, J.G. (1995). The Mechanical Properties of Three High
Strength Concretes Containing Silica Fume. ACI Material Journal, Vol. 92, No. 2,
March-April, 135-145.
18. Xie, J. Elwi, A.E. and MacGregor, J.G. (1996). Numerical Investigation of Eccentrically
Loaded High-Strength Concrete Tied Columns. ACI Structural Journal, July-August, 449
19. Xie, J. and Zahacy, T. (2011). Understanding Cement Mechanical Behavior in SAGD Wells.
World Heavy Oil Congress, Edmonton, Alberta. WHOC11-557.
20. Xie, J., (2014a), Analytical Modeling of Cement Mechanical Behavior in Thermal Wells.
World Heavy Oil Congress, New Orleans, USA. March 5-7. WHOC14-104.
21. Xie, J. (2014b), Post-Cracking Design for Thermal Well Cements. SPE Thermal Well Design
and Integrity Workshop. Banff, Alberta, Canada. November 18-20.
22. Xie, J. (2015), Reliability-Based Design and Assessment (RBDA) with Post-Cracking Design
Basis for Thermal Well Cements. World Heavy Oil Congress, Edmonton, Canada.
March 24-26. WHOC15-108.

9. ACKNOWLEDGMENTS
The funding for preparing this paper was provided by C-FER Technologies. The author would like
to acknowledge Dr. Feng Rong, Research Engineer at C-FER Technologies for assistance in the
numerical modeling.

Use of FEA and Radial Basis Functions for
Reliability-based Design and Assessment of
Tubular Connection Sealability
Jueren Xie, Thomas Dessein, Feng Rong
C-FER Technologies (1999) Inc., Canada
Abstract: One of the significant challenges in oil wells is to maintain adequate structural and
sealing capacities of casing and tubing connections. In particular, unconventional wells, such as
thermal and HPHT wells, require elevated standards to evaluate structural integrity and
sealability of tubular connections. This paper presents a Reliability Based Design & Assessment
(RBDA) methodology that characterizes system safety using a quantitative estimate of reliability.
In the RBDA approach, Finite Element Analysis (FEA) using Abaqus is employed to determine the
connection sealing capacities as functions of several design parameters, such as connection
geometry, material mechanical properties, and make-up and operational conditions. Based on the
FEA results to characterize the effect of design parameters and variations, an implicit limit state
function can be established using Radial Basis Functions (RBF). This paper also presents an FEA
example of a generic premium connections to demonstrate the use of the proposed methodology.
Keywords: Casing; Connection; FEA; HPHT; Leakage Resistance; Radial-Based Function,

RBDA; RBF, Reliability-Based Design Assessment; Sealability; Structural Integrity; Tubing.
1. Introduction
Oil wells are completed with casing and tubing strings consisting of many tubular joints (usually
12 to 14 m long) joined together by threaded connections. Tubular connections are one of the
most critical components of an oil well in terms of risk of wellbore failure (Xie 2006). Payne and
Schwind (1999) noted that, based on industry estimates, connection failures account for 85% to
95% of all oilfield tubular failures.
In recent decades, unconventional wells such as thermal and High Pressure and High Temperature
(HPHT) wells pose more significant challenges to well designs. Thermal wells typically operate
under cyclic high temperature in the range of 220°C to 350°C. For HPHT wells, high pressure
(greater than 69 MPa) and high temperature (greater than 150°C) conditions are present. The
tubular designs for unconventional well will require elevated standards to account for the plasticity
load conditions resulting from high pressure, high temperature and formation movement.
Unconventional wells typically use premium connections, which generally show superior
structural and sealing capacity over API round and buttress connections. Figure 1 shows a generic
premium connection which uses a buttress-type thread form to meet the structural requirements of
the full pipe body strength capacity, and incorporates a metal-to-metal radial seal for enhanced
sealability. Many premium connections also include a torque shoulder next to the seal region to
control make-up torques and provide additional sealability.

Figure 1. Schematic of a premium connection.
To date, deterministic design approaches have been used to ensure the structural integrity and
serviceability of tubular connections. In order to evaluate the sealability of connections, the
deterministic approach applies a leakage resistance criterion to the Finite Element Analysis (FEA)
results for the connection design with the worst geometry, minimum yield strength and worst
combination of loads. However, the deterministic design approach appears to be overly
conservative as it does not take into account the probabilities and uncertainties related to the
variability in tubular connection geometry, material properties (e.g. yield and tensile strengths),
and operational loads. To balance operational safety and cost, Reliability-Based Design and
Assessment (RBDA) is recommended for use to optimize the connection design and minimize the
risk of failure due to connection leakage (Xie et.al. 2012, 2014).
In RBDA, the probability of failure, Pf, is predicted by explicitly addressing the uncertainty in
capacity (e.g. resistance) and demand (e.g. load, deformation) through probability distributions. Pf
depends on the degree of overlap between these two distributions (Figure 2) ( i.e. cases where
demand exceeds capacity).
Connection sealability in unconventional wells, such as thermal, HPHT wells, is often associated
with various uncertainties related to the variability in casing/connection geometry (e.g. thread and
seal interferences and tapers), material properties (e.g. yield and tensile strength), cement
completeness, formation properties, and operational loads (e.g. temperature and pressure). Due to
the non-linear mechanical and material responses, it is virtually impossible to establish analytical
limit state functions that represent the structural and serviceability requirements for tubular
connections in HPHT and thermal wells. As such, Finite Element Analyses (FEA) are often used
to assist the establishment of numerical limit state functions.
This paper presents a RBDA methodology that characterizes system safety using a quantitative
estimate of reliability. In the RBDA approach, Finite Element Analysis (FEA) using Abaqus is
employed to determine the connection sealing capacities as functions of several design parameters,
such as connection geometry, material mechanical properties, and make-up and operational
conditions. Based on the FEA results of design parameters and variations, an implicit limit state
function can be established using Radial Basis Functions (RBF). This paper also presents an FEA
example for a generic premium connection to demonstrate the use of the proposed methodology.

Probability Density
Demand Capacity
(Load) (Resistance)
Figure 2. Illustration of capacity and demand interference.
2. FEA Model for Connection Sealability

FEA model of the connection should be developed to assess the following conditions:
• Make-up;
• Internal and external pressures;
• Axial tensile and compressive loads; and
• Curvature loading such as bending and shear.
This paper presents the connection model considerations taken into account when using the
commercial finite element analysis software, Abaqus v6.14 (2014).
Figure 3 shows the FEA mesh for a generic premium connection and an adjoining section of pipe
body. Note that due to the symmetry of deformation about the connection center under the
imposed loading (except for in the case of shear deformation which is not discussed here), only
half of the coupling is modeled.
At the coupling centre plane, symmetric boundary conditions are assumed. For axisymmetric
loading analyses, such as make-up, axial forces and pressure loading, the connection is modeled
using axisymmetric solid elements. However, for the curvature loading analysis, such as bending
and shear, the connection is modeled using axisymmetric solid elements with non-linear,
asymmetric deformation. As noted in the Abaqus documentation (2014), these elements are
intended for the non-linear analysis of structures which are initially axisymmetric but may
undergo non-linear, non-axisymmetric deformation, which is the case for the connection and load
scenarios examined here. Contact between the pin and coupling elements within the connection is
modeled using slidelines.

Figure 3. FEA mesh for a generic premium connection.
The tubular connection material response should be modeled using an elastic-plastic, non-linear
kinematic hardening constitutive relationship. A comprehensive material model should
incorporate both temperature-dependent and time-dependent characteristics of the casing material.
These modeling considerations are needed to properly capture material behavior variation with
temperature, stress relaxation and Bauschinger effects (i.e. a reduced yield stress upon load
reversal after the plastic deformation that occurs during initial loading).
Tubular connections should be modeled with sufficient mesh density, especially in critical areas
such as the thread root, seal and shoulder regions, to ensure accuracy and convergence of analysis
results. Over the seal region, a mesh density of 0.1 mm/element is recommended along the seal
direction, to ensure an accurate prediction of seal contact intensity. To accurately predict the
structural behavior of the connection threads under the various loading conditions considered, a
more refined mesh density (e.g. 0.04 mm/element) is suggested for the areas next to the thread
roots where relatively high stress/strain concentrations are expected to be present.
Connection make-up is modeled through the interference fit in the thread, seal and shoulder. The
interference amount in the threads and the seal is typically defined by the connection drawing.
The shoulder interference is defined, in this paper, in correlation with the make-up torque values.
For modeling the load conditions, the axial tensile/compressive forces are modeled by applying
global forces over the pin end section of the model; the external pressure is modeled by applying
the pressure loading on the OD surface of the connection, on the pin and coupling surfaces over
the thread region, and on the partial sealing region up to the anticipated maximum contact stress
point on the seal; and the internal pressure is modeled by applying the pressure loading on the ID
surface of the pin and coupling surface and on the torque-shoulder region.
Details of the FEA methodology for assessing tubular connections can be found in several
publications (Xie 2007; Xie 2011; Xie et al. 2011).
3. Limit State for Connection Sealability

A limit state function represents the relationship between capacity and demand parameters, as well
as differentiates between safe and failure domains:

𝑔(𝐶, 𝐷) = 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦 − 𝐷𝑒𝑚𝑎𝑛𝑑 (1)
The sealing capacity of a premium connection is determined by the magnitude and distribution of
the metal-to-metal contact stress that exists over the seal region under each load condition. As a
necessary requirement, the seal contact stress must be higher than the differential pressure across
the inner and outer surfaces of the casing in order to maintain sealability. Figure 4 presents an
illustration of seal contact stress in a premium connection. The following is the limit state for
sealability of premium connections:
𝑔𝑠𝑒𝑎𝑙𝑖𝑛𝑔 = 𝑆 𝐶 − 𝑆 𝐷 (2)
where 𝑆 𝐶 is the connection sealing capacity to resist leaking, and 𝑆 𝐷 is the differential pressure
demand. 𝑆 𝐶 is typically expressed as a function of the contact stress profile (i.e. magnitude,
length, and distribution) over the seal region, and 𝑆 𝐷 is often expressed in terms of differential
pressure of gas/liquid and sealing compound.
Note that to be consistent with ISO 13679 requirements (ISO 2002), the torque shoulder region is
not considered to be a part of the sealing mechanism for the leakage resistance assessment.
Figure 4. Illustration of seal contact stress distribution.

In a more generalized assessment approach, the sealing capacity of a premium connection can be
represented by the “Seal Contact Intensity” (fs), which is defined as the integration of the seal
contact stress (σc) over the effective seal length (LES):
𝑓𝑠 = ∫𝐿 𝜎𝑐 𝑑𝑥 (3)
𝐸𝑆
Based on Xie et al. (2012), a seal contact intensity value of equal to or greater than 250 N/mm is
considered sufficient for thermal wells where the fluid pressure is typically low (e.g. less than
15 MPa).
For HPHT wells, Xie (2013) recommended using a weighted area of seal contact stress for
evaluating connection sealability:
𝐿
𝑊𝑎 = ∫0 𝑃𝑐𝑛 (𝑙)𝑑𝑙 (4)
where Pc(l) is the seal contact pressure, L is the seal length, and n is a correlation exponent. Based
on test results, Murtagian et al. (2004) proposed values of 1.2 and 1.4 for the exponent n, for
connections with and without sealing compounds, respectively.

A premium connection might be considered to provide acceptable seepage resistance when the
weighted area of seal contact stress (Wa) is greater than some critical value Wac. Based on
experience, further review of the available data and incorporation of the ISO 13679 leakage rate
limit, the following equation for calculating the critical value of Wac was recommended for HPHT
tubular connections (Xie 2013):
𝑃𝑔𝑎𝑠 0.838
𝑊𝑎𝑐 = 0.01 � � (m·MPa1.4) (5)
𝑃𝑎𝑡𝑚
Surface roughness could also have an impact on the leakage rate for premium connections. The
effect of the roughness properties on the leakage rate could be studied through analytical
simulations and physical tests. Further studies are also required to develop suitable sealing criteria
for sour gas wells.
4. RBDA of Connection Sealability

Reliability analysis is performed using Monte Carlo simulation on the limit state functions defined
either explicitly or implicitly. Since FEA is used to determine the connection sealabilty, the limit
state function may not be explicitly defined as a function of the design variables. In the absence of
an explicit format, surrogate functions need to be developed to approximate the limit states using
techniques such as the Response Surface Method (RSM) (Rajashekhar 1993; Haldar 2000) or
Radial Basis Functions (RBF) (Buhmann 2003). The proposed RBDA approach for connection
integrity includes the following three steps:
FEA Sampling
Due to the large number of relevant variables in connection design and load conditions, a
significant number, possibly millions, of Monte Carlo simulations may be required to derive a
prediction of probability of failures. It is then impossible to perform detailed FEA for each Monte
Carlo simulation point. A more practical approach would be to perform the reliability analysis
based on selective FEA on parameters with more significant influence. The FEA sampling cases
can be defined through trial analyses.
Surrogate Function
A surrogate function can be developed based on the FEA data to facilitate the Monte Carlo
simulations. The surrogate function can be developed by using a linear Radial Basis Function
(RBF) interpolation and extrapolation from the FEA dataset. The RBF uses the Euclidian distance
between points to perform the interpolation (Buhmann 2003). To improve the accuracy of the
RBF interpolation, all input data are normalized to unit length using the range of the FEA data.
This method was configured to ensure that the surrogate function passed exactly through the FEA
data points (i.e. the nodal points). Deng (2006) noted that unlike the response surface method, with
RBF based approaches, it is not necessary to know the underlying relationship or to suppose a
relationship between the input variables and the output. The RBF is a universal approximator and
can be used to approximate linear or non-linear, implicit or explicit performance functions. The
RBF based Monte Carlo simulation is especially useful for reliability problems with implicit and
nonlinear performance functions where other reliability methods are not applicable.
According to Buhmann (2003), an RBF-based surrogate function is calculated from the weighted
sum of radial basis functions that typically use the n-dimensional Euclidian distance from the
interpolation point to each of the FEA input data points:

𝑁
𝑦(𝐱) = � 𝑤𝑖 𝜙( ‖𝐱 − 𝐱 𝑖 ‖) (6)
𝑖=1
where
𝑟𝑖 = ‖𝐱 − 𝐱 𝑖 ‖ = the Euclidian distance to the coordinates of FEA data point 𝐱 𝑖

𝜙(𝑟𝑖 ) = the radial basis function for 𝐱 𝑖
𝑤𝑖 = weight for 𝜙(𝑟𝑖 )
The surrogate function weights (wi) can be calibrated using the least squares method, because the
function is linear in wi (or an iterative neural network learning method) (Buhmann 2003;
Deng 2006).
Isight (2014) implemented a methodology to calibrate the shape function variable of a power
spline RBF, for minimizing the errors resulting from data interpolations/extrapolations during
cross-validation.
For the work presented in this paper, linear RBF was used for ease of implementation and
flexibility in performing Monte Carlo Simulation for the RBDA.
Monte Carlo Simulation
Monte Carlo simulation is implemented to determine the reliability of connection integrity by
establishing the portion of cases that would have the weighted area of seal contact stress less than
the threshold value. This method provides a means to establish the effect that changes to the
design (shifted mean, µ) or improved consistency (reduced standard deviation, σ) of a particular
design parameter would have on the reliability of the tubular connection.
RBDA Flow Chart

Figure 5 presents the flow chart for RBDA which is composed of FEA sampling, establishment of
surrogate function using RBF, and Monte Carlo Simulation.

Initial FEA of nominal case Redesign
connection
Trial FEA to identify key parameters

and define parametric FEA matrix
Parametric FEA to create

comprehensive database with high
resolution near failure surface
Calibrate RBF interpolation scheme
No Adequate
resolution
Yes
Monte Carlo simulations with

selected design distributions to Adjust design
calculate probability of failure (POF) tolerances
No or
Complete No Yes POF < threshold
Figure 5. RBDA flow chart.
5. Analysis Example
An analysis example is presented here to demonstrate the use of the proposed RBDA approach for
casing connections in HPHT application. The analysis example used a 177.8 mm, 34.2 kg/m P110
generic premium casing connection. The connection model includes the basic features common to
many of the premium connections currently used in HPHT well applications (e.g. buttress thread
form, axial torque shoulder, and radial metal-to-metal seal) so that the analysis results would be
reasonably representative for premium connections in the HPHT application.
5.1 Load Scenarios
Guidelines for qualification of tubing and casing connections have been established by ISO 13679,
“Petroleum and natural gas industries — Procedures for testing casing and tubing connections”
(ISO 2002). According to ISO 13679, tubular connections for HPHT applications should meet the
requirements of a Connection Application Level (CAL) IV.
The RBDA of the analysis example considered the response of the generic premium connection
subjected to combined load conditions (i.e. axial tensile/compressive forces, and internal/external
pressures) as specified by the ISO 13679 CAL-IV Test Series A category.
Figure 6 presents the Test Series A load path for a 177.8 mm, 34.2 kg/m P110 generic premium
connection, with a rated axial load capacity greater than or equal to the pipe-body in compression.
The load path was calculated based on guidelines provided by ISO 13679 (2002). There are a total
of 14 load points with various combinations of internal/external pressures, and tensile/compressive
forces included in the analysis.

80
100% VME Pipe Body Yield Envelope
95% VME Pipe Body Yield Envelope
4
60 Load Point 5 3
Internal
2
6
40
7
20
Pressure (MPa)
1
0 8, 9
14
-20
10 11 12 13
-40
External
-60
-80
-5000 -4000 -3000 -2000 -1000 0 1000 2000 3000 4000
Compression Tension
Axial Load (kN)
Figure 6. Test Series A load path for a connection with an axial load rating greater
than or equal to the pipe body in compression (ISO 2002).
5.2 Base Case Results
Figure 7 presents the contour plots of radial stress for the base case at make-up, Load Point 4
(axial tension and internal pressure) and Load Point 13 (axial compression and external pressure).
The figure shows that at make-up, the threads are quite uniformly engaged through the thread
length, and the contact stress is distributed over the seal length. At Load Point 4, the internal
pressure appears to enhance the contact stress at the seal. However, at Load Point 13, the seal
contact stress is reduced significantly under external pressure.

Figure 7. Contour plot of radial stress (MPa) at make-up (top),
Load Point 4 (middle) and Load Point 13 (bottom).
Figure 8 presents the weighted contact stress area Wa and the critical value of Wac, calculated using
Equations 4 and 5. Since the internal/external pressures vary through the load points, as shown in
Figure 6, the critical value of Wac also varies in Figure 8. As shown for this base case, the results
of the weighted area of seal contact stress are higher than the critical values for all load points.
However, the base case analysis suggests that the potential leakage scenarios could include:
• Load Points 2 through 4 for containing the internal pressure when the connection is
subjected to nearly the maximum internal pressure and the maximum axial tensile force;
and
• Load Points 10 through 14 for containing the external pressure, when the connection is
subjected to nearly the maximum external pressure and a high tensile force.
Among all the load points, Load Point 13 with a combination of axial compression force and
external pressure appears to be the most critical in terms of connection leakage resistance. As
such, Load Point 13 was chosen for the reliability assessment in this paper.

25
20
Seal - Wa (MPa1.4·m)
15
10
0
0 2 4 6 8 10 12 14
Load Point
Figure 8. Calculated values of the weighted area of seal contact stress

through various load points (Point 0 represents make-up).
5.3 Parametric Analyses
The following parameters were considered for the sensitivity analyses to facilitate the reliability
assessment of connection sealing capacity:
• thread and seal interferences,
• thread and seal tapers for pin and coupling,
• make-up torque, and
• material yield strength.
In defining the distribution of the varying geometric parameters, the tolerances specified by the
connection drawing were considered to cover a range of four standard deviations centered on the
mean value (i.e. μ - 2σ< x <μ + 2σ, where μ is the mean value and σ is the deviation). Such
defined range would cover up to 95.5% of all possible cases given a normally distributed variable.
Figure 9 presents a tornado plot of the effect of each variable on the weighted area of seal contact
stress, Wa, based on the trial FEA. As shown in the figure, the thread and seal interference are
identified as the top two primary variables. The other parameters are considered to be secondary
variables.
For the two primary variables, a full factorial study was performed with five treatments for each
variable with the following levels: μ – 3σ; μ – 1.5σ; μ; μ + 1.5σ; μ + 3σ. This procedure required a
total of 25 (i.e. 52) simulations for the two top influential variables. Approximately 200 FEA
simulations were performed to cover the sensitivities of all variables.
FEA results of Wa – Wac values for the full factorial FEA conducted on the two most influential
variables are presented as hollow circles in Figure 10. It shows that the weighted area of seal
contact stress decreases with the thread interference and increases with the seal interference.

Wa (m∙MPa1.4)
2.50 2.75 3.00 3.25 3.50 3.75 4.00 4.25 4.50
Seal Interference
Thread Interference
Yield Stress
Pin Thread Taper
Coupling Thread Taper
High
Torque
Low
Coupling Seal Taper
Pin Seal Taper
Figure 9. Tornado plot of trial sensitivity FEA results.
5.4 Surrogate Function and Reliability Analysis

Based on the FEA results discussed above, a surrogate function was established using a linear
RBF interpolation. Figure 10 presents Wa – Wac values interpolated and extrapolated using RBF
surrogate function for a range of thread and seal interference domain. The FEA dataset is shown as
hollow circles in the figure. The figure demonstrates that the surrogate function generates a
response that is smooth and produces a representative output in the extrapolation regions.
To evaluate the accuracy of the surrogate function, a “leave one out” cross-validation was
performed. The procedure tested the surrogate function at each FEA data point by removing that
point from the dataset and using that reduced dataset (i.e. dataset-1) to calibrate the surrogate
function. Figure 11 shows the comparison of all the FEA data points to the response from the
surrogate functions with reduced datasets. Interpolated/extrapolated data points represent
generated points based on interpolation/extrapolation from the reduced FEA dataset, using the
RBF method. The reasonably accurate correlation suggests that the proposed method for
developing the surrogate function was adequately accurate in the region of the assessment limit
(Wa-Wac=0) as a result of the finer resolution of the dataset in this region. In all cases, the call
between pass and failure is made correctly (i.e. above or below the assessment limit).
Note Isight (2014) offers calibration of power spline RBFs by minimizing the errors between the
FEA data points left out of the dataset and the corresponding interpolated/extrapolated results.

Figure 10. Wa - Wac (mm·MPa ) values using RBF surrogate function in thread
1.4
interference and seal interference domain.
Figure 11. “Leave one out” cross-validation of surrogate function.

Monte Carlo simulation was performed to determine the reliability of the connection sealability.
The results of the Monte Carlo simulation can establish the portion of connections that would have

a weighted area of seal contact stress lower than the limit established by Equation 5 based on the
probability functions of the design parameters.
The method presented in this paper provides a means to establish the effect that changes to the
design (i.e. shifted mean, µ) or the manufacturing consistency (i.e. reduced standard deviation, σ)
of any given design parameter would have on the reliability of the connection sealability. For
example, by simultaneously adjusting the mean values of the three most significant parameters
(i.e. seal interference, thread interference and yield stress), the effect on the probability of failure
due to connection sealability is shown Figure 12. The α value in the figure is the shift factor over
the standard deviation σ. If the means for these three design parameters deteriorate from the mean
value by one σ, the probability of failure will increase dramatically from 0.0008% (without shift)
to 0.18% (with the maximum shift), as shown in Figure 11.
0.20%
0.18%
0.16%
Probability of Failure
0.14%
0.12%
0.10%
0.08%
0.06%
0.04%
0.02%
0.00%
0 0.2 0.4 0.6 0.8 1
α
Figure 12. Effect on probability of failure due to mean value deterioration.
6. Platform for RBDA

The key objective for the proposed approach is to assist the designers/manufacturers to optimize
the connection design by adjusting design parameters, thereby minimizing the risk of failure due
to connection leakage. A platform can be developed to facilitate the RBDA.
Figures 13 shows the interface of the platform for calculating capacity, demand and reliability.
The data input for connection design in embedded in the capacity interface, which allows a user to
specify statistical distribution (e.g. distribution type, mean and standard deviation) of connection
design parameters (e.g. geometry, material property, and make-up torque). The distribution types
for the data input include normal, uniform, log-normal, weibull, and triangular. The reliability
calculation interface allows a user to perform calculations to determine the reliability and
probability of failure for connection sealability under internal and external pressures.

(2) Calculating Demand
(1) Data Input for Calculating Capacity (3) Calculating Reliability
Figure 13. RBDA platform for connection sealability.

The presented RBDA platform can be used for other connection designs, provided additional FEA
of various parametric cases are performed and the additional FEA data are stored in the platform
dataset. Designers/manufacturers can use this platform to adjust design parameters to reach a
targeted reliability for acceptance of the product.
7. Summary
This paper presents a methodology for RBDA of tubular connection sealability for oil wells. The
RBDA was based on FEA of tubular connections under various load conditions such as make-up,
axial tension and compression, internal and external pressures, and curvature loading. A surrogate
function is established using a linear RBF interpolation to facilitate the Monte Carlo simulations to
determine the reliability of connection sealability. An example analysis was performed to
demonstrate the use of the proposed methodology for RBDA of connection sealability for HPHT
well application.

The presented FEA-based RBDA methodology can be used by manufacturers to optimize the
connection design by adjusting the mean value and the distribution of key design parameter, to
reach a minimal risk of connection failure due to leakage.
8. References
1. Abaqus User Manual. 2014. Version 6.14.
2. API RP 5C3. 2008. Equations and Calculations for Casing, Tubing, and Line Pipe Used as
Casing or Tubing; and Performance Properties Tables for Casing and Tubing. 1st Edition.
3. API Spec 5CT. 2011. Specification for Casing and Tubing. 9th Edition, July.
4. Buhmann, M.D. 2003. Radial Basis Functions: Theory and Implementations. Cambridge
University Press, ISBN 978-0-521-63338-3
5. Deng, J. 2006. Structural Reliability Analysis for Implicit Performance Function Using Radial
Basis Function Network, International Journal of Solids and Structures 43. 3255–3291.
6. Haldar, A. and Mahadevan, S. 2000. Reliability Assessment Using Stochastic Finite Element
Analysis. John Wiley & Sons.
7. International Organization for Standardization (ISO). 2002. Petroleum and Natural Gas
Industries — Procedures for Testing Casing and Tubing Connections. ISO 13679:2002(E).
8. Isight User Guide. 2014.
9. Murtagian G.R., Fanelli V., Villasante J. A., et al. 2004. Sealability of Stationary Metal-to-
Metal Seals. Journal of Tribology. Vol. 126, No. 3. 2004: 591.
10. Payne M.L. and Schwind B.E. 1999. A New International Standard for Casing/Tubing
Connection Testing. SPE/IADC 52846. Presented at the SPE/IADC Drilling Conference.
Amsterdam, Holland, March 9-11.
11. Rajashekhar, M. and Ellingwood, B. 1993. A New Look at the Response Surface Approach
for Reliability Analysis. Structural Safety, Vol. 12, issue 3, pp. 205-220.
12. Xie J. 2006. Casing Design and Analysis for Heavy Oil Wells. Paper 2006-415, World
Heavy Oil Conference, Beijing, China, November.
13. Xie J. 2007. Analysis of Oil Well Casing Connections Subjected to Non-axisymmetric
Loads. Abaqus User’s Conference. Paris, France, May 22-24.
14. Xie J., Fan C., Tao G., Matthews, C. 2011. Impact of Casing Rotation on Premium
Connection Service Life in Horizontal Thermal Wells. WHOC11-558, March.
15. Xie J. 2011. Analysis of Strain Rate Impact on Make-up of Oilfield Premium Casing
Connections. 2011 SIMULIA Customer Conference, May.
16. Xie J. and Hassanein S. 2012. Reliability-Based Design and Assessment (RBDA) Method for
Thermal Wells. Paper No. WHOC-144, World Heavy Oil Congress.
17. Xie J. 2013. Numerical Evaluation of Tubular Connections for HPHT Applications. Baosteel
Conference, Shanghai, China.
18. Xie, J., Rong, F. and Dessein, T. 2014. Methodology for Reliability-Based Design and
Assessment of Tubular Connection Sealability in HPHT Wells. International Conference on
Oil Country Tubular Goods and Tubular String. Xi’an, China, May 22-23, 2014.

9. Acknowledgements
The work presented in this paper was supported by C-FER Technologies (1999) Inc., Canada.
The authors would like to sincerely acknowledge Dr. Maher Nessim, Chief Engineer at C-FER
and FCAE, for his technical guidance in the development of the reliability analysis model;
Dr. Nader Yoosef-Ghodsi, Senior Research Engineer at C-FER, for his technical review of this
study.

Simulation of Hydraulic Fracturing of
Unconsolidated Sands using Fully Coupled Poro-
Elastoplastic Models
Saurabh Puri, Juan Hurtado, Deepak Datye
Dassault Systemes Simulia Corp, 1301 Atwood Avenue, Suite 101W, Johnston, RI 02919, USA
Ganesh Dasari, Kevin Searles, Pablo Sanz

ExxonMobil Upstream Research Company, 22777 Springwoods Village Parkway, Spring, TX
77389
Abstract: Many deep water reservoir rocks are soft and consist of unconsolidated or weakly
cemented sands. Fractures may have to be created in these soft reservoirs for a variety of
purposes, such as the injection of fluids to maintain reservoir pressure, constructing Frac-Pack
sand control mechanisms, etc. It is now recognized that fracturing in unconsolidated sands
involves complex failure mechanisms and is very different from fracturing in hard rocks. The
simulation of these complex failure mechanisms can only be achieved with suitable models that
account for the coupling of inelastic soft rock deformation and fluid pressure changes. In this
paper, we discuss the implementation in Abaqus of a poro-elastoplastic constitutive model for soft
rock and its application to some practical problems. First we present a critical-state based
constitutive model for soft rock, which was specifically designed to describe the evolution of sand
and weakly cemented rock from the intact state to the fully damaged state. It includes pressure
dependency of the elastic response as well as the yield surface. A softening regularization scheme
is also introduced to minimize mesh sensitivity due to strain localization effects. The model is
implemented in Abaqus/Standard and Abaqus/Explicit using an implicit backward-Euler
integration scheme. Numerical examples are presented in the second part of the paper in the
context of applications of practical interest to the Oil and Gas industry. First, we demonstrate the
model’s ability to predict shear band formation and localization in a borehole stability
analysis. Second, we investigate the evolution of a fracture due to injection of fluids in
unconsolidated sand.
Keywords: Plasticity, Geomechanics, Shear bands
1. Introduction
Planning Oil & Gas exploration and development operations often requires predictive modeling of
the behaviors of rocks under loading, such as the loads applied or induced during drilling,
fracturing or production. Finite element methods, or other numerical methods, are often used with
rock mechanical constitutive models to simulate the behaviors of rocks so that critical information
can be obtained and analyzed to ensure operational success. In particular, weakly cemented rocks
exhibit complex deformation behaviors, such as localized deformation associated with borehole

breakout and formation liquefaction during injection. Simulating such complex behaviors
necessitates constitutive models that are able to reproduce a physically realistic evolution of the
material from the intact state to the fully damaged state. In this paper, a critical-state based soft
rock plasticity model that follows the work of Crook et al. (2006) is discussed. The model is
referred to as the soft rock plasticity model in this document.
The paper is organized as follows: the governing equations of the soft rock plasticity model are
summarized in Section 2. The model is first used to analyze shear band localization in a borehole
using Abaqus/Explicit in Section 3. Next, in Section 4, the model is used to study hydraulic
fracturing of unconsolidated sands using Abaqus/Standard. The paper ends with some concluding
remarks in Section 5.
2. Constitutive model for soft rocks

The governing equations of the soft rock plasticity model are described in this section. This poro-
elastoplastic constitutive model is based on the theory of critical state and assumes additive
decomposition of strain rate (𝜺̇ ) into elastic (𝜺̇ 𝑒𝑙 ) and plastic parts (𝜺̇ 𝑝𝑙 ).
𝜺̇ = 𝜺̇ 𝑒𝑙 + 𝜺̇ 𝑝𝑙 (1)
The elastic response is described by,
𝝈 = 𝑫𝑒𝑙 : 𝜺𝑒𝑙 (2)
where 𝝈 is the effective stress tensor, 𝜺𝑒𝑙 is the elastic strain tensor and 𝑫𝑒𝑙 is the fourth-order
elasticity tensor defined by Young’s modulus (𝐸) and Poisson’s ratio (𝜈). Experimental data on
soft rocks indicate that 𝐸 and 𝜈 are dependent on effective mean stresses. The following empirical
relations are used to define the evolution of 𝐸 and 𝜈 with effective pressure,
𝑛
𝑝+𝑝
𝑜
𝐸𝑟𝑒𝑓 � � 𝑝>0
𝐸=� 𝑝𝑟𝑒𝑓 +𝑝𝑜 (3)
𝑓𝐸𝑟𝑒𝑓 𝑝≤0
𝑛
𝑝𝑜
𝑓=� � (4)
𝑝𝑟𝑒𝑓 +𝑝𝑜
𝜈𝑚𝑎𝑥 + (𝜈𝑚𝑖𝑛 − 𝜈𝑚𝑎𝑥 )(1 − 𝑒 −𝑚𝑝 ) 𝑝>0

𝜈=� (5)
𝜈𝑚𝑎𝑥 𝑝≤0
where 𝐸𝑟𝑒𝑓 is the reference Young’s Modulus at pressure 𝑝𝑟𝑒𝑓 ; 𝑝 = −(𝜎11 + 𝜎22 + 𝜎33 )/3 is the
effective mean stress; 𝑓 is used to define the elastic modulus for tensile effective pressure as a
function of 𝐸𝑟𝑒𝑓 ; 𝑝𝑜 , n and m are material constants; 𝜈𝑚𝑎𝑥 is the value of Poisson’s ratio at 𝑝 = 0,
and 𝜈𝑚𝑖𝑛 is the value of Poisson’s ratio at 𝑝 = ∞. The above description of elastic constants is
shown schematically in Figure 1.

Figure 1. Variation of elastic constants with mean effective stress.
The plastic response of the material follows the critical state based constitutive model proposed by
Crook et al. (2006), modified to introduce an eccentricity parameter (𝑒0 ). The yield function
depends on the three stress invariants and is defined in Equation (6) . It intersects the hydrostatic
axis in both tension and compression, as shown in Figure 2.
1
𝑝𝑙 𝑝−𝑝𝑐 𝑛𝑦
Φ�𝝈, 𝜀𝑣 � = �(𝑒0 𝑝𝑐0 tan𝛽)2 + (𝑔(𝝈)𝑞)2 − (𝑝 − 𝑝𝑡 )tan𝛽 � � − 𝑒0 𝑝𝑐0 tan𝛽 = 0 (6)
𝑝𝑡 −𝑝𝑐
where 𝛽 is a material parameter related to the friction angle at low confining pressures, 𝑞 =
3
� 𝑺: 𝑺 is the effective Mises stress, 𝑺 = 𝝈 + 𝑝𝐈 is the deviatoric effective stress, 𝑝𝑐 is the
2
intercept of the yield surface with the positive 𝑝-axis in the 𝑞 − 𝑝 plane, 𝑝𝑡 is the intercept of the
yield surface with the negative 𝑝-axis in the 𝑞 − 𝑝 plane, 𝑝𝑐0 is the initial value of 𝑝𝑐 , 𝑛𝑦 is a
material constant that controls the shape of the yield function in the 𝑞 − 𝑝 plane, and
𝑔(𝝈) captures the observed transition of the yield surface in the Π-plane from rounded-triangular
shape to circular shape with an increase in pressure. 𝑔(𝝈) is defined as
𝛼
1 𝑟3
𝑔(𝝈) = � �1 + 𝑓(𝑝) �� (7)
1−𝑓(𝑝) 𝑞3
3 9
𝑟 = � 𝑺: 𝑺: 𝑺 (8)
2
𝑝𝑐0
𝑓(𝑝) = 𝑓0 exp �𝑓1 𝑝 � (9)
𝑝𝑐
where 𝑟 is the third stress invariant, and 𝑓0 , 𝑓1 and 𝛼 are material constants.

Figure 2. Yield Surface and plastic potential of the soft rock plasticity model in the
𝒒 − 𝒑 plane.
The function 𝑓(𝑝) should satisfy the following constraints to ensure convexity of the yield
surface (Bigoni and Piccolroaz, 2004).
1 3
0 ≤ 𝑓(𝑝) ≤ 𝛼≥
9𝛼−1 11
4−13𝛼 1/2 3
0 ≤ 𝑓(𝑝) ≤ � � 0<𝛼< (10)
4(𝛼−1)(9𝛼 2 −1) 11
𝑝𝑙
The evolution of the yield surface is governed by the plastic volumetric strain, 𝜀𝑣 , via
hardening/softening relations in the form
𝑝𝑙
𝑝𝑐 = 𝑝𝑐 (𝜀𝑣 ) (11)
𝑝𝑙
𝑝𝑡 = 𝑝𝑡 (𝜀𝑣 ) (12)
𝑝𝑙 𝑝𝑙
Typical curves of 𝑝𝑐 (𝜀𝑣 ) and 𝑝𝑡 (𝜀𝑣 ) are shown in Figure 3. The initial size of the yield surface
𝑝𝑙
(𝑝𝑐 |0 and 𝑝𝑡 |0 ) is determined from the initial volumetric plastic strain 𝜀𝑣 �0 .
Figure 3. Typical piecewise linear hardening/softening curves of the soft rock

model.

2.1. Flow rule
The evolution of the plastic flow is governed by a non-associated flow rule so that the plastic
strain rate is defined as
𝜕Ψ
𝜺̇ 𝑝𝑙 = 𝜆̇ (13)
𝜕𝝈
where 𝜆̇ ≥ 0 is the plastic multiplier and Ψ is the plastic potential defined as (see Figure 2)
1
𝑝𝑙 𝑝−𝑝𝑐 𝑛𝑦
Ψ�𝝈, 𝜀𝑣 � = �(𝑒0 𝑝𝑐0 tan𝜓)2 + (𝑔(𝝈)𝑞)2 − (𝑝 − 𝑝𝑡 )tan𝜓 � � − 𝑒0 𝑝𝑐0 tan𝜓 (14)
𝑝𝑡 −𝑝𝑐
where 𝜓 is a material parameter related to the dilation angle at low confining pressures. It should
be noted that adding eccentricity to the model ensures uniqueness of plastic flow at 𝑝 = 𝑝𝑐 and
𝑝 = 𝑝𝑡 , as shown in Figure 4.
Figure 4. Effect of eccentricity on the shape of the plastic potential in the q-p
plane.
2.2. Regularization of the softening process
Granular materials often exhibit strain localization with increasing plastic deformation. Post-
failure solutions from conventional finite element methods can be strongly mesh-dependent. In
order to mitigate mesh-dependence of the solutions, a regularization method is often used to
(𝑚)
introduce a micro-structural length-scale into the constitutive formulation. Let 𝑙𝑐 denote the
(𝑒)
characteristic width of a shear band or a crack band, 𝑙𝑐 the characteristic length of the element,

and (𝜀 𝑝 )(𝑒) the inelastic strain for the element; then the inelastic strain in the localization band,
(𝜀 𝑝 )(𝑚) , is defined as
(𝑒) 𝑛𝑟
𝑙
(𝜀 𝑝 )(𝑚) = (𝜀 𝑝 )(𝑒) min �� (𝑚)
𝑐
� , 𝑓𝑚𝑎𝑥 � (15)
𝑙𝑐
where 𝑛𝑟 is a material parameter and 𝑓𝑚𝑎𝑥 is a positive number used for bounding the magnitude
of regularization. The strain regularization method is strictly only valid when the characteristic
(𝑒) (𝑚)
length of the element is greater than the width of the localization band, i.e., 𝑙𝑐 ≥ 𝑙𝑐 .
2.3. Implementation in Abaqus
The poro-elastoplastic constitutive model, described in the previous subsections, was implemented
in Abaqus/Standard and Abaqus/Explicit using an implicit backward-Euler integration scheme.
This implementation was used for the numerical analyses described next.
3. Wellbore stability-thick walled cylinder test

Borehole collapse can lead to significant economic consequences, and therefore, its prediction and
prevention becomes important for the success of oil and gas projects. The behavior of rock along
the periphery of a borehole depends on its elastic, plastic, as well as failure properties. The failure
of rocks is often associated with the formation of shear bands and the localization of strains along
those bands. Reliable prediction of the formation of such shear bands is therefore important, along
with an accurate computation of the stress distribution in the rock strata in the neighborhood of the
borehole.
The deformation of a borehole can be studied by performing computational tests on thick cylinders
subjected to external pressure. Such thick cylinder tests were performed using Abaqus/Explicit
and are discussed in this section. The setup of the model is based on the numerical experiments
described in Crook et al. (2003).
3.1. Problem setup
A thick walled cylinder with an internal diameter (ID) of 10 mm and outer diameter (OD) of 30
mm was considered. Sketch of a typical model is shown in Figure 5. The tangential displacement
was constrained at four points on the outer boundary of the cylinder. These points are indicated by
blue dots in Figure 5. Pressure was applied on the outer boundary of the cylinder and was ramped
up linearly from 0 MPa to 3.52 MPa in 0.044 s.
Plane strain elements, CPE4R and CPE3, were used for discretizing the model. The effect of
element size on the results was also studied. Three different meshes with 320, 480 and 640
elements on the inner circle of the annulus were used. The soft rock plasticity model with constant
linear elastic properties was used in all the analyses. The material properties used in this example
are listed in Table 1. Variable mass scaling was used to speed up the solution time of the
Abaqus/Explicit analysis.

Figure 5. Sketch of the model used for testing.
Table 1. Material properties used in the wellbore stability problem.

Elastic properties
𝐸 (Pa) 𝜈
7 × 1010 0.3
Plastic properties
𝛽 𝜓 𝑛𝑦 𝑓0 𝑓1 (Pa−1 ) 𝛼 𝑒0
50o 48o 1.3 0.06 0.007 0. 0.001
Softening regularization properties

(𝑚) 𝑛𝑟 𝑓𝑚𝑎𝑥
𝑙𝑐 (m)
5 × 10−5 0.6 1.0
Hardening in hydrostatic compression

𝑝𝑙
𝑝𝑐 (Pa) 𝜀𝑣
2.5 × 106 0.0
2.52 × 107 0.002
1.0 × 108 0.003
Hardening in hydrostatic tension

𝑝𝑙
𝑝𝑡 (Pa) 𝜀𝑣
−2.5 × 104 0.
−2.7 × 104 0.002
−3.0 × 104 0.003
3.2. Results and discussion

As the analyses progress, the increasing pressure eventually leads to the formation of shear bands,
as shown in Figure 6. The criterion for the onset of instability is assumed to be based on the
increase in kinetic energy. For all the meshes studied, the kinetic energy increases substantially at
t=0.026 s, which corresponds to an external load of 2.08 MPa. It was concluded from this
observation that the external load at which the instability initiates is independent of the mesh.
However, some variation of the localization profile with element size was observed.
Figure 6. Localization in thick walled cylinder test; equivalent plastic strain (PEEQ)
profile is shown for the mesh with 640 elements on the inner circle at an external
load of 3.2 MPa.
4. Fracturing of unconsolidated sands

Unconsolidated or weakly cemented sand reservoirs experience significant compaction and this
compaction can lead to reduction in permeability and failure of completion equipment. In order to
mitigate these challenges, water or other fluids are injected as oil/gas is extracted to maintain
reservoir pressure. The optimum design of injection program requires thorough understanding on
how fractures develop in unconsolidated sand reservoirs. Limited number of experiments
conducted by various researchers show that fracturing in unconsolidated sands differs from
fracture growth in hard rocks because of poroelastic effects and elastoplastic material behavior. A
key result from past experimental program is that fracture growth and pressure distribution in
unconsolidated sands strongly depends on fluid rheology and matrix permeability. Thus the
objectives of this analysis are to study:
• failure modes and evolution of breakdown pressure in unconsolidated sands, and
• the shape of fracture for different mobility of injected fluids.

4.1. Problem setup
In this problem, a borehole in a rectangular domain was considered, as shown in Figure 7. Only
half of the domain was modeled due to symmetry of the geometry. Plane strain, pore pressure
elements, CPE4P, were used for discretizing the model. Water was injected into the well at a
prescribed flow rate using the pipe element, FP2D2, in Abaqus/Standard. Pore pressure on the left,
right and top surfaces was set to zero during the analysis. The top surface wass fixed in the
horizontal direction and symmetric boundary conditions in the vertical direction were applied on
the bottom surface at all times.
The porous elasticity model with soft rock plasticity, as described in Section 2, was used in this
𝑝𝑙 𝑝𝑙
example (material properties are listed in Table 2). The hardening curves of 𝑝𝑐 (𝜀𝑣 ) and 𝑝𝑡 (𝜀𝑣 )
are shown in Figure 8. The permeability of the unconsolidated sand was assumed to vary with
void ratio as shown in Figure 9. Two test cases with different values of permeability were
considered: a baseline case with lower permeability and a case with increased permeability. Initial
void ratio was set to 0.36 and the following initial stress state was prescribed:
𝜎𝑥𝑥 = −520 psi, 𝜎𝑦𝑦 = −670 psi
This analysis involves three steps. In the first step, surface loads, equal in magnitude to the initial
stresses, were applied on the top, left and right surfaces. The surface of the borehole was fixed in
the horizontal and vertical directions, and the bottom surface was fixed in the horizontal direction.
An equilibrium configuration was obtained using the *GEOSTATIC procedure. In the second step
the “construction” of the wellbore was simulated by removing the displacement constraints on the
borehole surface as well as the constraints in the horizontal direction on the bottom surface. In the
third step, water was injected into the borehole using pipe elements by prescribing a volumetric
flow rate at one end of the pipe; the pressure degree of freedom at the other end of the pipe was
coupled to the surface of the borehole using the surface-to-surface tie constraints. The prescribed
volumetric flow rate was ramped from zero to a constant value for the initial 1 sec of the step time
and was then kept constant for another 1 second. This constant value of the prescribed flow rate is
different for the two cases with different permeabilities.

Figure 7. Sketch of the model used for testing; inset shows the geometry of the
borehole.
Figure 8. Hardening curves used in the fracturing problem.

Table 2. Material properties used in the fracturing of unconsolidated sands
problem.
Porous Elasticity properties
𝐸𝑟𝑒𝑓 (psi) 𝑝𝑟𝑒𝑓 (psi) 𝑝𝑜 (psi) 𝑛 𝜈𝑚𝑖𝑛 𝜈𝑚𝑎𝑥 𝑚
1.5 × 105 0.0 100.0 0.01 0.2 0.28 0.003
Plastic properties
𝛽 𝜓 𝑛𝑦 𝑓0 𝑓1 (psi−1 ) 𝛼 𝑒0
67.1o 66.9o 1.25 0.5 1.5 × 10−6 0.25 0.001
Softening regularization properties

(𝑚) 𝑛𝑟 𝑓𝑚𝑎𝑥
𝑙𝑐 (in)
0.05 0.6 100.0
Figure 9. Permeability as a function of void ratio. k is the permeability in in/sec and

k0 is the base permeability corresponding to the void ratio of 0.18.
4.2. Results and discussion

Results for the fracturing problem are summarized in this section. Figure 10 shows the effect of
permeability on the borehole breakdown pressure. It is observed that the breakdown pressure
slightly increases on decreasing the permeability. The breakdown pressure (𝑃𝑏 ) of an elastic
medium with zero tensile strength is a function of initial stress state and is given by (Hubbert and
Willis, 1957)
𝑃𝑏 = 3σℎ − σ𝐻 − 𝑃 (16)

where σℎ is the in-plane minimum total stress, σ𝐻 is the in-plane maximum total stress and P is
the pore pressure. The breakdown pressure per the above equation is 890 psi. The computed
breakdown pressure is higher than 890 psi for the following reasons:
(i) Ductility of unconsolidated sands,
(ii) Absence of explicit fracture representation (e.g. cohesive zone elements), and
(iii) Finite amount of tensile strength prior to breakdown.
Figure 11 shows plastic strains and pore pressure at the end of injection. As can be seen, the
localization profile depends on the permeability of the formation. In the higher permeability case,
there is significant leak-off into the formation, whereas in the baseline case with lower
permeability there is less leak-off and more energy is available to propagate “fracture”. This
causes the “fracture” (represented by the localization of equivalent plastic strain, PEEQ) in the
lower permeability formation to extend further than in the higher permeability formation.
The stress path of an integration point close to the fracture and the borehole was analyzed and was
found that the material fails on the shear failure surface, rather than on the tensile failure surface.
Figure 10. Variation of breakdown pressure with permeability.

(a)
(b)
Figure 11. Equivalent plastic strain (a) and pore pressure (b) at the end of the
simulation. In both, (a) and (b), figure on the left side represents the lower
permeability case (baseline) and the figure on the right side represents the higher
permeability case.

5. Conclusions
A poro-elastoplastic constitutive model for weakly cemented rocks was implemented in Abaqus.
The model captures the variation of elastic properties and strength with mean effective stress. The
implementation was verified using simple problems for which analytical solutions may be
computed. The soft rock model was used to simulate wellbore stability and hydraulic fracturing
applications. The borehole model was able to capture the formation of shear bands due to
enhanced shear loading. The results from the simulation of fracturing in unconsolidated sands
show that fracturing is dominated by shear failure as opposed to Mode I tensile failure in hard
rocks. The results also show that higher ductility of unconsolidated sands leads to significantly
higher net fracturing pressure compared to that of hard rocks. It was also observed that breakdown
pressure in soft rocks is higher than the minimum stress, again due to ductility.
6. References
1. Bigoni, D., Piccolroaz, A., Yield Criteria for Quasibrittle and Frictional Materials,
International Journal of Solids and Structures, 41, 2855 – 2878, 2004.
2. Crook, T., Willson, S. M., Yu, J. G., Owen, D. R. J., Predictive Modeling of Structure
Evolution in Sandbox Experiments, Journal of Structural Geology, 28, pp. 729, 2006.
3. Crook, T., Willson, S.M., Yu, J.G., Owen, D.R.J., Computational Modelling of the Localized
Deformation Associated with Borehole Breakout in Quasi-brittle Materials, Journal of
Petroleum Science and Engineering, 38, 177–186, 2003.
4. Hubbert, M.K. and Willis, D.G., Mechanics of Hydraulic Fracturing, Petroleum Transactions,
AIME, 210, 1947.

Finite Element Verification of Non-Homogeneous
Strain and Stress Fields during Composite Material
Testing
Lars P. Mikkelsen
Composite and Materials Mechanics, DTU Wind Energy,
Technical University of Denmark, 4000 Roskilde, Denmark
Abstract: Uni-directional glass fiber reinforced polymers play a central role in the task increasing
the length of wind turbines blades and thereby lowering the cost of energy from wind turbine
installations. During this, optimizing the mechanical performance regarding material stiffness,
compression strength and fatigue performance is essential. Nevertheless, testing composites
includes some challenges regarding stiffness determination using conventional strain gauges and
achieving correct material failure unaffected by the gripping region during fatigue testing.
Challenges, which in the present study, has been addressed using the finite element method.
During this, a verification of experimental observations, a deeper understanding on the test
coupon loading and thereby improved test methods has been achieved.
Keywords: Composites, Experimental Verification, Fatigue, Scripting
1. Introduction
Glass fiber reinforced polymers are a lightweight material which compared to its weight and price
is a stiff and a strong material in the fiber direction. Glass fiber composites is a material
technology, which has made it possible for the wind turbine industry designing wind turbine
blades with a length of a little more than 80 m at present. A development making it possible
making onshore wind turbines, which regarding the cost of energy, can compete with conventional
energy sources. On the other hand, offshore wind turbines is still too expensive due to larger
installations and maintenance cost regarding a direct competition with conventional energy
sources and will properly need to grow with at least 50% in diameter in order to be competitive.
Therefore, there are intensive research activities focusing on making better but still cheap
composites materials. Those activities involve large amount of experimental material
characterizations where the main focus is on improving the materials stiffness, compression
strength and fatigue life.
One of the challenges testing composite materials are the highly anisotropic properties for the uni-
directional glass fiber reinforced polymers used for the load carrying parts in a wind turbine blade.
Anisotropy which are resulting in stiffness and strength properties more than 10 times larger in the
fiber direction compared with loading in the transverse direction and under shear. In addition, even
in the fiber direction the material is rather compliant compared with metallic materials. During
mechanical testing, both the anisotropy and the compliant material behavior causes some

challenges regarding load introductions during testing (Hojo et al., 1994; Mikkelsen and Bech,
2013; Xie and Adams, 1995). Challenges which based on numerical predictions are found to
results in highly non-homogenous stress field which subsequently is found to influence the
measured statically and fatigue strength significantly. An alternative to measuring the macroscopic
strength is by a micro-mechanical determination and understanding of failure mechanisms during
mechanical testing. Studies requiring exporting actually micromechanical structures determined
e.g. by micro x-ray CT scans (Emerson et al., 2015) to micromechanically based finite element
models.
Not only the strength properties are influenced by the non-homogeneous strain and stress fields.
Determining the material stiffness require a precise strain measurement. Due to the price and ease
of using, strain gauges are often used as the deformation measurement testing composites. Strain
gauges are measuring the deformation of the substrate through an electrical resistance change
along an approximately 5 µm thick and 30 µm wide conductive wire made of a 180 GPa stiff
constantan alloy printed as a measurement grid on a 45 µm thick backing polymeric film. Despite
the small thickness of the measuring grid, finite element simulations verify that the stiff grid
material introduce some strain inhomogeneity (Zike and Mikkelsen, 2014). A verification which
explains a difference of 2-5% on the measured composite material stiffness comparing
measurements based on strain gauges with clip-on extensometers or optical methods.
Subsequently, based on an intensive parametric study facilitated through the python scripting
language in the Abaqus CAE, an improved grid design has been proposed (Mikkelsen and Zike,
2014).
2. Strain gauges introduced non-homogeneity

An important material property, designing wind turbine blade with a length of over 80 m, is a high
material stiffness (Brøndsted et al., 2005). Designing this; a precise strain measurement is
essential. Strain gauges are considered one of the most reliable strain measurements devices and is
therefore, do to its low price and ease to use, by far the most common device for strain
measurements. Nevertheless, applied on a moderate stiff composite material with a stiffness in the
range of E ∈ [3;35]GPa , strain gauges is observed to overestimate the material stiffness with 1-
10% error depending on the stiffness and strain gauge grid length (Zike and Mikkelsen, 2014).
In the present study, this is studied on the cases, where a 1.5, 3 or 10 mm strain gauges is mounted
on both side of a 30mm thick, 30 mm wide and 100 mm long specimen. Due to symmetry, only
one quarter of the specimen is modelled. The chosen width and thickness values of the test
coupons are found sufficiently large such that the overall stiffness of the test sample is not affected
by the strain gauges. During the analysis, the elongation of the specimen is prescribed by a
prescribed elongation corresponding to an overall axial strain of 0.25%. Nevertheless, due to the
length, a prescribed force could equally have been chosen. The prescribed elongation, is in
agreement with the test standard, (ISO-527-1, 1997), where the material stiffness of a polymer or a
fiber reinforced polymer should be determined by the slope of the stress versus strain curve in the
strain range ε ∈ [0.05;0.25]% . From the finite element model, all the materials are found
deforming essentially elastically why a linear elastic finite element simulation solved using only
one deformation increment has been found sufficiently.

L = 1.5 mm
grid
30
Lgrid= 3 mm
Measurement error [%]

Lgrid= 10 mm
20
10
0
1 10 100 200
E [GPa]
material
Figure 1. Predicted measurements error on the material stiffness using a 1.5,
3 and 10 mm strain gauges, respectively.
Figure 1, show the finite element predictions of the erroneous of the strain gauges determined
material stiffness as a function of the actually stiffness. Physical strain gauges measure the strain
through a resistance change in the measurement grid. This resistance change are correlated to the
strain in the substrate through a gauge factor found by calibrating the strain gauge mounted on a
steel material. In the finite element model, the strain gauges output is defined by the elongation of
the measurement wire and virtually calibrated on a 200GPa stiff material. Similar to the
experimentally observations in (Zike and Mikkelsen, 2014), the strain is found to significantly
overestimate the material stiffness in the moderate material stiffness range ε ∈ [0.05;0.25]% .
Figure 2, show the strain field in the case of a pure epoxy at an overall straining of the material of
0.25% using a 3 mm strain gauge. Despite a thickness of the measuring grid on only 5 µm, the
non-uniform strain state is found to spread significantly into the measuring grid of the strain
gauges and thereby causing a under prediction of the strains. As a consequence of the
underestimated strains, an overestimate of the material stiffness is found as shown in figure 1.
Thus, the only 5 µm thick but 180 GPa conductive wire in the strain gauges measurement grid is
found to introduce significant strain inhomogeneity in the substrate material resulting in an
erroneous of the experimentally determined material stiffness. An erroneous validated in the
present work through a linear elastic finite element model.

Figure 2. Non-homogeneous strain field applying a 3 mm strain gauges on a 3 GPa
stiff polymer matrix.
3. Gripping introduced non-homogeneity

Uni-directional glass fiber reinforced composites has tensile and compression strength in the range
of 1-2 GPa while the shear and transverse tensile strength in the range of 50-100 MPa. In addition,
the axial stiffness of a uni-directional glass fiber composite material is around 40 GPa while the
clamping material has a stiffness of 200 GPa. Therefore, it is a challenge loading material test
coupons in a way such that the resulting material failure occurs in the gauge section. Especially
during fatigue loading, failure in the gripping regions is often observed. Therefore, the measured
strength and fatigue properties can often only be considered as conservative values making it
difficult comparing different materials and thereby optimizing the material components. Figure 3
show a typical gripping fatigue failure together with a numerical simulation of the corresponding
stress state during loading. The numerical finite element simulation is performed using a 3D linear
elastic model where the interaction between the grip and tabs are simulated by a fictional contact
with the friction coefficient of µ=0.4. Due to symmetry, only one eight of the model has been
modelled. The material test is performed using hydraulic grips, where it is possible independently
to prescribe the clamping and tensile force. Similar to the experimental setup, the clamping force
are just prescribed to be high enough avoiding the test coupons slipping out of the grips. Similar to
the real case, the test sample is loaded up though the grips until the tensile stress in the gauge
section are equal to 1 GPa. At this load-level the finite element simulation show high shear
stresses in the gripping region of the test coupons. Shears stresses approaching 60 MPa and locally

penetrating into the test material and thereby properly causing tab shear failure as the failure
shown on the picture in Figure 3.
Figure 3. Non- homogeneous shear strain state introduced by the gripping of a

tension-compression fatigue test specimen shown together with the observed
failure mechanism in the test coupon.

4. Scripting using the Python interface in Abaqus/CAE
In order to perform a parametric study, the Python scripting language interface to the Abaqus/CAE
pre- and post-processor gives great advances. In the flowing, a generic example is shown
including two Python scripts used by the author performing the parametric studies regarding pre-
and post-processing. Using the scripts it is possible to modify a large variety of parameters
defining geometry, materials, meshing etc. Eventually, the two scripts can be built together
resulting in only one rpt-file for the chosen output values from each studied case.
4.1 Pre-processing
## Default ABAQUS inputs
from part import *
from material import *
from section import *
from assembly import *
from step import *
from interaction import *
from load import *
from mesh import *
from optimization import *
from job import *
from sketch import *
from visualization import *
from connectorBehavior import *
NameModelBase='ModelName'
E1s=(1000,3000,10000,30000,100000,200000)
for i in range(0,6):
## Model in names
NameCAE='Model.cae'
NameModelin=NameModelBase
NameModel=NameModelBase+'-E1-'+str(E1s[i])
print NameModelin
Ncpus=20 # in the case of parallel processes
## Import model from other cae-file

mdb.openAuxMdb(pathName=NameCAE)
mdb.copyAuxMdbModel(fromName=NameModelin, toName=NameModel)
mdb.closeAuxMdb()
## Eventually Modify model

mdb.models[NameModel].materials['Tab'].elastic.setValues(table=((E1s[i],
0.3333), ))
## Generate job
modelJob=mdb.Job(atTime=None, contactPrint=OFF, description='', echoPrint=OFF,
explicitPrecision=SINGLE, getMemoryFromAnalysis=True, historyPrint=OFF,
memory=90, memoryUnits=PERCENTAGE, model=NameModel, modelPrint=OFF,
multiprocessingMode=MPI, name=NameModel, nodalOutputPrecision=SINGLE,
numCpus=Ncpus, numDomains=Ncpus, numGPUs=0, queue=None, scratch='',
type=ANALYSIS, userSubroutine='', waitHours=0, waitMinutes=0)
## The following lines submit the job, and wait for completion before continue
modelJob.submit(consistencyChecking=OFF)
modelJob.waitForCompletion()

4.2 Post-processing
## Default ABAQUS inputs
from part import *
from material import *
from section import *
from assembly import *
from step import *
from interaction import *
from load import *
from mesh import *
from optimization import *
from job import *
from sketch import *
from visualization import *
from connectorBehavior import *
from abaqusConstants import *
from viewerModules import *
from driverUtils import executeOnCaeStartup
# Model in names
NameModelBase='ModelName'
E1s=(1000,3000,10000,30000,100000,200000)
EpsX=0.3
for i in range(0,6):
NameModel=NameModelBase+'-E1-'+str(E1s[i])
ODBfile=NameModel+'.odb'
RPTfile=NameModel+'.rpt'
## Save parameters from simulation

session.viewports['Viewport: 1'].setValues(displayedObject=
session.openOdb(name=ODBfile))
odb = session.odbs[ODBfile]
# Make path
session.Path(name='Path-1', type=POINT_LIST, expression=
((0.0+EpsX, 0.0, 0.0), (1.0+EpsX,-1.0, 0.0)))
session.Path(name='Path-2', type=POINT_LIST, expression=
((0.0+EpsX, 0.0, 0.0), (0.0+EpsX,-1.0, 0.0)))
# Extracting data from path

session.viewports['Viewport: 1'].odbDisplay.setPrimaryVariable(
variableLabel='S', outputPosition=INTEGRATION_POINT, refinement=(COMPONENT,
'S11'))
pth = session.paths['Path-1']
session.XYDataFromPath(name='XYData-1', path=pth, includeIntersections=True,
projectOntoMesh=False, pathStyle=PATH_POINTS, numIntervals=10,
projectionTolerance=0, shape=UNDEFORMED, labelType=TRUE_DISTANCE)
pth = session.paths['Path-2']
session.XYDataFromPath(name='XYData-2', path=pth, includeIntersections=True,
projectOntoMesh=False, pathStyle=PATH_POINTS, numIntervals=10,
projectionTolerance=0, shape=UNDEFORMED, labelType=TRUE_DISTANCE)
x0 = session.xyDataObjects['XYData-1']
x1 = session.xyDataObjects['XYData-2']
session.writeXYReport(fileName=RPTfile+'1', appendMode=OFF, xyData=(x0))
session.writeXYReport(fileName=RPTfile+'2', appendMode=OFF, xyData=(x1))
del session.xyDataObjects['XYData-1']
del session.xyDataObjects['XYData-2']
del session.paths['Path-1']
del session.paths['Path-2']

5. Improved experimental procedures extracted from finite elements
studies
After a finite element model has been validated against experimental measurements, the model can
be used suggesting new improved measurements strategies. Based on experimental validation of
the finite element model used for the predictions presented in chapter 2 and 3, an extensive
numerical study has been performed using the Python interface shown in chapter 4. Based on
those simulations, improved experimental strategies have been suggested lowering the effect of
the non-homogeneity of the strain and stress fields on the experimental measurements. Those
improved test strategies will be presented below.
5.1 Strain-gauges
20
Reference
New design
Measurement error [%]
± 1% error
10
-5
1 10 100 200
E [GPa]
material
Figure 4. Improved strain gauges design of the 3mm strain gauge case
The lower of the two grid pattern in Figure 4, show an improved strain gauge design resulting in a
significant lower erroneous of the stiffness prediction for a large range of material stiffness. A
strain gauges design which has been presented in the patent application (Mikkelsen and Zike,
2014). Using a strain gauge grid pattern which includes reinforcement parts at the end of the strain
gauges it is possible to move the non-homogeneous strain field outside the measuring part of the
strain gauges as illustrated in Figure 5. The improved strain gauges design has not yet been
validated experimentally.

Figure 5. Moving the non-homogenous strain field outside the measuring grid.
5.2 Tensile testing

A way to decrease the shear stress level in the gripping region in a tensile test coupon is by using
a waisted shape as illustrated on the upper right figure in figure in An improved waisted shape of
the tension-compression test coupon is presented in the two top figure in Figure 6. In figure 6, a
comparison of tab shear stresses (left figures) and the splitting shear stresses (right figures) are
shown for the waisted (upper figures) and the rectangular (lower figures) test coupons geometries.
Comparing the stress states, it can be seen that the tab shear stresses are lowered significant but
with the drawback that there will be some splitting shear stresses. The optimal design will make a
good compromise between these two shear stresses depending on the actually shear strength of the
test coupons in the two loading directions. No matter what, the waisted test geometry has
experimentally been found to give improved measured material strength compared with the
rectangular ISO specimen.

Waisted shape
Rectangular shape
(a) Tab shear side view (b) Splitting shear top view
Figure 6. Comparing the shear stresses in waisted and a rectangular tensile-
compression test coupon.
6. Conclusion
Finite element simulations have been used validating and improving the material test procedure
regarding stiffness and strength/fatigue determination of composite materials. The work is related
testing composite materials for wind turbine blade. An application where a low material price and
weight together with high material stiffness and fatigue properties are essential designing larger
wind turbines with an resulting lower cost of energy. During the study, an improved strain gauge
grid design and a test coupon shape has been proposed.
7. Acknowledgement
This research was supported by the Danish Centre for Composite Structure and Materials
for Wind Turbines (DCCSM), grant no. 09-067212, from the Danish Strategic Research
Council (DSF).
8. References
1. Brøndsted, P., Lilholt, H., Lystrup, A., 2005. Composite materials for wind power turbine
blades. Annu. Rev. Mater. Res. 35, 505–538.
2. Emerson, M.J., Jespersen, K.M., Jørgensen, P.S., Larsen, R., Dahl, A.B., 2015. Dictionary
based segmentation in volumes, in: 19th Scandinavian Conference on Image Analysis,. pp. 1–
12.
3. Hojo, M., Sawada, Y., Miyairi, H., 1994. Influence of clamping method on tensile properties
of unidirectional CFRP in 0° and 90° directions — round robin activity for international
standardization in Japan. Composites 25, 786–796.

4. ISO-527-1, 1997. 527-1: Plastics—Determination of tensile properties—Part 1: General
principles.
5. Mikkelsen, L.P., Bech, J.I., 2013. Secondary stress effects during load introduction into
unidirectional composite test coupons, in: Thomsen, O.T., Sørensen, B.F., Berggreen, C.
(Eds.), 6th Int. Conf. on Composites Testing and Model Identification. Aalborg.
6. Mikkelsen, L.P., Zike, S., 2014. A strain gauge. Patent Application: EP14180022.7.
7. Xie, M., Adams, D.F., 1995. Effect of Specimen Tab Configuration on Compression Testing
of Composite Materials. Compos. Technol. Res. 17, 77–83.
8. Zike, S., Mikkelsen, L.P., 2014. Correction of Gauge Factor for Strain Gauges Used in
Polymer Composite Testing. Exp. Mech. 54, 393–403.

Advanced Fracture Modeling for Cuttings Re-injection
Jorge Garzon, Matias Zielonka, Kevin H. Searles and Pablo Sanz
ExxonMobil Upstream Research Company
22777 Springwoods Village Parkway, Houston, TX 77389
Abstract: An environmentally friendly and cost-effective method for waste-disposal operations

commonly used worldwide in the oil and gas industry is known as Cuttings Re-Injection (CRI). It
consists of injecting drill cuttings into a suitable geological formation through one or more
hydraulically-induced fractures. In this paper, some case studies assessing the risks associated
with CRI are introduced using recently co-developed capabilities in Abaqus Standard for
simulating fully-coupled hydraulic fracturing. Finite element models (FEM) are constructed using
material properties and in situ stress state based on well log information. Furthermore, models
are calibrated using data from actual field injection tests.
This new 3D simulation capability in Abaqus Standard overcomes some of the limitations of
traditional tools and gives engineers an opportunity to assess the impact of planned CRI
operations on predicted fracture geometry and pore pressure changes. These variables are very
important in understanding risk to formation top seal integrity and possible interaction with pre-
existing discontinuities, both of which may lead to loss of fluid containment and non-compliance
with environmental regulations.
Keywords: Cuttings re-injection, CRI, Hydraulic fracturing, Pore pressure, Injection,

Geomechanics
1. Introduction
In the late 1980s, the industry began large scale subterranean injection of slurrified drill cuttings in
Alaska and the Gulf of Mexico [1]. At that time, the process of injecting was done open hole and
most well-related wastes that could be rendered into slurry were injected for disposal. Other kinds
of materials produced in the drilling process were considered as useful construction materials
(sand and gravel). Currently, governmental regulations concerning injection are taken into account
although they may vary from country to country. In general, disposal of material wastes related to
drilling, completion and production operations such as oily cuttings, waste drilling fluid, oily

water, waste brines and other liquids, are covered by governmental regulation and are candidates
for downhole injection. Some other wastes associated with well operations (e.g. waste fuels, rig
debris) classified as industrial wastes may not be injected in some jurisdictions.
There are different approaches to managing drill cuttings waste. For example, common methods
used offshore include cuttings dehydration, thermal or mechanical treatment with onsite cuttings
processing, or shipping to shore for processing. Each method has its application under different
circumstances, but downhole injection can be the preferred method when feasible and is
appropriate due to several advantages:
1) No transportation risks; it does not rely on shipment of wastes, spill contamination
and weather related risks offshore
2) No future clean-up liabilities; when properly injected, waste is permanently
immobilized in subterranean formations with little risk of water contamination
3) No limit to location; this type of treatment can be performed on-shore and offshore
4) Favorable economics; costs can become lower in the right environment.
Cuttings Re-Injection (CRI) is a common practice in many regions including, North Sea, Gulf of
Mexico, Alaska, offshore California, Sakhalin Island and Eastern Canada. An ideal CRI project
goes through a well-planned and integrated process to guarantee adequate barriers to prevent
potential hazards (e.g., surface broaching). A detailed study (pre-screening and during any
injection program) is necessary to determine whether the geological setting is suitable to prevent
escape of waste. Some of the essential details needed include fracture size versus injection
volume and pressure and these can be obtained more accurately using advanced simulations. The
successful progress of an injection project can be evaluated with daily pressure monitoring and
model calibration (updated simulations). Guo et al. [2] described a successful example of
integrating all these components to estimate storage capacity and assess the operational integrity
of disposal wells.
CRI modeling requires a pseudo three-dimensional or, even better, a three-dimensional fully-
coupled hydraulic fracturing simulator to predict the geometry of injection induced fractures and
evaluate the risk of a broaching situation. Some developed software called MFrac commonly used
in industry [3]. Although this pseudo three-dimensional software is capable of modeling CRI, it
has these limitations: simulation of a single fracture, fixed number of batches, fracture geometry
fixed to an ellipse, ignoring stress buildup due to solids accumulation and no information about
pore pressure changes inside the rock. Despite these limitations, this software has proven useful to
estimate conservative bounds of the disposal domain [4-6]. The general consensus in the industry
is that multiple parallel fractures grow due to the injected cuttings and it is conservative to assume
that all the volume will propagate one long fracture. ExxonMobil Upstream Research Company
(URC) is working with SIMULIA® to overcome some of these limitations with recently co-
developed, three-dimensional, fully-coupled hydraulic fracturing capabilities for Abaqus Standard.
In this paper, two case studies assessing the risk associated with CRI are introduced using the co-
developed capabilities. All finite element (FE) models are constructed using material properties

and in situ stress state based on well log information. In addition, models are calibrated using data
from injection tests performed in the field.
Results from these simulations give an estimate not only of fracture geometry but also pore
pressures changes along the fracture and within the surrounding rock. This information is
essential to minimize the risk associated with breaching top seals and possible fracture
intersections or reactivation of any neighboring faults. Engineers need to ensure fluid containment
and compliance with environmental regulations.
2. Model set-up: Material Properties and FE Mesh
Finite element analysis (FEA) has been used for more than 40 years to simulate the mechanical
behavior of materials. It is a numerical technique for finding an approximate solution and it
consists of partitioning a problem into simpler parts, called elements. Theoretically speaking, if
these elements are infinitely small the approximation reaches the true solution of the problem, but
in practice the size of the elements depends on the computational power and simulation time.
Therefore, the numbers of elements and their size have to be such that a good approximation of the
solution can be obtained in a reasonable time.
CRI problems tend to have very large domains extending over 1 km in all directions, and good
approximations for fracture problems require very small elements (less than 0.1% of the crack
size). This makes the creation of these types of models very challenging. In the following models,
multi-point constraints are used to refine close to the fracture plane and maintain continuity
between elements.
Material properties and initial conditions come from a rock physics model (RPM) created based
on well log data (e.g. sonic, gamma ray, density, etc.). Some estimates of both include Young’s
modulus, Poisson’s ratio, porosity, permeability, fracture toughness, pore pressure, horizontal and
vertical initial stress, etc. The resolution of the well log data is much higher than the discretization
of the FEA model and thus, zoning and up-scaling methods are necessary to accurately assign
representative values to each element in the model. In this section the material properties and
initial conditions for two case studies are presented.
2.1 Case Study 1

In this model, a rock formation with dimensions 800x800x900 meters, as shown in Figure 2.1, is
simulated. The FE mesh created is only for half of the problem since a symmetry plane is
considered. The injection location is at 2,370 m TVD. The model consists of 641,000 hexahedral
elements with the smallest element having a size of 1.2 m.
The model has been discretized by 46 material layers with a higher resolution close to the injection
location. After creating the RPM based on the well log information, all material properties and
initial conditions have been up-scaled to the FE material layers as shown in Figure 2.2. The
colored rectangles correspond to the largest element size on the FE layer, the dark blue line is the

data at the well log resolution and the yellow line corresponds to the up-scaled data. Notice that a
constant value of the data is assigned to each FE layer.
Figure 2.1. Case Study 1: Finite Element Mesh. Symmetric model with 46 different
material layers for CRI injection.
Figure 2.2. Case Study 1: Material properties and initial conditions up-scaled to 46
finite element layers.
Young’s modulus, Poison’s ratio, initial stress and pore pressures are inserted directly into the
input file, permeability is used to compute hydraulic conductivity which is the value required for

the soils analysis. Additional material information needed for this type of model is the fluid
viscosity.
2.2 Case Study 2

The dimensions of this problem are 1,900x1,055x1,300 meters as shown in Figure 2.3. Again,
due to symmetry only half of the problem is simulated. The injection location is at 1,128 m TVD.
The model consists of 221,000 hexahedral elements with the smallest element having a size of 2.7
m.
Figure 2.3. Case Study 2: Finite Element Mesh. Symmetric model with 26 different
material layers for CRI injection.
This model has 31 material layers. A RPM was created based on the well log information and all
material properties and initial conditions have been up-scaled to the FE material layers as shown
in Figure 2.4. Notice that the first layer in this model is very large compared to other layers.
Since well log data was not available for that interval, constant values were assumed.
Permeability in this problem is not well-known but the calibration exercise explained in the next
section allows for adjusting to a value representative of the field conditions.
As done in previous example, Young’s modulus, Poison’s ratio, initial stress and pore pressures
are inserted directly into the input file. Permeability and fluid viscosity are used to compute
hydraulic conductivity. The CRI fluid considered has a viscosity expressed by Power-law
constants of n=0.53 and k=0.028 lbs secn/ft2.
In this section, the creation of two finite element models was presented. Material properties and
initial conditions prescribed at each element are derived from up-scaled rock physics models based

on well log data. To ensure a correct representation of the field conditions, the FE models are
calibrated using data from injection tests. This calibration process is described in the following
section.
Figure 2.4. Case Study 2: Material properties and initial conditions up-scaled to 31
finite element layers.
3. Model Calibration
Rock physics models (RPM) were created for two study cases to obtain material properties and
initial conditions. Although these RPMs have high resolution and tend to be a very accurate
representation of the field conditions, the coarsening and up-scaling process may deteriorate the
precision of the input data. Therefore, a calibration of the FE model is necessary to assure a good
representation of the problem. In this section, calibration is done by simulating an actual injection
test carried out in the field. Measured pressure data is recorded and used for comparison.
3.1 Case Study 1

In this example, an injection test of 216 minutes is simulated using the injection rate schedule
provided from field operations as input. All other material properties and initial conditions remain
as explained in the previous section, the injection fluid is sea water. The injection test consists of
a ramped-up injection interval, a shut-in interval and a constant injection period.
The simulated injection pressure results are shown in Figure 3.1. The gray line is the injection rate
schedule (input), the red line corresponds to pressure measured at the field and the green line is
inlet pressure output from the FE simulation. The field data is measured at surface conditions, but
it has been adjusted to reflect flowing bottom-hole conditions by considering hydrostatic pressure

and frictional losses. This allows for a more direct comparison with the FE simulation output.
The location of the pressure output in the FE simulation is taken at the injection point inside the
fracture (depth 2,370 m TVD). Additionally, snapshots with a contour representing the fracture
opening (units are meters) at specific injection times are shown.
In general, FE results show a very good match to the pressure measured at the field. A small
difference is noticed at the time interval 30-35 min. This difference is due to the time that it takes
to completely fill up the pipe with the injection fluid. This is not simulated at the moment and it is
assumed that the pipe is 100% filled before injection starts. Therefore, the breakdown pressure for
the FE simulation is reached at an earlier time.
Figure 3.1. Case Study 1: Pressure results from the injection test and fracture
opening at 3 different injection times.
At approximately 75 minutes after fracture closure (shut-in), the rate of inlet pressure decay
becomes dependent on formation permeability (transition from bi-linear to radial flow). Here, the
match between measured and simulated pressures suggests an accurate representation of rock
permeability used in the model. This type of post-closure response cannot be obtained in a pseudo
three-dimensional simulator, such as MFrac for example. After the fracture closes, this type of
simulator cannot perform computations to consider the pore pressure diffusion into the
surrounding rock formation. In addition, the FE results demonstrate a fairly good match with
measured propagation pressure. This provides further assurance that estimation of the initial
conditions considered in the model are reasonable.

For the aforementioned calibration example, no adjustments of material properties or initial
conditions were needed to obtain a good match to the field behavior. This injection test
simulation guarantees that the FE model has a good representation of the field characteristics and
can serve as a good approximation of the true behavior.
3.2 Case Study 2

In the previous case study, the measured surface injection pressure was corrected to flowing
bottom hole conditions by considering hydrostatic pressure and frictional losses. This was done to
facilitate comparing measured pressure to simulated inlet pressure at the modeled injection
location. In this case study, pipe elements are used to simulate Bernoulli fluid flow from the
surface to the injection location. This allows making a straight-forward comparison to the
pressure results at the surface without having to perform any manual computations. Figure 3.2
shows the FE model with the pipe elements. The contour corresponds to initial pressure
(hydrostatic). Note that pressure is zero at the surface (0.0 m TVD).
The injection test in this example consists of 50 minutes of injection duration at 4 bpm and 50
minutes of shut-in time. All material properties and initial conditions are the same as explained in
the previous section. A sensitivity study is also conducted for permeability since it is an unknown
in the problem. The injection fluid is assumed to be seawater.
For this particular example, the actual field injection test was carried out following 300 days of
injection, thus a fracture should have existed by the time of the test. To simulate that current state
of the rock formation, a set of cohesive elements close to the injection location was initialized to
represent existing damage. These elements are colored red in Figure 3.2. This initial damage may
not be representative of the actual fracture in the field, but it is large enough to dissipate all fluid
injected during this short injection test. The key for this simulation is to re-open a preexisting
fracture without inducing additional fracture propagation.
Results of this simulation are shown in Figure 3.3. The gray line is the injection rate schedule
(input), the red line corresponds to the measured surface pressure, the green and blue lines are the
inlet pressure outputs from the FE simulation for a 50 mD and 30 mD rock permeability
assumption, respectively. Both FE simulation pressure results are output at the pipe inlet (0.0m
TVD). These results can be compared directly to the field data measured at the surface, because
the pipe elements solve for hydrostatic pressure and frictional losses during the simulation. Also
shown are snapshots of a contour representing the fracture opening (units are meters) at specific
injection times for the 30 mD assumption.

Figure 3.2. Case Study 2: Initial pressure contour and description of pipe elements
and trajectory to injection location.
In comparison to the results from the prior case study, the FE results for this case are not as
accurate. This is mainly due to the inadequate discretization of the rock permeability. Although
this important rock characteristic is unknown, the results suggest that an assumption between 30 to
50 mD is sufficient to obtain a similar behavior as observed in the field.
The difficulty in this particular case study stems from the lack of information regarding rock
permeability. However, this calibration exercise is still very useful in finding discrete values that
seem to be representative of the field behavior. This helps to assure that the FE model is a good
representation of the field characteristic and serves as a good approximation of the true behavior.
4. Results and Conclusions

A calibrated fully-coupled FE model with reliable material properties and initial conditions is a
useful tool to investigate different scenarios for CRI. The FE model can predict crack surface
geometry, pressure changes at the fracture inlet and pressure diffusion into the surrounding rock.
This helps to assess the risk of compromising the integrity of the formation top seal and possibly
the reactivation and intersection of any pre-existing discontinuities (fracture and faults), both of
which may lead to loss of fluid containment and non-compliance with environmental regulations.
In this section, results for a CRI simulation are presented for the two study cases calibrated in the
previous section. A 100 day-injection duration is carried out for both examples. Figure 4.1 shows
the injection rate schedule which consists of 17 cycles of single-day injection and 5 days of shut-in
time.

Figure 3.3. Case Study 2: Pressure results from injection test and fracture opening
at 2 different injection times (30 mD assumption); units are meters.
Figure 4.1. Injection rate schedule used for CRI simulation.

4.1 Case Study 1
The material properties and initial conditions remain unchanged from the previous section. The
injection fluid is assumed to be sea water to facilitate a comparison with Case Study 2.
Simulation results show that the crack opens and propagates during injection cycles and closes
completely during the shut-in period. Figure 4.2 shows the evolution of the crack surface at
different days and contours correspond to the fracture opening width. The bottom right figure
illustrates the fracture damage variable after 97 days of injection. This damage variable represents
the rock that was damaged or fractured during the injection but may not be currently open. In this
example, most of the propagation of the fracture occurs during the early cycles, while re-opening
of the preexisting fracture dominates the later part of the injection.
After 100 days of sea water injection, the predicted fracture length is 52 m and the height is 34 m.
Another important insight obtained from this FE simulation is the pore pressure changes inside the
fracture and in the formation away from the fracture. Figure 4.3 shows the pore pressure evolution
for two locations: i) at the injection point (2,370 m TVD) and ii) 70 m perpendicular to the
fracture plane (inside the rock). The increase in pore pressure inside the fracture is substantial
during injection (80 MPa) and very similar after the first cycle. The decrease of pore pressure also
occurs rapidly; the pressure drops about 25 MPa in less than 2 days after shut-in. As expected, the
variations on the pore pressure are less significant away from the fracture. Figure 4.3 shows that
the pore pressure 70 m from the injection location experience minor changes.
Figure 4.4 shows two contour plots for pore pressure change in terms of percentage from its initial
value after 97 days of injection. Contour a) shows pore pressure changes on an isometric view of
the problem, notice that fracture propagates in the YZ plane. Pore pressure changes are more
significant close to the fracture and thus more significant pore pressure changes are observed on
the fracture plane. Contour b) shows a XZ cut-in plane at the injection point (2,370 m TVD).
This gives a better view of what happens inside the rock away from the fracture. The pore pressure
change at a radius of ~65 m around the fracture is insignificant.
4.2 Case Study 2

In this example, material properties and initial conditions remain the same as explained in
previous sections. In order to capture the contrast in permeability between element layers
representing sand and shale, permeability of 500 mD and 5 mD are assumed as upper and lower
bounds, respectively. The fluid injected is simulated as a non-Newtonian slurry with viscosity
given by the Power-law constants n=0.53 and k=0.028 lbs secn/ft2.
Similarly to previous case, the crack opens and propagates during injection cycles and closes
completely during the shut-in period. Figure 4.5 shows the complex evolution of the fracture
geometry during injection. Each injection cycle shows a different behavior. Locations that were

fractured in a previous cycle may not be re-opened in the current steps; instead fracture
propagation may occur elsewhere. Thus, fracture damage is a good indicator of the different paths
the fluid took during fracturing. In this example, fracture propagation is more significant in length
than height when compared to previous case. Even though it propagates upwards, there are
barriers that do not allow an important propagation towards the surface.
Figure 4.2. Case Study 1: Crack surface evolution, fracture opening and final
damage state.
Figure 4.3. Case Study 1: Pore pressure evolution at injection location (red) and at
70 m from injection location (blue).

Figure 4.4. Case Study 1: Pore pressure changes (%) after 97 days of injection (just
after last injection cycle); a) Isometric view, b) Cut-in plane at depth 2,370 m TVD.
The fracture propagates on the YZ plane.
Figure 4.6 shows the pore pressure evolution at the injection point (1,128 TVD) and 90 m away
from the fracture plane. Larger time steps were utilized in this simulation to save computer
memory. For this reason, the pore pressure evolution curves are not as smooth as in the previous
example.
The increase and decrease of pore pressure is significant close to the fracture. As you move away
from the fracture, the magnitude and fluctuation of the pore pressure rapidly decreases. Figure 4.7
shows two contour plots for pore pressure change in terms of percentage from its initial value after
97 days of injection. Contour a) shows on an isometric view that pore pressure changes are more
significant near the fracture. Contour b) shows a XZ cut-in plane at the injection point (1,128 m
TVD). The pore pressure change is insignificant at a radius of ~300 m around the fracture.
4.3 Conclusions
The size of the fracture in Case Study 1 is small compared to the fracture in Case Study 2. This is
mainly driven by the differences in the rock permeability and fluid viscosity. While sea water is
injected in the first case, viscous slurry is injected in the second example. Hydraulic conductivity
is the rate of flow through a cross-sectional area, and is a function of permeability, fluid density
and viscosity. The hydraulic conductivity of case 1 is about 2 orders of magnitude greater than in
case 2. This means that fluid can travel quicker in the first case. This can also be seen in the pore
pressure behavior, in case 1 pore pressure changes occur rapidly and it stabilizes in a few days.

Another observation is that pore pressure changes are greater in case 1 because the injection point
is at a deeper location, thus a higher compression has to be overcome to open a fracture.
Figure 4.5. Case Study 2: Crack surface evolution, fracture opening and final
damage state.
Figure 4.6. Case Study 2; Pore Pressure evolution at injection location (red) and at
90 m from injection location (blue).

Figure 4.7. Case Study 2: Contours showing pore pressure changes after 97 days
of injection (just after last injection cycle); a) Isometric view, b) Cut-in plane at
depth 1,128 m TVD.
Pore pressure and fracture geometry have been predicted for two CRI examples. Height and
length are important insights that assure fluid containment. Also contours of pore pressure change
have been presented to show the extent of perturbation of the formation rock. This can be used as
an overview of the trajectory of the injected fluid and amount of perturbation in terms of pressure.
If any important geologic features exist (e.g. pre-existing faults) within the region of interest, more
complex models can be created to understand the consequence of perturbing the pressures and
stresses at those particular features. This is future work and it is beyond the extent of this paper.
In this study filter cake buildup has not been considered. Filter cake is understood as a barrier
between the fracture and the rock formation, that reduces the rate (leak-off) at which the fracturing
fluid flows. This can be easily added to the CRI simulation by means of a user subroutine.
None of the current semi-analytical fracture modeling tools predicts injection-induced hydraulic
fracturing with a fully coupled 3D formulation. Enhanced fracture modeling capabilities are
needed to address increasingly complex business questions. More accurate models would lead to
optimized well designs, and would assist in acquiring permits for cuttings disposal operations
from regulatory bodies and further reduce risk of environmental impact.
5. Acknowledgments
The authors would like to thank ExxonMobil Upstream Research Company for permission to
publish this work.

6. Bibliography
1. Guo, Q., Abou-Sayed, A. S., and Egel, H. R., "Feeling the Pulse of Drill Cuttings
Injection Wells - A Case Study of Simulation, Monitoring and Verification in Alaska".
SPE Journal, 2007, 458-467.
2. Hofmann, Hannes, Babadagli and Zimmermann, G. “Hot water generation for oil sands
processing from enhanced geothermal systems: Process simulation for different hydraulic
fracturing scenarios”. Applied Energy, 113, 2014, 524-547.
3. Malachosky, E. and Lantero, DM., “Disposal and Reclamation of Drilling Waste”, U.S.
Patent No. 4,942,929 July 4, 1990.
4. Meyer, B.R. "Three-Dimensional Hydraulic Fracturing Simulation on Personal
Computers: Theory and Comparison Studies". SPE 19329, 1989, 1-18.
5. van der Hoorn, K., Heijnen, L.J., Gankema, M.E. and Nitters, G., “Hydraulic Fracturing
In Limestone: A Case Study of Two EGS Projects in The Netherlands",
PROCEEDINGS, Thirty-Seventh Workshop on Geothermal Reservoir Engineering,
January 30, 2012
6. Warpinski, N.R., Moschovidis, Z.A, Parker, C.D., and Abou-Sayed, I.S., "Comparison
Study of Hydraulic Fracturing Models - Test Case: GRI Staged Field Experiment No. 3".
SPE Production & Facilities, 1994. 7-18.

Experimental Validation of Simulation Capabilities
for Hydraulic Fractures Propagating in a Porous
Medium
Jing Ning, Gilbert Kao, Nikolay M. Kostov, Kevin H. Searles, Scott R. Buechler
and Pablo F. Sanz
22777 Springwoods Village Parkway, Houston, TX 77389
Abstract: The problem of a fluid-driven (hydraulic) fracture propagating in a porous medium is

studied using recently co-developed capabilities in SIMULIA® Abaqus Standard (“Abaqus”) for
fully-coupled simulation of hydraulic fracturing. The hydraulic fracturing simulation capabilities
are validated against lab-scale experimental results. A Polyaxial Test Cell (PTC) is set up for
controlled hydraulic fracturing in fabricated rock samples allowing different testing conditions of
confining stress, injection rate, and rock and fluid properties. Two representative regimes of
fracture propagation (toughness/storage and viscosity/leak-off) are selected to compare against
numerical simulations carried out in Abaqus. Fracture propagation and fracturing fluid flow are
modeled using coupled pressure/deformation cohesive elements, while the porous medium
deformation and pore-fluid flow in the rock are modeled with coupled “pore-
pressure/deformation” continuum finite elements. Injection pressure response and fracture and
leak-off geometries agree well with corresponding experimental results. The experimental
validation of the newly co-developed hydraulic fracturing simulation capabilities within Abaqus
provides confidence for accurately modeling field-scale, fluid-driven fracturing applications
including drill cuttings and water injection, stimulation of unconventional reservoirs, lost returns
and other drilling operations.
Keywords: Geomechanics, fluid-driven fracturing, pore pressure, leak-off, cohesive elements
1. Introduction
Hydraulic fracturing has significant application for the oil and gas industry and has been a very
active research area over the past several decades. From a simulation viewpoint, there are
substantial technical challenges in coupling the physics of fluid flow, poromechanics, and fracture
mechanics. Solving hydraulic fracturing problems not only requires a deep fundamental
understanding of the physical mechanisms, but also the ability to study these problems at field-
representative time and length scales. Many attempts have been made to address the problem
using different numerical approaches (Ortiz and Pandolfi, 1999; Detournay et al., 2006; Garagash
2006). However, most are based on research codes that are impractical for representing actual
reservoir conditions and often lack the computational capabilities (e.g., cluster parallel computing)
required to solve field-scale problems (Dahi-Taleghani and Olson 2011, Abbas et. al 2014).
Traditional industry tools routinely make assumptions and simplify the physics in order to model

the problem at applicable scales, often providing unrealistic predictions and non-physical behavior
(Clearly 1980; Meyer 1989; Warpinski et al., 1994). To address this challenge, fully-coupled
hydraulic fracture simulation capabilities were co-developed between ExxonMobil Upstream
Research Company and Dassault Systems Simulia Corporation. These capabilities leverage the
existing Abaqus non-linear soils consolidation analysis procedure, and incorporate two
formulations for fracture modeling, i.e., cohesive elements (CZM) and the extended finite element
method (XFEM).
Zielonka et al. (2014) described the aforementioned formulations in great detail and provided
validation against approximate analytical solutions defining a parametric space of four limiting
propagation regimes for hydraulically driven fractures (Figure 1). These semi-analytical solutions
exist in the form of regular asymptotic expansions (Bunger et al., 2005; Detournay et al., 2006;
Garagash, 2006; Hu and Garagash, 2010; Garagash et al., 2011; Peirce and Detournay, 2008;
Savitski and Detournay, 2002) and result from assuming: (i) an infinite domain, (ii) material fully
impermeable, (iii), material linear elastic, (iv) linear-elastic fracture mechanics, and (v) Carter’s
leakoff model (Howard et al., 1957; Charlez 1997). Good agreement between the Abaqus and
asymptotic analytical solutions served as a first pass of validation for successful development of
fully coupled simulation capabilities (example shown in Figure 2).
Figure 1. The parametric space representing the four limiting propagation regimes of hydraulically
driven fractures (Garagash et al., 2011; Zielonka et al., 2014).

2a. 𝑝(𝑟 = 0.01, 𝑡) 2b. 𝑝(𝑟, 𝑡 = 𝑡𝑚𝑎𝑥 ) 2c. 𝑝(𝑟 = 0.01, 𝑡) 2d. 𝑤(𝑟, 𝑡𝑚𝑎𝑥 )
Figure 2. Validation examples for the M-vertex propagation regime (after Zielonka et al., 2014).
In the present work, we extend validation of the co-developed simulation capabilities to the
laboratory scale. Results from an in-depth experimental program are presented to demonstrate
good test repeatability and provide insight across a broad range of rock and fluid properties, and
in-situ stress conditions. The high quality of test results provide a suite of benchmark experimental
tests for model validation. The modeling approach using Abaqus is presented along with a
comparison between the numerical simulations and physical experiments for two representative
propagation regimes.
2. Experiments
Unique experimental capabilities have been developed in-house based on a Polyaxial Test Cell
(PTC) apparatus for controlled hydraulic fracturing in rocks. The experimental test data helps to
better understand and characterize the fundamental physical mechanisms of hydraulic fracture
initiation and propagation. Three-dimensional loads can be applied to represent the in-situ stress
conditions that reservoir rocks may experience. The computer controlled hydraulic system can
apply up to 1 million lbs of force and 20,000 psi of injection pressure. The testing workflow
begins with internally standardized procedures for fabricating large-scale hydrostone samples
having dimensions of 12 in. x 12 in. x 18 in. (see Figure 3a). A through-drilled or cast-in-place
borehole is then created in the hydrostone with a 2 in. section in the center representing an “open-
hole completion” (Figure 3b). Dry samples are loaded into the polyaxial load platens and viscous
fluid with colored dye is then injected into the hydrostone through the wellbore in a controlled
manner. Pressure responses and acoustic emissions are collected in real-time for data analysis
purposes. The post-test analysis (Figure 3d) enables the ability to compare numerical results
(pressure response and fracture/leak-off geometry) to experimental data by reconstructing the 3-D
fracture geometry from thinly sliced hydrostone sections.
A suite of parametric experimental studies covering a wide range of rock and fluid properties, as
well as in-situ stresses, was carried out as a benchmark. These test results demonstrate good
repeatability and provide insight into field-scale applications where wellhead pressure is often the
only measured data available. Two test cases representing fracturing in tight rock (referred to here
as the “no leak-off” case) and permeable reservoir rock (referred to here as the “leak-off” case) are
chosen for experimental validation of the Abaqus simulation capabilities.

Figure 3a. Rock Fabrication. Figure 3b. Borehole Creation.
Figure 3c. Sample Loading and Fracturing. Figure 3d. Post-test Analysis.
3. Model
The representative model used for simulating the PTC experiments is shown in Figure 4a. A 3-D
model is used for simulating the “no leak-off” case while a 2-D model is used for simulating the
“leak-off” case. This is due to the increased computational cost associated with the added
complexity of capillary pressure modeling described in greater detail below. A half-symmetry
assumption is used for the finite element models, and in-situ stresses of 1000 psi, 500 psi, and
1000 psi are applied to the surfaces (in x, y, z, orientation). The 3-D fracture is expected to
propagate in the X-Z plane with the fracture opening against the minimum stress direction (SY,
Figure 4a). The 2-D fracture model is based on a plane strain assumption by modeling only a
small section in the Z direction (highlighted by the dashed blue line in Figure 4a). Fluid is injected
into the borehole and hydrostone through pipe elements in both the 2-D and 3-D models.
Fracture propagation and fracturing fluid flow are modeled using coupled pressure/deformation
cohesive elements in Abaqus, while the porous medium deformation and pore-fluid flow in the
surrounding formation are modeled with coupled “pore-pressure/deformation” continuum finite
elements (Abaqus User’s Manual Version 6.14). Pipe elements are used to connect the fluid flow

from the injection inlet to the hydrostone. The overall number of elements is 12,349 and 275,544
for the 2-D and 3-D models, respectively (see Figure 4b).
The input parameters for the fluid and hydrostone properties, including Young’s modulus and
fracture toughness, are derived from measurements made directly on hydrostone core plugs taken
from the larger test samples during the fabrication process. The intent of this modeling approach is
to validate the newly co-developed numerical capabilities in Abaqus by incorporating physically
measured input parameters and representing the full physics, and not simply “tuning” the models
to achieve desired results.
𝑆𝑌
𝑺𝑿 𝑆𝑌
3D to 2D 𝑺𝑿
𝑆𝑍
Figure 4a. 3-D Model Schematic. Figure 4b. 2-D Model Schematic.
Figure 4c. 3-D Mesh. Figure 4d. 2-D Mesh.

4. Results
The PTC results for the fracture geometry and fluid leak-off profile in the hydrostone vary
depending on the specific testing conditions (e.g., fracturing fluid viscosity and formation
permeability). The first validation, completed on a Super-X hydrostone sample (permeability of
~0.28 mD), fractured with 13 mL of dyed, high viscosity fluid (35,000 cSt) at an injection rate of
2.5 ml/min. The resulting fracture profile is considered the “low leak-off” case in which most of
the injected fluid drives fracture propagation instead of leaking off into the surrounding formation
through the fracture surface area (see details in Figure 5). Due to the low leak-off behavior in this
test, the hydrostone is modeled in Abaqus without any poro-mechanics and capillary effects
(imbibition and drainage behavior) being modeled in the surrounding formation.
The matched results for the pressure response and fracture (leak-off) geometry between Abaqus
and the PTC test are presented in Figure 5. The differences in the breakdown pressure may
potentially be due to system compressibility (not accounted for in the Abaqus modeling), rock
heterogeneities, and mesh sensitivity. The Abaqus simulated injection pressure during fracture
propagation and fracture closure (shut-in) agrees very well with experimental results. Comparisons
of fracture extent associated with the pressure response can be seen in Figure 5b. The top figure is
the experimental leak-off profile within a thin section representing the X-Y plane. The thin red
line goes through the 3/8” wellbore and shows a 7 in. bi-wing fracture with a very limited leak-off
profile, as can be seen by the red dye. The bottom figure presents the 3-D finite element modeling
results demonstrating a similar final fracture length (~8 in. versus ~7 in.) and leak-off profile.
~7”
~8”
5a. Pressure response: Test data vs Abaqus model results. 5b. Fracture geometry.
Figure 5. Fracture results comparison for no leak-off case.

The second validation case is for “high leak-off” category in which an significant amount of leak-
off occurs during the experimental test. The selected test is carried out on a hydrostone sample
(permeability of ~15 mD) fractured with 200 ml of red-dyed fluid having a lower viscosity (1332
cSt) at an injection rate of 5 ml/min. The main difference in the results of the “no leak-off” and
“high leak-off” cases is the presence of capillary pressure effect.
Unlike typical saturated reservoir conditions, the PTC experimental tests are carried out on dry
hydrostone. In the “no leak-off” case, this would not require capturing the capillary effects as no
significant injected fluid is being imbibed into the formation due to capillary pressure. However,
for the second validation case, this can be found to be a significant factor in the observed fracture
propagation and pressure response, and needs to be accounted for in the model. As a result, a 2-D
model with more mesh refinement is adopted for accurately resolving the saturation front within
the model.
To further illustrate the impact of capillary effects, two modeling results are presented in Figures
6a and 6b for comparison. Figure 6a shows the fracture propagation in a fully saturated formation
without capillary pressure effects while Figure 6b shows the fracture propagation in a formation
with partial saturation. Notice that the fracture propagates at a much slower speed within a
partially saturated formation because a greater fraction of the injected fluid is used to fill
remaining pore space in the formation. This results in a reduced volume of injected fluid to drive
the fracture forward.
5 𝑠𝑒𝑐 10 𝑠𝑒𝑐 20 𝑠𝑒𝑐
Figure 6a. Fracture propagation in fully saturated formation.
5 𝑠𝑒𝑐 10 𝑠𝑒𝑐 20 𝑠𝑒𝑐 30 𝑠𝑒𝑐 40 𝑠𝑒𝑐 50 𝑠𝑒𝑐 60 𝑠𝑒𝑐 70 𝑠𝑒𝑐 80 𝑠𝑒𝑐 90 𝑠𝑒𝑐 100 𝑠𝑒𝑐
Figure 6b. Fracture propagation in a formation with partial saturation.

The matched results between lab-scale experiments and numerical modeling for the “high leak-
off” case are shown in Figure 7. Similar to the “no leak-off” case, the propagation pressure
prediction agrees well with the test result for the majority of the 20 minutes of injection. The final
fracture length and leak-off profile are shown in Figure 7b; as expected, significant leak-off occurs
resulting in a smaller fracture length compared to the “no leak-off” case (for reference 200 ml was
injected in the “high leak-off” case compared to 13 ml in the “no leak-off” case). The agreement
of leak-off profile for both the Abaqus simulation and experimental results further validates the
co-developed capabilities for solving more complex, fully coupled problems involving
hydraulically-driven fracturing.
~4”
~5”
7a. Pressure response. 7b. Fracture geometry.

Figure 7. Fracture results comparison for leak-off case.
The agreement between models and physical experiments validates the implementation of
numerical capabilities for solving the fully coupled hydraulic fracturing problem and establishes
confidence in ability to model physical behavior over a wide spectrum of conditions. It fills the
gap between analytical problems having known solutions and complex, large-scale applications
where pressure response is the only available downhole indicator. Fracture modeling using the
advanced capabilities currently available in Abaqus helps to identify and develop solutions to
address certain subsurface challenges such as fracturing in tight rocks and fracture behavior for
lost returns during drilling operations.
5. Conclusions
This work is intended to further validate the Abaqus hydraulic fracturing capabilities co-developed
between ExxonMobil Upstream Research Company and Dassault Systemes Simulia Corporation

using two representative categories of lab-scale experiments. Good agreement between the Abaqus
and experiment results for two benchmark problems establishes a foundation for these new
technologies to support oil and gas problems involving fluid-driven fractures such as cuttings
disposal, wellbore integrity and lost returns, water injection, and unconventional stimulation. For
these problems, traditional fracture modeling tools currently struggle to model the key physical
mechanisms at field scales.
6. References
1. Ortiz, M., Pandolfi, A., “Finite-Deformation Irreversible Cohesive Elements For Three
Dimensional Crack-Propagation Analysis,” Int. J. Numer. Meth. Engng. Vol. 44, pp. 1267-
1282, 1999.
2. Hu, J., Garagash, D. I., “Plane-Strain Propagation Of A Fluid-Driven Crack in a Permeable
Rock With Fracture Toughness,” J. Eng. Mech., vol. 136(9), pp. 1152–1166, 2010.
3. Detournay, E., Adachi, J., Garagash, D. I., Savitski, A., “Interpretation and Design of
Hydraulic Fracturing Treatments,” United States Patent No. US 7111681B2, 2006.
4. Dahi-Taleghani, A, Olson, J.E., Numerical Modeling of Multistranded-Hydraulic-Fracture
Propagation: Accounting for the Interaction Between Induced and Natural Fractures, SPE-
124884, 2011.
5. Abbas S., Gordeliy E, Peirce A., Limited Height Growth and Reduced Opening of Hydraulic
Fractures due to Fracture Offsets: An XFEM Application, SPE 168622, 2014.
6. Clearly M. P., “Comprehensive Design Formulae for Hydraulic Fracturing,” Society of
Petroleum Engineers SPE-9259-MS, 1980.
7. Meyer B.R. “Three-Dimensional Hydraulic Fracturing Simulation on Personal Computers:
Theory and Comparison Studies,” Society of Petroleum Engineers, SPE 19329, 1989.
8. Warpinski N.R., Moschovidis Z.A., Parker C.D., Abou-Sayed I.S., “Comparison Study of
Hydraulic Fracturing Models,” Society of Petroleum Engineers, SPE-25890-PA, 1994.
9. Bunger, A. P., Detournay, E., Garagash, D. I., “Toughness-Dominated Hydraulic Fracture
with Leak-Off,” International Journal of Fracture, vol. 134, pp.175-190, 2005.
10. Garagash, D. I., “Plane-Strain Propagation Of A Fluid-Driven Fracture During Injection and
Shut-In: Asymptotics Of Large Toughness,” Engineering Fracture Mechanics, vol. 73, pp.
456-481, 2006.
11. Garagash, D. I., Detournay, E., Adachi, J., “Multiscale Tip Asymptotics in Hydraulic Fracture
with Leak-Off,” J. Fluid Mech., vol. 669, pp. 260–297, 2011.
12. Peirce, A., Detournay, E., “An Implicit Level Set Method for Modeling Hydraulically Driven
Fractures.” Comput. Methods Appl. Mech. Engrg, vol. 197, pp. 2858–2885, 2008.
13. Savitski, A. A., Detournay, E., “Propagation Of A Penny-Shaped Fluid-Driven Fracture In An
Impermeable Rock: Asymptotic Solutions,” International Journal of Solids and Structures,
vol. 39, pp. 6311-6337, 2002.
14. Howard G. C., Fast. R., Carter R.D. “Optimum Fluid Characteristics for Fracture Extension,”
Drilling and Production Practices, pp. 261-270, 1957.

15. Charlez P.A., “Rock Mechanics, Volume 2, Petroleum Applications,” Editions Technip, 1997.
16. M. G. Zielonka, K. H. Searles, J. Ning and S. R. Buechler. Development and Validation of
Fully-Coupled Hydraulic Fracturing Simulation Capabilities. 2014 Simulia Community
Conference, Providence RI
17. Abaqus User’s Manual, Version 6.14, Dassault Systèmes Simulia Corp., Providence, RI.
2014.
7. Acknowledgement
The authors would like to thank the Friendswood lab staff for carrying out the experiments and
ExxonMobil Upstream Research Company for permission to publish this work.

Dynamic Hydraulic Fracture Modeling for Wellbore
Integrity Prediction in a Porous Medium
Nikolay M. Kostov1, Jing Ning1, Shekhar V. Gosavi2, Piyush Gupta3, Kaustubh
Kulkarni1 and Pablo F. Sanz1
1
2
ExxonMobil Development Company
3
ExxonMobil Upstream Research Company 2014 Summer Intern
Abstract: We present a workflow for modeling dynamic growth of hydraulically induced fractures
in a porous medium in the near-wellbore region. The focus is on the early time fracture growth
behavior and the stress state near the wellbore as a result of it. The model is based on the
Cohesive Zone Model (CZM) elements with pore pressure degrees of freedom, which were co-
developed with ExxonMobil Upstream Research Company and recently implemented into the
Abaqus finite element package. In addition to including the physics of dynamic fracture growth
and pore pressure diffusion, the model allows for fluid leak-off from the wellbore and fracture
faces into the porous medium, as well as an arbitrary injection schedule. The dynamic fracture
growth and interaction between the stress concentration region near the tip and the wellbore
result in a non-linear behavior of the near-wellbore stress state during early time fracture
propagation. This analysis allows for a recommendation of optimum fracture width for a
corresponding maximum increase in wellbore integrity for the target formation. The model also
provides an effective tool to improve the treatment selection (e.g., particle size distribution for
effective treatment), as well as field diagnostics for various lost circulation treatments and lost
returns events. The model is incorporated into a fully automated workflow which allows the user
to setup the analysis without any previous experience with Abaqus.
Keywords: CZM, drilling and subsurface, hydraulic fracturing, lost circulation, lost returns, oil
and gas, wellbore integrity
1. Introduction
Rotary drilling requires the use of a circulating fluid. While this fluid has a variety of vital roles,
of particular importance is the ability to carry the formation cuttings, which are generated at the
bit, to the surface. However, due to the hydrostatic head of the fluid at great depth and the
dynamic pressure from the fluid motion, the forces acting on the wellbore faces can become high
enough to induce mode I tensile fractures. When a large enough volume of fluid is lost to these
fractures, the ability to carry the cuttings to the surface diminishes and this can prevent further
drilling.
In this paper, we investigate the early time initiation and dynamic growth of such hydraulically
induced fractures in a porous formation. The analysis of the changing width and length of the
fracture and the changing stress state in the near-wellbore region can help us understand how to

prevent such fractures. In addition, accurate prediction of the fracture geometry is important for
the selection of the treatment needed to seal the fracture to prevent further drilling fluid loss.
We use a symmetric 2D plane strain model of the wellbore with Cohesive Zone Model (CZM)
elements representing the induced fracture plane. This allows modeling the transient behavior of a
bi-wing hydraulically induced mode I fracture in a porous medium while keeping the number of
degrees of freedom manageable. This is of prime importance, as a large number of runs are
required to evaluate different formation and fluid properties. We have developed an automated
workflow which can generate the Abaqus input files, including an optimized mesh, execute the
analysis and present the user with the post-processed results of interest.
Most of the results which these new Abaqus capabilities can generate have not been available
through static analysis. The transient effects of the pore pressure diffusion and the interaction
between the stress concentration around the moving fracture tip and the borehole all play a
significant role on the tangential (hoop) stresses at the borehole face. The analysis results allow
recommending the ideal fracture width at the mouth for which the tangential stresses around the
borehole are maximized. This increase in stresses at the wellbore can prevent further fracture
initiation.
Traditionally, there are two models used to estimate the wellbore pressure at which fracture
initiation occurs from a wellbore in a porous medium. The first one is the Hubbert-Willis model
(Equation 1) which considers the stress state around an infinite cylindrical borehole with two far
field stresses and formation pore pressure acting on it, but does not consider the effect of fluid
diffusion into the formation (Hubbert, 1957). The second one is the Haimson-Fairhurst model
(Equation 2) which adds the steady state effect of increasing the near-wellbore pore pressure
(Haimson, 1967). These are usually considered the upper and lower limit, respectively, for the
fracture initiation pressure. However, because of uncertainties in the material properties and the
irregularity of the borehole shape, often, more conservative estimates are used in practice.
𝑃𝑖 = 𝑇 + 3 𝜎min − 𝜎max − 𝑃𝑝 Eq. 1
𝑇+3 𝜎min − 𝜎max −2𝜂𝑃𝑝

𝑃𝑖 = Eq. 2
2(1−𝜂)
Here 𝑃𝑖 is the fracture initiation pressure (i.e., wellbore pressure at which fracture initiation
occurs), T is the tensile strength of the rock, 𝜎min and 𝜎max are the minimum and maximum far
field stresses, 𝑃𝑝 is the initial pore pressure and η=α(1-2ν)/2(1-ν), with ν being the Poisson ratio
and α the Biot parameter of the rock. A compression-positive sign convention is used for stresses
and pressures.
A map of the fracture initiation pressure (nondimensionalized by𝜎min ) as a function of the
formation pore pressure (𝑃𝑝 ) and the maximum far field stress (𝜎max ) for each of the two models
is shown in Figure 1.
While these two approaches give reasonable estimates for the fracture initiation pressure, they are
still not capturing the time-dependent effects of fluid diffusion. More importantly, they do not
provide any information about the fracture dimensions, or the wellbore stresses, once a fracture
has been induced. To the authors’ knowledge, there are no analytical models which can predict the
near-wellbore stresses during fracture growth.

Hubbert-Willis Fracture Initiation Pressure nondimensionalized by σmin as a function of pore pressure and
maximum far field stress. We assume no tensile strength.
Haimson-Fairhurst Fracture Initiation Pressure nondimensionalized by σmin as a function of pore pressure

and maximum far field stress. We assume no tensile strength and η=0.3.
Figure 1. Hubbert-Willis (a) and Haimson-Fairhurst (b) models for fracture initiation

2. Model
The proposed method for modeling the stresses near the wellbore during the dynamic growth of
the hydraulically induced fracture will be described next. ExxonMobil Upstream Research
Company has been working with Simulia on developing new hydraulic fracturing capabilities into
Abaqus. The coupled pore pressure-displacement cohesive elements which model tangential flow
within the gap have been used for our analysis (Abaqus Manual 2014). We use a half-symmetry
plane strain 2D model to keep the degrees of freedom to a minimum while capturing the most
significant physics. For the formation with pore pressure, we used the CPE4P elements. For
modeling the fluid flow and pressure loss from the surface to the depth of fracture initiation, we
use fluid flow elements. The fluid is injected to both the borehole (pore pressure-displacement
elements) and fracture plane (CZM with pore pressure elements). A sample model showing the
mesh and highlighting the connection between the fluid elements and wellbore can be seen in
Figure 2.
Figure 2. Model mesh
The smaller computational cost of a 2D mesh allows modeling a large number of cases with
varying formation and fluid properties. This way we can evaluate the envelope of plausible field
scenarios. Finally, we can generate a range of expected values for fracture initiation, and more
importantly, the range of fracture geometries created can be studied. This is of great help for
drilling, as any materials which can be used for plugging the fractures can be sized and ordered in

advance to have ready while drilling. This can also greatly reduce the cost of drilling as it reduces
the non-productive time which occurs while waiting on delivery of products once drilling begins.
The model has been benchmarked with known semi-analytical solutions for fracture propagation
(Zielonka et al., 2014) and has also been validated with laboratory scale experiments of hydraulic
fracturing (Ning et al., 2015).
3. Workflow
The drilling engineer can quickly become familiar with the workflow and start executing the
analysis as it allows a separation between the user and the model generation and execution. The
process begins with the user filling out a template of material properties for the formation and the
fluid. Some of the properties which are not well known in advance can be left blank and empirical
models are used internally to estimate them. In addition, it is possible to specify value ranges
instead of individual values. For example, if the user specifies three values for Young’s modulus
and two values for Poisson’s ratio, the workflow will generate the input files for all the six
combinations. After the input values are specified, all the input files needed are generated
automatically. If more than one analysis is needed, the cases are submitted in parallel on a
computing cluster. After the Abaqus runs are complete, the results are automatically post-
processed using Abaqus post-processing scripts and various graphs of interest to the drilling
engineer are generated. The user has direct access to these plots. Advanced users can also access
the raw data.
This automated workflow allows a much better use of the engineer’s time, as the user spends very
little time on preparing and running the analysis, and also allows users who are unfamiliar with
Abaqus to perform the task. A schematic illustrating the workflow is shown in Figure 3.
Figure 3. Model generation workflow.

4. Results
Solving for the near-wellbore stresses during dynamic fracture growth provides a wealth of
information which was not accessible before. We are able to analyze the width and length
evolution of the fracture for a given set of conditions, and also look at the pressure response, the
tangential stresses at the wellbore, and the stresses around the fracture tip. In addition, we can
model the stresses after the fracture has been arrested and the fluid has diffused into the formation,
allowing the pore pressure to return to a level close to its original value. Figure 4 shows the
tangential stresses (σθθ) around the wellbore after the fracture has grown to some length. Figure 5
shows the pressure response in the wellbore during the fracture initiation and propagation for three
different injection rates. This type of plot is automatically generated after the analysis. Figure 6
shows the width evolution at the fracture mouth for these injection rates. Figure 7 shows the length
evolution for the same three fractures.
A variety of other data can also be presented which was not possible to gather before, e.g. the time
for the fracture to grow to a given length or the dynamic aspect ratio of the fracture. In addition,
the value and location of the maximum and minimum tangential stress at the wellbore during
fracture evolution can be predicted. Knowing the minimum (least compressive) tangential stress at
the wellbore, and how it changes with fracture width and length, allows us to determine the ideal
particle size to use for plugging the fracture. In addition, it allows estimating the pressure needed
to initiate another fracture.
Figure 4. Tangential stress σθθ around the fracture and wellbore. Fracture width exaggerated ten times.

Figure 5. Pressure response for three different injection rates.
Figure 6. Fracture width at mouth evolution for the three different injection rates.

Figure 7. Fracture length evolution for the three different injection rates.
Contrary to some existing models, we are not assuming a given fracture length, but allowing for
the fracture to propagate at a width and length combination governed by the complex interaction
between rock deformation, fracture fluid pressure, and fluid leak-off.
5. Conclusions
We presented a workflow which allows predicting the particle size needed to plug a dynamically
growing fracture. The size is chosen such that it maximizes the wellbore tangential stresses at the
wellbore faces. The workflow is based on an automated generation of Abaqus input files from a
user supplied list of material properties and field data. The model is generated, executed, and post-
processed completely autonomously, minimizing user processing time. It also allows drilling
engineers who are not familiar with running Abaqus to be able to perform the analysis.
The fracture modeling captures dynamic fracture growth in a porous formation utilizing CZM
elements for the fracture plane. These capabilities, unique to Abaqus, allow for a more accurate
prediction of the fracture geometry and the pore pressure dependent transient stresses near the
wellbore.
6. References
1. Hubbert, M.K., Willis, D.G. Mechanics of Hydraulic Fracturing. Transactions of Society of
Petroleum Engineers of AIME, 1957, v. 210, pp. 153-168.

2. Haimson, B., Fairhurst, C. Initiation and Extension of Hydraulic Fractures in Rocks. SPE
1710. SPE Third Conference on Rock Mechanics. Austin, Texas, 1967.
3. Ning, J., Kao, G., Kostov, N., Searles, K. H., Buechler, S. R., and Sanz, P. F. Experimental
Validation of Simulation Capabilities for Hydraulic Fractures Propagating in a Porous
Medium. 2015 Simulia Community Conference, Berlin, Germany.
4. Zielonka, M. G., Searles, K. H., Ning, J., and Buechler, S. R. Development and Validation of
Fully-Coupled Hydraulic Fracturing Simulation Capabilities. 2014 Simulia Community
Conference, Providence RI
5. Abaqus 6.13 Online Documentation. April, 2013.
7. Acknowledgement
We would like to thank ExxonMobil Upstream Research Company and ExxonMobil Development
Company for permission to publish this material.

A Numerical Investigation on the Bedding
Resistance of Laterally Displaced Pipelines
T. Gerlach, M. Achmus, and M. Narten
Institute for Geotechnical Engineering
Leibniz University of Hannover, Germany
Abstract: The soil resistances in axial and lateral direction strongly affect the behavior of earth-
buried district heating pipes under variable operating temperatures. The state of knowledge
regarding these resistance forces is summarized, and it is shown that almost no information exists
regarding the cyclic effects on lateral soil resistances. However, for the safe design of a district
heating network, the forces that soil can exert on the structure due to relative soil-pipe
displacements are of particular interest. So, a numerical model has been developed, using the
concept of hypoplasticity as an advanced constitutive model for noncohesive, granular materials.
Additionally the intergranular strain overlay comes into use, enabling a realistic prediction of soil
behavior under cyclic loading. The material model is included using the Abaqus umat interface.
Results gained from the finite element model are compared to results from an experimental test
with cyclic lateral loading of a pipe. Due to the satisfying agreement, numerous variations on
loading device have been done. The investigations give a first insight into the behavior of cyclic
laterally loaded district heating pipes.
Keywords: Abaqus, hypoplasticity, intergranular strain concept, soil-structure-interaction, slave-
master-concept, district heating networks, pipelines.
1. Introduction
District heating is a concept where the heat is produced in a centralized location and transported to
the customer via pipelines. The heat is taken from combined heat and power, biomass, geothermal
heating, heat pumps, or solar heating plants and can be used for space and water heating. Due to
the high energy efficiency of district heating with combined heat and power, the supply networks
are continually being expanded. In 2006 district heating had a market share of around 14% in the
residential building sector of Germany. The connected power capacity was around 52.729MW
(AGFW, 2006).
For the design of safe district heating systems, the occuring forces that the soil can exert on the
pipeline have to be determined. During the operation of district heating networks, changes of the
media temperature lead to periodic varying loading conditions. Hereby, the free movement of the
pipe, which tends to elongate and shorten under temperature load, is hindered by the adjacent soil.
The resistance of the soil can be divided into axial friction resistance and lateral bedding
resistance. Within this paper the focus is set on the latter mentioned. While the bedding resistance
under lateral monotonic displacement has already been investigated, the knowledge concerning
cyclic influences is limited.

In a research project funded by the German Federal Ministry for Economic Affairs and Energy,
the cyclic features of lateral soil resistances on pipes are investigated. Herein, experimental tests
and numerical simulations are carried out. This paper firstly outlines the state of knowledge
regarding the lateral resistances acting on district heating pipes. Then a numerical model for the
calculation of lateral soil resistances under monotonic and cyclic loading is presented. First results
of experimental tests are used for calibration and validation of the numerical model. Finally a
parametric study on the impact of cyclic displacement amplitude and number of cycles is
presented.
2. State of Knowledge
The system under consideration and the denominations are depicted in Figure 1. The problem to
be dealt with is to predict the lateral resistance force FB or the average bedding pressure pm
dependent on the lateral deflection y.
Figure 1. System and denominations for the problem of lateral soil resistance
The first publication on this subject was presented by Audibert and Nyman. They carried out a
laboratory test series where the influence of soil density, pipe diameter and embedment ratio was
investigated. A detailed description of the test setup can be found in (Audibert, 1975). An air
dried, clean medium sand, termed Carver sand, was used for testing. The scope of laboratory
investigations included three different diameters, the distinction between loose and dense sand and
the variation of cover ratio (depth of cover/pipe diameter). Additionally, one field test was carried
out. As a schematic result, Audibert and Nyman pointed out two kinds of failure mechanisms
dependent on the embedment depth. This was possible due to a transparent plexiglass window,
which made the inspection of the soil movement possible. For a better view, in some tests thin
layers of charcoal powder were added. It was found that for shallow burials, less than three times
of the pipe's diameter, two wedges are formed. A passive wedge can by observed in front of the
pipe (in direction of displacement). Behind the pipe, a smaller, nearly vertical wedge is forming an
active zone. With increasing burial depth, the lower bounds of the wedges become steeper and
between the two wedges a seperating zone forms, in which no vertical movement of the soil is
observed, compare Figure 2a. With further increase of depth, the passive wedge degenerates into a

confined zone of soil flow. At this state no disturbance at the surface occurs, as it is illustrated in
Figure 2b.
Figure 2. Failure mechanisms according Audibert & Nyman (Audibert, 1977)
Another goal was to find an analytical representation for the soil resistance-displacement
relationship. A nonlinear curve which represents the relation of bedding resistance p over
displacement y from the initial state to the maximal displacement yu was found. At yu, the soil
reaction force reaches its maximum value pu. According to Audibert and Nyman, the normalized
bedding resistance p̃ over the normalized displacement ỹ can be approximated by:
ỹ
p� = (1)
0.145 + 0.855 ỹ
The two normalized parameters are defined as:
p y
p� = and ỹ= (2)
pu yu
So, to predict the bedding resistance for any state of displacement the maximum bedding
resistance pu and the maximum displacement yu are needed. The maximum of bedding resistance
can be obtained from:
pu = γ ∗ H ∗ Nu (3)
Herein, γ is the soil unit weight, H the embedment depth and Nu is a bearing capacity factor found
by Brinch Hansen (Hansen, 1961). This factor can be taken from a chart, depending on the angle
of internal friction and the cover ratio. Audibert and Nyman found a good agreement between their
experimental laboratory and field tests carried out by Brinch Hansen. Also, for the derivation of
the maximum displacement yu an approximation was determined. Here, the relation between
maximum displacement and depth of embedment (in percent) is plotted over diameter. For a
diameter of D = 1 m the maximum displacement yu is up to 6 % of the depth of embedment H.

With increasing diameter, the ratio of yu/H is regressive, and a lower limiting value for the ratio
yu/H can be observed. For loose and dense sand, the values were found to be yu = 0.015H and yu =
0.02H, respectively. This is reported to be in close agreement with the results of Das (Das, 1975),
who investigated the subgrade reaction on vertical anchor plates.
Further experimental work was carried out by Trautmann and O' Rourke (Trautmann, 1985). A
total of 30 lateral loading tests were performed. The relation between normalized bedding
resistance p̃ over normalized displacement ỹ was found similar to Audibert and Nyman, while the
bearing capacity factors are much lower.
Achmus (Achmus, 1995) investigated the bedding resistance of laterally displaced pipes
numerically. Within his finite element modeling, he used the incremental-elastic material law of
Duncan and Chang (Duncan, 1970). Assuming plane strain conditions, the influences of cover
ratio, distance to the border of the trench, and pipe diameter were investigated. It is pointed out
that the diameter of the pipe obeys a functional relation of the bearing capacity factor. This gives
an explanation for the differing values of the bearing capacity factor as determined by Audibert
and Nyman, who mainly investigated pipes with diameters of D = 25mm and D = 60mm, and
Trautmann and O' Rourke, who used pipes with D = 102mm and D = 324mm.
The calculated bearing capacity factors for different deposit densities, burial depth ratios and
nominal pipe diameters are shown in Figure 3.
Figure 3. Bearing capacity factors acc. to Achmus
Regarding the development of lateral bedding resistances on pipes with un- and reloading and
with repeated (cyclic) loading, the state of knowledge is very limited. It is generally known that
foundations or structures in soil under cyclic loading can exhibit an accumulation of
displacements. With un- and reloading a hysteresis loop is observed, and the un- and reloading
stiffnesses may change with the number of load cycles. However, at the time being, for cyclic
lateral displacements of piles an accurate prediction of the bedding resistance is not possible.

3. Experimental Investigations
Experimental tests were carried and shall be used for the calibration and validation of the
numerical model. A sketch of the test device is shown in Figure 4. The testing pit has a width of
2.04m, a length of 2.96m and a depth of 1.57m. Two of the walls are made of concrete, while the
other two are soldier pile walls, so that the assumption of a stiff enclosure seems appropriate. The
sand is poured in layers and compacted with a vibratory device. A deposit density of Dr = 0.72
was measured in laboratory and field testing. After the first layer with a height of 0.4m is
prepared, a concrete filled steel tube with an outer diameter of D = 0.27m is placed on top.
Figure 4. Sketch of the testing device
The load equipment is orthogonally welded to the pipe. It consists of a hydraulic press which is
connected to the pipe by a steel tube, strengthened by two flanges. In the second layer, four
transducers are installed to measure the occurring earth pressures. Sand is then filled up to a cover
ratio of 3 and densified. Afterwards, the measuring instruments are calibrated to zero, followed by
the displacement-controlled loading of the pipe. The experimental result can be seen in Figure 5.
The peak strength has been surpassed within the primary displacement to y = 4.5cm. Afterwards, a
total of 40 displacement cycles with a constant amplitude of 1.25cm are applied. The mean
displacement is increased in steps of 0.5cm from 3.75cm to 5.25cm at intervals of ten cycles so
that the first ten cycles are in the range between 2.5cm and 5.0cm. Figure 5 shows that the bedding
pressure at maximum displacement decreases only slightly in each cyclic loading phase. The
(negative) maximum bedding pressure on the left side of the pipe first increases with the number

of load cycles, until almost the same value as on the right side is reached. In subsequent cycles, the
maximum bedding pressure remains constant or increases slightly, as it does on the right side of
the pipe.
Figure 5. Force-displacement-curve from experimental test
4. Numerical Investigations
The primary loading and the first load level were computed using Abaqus, assuming plane strain
conditions. However, it must be taken into account that the loading tube might hinder the free
movement of soil around the pipe. It is questionable whether the length of the pit is long enough to
avoid an influence of the vertical walls parallel to the loading tube. This should be investigated in
the future using a three dimensional model. Nevertheless, plane strain conditions are deemed
appropriate for the given problem as the main influences on the bedding resistance, geometrically
and materially, can be investigated. For the developed model, the element CPE8 comes into use.
This 8-node element is full integrated (9 Gaussian points) and uses quadratic shape functions. In
contrast to elements with first-order (linear) interpolation, these second-order elements are capable
of representing all possible linear strain fields. Additionally the shape approximation of curved
structures is much better, because the line between two nodes is not necessarily straight. Thus, a
more accurate solution may be obtained with fewer elements in contrast to first-order elements.
Within the soil layers a size bias is used so that the size of the element is increasing with the
distance from the pipe. Hence, the region near the pipe can be discretized finer than the elements
at a larger distance where the mesh density can usually be reduced because of lower stress
gradients. The pipe is assumed to be very stiff and thus not deformable, allowing it to be
approximated by an analytical circular surface. This can be done easily in Abaqus and comes with

a few advantages. The curved geometry of a pipe can be modeled exactly with analytical surfaces,
using curved line segments. The result is a smoother surface description, which can reduce contact
noise and provide for a better geometry approximation. The reference node, whose motion
governs the motion of the surface, is used to apply a prescribed pipe displacement. To take into
account the field installation process four geostatic load steps are computed, where the soil gets
activated in layers, so that the underlying soil would be prestressed.
A relative displacement between the pipe and the surrounding soil occurs under lateral pipe
displacement. To enable the possibility of relative slipping within the finite element model, a
contact condition is defined. For the contact with an analytical rigid surface, the node to surface
option must be used. In this case, the contact conditions are established such that each slave node
interacts with a point of projection on the master surface. The frictional contact is established
using the simple friction law according to Coulomb. The classical Couloumb's friction law
considers the stick-slip phenomenon, which describes the transition between stick and slip states,
depending on the magnitude of applied contact pressure. In contrast, Abaqus employs a nonlocal
friction model, where the full mobilization of the limit frictional stress is dependent both on the
relative displacement between the pipe and soil as well as on the friction coefficient. The relative
displacement after which the maximum shear stress is reached, is denoted as the critical shear
displacement, limiting the elastic range. If the critical shear displacement is exceeded, the surfaces
slide over each other and the friction changes from static friction to kinematic friction. For the
calculations a critical shear displacement of scrit = 1.0mm and a coefficient of friction of µ = 0.42
(corresponding to two third of the angle of internal friction) has been chosen for the soil-pipe as
well as for the soil-wall interaction.
The constitutive law of hypoplasticity is used to describe the soil behaviour. An implementation of
hypoplasticity as an Abaqus compatible user material (umat) is employed (Gudehus, 2008). The
hypoplastic material law is a rate-type model which accounts for the complex non-linear soil
behavior by just one (rather sophisticated) tensorial equation. In its original form, the hypoplastic
material law needs eight parameters (von Wolfferdorff, 1996). It uses the stress state and the void
ratio of a non-cohesive material as state variables and accounts for stress-dependence of soil
stiffness and soil strength, thereby reflecting contractancy or dilatancy of soils under shearing in a
realistic manner. The material law in the original form was found to overestimate the
accumulation of soil deformations under cyclic loading.
Figure 6. Cyclic stress-strain relation: Hypoplastic model without (right) and with
(left) intergranular strain (Niemunis, 1997)

Therefore, the intergranular strain concept was developed, which cares for a realistic simulation of
the behavior under cyclic loads (see Figure 6). However, another five parameters are necessary to
define the behavior with consideration of intergranular strain (Niemunis, 1997). The needed input
parameters for the extended hypoplastic model were determined through laboratory testing,
including cyclic triaxial tests. They have been slightly adjusted to fit better with the experiment.
The used parameters are summarized in Table 1.
Table 1. Input parameters for the hypoplastic model with intergranular strain
φc hs n ed0 ec0 ei0 α β mR mT R βr χ
(°) (MPa) (-) (-) (-) (-) (-) (-) (-) (-) (-) (-) (-)
29.0 4000 0.29 0.443 0.898 1.08 0.1 2.0 6.0 3.0 0.00008 0.8 5.0
Before the backcalculation of the experimental data was done, the developed model was checked
for accuracy. Since the quality of mesh discretization has an important impact on the calculated
results, a study on mesh fineness was performed. However, a more accurate calculation is
associated with an increase in calculation time. The chosen mesh is displayed in Figure 7, for
which the element surfaces along the pipe's circumference have a length of 1.4cm. Taking into
account the quadratic shape functions, the pipe discretization angle is 2.81°.
Figure 7. Finite element mesh used for the calculations

As can be seen in Figure 8, the numerical results are in a fair overall agreement with the
experimental results. The numerical model seems thus capable to properly describe the main
features of the system behavior.
Figure 8. Comparison between experimental and numerical results
To get a better insight on the impacts of cyclic pipe displacement on the occurring bedding
resistance, a parametric study of the loading was performed. Based on the resulting force-
displacement curve of the presented experimental test, a mean displacement of y = 0.6cm was
selected to investigate the behavior of cyclic loading prior to reaching the peak strength. The
amplitude of displacement was chosen to be 0.25cm, 0.50cm, 0.75cm, and 1.25cm. A number of
16 load cycles were applied for each displacement amplitude. The results of the study are plotted
in Figure 9. Herein, the reduction factor denotes the reaction force at the peak of each cycle
normalized w.r.t. the reaction force after the primary loading.
As can be seen, for all amplitudes, the reaction force decreases by about 15% after the first
reloading cycle. For the smallest amplitude, the reaction force reaches a constant value after the
second cycle. For greater amplitudes, the degradation occurs slower. Note that the given values for
the normalized reaction force are related to the maximum displacement of each calculation and not
to peak strength. So the degradation of the peak strength may differ from these values.

Figure 9. Reduction factor for varying amplitude of displacement and cycle number
5. Conclusion
A numerical model with an advanced material law was developed, which predicts the behavior of
a cyclic laterally loaded pipe buried in sand. Results derived with this model are compared to
results from an experimental test in which cyclic lateral displacements were applied on a pipe
section. Good agreement regarding the basic features of the system behavior was observed. This
shows that the model is capable of reflecting the main features of the complex system behavior. A
parametric study was carried out, showing the impacts of displacement amplitude and number of
cycles. Herein the amplitude could be identified for being more influencing on the pipes reaction
force. Nevertheless, more research is needed for a more accurate prediction of the pipe behavior in
operation and with that to reach an optimized design of such pipelines.

6. References:
1. Achmus, Martin: Zur Berechnung der Beanspruchungen und Verschiebungen erdverlegter
Fernwärmeleitungen, Universität Hannover, Ph.D. thesis, 1995 (in German).
2. Arbeitsgemeinschaft für Wärme und Heizkraftwirtschaft - AGFW - e.V., Branchenreport,
2006 (in German).
3. Audibert, Jean M. ; Nyman, Kenneth J.: Coefficients of subgrade reaction for the design of
buried piping. In: Structural Design of Nuclear Plant Facilities ASCE, S. 109–141, 1975.
4. Audibert, Jean M. ; Nyman, Kenneth J.: Soil restraint against horizontal motion of pipes. In:
Journal of the Geotechnical Engineering Division 103, Nr. 10, S. 1119–1142, 1977.
5. Das, Braja M. ; Seeley, Gerald R.: Load-displacement relationship for vertical anchor plates.
In: Journal of Geotechnical and Geoenvironmental Engineering 101, Nr. ASCE 11402
Proceeding, 1975
6. Duncan, James M. ; Chang, Chin-Yung: Nonlinear analysis of stress and strain in soils. In:
Journal of the Soil Mechanics and Foundations Division 96, Nr. 5, S. 1629–1653, 1970.
7. G. Gudehus, A. Amorosi, A. Gens, I. Herle, D. Kolymbas, D. Mašín, D. Muir Wood, R.
Nova, A. Niemunis, M. Pastor, C. Tamagnini, and G. Viggiani. The soilmodels.info project.
International Journal for Numerical and Analytical Methods in Geomechanics, 32(12):1571-
1572, 2008.
8. Hansen, J B. ; Christensen, NH: The Ultimate Resistance of Rigid Piles Against Transversal
Forces; Model Tests with Tranversally Loaded Rigid Piles in Sand. Geoteknisk Institut, 1961.
9. Niemunis, A ; Herle, I: Hypoplastic model for cohesionless soils with elastic strain range. In:
Mechanics of Cohesive-frictional Materials 2, Nr. 4, S. 279–299, 1997.
10. Trautmann, Charles H. ; O’Rourke, Thomas D.: Lateral force-displacement response of buried
pipe. In: Journal of Geotechnical Engineering 111, Nr. 9, S. 1077–1092, 1985.
11. Wolffersdorff, P-A von: A hypoplastic relation for granular materials with a predefined limit
state surface. In: Mechanics of Cohesive-frictional Materials 1, Nr. 3, S. 251–271, 1996.

Pipe Flexibility Analysis and the Development of
PCS – Pipe Calculation System for Abaqus
Carlos Eduardo A. de C. e Sousa
RMC Consultoria, Brazil
Abstract: The current framework for piping stress analysis is based on a simplified calculation
method directly derived from experimental research performed over 60 years ago in the 1940s
and 1950s. This framework was originally intended for hand calculations and, apart from minor
changes and amendments, has been successfully employed by piping engineers since its
development. As computational power increases and finite element analysis (FEA) becomes
accessible for piping engineers, it has become clear that this framework is not well suited for
complex FEA evaluation of piping. Advanced FEA procedures enable engineers to perform in
depth evaluation of piping systems that are extremely difficult or even impossible to evaluate
through traditional methods. Contrasting to traditional methods, a FEA simulation allow
engineers to evaluate creep-fatigue interactions, advanced material models, complex loadings,
complex geometries, complex support conditions, as well as the inherent evaluation of fatigue
stress intensification factors. A major difficulty for performing piping FEA simulation is that
currently there are no high performance FEA packages capable of efficiently generating the
required analysis model for complex piping systems. This paper presents a comparison of Abaqus
results and fatigue tests performed by Markl in 1952, as well as the development of PCS - Pipe
Calculation System, a piping flexibility analysis application for Simulia Abaqus.
1 Introduction
Industries such as Oil & Gas, Chemical, Petrochemical, Nuclear, Naval, Mining and Pulp & Paper
employ pipes as a way of moving fluids between two separate points. On petrochemical plants, it
is possible to identify several acres of piping systems of varying complexity, from the simplest
layouts, for long pipelines interconnecting different units, to the extremely complex layouts, for
piping inside high temperature process units. Pipework represent up to 35% of the materials, 40%
of the erection labor and 48% of the engineering man-hour employed on a process plant erection
project (Peng & Peng, 2009). As a perspective, it is important to point out that a refinery
construction costs can reach figures as high as 15 billion US dollars (Kaiser & Gary, 2013).
Piping systems are typically composed by the combination of a large number of fittings such as
bends, tees, flanges, valves etc. In addition, the piping layout supports locations, and the nature
and position of the installed fittings, play a major role on the mechanical response of such system.
Final piping layout and support design are highly dependent of its global mechanical evaluation
and should be properly defined during design phase. When necessary, several layout iterations
may be needed until a satisfactory solution is found. These iterations are time consuming and all

changes must be propagated to other disciplines involved in the project (i.e.: civil works,
stationary equipment, rotary equipment, instrumentation, etc.). (Basavaraju & Sun, 2004)
Even though during its operational life a piping system may experience several loading conditions,
one of the most important is thermal expansion. The importance of this loading condition
increases along with pipe lengths and temperatures, in many cases dictating the pipe route that
must be followed.
A summary of some of the possible piping loads is presented on Figure 1.
Soil (Buried)
External
Pressure
Vaccum
Static
(Sustained) Dead (Own, Attachments,
Loads Insulation, etc)
Weight
Live
Internal Pressure
Primary Reversing (Level B/C,

earthquake, wind)
Occasional
Not Reversing
(Mechanical)
Vibration (equipment,
Dynamic Repeating
pressure pulsations, etc)
Loads
Level D earthquake
Piping Stresses
Loss of Coolant Accident

Accidental
(LOCA)
Thermal
Internal Pressure Aircraft Impact
Expansion
Anchor/Support Relative Seismic
Movement Motion
Secondary
Building
Settlements
Thermal
Cyclic
Expansion
Stress
Concentration
Peak
Radial Gradient
Temperature
Figure 1. Piping Stresses and Loads

Adapted from (Avrithi & Ayyub, 2010)

Figure 2 presents some possible failure modes that can be associated with these loading conditions
based on type of loading and/or environmental condition.
Platic Collapse Primary Stresses
Primary Stresses
Buckling
Do/t>75
Ductile
Secondary
Ratcheting
Cyclic Stresses
Secondary Stresses
Time Dependent
Piping Failure
Creep
Behavior of Steel
T > 900 ºF
Primary Stresses
Burst
(Pressure)
Machine
Vibration
Brittle High Cycle Fatigue
Flow Induced
Vibration
Fatigue
Peak Stresses
Secondary
Low Cycle Fatigue
Stresses
Primary
(Earthquake,
Pressure)
Figure 2. Failure modes for piping and their cause

Adapted from (Avrithi & Ayyub, 2010)
Technological advances in chemical processes and improvements in power plants energy
efficiency are often limited by the operating pressure and temperature to which their piping plants
are able to safely withstand. During the design phase, increasing process temperature and pressure
may be accomplished through the use of advanced materials, or the development of new ones.
Usually these solutions are not cost effective or even commercially available.
Another way of improving the allowable pressure and temperature of the pipe system is through
its accurate stress analysis. Improvements in stress analysis accuracy would also represent
economical gain through the rational use of materials and preventing unnecessary changes in
layout, as well as substantially improving the piping system operational safety and reliability.
At the present time, the vast majority of industrial pipe systems are mechanically evaluated
through a simplified stress analysis procedure called “Piping Flexibility Analysis”.

2 Background
2.1 Piping flexibility analysis
Even though a stress analysis represents a deeper and much more comprehensive evaluation of a
piping system than a piping flexibility analysis, still today many pipe engineers regard them to be
equivalent (Peng & Peng, 2009).
In order to ensure safety against piping failure or its attached structures, piping flexibility analysis
aims at reducing pipe stresses and loads resulting from its operation. (Markl, 1955).
A set of papers published by A. R. C. Markl (Markl, 1947, 1950, 1952, 1955) in the late 1940’s
and early 1950’s is arguably the most significant contribution to the current development of piping
flexibility methods. (Koves, 2000)
A piping flexibility analysis is performed considering the pipe system as a set of straight beams.
Some fittings, such as bends tees, present a reduction on their stiffness due to cross section
ovalization. In order to correct its results, a set of flexibility factors is included in the analysis.
This flexibility factors act as knock down factors to the stiffness whenever required.
Stress intensification factors (SIFs) need also to be included in the analysis in order to account for
cyclic stresses. SIFs relate the fatigue life of a component with respect to a girth weld on a straight
pipe, subjected to displacement loads.
2.2 Piping flexibility analysis inconsistencies
Even though the above framework allowed for safe design of many piping systems during the last
six decades, several authors have raised questions regarding many aspects of its application
(Hinnant & Paulin, 2008; Moore, Rodabaugh, Gwaltney, & Mokhtarian, 1987; Rodabaugh, 1987;
Scavuzzo, 2006; E. A. Wais, Rodabaugh, & Carter, 2002; E. Wais & Rodabaugh, 1998a, 1998b)
Flexibility factors and SIFs for pipe bends and welding tees presented on (ASME-B31.3, 2012).
Code standards usually provide SIFs and flexibility factors for a small amount of simple
geometries. Whenever a certain geometry is not clearly permitted, the designer is required to
develop this factors through testing (ASME-B31J, 2008) and since the geometries provided by the
Codes are very simple, this is often the case. Due to the high costs of performing such tests, many
users perform FEA of the component in order to determine the required factors.
Currently no piping Code present comprehensive rules for FEA evaluation, and it is not unusual
for users resort to pressure vessel codes such as (ASME-VIII-2, 2013; BS-EN-13445-3, 2012).
Mixing of piping and pressure vessel Codes is not permitted by the Codes themselves, but is an
industry-accepted practice. Both types of Codes have very different design criteria and this mixing
leads to an inconsistent and possibly unsafe design (Peng, 2013).
Accurate determination of the flexibility factors are of major importance. Flexibility factors are
directly used on the determination of support loads, nozzle loads and, along with SIFs, in the
determination of the fatigue life of the system. The piping Codes provides flexibility factors
calculation formulas for pipe and mitered bends only, all other fittings have either flexibility
factors of 1, or require the flexibility factor to be determined by testing.

It is clear that the Codes flexibility factor of 1 for tees and branch connections, for instance, might
result on overestimation of the internal moments by a factor of 9 (Rodabaugh, 1987) or up to a
factor of 20 (PRG, 2008). Inaccuracies on flexibility factors of other fittings may lead to errors of
the same magnitude, while it is not possible to assure that these errors will lead to safe design.
Stress intensification factors where determined with respect to girth welds produced in the 1950’s.
Recent studies have shown that these Code values may need revision (Hinnant & Paulin, 2008).
These inconsistencies may lead to unnecessary re-routing of piping, incorrect estimation of loads
on sensitive equipment nozzles (turbines, pressure vessels, pumps, compressors, etc.).
Table 1. Stress intensification factor (ASME-B31.3, 2012)
Flexibility Stress Intensification Factor Flexibility
Factor, Out-of-Plane, In-Plane, Characteristic,
Description Sketch
𝑘 𝑖𝑜 𝑖𝑜 ℎ
Welding elbow 1.65 0.75 0.9 𝑇�𝑅1

or pipe bend ℎ ℎ2⁄3 ℎ2⁄3 𝑟2 2
Welding tee in 0.9

accordance 1 3� 𝑖 + 1�
with B16.9 ℎ2⁄3 4 𝑜 4
3 Using Abaqus for Markl’s Fatigue Tests

An alternative to avoid the inconsistencies presented on item 2.2 would be to perform the full
piping stress analysis according to a modern pressure vessel Code such as (ASME-VIII-2, 2013;
BS-EN-13445-3, 2012).
For a large piping system, stress analysis may become extremely time consuming and the proper
use of specialized elements such as Abaqus ELBOW31 is of paramount importance. The
performance of this elements have been evaluated at (Simulia, 2014).
In order to evaluate the capability of Abaqus specialized element ELBOW31 and the linear
quadrilateral element S4R against the fatigue test results performed by (Markl, 1952), a set of 10
models was generated. General model dimensions are indicated in Figure 3. The fatigue evaluation
of straight pipe, long radius elbow and ASME B16.9 tee were performed according to the ASME
master S-N Fatigue curve (ASME-VIII-2, 2013). No safety factor on the Master Fatigue curve was
considered. The cross section properties were taken according to the nominal dimensions of 4”
STD pipe. Material properties according to ASTM A-106 Gr. B. For the straight pipe and long
radius elbow models, both elements ELBOW 31 and S4R were evaluated. Only S4R element was
evaluated for the tee model. An “opening” displacement of 25 mm on the In-Plane and Out-of-
Plane directions were applied to the free edge of each evaluated system.

Figure 3. Markl’s fatigue test general dimensions
Fig. 2 from (Markl, 1952)
3.1 Straight pipe ELBOW31 and S4R analysis
The analysis results for the straight pipe model displayed extremely good agreement to Markl’s
original results (Figure 5). All results from Abaqus were found to be within the test data scatter.
A sample analysis result for the S4R models is presented on Figure 4.
Figure 4. Sample straight pipe FEA results (S4R)

Abaqus ELBOW31
Abaqus S4R
Figure 5. Fatigue results for butt-welded joints in straight pipe

3.2 Long radius ELBOW31 and S4R analysis
The analysis results for the long radius pipe model displayed extremely good agreement to
Markl’s original results (Figure 7 and Figure 8). All results from Abaqus were found to be within
the test data scatter.
Figure 6. Sample long radius FEA results (S4R)

Abaqus ELBOW31
Abaqus S4R
Figure 7. Fatigue results for long radius elbows, in-plane

Abaqus ELBOW31
Abaqus S4R
Figure 8. Fatigue results for long radius elbows, out-of-plane


3.3 ASME B16.9 Welding tee S4R analysis
Result analysis for the welding tee model presented good agreement to Markl’s original results for
the out-of-plane loading condition and Abaqus results were within the test data scatter (Figure 11).
FEA results for the in-plane loading conditions were below the reference line (Figure 10). Further
analysis is required, but the probable cause for the adverse result is a geometric difference between
the crotch radius of Markl’s welded tee and the one modeled on the present study.
Figure 9. Sample welding tee FEA results (S4R)
Abaqus S4R
Figure 10. Fatigue results for welding tees, in-plane


Abaqus S4R
Figure 11. Fatigue results for welding tees, out-of-plane

4 Pipe Calculation System

4.1 Introduction
Finite element simulation on large industry piping systems required the development of an
application for Abaqus to improve speed and ease the geometry creation process. The developed
application is called Pipe Calculation System (PCS). PCS focuses in select industries such as Oil
& Gas, Chemical, Petrochemical, Nuclear, Naval, Mining and Pulp & Paper.
Popular software for piping flexibility analysis currently available to end-users are Caesar II
(Intergraph), Triflex (Piping Solutions) and Rohr2 (Sigma). The first two software are based on
the calculation criteria originally developed by (Markl, 1955) and described on item 2, while the
last one is able to perform mixed flexibility and FEA calculations. Intergraph estimated that
Caesar II currently controls 75% of the piping stress analysis market share (Intergraph, 2012).
Improvements introduced by the use of PCS can be seen on the overall analysis quality, where
finite element simulation provides better and more reliable results compared to regular piping
flexibility. This allows the pipe system to be designed with a proper estimation of its lifespan,
reducing the probability of unforeseen maintenance plant stops due to pipe failure. Additionally,
PCS reduces design time with its intuitive user-interface by the use of up-to-date, consistent and
reliable engineering requirements, together with easy and effective reporting with state-of-art
technology.

4.2 Development
Developing such application requires merging of different expertise: a wide knowledge of the
traditional framework of piping flexibility analysis; proficiency on Abaqus CAE its qualities and
weaknesses regarding its use for piping flexibility analysis; knowledge on software development
and Abaqus integration. With these set of skills, it is possible to define requirements and changes
that must be implemented on Abaqus, in order to develop a suitable application for modern piping
flexibility analysis to the end-user.
These expertise can be extended to deeper issues. Defining PCS requirements contemplates
awareness of several software currently available with their strengths and weaknesses, along with
how the users expect applications to react to input data and how output data should be presented.
Comprehending Abaqus can be broken down to an understanding of its internal structures,
documentation, mesh generation capabilities and simulations execution. On the developer’s end,
integration of the application with Abaqus needs to benefit from all features already available on
the software, minimizing development costs and providing foundation to PCS’ results.
4.2.1 PCS Features
PCS creates an interface between the user and Abaqus in order to create the pipe system geometry,
define its loads and properties, also incorporating proprietary tools previously developed by RMC
to integrate PCS with different file formats broadly used. The application is also scriptable through
the python interface already provided by Abaqus and a seamless user interface (UI) with Abaqus
allows users to make use of the latest features developed by Simulia.
• Geometry creation
PCS allows a simplified creation of specific pipe geometries from the
application’s smart UI as pipes, reductions, valves and supports. Setting of
material properties can be achieved by selecting it from the material database
provided (). This sets the required properties such as temperature and diameter
for each element modeled. This database includes properties and fittings
dimensions from several piping and pressure vessels standards. The material
database can also be edited, in order to modify existing values or include new
ones.
• Third party file import
Previously developed pipe geometry, created on several common piping
software, can be imported as seen on Figure 14. Integration to Intergraph Caesar
II binary file (C2) and neutral file (CII) is available. It is also possible to import
almost any 3D geometry created by all major 3D pipe design software through
PCF files.
Imported data depends on the file type and amount of usable data it contains.
For instance, Caesar II files allows importing geometry, temperatures, pressures
and load cases, which will be automatically included in the analysis of the
piping system. PCF files allow importing only basic geometry information.

Figure 12. Pipe creation window Figure 13. Pipe properties window
Figure 14. Caesar II Imported Model

• Mesh
PCS automatically optimizes 1D meshes by generating elbow type elements,
which have relatively high performance and low computational cost (Simulia,
2014). 2D meshes are also automatically generated under user request.
• Constraints
Before simulation starts all required constraints, such as elbows MPCs and
support connectors, are applied to the model. Including constraints in a large
piping flexibility analysis is a tedious effort that would consume considerable
amount of time. This feature benefits user experience also reducing time and
minimizing modeling errors.
• Reports
In order for a flexibility analysis to be checked and reviewed by a piping
engineer, it has to be presented in a way that current professionals can readily
evaluate, without representing a major change on the current project workflow.
PCS reports are generated in just a few user clicks. Information includes basic
model data, restraint loads, displacements, stresses and allowable stresses.
4.3 Sample Results
Consider a sample piping system consisting of four parallel air-coolers. Each air-cooler has four
inlet nozzles that are connected to the process unit through a 20” diameter, ¼” thick, ASTM A672
B60 pipe header. The final section of the feed pipe consists of an extremely flexible pipe
arrangement in order to reduce piping loads on the equipment nozzles. The fluid is a hydrocarbon
gas at 450 °C.
The PCS model consisted of 142053 linear quadrilateral elements of type S4R, geometrically non-
linear and the material model was linear elastic. The evaluation of the four load steps took less
than 6 minutes on a standard desktop computer (i7-4790 @ 3.6 Ghz), running Abaqus Standard on
a single processor.
Figure 15 present PCS stress results for the load step WNC+P1+T1, which consists of loadings
from piping weight with no contents (WNC), pipe pressure (P1), and pipe temperature (T1). These
results were compared to the ASME B31.3, calculated using Coade Caesar II 5.1, a standard pipe
flexibility analysis software.
The stresses computed with Caesar II converted into cycles through Markl’s methodology were
compared to cycles calculated according to ASME master S-N curve. The master S-N curve
approach resulted on a fatigue life 20% higher than Markl’s. Differences of up to 120% on nozzles
reaction forces were found between the two approaches. These differences may be attributed to
inaccuracies of ASME B31.3 flexibility factors that are of great importance on compact regions
with several pipe fittings, such as the nozzle region of this model.
The Caesar II calculation, indicates that the piping system needs to be redesigned due to
overstress, while calculation through PCS and ASME VIII div. 2 part 5 clears the system for safe
operation.

Figure 15. Sample air-cooler model
5 Conclusions
On the present paper, a comparison of fatigue analysis on Simulia Abaqus and historical results
from (Markl, 1952) is performed.
Elements ELBOW31 and S4R were used in the FEA models while the ASME master fatigue
curve (ASME-VIII-2, 2013) was used to determine the expected life of each component.
Results show that it is possible to obtain extremely satisfactory results from using Abaqus for the
fatigue analyses of regular piping components.
Further analysis will be required for the welding tee under in-plane loading model, which was the
only analysis case where results did not fall within Markl’s data scatter. A probable cause for this
result inconsistence is a geometric difference between the crotch radius of Markl’s welded tee and
the one modeled on the present study.
This paper also introduces PCS - Pipe Calculation System, a piping flexibility analysis application
for Simulia Abaqus, a tool for fast modelling and result analysis of finite element piping systems.

A sample PCS calculation for an air-cooler pipe assembly is provided and some general results are
evaluated. These results indicate a clear advantage on using advanced finite element models to
evaluate common pipe flexibility analysis problems.
General use of advanced finite element software on piping stress analysis will enable users to
achieve better understanding of the mechanical behavior of their piping systems, reducing
calculation uncertainties, increasing fatigue life and allowing for proper evaluation of non-
standard fittings. Additionally, the use of these techniques will make it possible for the piping
industry to take advantage of the latest developments and trends of the simulation community.
6 References
1. ASME-B31.1,"Power Piping", New York, N.Y., (2012).

2. ASME-B31.3,"Chemical Plant and Petroleum Refinery Piping", New York, N.Y., (2012).
3. ASME-B31.4,"Pipeline Transportation Systems for Liquids and Slurries", New York, N.Y.,
(2012).
4. ASME-B31.8,"Gas Transmission and Distribution Piping Systems", New York, N.Y.,
(2010).
5. ASME-B31J,"Standard Test Method for Determining Stress Intensification Factors (i-
Factors) for Metallic Piping Components", New York, N.Y., ASME, (2008).
6. ASME-VIII-2,"ASME Boiler and Pressure Vessel Code (BPVC), Section VIII, Division 2:
Alternative Rules", New York, N.Y., (2013).
7. Avrithi, K., & Ayyub, B. M.,"Load and Resistance Factor Design (LRFD) of Nuclear
Straight Pipes for Loads That Cause Primary Stress",Journal of Pressure Vessel Technology,
132, 021101, (2010), doi:10.1115/1.4000976.
8. Basavaraju, C., & Sun, W. S.,"CHAPTER B4 Stress Analysis of Piping Sytems",In Piping
Handbook (pp. 107–214), (2004).
9. BS-EN-13445-3,"Unfired pressure vessels. Design", (2012).
10. BS-EN-13480-3,"Metal industrial piping - Part 3: Design and calculation", (2012).
11. Hinnant, C., & Paulin, T.,"Experimental Evaluation of the Markl Fatigue Methods and
ASME Piping Stress Intensification Factors",In Volume 1: Codes and Standards (pp. 97–
113), ASME, (2008), doi:10.1115/PVP2008-61871.
12. Intergraph,"Intergraph Cadworx & Analysis Solutions Flyer", (2012), Retrieved February
10, 2015, from http://www.intergraph.com/assets/pdf/Flyer.pdf.
13. Kaiser, M. J., & Gary, J. H.,"A Review of Refinery Markets and Cost Estimation", 597–630,
(2013).
14. Koves, W. J.,"Process Piping Design: A Century of Progress",Journal of Pressure Vessel
Technology, 122(August 2000), 325, (2000), doi:10.1115/1.556199.
15. Markl, A. R. C.,"Fatigue Tests of Welding Elbows and Comparable Double-Mitre
Bends",Transactions of the ASME, 869––879, (1947).
16. Markl, A. R. C.,"Fatigue Tests on Flanged Assemblies",Transactions of the ASME, 78––87,
(1950).
17. Markl, A. R. C.,"Fatigue Tests of Piping Components",Transactions of the ASME, 287––
303, (1952).
18. Markl, A. R. C.,"Piping-Flexibility Analysis",Transactions of the ASME, 127–149, (1955).

19. Moore, S. E., Rodabaugh, E., Gwaltney, R., & Mokhtarian, K.,"Review and Evaluation of
Design Analysis Methods for Calculating Flexibility of Nozzles and Branch Connections",
(1987).
20. Peng, L. C.,"Understanding Piping Code Stress Evaluation Paradoxes And ASME B31.3
Appendix P", (2013), Retrieved December 20, 2014, from
http://www.pipestress.com/papers/piping-code-paradoxes-b31-3-p.pdf.
21. Peng, L. C., & Peng, T. L.,"Pipe stress engineering", New York, NY ASME Press, (2009).
22. PRG,"Pipe Stress Errors - When You’re Off by 10 Times! Evaluating Pipe Stress Errors
Due to Incorrect SIFs and Stiffnesses", (2008), Retrieved January 07, 2015, from
http://www.paulin.com/WEB_FESIF.pdf.
23. Rodabaugh, E. C.,"WRC-329 - Accuracy of Stress Intensification Factors for Branch
Connections",Welding Research Council, (1987).
24. Scavuzzo, R. J.,"Effect of Loading on Stress Intensification Factors",Journal of Pressure
Vessel Technology, 128(1), 33, (2006), doi:10.1115/1.2148420.
25. Simulia,"Parametric Study of a Lner Elastic Pipeline Under In-Plane Bending",In Abaqus
6.14 - Example Problems Guide - Volume I: Static and Dynamic Analysis (pp. 1.1.3–1–
1.1.3–7), (2014).
26. Wais, E. A., Rodabaugh, E. C., & Carter, R.,"Investigation of Torsional Stress
Intensification Factors and Stress Indices for Girth Butt Welds in Straight Pipe",In Design
and Analysis of Piping, Vessels, and Components (Vol. 2002, pp. 83–89), ASME, (2002),
doi:10.1115/PVP2002-1266.
27. Wais, E., & Rodabaugh, E. C.,"TR-110755 - Stress Intensification Factors and Flexibility
Factors for Pad-Reinforced Branch Connections",EPRI, (1998a).
28. Wais, E., & Rodabaugh, E. C.,"TR-110996 - Stress Intensification Factors and Flexibility
Factors for Unreinforced",EPRI, (1998b).

Design Optimization of Curved Fir-Tree Root of
Steam Turbine Using Computer Aided Optimization
and Finite Element Analysis
TAN Shangjiong, SHU Weilin, WANG Gongyi, CHENG Kai, YANG Jiandao
Shanghai Turbine Works Co., Ltd, Shanghai 200240, China
Abstract: This paper is concerned with optimization of the design of a turbine blade curved
fir-tree root and rim by integrating a knowledge of computer aided optimization (CAO) and finite
element analysis (FEA). The optimization software Isight and FEA software Abaqus are applied in
an effort to optimize the shape of the curved fir-tree root of a given turbine blade. During this
process, root profile from axial perspective is fixed, while attention is devoted to assess the effect
of geometrical feature from the radial perspective. Several critical geometrical features such as
the radius of root centerline, (x, y) coordinate of center point of the centerline, and distances from
blade center to leading and trailing edge of the platform, are taken as design variables and the
maximum equivalent stress of root and rim as objective function. Two optimization algorithms are
employed in this paper. Compared with the original structure, the position of the maximum
equivalent stress after optimization hardly changes while the maximum equivalent stress of root is
reduced significantly, resulting in better operation reliability of turbine blade. A product model
from Shanghai Electric (Group) Corporation (SEC) is used as a preliminary design in the
optimization.
Key Words: Steam turbine blade, Curved root, Finite element analysis, Multi-Island Genetic
Algorithm, Downhill Simplex

1. Introduction
Last stage long blade is a core part of steam turbine, it directly influences the performance of
steam turbine and represents the most advanced level of steam turbine technology. During the
operation of steam turbine, blade root suffers from huge centrifugal force generated by all the
rotating blade. Take a 3000rpm last stage blade for example, the centrifugal force root bears
will be as large as the weight of a train. As a result, root design is of great importance in blade
development. Generally, fir-tree root is a root style which is widely used in last stage blade.
Compared to Tee root, fir-tree root has more pairs of contacts, resulting in smaller stress on
each contact pair. Curved fir-tree root, as an improved version, is more advanced and has
obvious advantages. Compared to traditional side entry straight fir-tree root, curved root has
larger contact areas, thereby has smaller stress, loading the same centrifugal force. With the
development of modern steam turbine, there is a need to make longer blades, nowadays long
blades can be longer than 1200mm, and even longer. With the ascending problems in root
design process, it is quite urgent to study the shape optimization of blade root.
There is plenty of research that concerns the optimization of steam turbine blade root at
present. The urgent optimization issues to tackle involve static stress, strain and durability.
Most of these issues are strongly nonlinear and multi-peak problems. Among the numerous
optimization algorithms, Genetic Algorithm plays a representative role of non-classical
optimization algorithm, which has received some attention for these problems recently. It has
so many advantages as simple currency, strong robustness and parallel processing possibility,
which can easily overcome the limitation of traditional optimization. But there are still some
practical disadvantages, such as poor local search, premature convergence, slow convergence,
etc. (Hu, 2014). Thus, an improvement is made and Multi-Island Genetic Algorithm is
employed in this paper. Although MIGA has advantages that TGA does not have, there is still
one disadvantage it cannot overcome. Good global search capability calls for high
computational cost, resulting in this optimization algorithm be not well-suited for long
running problems. For some engineering optimization problems like in this paper, fast design
improvement is much more important than finding a global optimum solution. Thereby,
another optimization algorithm Downhill Simplex is also employed in this paper. The results
of both optimization algorithms are compared and analyzed respectively.
Blade root shape optimization is a topic which is studied by a large number of scholars
(Zhang, 2012; Wu, 2012). According to JS Rao and B Kumar’s paper (J.S. Rao, 2012), a
turbine blade with a curved entry root shape is optimized to achieve minimum plastic strain.

Three different DOE methods (Latin Hypercube, Hammersley and Box-Behnken) are chosen
in this optimization study. In W Song’s paper (Song, 2002), intelligent CAD system and finite
element method are employed to realize the automation and optimization of the design of a
turbine blade fir-tree root. During the optimization, OPTIONS software package is employed
and Genetic Algorithm is chosen. However, past research mainly studied the optimization of
root profile from the axial perspective, there is few research on the shape optimization of
curved root from the radial perspective. According to practical experience, parameters in this
perspective influences the stress distribution significantly. In this paper, root profile from
axial perspective is fixed, while attention is devoted to examining the effect geometry features
from the radial perspective. During this process, optimization software Isight is employed to
manage the data flow and control the optimization parameters, Finite element software
Abaqus is employed to carry on finite element analysis of blade-groove model. Several
geometric parameters are chosen and optimization objective is to minimize the max Mises
stress on blade root.

2. Problem configuration
The optimization of blade root shape essentially belongs to constrained optimization

problems. With highly nonlinear characteristics as geometry discontinuity and contact
nonlinearity, this problem often has a large number of local optimums. Typical blade-groove
contact model includes a blade, corresponding part of groove and the contact of them.
The optimization proceeding of the problem is shown in Figure 1. As the decrease of
max Mises stress means the reliability of turbine operation, the max Mises stress of both root
and groove is an important indicator, especially for last stage blades. Once the root profile
from axial perspective is fixed, the stress distribution is mainly influenced by geometry
features from radial perspective, such as radius of root centerline, (x, y) coordinate of center
point of the centerline, and distances from blade center to leading and trailing edge of the
platform. Furthermore, According to present experience, stress of groove is usually smaller
than that of corresponding root, thereby only the max Mises stress of root is taken as the
optimal objective, for the purpose of reducing the complexity of this problem.
Figure 1. Optimization proceeding of the problem
3. Integrated optimization based on MIGA & Downhill Simplex
In this paper, two optimization algorithms are employed to carry on optimization of root

shape. The global optimization is mainly for specifying the standard values of geometry
parameters of root from radial perspective under the circumstance of minimum max Mises
stress on root, while another algorithm is mainly for quick improvement of blade root design.
The corresponding two algorithms are Multi-Island Genetic Algorithm and Downhill Simplex,
specifically.
In the Multi-Island Genetic Algorithm (MIGA), like other genetic algorithms, each
design point is perceived as an individual with a certain value of fitness based on the value of
the objective function and constraint penalty. An individual with a better value of the
objective function and penalty has a higher fitness value. Each population presents a set of
design points, which is altered by genetic operations as follows: selection, crossover and
mutation. In the selection process, a new population of individuals is selected according to the
survival of the fittest theory; in the crossover operation, two individuals are crossed at 2
points and the genes between those points are swapped in the two chromosomes resulting in
two new individuals. In the mutation operation, similar to the biological mutation, a randomly
selected gene is changed to further increase the variability of population and avoid stagnation
in the evolution process. Meanwhile, the best individuals called elites are preserved without
alternation, to make sure that the best genetic material is carried over to child population
(Isight Users Manual).
The main feature of Multi-Island Genetic Algorithm that distinguishes it from traditional
genetic algorithms is the fact that each population of individuals is divided into several
sub-populations called "islands". All traditional genetic operations are performed separately
on each sub-population. Some individuals are then selected from each island and migrated to
different islands periodically. This operation is called "migration" (Meng, 2010). Two
parameters control the migration process: migration interval which is the number of
generations between each migration, and migration rate which is the percentage of individuals
migrated from each island at the time of migration (Chen, 2008).
The Downhill Simplex method is a geometrically intuitive algorithm. A simplex is
defined as a body in n dimensions consisting of n+1 vertices. Specifying the location of each
vertex fully defines the simplex. This method requires that you define an initial simplex by
specifying N + 1 starting points. That is to say, you must judge the best location from which
to start. Starting with the initial simplex defined by the points, the downhill simplex method
performs a series of reflections. A reflection moves from a point on the simplex through the
opposite face of the simplex to a point where the function f is smaller. The configuration of
the reflections conserves the volume of the simplex, which maintains the non-degeneracy of

the simplex. However, sometimes expansion will occur during reflection to make
convergency faster. Conversely, shrinking of simplex can also occur, allowing it to settle into
a minimum or pass through a small opening like the neck of an hourglass (Isight Users
Manual).
This method has the highest probability of finding the global minimum when it is started
with big initial steps. However, for complex hyper-dimensional topographies, the method can
break down. Besides, the result of this method can vary due to different initial geometry.
Thereby, selection of starting point is quite important when employing this method.
As we can see from the comparison above, Multi-Island Genetic Algorithm is a method
for global optimization, which is not well suited for long running problems, but can find a
global optimum solution robustly; while Downhill Simplex is an exploratory technique
method for quick design improvement, which is well suited for long running problems, but
may result in the poor local search. Optimization parameters for each algorithm are shown in
Table 1 and Table 2, respectively.
Table 1. Optimization parameters of Downhill Simplex

Option Value
Initial Simplex Size 0.1

Max Iteration 100
Max Failed Runs 5
Failed Run Penalty 1.0E30
Failed Run Objective 1.0E30
Table 2. Optimization parameters of MIGA

Option Value
Sub-Population Size 10
Number of Islands 10
Number of Generations 10
Rate of Crossover 1.0
Rate of Mutation 0.01
Rate of Migration 0.01
Interval of Migration 5
Elite Size 1
Rel Tournament Size 0.5

Penalty Base 0.0
Penalty Multiplier 1000.0
Penalty Exponent 2
Default Variable Bound (Abs Val) 1000.0
Max Failed Runs 5
Failed Run Penalty Value 1.0E30
Failed Run Objective Value 1.0E30
4. Employing the derived approach for a model problem
4.1 Optimization strategy
The optimization approach introduced in this paper is employed for a product model
from SEC. Figure 2 shows the base design. The product model chosen is a free standing blade
to get rid of the difficulty in assembly. Besides, to make model assembly and mesh generation
easier, transition arc of middle part is simplified. Figure 3 shows the view from radial
perspective, in which main parameters are labeled. The parameters have fixed values except
three design variables: Rin, LengthLeft and RootOffset. In this model, shape of airfoil is fixed,
while shape of groove changes with the corresponding changes of the shape of root. To make
sure that the platform can cover the base section of airfoil, range of three design parameters
has some limitations as listed in Table 3.

Figure 2. Base design
Figure 2. Radial view of platform and root
Table 3. Range of design variables

Design Variable Initial Low bound Up bound
Rin (mm) 300 300 600

LengthLeft (mm) 100 100 115
RootOffset (mm) 50 35 50
During the optimization, the Downhill Simplex algorithm is first employed to achieve a
fast improvement, then a global optimization is carried on employing Multi-Island Genetic
Algorithm.
4.2 Integration of optimization algorithms and finite element analysis
A commercial optimization software would be preferable for integrating optimization

algorithms and finite element analysis. Isight is such an optimization software combining
industrial optimization design and automatic analysis. The integration of Isight and other
software does not have a uniform standard and method. So it is important to further study the
interface to connect Isight and FEA software. In this paper, interface is developed between
Isight and UG /Abaqus. In general, the integration of Isight and UG/Abaqus requires the
definition of data transfer, such as importing, exporting and data processing. Figure 4 shows a
typical integrated platform, which can monitor the whole optimization process and
accomplish the optimization design. The Optimization1 module controls the optimization
algorithm and value of design variables, GoUG module updates geometry models based on
the received design variables, and GoAbaqus module is in charge of pre-processing, solving,
and post-processing, these functions are realized by secondary development of Abaqus using

Python language. Then the output of module GoAbaqus is transferred to module
DataExchanger, which picks the max Mises stress and passes it to module Optimization as an
objective function.
Figure 4. Integrated platform of Isight and UG/Abaqus
4.3 Finite Element Model
To model the real operation condition of the blade, a one sector model is considered and
the constraint of cyclic symmetry is applied. All three degrees of freedom of the bottom face
of groove is fixed. To make it easy to assembly, the part of airfoil is tied to up face of the
middle part. Then, general contact is set on the corresponding areas of root and groove. All
the boundary conditions mentioned above are illustrated in Figure 5. At last, rotational body
force is applied on the whole model.
Figure 5. Boundary conditions illustration

Figure 6. Mesh modeling
In general, finite element analysis is computational expensive, there should be a balance

between accuracy of result and computational cost, especially time consumption. So
compromise is made by adjusting the density of mesh. In such a model, mesh is really rough,
the element size and mesh quality can be seen from Figure 6. Although mesh quality is
relatively low, the result of analysis can tell the difference between the original design and the
design after optimization. Mesh parameters are listed in Table 4.
Table 4. Mesh parameters

Position Element size Element type
Root 4
Linear tetrahedral
Groove 4
element (C3D4)
Airfoil 10
Mesh quantity
409376 elements, 80202 nodes
(Base Design)
Finite element software Abaqus supports batch running of the analysis and this allows
the analysis to be embedded into the overall optimization loop. Thereby, the whole process
from importing geometry, assembling, meshing, setting boundary constraints, applying loads
and post data processing are finished automatically by calling a script. During which, the
geometry file is generated by module GoUG automatically, and the output data of module
GoAbaqus is transferred to module DataExchanger.

4.4 Optimization Results
Two optimization algorithms are applied in this optimization task, so the results of them
are compared and analyzed respectively.
4.4.1 Optimization by Downhill Simplex
Figure 7(a) illustrates the history of max Mises stress changes with optimization loops.
Predefined loop number is 500, but the optimization finished after 200 loops. The max Mises
stress of original design is as large as 982MPa, while it changed to 823MPa, with a reduction
of 159MPa, 16.2% relatively, which is really significant. Figure 7(b) (c) (d) illustrates the
history of three design variables during optimization, from which we can see there is no
obvious regularity of LengthLeft, nor is RootOffset before convergence. However, the
variable Rin tends to convergent after 40 loops. Obviously, the optimum solution achieved is
a local search result, which is far away from the global optimum solution. However, this
solution is easily reached, with a significant reduction in max Mises stress, which is enough
for an engineering problem. What’s more important, the optimum solution is reached in about
40 loops, spending 13.3hours.

Figure 7. History of parameters during Downhill Simplex process
4.4.2 Optimization by MIGA
Due to time and computational resource limitation, the whole optimization process by

MIGA is not finished up to now. However, the result so far is worth discussing. Figure 8(a)
illustrate the history of max Mises stress changes with optimization loops, it changes from the
original 982MPa to 580MPa, with a reduction of 402MPa, 41.2% relatively. Figure 8(b) (c) (d)
illustrates the history of three design variables during optimization, which is obviously
different from that of Downhill Simplex. The variable LengthLeft tends to converge around
the value of 105, while the variable RootOffset tends to converge around the value of 43.
However, the variable Rin does not show any obvious convergency after such loops. The
optimization of MIGA is time and computational costly. The result so far is reached in 430
loops, spending 143.3 hours.

Figure 8. History of parameters during MIGA process
Table 5 shows the comparison between the results of Downhill Simplex and MIGA.
Downhill Simplex is well-suited for long-running optimization problems, while MIGA is not

well-suited for that. Downhill Simplex takes shorter time to accomplish a fast design
improvement for engineering development, while MIGA takes more time and computational
resource to find a global optimum solution, which is suitable for research purpose.
Table 5. Result comparison between Downhill Simplex & MIGA

Downhill Simplex MIGA
Cpus 16 16
Loops 40 430
Time consumed 13.3hours 143.3hours
Optimum solution 823MPa 580MPa
Percentage of reduction 16.2% 41.2%
5. Optimization design verification
It has been shown from previous preliminary optimization work that it is possible to
reduce max Mises stress significantly by adjusting geometry parameters from radial
perspective. However, the mesh shown in the previous chapters is really rough. To find out
the real difference between designs before and after optimization, a fine mesh model is
employed and precise finite element analysis is carried on in this chapter.
Three cases are studied: the original prototype, the optimized model after Downhill
Simplex optimization, and the optimized model after Multi-Island Genetic Algorithm
optimization. The geometry parameters of these models are listed in Table 6.
Table 6. Geometry parameters of verification models

LengthLeft Rin RootOffset
(mm) (mm) (mm)
Original prototype 100 300 50

Downhill Simplex 101 334 49
MIGA 101 306 42
Finite element analysis is carried on using finite element software Abaqus, as well.
Boundary conditions and load are the same as illustrated above. Fine mesh is generated by
pre-process software Ansa, and the mesh is illustrated in Figure 9. Airfoil, middle part, root
and groove are all meshed by structured mesh, with the element type of C3D8I (8 node linear

incompatible element). Mesh near the contact surface is especially refined both to speed the
calculation, and to make result more accurate.
Fig 9. Fine mesh for design verification
Mises stress contours of root are illustrated in Figure 10, Figure 11, and Figure12,
respectively. The up contour in each figure shows the stress distribution in convex side, while
the down contour shows the stress distribution in concave side. In the case of original
prototype, max Mises stress of root is as big as 924MPa, exceeding the yield limit of blade
material. What’s more, in this case, stress in the convex side is much bigger than that in the
concave side, which means the load distribution is non-uniform. In the case of Downhill
Simplex, max Mises stress of root is 876MPa, which is a bit smaller than that of original
design. Such a stress reduction can easily satisfy the demand of engineering application.
However, the stress difference between convex and concave side is still large. In the case of
Multi-Island Genetic Algorithm, max Mises stress of root is 645MPa, which is much smaller
than that of the other two cases. Besides, seen from the color, values of max stress of convex
and concave side are close, which means the load distribution on the root is more reasonable
in this case. The results are sum up in Table 7.
Fig 10. Mises stress contour of root (Original prototype case)

Fig 11. Mises stress contour of root (Downhill Simplex case)
Fig 12. Mises stress contour of root (MIGA case)
Table 7. Verification result summation

Original Downhill
MIGA
prototype Simplex
Max Mises stress (MPa) 924 876 645

Stress reduction (MPa) 0 48 279
Stress reduction rate 0 5.19% 32.19%
6. Conclusions
According to the result of optimization by MIGA, max Mises stress is most sensitive to
parameter RootOffset. The process tends to converge after 430 loops, but final convergency is
not reached due to time limitation. The optimum solution (maybe the global optimum
solution), max Mises stress=645MPa, can be reached when LengthLeft =101, Rin =306 and
RootOffset =42, according to the FEA result of verification.
According to the result of optimization by Downhill Simplex, max Mises stress does not
show any obvious sensitivity to any single parameter. But the result is converge to 823MPa,

when LengthLeft = 101, Rin = 334 and RootOffset = 49, according to the FEA result of
verification.
The max Mises stress is reduced after optimization significantly, proving that the
technical route of the optimization is reasonable. The algorithm of Downhill Simplex is
suitable for long-running problems. In this paper, it accomplishes a quick improvement in
design within 13.3 hours, with a result significant for engineering application. The algorithm
of Multi-Island Genetic Algorithm is able to find the global optimum solution robustly, which
is much better than that of Downhill Simplex. However, the MIGA method requires more
time and computational resource. In simple engineering problems, the time of several hundred
hours is acceptable, while for complex problems, there must be a compromise between
efficiency and accuracy.
7. References
[1] SONG Wenbin, Andy Keane, Janet Rees, Atul Bhaskar, Steven Bagnall. Turbine Blade
Fir-tree Root Design Optimisation Using Intelligent CAD and Finite Element Analysis [J].
Computers & Structures. 2002, 80: 1853-1867;
[2] KONG Haipeng, LI Ni, SHEN Yuzhong. Adaptive Double Chain Quantum Genetic
Algorithm for Constrained Optimization Problems [J]. Chinese Journal of Aeronautics. 2015,
28(1): 214-228;
[3] GAO Xiaoxia, YANG Hongxing, LU Lin, Prentice Koo. Wind Turbine Layout Optimization
Using Multi-population Genetic Algorithm and A Case Study in Hong Kong Offshore [J].
Journal of Wind Engineering and Industrial Aerodynamics. 2015, 139: 89-99;
[4] HU Xingzhi, CHEN Xiaoqian, ZHAO Yong, YAO Wen. Optimization Design of Satellite
Separation Systems Based on Multi-Island Genetic Algorithm [J]. Advances in Space
Research. 2014, 53: 870-876;
[5] CHEN Hong, Ooka Ryozo, Kato Shinsuke. Study on Optimum Design Method for Pleasant
Outdoor Thermal Environment using Genetic Algorithms and Coupled Simulation of
Convection, Radiation and Conduction [J]. Building and Environment. 2008, 43(1): 18-30;
[6] MENG Xiangzhong, Ni Jinping, ZHU Yanbo. Research of Active Vibration Control Optimal
Disposition Based on MIGA and NSGA-II [C]. Sixth International Conference on Natural
Computation, Yantai, China, 2010;
[7] J.S. Rao, B. Kumar. 3D Blade root shape optimization [C]. 10th International Conference on
Vibrations in Rotating Machinery, 2012, 173-188;
[8] WANG Mingfang, WANG Mei, TAN Junshan, Application of MIGA to Optimal Disposition
of Sensors in Active Vibration Control [J]. Journal of Vibration, Measurement & Diagnosis.
2008,28(1):62-65,79;
[9] ZHANG Minghui, ZHANG Di, XIE Yonghui. Design Optimization of Fir-tree Root and Rim
for Control Stage Blades Considering 3D Thermo-elastic Contact Problems [J]. Journal of
Chinese Society of Power Engineering. 2012,32[7]:501-507,516;

[10] WU Jun, ZHANG Di, MA Dandan, ZHANG Minghui, XIE Yonghui. Optimization for
Turbine Blade with Fir-tree Root and Rim by Three-Dimensional Contact Finite Element
Method [J]. Journal of Xi’an Jiaotong University. 2012,46(5):25-31;
[11] Abaqus Users Manual, Version 6.14, Dassault Systémes Simulia Corp., Providence, RI;
[12] Isight Users Manual, Dassault Systémes Simulia Corp., Providence, RI.

Increasing the efficiency of offshore rigid pipeline
lateral buckling assessments using a dedicated GUI
and Isight
Carlos Charnaux, Samuel Paul, and Graeme Roberts
Subsea 7
Abstract: The design phase for offshore rigid pipelines can be lengthy and demanding. A
significant number of engineering hours can be burned before the various analyses attain target
results. The assessment of the lateral buckling phenomenon is one aspect of the pipeline design
that requires numerous finite element simulations. The purpose of such simulations is to
determine if the pipeline has a genuine tendency for lateral buckling and, if so, whether control of
the buckling behavior is required to ensure that the pipeline stresses and strains are within
allowable limits.
The process map for finite element based lateral buckling assessments follows a distinct procedure
distributed into four generic steps: (1) gathering input information, (2) constructing finite element
models, (3) running the simulations and (4) results post-processing. Typically, the process map is
followed manually, and requires engineering analysts to develop quite complex FE models using
typed data input with subsequent use of batched Python scripts to acquire model results, verify
and validate the analysis runs and make tabular and graphical output.
Depending on project requirements and other circumstances such as availability of license tokens
and disk space, the simulation process can be relatively inefficient and places a limit on the
number of analysis runs that can be made and, therefore, the range of input that can be assessed.
The latter point is important because the results of lateral buckling assessments depend heavily on
input uncertainties (e.g. pipe-soil interaction data) and limiting the number of analysis runs may
lead to conclusions that are not based on sound engineering judgment.
In modern offices, it is rare to find engineers who are conversant and comfortable with developing
batch processes or writing computer code that would enable a large range of jobs to be completed
and post-processed efficiently. However, Windows-based applications such as the user-friendly
Isight program allow engineering processes to be automated in order to enhance the results fields
and to better understand the influences of the uncertainties of some input parameters.

Subsea 7 has developed a Graphical User Interface (GUI) that works with Isight and which fully
automates lateral buckling assessments, from the input gathering stage through model generation,
post-processing, graphical output and model verification.
This paper demonstrates the current status of the automation process which is built around the
“Splatter” GUI, Isight workflows and in-house Python scripts. The process enables engineers
with relatively little experience of Abaqus to create, analyze and post-process Abaqus models for
lateral buckling, pipeline walking and on-bottom roughness assessments.
Keywords: Design of Experiments, Exploration, Lateral Buckling, Optimization, Pipeline, Pipe-

Soil Interaction, Post-processing, Probabilistic Design, Scripting.
1. Introduction
Subsea steel rigid pipelines may be required to transport hydrocarbons that are at elevated
temperatures and pressures. Pipelines with these operating characteristics, and which are surface-
laid on the seabed, may be susceptible to the formation of global lateral buckles. The lateral
buckles are the result of a combination of the operational characteristics and the interaction
between the pipe and the seabed. In simple terms, the elevated temperature and pressure introduce
an expansion tendency along the pipeline’s length but this tendency is resisted by pipe-soil
interaction (Figure 1). The pipe-soil interaction generates a compressive force within the pipeline.
The compressive force builds up transiently with the pressure and temperature until either the full
operating condition is reached, and the compressive internal force in the pipeline is not high
enough to initiate buckling, or the pipeline’s capacity to resist lateral buckling is exceeded.
Figure 1. Pipeline expansion due to increase in temperature and pressure and soil
compressive resistance
A typical lateral buckle formation sequence and shape is shown in Figure 2. Were the subsea
pipeline to be fully trenched and buried, then the risk would be of upheaval rather than lateral
buckling. In this paper, only lateral buckles are discussed. Designing for upheaval buckling
requires a somewhat different design approach.
One of the main objectives during the design process for lateral buckling is to ensure that the
stress and strain levels at the buckle crowns remain within allowable limits, such as the ones
recommended in the DNV OS-F101 standard [1]. In the event that the allowable limits are
exceeded, engineered buckling control measures are required. The purpose of the control
measures is to initiate buckling at specific locations along the pipeline length, control the amount
of pipeline that feeds-in to the buckle sites and produce post-buckle deflected shapes in which
stress and strain are within the allowable limits.
An increased number of controlled buckle initiation sites results in a reduced reliability of buckle
formation at each site. The pipeline design engineer has the responsibility of determining the

balance between the number of buckle initiation sites and the reliability of the triggering strategy.
Each engineered buckling initiator results in an additional fabrication and installation cost that
must be borne by the project.
Figure 2. Typical lateral buckle formation

Lateral buckling assessments may become long and complex depending on the project
characteristics and client requirements. The assessment process consists of several stages that
require engineering expertise and time, most notably:
• Data interpretation;
• Constructing FE models;
• Running simulations;
• Post-processing FE output.
The objective of this paper is to show the work performed by Subsea 7 to enhance the lateral
buckling design process by using in-house tools and Simulia’s Isight process automation software.
Automating analyses and post-processing has enabled pipeline design engineers working on recent
deep water, high-pressure high-temperature pipeline projects to assess a wider range input
parameters and better optimization of lateral buckling control strategies than was previously
possible.

2. Lateral buckling analysis process map
Pipelines that transport product with high temperature and pressure may be prone to lateral
buckling. The assessment of a pipeline for the lateral buckling phenomenon is performed in four
distinct stages, as follows:
• Screening assessment to determine the pipeline’s tendency to buckle;
• Virtual Anchoring Spacing (VAS) assessment;
• Global lateral buckling assessment;
• Pipeline walking assessment (axial ratcheting).
The SAFEBUCK JIP [5] has resulted in guidelines for lateral buckling design. The guidelines
include a process map, reproduced in Figure 3. The process map does not mention the specific
stages noted above but gives a general overview of the decision-making process to be used for the
assessment of pipelines that have a lateral buckling tendency.
Figure 3. SAFEBUCK JIP [5] recommended lateral buckling design process map
A “Basis of Design” document is published at the beginning of a project in which is contained all
the relevant design information. Pipeline design is non-determinate because of uncertainties
associated with data collection and application. The main uncertainties relate to:
• Pipe-soil interaction;
• Embedment;
• Environmental data (current speeds, wave heights and periods, etc.);

• Pipe dimensional tolerances;
• Material certificates.
The pipeline design engineer has to deal with these uncertainties, be able to interpret them, select
the useful values and perform engineering with the selected data. It is only after data
interpretation that calculations may be performed. Data is not always available at the project start
and, consequently, significant re-work has to be done during the mid or late project stages.
Screening assessments for lateral buckling are carried out once data is interpreted. Screening
defines the tendency or susceptibility of the pipeline to buckling. Analytical calculations are
performed to determine the compressive effective axial force distribution along the pipeline
assuming no buckles are formed. A comparison is made between the effective axial force and the
buckling limit force (the buckling limit force proposed by Hobbs [3, 4] is used most often for this
purpose). If the maximum compressive effective axial force exceeds the Hobbs’ force then the
pipeline is prone to buckling, as indicated in Figure 4. Different pipe-soil resistance values are
implemented to assess the full spectrum of possible outcomes. Finite element modelling is not
required for this manual work.
Figure 4. Post-buckle effective axial force along a 40km pipeline route
A pipeline that is susceptible to lateral buckling may have buckles that are either “uncontrolled” or
“controlled” with buckles formed either naturally on seabed or by placing engineered buckle
initiators along the route, respectively.
Uncontrolled buckles are triggered by the vertical out-of-straightness of the pipe as it is laid on an
uneven seabed or by horizontal out-of-straightness of the pipe as it is laid down from the
installation vessel. Engineered buckle sites need to introduce a stronger buckling tendency to the
pipeline than any residual tendency from initial out-of-straightness. Examples of engineered
buckle initiators include distributed buoyancy modules or sleepers (Figure 5).

Figure 5. Engineered buckling initiation strategies include distributed buoyancy
modules (left) and dual sleepers (right)
The way forward, when deciding if pipeline buckling needs to be controlled, is to check the
tolerable “Virtual Anchoring Spacing” (VAS) for the pipeline.
Virtual anchors along a pipeline’s length are locations where there is zero axial displacement.
From these locations, pipeline feeds in both directions towards buckle locations. Hence, for the
purpose of analysis the pipeline can be divided into a series of short lengths with fixed boundary
conditions restraining the end of each length (Figure 4). The purpose of the analysis is to
determine the post-buckle stress and strain at the crown of the buckle. The assessment is
performed using Abaqus non-linear finite element models.
VAS models consist of a pipeline, modelled with PIPE31 elements, laid on a flat seabed modelled
as an analytical rigid surface (Figure 6). The pipeline ends are fully restrained and an
imperfection is introduced from which the buckle initiates.
The tolerable VAS length is found by increasing the model overall length until the stress and
strain at the buckle crown are on the allowable limit. The pipeline can be left uncontrolled only in
the event that the tolerable VAS length exceeds the likely buckle spacing predicted by a global
model or manual calculations.
Finding tolerable VAS values may be a long-winded and repetitive process, depending on the
pipeline length and characteristics of the system (wall thickness changes, P&T variation etc.)
Long pipelines can have tolerable VAS values that vary along the route. Work is done in the
detailed design stage to minimize the number of engineered initiator interventions. This type of
optimization can increase significantly the scope of work during pipeline design.

Figure 6. VAS finite element model
A global lateral buckling assessment is generally required regardless of whether uncontrolled or
controlled buckle formation is acceptable. An Abaqus finite element model is built including the
whole pipeline length, using PIPE31 elements, and the seabed is formed of an undulating
analytical rigid surface (Figure 7). The model is numerically more expensive than VAS models
because of its length and the extent of the master-slave contact. The aim of the global model is to
observe the buckle formation along the pipeline and to assess the efficiency and reliability of the
engineered triggering strategy. The pipeline integrity is assessed by checking if the post-buckling
stress and strain levels along the pipeline are within allowable limits.
The assessment is performed using a selection of pipe-soil interaction combinations and different
buckle initiation strategies in order to optimize the design. When the results from the sensitivity
cases are acceptable the buckle initiation strategy and pipeline design for lateral buckling is
accomplished.
Figure 7. Typical Global Finite Element Lateral Buckling model

Lateral buckling models, particularly for global modelling, have any number of input parameters
and variables that add significant complexity to model debugging and running. The analysis can
be surprisingly sensitive even to minor changes in the input and, furthermore, each pipeline design
is unique.
Pipeline walking described by Carr et al [2] is the last stage of “lateral buckling” assessment.
Pipeline walking is a phenomenon that ratchets and accumulates the axial displacement of the
pipeline towards one end or between buckle sites over the design life (Figure 8). The objective of

the pipeline walking analysis is to decide whether the pipeline is stable or not and to derive the
loads that must be resisted by structures that are designed to stabilize the pipeline against walking.
Figure 8. Walking through interaction of buckle sites

Typical anchoring system are suction piles, gravity based anchors and driven piles (Figure 9).
These can be tied to the pipeline either at the ends or at mid-points.
Figure 9. Suction pile (left) and gravity base (right) anchoring mitigations
Pipeline walking finite element models are derived from the global lateral buckling models.
Transient temperature and pressure profiles are applied, which adds a significant number of load
steps. The pipeline walking analysis is often a lengthy process.

3. Automating Lateral Buckling and Pipeline Walking Assessments
3.1 Introducing Splatter
For surface-laid pipelines with lateral buckling tendency, the assessment of the post-buckle
behavior may be the most time-consuming activity.
The process workflow clearly sets out the large number of analytical and numerical calculations
that must be completed to demonstrate that the pipeline integrity is maintained for all reasonable
and foreseeable scenarios.
Commencing the lateral buckling assessment may be a challenge because the full set of input data
may not be available and assumptions may have to be made. Consequently, when proper data is
available, models may have to be re-analyzed.
Constructing Abaqus models for lateral buckling assessments is generic and logical. The model-
building process can be improved by developing tools to make it more efficient. Most engineers
involved with lateral buckling assessments will have their own working methods to derive and
input data for analysis.
Subsea 7 has recently supported the development of an in-house tool to enhance its lateral
buckling analysis and design. The Subsea Pipeline Lateral Buckling model builder (Splatter)
consists of an Excel-based Graphical User Interface (GUI) through which users are able to build,
run and post-process Abaqus lateral buckling and pipeline walking models. Although the GUI is
Excel-based, the pre- and post-processing is dependent on Python scripts.
There are numerous reasons for developing Splatter and the most important are that the finite
element model building process can be optimized by having a simplified and dynamic user
interface and debugging analysis data can be minimized by reducing the number of user errors.
Splatter presents an opportunity to standardize lateral buckling models which consequently
enhances the time spent on model checking and the quality of the delivered results. The dynamic
user interface can avoid mistakes by having live visual checking functions. The standardization is
extended from the model building process through to results post-processing. The analysis output
is formatted to ensure client satisfaction.
Splatter inherits the best practice of experienced Abaqus users in the pipeline design section to
ensure that information legacy is retained and passed to future engineering generations.
Input files are created that are easy-to-read and fully parameterized. With fully parameterized
input, re-work to accommodate data changes is minimized and integration with Simulia’s Isight
software is straight-forward.
3.2 Splatter workflow
The lateral buckling analysis process has several stages and it is a repetitive task but it may not be
easy to follow for inexperienced Abaqus users or engineering novices. Splatter was programmed
to ease the model building process but it has since been found to be an invaluable tool for learning
the subject. The lateral buckling FE model assessment process is illustrated in Figure 10.

Figure 10. Actual typical lateral buckling finite element assessment process
As illustrated, an input text file is manually generated and a static analysis is run using
Abaqus/Standard. The Abaqus Odb file is interrogated using three in-house Python scripts. The
first script performs design code checks and writes the interaction ratios back to the Odb file for
contour plotting. The second script extracts requested output variables and writes the output in
tabular text format for provision to clients. The third script gathers the information and generates
high quality graphs outside of Abaqus/Viewer.
Splatter adds a new layer to the typical process, as illustrated in Figure 11. The GUI generates a
high-level keyword input file which is clear, and easier to interpret than an Abaqus input file. The
high-level keyword file drives Splatter’s Model Builder Python script which generates the full and
detailed Abaqus input files. The GUI facilitates the process.
Figure 11. Splatter typical lateral buckling finite element assessment process
The major advantage of the high-level keyword file is that it is fully parameterized and can be
easily handled by the Isight Simcode application component. Therefore, all process components
(Figure 12) can be used in conjunction with Splatter allowing the assessment of a wide range of
variables and leading to a deeper understanding of the uncertainties surrounding lateral buckling.
Figure 12. Splatter inserted in the Simulia Isight process

The Splatter GUI automates the post-processing stage which is an extensive and fundamental part
of lateral buckling design and one on which engineers may spend any number of hours. A single
tool to post-process model results allows users to build an identity for the company by
standardizing results and enhances communication with clients.
3.3 Splatter GUI and its capabilities

The GUI, illustrated in Figure 13, follows closely the lateral model building workflow. It
combines the construction, analysis and post-processing of lateral buckling simulations.
Figure 13. Splatter Graphical User Interface
The user completes nine modules (forms) prior to running the simulation (Figure 14). They are:
• Analysis module: controls the analysis type, model length, analysis steps, and analysis
controls.
• Pipe section module: controls pipe dimensions, pipe length discretization, material
selection, and mesh refinement. Single pipe, sliding pipe-in-pipe (PiP), and swaged PiP
can be defined.
• Operating profiles module: temperature, pressure and density profiles can be added in
this module. It accepts constant, linear and non-linear user-defined profiles.
• Seabed and out-of-straightness module: flat or undulating seabed profiles can be defined
in this module. Different initial pipeline out-of-straightness can be defined for automatic
implementation by Splatter, e.g. single imperfection, sinusoidal and as-laid horizontal
out-of-straightness.

• Soils module: controls the pipe-soil interaction parameters. Axial and lateral friction
with linear or non-linear responses can be applied by specifying Coulomb friction or
input to a pipe-soil interaction subroutine. Different embedment curves may be defined.
• Triggering module: defines the man-made triggers to be used along the pipeline. These
can be buoyancy modules or sleepers. The module can be also used to place crossing
supports and free span mitigation such as grout bags or concrete mattresses.
• Lay tension module: defines the residual lay tension from pipeline installation. It can be
a constant value or variable along the pipeline route.
• Materials module: defines the material properties to be used for the pipelines and its
coating.
• Anchors module: defines the anchoring used to mitigate pipeline walking displacements.
Anchoring can be defined at any number of locations along the pipeline route and the
anchors can have a linear or non-linear response, the latter may be required to simulate
slacking of anchor chains.
Figure 14. Four of the nine Splatter modules

The Splatter GUI incorporates live diagrams that show pipe length, geometry distribution, mesh
refinement, trigger locations, initial triggering shapes and other information to aid the user and
help instantly verify the models, prior to any analysis. The number of available Abaqus tokens
can be checked on-line and model syntax checks can be made prior to the final run. The analysis
can be started locally with a user defined number of CPU’s and analyses can be suspended and
terminated from inside the GUI.
The user interface is mature. However, based on users’ input and comments the software has been
refined and an on-bottom roughness module is under development. The objective of this new
module is to build local modal analyses that may be necessary to augment input to vortex induced-
vibration (VIV) fatigue assessments. When this piece of the puzzle is in place Splatter will

become a single interface that is able to deal with lateral buckling, pipeline walking, on-bottom
roughness and free span assessments.
4. Integration with Simulia’s Isight

Simulia’s Isight is a Process Integration and Design Optimization (PIDO) application. Isight has
been heavily used in the automotive and aerospace industries and it is starting to get some renown
and traction in the oil and gas sector. Subsea 7 is engaged with the development of analysis tools
that benefit directly from Isight’s capabilities.
Splatter is one of several applications that are under development which will interface with Isight
to increase the efficiency of analysis processes and enhance the quality of the results. In essence,
the aim is to move from deterministic towards probabilistic pipeline assessments. Splatter
enhances model-building efficiency and results’ post-processing but it still only allows running
one case per time and the extraction of deterministic results. But because Splatter is fully
parameterized it is possible to work in close partnership with Isight and Splatter’s architecture was
developed with Isight integration as a major objective. The high-level keyword file aimed to
establish a quick and simple implementation of lateral buckling models inside the Isight Simcode
application component. The Simcode component per se is enough to pre-process, run and post-
process the lateral buckling analysis. Once the process is set, it is straightforward to run several
analyses driven by Design of Experiments (DOE), Exploration, Monte-Carlo and Six-Sigma
process components.
Isight process components have already been invoked by Subsea 7 for lateral buckling analyses
and the following aspects will be presented briefly in this paper:
• Tolerable Virtual Anchoring Spacing assessment;
• Welds stress ranges due to lateral buckle formation assessments;
• Global model probabilistic assessment.
Future papers will describe Subsea 7’s ongoing work with Isight in pipelines and other disciplines.
4.1 Tolerable VAS analysis using Isight
Tolerable VAS modelling is a repetitive process that requires a number of engineering hours in
order to find the maximum tolerable virtual anchoring spacing (VAS) that a buckled pipeline can
have whilst remaining within integrity limits.
Tolerable VAS analysis is performed for a specific set of pipeline data, pipe-soil interaction
parameters and operational conditions. VAS FE models are constructed, run and the resultant
stresses and strains compared to the allowable limits. Graphical output is helpful to confirm the
validity of the models and results. The pipeline length is incremented or decremented until the
allowable limits are reached. Within each simulation there are several steps that can be automated
with Isight.
In the following example, a single sleeper is being used in a VAS pipeline model to initiate lateral
buckling. The buckle has to occur at the middle of the VAS pipeline length to maximize feed-in
towards the buckle crown.

A suitable Isight process component to drive the tolerable VAS analysis is the “Exploration”
process. The analysis results in this example had linear responses and “Hooke-Jeeves” was
chosen as a suitable optimization technique. The Exploration component is set up with the
objective of making the DNV interaction ratio be equal to 1.0. The Isight process used in this
assessment is illustrated in Figure 15.
Figure 15. Isight process for tolerable VAS seeking

Two application components are used in the Isight process, namely Calculator and Simcode. The
Calculator component is used to calculate the position of the single sleeper in the model, which in
this case is always in the middle of the VAS pipeline length. The calculator component accesses
the pipe length input variable, divides it by two and uses this value as input for the Simcode
component.
The Simcode component is where Splatter software is implemented. The Splatter high-level
keyword file is read as ‘general text’ in the input tab (Figure 16). With this file, any variables from
a lateral buckling model can be parameterized.
Figure 16. Splatter Simcode input tab

The tolerable VAS execution sequence can be controlled by using the Simcode Command tab. A
sequence of Command prompts using ‘Windows Batch’ mode can be given to execute: Splatter
input file model building (1st line), analysis run (3rd line), calculating DNV interaction ratios and
generating the “DNVResults.txt” output file (4th line), extracting resultant information in text
format (5th line), and creating results plots (6th line), as illustrated in Figure 17 . The commands
call the execution of in-house Python scripts which are included in the Splatter package that is
rolled-out to design teams.
Figure 17. Splatter Simcode Command tab

The required output that determines the tolerable VAS is the maximum DNV interaction ratio
along the pipeline. The value is written in the “DNVResult.txt” file, which is read as an output
variable in the Simcode output tab (Figure 18). The additional extracted information is not
specifically read as output but may be saved in the Isight database and can be checked inside the
Isight Runtime Gateway. The same applies for the generated plots which are stored within the
Isight database.
In this specific case, the analysis is started and the Exploration process component drives the pipe
length. The pipeline length starting value was 800m which resulted in a code check value lower
than 1.0, therefore, the pipeline length may be increased, on this occasion up to 2900m at which
the interaction ratio exceeded 1.0. The iterations continued until the interaction ratio was satisfied.
A total of seven cases were run for this analysis. With a less tight error tolerance, the results could
be acceptable with four runs. Figure 19 shows the Runtime Gateway process run, and the
interaction ratio (DCC) variable is shown until convergence is achieved.
The extracted result files and plots are accessible through the database and the standard result
output is shown in Figure 20.

Figure 18. Splatter Simcode Output tab
When analyses were performed manually, three to five iterations were required to find only an
approximately correct value. It would be time-consuming to complete this task and error-prone
too. With the model prepared within Isight, possible because of the parameterization, it was only
required to prepare the model once. After this initial setup, only 40 minutes were required to run
the models, post-process the data and build graphs for all the iterations and to obtain a precise
solution. Were the same process to be performed manually, it would take at least one day to
complete and the results would only be approximate.
There may be cases for which the tolerable VAS varies along the pipeline route, depending on
combinations of pipe wall thickness and pipe-soil interaction characteristics, for example. Using a
“Design of Experiments” process component it would be possible to run all combinations
automatically by encapsulating the “Exploration” process.
Extending the concept, the same process could be used to run all combinations of input parameters
and create a large internal database. From such a database, approximations could be created and
results for tolerable VAS could be found instantly. Subsea 7 is working towards the formulation
of such a database but the number of input variables is significant and it is expected to take a long
time to achieve this objective.

Figure 19. Runtime Gateway results
Figure 20. Standard plots generated by Splatter and Isight

4.2 Other processes using Isight
The tolerable VAS process using Isight is one example of a development made by Subsea 7. The
tolerable VAS process is a basis for other processes involving lateral buckling assessments or
general pipeline analysis. For example, determining the maximum End of Life (EOL) stresses and
associated stress ranges along a pipe length is important for fatigue evaluation and Engineering
Critical Assessment.
The Isight process to assess Pipe-in-Pipe system EOL stresses and stress ranges is shown in Figure
21. The process component used is a “Design of Experiments” (DOE). At this stage of design the
tolerable VAS is already calculated and fixed and for this single VAS length it is required to
assess different full and partial temperature and pressure shutdown combinations for inner and
outer pipes of the PiP system. The DOE feeds input information to a Simcode component, with
Splatter integrated into the workflow. There is an additional Python script in the Simcode
command tab that extracts the EOL stress and stress range for each node along the pipeline. The
extracted data is used as input for the following Excel components that will post-process the data
in tabular and graphical (Figure 21) formats to facilitate the identification of the maximum stress
values. The Excel components access data combinations for inner and outer pipes sections, and
also design and operating system conditions. For this case, the combinations resulted in a total of
20 analysis runs with 80 combinations of results to be assessed. Incorporating Isight in the process
enabled running and post-processing of all results to be performed in a period of 5 hours.
Figure 21. EOL stress and stress ranges calculation Isight process and typical
graphical response in Excel components
Subsea 7 is benefiting from Isight with regard to probabilistic assessment of global lateral
buckling models. Currently in the industry, probabilistic assessments are based on analytical
(manual) calculations due to hardware limitations. It is only recently that finite element models
with Isight or other optimization tools have been used to calculate the probability and reliability of
buckle formation. However, this type of assessment is still being done with a mixture of
analytical methods and finite element VAS models.

Subsea 7 is developing probabilistic assessments using full route-length models including all man-
made triggers, initial out-of-straightness variations, and seabed unevenness. The reasoning for
using global models is that pipeline-seabed interaction, as-laid horizontal and vertical out-of-
straightness and buckle-site interaction may affect significantly the overall response.
The process built to assess this type of analysis uses the “Monte Carlo” process component to
drive the Splatter Simcode (Figure 22).
The pipe-soil axial and lateral resistance and pipe as-laid initial out-of-straightness are defined in
the Monte Carlo component as probabilistic distributions. In the Simcode command tab there is
an additional Python script that is able to count the number of buckles per run. This is required to
demonstrate the reliability of the engineered buckle initiators in the lateral buckling strategy.
The code check interaction ratios are verified along the route to ensure that for all possible
combinations the pipe integrity remains within the allowable limits. The results output will
include probabilistic distributions.
Figure 22. Global finite element lateral buckling model probabilistic assessment
The Isight probabilistic assessment is under development and initial results are expected by early
2015. Initial assessments demonstrated that hardware will not be a limiting criterion. The
performance in the later development stages will dictate if this probabilistic processing can be
maturely offered and used in future projects, with their inevitably challenging deadlines.
The potential is that this kind of assessment could drive pipeline design engineering to a new
level. There is strong demand to reduce project costs. The development of probabilistic solutions
could enable designs to achieve target safety levels whilst reducing unwarranted over-
conservatism.

5. Conclusions
• Lateral buckling of subsea pipelines may incur a high engineering cost during the
detailed design stage, depending on the complexity of the production requirements,
terrain and pipeline characteristics. Pipeline design engineers rely on finite element
analysis to assess the lateral buckling phenomenon.
• Subsea 7 has developed the Splatter software to enhance the efficiency in constructing,
running and post-processing Abaqus lateral buckling models.
• Splatter standardizes finite element models and output including graphical interpretation,
based on engineers’ experiences and best practice.
• Splatter is fully parameterized and the architecture works easily with Simulia’s Isight
software.
• Isight workflows have been implemented to determine tolerable pipeline VAS lengths as
a basis for all types of lateral buckling models.
• Automation with Isight can result in significant hours saved compared to manual
processing.
• Isight allows a wide envelope of input variables to be assessed resulting in a richer set of
results compared to deterministic assessments that are frequently used in the industry.
• The next step will be the probabilistic assessment of global finite element models which
may result in output that demonstrates realistic pipeline responses and which fully satisfy
target safety levels whilst minimizing unwarranted over-conservatism.
6. References
1. DNV, “Submarine Pipeline Systems - DNV-OS-F101”, Norway, 2013
2. Carr M., Sinclair F., Bruton D., “Pipeline Walking – Understanding the Field Layout
Challenges, and Analytical Solutions Developed for the SAFEBUCK JIP”, OTC 17945, May
2006.
3. Hobbs R.E., “In Service Buckling of Heated Pipelines”, Journal of Transportation
Engineering, Vol. 110, No 2, March 1984.
4. Hobbs R.E. and Liang F., “Thermal Buckling of Pipelines Close to Restraints”, Eighth
International Conference on Offshore Mechanics and Arctic Engineering, The Hague,
Volume 5, pp 121-127, 1989.
5. SAFEBUCK JIP, “Safe Design of Pipelines with Lateral Buckling – SAFEBUCK III”, United
Kingdom, 2011

Finite Element Simulation of Rotary Steerable
Drilling Systems Advance Understanding of
Damaging Downhole Shock and Vibration
Nader E. Abedrabbo, and Liam A. Lines
Weatherford International
nader.abedrabbo@weatherford.com
Abstract: The latest generation of rotary steerable drilling systems make it possible to reach
hydrocarbons located deeper and in more hostile environments than ever before. These systems
give the directional driller, standing some 10,000ft above on the rig floor, the ability to steer the
well path in real-time to reach the elusive ‘pay zone’. Rotary steerable drilling systems, which are
located at the end of the drilling assemblies, are exposed to extreme conditions. In addition to
temperatures in excess of 390°F and pressures up to 25,000 psi, they must also convey
compressive and bending loads, torque and rotation to the drill bit from the surface, all while
controlling the 3D trajectory of the wellbore.
Shock and vibration sensors highlight the severity of the downhole environment, but the root cause
of vibration is difficult to determine based on sensor data alone. Industry standard predictive
analysis techniques are incapable of detecting the effects of contact dynamics and real-world
testing has shown that these are among the most damaging to downhole drilling tools,
significantly reducing efficiency of the drilling process. With some deepwater rig rates exceeding
$500,000 per day, the most successful systems are those that drill fastest with the lowest number
of downhole failures.
Advanced finite element transient dynamic analysis and the manner in which it yields new insight
into downhole shock and vibration shows that a combination of 1D beam elements and real 3D
part geometries is critical to accurately predict the drilling assembly’s dynamic signature, while
still allowing for a timely solution process. This analysis enables the complex drilling assembly to
be modelled within the confines of the wellbore. The analysis also assesses dynamics associated
with rotating contact, such as lateral and torsional vibration, as well as the excitation of the
drilling assembly’s structural resonant modes. Transient dynamic analysis enables drilling
assemblies and downhole tools to be designed from the outset to minimize contact dynamics and
avoid excitation of damaging structural resonances. In addition to improved design of downhole
tools, the results of this analysis also show the drilling team what drilling parameters are optimum
to minimize and mitigate shock and vibration while drilling.
Keywords: Vibration, Drilling Dynamics, Contact Dynamics, Rotary Steerable, Downhole Tools,
Bottom Hole Assemblies, Oil & Gas.

1. Introduction
Rotary Steerable Systems (RSS) have become an essential part of today’s drilling marketplace.
They build upon the advantages of conventional directional drilling systems through offering
improved directional control, better well bore quality and the capability to drill extended reach
wells - typically where the horizontal displacement to vertical depth ratio for the well exceeds 2:1.
A conventional RSS assembly is connected directly to the drill bit and uses either a point or push
the bit steering system to control the trajectory of the wellbore while continuously rotating. It is
not uncommon for these assemblies to undergo high rotation speeds (in excess of 200 rpm) to
maximize drilling rates and allow for proper hole cleaning. The particular RSS discussed in this
paper comprises a drill bit with polycrystalline diamond compact (PDC) cutters, a fulcrum or pivot
stabilizer which contacts the formation and a hydraulic non-rotating steering unit.
Drilling tools operate in some of the most hazardous environments known to man with pressures
exceeding 25,000 psi and temperatures in excess of 390°F not to mention the corrosive and
abrasive formations and fluids that they are exposed to. As a result, drilling assemblies are
extremely expensive to develop and downhole damage can result in costly repair bills and non-
productive time, especially on a deepwater off-shore well where rig rates approach $500,000 per
day.
There is always a fine balance between optimizing drilling parameters to avoid tool damage and
drilling as fast as possible. These two usually cause conflict and understanding the limitations of
the drilling system and the point at which it becomes in imminent danger of failure are critical.
Typically, drilling companies rely on advanced sensors to determine the severity of the downhole
environment and the likelihood of tool failure, however, a significant effort within the industry is
seeing wider adoption of advanced modelling techniques to better understand what goes on
downhole, enabling improved operating techniques and improved drilling assembly design.
Typically, downhole tool damage and poor drilling efficiency is a result of shock and vibration.
Vibration reduces energy that would otherwise be used to shear and crush the rock. Unintended
dynamic behavior of the drill-string can have costly consequences both in terms of reduced rate of
penetration (ROP) and damage to drillstring components (Vandiver et al. 1990; Christoforou and
Yigit 1997; Spanos et al. 2002; Zhu et al. 2011). This study focuses on the application of advanced
modelling techniques to simulate and understand two lesser known vibration modes which are a
source of significant problems downhole: rotational dynamics especially high frequency torsional
oscillations (HFTO); and lateral shock and vibration where whirl (both forward and backward) is
of particular interest. HFTO and whirl typically go unaddressed due to the requirement for
advanced downhole sensors and modelling techniques outside of industry standard methods.
In this paper, an advanced finite element analysis model of two BHA systems was created in
Abaqus. The two BHAs are identical except for the placement of an extra stabilizer at a critical
position along the BHA. The effect of the extra stabilizer is shown to reduce the critical lateral
displacements and accelerations of the BHA in general and for other critical tools along the BHA
as well. Numerical results are compared to real experimental data for verification.

2. Conventional BHA Analysis
Drilling assemblies can range from simple to complex. A typical drilling assembly connects above
the drill bit and consists of a steering tool to control wellbore trajectory, a navigation tool for
obtaining inclination and azimuth, logging tools for determination of petrophysical and
lithographical formation properties, a telemetry tool which sends downhole data to surface by
electromagnet and /or mud pulse telemetry, and drill collars / heavy weight drill pipe used for
transition to the drill pipe above. BHAs can range from 10’s to 100’s of feet long and there is an
endless variety of BHA configurations depending on application.
Performing computer based BHA analysis is a vital part of any drilling campaign. It can be used to
detect damaging downhole conditions, both dynamic and static, that may occur during the drilling
process and gives insight into optimum stabilizer placement. Modelling can also be used as a tool
to analyze past failures and diversions from planned well trajectories. In both cases, computer
modelling offers a more cost effective alternative to performing experimental tests for each BHA
configuration.
Standard analysis methods developed for BHA analysis utilize a mixture of rigid-flexible
multibody mechanical systems using the component mode synthesis method (Pogorelov et al.,
2012).
BHA analysis has two main purposes:
1. Static Analysis: BHAs typically pass through highly curved sections of the well. These
impart cyclic bending moments in the BHA which could result in gross overload or
fatigue damage to downhole tools and the threaded connections between them. Static
BHA analysis calculates the equilibrium state of the BHA due to the effects of gravity,
hole curvature and axial compressive loads or weight on bit (WOB). Important results for
this type of analysis include: The shape of the deformed BHA, location of contact points
between the BHA and the wellbore and stress and bending moment distributions
throughout the BHA. The BHA and well plan can then be optimized quickly and
efficiently to minimize the chance of downhole tool damage. Figure 1 shows example
results of static analysis. The plot shows the deformed BHA cross section, contact forces,
bending moments and the von Mises stresses along the BHA length, respectively.

Figure 1. BHA Static Analysis Results.
2. Vibration Analysis: Industry standard analysis techniques such as linear vibration and
forced frequency response enable resonant modes to be determined and critical operating
parameters to be approximated. Linear vibration analysis of the BHA is performed using
kinematic and forced harmonic excitations. Vibration sources, such as the cutting
processes of the drill bit, contact point interaction with wellbore wall, rotation of
eccentric parts and operation of BHA components (mud motors, RSS assemblies, etc.)
are accounted for. The implemented analysis procedure, which is based on linearization
of the equations of motion, is widely used and with relatively low computational expense.
Figure 2 shows simulation results of the maximum lateral displacements as extracted
from frequency domain vibration analysis of a sample BHA. Utilizing the forced
frequency responses, the developed model has the ability to detect dangerous rotation
speeds in the BHA which can cause damaging downhole vibrations though excitation of
critical structural resonances.
Figure 2. Results of Linear Vibration Analysis.

In the past, historical and empirical data was heavily relied on to determine BHA suitability.
Nowadays even the most fundamental analysis methods as previously described have proved
invaluable and have been a major step in the right direction; however, there are some significant
limitations. Typically, these limitations are overcome by over stabilizing the assembly, which is a
common practice in rotor-dynamic problems. In the imperfect world of oil exploration,
stabilization - specifically the number of stabilizers and their clearance with the drilled borehole -
is under close scrutiny by drilling companies. Higher levels of stabilization are commonly
considered to increase the chance of getting stuck in hole when drilling mobile or swelling
formations. In addition, stabilizers can restrict the flow of rock cuttings and cause pressure spikes,
which can damage formations and downhole tools. Stabilizers may also increase torque and drag
putting higher loads on surface power systems and reducing the depth or lateral extension of the
well. Particular focus recently has turned to efficient and effective stabilization rather than
maximum stabilization and in order to evaluate this, more advanced time domain based simulation
techniques are required.
3. Motivation for Time Domain Analysis

The main advantage of frequency domain vibration analysis is the fast solution time in which the
analysis can be performed. Each analysis case can be completed within minutes. The fast solution
time is particularly useful as linear vibration analysis typically requires the response behavior to be
modelled at multiple WOBs and RPMs for a number of locations in the well.
Aside from the relatively fast solution time, linear vibration has some notable limitations:
• Frequency domain analysis is performed using linearized (simplified) equations of
motion, therefore, time varying or transient effects, such as those resulting from contact,
are not possible to determine from frequency domain vibration analysis.
• Due to the analysis methods used (frequency domain), time varying contact with the
wellbore boundary is not possible. Therefore contact regions have to be manually
imposed and are fixed during linear vibration analysis.
• Limited outputs are available: lateral, axial and torsional displacements only.
• No acceleration data is available.
• The analysis is sensitive to the location and amplitude of excitations applied to the BHA.
It is extremely difficult to determine what the true input excitations should be, especially
when multiple contact points exist and limited sensor data is available.
In contrast to frequency domain techniques, time domain analysis allows contact between the
BHA and borehole to occur “naturally”. These may change during the course of the analysis. The
advantages of time domain analysis are:
• The model is analyzed using nonlinear equations of motion, accounting for real physics.
• Forces occur naturally due to contact with the wellbore rather than relying on manual
definition by user.

• A large set of output data, e.g. displacements, accelerations (lateral, torsional and axial)
are available.
• Special contact formulations can be applied to account for drill bits & stabilizers (bit-rock
interaction).
• Results can be easily correlated to real world sensor data – vibration severity is generated
in terms of acceleration which is directly measured by downhole sensors.
• Interpretation of results can highlight the presence of contact dynamics such as backward
whirl and the interaction/coupling between dynamic modes.
• Time domain analysis is based on real-physics and therefore is far more representative of
the downhole environment.
There are, however, some notable drawbacks:
• The solution time can be considerably higher than frequency domain analysis (hours
compared to minutes).
• It requires the user to have a special skill set, but in some circumstances the process can
be semi-automated.
• It is not suitable for wide distribution; only used in special cases when time allows for
extra solution time.
4. FEA Time Domain Analysis of BHA Systems

Time domain analysis was conducted on two BHAs. Both consisted of an 8.5 inch diameter PDC
drill bit, point-the-bit RSS, integrated directional sonde (IDS), multi-frequency resistivity (MFR)
tool, and drill collars. The only difference between the two BHAs is the addition of a stabilizer to
the top end of the MFR (Figure 3). By analyzing these two systems using the time domain finite
element analysis a better understanding of the effects of stabilization can be achieved on the
overall stability of the test BHA.
Figure 3. Test BHAs: (a) Case 01: Non-Stabilized MFR Model; (b) Case 02: Stabilized MFR Model.

There are several challenges in efficiently applying the finite element analysis to the BHAs
described in Figure 3. The ratio of the model’s length to the typical diameter is in the order of 105,
which prohibits using shell and solid elements. Finite element analysis using Abaqus provides
several tools for modelling the components of the BHA. The conference paper by Abedrabbo et al.
(2012) provides details of the methods and requirements for modelling BHA systems. There are
several options for creating the components of the BHA model:
4.1 Beam Only FEA Models:

In this method, all the BHA components are modeled using simplified beam elements (beam in
space). Beam elements provide a good approximation to the geometry and stress distribution,
resulting in an acceptable combination of accuracy and solution time. However, beam elements
are most suited to static BHA analysis (no rotation applied) and frequency domain analysis. Due
to the fact that beam elements are represented by a line in space where the cross section is a
homogeneous circular section, they are not capable of capturing the complex contact dynamics
that occur with the wellbore during rotation. This is particularly important for BHA parts with
non-uniform geometry such as stabilizers. Figure 6a is the actual geometry of the stabilizer as
created in a CAD program while Figure 6b is the graphical representation of the beam elements
with the beam profile artificially rendered. As seen, the stabilizer representation using beam
elements does not capture the complex contact geometry of the real part, and hence, the time
domain response would not be realistic.
Figure 6. a) Actual pivot stabilizer geometry; (b) Beam element representation.
Figure 7 shows the lateral displacement of the beam-only model at the center of the RSS shaft
which is a critical design point (Figure 5). The analysis does not exhibit the expected oscillations
due to the rotation of BHA with time. After an initial lateral displacement due to application of
gravity on the model, the beam rests on the bottom of the wellbore and rotates without generating
any excitations. Although tangential friction is included, the model is unable to replicate the
complex contact imposed by the blades of the stabilizers due to the smooth representation of the
beam surface.

Figure 7. Lateral displacement results for beam elements at center of RSS shaft using time domain
method.
4.2 Solid or Shell Element Models:

In order to be able to capture the true effects of the stabilizer blades as they make contact with the
wellbore, the components of the BHA must use solid or shell elements. These element types, with
careful application, can capture the real contours of the part. However, although using full 3D
solid or shell elements would provide the highest quality of results for the model, the fact that the
BHA is very long compared to the required element sizes, the degrees of freedom of the model
(total number of nodes and elements) would be substantial. This would render the solution time
and computational resources unfeasibly high.
4.3 Mixed Beam-Solid Models:

An optimized strategy to perform time domain finite element analysis is to adopt a mixture of
beam and solid (or shell) elements. Beam elements can be used for uniform parts of the BHA, e.g.
drill pipes and drill collars, without any loss of reliability. Parts that have more complex non-
uniform geometry, e.g. stabilizers, can be created using solid or shell elements. Kinematic
coupling is then used to connect the different types of elements so that motion and excitations are
transferred correctly. This method provides the best compromise, important contact information is
captured by the real part geometry and analysis time is improved by using beam elements
wherever possible.
Figure 9 shows the BHA model with the locations where 3D geometry was used to represent
modelled parts. The other parts in the model were represented with beam elements.

Figure 9. BHA model with location of 3D geometry parts.
5. FEA Model Specifics

The following are details of the finite element model used in the time domain analysis:
• The 3D parts with complex geometry were modeled as discrete rigid bodies with a
mixture of triangular and quadrilateral elements. The rigid element assumption is
adequate for this analysis as plastic deformation is not anticipated in the simulation.
• Uniform pipe parts were modeled using Timoshenko beam (shear flexible) elements.
• The wellbore was modeled as an elastic body with properties of mild steel. This provided
some degree of energy absorption (damping) to the model. A wellbore that is 0.5 in
larger than the drill bit and stabilizers was used in the analysis which is considered a
worst-case scenario for clearance between BHA and the wellbore.
• The model consists of 234,761 nodes (1,408,566 DOF) and 237,141 elements.
• The BHA was modeled in two orientations:
o Horizontal where gravity acts perpendicular to BHA length; and
o Near vertical with 1.0° inclination.
• RPM was applied at the top of BHA where it was ramped quickly to the desired value
(shown in Figure 11).
• An exponential decay friction model (Static-Kinetic with Exponential Decay) was used.
• Boundary conditions applied were as follows:
o Axial movement of the bit was restricted.
o All other nodes were free.
o An elastic bushing was applied at the top of the model to simulate the stiffness
of the drill pipe to surface.
• The dynamic explicit analysis method with double precision was used (minimal mass
scaling used).

• Utilizing a server machine with 24 CPUS, the model solution time ranged between 12-14
hours (depending on model).
• Output frequency for simulation animations was 100Hz.
• Output for the required nodes of interest was recorded at 1000HZ. In this paper, results
are shown only from 0 to 250Hz frequency as this is the effective bandwidth of the
dynamics sensor (Lines et al., 2014).
6. Time Domain Analysis Results

6.1 Horizontal Orientation
Figure 5 shows the lateral vibrations (y-axis) for both analyzed models (MFR with and without
stabilizer) as recorded at the center of the RSS shaft. Figure 6 shows the results obtained for both
models at the end of the MFR unit. In horizontal orientation gravity is applied along the Y-axis.
As seen from the two figures, initially, gravity acts to pull the BHA to be in contact with the
wellbore. However, as contact between the non-uniform geometries and the wellbore progresses,
the RSS shaft begins to oscillate. Although this oscillation cannot be completely eliminated, it is
beneficial if it is reduced as much as possible to minimize the workload on the RSS steering unit
and to improve directionality stability. In the model without the added MFR stabilizer it can be
observed that the RSS shaft exhibits larger amplitude oscillations. Although the stabilized MFR
was located more than 45 ft above the RSS unit, the effect on the overall stability of the system is
apparent.
Figure 5. Lateral vibrations (y-axis) for horizontal BHA case at middle of RSS point.

Figure 6. Lateral vibrations for horizontal BHA case at end of MFR unit. The larger lateral displacement
of the MFR without stab case is due to MFR sagging as a result of gravity.
Figure 7 shows the angular accelerations (torsional vibration) at the center of the RSS shaft.
Figure 8 shows the FFT (PSD) of the torsional accelerations as show in Figure 7. From the FFT
plot, we notice that the dominant frequency is at 59.5 Hz, which corresponds well with observed
field data for similar BHA as shown in Lines et al. (2013) (see Figure 12).
Figure 7. Angular accelerations at center of RSS shaft.

Figure 8. FFT of angular accelerations at center of RSS shaft.
6.2 Near Vertical Orientation:

In the near vertical orientation, the damping effect provided by gravity on the MFR is no longer
present. Initially the MFR oscillates at small amplitudes within the wellbore. As momentum and
energy levels increase, the MFR enters a resonant frequency and quickly transitions into a
destructive vibration mode. Figure 9 shows the results at the middle of the RSS shaft while Figure
10 is a comparison of the lateral vibrations near the end of the MFR. The figures show the extreme
lateral vibration resulting from the un-stabilized MFR and how, in contrast, the addition of a single
stabilizer significantly reduces vibration amplitude.

Figure 9: Lateral vibrations at middle of the RSS shaft for vertical orientation.
Figure 10: Lateral vibrations at end of MFR unit for vertical orientation.
6.3 Whirl Detection:

One of the main objectives from time domain analysis is to investigate if finite element models are
capable of detecting whirl, both forward and backward. Our area of interest for whirl detection
and elimination is at the RSS stabilizer (as shown in Figure 3) but the technique is equally
applicable to all rotating elements which contact the stationary borehole. By extracting the lateral
displacements for the RSS pivot stabilizer and plotting them versus time, the presence of whirl can
be clearly observed.
Figure 11 shows the plot of backward whirl initiation for case #1 (without MFR stabilizer) and the
full history for case #2 (with MFR stabilizer). Both the applied RPM rotation and the nodal history
rotational direction for backward whirl are indicated. From the plot several things are noticed:
• Backward whirl is clearly visible in the analysis for Case #1 (MFR without stabilizer) as
indicated both by the circular pattern of the nodal history and also by the fact that the
nodal rotational history is opposite that of the applied RPM.
• Adding a stabilizer to the top MFR (Case #2) has been effective at eliminating the
dangerous backward whirl. The stabilizer rolls benignly at the bottom of the borehole.

Figure 11: Lateral displacement history at center of RSS pivot stabilizer for Case #1 and full nodal
history for Case #2.
7. Experimental Results
Weatherford has several reliable downhole vibration sensors capable of detecting several modes of
downhole vibration and transmitting pertinent data to the surface in real-time (Lines et al. 2014).
During a drilling study with a BHA similar to the one discussed here, the downhole dynamics
sensor detected HFTO. Figure 12 shows a rapid sample captured at the initiation of HFTO while
drilling with an 8.5-in bit through a formation of chalk and marl (both carbonates). Note the
presence of a strong 60 Hz frequency content in tangential accelerometer. This data correlates well
with the 59.5 Hz torsional vibration frequency predicted by the analysis as shown previously
(Figure 8).
Figure 12: Rapid sample data showing HFTO.

8. Conclusion
Abaqus nonlinear finite element analysis capabilities provided a very powerful tool in predicting
the dynamic behavior of downhole drilling tools. The ability to integrate real 3D parts in the
analysis enables the simulation of real physics, especially those relating to contact dynamics.
Time domain analysis of the BHA, although computationally expensive compared to industry
standard techniques, offers valuable insights that linearized frequency-domain vibration analysis is
incapable of. Clear indication of lateral vibration, damaging whirling modes and HFTO at areas of
interest can be predicted before drilling an actual well. It was also shown that the numerical
predictions matched experimental results for the same system.
Future improvements on the modelling methods developed and presented here include adding rock
removal due to contact (hole enlargement), adding fluid-structure interaction (FSI) to gauge the
damping effects of fluid and modelling of nonsymmetrical parts. Also, analysis automation is a
priority in order for wider user participation.
9. References
1. Abedrabbo, N., Ring, L. and Gandikota, “Automation of Static and Dynamic FEA Analysis of
Bottomhole Assemblies”, in 2012 SIMULIA Customer Conference, Rohde Island, USA.
2. Christoforou, A.P. and Yigit, A.S. “Dynamic Modelling of Rotating Drill Strings with
Borehole Interactions”. Journal of Sound and Vibration, 206 (2): 243-260, 1997.
3. Lines, L. A, Stroud, D. R. H., Coveney, V. “Torsional Resonance - An Understanding Based
on Field and Laboratory Tests with Latest Generation Point-the-Bit Rotary Steerable System”.
Society of Petroleum Engineers, 2013. doi:10.2118/163428-MS.
4. Lines, L. A., Mauldin, C. L., Hill, J. W., & Aiello, R. A. “Advanced Drilling Dynamics
Sensor Allows Real-Time Drilling Optimization, Damage Prevention and Condition
Monitoring of RSS and LWD BHAs”. Society of Petroleum Engineers, 2014.
doi:10.2118/170586-MS.
5. Pogorelov, D., Mikheev, G., Lysikov, N., Rin, L., Gandikota, R., and Abedrabbo, N. “A
Multibody System Approach to Drill String Dynamics Modeling”. Proceedings of the ASME
2012 11th Biennial Conference on Engineering Systems Design and Analysis, ESDA2012,
Nantes, France.
6. Spanos, P.D., Payne, M.L., and Secora, C.K. “Bottom-hole Assembly Modeling and Dynamic
Response Determination”. Journal of Energy Resources Technology, 119 (3): 153-158, 1997.
doi: 10.1115/1.2794983.
7. Vandiver, K.J., Nicholson, J.W., and Shyu, R.-J. “Case Studies of the Bending Vibration and
Whirling Motion of Drill Collars”. SPE Drill Eng. 5(4): 282-290. SPE-18652-PA, 1990. doi:
10.2118/18652-PA.
8. Zhu, W., & Di, Q. “Effect of Prebent Deflection on Lateral Vibration of Stabilized Drill
Collars”. Society of Petroleum Engineers, 2011. doi:10.2118/120455-PA.

Abaqus Analysis Methods on Highly Restrained
Pipeline with Soil Berm Formation
Shulong Liu*, Emil Maschner, Teng Zhang, John Smyth, John Li
Wood Group Kenny, Compass Point, 79-87 Kingston Road, Staines-upon-Thames, TW18 1DT,
UK.
*Shulong.Liu@woodgroupkenny.com
Abstract: Wood Group Kenny has developed an advanced pipe-soil interaction subroutine to
accompany the analysis of pipeline lateral buckling and axial walking in Abaqus. The pipe-soil
interaction subroutine takes into consideration both variations in axial friction and the lateral soil
berm formation mechanism in the analysis of lateral buckling and axial walking of pipelines
subject to cyclic loading. The lateral soil berm formation scheme accounts for differential berm
growth on the soil berm front and rear faces, berm resistance accumulation, mobilization distance
and the residual ‘sweep’ friction variation in production operation cycles. The subroutine
monitors nodal displacement history around a buckle periphery, and has been shown to simulate
realistic, cyclic stress ranges, soil berm breakout mechanism during rare high temperature
/pressure loading events and the influence of lateral soil restraint in global pipeline walking
studies.
Due to the complexity of the berm formation mechanism, it has been found that small differences
in the initial lateral buckle profile generated using various Abaqus analysis methods can have a
significant impact during subsequent cyclic loading. This can be important as it has the potential
to alter the berm accumulation profile around the developing lateral buckle; and more
importantly impact on the pipeline behaviour with number of cycles. Sensitivity checks, using
various numerical analysis methods available to Abaqus users, to identify the actual pipeline
displacement behaviour in production operation cycles and following mid-life or late life hot
oiling events, are compared in this paper. A buckle formation sensitivity check using various
analysis methods, including static analysis with stabilizer values, static analysis with various
damping values, and the less-favoured static/ quasi dynamic analysis with RIKS method has been
performed.
Keywords: Abaqus FEA, Pipe-Soil Interaction, Soil Berm Formation, Pipeline Lateral Buckling,
Cyclic Loading, Statics, Dynamics
1. Introduction
An on-bottom submarine pipeline system subject to high pressure and temperature loading is
susceptible to Euler buckling in the horizontal plane, i.e. lateral buckling, and indeed a planned or
controlled lateral buckle can be a safe and effective way to accommodate the thermal expansion of
a hot pipeline. However, as high stresses and strains can develop in such buckles a simple stress
based design approach will not suit the design of pipelines in deepwater. Accordingly, within

current oil and gas guidelines the conventional load limit control maybe replaced by a strain limit
control (SAFEBUCK III, 2011).
In theory, a single analysis of the entire pipeline could demonstrate that all failure modes will be
avoided, but in practice there are considerable uncertainties, including variable pipe-soil
interaction and as laid out-of-straightness that need to be taken into account. Thus choosing one
input to be conservative for one a failure mode can underestimate the effect of another. The pipe-
soil interaction model presented in SAFEBUCK III Design Guideline Appendix B, most notably
shows that lateral buckling, axial walking, route curve pull-out and flowline anchoring are all
extremely sensitive to the pipe-soil interaction forces. Consequently this presents significant
uncertainties to predicting the operational displacement response of pipelines to ensure their
through-life integrity.
In practice, analytical models may be used for simple configurations such as an exposed pipeline
on an even seabed and a buried pipeline. However, for more heavily loaded pipelines on soft or
uneven seabed with significant soil restraint advanced finite element methods are required. Indeed
industry codes and standards emphasis that lateral buckling in detailed design of offshore pipelines
must be modelled using a non-linear finite element method with quality assurance and verification
(DNV-RP-F110 2007, DNV OS F101 2013 and SAFEBUCK III 2011). Many commercial finite
element analysis (FEA) packages are available, including Abaqus, which has the capability to
model snap buckling behaviour and the appropriate non-linearity including non-linear geometry
response, material non-linearity and non-linear elastic-plastic pipe-soil interaction forces with user
defined pipe-soil interaction subroutines. These capabilities present Abaqus users with an
opportunity to meet industry requirements in running cost effective simulations involving variable
complex input parameters to ensure the operational displacement integrity of oil and gas pipelines.
There is still however a gap between the engineering needs of the oil and gas industry and
established engineering theory that has been written into many functions and subroutines
employed in FEA programs such as Abaqus. The oil industry is a fast growing sector, and subsea
oil field development needs state-of-art advanced simulation model with reliable and accurate
prediction that produces the right results first time. Wood Group Kenny has developed a pipe-soil
interaction subroutine (PSI), written in FORTRAN, that can be embedded within an Abaqus
global buckling FE model. The PSI subroutine has been successfully validated for the analysis of
advanced pipeline lateral buckling problems in recent Wood Group Kenny projects.
The advanced pipe-soil interaction subroutine is to accompany pipeline lateral buckling and axial
walking analyses in Abaqus. The PSI subroutine takes into consideration both axial friction
variation and the lateral soil berm formation mechanism for pipeline lateral buckling and walking
analyses subject to production operation and pipeline service cyclic loadings. The lateral soil berm
formation takes into account differential berm growth on the soil berm front and the rear faces,
berm resistance accumulation and mobilisation distance and the residual ‘sweep’ friction variation
in loading cycles. The subroutine can monitor nodal displacement against its past displacement
history around the buckle periphery. With regards to through life oil and gas pipeline design the
subroutine has been shown to simulate realistic, cyclic stress ranges, soil berm breakout
mechanism during rare high temperature /pressure loading events and the influence lateral soil
restraint in global pipeline walking studies.
Due to the complexity of the berm formation mechanism it has been found that small differences
in the initial lateral buckle profile generated using various Abaqus analysis methods can have a
significant impact during subsequent cyclic loading. This can be important as it has the potential

to alter the berm accumulation profile around the developing lateral buckle and increasingly
impact on the analysed pipeline behaviour with number of cycles. To identify the actual pipeline
displacement behaviour in production operation cycles and following mid-life or late life hot
oiling events, sensitivity checks using the various numerical analysis methods available to Abaqus
users are presented and compared in this paper. The analysis methods employed for the buckle
formation sensitivity checks include static analysis with default ‘stabilize’ value, static analysis
with specified ‘stabilize’ values, static analysis with ‘direct damping factor’ values, static analysis
employing RIKS method and dynamic analysis.
Illustration of lateral buckle formation and soil berm development

To aide understanding Figures 1, 2 and 3 illustrate the typical soil berm formation process
occurring with cyclic start-up / shutdown operation (heat-up/cool-down). It should be noted that
the initial lateral buckle is formed as a consequence of pipeline out of straightness and high
pressure and temperature induced compression loading. The Figure 3 hysterisis plot shows an
initial soil breakout resistance to be overcome by the embedded pipe upon first loading which
results in the initial buckle formation.
6
Displaced pipe profile upon heat-up

5
Recovered pipe profile upon cool-down
4
Lateral Displacement (m)
-1
-2
980 1000 1020 1040 1060 1080 1100 1120 1140 1160 1180
Distance along pipeline (m)
Figure 1. Lateral buckle profiles seen during heat-up and cool-down

Figure 2. Continuum modelling of soil berm accumulation as a function of soil
type, pipe weight, embedment and the extent of cyclic movement (S. Yu and I.
Konuk, 2007)
Figure 3. Illustration of first load breakout during buckle formation and subsequent
development on the advance and retreat pipe faces during cyclic operation
(SAFEBUCK III, 2011)
2. Methodology of the analysis

Pipeline global buckling and eventually local collapse of pipelines laid in deepwater seabed can
occur for a variety of reasons including excessive strain with associated stress/strain concentration
sites, fatigue, weld fracture aggravated by; high temperature, high pressure, product slugging and

the pipeline-seabed interaction. In this section, static analysis and dynamic analysis methods
suitable for global pipeline buckling and which are readily available to Abaqus users and are
reviewed (Abaqus User Manual).
Pipeline-soil interaction in production operation cycles is an important factor that can dominate
the pipeline fatigue and bending strain conditions. Results using the advanced Wood Group Kenny
pipe-soil interaction FORTRAN subroutine (PSI) are presented in this study.
2.1 Overview of the suitable analysis methods in Abaqus for global pipeline
buckling
Results presented in the paper were generated as an outcome of an offshore subsea pipeline design
project completed by Wood Group Kenny, Detail comparison of the results obtained using Abaqus
with other commercially available FEA tools is outside the scope of this study, and thus the
effectiveness of this analysis method in other commercially available tools is not reviewed.
Given the complexity of including inertial effects upon first loading a static stress analysis can be
used in Abaqus. In a static stress analysis time-dependent material effects such as creep, swelling
and viscoelasticity are ignored and conversely rate-dependent plasticity and hysteretic behaviour
for hyper-elastic materials can be taken into account.
To ensure as accurate and representative as possible results from FEA, a sensitivity study
employing adaptive automatic stabilisation with default stabilisation energy tolerance, ‘damping
factor’ based on the dissipated energy fraction, ‘direct damping factor’ and RIKS method has been
performed. While the model is stable, viscous forces and, therefore, the viscous energy dissipated
are very small, thus, additional artificial damping has no effect. If a local region becomes unstable
the local velocities increase and, consequently, part of the strain energy released is dissipated by
the applied damping. The ‘damping factor’ is determined in such a way that the dissipated energy
for a given increment is a small fraction of the extrapolated strain energy compared with the
previous increment. The fraction is called the dissipated energy fraction and, if a value is not
specified by the user, a default value of 2.0 × 10-4 is applied.
The RIKS method is generally used to predict unstable, geometrically nonlinear collapse of a
structure. For post buckling problems involving loss of contact (in this case the momentary loss of
contact between the pipelines and seabed as the pipeline buckles) the RIKS method will usually
not work unless inertia or viscous damping forces, such as those provided by dashpots are
introduced in a static or dynamic analysis to stabilize the solution.
The choice of appropriate method depends on the details of the analysis and the problem at hand,
below is a brief description of each of the methods available to Abaqus users (Abaqus User
Manual).
1. Adaptive automatic stabilization with the default stabilization energy tolerance in static
stress analysis
Adaptive automatic stabilisation with default stabilisation energy tolerance can be defined in
Abaqus using the following keywords:
*STATIC, STABILIZE
The default accuracy tolerance used by the adaptive automatic stabilisation scheme is 0.05. If
the accuracy tolerance is not specified but a dissipated energy fraction with the default value

of 2.0 × 10-4 is used, the adaptive automatic damping algorithm will be activated
automatically with an accuracy tolerance of 0.05.
Whenever automatic stabilisation is used in a problem, the following criteria should be
checked to ensure that an accurate solution is obtained:
I. For a ‘damping factor’ calculated using the dissipated energy fraction it
necessary to check the factor to ensure that a reasonable amount of damping has
been applied.
II. Compare the viscous forces (VF) with the overall forces in the analysis and
ensure that the viscous forces are relatively small compared with the overall
forces in the model.
III. Compare the viscous damping energy (ALLSD) with the total strain energy
(ALLIE), and ensure that the ratio does not exceed the dissipated energy fraction
and adopts a reasonable value
IV. The ‘damping factor’ is neither too small, thus not controlling the instability, or
too large, thus leading to inaccurate results.
2. ‘Damping factor’ based on the dissipated energy fraction in static stress analysis
The non-default dissipated energy fraction for automatic stabilisation can be defined directly
using the following options:
*STATIC, STABILIZE=dissipated energy fraction
As mentioned, the ‘damping factor’ is determined using the dissipated energy fraction. The
‘damping factor’ calculated from the dissipated energy for a given increment is a small
fraction of the extrapolated strain energy compared with the previous increment.
3. ‘Direct damping factor’ in static stress analysis
The ‘damping factor’ can be specified directly in the static analysis as follows:
*STATIC, STABILIZE, FACTOR=damping factor
A reasonable estimate for the ‘direct damping factor’ is generally quite difficult to make
unless a value is known from the output from previous runs. The ‘damping factor’ depends
not only on the amount of damping but also on mesh size and material behaviour.
Unfortunately, there is no guarantee that the value of the chosen ‘damping factor’ is optimal
or even suitable in some cases.
During analysis it may, in some situations, be necessary to increase the ‘damping factor’ thus
relaxing the convergence criteria if the system is unable to converge. Conversely it may be
necessary to decrease the ‘damping factor’ if a high ‘damping factor’ is distorting the
solution. The former would require the user to rerun the analysis with a larger ‘damping
factor’, while the latter would require the user to perform post-analysis comparison of the
energy dissipated by viscous damping to the total strain energy. Therefore, obtaining an
optimal value for the ‘damping factor’ is a manual process requiring trial and error in the
‘direct damping factor’, until a converged solution is obtained and the dissipated stabilisation
energy is sufficiently small.
4. The RIKS method in static stress analysis

The RIKS method is generally used to predict unstable, geometrically nonlinear collapse of a
structure. It is often used to speed up the convergence of ill-conditioned or snap-through
problems that do not exhibit instability.
The RIKS algorithm cannot obtain a solution at a given load or displacement value since
these are treated as unknowns, rather the simulation terminates at the first solution that
satisfies the step termination criterion.
To obtain solutions at exact values of load or displacement, the solution must be restarted at
the desired point in the step and a new non-RIKS step must be defined. Since the subsequent
step is a continuation of the RIKS analysis, the load magnitude in that step must be given
appropriately so that the step begins with the loading continuing to increase or decrease
according to its behaviour at the point of restart. If a RIKS analysis includes irreversible
deformation such as plasticity, the point of restart should be ensured to occur at a point in the
analysis where the load magnitude is increasing if plasticity is present.
As the pipeline buckles, a momentary to permanent loss of contact between the pipeline and
seabed surface take place, the RIKS method is not able to analyse this kind of problem
(SIMULIA Release Authorization and Stipulations, 2013). Consequently the analysis of
global pipeline buckling and post-buckling problems may not be so well defined by the RIKS
method.
5. Dynamic Analysis
Abaqus offers several methods for performing dynamic analysis of problems in which inertia
effects are considered. Dynamic analysis is outside the scope of this project; hence it will not
be discussed further in detail. However, for the benefit of the reader, a transient dynamic
analysis is defined using the keyword
*DYNAMIC
This option is used to provide direct integration of a dynamic stress/displacement response in
Abaqus and is generally used for nonlinear cases. The maximum time increment allowed is
the time period of the step divided by 100, including transient analysis. Time period of the
step and the loading history within the time step are found to be critical issues for dynamic
analysis of the pipeline global buckling. This is very difficult to measure on site, and to be
verified thereafter.
2.2 Soil berm formation verification
The berm formation mechanism is defined within the PSI subroutine in terms of berm resistance,
berm width in front and rear of the berm peak location (mobilisation distance) and the mid-sweep
resistance between the front and rear berms. Typically these cyclic soil berm parameters are
determined from field data and testing curves supplied by specialist geotechnical consultancies. In
certain less critical cases they can be approximated using the methodology described in
SAFEBUCK III.
Within the WGK sub-routine the parameters are calculated independently for the advance and the
retreat directions for individual nodal locations along the pipeline, and are functions of the
production cycle number (counted from the latest berm breakout in a given direction) and the
displacement of the pipeline between the berm in the front and the rear.

The berm strength, berm mobilisation and mid-sweep resistance are written into the PSI
subroutine as functions of both the cycle number and the displacement history using the original
measured data from laboratory. These functions which represent the berm data sampled at
different levels of cyclic displacement with number of cycles; say 5.0 diameters (5D), 1.0 diameter
(1D) and 0.1 diameters (0.1D) displacement. The mathematical curve fits used in the WGK PSI
subroutine presents a reasonable level of agreement with the original supplied data.
As part of the validation process, a single element model and a full model are used to verify that
the berm response is adequately modelled by comparing the interpreted FE model and the given
curves. The results generally support the modelling scheme (Figures 4 to 6); however, it appears
that the berm mobilisation may lag the expected response by 1 cycle (Figure 6). Whilst not
believed to be critical it is considered that the 1 cycle lag may help to stabilise the buckle growth
slightly earlier and hence result in a slightly stronger berm under hot oil conditions.
The berm resistance, mid-sweep resistance and the berm mobilisation curves are shown in Figures
4 to 6. In Figures 4 through to 6, the displacement is unified to dimensionless by dividing by the
pipe diameter. The data points marked _Given are associated with the data used in the PSI
subroutine as an input for the FEA. The curves marked _FEA represent the data generated in the
Abaqus FE computational procedure from a user defined formula and outputted from the Abaqus
FE model.
Figure 4. Plot of showing the variation in berm resistance with pipeline sweep
distance and cycle number

Figure 5. Plot of showing the variation of the mid-sweep residual resistance with
pipeline sweep distance and cycle
Figure 6. Plot showing the variation in the berm mobilization distance with pipeline
sweep distance and cycle number
Figure 7 illustrates at the apex of a lateral buckle the cyclic build-up of soil berm friction in the
first 20 cycles of operation. As can be seen from initial buckle formation (cycle 0) the buckle
profile continues to grow for some 11 cycles of heat-up / cool-down operation continuously

breaking out of the small soil berm on each heating step. From cycle 11 the buckle form ceases to
grow further in the advance direction enabling the soil berm to develop with subsequent cycles.
The cyclic displacement pattern and the lateral reaction of the pipe against the accumulated soil
berm resistance at the buckle apex are also shown in Figure 7.
The theory and pipeline engineering associated with the Wood Group Kenny pipe-soil interaction
simulation subroutine is outside the scope of this paper, and consequently detailed discussion of
the development methodology of the subroutine is will not be presented here.
Figure 7. Typical results plot at the apex of a lateral buckle showing the pipe
reaction against accumulated soil berm resistance and cyclic displacement.
3. Results from the analysis

Given the problem at hand and after reviewing the suitability of each of the built-in Abaqus
analysis methods for analysing global buckling of offshore pipelines (refer to Abaqus User
Manual), the static analysis with damping based on the dissipated energy fraction was used to
achieve the project deliverables. The other methods discussed earlier in the paper are employed in
this study as a sensitivity check.
In this study, initial concerns arose over different post hot-oil results seen from cases with similar
first load responses. To illustrate the effect of the different Abaqus analysis methods on the results
of pipeline global buckling analysis, a 10” pipeline on a flat seabed with a VAS length of 2160m
and an initial imperfection of 1m at the planned buckle apex is used. The pipeline lateral profile is
established in first production (first load), this initial profile will be further stabilized in production

heat-up and cool-down operation cycles. After 20 cycles of normal production operation, the
pipeline will be serviced with hot oil causing it to push into the yielding soil berm and impose
high bending strain on the pipeline.
The uses of the analysis methods, i.e. either static or dynamic, vary only in the first production
operation. The effect of each of the analysis methods is expected to be isolated in the first
production operation and then subsequently the effect of the various methods is propagated to the
hot oiling step through the 20 production operation cycles. In pipeline analysis, the minimum and
maximum strains are calculated and compared at all the peripheral integration points on elements
at various locations along the pipeline. Note that along the pipeline the minimum strain is not
necessarily a compressive strain and the maximum strain is not necessarily a tensile strain,
however for convenience, in this paper, the minimum strain is termed the maximum compressive
strain and the maximum strain is termed the maximum tensile strain. The strain value at the buckle
apex refers to the longitudinal maximum tensile/compressive component of mechanical strain
along the pipeline axial direction, unless otherwise stated.
3.1 Sensitivity check of the ‘stabilize’ value in static analysis
Results presented in this section demonstrate the effect of the ‘stabilize’ values in Abaqus static
analysis for the prediction of pipeline behaviour in production operation and during service. The
static analysis at first load is set with the adaptive automatic stabilisation, for all other cases the
‘stabilize’ function takes values ranging from 10-1 to 10-10.The adaptive automatic dissipated
energy fraction is set adopt a default ‘stabilize’ value of 2×10-4 in Abaqus static analysis (refer to
Abaqus User Manual). Results from this case are used, in this study, as the base case and labelled
as DEFAULT.
The effect of varying the ‘stabilize’ value from 10-1 to 10-10 on the lateral displacement and the
maximum compressive strain at buckle apex are summarised in Table 1. It should be noted that the
percentage difference presented in Table 1 and subsequent tables represents the result (e.g. the
displacement at the buckle apex) under a specific condition compared with the equivalent from the
base case (which takes a ‘stabilize’ value of 2×10-4).
Table 1. Sensitivity check when varying the ‘stabilize’ value in static analysis
(displacement at the buckle apex)
‘Stabilize’ Value First Load Diff% Cycle 20 Diff% Hot Oiling Diff%
1E-10 4.84 -1% 6.42 1% 8.32 -2%
1E-07 4.75 -3% 6.22 -2% 8.54 1%
1E-05 4.87 -1% 6.28 -1% 8.50 1%
Default (2E-04) 4.91 0% 6.37 0% 8.45 0%
1E-03 4.94 1% 6.59 3% 8.69 3%
1E-02 4.97 1% 6.36 0% 8.72 3%
1E-01 4.82 -2% 6.31 -1% 8.56 1%
Table 2. Sensitivity check when varying the ‘stabilize’ value in static analysis
(maximum compressive mechanical strain at buckle apex)
‘Stabilize’ Value First Load Diff% Cycle 20 Diff% Hot Oiling Diff%
1E-10 -0.279% -1% -0.213% -2% -0.414% -31%
1E-07 -0.276% -2% -0.197% -10% -0.386% -36%

1E-05 -0.282% 0% -0.210% -4% -0.516% -14%
Default (2E-04) -0.281% 0% -0.218% 0% -0.600% 0%
1E-03 -0.285% 1% -0.214% -2% -0.362% -40%
1E-02 -0.283% 1% -0.205% -6% -0.384% -36%
1E-01 -0.271% -4% -0.195% -11% -0.351% -42%
Figure 8 shows a plot of the displacements at the buckle apex presented in Table 1, the x-axis
adopts a logarithmic scale, lg(X). Table 1 and Figure 8 show that when varying the ‘stabilize’
value from 10-1 to 10-10, the lateral displacement at the buckle apex varies by 3% when compared
with the default setting in all the presented pipeline production operation and pipeline service
steps. The percentage difference of 3% in the lateral displacement at the buckle apex results in up
to percentage differences of 4%, 11% and 42% in the longitudinal mechanical compressive strain
in the first production operation, production operation cycle 20 and the hot oiling step,
respectively.
Sensitivity Check of the Dissipated Energy Fraction (Stabilize
Value) in Static Analysis
9
8.5
8
7.5
Displacement (m)
7
6.5
6
5.5
5
4.5
4
-10 -8 -6 -4 -2
Common Logarithm of the Dissipated Energy Fraction (lg(x))
First Production Cycle 20 Hot Oiling
Figure 8. Sensitivity check when varying the ‘stabilize’ value in static analysis
(displacement at buckle apex)

Sensitivity Check of the Dissipated Energy Fraction in Static
Analysis
-0.1%
Longitudinal Strain Component (%)
-0.2%
-0.3%
-0.4%
-0.5%
-0.6%
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1
Common Logarithm of the Dissipated Energy Fraction (lg(x))
Figure 9. Sensitivity check when varying the ‘stabilize’ value in static analysis
(mechanical strain at buckle apex)
The results for the hot oiling step illustrated in Figure 9 show that as far as the maximum
longitudinal compressive mechanical strain is concerned in the analysis, the choice of the
‘stabilize’ value and the associated sensitivity check becomes very important in the analysis of
pipeline global buckling subject to production operation cycles and further high temperature
events.
The pipeline lateral displacement profiles and the maximum tensile strain profiles along the
pipeline in the buckle region at first production, production cycle 20 and the hot oiling step as
varying ‘stabilize’ values using Abaqus static analysis are illustrated and compared in Figures 10
and 11. Prod, C20 and Oil are used to denote the results from the first production, production
cycle 20 and the hot oiling cases, respectively. This notation coupled with 1E_3 and 1E_5
represent the results from each respective case with ‘stabilize’ values of 10-3 and 10-5, respectively.

Stabilize Value (Energy Dissipation Fraction) Sensitivity Check
10
8
0
980 1000 1020 1040 1060 1080
-2
Pipe Length (m)
Prod_Stabilize_1E-3 Prod_Default Prod_Stabilize_1E-5
C20_Stabilize_1E-3 C20_Default C20_Stabilize_1E-5
Oil_Stabilize_1E-3 Oil_Default Oil_Stabilize_1E-5
Figure 10. ‘Stabilize’ value sensitivity check in static analysis (pipeline ½ lateral
profiles)
Satbilize Value (Energy Dissipation Fraction) Sensitivity Check

0.6%
Max. Longitudinal Mechanical Strain (%)
0.5%
0.4%
0.3%
0.2%
0.1%
0.0%
980 1000 1020 1040 1060 1080
-0.1%
Pipeline Length (m)
Prod_Max_1E-3 Prod_Max_Default Prod_Max_1E-5

C20_Max_1E-3 C20_Max_Default C20_Max_1E-5
Oil_Max_1E-3 Oil_Max_Default Oil_Max_1E-5
Figure 11. ‘Stabilize’ value sensitivity check in static analysis (maximum tensile
strain profile)

Figures 10 and 11 show that at the first production load step and production cycle 20, both the
pipeline lateral profiles and maximum longitudinal tensile strain component profiles have good
agreement along the pipeline length in the buckle region when varying the ‘stabilize’ values. In the
hot oiling step the pipeline lateral profiles show a reasonable good level of agreement, with
varying ‘stabilize’ values. However, the maximum longitudinal tensile strain component profiles
are almost identical away from the buckle crown, but largely have deviated at the narrow buckle
crown region from about 0.32% to 0.52%.
Under the hot oiling condition, a content temperature of 83˚C is assumed. As a higher temperature
event is encountered, the gap of the predicted strain values may be magnified, suggesting that a
temperature sensitivity check is needed to identify the effect of temperature in conjugation with
varying the ‘stabilize’ value in the analysis of pipeline in deepwater subject to high temperature
and high pressure.
3.2 Sensitivity check of the ‘direct damping factor’ in static analysis.
The effect of the ‘direct damping factor’ in Abaqus static analysis has been examined directly by
varying the damp factor value from 10-1 to 10-8.
Tables 3 and 4 summarise the lateral displacement and the maximum compressive strain at the
buckle apex in first production, production cycle 20 and hot oiling steps, with the ‘damping factor’
value varying from 10-1 to 10-8.
Table 3. Sensitivity check when varying ‘damping factor’ in static analysis (lateral
displacement at buckle apex)
‘Damping factor’ Value First Load Diff% Cycle 20 Diff% Hot Oiling Diff%
1E-08 4.21 -14% 6.56 3% 8.56 1%
1E-07 4.74 -3% 6.64 4% 8.49 0%
1E-06 4.78 -3% 6.69 5% 8.64 2%
1E-05 4.81 -2% 6.23 -2% 8.09 -4%
1E-04 4.78 -3% 6.30 -1% 8.31 -2%
1E-03 4.74 -3% 6.57 3% 8.51 1%
1E-02 4.68 -5% 6.17 -3% 8.00 -5%
1E-01 4.49 -9% 6.14 -4% 7.93 -6%
Table 4. Sensitivity check when varying ‘damping factor’ in static analysis

(longitudinal maximum mechanical strain at buckle apex)
‘Damping factor’ Value First Load Diff% Cycle 20 Diff% Hot Oiling Diff%
1E-08 -0.267% -5% -0.233% 7% -0.374% -38%
1E-07 -0.273% -3% -0.223% 2% -0.378% -37%
1E-06 -0.279% -1% -0.229% 5% -0.388% -35%
1E-05 -0.279% -1% -0.205% -6% -0.552% -8%
1E-04 -0.275% -2% -0.199% -9% -0.414% -31%
1E-03 -0.275% -2% -0.221% 1% -0.370% -38%
1E-02 -0.274% -2% -0.193% -11% -0.417% -31%
1E-01 -0.283% 1% -0.213% -2% -0.456% -24%
The results presented in Table 3 shows that when varying the ‘damping factor’ values ranging
from 10-1 to 10-8, the lateral displacement at the buckle apex varies with maximum values of 14%,

5% and 6% when compared with the base case for the first load, cycle 20 and the hot oiling steps,
respectively.
From Table 4 it is observed that the longitudinal maximum compressive mechanical strain at the
buckle apex varies up to 5% and 11% when compared with the base case first load, cycle 20 steps
respectively. A percentage difference of less than 6% in lateral displacement, results in up to 38%
percentage difference in the longitudinal mechanical compressive strain in the in the hot oiling
step.
Figures 12 and 13 show plots of the lateral displacement and the maximum compressive strain at
the buckle apex displayed in Tables 3 and 4.
Sensitivity Check of the Direct Damping Factor in Static Analysis
9
8.5
8
7.5
Displacement (m)
7
6.5
6
5.5
5
4.5
4
-8 -7 -6 -5 -4 -3 -2 -1
Common Logarithm of the Damping Factor (lg(x))
Figure 12. Sensitivity check when varying ‘damping factor’ in static analysis
(lateral displacement at apex)

Sensitivity Check of the Direct Damping Factor in Static Analysis
-0.1%
Longitudinal Stain Component (%)
-0.2%
-0.3%
-0.4%
-0.5%
-0.6%
-8 -7 -6 -5 -4 -3 -2 -1
Common Logarithm of the Damping Factor (lg(x))
Figure 13. Sensitivity check when varying ‘damping factor’ in static analysis
(longitudinal maximum compressive strain)
From Figure 13 it is interestingly to see like Figures 9 and 11 in which the ‘stabilize’ value is
varied, that varying the ‘damping factor’ value results in higher strain values being predicted at
certain data points. The higher strain value predicted confirms, that as far as the maximum
longitudinal compressive mechanical strain is concerned, the choice of the ‘damping factor’ value
is critical in the analysis of global buckling of offshore pipelines subject to production operation
cycles and midlife higher temperature events.
As outlined in the analysis review section of this paper and as the Abaqus user manual suggests,
the energy dissipation associated with the value for ‘direct damping factor’ used in subsequent
steps of the analysis should be checked to ensure conservation of energy. For this reason the
‘stabilize’ value method, unless there is serious convergence issue with using ‘stabilize’ value, has
been preferred in this analysis over the ‘direct damping factor’ method.
The pipeline lateral profiles and the longitudinal maximum tensile strain in the first production,
production cycle 20 and hot oiling steps along the pipeline axial direction resulting from varying
‘damping factor’ values are plotted in Figures 14 and 15, respectively. The suffixes Factor_1E_3,
Factor_1E_4 and Factor_1E_5 are used in the legend to differentiate between the data sets,
representing ‘damping factor’ values of 10-3, 10-4 and 10-5, respectively.

Direct Damping Factor Value Sensitivity Check
10
8
0
980 1000 1020 1040 1060 1080
-2
Pipe Length (m)
Prod_Factor_1E-3 Prod_Factor_1E-4 Prod_Factor_1E-5

C20_Factor_1E-3 C20_Factor_1E-4 C20_Factor_1E-5
Oil_Factor_1E-3 Oil_Factor_1E-4 Oil_Factor_1E-5
Figure 14. Sensitivity check when varying ‘damping factor’ (Pipeline ½ lateral
profile)
Direct Damping Factor Value Sensitivity Check

0.6%
Max. Longitudinal Mechanical Strain (%)
0.5%
0.4%
0.3%
0.2%
0.1%
0.0%
980 1000 1020 1040 1060 1080
-0.1%
Pipeline Length (m)
Prod_Max_1E-3 Prod_Max_1E-4 Prod_Max_1E-5
C20_Max_1E-3 C20_Max_1E-4 C20_Max_1E-5
Oil_Max_1E-3 Oil_Max_1E-4 Oil_Max_1E-5
Figure 15. Sensitivity check when varying ‘damping factor’ (maximum tensile strain
profile)

The results presented in Figure 14 and 15 illustrate that for the first production load step and the
production cycle 20, both the pipeline lateral profiles and maximum longitudinal tensile strain
component profiles have good agreement under the varying ‘damping factor’ values. The results
generated for the hot oiling step show that the pipeline lateral profiles have reasonable good
agreements in general under directly varying the ‘damping factor ' values. Interestingly, the
maximum longitudinal tensile strain component profiles are almost identical away from the
primary buckle region, but have largely deviated at the buckle crown narrow region from
approximately 0.32% to 0.52%.
3.3 Effect of the analysis methods
In addition to the Abaqus analysis methods outlined earlier a first load analysis employing the
RIKS method and a dynamic analysis, was undertaken. The initial first load buckle profiles were
seen to exert a significant impact on final hot oil strains following cyclic analyses. The 20 cycle
operation displacements and strains and the subsequent hot oil displacements and strains are
presented in Tables 5 and 6.
It is common practice, in global pipeline buckling analysis, that a stabilize value of around 10-4
and a ‘damping factor’ value of around 10-3 be used. These values are favoured so as to achieve a
reasonable balance between the computational time (and its associated cost!) and the user
accuracy requirement. To increase comparability between the results under the different cases
analysed, a ‘stabilize’ value of 2×10-4 (base case) and a ‘damping factor’ value of around 10-3 are
used to compare results with results emerging from RIKS method and dynamic analyses.
As mentioned earlier, in RIKS analysis, the exact results cannot be achieved directly in its own
step. Rather a non-RIKS step in the first production is used to continue the RIKS step to achieve
the results under a specified load condition. In the first production operation, a non-RIKS step has
been used to continue a RIKS run under two scenarios, the first being a non-RIKS step is used to
continue a RIKS step before the RIKS run reached a peak on the monotone increasing stress-strain
curve, and in the other on the peak of the monotone increasing stress-strain curve. These are
labelled RIKS1 and RIKS2 respectively in Tables 5 and 6. These two scenarios have been
considered to check the reliability of results generated when a RIKS step is manually continued by
a non-RIKS analysis step analysis. It should be noted that, if a non-RIKS step is used to continue a
RIKS step to early on the monotone increasing stress-strain curve, the advantages of RIKS
analysis may not be accurately of sufficiently portrayed. Conversely, if a non-RIKS step is used to
continue a RIKS step too far along the monotone increasing stress-strain curve, the results might
be distorted when there is large plasticity in the RIKS analysis step.
Figure 16 compares both RIKS generated buckle profiles against the lateral profile generated
using the project ‘Stabilize’ value with an initial 1m out of straightness. As can be seen the RIKS
method generates lateral profiles of larger amplitude than with the Abaqus ‘Stabilize’ method with
shallower secondary lobes. As seen in Table 6 the initial RIKS profiles carried through 20 cycles
of operation can give rise to significant differences in the final berm breakout strain levels.
Figure 17 illustrates that the choice of initial pipeline out of straightness in buckle formation
sensitivity analyses can also give rise to different initial buckle profiles impacting on final strain
levels. The comparative FE runs without system damping using another FE package shows
smaller levels of initial out of straightness giving rise to more dynamic buckle developments tend
towards the RIKS profiles whilst larger 1m out of straight imperfections develop the prominent
secondary lobes seen using the Abaqus ‘Stabilise’ value.

10
8
-2
980 1000 1020 1040 1060 1080 1100 1120 1140 1160 1180
Pipe Length (m)
Default_Prod RIKS1_Prod RIKS2_Prod
Default_Oiling RIKS1_Oiling RIKS2_Oiling
Figure 16. Variation seen in initial first load and 20 cycle hot oil buckle profiles using
1&2
RIKS and default ‘Stabilize’ parameters
6
5
1m x 100m OOS
0.2m x 200m OOS

4
Lateral displacement (m)
-1
-2
980 1000 1020 1040 1060 1080 1100 1120 1140 1160 1180
Distance along pipe (m)
Figure 17. Un-damped first load buckle profiles seen with different levels of initial
pipeline out of straightness
Note that the smaller more dynamic buckle form generated with lower 0.2m levels of initial OOS
compares more closely with Abaqus RIKS simulations undertaken with 1m initial OOS.

The lateral displacement and the maximum compressive strain component values at the planned
buckle apex are shown in Tables 5 and 6.
Table 5. Effect of the Abaqus analysis methods to predicted lateral displacement at
apex
Method First Load Diff% Cycle 20 Diff% Hot Oiling Diff%
Default Stabilize (2×10-4) 4.91 0% 6.37 0% 8.45 0%
‘Damping factor’ (1×10-3) 4.74 -3% 6.57 3% 8.51 1%
RIKS1 5.61 14% 6.90 8% 9.07 7%
RIKS2 6.84 39% 6.65 4% 8.31 -2%
Dynamic 6.10 24% 6.02 -5% 6.72 -20%
Table 6. Effect of the Abaqus analysis methods to predicted longitudinal maximum

compressive mechanical strain at apex
Method First Load Diff% Cycle 20 Diff% Hot Oiling Diff%
Stabilize Default (2×10-4) -0.281% 0% -0.218% 0% -0.600% 0%
‘Damping factor’ (1×10-3) -0.275% -2% -0.221% 1% -0.370% -38%
RIKS1 -0.285% 1% -0.227% 4% -0.462% -23%
RIKS2 -0.577% 105% -0.548% 151% -1.543% 157%
Dynamic -0.310% 10% -0.293% 34% -0.636% 6%
Table 5 shows that when a non-RIKS step is continued from a RIKS step in the increasing part of
the stress-strain curve, the case labelled RIKS1, the analysis overestimates the lateral displacement
at the buckle apex by 14%, 8% and 7% in the first production, Cycle 20 and hot oiling steps,
respectively. The percentage difference of longitudinal mechanical strain compared with the base
case in the first production, Cycle 20 and hot oiling step are 1%, 4% and -23%, respectively.
If the RIKS step is continues by non-RIKS step on the peak of the increasing part of the stress-
strain curve (RIKS2) the lateral displacement at buckle apex is overestimated by 39%, 4% and -2%
in the first production, Cycle 20 and hot oiling steps, respectively. The percentage difference of
longitudinal mechanical strain compared with the base case in the first production, Cycle 20 and
hot oiling step are all over 100%, which places doubt on the results emerging from RIKS2 method.
Results emerging from DYNAMIC analysis show a percentage difference for the longitudinal
mechanical strain compared with the base case in the first production, Cycle 20 and hot oiling
steps of 10%, 34% and 6%, respectively. It is worth noting that the dynamic analysis results may
be affected by temperature and pressure loading patterns at locations along the pipeline necessary
for the buckle to form, which cannot be verified at the current stage of the study. Further study
into dynamic analysis is suggested.
4. Discussions and Conclusions

Analysis methods available to users in Abaqus suitable for global pipeline buckling have been
assessed in this study using the Wood Group Kenny advanced pipe-soil interaction subroutine.
The pipe-soil interaction subroutine simulation strategy accounts for the berm formation
mechanism during production operation cycles and midlife pipeline services.

Varying the ‘stabilize’ value in a static analysis with a specified ‘stabilize’ value, results conclude
that the lateral displacement and the longitudinal maximum strain at the apex in the first
production and production operation cycle 20 show a good level of agreement. The longitudinal
mechanical compressive strain can be underestimated up to 42% in the hot oiling step following
20 production operation cycles when compare with the base case, which adopts the default
‘stabilize’ value of 2×10-4.
Varying the ‘damping factor’ value in static analysis, it is observed that the lateral displacement
and the longitudinal maximum strain at the apex in the first production and the production
operation cycle 20 also show good agreement. Compared with the base case, the longitudinal
mechanical compressive strain can be underestimated up to 38% in the hot oiling step following
20 production operation cycles.
Static analysis with RIKS method showed convergence issues, this is likely to be due to the
momentary or permanent loss of surface contact between pipeline and soil surfaces when a buckle
occurs. The computational runs using RIKS method have been conducted by a trial and error
procedure. The quality of the predicted results may rely on a good educated guess of the point on
the monotone increasing stress-strain curve where a non-RIKS step should continue the RIKS
step.
The pipeline lateral profile from the dynamic analysis failed to present good agreement with the
equivalent results emerging from a static analysis. Further study to understand the global buckling
dynamic behaviour of pipelines in deepwater subject to high temperature and high pressure events
is much needed, especially as the complexity of offshore pipeline projects is ever increasing.
Motivation for this study and subsequently the development of the PSI subroutine was triggered
by concerns expressed over different post hot-oil results seen from cases started with rather similar
first load responses. The large differences in analysis results under hot oiling may indicate that a
good quality berm formation strategy is need, for projects related to the oil and gas industry. The
subroutine presented in this paper is one such solution.
The accumulated difference in cyclic analyses resulting from using inappropriate Abaqus analysis
methods at the first load step may result in differences in the berm formation. As the pipeline is
fully locked-in to the soil berm, the berm resistance experienced by the pipeline may not actually
be the available berm resistance. The difference in berm formation resulting from a cumulative
error as an outcome of using a certain analysis method may not be immediately apparent to the
user. In hot oiling events, the pipeline buckle crown further breaks out from the berm and may
meet and have to overcome the diverged cumulated berm surrounding the pipeline.
This study has highlighted the need for using advanced pipe-soil interaction model to account for
the berm formation mechanism in deepwater environment related to the oil and gas industry. The
development and subsequent validation of the Wood Group Kenny advanced pipe-soil interaction
subroutine in recent projects has offered opportunity to look into this phenomenon, the detailed
theory and development methodology of this subroutine will be reported in due course in a further
publication.
5. Acknowledgements
This study was carried out under the support of Wood Group Kenny at Aberdeen and London
offices in UK, Galway office in Republic of Ireland and Melbourne office in Australia, which is

appreciated. Special thanks to Dr. Mahmoud H. M. Ahmed for his contribution in finalising the
paper manuscript.
6. References
1. ABAQUS User’s Manual, Version 6.12, SIMULIA, Dassault Systèmes Simulia Corp.,
2012.
2. SIMULIA Release Authorization and Stipulations, Dassault Systèmes Simulia Corp,
September 3, 2013.
3. Safe Design of Pipeline with Lateral Buckling Design Guideline, SAFEBUCK III
5087471/01/A, 2011.
4. Global Buckling of Submarine Pipelines, DNV Offshore Codes Recommended Practice
DNV-RP-F110, Det Norske Veritas, 2007.
5. Submarine Pipeline Systems, Offshore Standard DNV-OS-F101, Det Norske Veritas,
2013.
6. E.J. Chin, D. Scholtz, E. Shim, E. Gerginov, “The Effects of Soil Berm Formation on
Pipeline Fatigue Response at Buckle sites” Deep Offshore Technology (DOT)
International Conference, Houston, Texas USA, 2013
7. S. Yu and I. Konuk, “Continuum FE modelling of lateral buckling” OTC 18934, 2007.

HIGH-TECH

Achieving Higher Productivity on Abaqus
Simulations with HPC Clustering
Technologies
Pak Lui, Gilad Shainer, David Cho, Scot Schultz, Richard Graham
HPC Advisory Council
Abstract: Engineers from wide ranges of industries face an ever increasing need to run virtual
tests to simulate complex models for improving reliability as well as reducing product
development time and costs. Abaqus Unified FEA is designed to tackle these challenges by
producing high-quality realistic simulation solutions, while delivering high performance in
productivity by utilizing efficient use of modern compute cluster available in High Performance
Computing (HPC). Many organizations can now deploy with HPC cluster to process such
complex analyses on Abaqus with computer cluster to render such high-quality models, while
reducing simulation time from days to just hours. Behind this type of computational and clustering
technologies that enables Abaqus to perform, it involves complex calculations and data exchanges
among computational systems. The HPC Advisory Council has performed a deep investigation on
Abaqus to evaluate its performance and productivity capabilities and to explore potential
optimizations. This study presents the techniques and profiling results to further understand
Abaqus dependencies on the CPUs, network, and the other underlying hardware and software
components. The paper will review the effects by comparing various components using different
simulation models on Abaqus.
Keywords: Abaqus/Standard, Abaqus/Explicit, clusters, High Performance Computing, HPC,

Benchmarking, Ethernet, Tuning, CPU, Power, network, InfiniBand, Turbo Mode, MPI Profiling,
IO Profiling
1. Introduction
The HPC Advisory Council has performed a deep investigation on Abaqus Unified FEA to
evaluate its performance and productivity capabilities and to explore potential optimizations. This
study presents the techniques and profiling results to further understand Abaqus dependencies on
the CPUs, network, IO, and the other underlying hardware and software components. The paper
will review the effects by comparing various components using different simulation models on
Abaqus.
1.1 Abaqus Unified FEA
Abaqus Unified FEA product suite offers powerful and complete solutions for both routine and
sophisticated engineering problems covering a vast spectrum of industrial applications. The
Abaqus analysis product provides tools for simulating nonlinear finite element analysis (FEA),
advanced linear and dynamics application problems. Abaqus/Standard is a General-purpose FEA

that includes broad range of analysis capabilities. Abaqus/Explicit is a nonlinear, transient,
dynamic analysis of solids and structures using explicit time integration.
1.2 High Performance Computing
High-performance computing (HPC) is a crucial tool for engineering design and manufacturing. It
is used for computer-aided engineering (CAE) from component-level to full vehicle analyses:
crash simulations, structure integrity, thermal management, climate control, engine modeling,
exhaust, acoustics and much more.
HPC helps drive faster time-to-market, significant cost reductions, and tremendous flexibility. The
strength in HPC is the ability to achieve best sustained performance by driving the CPU
performance towards its limits. The motivation for high-performance computing in the automotive
industry has long been its tremendous cost savings and product improvements – the cost of a high-
performance compute cluster can be just a fraction of the price of a single crash test, and the same
cluster can serve as the platform for every test simulation going forward.
The recent trends in cluster environments, such as multi-core CPUs, GPUs, and new interconnect
speeds and offloading capabilities are changing the dynamics of clustered-based simulations.
Software applications are being reshaped for higher parallelism and multi-threading, and hardware
is being configured for solving the new emerging bottlenecks, in order to maintain high scalability
and efficiency.
1.3 HPC Cluster
Abaqus simulations can be carried out on a single server system or a cluster of compute servers
known as high-performance computing (HPC) cluster. The HPC cluster is based on industry-
standard hardware connected by a private high-speed network. The main benefits of clusters are
affordability, flexibility, availability, high-performance and scalability. A cluster uses the
aggregated power of compute server nodes to form a high-performance solution for parallel
applications such as Abaqus. When more compute power is needed, it can sometimes be achieved
simply by adding more server nodes to the cluster.
The areas which HPC clusters are architected have a huge influence on the overall application
performance and productivity are the number of CPUs, usage of GPUs, the storage solution and
the cluster interconnect.
In the subsequent sections of this paper, we are going to explore each of those components of a
cluster that can enable higher performance.
2. Base Configuration
The configuration that will be used to demonstrate the performance differences are as follows,
unless otherwise stated.
2.1 Hardware Configuration
The current study was conducted at the HPC Advisory Council Cluster Center [1] on the cluster
comprised of HP ProLiant SL230s Gen8 4-node cluster, each node with Dual Socket Intel®
Xeon® 10-core CPUs E5-2680 v2 at 2.80 GHz, Mellanox Connect-IB 56Gb/s FDR InfiniBand
adapter, and with 64GB of 1600MHz DDR3 memory. The nodes were connected into a network

using a Mellanox SwitchX® SX6036 36-Port VPI switch which supports 10Gb/s and 40Gb/s
Ethernet and 56Gb/s FDR InfiniBand. The Operating System used was RHEL6.2, the InfiniBand
driver version was OFED 2.1-1.0.0, and the File System is shared over NFS, which provides 1TB
SATA hard drive. The MPI library used was Platform MPI 9.1.
References to the previous performance studies [4][5][6][9][10] on Abaqus Unified FEA by the
HPC Advisory Council can be found at the Best Practices [8] section of the HPC Advisory
Council web site [1].
The Abaqus version was 6.13-2. The Abaqus/Standard benchmark workload was the test
simulation S2A that has 474,744 DoF. This description of the benchmark is provided on the
Abaqus Performance Data webpage [3]. The benchmark utilizes the nonlinear static analysis of a
flywheel with centrifugal loading. The flywheel is meshed using 1st order hexahedral elements of
type C3D8R and uses an isotropic hardening Mises plasticity material model. The nonlinearity in
this problem arises from localized yielding in the vicinity of the bolt holes.
2.2 Benchmark Cases
The Abaqus/Explicit benchmark workload used is the test simulation E6. The E6 is the benchmark
that consists of many concentric spheres with some spacing between each sphere. The Concentric
Spheres with 244,124 elements, as shown in Figure 1. The spheres are put into a general domain
and forces are exerted from the outer sphere that creates some complex interactions between the
spheres from within. The E6 model is an example of a memory bandwidth bound problem.
Figure 1: Abaqus/Standard Benchmark workloads: S2 Flywheel (left) and Abaqus/Explicit

workload E6 Concentric Spheres (right)
2.3 Performance Rating
The performance rating is the metric that will be shown; the value of performance rating is
calculated based on the amount of jobs can be run in a day from the actual runtime achieved.
We measure the total runtime of each dataset that is accumulated with the time in seconds for the
input file preprocessor, solver phases. Then we take the total runtime and convert it into a value
that represents the performance metric rating based on the number of analyses run that can be
achieved during a 24 hour period. The metric we display is effectively becomes the number of
jobs can be run per day. This rating system provides a much clearer representation when observing
the scaling progression in a multi-core and multi-node environment because it can clearly shows
the additional amount of workload can be achieved by increasing the number of systems involved
in the simulation. It also becomes useful when comparing environment of similar system
components.

It is also to note that in the traditional HPC environment, it is a common practice to utilize all of
the CPU cores and hardware resources available on the compute nodes to run exclusively for a
single workload. By having dedicated compute nodes for a particular parallel job, the underlying
hardware would not run into resource contention or interference with other processes, it is
regarded as a good approach for achieving highest performance for a parallel job. Thus, the
comparisons made in the subsequent sections will be on a per-node basis, enough though the
actual CPU cores available on a node may vary.
3. Analyses on Turbo Mode and CPU Frequencies

3.1 MSR Tools and CPU Governor
To show the cost and benefit of turbo mode and higher CPU clock frequency, we are relying on
the software tools which allow us to boost and throttle the CPU clock frequency and toggling the
turbo mode dynamically in the OS. The software tools that we used for modifying turbo mode
dynamically in the OS is a kernel tools called “msr-tools”[7]. MSR stands for Machine Specific
Register.
Another method to dynamically change the CPU clock relies on the kernel module called CPU
governor [2] which is available in the OS. This tools facilitates our benchmark testing with the
different CPU speeds as we would not need to acquire the actual CPU processors that runs that
particular clock rates.
3.2 Effects of CPU Frequency
4%
25% 5% 23%
Figure 2: Turbo Mode Effect on Abaqus/Explicit Performance

The data labeled with “2800MHz-Turbo Off” is our performance baseline since it is the type of
processors that we have installed on all the compute nodes in the HPC cluster. The data labeled
“2000MHz-Turbo Off” is the comparison run that is conducted by lowering the CPU clock speed
using the CPU governor available in the OS. The reason that we are interested in scaling down to
2800MHz to 2000MHz is to observe the performance difference between the 2, in order to judge
whether it makes sense to use a CPU SKU with the lower clock frequency.
We observe that there is some modest dependency on CPU frequency by showing the performance
difference using the Abaqus/Explicit benchmark case, as shown on Figure 2. There is a 23% of the
performance difference between 2000MHz vs 2800MHz. However, the tradeoff of having a higher

clock frequency that delivers the performance is the power consumption. There is about 51% of
the power additional power needed to run from 2GHz to 2.8GHz CPUs. This implies that there is
actually some benefit of using higher clock rate CPUs on Abaqus workload, as generally there are
CPU models available at higher clock rate, but at a higher price point premium.
The similar behavior on CPU frequency and Turbo mode have exhibit with the Abaqus/Standard
test case on Figure 3. While there is about 25% of the improvement at 51% of higher power
utilization by enabling higher clock frequency from 2000MHz to 2800MHz.
51% 18% 51% 18%
Figure 3: Power Consumption of Enabling CPU Turbo Mode on Abaqus/Standard and

Abaqus/Explicit
3.3 Effects of Turbo Mode
On the other hand, enabling Turbo Mode can only provide very marginal difference of
performance for Abaqus/Explicit. Enabling Turbo mode can provide approximately 4% of
performance gain for CPU cores running at 2800 MHz, which resulted in 18% of higher power
usage as a tradeoff for the additional performance. It is clear that turbo mode only provide
marginal additional benefit for Abaqus/Explicit, while the power consumption of turbo mode
overweighs the benefit.
Similarly for Abaqus/Standard, it is also observed a 5% gain in performance when using Turbo
Mode, which results in a tradeoff of higher power consumption by 18%.
4. Impact of Interconnect on Abaqus Performance

The cluster interconnect is very critical for efficiency and performance of the application in the
multi-core era. When more CPU cores are present, the overall cluster productivity increases only
by the presence of a high-speed interconnect. We have compared the elapsed time with Abaqus
using 1Gb/s, 10Gb/s and 40Gb/s Ethernet, and 56Gb/s FDR InfiniBand.
Abaqus FEA software utilizes Message Passing Interface (or MPI) for cluster or node-to-node
communications. MPI is the de-facto messaging library for high performance clusters. It relies on
fast server and storage interconnects in order to provide low latency and high messaging rate.
Performance demands from the cluster interconnect increase dramatically as the simulation
requires more complexity to properly simulate the physical model behavior.

4.1 Ethernet
The 10GbE and 40GbE are the newer standards used in the data center environment. The Ethernet
interconnects are typically used in the data center to provide a cost-effective solutions for
transferring data between the compute nodes and storage systems. While these data rates provides
faster throughput than 1GbE that is typically available on server systems, the 10GbE and 40GbE
are not able to deliver good performance in the HPC environment, as majority of the
communication that takes place in the HPC run are not point-to-point communication, but rather
one-to-many and many-to-many group communications that may overwhelm the Ethernet network
even at small scale.
4.2 InfiniBand
This network interconnect has a scalable architecture for delivery performance on large high-
performance clusters environment, specifically designed for use by the largest supercomputers in
the world. By providing low-latency, high-bandwidth and extremely low CPU overhead,
InfiniBand has become the most deployed high-speed interconnect for HPC clusters, replacing
proprietary or low-performance solutions. The InfiniBand Architecture (IBA) is an industry-
standard fabric designed to provide high-bandwidth, low-latency computing, scalability for tens of
thousands of nodes and multiple CPU cores per server platform and efficient utilization of
compute processing resources.
Figure 4 shows the Performance Rating for the InfiniBand and Ethernet interconnects for a range
of node counts for the benchmark cases on the same set of the aforementioned cluster of servers.
28%18% 34% 30%

183% 142%
Figure 4: Interconnect Comparisons between 1/10/40Gb Ethernet and FDR InfiniBand

4.3 Performance Comparisons on the Network Interconnects
The result shows that InfiniBand delivered better scalability in application performance, which
results in faster run time which translates into higher performance rating, which means more jobs
per day. The 56Gb/s FDR InfiniBand-based simulation performance was 142% higher than 1GbE,
over 34% higher than 10GbE and over 30% higher than 40GbE on a Abaqus simulation which
runs on 4 nodes or 80 MPI processes.
Likewise for the Abaqus/Explicit case, FDR InfiniBand enables higher cluster productivity among
the network interconnects being tested. InfiniBand reduces the runtime by 183% versus 1GbE, and
delivers higher performance by 28% versus 10GbE, and 18% higher performance versus 40GbE.

While 1Gb, 10Gb, and 40Gb Ethernet showed some loss of performance or increase in run time
beyond 2 nodes or 40 MPI processes, InfiniBand demonstrated good scalability throughout the
various tested configurations.
4.4 Time Ratio on Different Network Interconnects
InfiniBand can speed up the communication time, which is not only due to the throughput
available, but it provides a much simpler and less congested communications that is more suitable
for the HPC environment. By conducting the MPI profiles among the runs from different network
interconnect, we captured the amount of time that different network interconnect would use to run
the same workload. We found that FDR InfiniBand consumes about 37% of total runtime at 4
nodes/80 MPI processes, while Ethernet solutions consume from 53% to 72% at 4 nodes. This is
shown on Figure 5.
Figure 5: Time Ratio between Computational (User) Time and Communication (MPI) Time
5. MPI Profiling
By inspecting the output from the MPI profiler, it is observed that the majority of the MPI
messaging appears to be concentrated on the small message buffer sizes. We can then understand
that the underlying communication patterns that make Abaqus scalable on the low latency
interconnect.
5.1 Time Spent on MPI Communications
By profiling the communication, we can observe in the type of MPI calls that are involved. We
can see the type of communication that Abaqus engages the MPI processes. In Figure 6, the MPI
profiler shows FDR InfiniBand spends approximately 54% of the time during the job run in the
MPI_Gather, 9% time MPI_Allreduce, and approximately 9% of the time in MPI_Scatterv on a 4-
node, 80 processes job. The MPI profile shows Abaqus/Explicit would majority of time in MPI
collective operations. The gather algorithm would gather values from a group of MPI processes,
while scatter would do the reverse. The MPI_Allreduce would do a combination of both Scatter
and Gather, with a preset operation, such as the sum operation. This clearly shows that many of
the processes are engaged in a group area of communication.
Similarly, when taking a look at the communication pattern on a slower, less efficient network
interconnect, we can compare the two and identify in the areas that would help by the higher

performance interconnect. The time spends on the 1GbE network is about 43% for MPI_Gather,
13% for MPI_Irecv, 14% for MPI_Scatterv.
Figure 6: Profiling Differences of MPI Communications between 1GbE and FDR InfiniBand
There are difference in communication patterns between Abaqus/Standard and Abaqus/Explicit.
Abaqus/Standard shows high usage for testing non-blocking messages, for instance, MPI_Gather
dominates the MPI communication time in Abaqus/Explicit. For Abaqus/Standard, majority of the
time is spent on MPI_Test. This is shown on Figure 7.
Figure 7: Communications Profiled on Abaqus/Standard and Abaqus/Explicit

In the MPI profiling, it shows that Abaqus uses MPI library for communications between the
application and the networking layer, it requires a scalable and efficient send-receive semantics as
well as good scalable collectives operations.
InfiniBand provides an effective method for communication called Remote Direct Memory
Access, or RDMA. With RDMA, InfiniBand can write directly to the pinned system memory of
the remote systems, which reduces the CPU involvement in handling of the networking traffic. By
freeing up the network traffic from CPU, can focus on computation which allow better
performance in computation. On the other hand, the collision in the network in the TCP protocol

in the Ethernet network would lead more overheads for CPU, which can translate into higher
network latency, reducing the cluster efficiency and scalability.
5.2 MPI Message Sizes
Profiling the MPI communications that is taking place during the job run can show the application
uses MPI to communicate between the MPI processes. By knowing the MPI message sizes and
communication patterns, we can estimate the benefits of using a network interconnect. The profile
shows us that Abaqus/Standard uses small and medium MPI message sizes for communication.
We see that most of the MPI messages are appeared in the small message sizes under 256 bytes.
For the most time consuming MPI calls, MPI_Recv messages are concentrated 4KB. Majority of
the MPI_Bcast messages are in buffers that are less than 16B. For MPI_Allreduce, most messages
are less than 256B.With the MPI profiling tool, we have conducted some profiling samples that
include the MPI message size distribution in Figure 8.
We analyzed that majority of the MPI messages fall into a 0 to 256 Byte range and the total of all
MPI message sizes fall in the 0-64KB range which are small message sizes. The overall MPI
communications collected was approximately 30GBs from all MPI ranks for about 50 seconds by
all 80 MPI processes. This amount of MPI traffic is low when InfiniBand can handle about
6.4GB/sec with FDR InfiniBand throughput. With Ethernet (1GbE) we can easily saturate the
network when running two or more datasets since one analysis consumes about 57% of the total
bandwidth of the GigE network. Knowing the MPI message size can be useful when evaluating
new interconnect technologies.
The small messages are typically used for group communications for coordinating the processes.
Most of those MPI messages fall between 0B to 64B, and 65B to 256B. There is some
concentration of medium size messages in the range of 64KB to 256KB. On the other hand,
Abaqus/Explicit shows a wide distribution of small message sizes. The small messages peak in the
range from 65B to 256B.
Figure 8: Message Sizes used for MPI communications during Abaqus/Standard and
Abaqus/Explicit job run
Many HPC applications are based on communications patterns that use many small messages
between parallel processes within the job. It is critical that the network interconnect used to
transport these messages provides low latency and high message rate capabilities to assure that

there are no bottlenecks to the application. The InfiniBand architecture can deliver over much
more messages per second to the network than the Ethernet solutions. This increase assures that
there are no message rate limitations for applications and that the multiple cores on the server will
communicate to other machines as fast as the cores are capable, without any slowdown from the
network interface.
6. Observations on File I/O

The file I/O can also be an important element in Abaqus performance.
For Abaqus/Standard, it is observed on that approximately 27 to 46% of the total time on
throughput. There is a mixture of I/O that is accessed to the local file system for scratch, or the
system shared memory space like /dev/shm, and some are accessed over the central file system,
such as NFS or parallel file systems, such as Lustre, PVFS, FhGFS and others.
All of the processes are involved in the disk I/O activity. Most of the I/O activities are write
activities, which accounts for approximately 83% are in the range of 1KB to 8KB sizes. There is
also a reasonable amount of system calls that are made for I/O reads, approximately 72.5% are
1KB to 8KB sizes. It also appears from the I/O profiling that a period of time there it has no I/O
activities, it is assume that the time period is used for running the solver. Figure 9 shows the write
and read activities for the Abaqus/Standard cases. Therefore, a decent file storage that provides
good IOPS and throughput is essential for Abaqus/Standard performance. Additional findings
were reported on the previous HPC Advisory Council Best Practices [5,8]
Figure 9: Write and Read Throughput History for an Abaqus/Standard test case
On the hand, Abaqus/Explicit shows little time for I/O activities compared to the overall run time,
as depicted on Figure 10. The previous HPC Advisory Council Best Practices studies [8] on
Abaqus/Explicit [6], it is found that the first MPI rank, also known as rank 0, handles most of the
file I/O activities. Only approximately 134MB is written and 10MB is read by the rank 0 process.
The rest of the MPI processes are having even less disk I/O activities; the activities are generally
used for loading the shared libraries for Abaqus/Explicit.
Given the IO profiling it shows the benchmark dataset only exhibited very limited IO activities. It
is not clear whether or not the lack of disk activity is due to the nature of the benchmark being
performed for Abaqus/Explicit.

Figure 10: Write and Read Throughput History for an Abaqus/Explicit test case
7. System Architecture
HPC systems typically get periodic update by the CPU refresh cycle. When the new generation of
CPU architecture is introduced, there the performance is expected to be increased. We have
conducted a study on the performance improvement by the different system architecture due to the
CPU generations for the over time.
7.1 Descriptions of System Architecture Compared
To conduct the performance comparison tests, the following system configurations were used:
• WSM: Each “Westmere” system used the HP ProLiant SL2x170z systems, with a dual-
socket 6-core Intel Xeon X5670 running at 2.93GHz, 1333MHz DIMMs, and Mellanox
ConnectX-2 QDR InfiniBand. [10]
• SNB: Each “Sandy Bridge” system used the HP ProLiant SL230s Gen8, each with a
dual-socket 8-core Intel Xeon E5-2680 running at 2.7GHz, 1600MHz DIMMs, and
Mellanox Connect-IB FDR InfiniBand.[9]
• IVB: Each “Ivy Bridge” system used the aforementioned HP ProLiant SL230s Gen8,
each with a dual-socket 12-core Intel Xeon E5-2680 v2 running at 2.8GHz, 1600MHz
DIMMs, and Mellanox Connect-IB FDR InfiniBand [4]
• HSW: Each “Haswell” system, each with a dual-socket 14-core Intel Xeon E5-2697v3
running at 2.6GHz, 2133MHz DIMMs, and Mellanox Connect-IB FDR InfiniBand [11]
22%
13%
73%

Figure 11: Abaqus Performance on system architecture based on the CPU technology over
time
7.2 Differences among System Architecture
By comparing to system architecture of different generations, it becomes clear that the latest
generation of system based on Intel Xeon E5-2697 v3 (Haswell) cluster can deliver gain of up to
22% higher performance over the older Intel Xeon E5-2680 v2 (Ivy Bridge) at 4 nodes.
Likewise, the Intel Xeon E5-2680 v2 (Ivy Bridge) cluster outperforms the Intel Xeon E5-2680
(Sandy Bridge) cluster by up to 13%, it also outperforms Intel Xeon X5670 (Westmere) cluster by
up to 73%. There is a wide gap in performance improvement between Westmere and Sandy
Bridge platforms is most likely because of the improvement in the system architecture beginning
in Sandy Bridge. Notably, the Ivy Bridge architecture enables the PCIe device to run at its
maximum throughput and lowest latency. Thus hardware such as InfiniBand can operate in the
PCIe Gen3 standard, which is able to provide higher throughput over the old PCIe Gen 2 standard.
Among the rest of the changes, there are also an increase in CPU core processing, memory
bandwidth which provide the necessary network throughput for the network. The results are
shown in Figure 11.
8. Software Releases
Another interesting aspect of the component that make a difference in performance is the version
of Abaqus FEA software that is used. Figure 12 shows that Abaqus/Explicit would perform faster
than with the 6.13-2 over the previous version of 6.12-2 by 8% at 4 nodes / 80 CPU cores.
However, the difference in Abaqus/Standard with the version 6.13-2 over the older version of
6.12-2 is not as clear. Marginal gain can be seen on the dataset tested.
8%
Figure 12: Performance Differences in Software Versions between Abaqus 6.12-2 and 6.13-2
9. Conclusion
HPC clustering technology allows Abaqus Simulations to achieve higher productivity by minimize
time-to-solution by deploying high performance systems to run simulation in parallel. The
components in the servers such as CPU, network, and IO can make a great influence on the
Abaqus performance. We have demonstrated that the performance improvement is providing

through over the different CPU generations which also introduce some system architecture. By
adjusting the different frequencies and toggling the Turbo mode setting, we have demonstrated the
benefit of increased Abaqus productivity with higher CPU clock and the tradeoff of increase
power consumption for running the simulation.
HPC cluster environments impose high demands for connectivity throughput, low-latency, low
CPU overhead, network flexibility and high-efficiency in order to maintain a balanced system in
order to achieve high application performance and scaling. Low-performance interconnect
solutions, or lack of interconnect hardware capabilities will result in application performance and
result in an extended time-to-market process.
It is to note that these two benchmark workload obtained from the Abaqus Performance
Benchmarking web site shown in this paper are relatively small compared to the typical problems
that are employed in real world Abaqus simulations which would require the use of an HPC
cluster. Therefore, when evaluating the system performance of HPC cluster, it would worth the
time and effort to use benchmark workload that has larger problem size to run with Abaqus on
similar computer system hardware.
10. References
1. HPC Advisory Council – http://www.hpcadvisorycouncil.com
2. IPM profile - http://ipm-hpc.sourceforge.net/
3. Abaqus performance Data http://www.3ds.com/support/certified-hardware/simulia-
system-information/abaqus-613/performance-data/
4. Abaqus 6.13-2 Performance Benchmarking and Profiling (Ivy Bridge) –
http://hpcadvisorycouncil.com/pdf/Abaqus_Analysis_and_Profiling_H-IVB.pdf
5. HPC Advisory Council – Abaqus/Standard I/O Profiling –
http://hpcadvisorycouncil.com/pdf/Abaqus%20Implicit%20IO%20Performance%20Anal
ysis.pdf
6. HPC Advisory Council – Abaqus/Explicit I/O Profiling –
http://hpcadvisorycouncil.com/pdf/Abaqus%20IO%20Performance%20Analysis.pdf
7. MSR Tools - https://github.com/01org/msr-tools
8. HPC Advisory Council Best Practices - http://hpcadvisorycouncil.com/best_practices.php
9. Abaqus 6.12 Performance Benchmarking and Profiling (Sandy Bridge) –
http://hpcadvisorycouncil.com/pdf/Abaqus_Performance_Analysis_Intel_E5_2680.pdf
10. Abaqus Performance Benchmarking and Profiling (Westmere) –
http://hpcadvisorycouncil.com/pdf/Abaqus%20Performance%20Analysis_Intel_x5670.pd
f
11. Lui, P., “Achieving Higher Productivity on Abaqus Simulations for Small and Mid-size
Enterprises with HPC and Clustering Technologies”. SIMULIA West Regional User
Meeting, San Jose, CA, 2014.
(http://hpcadvisorycouncil.com/pdf/141029a%20SIMULIA%20West%20RUM%20-
%20Pak%20Lui.pdf)

INDUSTRIAL EQUIPMENT

Robust and Optimal Strain Gauge Positions for
Blade Vibration Testing
P. Teufel and J. Poland
ABB Turbo Systems AG and ABB Corporate Research, Switzerland
Abstract: The maximal dynamic stresses on turbo machinery blades used for predicting end of life
due to high cycle fatigue (HCF) are obtained from the correlation between an FE modal analysis
and dynamic strain measurements. Still today it is rather common to use sets of strain gauges to
measure the dynamic strains.
A particular challenge in preparation of blade vibration measurement is the selection of locations

and directions for strain gauges to be mounted, for capturing dynamic blade behavior of various
mode-families.
The paper presents an automatic selection strategy of an optimum gauge placement using greedy
hill climbing and tabu search, considering following criteria:
• gauge performance,
• distinguishability of neighboring modes,
• accuracy regarding positioning robustness,
• measurement robustness regarding possible gauge failure and
• installation constraint regarding gauge size and available space.
The proposed automatic procedure is implemented in Python and can be applied for any Abaqus
FE-modal analysis using any meshing techniques and element types. The placement of strain
gauges and measured strains are written in a new Abaqus odb. The strain distribution within the
actual gauge size is used as an quality indicator regarding the strain gradient. Thanks to this new
developed tool, it is possible to obtain optimum gauge performance under additional robustness
criteria with a minimized total design process cost regarding HCF.
Keywords: strain gauge, high cycle fatigue, blade vibration, automatic pre- and post-processing,
Abaqus scripting language, Python, optimization, hill climbing, tabu search

1. Introduction
High Cycle Fatigue (HCF) is one of the major factor, that can cause blade failure in
turbomachinery. In the design process, modal, forced response and fluid structure interaction (FSI)
analyses are carried out in order to prevent HCF damage. However, the actual blade dynamics
behavior cannot be fully predicted due to uncertainties e.g. aerodynamic damping. A blade
vibration measurement is required for certifying the blade life regarding HCF.
The blade dynamic behavior can be observed by a non-intrusive method using tip timing or an
intrusive method using strain gauges. Both methods are used for blade vibration testing at ABB
Turbo Systems. A big advantage of using strain gauges is to have a real time data processing using
FFT.
The particular challenging in preparing blade vibration measurement is the selection of the gauge
positions and directions to capture all modes of interest. This paper presents an automatic selection
strategy based on a pre-existing method used in ABB Turbo Systems (Szwedowicz, 2012). The
application of the pre-existing automatic tool is limited to a “rectangular” blade surface meshed by
a structured meshing technique using hex elements. The use of the pre-existing tool can be then
counterproductive due to the effort needed for the particular requirement, although it increases the
quality of the HCF design process.
Given a blade geometry of a certain complexity e.g. a turbine blade with damping wire or limited
accessibility for mounting high-temperature strain gauge using flame spray (cf. Figure 1) the
requirement of the pre-existing tool cannot be fulfilled with acceptable effort. The position and
orientation of strain gauges have to be then manually determined. This leads sometimes to
dissatisfying results.
Figure 1. High-temperature strain gauge installation for blade vibration

measurement
For these reasons, a new approach has been proposed to determine the optimum placement of
strain gauges considering following criteria:
• gauge performance,

• distinguishability of neighboring modes,
• accuracy regarding positioning robustness,
• measurement robustness regarding possible gauge damage and
• installation constraint regarding strain gauge size.
2. Optimum gauge position for a mixed flow turbine blade vibration

measurement
To illustrate the developed approach, an Abaqus modal analysis of a mixed flow turbine wheel is
carried out. The cyclic symmetry model was meshed with tetrahedral elements using free mesh
technique. A good model quality can be still achieved with very fine tetrahedral elements using
free meshing technique, which is much less time consuming. The modelling effort is significantly
reduced compared to the modelling using structured technique. Due to the high-temperature
operating, flame-sprayed resistance strain gauges are used for the blade vibration measurement.
The applicable regions are limited to the installation constraint regarding accessibility of flame
spray. The set of strain gauges can be placed on the highlighted regions (two small regions on the
suction side and one small region on the pressure side), cf. Figure 2. The highlighted regions are
defined as sets or surfaces within the Abaqus odb.
2
3
1
Figure 2. Applicable high-temperature strain gauge regions

The optimum placement of 4 strain gauges for capturing dynamic strains of 22 mode-families of
nodal diameter 2- 5 (number of vibration modes of interest = 82), with natural frequencies below

fmax, are requested for blade vibration measurement, cf. the blue highlighted area in Figure 3.
Despite up to 10 telemetry channels available for the blade vibration measurement, only a small
number of gauge placements are determined to have at least two redundant signals of different
blades for capturing amplitude magnification. If the natural frequencies of neighboring mode-
families are close to each other, there is a certain risk that the measured modes cannot be clearly
distinguished by using the frequency level. In this case the measured strains at the same blade are
used to distinguish the neighboring modes. In the given example , the circled mode combinations
in Figure 3 have to be distinguished using a suitable mode identification criterion based on the
dynamic strains.
Figure 3. Dispersion diagram and the critical mode-combinations to be

distinguished
3. Automatic selection strategy of optimum gauge position

The new automated selection approach is implemented in Python. Python is the standard
programming language for Abaqus products and can be used to extend the functionalities that are

not directly available in Abaqus (Simulia, 2014). As the required functionality for strain gauge
application is missing, Python is used here.
The optimum gauge performance criterion considered here relates to transmission coefficient,
which is estimated by the relation between the maximum Mises stress and the potential strain of
the underlying material during the vibration. Therefore, the optimum gauge performance is
associated to the dynamic strain. The dynamic strain data from Abaqus can be exported by using
“Pickle” method. The further data preprocessing can be carried out outside Abaqus.
3.1 Data Preprocessing
Compared to the pre-exiting tool, which can be used only for the structured mesh and a
rectangular-like application region, data preprocessing needs to be changed significantly. The flow
of data preprocessing is shown in Figure 4. Contrary to the pre-existing method, the new proposed
strategy considers all nodes on the highlighted surface in Figure 2 as a center of a strain gauge.
Thus, the new tool can be applied for any shape of surfaces or sets of application regions.
Get E(Re,Im)
Get mesh relation on

application regions
Assign applicable
nodes Abaqus
Calculate unit normal

vector at nodes
Calculate Egauge
Normalize Egauge
Determine mode
combination Python
Figure 4. Data preprocessing flow
As a first step, the real and imaginary part of the nodal strain component are obtained from the
cyclic symmetry FE modal analysis. The average nodal complex strain tensor 𝑬𝒄 (k) in the spatial
Cartesian system of FE model is expressed by

𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚
𝜀𝑥𝑥 + 𝑖 ∙ 𝜀𝑥𝑥 𝜀𝑦𝑥 + 𝑖 ∙ 𝜀𝑦𝑥 𝜀𝑧𝑥 + 𝑖 ∙ 𝜀𝑧𝑥
𝒄 (k) 𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚
𝑬 = �𝜀𝑥𝑦 +𝑖∙ 𝜀𝑥𝑦 𝜀𝑦𝑦 +𝑖 ∙ 𝜀𝑦𝑦 𝜀𝑧𝑦 + 𝑖 ∙ 𝜀𝑧𝑦 �,
𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚
𝜀𝑥𝑧 +𝑖∙ 𝜀𝑥𝑧 𝜀𝑦𝑧 +𝑖 ∙ 𝜀𝑦𝑧 𝜀𝑧𝑧 + 𝑖 ∙ 𝜀𝑧𝑧
where k is a node on an applicable blade region. The application region can be then reconstructed
using the existing mesh information from Abaqus e.g. nodal coordinates, element connectivity.
Considering the installation constraint regarding gauge size, the boundary lines of the application
regions are determined by finding the boundary nodes and elements from the mesh relation
derived from Abaqus e.g. quadratic order elements at boundary always have at least one middle
node, that belongs to one single element in application surface. Nodes lying in the gap between the
boundary and the offset of a half of gauge foil are not considered for mounting. In this given
example, the nodes highlighted in red cannot be used for the instrumentation, c.f. Figure 5.
Figure 5. Region assignment for possible gauge-installed region
The nodal strains derived from Abaqus are in the global Cartesian coordinate system. To calculate
the dynamic strain in gauge directions, the normal vector at each node derived from a weighted
average of the adjacent faces is required. Figure 6 shows the validation of the calculated node
normal of different mesh type. The validation can be carried out by using Python to create a
normal at each applicable nodes, which can be compared with the adjacent element normal
queried from Abaqus.

Figure 6. Element face normal and node normal
𝑐 (𝑘,
The complex dynamic nodal strains 𝜀𝑚 𝜑 ) along unit directional vectors 𝑟⃗(𝜑) about a node
normal axis can be then calculated by the equation below.
𝑐 (𝑘, 𝑻
𝜀𝑚 𝜑 ) = �𝑟𝑥 (𝜑), 𝑟𝑦 (𝜑), 𝑟𝑧 (𝜑)� ∙ 𝑬𝒄 (k) ∙ �𝑟𝑥 (𝜑), 𝑟𝑦 (𝜑), 𝑟𝑧 (𝜑)�
Figure 7 illustrates the measured nodal strain for all possible gauge orientation. It shows, that the
value of dynamic strain is depending on the orientation of strain gauges, i.e. the mode vibration
can be well detected from certain gauge directions.
Figure 7. Measured strains along unit vectors of each 𝝋 about a nodal normal axis

The expected strain to be measured 𝜀𝑚 (𝑘, 𝜑 ) at node 𝑘 in direction 𝜑 is then the absolute value. It
is common that all vibration modes have different maximum response, which can cause a different
trade-off among mode families. To prevent the complex weighting, the nodal dynamic strains
normalized by the maximum dynamic strain of each mode are used for the optimization.
3.2 Optimization
A number of candidates is dependent on the number of applicable nodes and the number of
direction vectors �𝒓⃗(𝝋) about a node normal axis. For the given example 11357 nodes lying on the
highlighted surface, can be used for instrumentation. Each node has 18 measuring axes (𝝋 =
𝟎°, 𝟏𝟎°, ⋯ , 𝟏𝟕𝟎°). The total number of candidate combinations using 4 strain gauges is
𝟏𝟏𝟑𝟓𝟕 × 𝟏𝟖 (𝟐𝟎𝟒𝟒𝟐𝟔)!
� �= == 𝟕. 𝟐𝟕𝟔𝟒 ∙ 𝟏𝟎𝟏𝟗
𝟒 𝟒! 𝟐𝟎𝟒𝟒𝟐𝟐!
Due to such a big number of candidates greedy hill climbing technique is used for finding an
optimum gauge placement. The algorithm has a tendency to converge to local optima. Therefore,
multiple run with different random start value are carried out. Moreover, to improve the
performance the algorithm is extended with tabu search (Glover, 1999).
By the preprocessing described in the previous section, each candidate 𝒊 ∈ {𝟏 … 𝑵} is associated
with a normalized strain sensitivity 𝑨𝒊,𝒋 ∈ [𝟎, 𝟏], referred to as mode coverage in the following, for
each mode 𝒋 ∈ {𝟏 … 𝑴}, where 𝑵 is the number of candidates and 𝑴 is the number of vibration
modes. The coverage of one candidate for one mode is subject to some uncertainty, since there are
simulation inaccuracies, and in the experiment, the placement of the strain gauges may be inexact.
Hence, the coverage values 𝑨𝒊,𝒋 are replaced by robust counterparts by taking the minimum over
some segment rpos around 𝒙𝒊 , cf. Figure 8:
� 𝒊,𝒋 =
𝑨 𝐦𝐢𝐧 𝑨𝒌,𝒋
𝒌: ‖𝒙𝒊 −𝒙𝒌 ‖≤𝒓𝒑𝒐𝒔
Figure 8. Position robustness

Then the basic problem to be solved translates to selecting a subset of number of strain gauges 𝒏
from the candidate set while achieving the best and robust overall coverage, where the basic

definition of overall robust coverage (objective function 𝒇 to be maximized) is given by the least
covered mode:
max𝐼⊂{1…𝑁},|𝐼|=𝑛 𝑓(𝐼), where
𝑓(𝐼) = min max 𝐴̃𝒊,𝒋 .

1≤𝑗≤𝑀 𝑖∈𝐼
Strain gauges that are physically too close together cannot be placed on the same blade, this gives
rise to a constraint. A candidate 𝑖 can be selected only if there is no other candidate within some
radius 𝑟 on the same blade. Depending on the setup of the blade vibration measurement, there are
𝑛𝑏 ≥ 1 blades available for strain gauge placement. Denoting the center position of a candidate 𝑖
by 𝑥𝑖 , candidate 𝑖 can be selected only if
|{𝑘 ∈ 𝐼 ∖ {𝑖}: ‖𝑥𝑖 − 𝑥𝑘 ‖ ≤ 𝑟}| ≤ 𝑛𝑏 − 1.

In order to solve the real problem, mode distinguishability and measurement robustness regarding
possible gauge damage need to be respected, giving raise to complexity of the optimization
problem.
3.3 Post processing
After the optimization, the placement of strain gauges with the actual size can be visualized in
Abaqus viewer. The nodal dynamic strain in gauge orientation is used to calculate a transmission
factor. If the strain distribution along the gauge segment is constant or linear, no error is induced.
Apart from that the average strain over the gauge length has to be performed in order to correct a
transmission factor, which is calculated by using the nodal dynamic strain at the gauge center
(Dally, 2005). The possible error due to strain gradient can be obtained from the strain distribution
within gauge size, cf. Figure 9.

Figure 9. Strain gauge placement and strain distribution within gauge size
4. Result
In this section the results of the variants of the optimization problem are presented. It must be
realized that the selection of the optimum placement of strain gauges generally involves
compromises because of typically conflicting objectives.
4.1 Robust mode coverage
Starting with the optimization of the basic problem, only the strain sensitivity (robust mode
coverage) of vibration modes of interest are taken into account. Finding the optimum mode
coverage seems to be a straightforward task, since the same objective function values were
achieved by using 20 different random start values, c.f. Figure 10. The optimum solution is found
within 2 minutes.

Figure 10. Optimization of mode coverage of the given example
The objective function value returns the minimal mode coverage of all interested vibration mode.
The best possible objective function value is 1.0, if the strain gradient is 0. With this problem
setup the minimum overall mode coverage obtained from the optimization is 0.67 (f3.2) and the
average mode coverage is 0.87, cf. Figure 11.

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
f1 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11 f12 f13 f14 f15 f16 f17 f18 f19 f20 f21 f22
Best mode coverage Best mode coverage Best mode coverage Best mode coverage
Figure 11. Mode coverages using 4 strain gauges considering one gauge per mode
4.1.1 Strain gauge failure robustness
Depending on the test operation turbocharger blade can rotate at a rotational speed of 60000 rpm
up above. Due to such a huge load strain gauges may be damaged during the test operation and
cannot deliver the measurement signal. In order to protect against these failures, one may request
good coverage of at least two or even more strain gauges per mode. Figure 12 shows the mode
coverage of the extended basic problem that requires at least two strain gauges per modes. The
optimization focuses on the second best mode coverage. Therefore, the best mode coverage fall off
in quality. The minimum of best mode coverage of the best coverage is 0.35 (f4.2) and the
minimum of second best mode coverage is 0.34. The average of best mode coverage is 0.555 and
the second best coverage is 0.546. The normalized strain of the best and the second best mode
coverage have almost the same level. Therefore, a trade-off parameter between the best mode
coverage and second best mode coverage is introduced for improving the best mode coverage.

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
Best mode coverage Best mode coverage Best mode coverage
Best mode coverage 2. Best mode coverage 2. Best mode coverage
2. Best mode coverage 2. Best mode coverage
Figure 12. Mode coverage of the best and the second best strain gauges without
weighting
The weighting between the best and second best mode coverage has to be suitable selected in
order to get an acceptable trade-off for a robustness regarding gauge damage without a big loss of
mode coverage. The algorithm can now focus on the best mode coverage as desired. The second
best mode coverage still have an acceptable level. Figure 13 shows the enhancement of the best
mode coverage compared to Figure 12. The second best signal coverage of 3 vibration modes have
instead a poor quality.

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
Best mode coverage Best mode coverage Best mode coverage
Best mode coverage 2. Best mode coverage 2. Best mode coverage
2. Best mode coverage 2. Best mode coverage
Figure 13. Mode coverage of the best and the second best strain gauges using a
proper weighting between the best and the second best mode coverage
4.1.2 Distinguishability of neighboring modes
If modes are very close in frequencies, it is important to be able to distinguish them by a pair of
strain gauges, where one of them has a good coverage of the first and bad coverage of the second
mode, and the opposite for the other strain gauge. This is considered in the problem setup in two
alternative ways: strong and soft neighborhood distinguishability criteria. With the strong
neighborhood distinguish criterion, each coverage is directly penalized with the neighboring
coverages. Thus, a strain gauge for one mode is always selected because it covers the mode well
and not the neighbors. With the soft criterion in contrast, a strain gauge for one mode may be
selected just for good coverage, while another strain gauge may be responsible for
distinguishability to a neighbor.
Figure 14 shows the proposed optimum placement using strong neighborhood distinguishability
criterion. For the given example we are going to look at the nodal diameter 4 in detail. Each
neighboring mode combination can be well captured from at least one specific strain gauge. The
first combination consists of mode families 7 and 8. Strain gauge 3 and 4 are responsible for mode
families 7. Strain gauge 2 can capture the mode families 8 well. The Neighboring modes can be
well distinguished. However, due to the very strong criterion the minimum mode coverage drops
now to 0.3 (f1.4). The average mode coverage is 0.76. The optimization result shows that the mode
coverage can be suffer from the strong neighborhood criterion.

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
1.4 2.4 3.4 4.4 5.4 6.4 7.4 8.4 9.4 10.4 11.4 12.4 13.4 14.4 15.4 16.4 17.4 18.4 19.4 20.4
strain gauge 1 strain gauge 2 strain gauge 3 strain gauge 4
Figure 14. Optimization result using strong neighborhood criterion

The soft neighborhood criterion is aimed at a compromise in neighborhood distinguishability in
return of an increase in mode coverage. The result obtained from the optimization is shown in
Figure 15.The most neighboring mode can also be well distinguished. Only the mode families 15-
18 cannot be as clearly distinguished as the result obtained from the strong neighborhood criterion.
The minimum mode coverage is now 0.45 and the average coverage is 0.79.

1.2
0.8
0.6
0.4
0.2
0
1.4 2.4 3.4 4.4 5.4 6.4 7.4 8.4 9.4 10.4 11.4 12.4 13.4 14.4 15.4 16.4 17.4 18.4 19.4 20.4
strain gauge 1 strain gauge 2 strain gauge 3 strain gauge 4
Figure 15. Optimization result using smooth neighborhood criterion
5. Conclusions
The automatic procedure for the optimum and robust placement of strain gauges related to blade
vibration measurement has been developed based on the pre-existing tool used in ABB Turbo
Systems. The automatic selection method is implemented in Python and can be now applied for
any blades using Abaqus FE-modal analysis. The placement of strain gauges and expected strain
to be measured are written in a new Abaqus odb. The strain distribution within the actual gauge
size is used as an quality indicator regarding the strain gradient.
Thanks to this new developed tool, it is possible to obtain optimum gauge performance under
additional robustness criteria with a minimized total design process cost regarding HCF.
6. References
1. J. Szwedowicz, S. M. Senn and R. S. Abhari, “Optimum strain gage application to bladed
assemblies”, ASME J. Turbomachinery, volume 124(4), 2002
2. Simulia, “Abaqus/Scripting Reference Guide”, 2014
3. F. Glover and M. Laguna, “Tabu Search”, Kluwer Academic Publishers, 1999
4. J. W. Dally J and W. F. Riley, “Experimental stress analysis”, 4th Edition, 2005

7. Acknowledgement
Thanks to Armin Schmid and Dr.-Ing. Ulrich Schaber from ABB Turbo Systems for contributing
to the success of this project with their considerable experience.

Prediction of Tin-Whiskers Generation
Using Stress and Mass-Diffusion Analysis
Takeshi Terasaki
Hitachi, Ltd., Hitachi Research Laboratory
Abstract: A spreading environmental awareness has prompted a global movement towards lead-
free electronic products, in which lead-free tin plating is included. Lead-free tin plating provides
good solderability equal to tin-lead plating, however the spontaneous whisker growth on lead-free
tin plating is a serious reliability issue. It's imperative to clarify the whisker generation and
growth mechanism. So we have developed a multi-scale simulation technique for calculating tin
atom diffusion caused by stress gradient induced by temperature change and crystal orientation
distribution in polycrystalline tin plating. This multi-scale simulation technique uses molecular
dynamics (MD) simulation and a finite element analysis (FEA). The FEA simulates macro-scale
stress and stress-induced mass-diffusion, and the MD simulates nano-scale atomic diffusion. The
stress analysis model considers elasticity anisotropy, thermal-expansion anisotropy, and the
crystal orientation of tin. A thermal cycling test was conducted to induce whisker generation on
tin-plated specimens, and the crystalline orientations around the whiskers were evaluated using
electron back-scattering diffraction pattern (EBSP) measurement. The hydrostatic pressure
distribution and tin-atomic-density distribution in the specimens were calculated using our
technique. The whisker locations corresponded to the areas of higher tin-atomic density and lower
hydrostatic pressure on the tin-plated surface, indicating that our technique can predict whiskers
generation in tin platings.
Keywords: Tin whisker, Tin plating, Electronic product, Lead-free, Finite element analysis,
Atomic diffusion, Stress gradient, Crystal orientation, Elasticity anisotropy
1. Introduction
Electronic systems, which include the electronic components, form the foundation for today's
convenient and comfortable world. Tin–lead plating is commonly used for electronic components
because it has excellent solderability and tin whisker resistance. Tin whiskers are microscopic
single-crystal metal fibers which grow from tin platings. They can grow at random and create
conductive paths between unintended leads on electronic components. The electrical short-circuits
may seriously interfere with performance of electronic systems. However, increasing
environmental awareness is leading to restrictions on the use of lead in electronic products
worldwide. Many producers are thus using pure tin or tin alloy plating (hereinafter collectively
referred to simply as “tin plating”) instead of conventional tin–lead plating because it has the same
solderability as tin–lead plating although it is more susceptible to tin whisker generation and
growth.

The major factor in the whisker growth is apparently compressive stress generated in the tin
plating. This compressive stress induces diffusion and local concentrations of tin atoms, which lead
to generation and growth of tin whiskers. Sources of this compressive stress include formation of
inter-metallic compounds at the tin/lead-frame interface (Lee, 1998), oxidation of tin (Barsoum,
2004), and cyclical changes in temperature (Zhao, 2006). The three-dimensional finite element
analysis (FEA) model established by Zhao et al.—which considers elasticity anisotropy, thermal-
expansion anisotropy, and β -tin plasticity—is an effective method for evaluating whisker growth
caused by cyclical changes in temperature (Zhao, 2006). However, it does not provide information
about the relationship between the strain-energy density and tin-atomic localization induced by
atomic diffusion. Tin-atomic localization, that is, the change in the atomic density distribution of
the tin, is a key factor in tin whisker growth.
A simulation technique for evaluating the atomic density distribution of tin was developed by our
group (Terasaki, 2009). It uses molecular dynamics (MD) simulation for nano-scale atomic
diffusion analysis, FEA for macro-scale stress and mass-diffusion analysis, and x-ray diffraction
(XRD) for crystalline orientation measurement. We investigated the possible use of this technique
to predict the locations of generated whiskers (Terasaki, 2012). Thermal cycling tests were
performed to induce whisker generation on tin-plated specimens. The crystalline orientations of the
tin film were evaluated in the vicinity of the whiskers using electron back-scattering diffraction
pattern (EBSP) measurement. The simulation results indicated that the areas of higher tin-atomic
density and lower hydrostatic pressure on the tin-plated surface were consistent with the locations
of generated whiskers.
2. Test procedure
2.1 Thermal cycling test
Test pieces were fabricated from a lead frame constructed of 125- µm -thick commercial copper
material plated with pure tin to a thickness of 10 µm . The crystal grains in the tin plating were β -
tin. After plating, the test pieces were kept at 150°C for one hour, and then a thermal cycling test
was conducted. The conditions for the test were −40°C to 85°C, 60 minutes per cycle (20 minutes
of holding at low temperature, 10 minutes of heating, 20 minutes of holding at high temperature,
and 10 minutes of cooling). There were 1000 cycles.
2.2 Crystal orientation measurement
After the thermal cycle test, we examined the surface of the tin plating with a scanning electron
microscope (SEM) and selected a 100 µm × 100 µm region containing multiple whiskers to
measure the crystal orientations of the tin grains. Because the tin-plated surface after the thermal
cycle test had protrusions and depressions, we shaved 6 µm off the surface of the plating in the
horizontal direction using a rotating microtome to enable evaluation of the crystal orientations in
the target region by EBSP measurement. We used a TSL MSC-2200 orientation-imaging
microscopy system operating at an acceleration voltage of 20 kV for the EBSP measurement.
EBSP measurements gave the idetified crystal orientations as Bunge Euler angles.

3. Calculation procedure
3.1 Finite element analysis model
For the macro-scale stress and mass-diffusion analysis, we used ABAQUS/Standard version 6.9.
An in-house general preprocessor was used to build the FEA model shown in Figure 1. Since the
crystal grains in the tin plating had a columnar organization that was as thick as the plating, the
grains were modeled as columnar crystals. Their shape was derived from the EBSP measurements.
The entire analysis model was constructed in an orthogonal coordinate system (x-y-z), with the tin
plating surface in the x-y plane and the plating thickness in the z direction. The analysis region was
102 µm × 105 µm . A 0.1- µm -wide region was set at the crystal grain boundaries. The thickness
of the copper lead fame was 100 µm , which is close to the actual thickness. The element
segmentation was done using 3D cubic elements with eight nodes (C3D8) to evaluate volume
fractions of relative tin atomic density at grain boundaries (Terasaki, 2009), resulting in an analysis
scale of 219,472 nodes and 206,130 elements.
105 µm
Columnar crystal
102 µm
Grain boundary
(0.1 µm wide)
y Copper
z 10 µm lead frame
x
Figure 1. FEA model for stress and mass-diffusion analysis.
3.2 Macro-scale stress analysis
The crystals in the tin plating are β -tin. Since β -tin has an orthotropic body-centered tetragonal
structure with c-axis about half the length of its a-axis, orthotropic elasticity and thermal expansion
orthotropy are significant (Revelo, 1997). The elastic constants of β -tin are listed in Table 1
(Allard, 1969), where 1, 2, and 3 indicate the perpendicular components of the a, b, and c axis
directions and 4, 5, and 6 indicate the respective cross-section components for the 12, 23, and 31
directions. The coefficient of thermal expansion (CTE) of β -tin is α 1 = α 2 = 15.8×10−6 /°C, and
−6
α 3 = 28.4×10 /°C (Zhao, 2006). A local orthogonal coordinate system (1-2-3) was defined for
each column-modeled crystal, and the Euler angle determined by EBSP measurement was set as
the angle of rotation of the local orthogonal coordinate system relative to the overall orthogonal
coordinate system.

The copper lead frame was assumed to be an isotropic elastic body with a Young's modulus of 127 (3)
GPa, a Poisson ratio of 0.3, and a CTE of 17.0×10−6 /°C. Since the macro CTE of the tin plating
was higher than that of the copper lead frame, an increase in temperature created compressive
stress in the tin plating. As mentioned above, this compressive stress is apparently the major factor
in whisker growth. Accordingly, we subjected the FEA model to a thermal load produced by a
uniform 100°C (i.e., 25°C to 125°C) rise in temperature.
Symmetric boundary conditions were applied to two cross sections of the 3D model (the lower left
area in Figure 1). Two other cross sections (the upper right area in Figure 1) were modeled under a
node-coupling boundary condition. The tin-plated surface was modeled under a free-boundary
condition. The bottom of the lead frame was fixed in the z direction.
Table 1. Material constants of tin crystal.

Elastic tensor
C11, C22 C33 C44, C55 C66 C12 C13,C23
component
Material
72.30 88.40 22.03 24.00 59.40 35.78
constant (GPa)
3.3 Macro-scale diffusion analysis
Abaqus/Standard provides the general mass-diffusion analysis capabilities. The governing

equations for mass-diffusion are an extension of Fick's equations, which allow for mass diffusion
driven by gradients of temperature and pressure (Abaqus Users Manual):
 ∂φ
J = − sD ⋅  + κ s
( (
∂ ln θ − θ z ))
∂p 
+κp  ,
 ∂x ∂x ∂x 
(1)
where J is concentration flux in the diffusing phase; s is solubility; φ is the “normalized
concentration, i.e., φ = c s , where c is the mass concentration; D(c ,θ , f ) is diffusivity;
κ s (c ,θ , f ) is the “Soret effect” factor (diffusion driven by temperature gradient); θ is
temperature; θ z is absolute zero on the temperature scale used; κ p (c ,θ , f ) is the pressure stress
factor (diffusion driven by gradient of equivalent pressure stress p); and f is any predefined field
variable. The boldface letters represent tensors.
In general, atoms diffuse at grain boundaries because the coefficient of grain-boundary diffusion is
much higher than the coefficient of intra-particle diffusion. The stress-induced atomic flow along a
grain boundary, J α , is defined as a function of the stress perpendicular to the grain boundary, σ v
(Needleman, 1980):
Dα ∂µ ,
J α ( xα ) = −
κT ∂xα
µ = µ0 − σ v Ω , (2)

where Dα is the coefficient of grain-boundary diffusion, κ is Boltzmann's constant, T is absolute
temperature, xα is a local coordinate system along the grain boundary, µ is the chemical
potential, µ 0 is the reference potential, and Ω is atomic volume. Assuming a proportional
relationship between the stress perpendicular to the grain boundary, σ v , and hydrostatic pressure p
makes it possible to use the mass-diffusion analysis function of the ABAQUS/Standard software
for grain-boundary diffusion analysis. Hydrostatic pressure is written as
p = − trace(σ ) 3 ,
where σ is a stress tensor. (4)

Mass-diffusion analysis based on Equation 1 was conducted using the FEA models shown in
Figure 1. The values of p were obtained from the stress analysis described in the previous section.
Steady-state analysis was performed because the holding time used in the thermal-cycling test (20
minutes) was sufficient for stabilizing the diffusion behavior of tin atoms. The initial concentration
was 100 %, and the solubility was 1. Since a uniform thermal load was used in the modeling, the
temperature gradient effect was ignored, i.e., κ s = 0 .
3.4 Nano-scale atomic diffusion analysis
To conduct mass-diffusion analysis by FEA, we needed the value of diffusion coefficient D for
solid-state β -tin. Since this value has not been reported, we estimated the values of the self-
diffusion coefficient and grain-boundary diffusion coefficient of β -tin by MD simulation. To
simplify the mass-diffusion analysis, D was assumed to be isotropic.
The number of atoms in the bulk model was assumed to be 1×104, and that in the grain-boundary
model was assumed to be 2×104. To take into account the diffusion of copper atoms from the
copper lead frame, we assumed tin with 1 atomic percent of copper as the bulk model. The
boundary between grain A and grain B was assumed to be Σ 5(110). Periodic boundary conditions
were applied in the x, y, and z directions in each model. To take into account the dependence of
stress on the diffusion coefficient, we calculated the diffusion coefficients at 20°C under
compressive stress σ x normal to the grain boundary acting on the system at 0, 50, and 200 MPa.

4. Test results
4.1 Whisker growth locations

An SEM photograph of a tin-plated surface after the thermal cycle test is shown in Figure 2. The
protrusions and depressions correspond to tin columnar crystals. We consider the protrusions to be
very short whiskers. Five very short whiskers are evident in the 102 µm × 105 µm region. The
whisker locations are indicated by yellow arrows. Five depressed areas, indicated by pink arrows,
are evident near the whiskers. From the adjacency of the whiskers and the depressions, we can
infer that tin atoms from the depressions fed the growth of the whiskers.
102 µm
Whisker
Depression
105 µm
Figure 2. SEM image of tin-plated surface after thermal cycling test.

4.2 Crystal orientation measurement
The crystal orientations within the tin plating shown in Figure 2 were evaluated by shaving a 6- µm
layer off the surface of the plating and then performing EBSP measurement, as mentioned above.
The shapes and crystal orientations of the measured crystal grains are shown in Figure 3. There
were 215 crystal grains within the measured area. The whisker growth areas (enclosed by light
yellow lines in the figure) include many pink-colored crystal grains, which have an orientation
close to (001). The depressed regions enclosed by the pink lines include blue or green grains,
which respectively have orientations of close to (110) or (100). We performed a stress analysis that
took crystal orientation into account.
As previously mentioned, β -tin has a large CTE in the c-axis direction. The CTE of the (001)-
oriented crystal grains in the planar direction of the tin plating was about the same as that of the
copper lead frame because the c-axis of the crystal grains is in the depth direction of the tin plating.
Therefore, a temperature change produced very little stress in the tin plating. The c-axes of the
(110)- and (100)-oriented crystals, however, were in the plane of the tin plating. Therefore, the
CTE of the grains was greatly different than that of the copper lead frame, and a change in
temperature produced large stress in the tin plating.
Whisker
Depression
110
β-Sn
001 100
Figure 3. Shapes and crystal orientations of measured crystal grains.

5. Calculation results
5.1 Diffusion coefficients
The diffusion coefficients estimated by MD simulation are illustrated in Figure 4. The diffusion
coefficients at the grain boundaries were much higher than those in the grains, indicating that
compressive stress increases the diffusion coefficient. Although it is believed that grain-boundary
diffusion is a more dominant factor in tin whisker growth than bulk diffusion, the MD model can
quantify the difference between grain-boundary diffusion and bulk diffusion. The simulated
diffusion coefficients were used in the following mass-diffusion analysis, and κ p at 1 was
estimated roughly from the MD calculation results.
1.E-20
1.E-20
Diffusion coefficient (m2/s)
1.E-30
1.E-30
1.E-40
1.E-40 GB: Σ5 (110)
1.E-50
1.E-50
bulk
1.E-60
1.E-60
00 50
50 100
100 150
150 200
200 250
250
Applied compressive stress σ x (MPa)
Figure 4. Simulated diffusion coefficients in bulk and at grain boundary.

5.2 Hydrostatic-pressure distribution
Figure 5 plots the hydrostatic pressure distribution on the tin-plated surface obtained from the
stress analysis. The orange and red indicate regions in which there was hydrostatic pressure of
from 40 to 50 MPa and stronger compressive stress. The light blue and dark blue indicate regions
in which there was hydrostatic pressure of from 0 to 15 MPa and weaker compressive stress. The
black indicates regions in which there was negative hydrostatic pressure and tensile stress of near
zero. The whisker growth locations and depressed areas indicated by the arrows in Figure 2 are
respectively marked by light yellow and pink polygons in Figure 5.
The whisker locations are consistent with regions where the hydrostatic pressure was 10 MPa or
less. The depressed areas are, with one exception, regions of from 20 to 30 MPa hydrostatic
pressure that lie between regions of weak hydrostatic pressure of 10 MPa or less. Because whisker
growth results from atomic diffusion driven by compressive stress, the crystal grains in the
depressed areas supply the atoms required for whisker growth. It is thus clear that stress analysis
that takes into account the anisotropy of elasticity and CTE of the tin-crystal grains and the crystal
shape can be used to predict the locations of whiskers that grow during a thermal cycle test.
50
40
30
20
Whisker
10
0 Depression
(MPa)
Figure 5. Hydrostatic-pressure distribution on tin-plated surface.

5.3 Tin-atomic-density distribution
Figure 6 plots the tin-atomic-density distribution on the tin-plated surface obtained from the
diffusion analysis. Comparing it with the hydrostatic pressure distribution shown in Figure 5, we
see that regions of high hydrostatic pressure (30 MPa or more) correspond to the low concentration
areas and that regions of low hydrostatic pressures (10 MPa or less) correspond to the high
concentration areas.
In Figure 6 the locations of whiskers and depressions are respectively marked with light yellow
and pink polygons. At the whisker positions and in their vicinities, there are portions where the tin-
atomic density was high, which confirms that tin-atomic diffusion driven by compressive stress is
the mechanism for whisker growth. In the depressed areas, the tin-atomic density was not low, but
the tin-atomic density in the vicinity of the depressed areas was low. Therefore, we can conclude
that regions of high compressive stress near whisker locations supply atoms for whisker growth.
145
135
125
115
105
Whisker
95
Depression
(%)
Figure 6. Tin-atomic-density distribution on tin-plated surface.
6. Conclusion
We have shown that a developed simulation technique can be used to predict the locations of
whiskers that grow on tin plating during a thermal cycle test. We first fabricated tin plating on a
copper lead frame, performed a thermal cycle test, observed the whisker formation, and evaluated
the crystal orientations of the crystal grains in the region of whisker growth by EBSP
measurement. Next, we used the crystal orientation measurements to perform a stress-diffusion
analysis, which showed that the areas of low hydrostatic pressure and high tin-atom density were
consistent with the locations of whisker formation.

7. References
1. Allard, S., “International Tables of Selected Constants Vol. 16, Metals”, Thermal and
Mechanical Data, Pergamon Press, Oxford , 1969
2. Barsoum, M. W., Hoffman, E. N., Doherty, R. D., Gupta, S. and Zavalaingos, A., “Driving
force and mechanism for spontaneous metal whisker formation,” Phys. Rev. Lett., vol. 93,
article number 206104, 2004.
3. Lee, B.-Z. and Lee, D. N., “Spontaneous growth mechanism of tin whiskers,” Acta Mater.,
vol. 46, no. 10, pp. 3701–3714, 1998
4. Needleman , A. and Rice, J.R., Acta Metallurgica, vol. 28, p. 1315, 1980.
5. Ravelo, R. and Baskes, M., “Equilibrium and thermo-dynamic properties of grey, white, and
liquid tin,” Phys. Rev. Lett., vol. 79, no. 13, pp. 2482–2485, 1997.
6. Terasaki, T., Iwasaki, T., Ookura, Y., Suzuki, T., Kato, T., Nakamura, M. and Hashimoto, T.,
“Evaluation of Tin-whisker Growth During Thermal-cycle Testing Using Stress- and Mass-
diffusion Analysis”, Proc 59th Electronic Components and Technology Conf, San Diego,
CA, pp. 277–284, 2009.
7. Terasaki, T., Kato, T., Iwasaki, T., Ookura, Y., Nakamura, M., Ishii, H. and Yamamoto, K.,
“Prediction of Tin-Whiskers Generation during Thermal Cycle Test Using Stress and Mass-
Diffusion Analysis”, Proc 62nd Electronic Components and Technology Conf, San Diego,
CA, pp. 1183 - 1189, 2012.
8. Zhao, J.H., Su, P., Ding, M., Chopin, S. and Ho, P. S., “Microstructure-Based Stress
Modeling of Tin Whisker Growth”, IEEE Transactions of Electronics Packaging
Manufacturing, vol. 29, no. 4, pp. 265–273, 2006.
9. Abaqus Users Manual, Version 6.9, Dassault Systemns Simulia Corp., Providence, RI.

Strength and Fatigue Life Analysis of Leaf Springs
used in Heavy Duty Trucks
Mehmet Bakır, Murat Sıktaş, Serter Atamer

Mercedes-Benz Türk A.Ş.
Abstract: A noticeable number of weight and cost reduction projects are performed with the help
of highly improved computer aided engineering (CAE) tools within the vehicle engineering
development activities. In this paper, a complete optimization study is applied to a leaf spring with
5 leaves used in the rear axle suspension systems of 6x4 heavy duty trucks by reducing the number
of leaves down to 4 together with weight and cost reductions.
A brief overview of the project is as follows; the stiffness of the leaf spring is calculated with in-
house software based on mathematical calculations using the thickness profile of the leaves. Then
the results of these calculations are compared with non-linear elastic leaf spring calculations
which are conducted with Abaqus. This elastic leaf spring finite element (FE) model is transferred
into multi body simulation (MBS) model, which is established with Simpack, in order to determine
the forces acting on the leaf spring. Using the results of the MBS calculations, which are time
histories of the internal forces and moments on the leaf spring, the FE simulations are performed
with Abaqus. Stress results are not always enough to conclude that the leaf spring will preserve its
integrity until the desired lifetime although they may give a rough estimation. Concurrently,
without fatigue life analysis it is not correct to make a final decision. Therefore, as the final step of
the project, a multi-channel fatigue life calculation process is performed to assess the durability
characteristics of the new design, showing that the weight and cost reduction is achievable
without worsening the fatigue life of the leaf spring.
Keywords: Damage, Dynamics, Elasticity, Fatigue Life Analysis, Finite Element Analysis, Multi-
Body Dynamics, Optimization, Suspension.
1. Introduction
Multiaxial loads are acting on the leaf spring assembly in a truck structure and therefore it has
great influence on the drivability of the truck. Because of this reason, the new design with reduced
number of leaves (Figure 1) should withstand all forces like the current design assuring same
strength and fatigue characteristics. In other words, it should be lower in cost and weight but at the
same time the new leaf spring assembly should, however, meet durability requirements. This fact
brings great responsibility to the development process of the new designed leaf springs. Hence, a
complete optimization process [11] is applied for the entire design cycle, the results of which are
described in this paper.

Figure 1. General overview of both leaf springs
2. Calculation Steps
Rather than a typical calculation process based on static finite element analysis, this study has a
distinctive point of view on the durability assessment of a leaf spring because various calculation
methodologies are applied for a comprehensive calculation process. All steps in Figure 2 are
necessary for the optimization process of the leaf spring. They are to be explained within the
following sections.
Figure 2. Flow of the Calculation Steps

2.1. Basic Checks with BLAFES
BLAFES [1] is a FORTRAN based in-house developed software for dimensioning and calculation
of leaf springs based on mathematical calculations. The main aim of this software is to have a
rapid calculation of the stiffness of the leaf spring under vertical loading. Additionally a general
overview of the stress results is also achieved. Having the stiffness values and the general
overview of the stress results available in hand at the first step, is very important to make the first
assessment within the design phase. This prevents time loss from the beginning. Therefore, this
significant first step is helpful in order to reach a feasible design to implement within the
following steps of the calculation flow. The new design should have expected stiffness values like
the currently used original leaf spring. Hence, a similar or a predictable safety, comfort and
dynamic vehicle behaviors are accomplished.
The reference stiffness value of the original leaf spring, which is the one from the producer, is
3636.3 [N/mm]. The stiffness value calculated by BLAFES is shown in Figure 3 and found as
3595.7 [N/mm] (Equation 1). The results indicate that the stiffness values from the producer and
from the simulation of the leaf spring are matching each other. The comparison here is done in
order to see how much difference there is between the simulation value and the producer’s value.
After the comparison, it is found that there is only 1 % difference between the stiffness values,
which indicates that the model verification is fulfilled. Having the basic stiffness check
accomplished, it is decided to continue with further steps through the calculation flow shown in
Figure 2.
Figure 3. Stiffness results of the original leaf spring
144400[N ]  N  (1)
koriginal = ≅ 3595.7  
40.16[mm]  mm 

The stiffness value of the new designed leaf spring calculated with BLAFES is shown in Figure 4
and found as 3882.4 [N/mm] (Equation 2). In comparison to the one of the original leaf spring, it
can be observed that there is approximately 8 % increase in the stiffness value which is already
expected and considered to be within the required tolerances. A similar stiffness comparison
method is also done both in elastokinematic and FE-calculations in the next steps.
Figure 4. Stiffness results of the new designed leaf spring
115000[N ]  N  (2)
knew _ designed = ≅ 3882.4  
29.62[mm]  mm 
2.2. Elastokinematic Calculations

The aim of this step is to make another check of the stiffness value with a simplified finite element
model. More importantly another purpose of this step is to prepare a leaf spring model which is to
be used in the subsequent multi body simulations (MBS).
Using the unloaded geometry data of the leaf spring, the finite element model is created with a
number of bar elements in Medina [2] along the length of the leaves. The characteristics of each
leaf are defined in MakSim (patent applied for) which is an in-house software to create Abaqus
input files to define section properties according to section dimensions of leaf spring.

Figure 5. MakSim’s tasks during simulation process
This final model is calculated with Abaqus [4] in which a vertical loading is applied. The stiffness
value is once more checked in this point. There is nearly 8 % of stiffness increase of the new
designed leaf spring in compare to the original one according to elastokinematic calculation results
which is very close to the comparison based on BLAFES results.
After the result of the elastokinematic model (Figure 6) is verified, the model is converted into a
data which can also be imported and used in MBS. As a final step of the elastokinematic part, the
model is delivered to the MBS.
Figure 6. Overview of an elastokinematic model
2.3. MBS Calculations

The aim of this calculation step is to build up an accurate multi body simulation model and to
calculate the load collective on the leaf spring.
First, wheel forces are measured from a test vehicle at the torture track using wheel force
transducers (Figure 7). Torture tracks are circuits of various sizes, skid pans, bad road surfaces etc.
The advantage of torture track testing is to shorten the total durability testing time. There is also
digitalized road data which is the measured road profile of these tracks. With a torture track
calculation, this road data is applied to the full vehicle simulation model and a dynamic simulation
of the ride of the vehicle on this track is performed. For each wheel 6 channels are measured

(Forces in x, y, z directions and moments in x, y, z directions). Spring deflections and
accelerations are also measured from the vehicle in order to compare the MBS model. The
measured forces and the moments are applied to MBS model as input values.
Figure 7. Wheel force measurements
Secondly, the multibody simulation model is prepared. At this part, only the rear axle model is
modeled as the focus is the leaf springs in the rear axle system. Axles, suspensions, leaf springs,
longitudinal control arms, triangular control arms, anti-roll bar are modelled including all non-
linear characteristics of these elements within the MBS model of the rear axle in SIMPACK [3]
(Figure 8).
Figure 8. Rear axle multi body simulation model
The comparison of the MBS model is also shown as measurements and calculation results (Figure
9). Spring deflection and acceleration data, which are also measured during wheel force
measurements, are used to compare the calculated spring deflection and accelerations. As it can be
observed from the Figure 9, very close results are achieved from the measured and calculated
spring deflections.

Figure 9. Comparison between measured and calculated spring deflections
After the MBS model comparison, the measured wheel forces are applied to MBS model and
torture track calculation is performed. Two separate MBS models are built, one of which contains
the original design of the leaf spring with 5 leaves whereas the other one includes the new design
with 4 leaves. Calculations with both MBS models are performed in which the load collectives on
the leaf springs are calculated as time histories of the internal forces (cutting forces) and moments
acting on the leaf springs. In addition, level crossing and range pair counting evaluations of the
calculated forces are conducted.
Figure 10. Level crossing counting of two leaf springs (Force in Z direction)
Figure 10 shows the results of level crossing counting of the forces acting on the front connection
point of each spring in vertical direction. The results indicate that the forces acting of the new
designed leaf spring has a more intensive level crossing distribution than the forces acting on the
original one.

Figure 11. Range-pair counting of two leaf springs (Force in Z direction)
Figure 11 shows the range pair counting of the forces acting on the front connection point of each
spring in vertical direction. The results here also indicate that the forces acting of the new
designed leaf spring has a more intensive range pair distribution than the forces acting on the
original one, which shows clearly that the new designed leaf spring undergoes a heavier loading.
The calculated time histories of the internal forces and moments on the leaf springs are used
within the further analysis steps the calculation process.
2.4. FEM Calculations
Unloaded leaf spring geometry is used for finite element modeling in Medina. Clamping plate and
the U-bolts are also modeled and mounted on the leaf spring model in order to reach more
accurate results. HEXA elements are used to model the leaves of the spring and for the sandwich
plates between the leaves. TETRA elements are used to model the clamping plate and the U-bolts.
The original leaf spring model with 5 leaves has 170731 nodes and 121647 elements and the new
designed leaf spring with 4 leaves has 134899 nodes and 93623 elements. Same material is used
for both leaf springs and that means the differences are purely geometric. Figure 12 shows a
general view of the calculation model. The calculations are performed with Abaqus approximately
in 2 hours with 8 processors at 2.93 GHz.

Figure 12. Unloaded original leaf spring finite element model
Contacts between the leaves are defined and boundary conditions are also applied in the model.
The definition of the contacts and boundary conditions is essential to have a base model to work
on; however, applying the load cases has also a great importance to reach accurate results. One
other reason for the importance of the load cases is that the same finite element model is also used
for the subsequent fatigue life calculation step with FEMFAT [5] using the stress results which are
calculated in Abaqus and the internal forces and moments coming from MBS. A deformed shape
of the leaf spring assembly, at one of the critical time points with high vertical loading, can be
seen in Figure 13.
Figure 13. Deformed shape of the leaf spring assembly
As a first step of the analysis, pre-tension forces are applied to each U-bolt for clamping. After
assuring that the pre-tension works correctly in the simulation, next step is to apply the forces and
moments acting on the leaf spring. For the calculations and comparison based on stress values, the
applied forces and moments are selected from the complete time histories from MBS. For this
purpose, the values of the forces and moments at the time point, at which the maximum vertical
force is observed, are used. The forces are applied on both fixing sides of the spring. Simulations
are performed by Abaqus. Pre- and post- processing is done with Medina. One outcome from the
FE-simulations is stiffness results. There is roughly 7 % stiffness increase of the new designed leaf
spring in compare to the original one matching very well with the previous BLAFES and
elastokinematic calculations results. Another outcome is the stress results. Therefore, principal
stress results (first and the last leaf in detail) are shown in Figure 14. The stress and the damage
values within the figures throughout the paper are scaled to the maximum value in the
corresponding figure showing the maximum as 1.00 and the others scaled in percentage (%)
values in relation to the maximum.

Figure 14. Principal stresses on the original leaf spring
In Figure 15, a comparison of the original and the new designed leaf spring is shown. It can easily
be observed that, although there is an increase in forces and moments coming from MBS for the
new designed leaf spring, the stresses are decreasing in all leaves of the new design. This is a
result of the optimized design with higher thickness distribution; however, a fatigue life
calculation step should be performed to have an accurate conclusion regarding the durability
characteristics of the new designed leaf spring.

Figure 15. Comparison of the principal stresses of the leaf springs
Preparation for fatigue life calculation is also done within the FE-calculation step. In order to
obtain a comparison of the maximum stresses, the time step with maximum forces and moments in
vertical direction is applied to the model and the results are shown Figure 15. However for the
fatigue life calculations, maximum forces and moments of time histories in each direction should
be applied to have necessary input data for channel-max FEMFAT calculations.
2.5. FEMFAT Calculations

Stress results are not always enough to make a decision that confirms the sufficient durability
characteristics of the new design. As the final step of the analysis, fatigue life calculation is
performed on both leaf springs (the original and the new design). From the modeling side, same
models which are used for FE-analysis are also used for FEMFAT calculations.
Channel-max calculation type is selected in FEMFAT for multi-channel fatigue life calculation
based on static superposition approach [5, 6, 7]. Clamping areas including the bolting holes are
excluded from the fatigue life calculation evaluations. 12 channels (Figure 16) are defined as
inputs. Force-time histories, which are calculated in the preceding MBS calculations, are used as
input values in each corresponding channel together with the stress results for the maximum
values of the forces or moments in the time history.

Figure 16. FEMFAT channel-max model
12 force-time history tables are taken from MBS. One example is shown in Figure 17. Each force-
time history input is coupled and superposed with the corresponding stress results so that the stress
time histories are calculated. Using these stress time histories, a fatigue life analysis with static
superposition is then performed by calculating the cumulative fatigue damage according to the
relative Palmgren-Miner rule (modified by Haibach) [8, 9, 10]. That means, for each force and
moment directions, 12 separate load cases with unit loads should be analyzed in the previous finite
element calculations. These results should be used as input in each corresponding channel
separately in channel-max calculations (superposition of the individual stress components at every
point in time) [4].
Figure 17. Sample of a force-time history in Fz

As it is already explained within the previous sections of this paper, the main reason which makes
the results of this particular fatigue life calculation important is; although the stresses on the new
design is lower than the original spring, the range pair calculations show that higher forces are
acting on the new design (Figure 11). Therefore, the comparison of the damage values plays an
incredibly essential role on the accurate assessment of the durability characteristics of the spring.
Hence, the results for each leaf of both original and new leaf spring are illustrated in a comparison
which has both stress and damage values in Figure 18 and 19.

In case of the original design, both the spots and the order of the maximum stress and damage
values are similar, and the most critical point is observed on the 1st (first or the lowest) leaf
(Figure 18).
Figure 18. Comparison – original design
The spots with the maximum stress values and maximum damage values are matching each other
in general for the new designed leaf spring (Figure 19). However, the order of the maximum stress
values and maximum damage values are different (Figure 19). It can be seen that, based on the
stress results the most critical point by far is observed on the 3rd leaf (counting from bottom
upwards), whereas the damage results indicate that the maximum damage value is reached on the
last (4th or the uppermost) leaf. A comparison based on the maximum stress values of the original
and new designed leaf spring would lead to an outcome indicating the new design is superior as
the maximum stress value on the new design is significantly lower than the one on the original
design. Without a fatigue life calculation, such a reduction in the maximum stresses would be
interpreted as a great improvement of the durability characteristics. However, the results of the
fatigue life calculations draw a completely different conclusion. The maximum damage values
observed on the original and the new leaf spring is quite close to each other, meaning that the
fatigue life of both leaf springs is expected to be very similar. The results show that, with the new
designed leaf spring with reduced number of leaves, the aimed weight and cost reduction can be
achieved without worsening the durability characteristics.

Figure 19. Comparison – new design
The results obtained in this study shows clearly that the stress results do not always represent the
durability characteristics of such a component like leaf springs working under highly variable
loading conditions. For a comprehensive durability assessment, a multi-channel fatigue life
calculation should be performed.
3. Summary/Conclusions
The aim of this paper is to present a methodology for a comprehensive durability assessment of
leaf springs with CAE methods, illustrated with a case study in which a new designed cost and
weight optimized spring with reduced number of leaves can withstand the forces coming to it and
preserve its integrity like the original one. The following statements can be concluded from this
project:
• Within the scope of this project, a specified calculation flow is followed in order to reach
an accurate comparison between the original and new designed leaf spring.
• After the stiffness and stress results under vertical loading of the leaf spring is calculated
with in-house software (BLAFES) based on mathematical calculations, elastokinematic
analysis is performed to compare the stiffness values and to prepare a base model for
MBS.
• Two separate MBS models are prepared both for the original and new designed leaf
spring. The internal forces and moments on the leaf springs are calculated based on the
wheel force measurements from a test vehicle at the torture track.
• FEM calculations with Abaqus show that, compared to the original design the maximum
stresses on the new designed leaf spring decreases up to 20 %.

• Fatigue life calculations with FEMFAT show on the other hand similar damage results on
both original and new designed leaf springs, which means, comparable fatigue life values
are to be expected.
• The results show that with the new designed leaf spring with reduced number of leaves
the aimed weight and cost reduction can be achieved without worsening the durability
characteristics.
• For a comprehensive durability assessment of leaf springs which work under highly
variable loading conditions, a multi-channel fatigue life calculation should be performed.
4. References
1. BLAFES, Version 1.6, User’s Manual, Jurgen Fischer - Technische Beratung, 2013
2. Medina, Version 8.3.2, User’s Manual, T-Systems, Germany, 2013
3. SIMPACK, Version 8.9, User’s Manual, Dassault Systémes Simulia Corp., Providence, RI.
4. Abaqus, Version 6.14, User’s Manual, Dassault Systémes Simulia Corp., Providence, RI.
5. FEMFAT, Version 5.1, User’s Manual, Engineering Center Steyr, Austria, 2014
6. Atamer, S., Barenbrock, D., Ageorges, C, Sonsino, C.M., “Consideration of the Resonance
Effects in the Fatigue Analysis of the Body in White” ATZ Worldwide 110(9):26-30, 2008,
doi: 10.1007/BF03225205
7. Atamer, S., “Schädigungsberechnung zur Lebensdauerprognose von Rohbaukarosserien unter
Berücksichtigung von dynamischen Effekten“ Ph.D. thesis, Mechanical Engineering
Department, Technical University Darmstadt, Germany, ISBN 978-3-8440-0991-0, 2012
8. Haibach, E., “Betriebsfestigkeit: Verfahren und Daten zur Bauteilberechnung“, 3. Edition
Springer Verlag Berlin Heidelberg New York, 2006
9. Sonsino, C.M., “Principles of Variable Amplitude Fatigue Design and Testing”, In: Fatigue
Testing and Analysis under Variable Amplitude Loading Conditions, ASTM STP 1439, 3-23,
2005, doi: 10.1520/STP11294S
10. Sonsino, C.M., ”Fatigue Testing Under Variable Loading”, Int. Journal of Fatigue 29, 1080-
1089, 2007
11. Bakir, M., Siktas, M., and Atamer, S., "Comprehensive Durability Assessment of Leaf
Springs with CAE Methods," SAE Technical Paper 2014-01-2297, 2014, doi: 10.4271/2014-
01-2297.

Coupled Analysis of Fracture Mechanics and
Piezoelectricity in Active Layers in the Abaqus Code
Operated through the Isight Tool
Eugenio Brusa1, Mehdi Mohammadzadeh Sari1, Cosima Fiaschi2, Mauro Parodi2
1
Dept. Mechanical and Aerospace Engineering, Politecnico di Torino
Corso Duca degli Abruzzi, 24 – 10129 Torino, ITALY
2
Exemplar srl
C.so Vittorio Emanuele II, 161 – 10139 Torino – Italy
Abstract: This paper describes a new computational approach aimed at investigating the crack
propagation inside smart structures equipped with surface bonded piezoelectric layers, when the
electromechanical coupling due to piezoelectric phenomenon is exploited. The Abaqus code is
used to perform a prediction of both the fracture mechanics and the coupled response of the
structure, through a suitable connection between the two solution environments, being provided
by the Isight tool. A preliminary analysis is shown and some significant results are proposed.
Keywords: Composites, Coupled Analysis, Crack Propagation, Dynamics, Fracture, Piezoelectric

Structures, Smart Materials.
1. Introduction
Piezoelectric materials are exploited in several smart structures for actuation, sensing, energy
harvesting and health monitoring purposes [1]. An increasing demand of industry concerns a
reliable prediction of failure mechanism of piezoceramic layers in case of fracture, fatigue and
creep. In some application like actuation, vibration energy harvesting and health monitoring this
task looks extremely relevant for the design activity, because of loading conditions applied and of
functions provided by the smart material. A key issue is currently the analysis of crack
propagation through the piezoelectric layer while its active behavior is exploited [2]. Analytical
methods already proposed in the literature may be ineffective when geometry of structure is rather
complicate, or structural behavior is nonlinear, like in case of a large vibration amplitude [3].
Therefore a reliable computational approach to cope with those cases is a goal of the current
research activity performed in this field.
Basically crack propagation has to be investigated in association with the piezoelectric
phenomenon either in condition of direct or indirect effect, i.e. when mechanical strain induces an
electric charge distribution within the material or electric field applies a mechanical strain through
the electromechanical coupling. The Finite Element Method (in the following simply FEM) is
highly recommended as a recognized approach also by some technical standards where ‘design by
rules’ is substituted by a ‘design by analysis’, during an advanced step of design operation.
Commercial codes provide nowadays the numerical tools for a straight analysis of the

electromechanical coupling between strain and voltage in piezoelectric continuous media, as well
as crack behavior is described by calculating the so–called ‘Stress Intensity Factor’ (SIF) within
the region around the tip and the ‘J–integral’ is used to calculate the energy release rate of crack
[4]. Those calculations allow predicting even the crack path. In case of a coupled system the
fracture mechanics has to be associated to local effects induced by the piezoelectric phenomenon.
The Abaqus code is daily used to investigate both the fracture mechanics [5] and the piezoelectric
effect, but these analyses are performed separately. In present case numerical solutions of the two
above mentioned problems were connected and coupled by means of the ISIGHT tool. This
approach allowed investigating in details the reaction applied by the electromechanical coupling to
the crack propagation at the tip. A complete procedure was defined and a numerical tool was built
up as they are herein described together with some preliminary numerical results.
2. Coupled behavior of smart piezoelectric structures

To clarify the goal of the proposed analysis it can be briefly summarized herein the context of
applications and the cases to be investigated. In active vibration control some piezoelectric layers
are surface bonded on a main structure or embedded inside. They are either used as an actuator to
apply a control action, by exploiting the imposed electric field, or as a sensor, by resorting to
direct piezoelectric effect, which allows monitoring the mechanical strain of the structure. In both
roles very often actions are not so large to induce a severe failure of piezoelectric layers, in regular
operation. In fact, in case of structural health monitoring (SHM) goal is detecting a damage
occurring inside the structure and piezoelectric layers might be exposed to a failure, if crack
reaches the interface between smart material and main structure. Moreover, vibration energy
harvesting (VEH) is aimed at recovering some amount of energy being usually dissipated by
resorting to a suitable conversion into electric charge. It can be then used as an autonomous power
supply to operate some sensor. In this case very often vibration amplitude is sufficiently large to
increase the risk of crack nucleation either on the structure or directly on the piezoelectric, thus
making prediction of its propagation critical for the reliability of the whole system. These
applications of piezoelectric transducers were fairly recently proposed and they made the issue of
describing the crack behavior in presence of electromechanical coupling a relevant topic of
mechanics of smart materials. This goal is even more strategic at microscale, as in micro–electro–
mechanical systems (MEMS).
According to the above mentioned issues, crack propagation is currently analyzed in:
• piezoelectric structures, without any other structural support, in case of both the roles of
actuator and sensor, even in self–sensing configuration;
• damaged structures with surface bonded sensors for SHM, to detect an early transition of
crack from the structure to the piezoelectric layer;
• in VEH, to analyze the crack propagation rate and its path through the composite structure.
A suitable computational process was therefore assessed to cover all the above mentioned cases.

3. Phenomenology of crack propagation in piezoelectric layers
It is known that crack propagation is effect of mechanical load applied to material surrounding the
crack tip, being the rate of propagation depending on the energy balance and on the opening mode
in fracture (I,II,III). Path strictly depends on load, constraints, material and mode. State of stress
and crack length are both relevant for propagation, therefore in the literature the ‘Stress Intensity
Factor’ (SIF), namely K, was defined to consider both those contributions to predict the
propagation and in particular its stability, in comparison to the material toughness, for a given
fracture mode, being referred to as Kc. Rate of propagation is finite until that SIF is lower than
toughness, but as soon as it overcomes this value, crack proceeds very fast, in practice unstably
[4].
σyy
σxx
τxy τrθ σ
y rr
r σθθ
Crack θ
Figure 1. Description of stress components around the crack tip.

Basically, for simple geometries, SIF is defined by some analytical approach in a local reference
by using polar coordinates as the distance from the crack tip, r, and its angular location, θ, with
respect to the crack line and to the opening mode, e.g. SIF (K) and stress components (σij) are
related as follows :
𝐾𝐼
𝜎𝑖𝑗 = 𝑓(𝜃) (1)
√2𝜋𝑟
From the computational point of view crack tip is corresponding to a singularity, i.e. to a very
sharp notch, which needs to be analyzed by a suitable numerical approach, being able to follow the
opening mode and to recognize the non–holonomic (variable with both the configuration and time)
compatibility of boundary conditions associated to the displacements inside the material. Energy
released rate of crack is usually computed by resorting to the J–integral (JI). It describes the
energy of a region around the crack tip limited by a selected contour as C1 and C2 in Fig.2. This
integral is applied to the total length of contour C, it is defined by a local coordinate s and by a
orthogonal versor n, and includes the density of elastic energy associated to strain, We, the stress
applied to the contour, σij, and the corresponding mechanical displacement, u.

ds 𝝏𝝏𝝏𝝏𝒊𝒊
C1 ni 𝑱𝑱 = � 𝑾𝑾𝒆𝒆𝒆𝒆 𝒅𝒅𝒅𝒅 − σ
𝑪𝑪
𝑻𝑻ij𝒊𝒊𝒊𝒊 𝒅𝒅𝒅𝒅
𝒅𝒅𝒅𝒅 𝒄𝒄
Mechanical System
Crack tip 𝝏𝝏𝝏𝝏𝒊𝒊

σij𝒊𝒊𝒊𝒊
𝑱𝑱 = � 𝑾𝑾𝒆𝒆𝒆𝒆 𝒅𝒅𝒅𝒅 + �𝑫𝑫𝒊𝒊 𝑬𝑬𝒊𝒊 ∙ 𝒏𝒏𝒊𝒊 −𝑻𝑻 𝒅𝒅𝒅𝒅 �
𝝏𝝏𝒙𝒙 𝒄𝒄
𝑪𝑪
C2
ElectroMechanical System
Figure 2. Definition of J-integral applied to fracture mechanics of a mechanical

system and to piezoceramic material.
Crack propagation can be investigated by resorting to the well known relation between J–I,
referred to as J, and the energy release rate, G [4]:
∂W
𝐽=G=− (2)
∂a
where W describes the total energy associated to the crack system, while a is the crack length.
Some authors ([2],[3]) investigated the role of piezoelectric effect upon the crack propagation
through analytical approaches, by including into the J–I the electromechanical energy associated
to the electric field (Fig.2). Electromechanical coupling in piezoelectric material is described by
the coupled constitutive laws of material as follows:
𝑇𝑖𝑗 = 𝐶𝑖𝑗𝑘𝑙 (𝑆𝑘𝑙 − 𝑆 ′ 𝑘𝑙 ) − 𝑒𝑘𝑖𝑗 𝐸𝑘

(3)
𝐷𝑖 = 𝑒𝑖𝑗𝑘 �𝑆𝑗𝑘 − 𝑆 ′𝑗𝑘 � + 𝜖𝑖𝑗 𝐸𝑗 + 𝑃𝑖
According to the Standards of Piezoelectricity [1], symbols are defined. Mechanical stress
components are Tij (instead of σ to avoid any misunderstanding with surface electric charge) and
strains are Skl, being S′kl the residual strain induced by polarization Pi. Cijkl are the elastic
coefficients, being ekij the piezoelectric coefficients. Ek is the electric field and Di is the electric
displacement along a given direction i, while ∈ij is the dielectric permittivity of material. The main
problem in case of industrial applications is that geometry sometimes does not allow an easy
prediction of coupled behavior in presence of crack. Moreover, in many commercial Finite
Element codes the two analyses, dealing with fracture mechanics and piezoelectric effect, are
usually run separately in different tasks of numerical solution.
Behavior of material around the tip is strictly important to define the propagation rate, the path and
the stress distribution. It is known that in ductile material plastic behavior at tip may decrease the
propagation rate and affects the stress distribution, since a local yielding effect occurs. A sort of
stop–and–go transition occurs in propagation, being due to material hardening around the tip, if
opening effect is sufficiently large. In case of active piezoelectric layer electric charge distribution
changes with opening, because it is coupled with the mechanical strains and new free surfaces are

generated. This effect may introduce an electromechanical attraction between the two crack lips,
being depending on the electric nature of crack. It can be permeable, impermeable or semi-
permeable to the electric charge. Around the tip stress concentration may even affect the electric
field and can induce some local electromechanical actions, which superpose to the mechanical
behavior. All those phenomena may have a large influence upon the crack propagation, although
they depend on the direction of piezoelectric polarization and of crack, respectively.
4. Numerical method
A first activity was herein performed to investigate the case of Linear Elastic Fracture Mechanics
(LEFM) of coupled metallic and piezoceramic specimens. According to the literature this is a
main task to be considered, although under certain operating conditions metallic structure might
undergo yielding, being herein neglected. The literature proposed some numerical approaches to
investigate the fracture behavior as Finite Elements, Boundary Elements, Meshless methods and
the so–called Extended Finite Element Method (XFEM), being available in the Abaqus code. The
XFEM extends the classical Finite Element Method (FEM) approach, because it allows analyzing
discontinuities in the displacement field, like those due to the crack opening. Therefore the main
goal of numerical simulation was coupling the analysis of crack propagation, being based on the
XFEM, which is associated to the computation of SIF, JI and crack path, to the prediction of the
electric charge distribution, voltage and related stress occurring within the piezoelectric material.
A sequential approach was implemented, i.e. electromechanical coupling was investigated by
applying the mechanical load and predicting a preliminary crack propagation, then computing the
displacements around the tip, thus starting an analysis of piezoelectric phenomenon to calculate
the stress induced around the tip. Mechanical loading conditions were then updated and procedure
was iterated up to the occurring of an unstable propagation, when it could be reached.
4.1 Use of the Abaqus code
The Abaqus code is used to perform the prediction of crack propagation and to investigate the
piezoelectric effect. It is applied first to model the geometry of the structure and to define and
locate the initial crack. Properties of materials are inputted, but features are different for metals
and piezoceramics. Elastic properties and mechanical strength are provided to the code in both
cases, but piezoelectric material needs that orientation is assigned, together with piezoelectric
coefficients. Volumes of crack and structure, respectively, are merged as separated parts inside the
model, then meshed, as soon as element type, shape and mesh technique are selected. A dynamic
implicit solution is used to predict the crack opening, as a nonlinear geometric solution whose
time steps time are kept sufficiently small to assure the numerical convergence, within a given
maximum number of increments. Crack is created by Abaqus code through the XFEM approach,
to predict the propagation for a given loading condition and known constraints. Contour integral
option is used to calculate JI and SIF. Results consist for the fracture mechanics module of crack
growth maps, crack angles and crack propagation rate. A dedicated module of the code deals with
the electromechanical behavior of piezoelectric layers. Finite elements with piezoelectric
properties are associated to numerical solutions which include both the stress and strain analysis
and the distribution of voltage and electric charge, respectively, for given boundary conditions. In
particular, constraints in this case include mechanical inhibition of some degree of freedom
(displacement and rotation) and similar boundary conditions applied to voltage and charge.

Loading conditions consist of static and more often dynamic mechanical actions and of applied
electric fields. A complete electromechanical coupling is therefore created. This allows converting
the mechanical energy into an electric storage and electric energy into a mechanical work.
4.2 Use of the Isight tool
As it was previously stated the Abaqus software is unable to perform a prediction of the fracture
mechanics and of an active piezoelectric material by taking into account its electromechanical
coupling. A suitable procedure to create a link between the fracture mechanics and the
piezoelectric analysis could be tested by resorting to an integrated model, being developed into the
Isight environment. This tool is currently used to combine cross–domain models to provide a
unique simulation process flow. Execution is automated, since the Isight tool can manipulate and
transfer numerical data between two process steps and run multiple simulations. In this case it
allowed introducing all the required piezoelectric properties inside the process flow performed to
predict the calculation of JI and SIF and the crack propagation when piezoelectric layers react to
the mechanical loading because of the electromechanical coupling.
5. Development of the simulation flow

Structure is first modeled inside the Abaqus code, by providing all the mechanical properties of
material and relevant information about crack (load, boundary conditions, geometry, initial
length), then it is meshed. To input the electrical properties of the piezoelectric layer the same
geometry is modeled as a separated case, through some piezoelectric elements. Prediction of
fracture in presence of piezoelectric phenomenon is performed by running the two analyses as a
sequence, thus making the two above mentioned models interacting each other, inside the Isight
tool. As Fig.2 shows the tool developed computes the SIF and the JI by resorting to a
concatenation of computational loops.
5.1 Use case 1: piezoelectric operated as a sensor or as a vibration energy
harvester
When piezoelectric layer just behaves like a sensor, i.e. no external voltage is applied, model of
the cracked structure described into the Abaqus code is transferred into the Isight tool in
subroutine ‘Crack’ which computes SIF and JI, as it does in case of a metallic structure. Results
are then collected by ‘Post–Crack’ module and used in ‘Piezo’ section, together with the second
model including the piezoelectric elements. Displacements induced by the mechanical actions
applied to the piezoelectric layer and the voltage distribution is computed by through the typical
solution performed in case of dynamic behavior of a piezoelectric material. It is appreciated that
the new configuration of voltage actually induces a local piezoelectric effect upon material, which
affects the boundary conditions of load around the crack tip. Therefore displacements just
computed are written (‘Write’) and used (‘SubMod’) to refine both the SIF and JI previously
calculated, which now include the electromechanical coupling. Subroutine ‘SIF_F’ is run to show
the main numerical results. This procedure is applied for each load step, by following an iterative
solution. Convergence is assured step by step.

5.2 Use case 2: piezoelectric operated as an actuator
When strain inside the piezoelectric layer is induced by the electric field applied by a control
system, i.e. piezoelectric layer behaves as an actuator; option ‘voltage driven’ is followed. Task
‘Model’ computes the loads applied to the structure as a consequence of the electric field
excitation, and both the preliminary SIF and JI are found as in option ‘Force driven’. Then
correction due to the capability of sensing still present in the layer is calculated by ‘SubModV’
and corrected SIF and JI are shown through the ‘SIF_V’ subroutine.
Fracture analysis without Piezoelectric effect Local analysis of

piezoelectric effect without Fracture piezoelectric
analysis effect and
correction of
Main
SIF and J-Integral
loop
FORCE Inner
UPDATE loop
Force
GEOMETRY
driven Crack Post-Crack Piezo PostPiezo Write SubMod SIF_F
UPDATE
Voltage
driven Model PiezoV PostPiezo WriteV SubModV SIF_V
Figure 3. Toolbox developed to perform the prediction of fracture mechanics with

piezoelectric effect.
5.3 Prediction of crack propagation in case of piezoelectric material

In the composite structure it was basically assumed that crack first occurs on the main metallic
layer, then through the piezoelectric one, if conditions are compatible with propagation. Prediction
of crack propagation requires a different implementation of the Isight tool above described. In this
case propagation is investigated in terms of direction and speed. To provide those results the
piezoelectric phenomenon is considered both in terms of applied strain when crack opening is
driven by voltage and in terms of change of shape of piezoelectric layers. As Fig.4 describes the
crack propagation model developed inside the Isight code includes some components. Properties
of piezoelectric material are introduced by the module “Create Piezo Model”, voltage is
consequently applied and resulting displacements are recorded by running module “Run Piezo”.
Geometry of structure is changed by the actuation voltage. A shrinking effect occurs in the layer,
which affects the geometry of the cracked structure. This change is predicted in “Grab Ux” and
inputted into the XFEM module. As in case of metal the XFEM provides the computation of crack

path and propagation, once that actual geometry, shape and loading conditions are known. Crack
propagation and geometry of structure are step by step recorded by “Status” module. The
numerical loop is closed so as for increasing voltage the crack propagation is computed by an
iterative solution at each step of voltage gradually imposed, until that final rupture occurs.
MAIN
TASK
Create Loop
Piezo
Model
Run Grab UX Run XFEM Status

Piezo
Figure 4. Isight model for the analysis of crack propagation inside the piezoelectric
material
6. Numerical simulation and tests

The simulation flow above described was preliminarily validated on a case of literature to check
its consistency [6]. Once that this action was performed, the numerical tool was used to investigate
the main cases described in above Section 2. Two main cases of fracture modes were analyzed as
are shown in Fig.5. A typical opening mode I of fracture was excited in specimen 1, since crack is
perfectly located at the middle span of the composite beam and load is aligned with its length.
Specimen 2 is suitable to investigate a mixed mode operation in which modes I and II are
superposed. Dimensions were suggested directly by [4]. In a first set of simulations material was
completely assumed to be piezoelectric PZT (d31=−150 pC/N, E=115 GPa; v=0.34; ρ=7.8 kg/m3),
while in a second one a composite with steel (E=206 GPa; v=0.3; ρ=7.89 kg/m3) was tested.
■ SIF and J–Integral: Results show that in case of Mode I (1 in Fig.5) and single piezoelectric
layer, crack propagates directly inside the dielectric material. The SIF grows up with both the
applied mechanical load and crack length, as in steel, when electric field is weak, being induced
only by crack opening and structure bending. J-integral grows nonlinearly with the SIF value. In
fact, when load is generated directly by the electric field applied to the electrodes of piezoelectric
layer, which are the upper and lower surfaces of the structure, the SIF increases with applied
voltage, but only up to almost one half the total thickness of the structure (Fig.6). When crack
length is longer than this value, it gradually decreases. This result is confirmed by the values of J-
integral which slightly grows up with the SIF when crack is fairly short, then decreases. Moreover,
against any intuitive interpretation J-integral is negative. Somehow energy exploited by the

piezoelectric effect to apply strain to the material is dominant on the release of energy coming
from the crack opening. This result can be motivated by expression of J-integral for the
electromechanical coupled system described in Fig.2. Those trends were confirmed in mixed mode
analysis, although sign of J-integral may depend on the dominant fracture mode in that case. In
case of composite specimen with surface bonded piezoelectric layer those results were even more
evident. J-integral and the SIF decrease with crack length after reaching a maximum peak, if crack
is started from the steel and propagates across the metallic portion of the smart system first.
P 20 P
PZT PZT
40
40
Piezo
20
20
STEEL STEEL
80 80 20 120
Steel
180 180
1 2
Figure 5. Test cases analyzed in numerical simulations (units are mm)
Some typical effects occurring in operation of piezoelectric structures were detected by numerical
simulations. They are herein described by remarking only some peculiar behaviors.
Min length Higher voltage F V

60
- 0,01
Automatic trend detection FORCE DRIVEN VOLTAGE DRIVEN
J - integral [N/mm]
SIF [MPa√mm]
SINGLE MODE
- 0,03 Max length 40
- 0,05 20
Sensitivity to crack Sensitivity to voltage MIXED MODE
length variation
Lower voltage
- 0,07
10 30 50 70 90 10 15 20 25 30
Voltage [V] Length [mm]
Figure 6. Relevant numerical results in voltage driven fracture of piezoelectric layer
■ Crack propagation and rupture: Simulations performed show that mechanical load induces a
crack propagation which might grow up to the rupture of material. By converse when voltage acts
as driving load, crack propagation stops when a certain critical length is reached. To start again the
propagation a larger electric field is required than that it was applied as crack was stopped. A sort
of ‘barrier effect’ is due to the boundary conditions imposed by the piezoelectric effect at crack
tip. In case of a mixed mode (Fig.6), if mechanical load is applied, crack changes direction, but
deviation is less evident when load is driven by voltage. In composite structures with a metallic
substrate equipped with a surface bounded piezoelectric layer the crack propagates along a straight
line under both mechanical force and voltage excitation in single mode, but in voltage driven
condition propagation stops when crack tip reaches the interface between materials. This effect is
more evident in mixed mode, because propagation changes its direction at the interface under
mechanical action and breaks both materials, while in electromechanical actuation crack never
breaks the piezoelectric layer, being stiffened by the applied voltage.

7. Conclusion
Crack propagation is fairly different across the steel and piezoceramic material, respectively,
because of their properties. In passive configuration a crack propagation through the steel easily
breaks the piezoceramic, if only a mechanical force is applied, speed of propagation increases and
might make unstable the fracture. By converse when bending is driven by voltage a sort of barrier
effect is opposed by the piezoelectric layer because of a superposition of phenomena. Local
distribution of stress around the tip is greatly affected by the piezoelectric coupling, which can
reduce the capability of material to allow cracking. In some cases stretching of piezoelectric layer
increases its stiffness and apparent toughness, by reducing the crack propagation. Proximity of
crack tip to free surface with concentrated electric charges seems to be favorable to reduce the
crack propagation. A relevant result of this work is that both the electromechanical coupling effect
and the crack direction in propagation are detected through the proposed procedure developed by
resorting to the Isight program. Nevertheless, experimental validation is still to be performed. Due
to high expenses related to the design and construction of the piezoelectric specimens it is worthy
noticing that the numerical tool developed allows a preliminary oriented design of experiments.
8. References
[1] Moheimani, S., Fleming, A., Piezoelectric transducers for vibration control and damping,
Springer, 2006
[2] McMeeking, R., “Crack tip energy release rate for a piezoelectric compact tension specimen”,
Eng. Fract. Mech, 64 (1999), pp.217–244
[3] Viola, E., Boldrini, C., Tornabene, F., “Non-singular term effect on the fracture quantities of a
crack in a piezoelectric medium”, Eng. Fracture Mechanics, 75 (2008), pp.4542–4567
[4] Recho, N., Fracture mechanics and crack growth, John Wiley, 2012
[5] Venkatachalam, G. et al., “Determination of J-integral and stress intensity factor using the
commercial FE software ABAQUS in austenitic stainless steel (AISI 304) plates”, Int. J. Adv.
Man. Technology, 1-4 (2008)
[6] Mohammadzadeh Sari, M., “Numerical tools for fracture mechanics prediction of passively
and actively coupled structural systems”, Ph.D. Thesis, Politecnico di Torino, Italy, 2014.

XFEM Simulation of Radial-Median Crack Evolution
in Knoop Indentation of Brittle Materials
Felix Rickhey*, Jin Haeng Lee**, and Hyungyil Lee*
* Dept. of Mechanical Engineering, Sogang University, Seoul, S. Korea
** Division for Research Reactor, Korea Atomic Energy Research Institute, Daejeon, S. Korea
Abstract: In sharp indentation of brittle materials, a complex three-dimensional stress field

evolves in the material beneath the indenter and may cause cracking. During unloading, a tensile
residual stress field drives the crack to its final semi-elliptical shape. The so-induced crack can be
used for e.g. crack growth experiments, but for which the initial crack size and shape have to be
known. For the analysis of the crack evolution in a material subject to a complex stress field,
which changes with position, time and crack advancement, the extended finite element method
(XFEM) represents a useful tool. We analyze the formation of the crack occurring during Knoop
indentation of brittle materials. Issues associated with XFEM in Abaqus/Standard are discussed
for the problem at hand, and suggestions are made with regard to meshing, the artificially
attributed viscosity parameter, and the damage-initiating stress threshold. Finally we discuss the
influence of material properties on the fracture toughness evaluation. Results can be used for
predicting fracture toughness and final crack shape.
Keywords: Knoop Indentation; Brittle Materials; Fracture Toughness; XFEM
1. Introduction
The indentation test has become a popular method for determining the fracture toughness of brittle
materials. Its main advantage over other standard methods is that it can be more or less directly
applied to structures with a flat surface; specimen preparation is thereby reduced to a minimum,
and pre-cracks are not required.
In Knoop indentation of brittle materials, a median crack is initiated during loading (high loads) or
during unloading (lower load range). While the crack grows due to the increasing applied load
during loading, during unloading it is driven by a growing residual stress field caused by the
increasing mismatch between the plastic zone and the outer elastic zone, which is trying to reduce
in size. For fracture mechanics analyses, the Knoop indentation crack has usually been treated as a
2D elliptical crack subject to a point load at the center of the indentation (Keer 1986). The
probably most cited work on Knoop indentation cracking was conducted by Marshall (1983). His
analysis was based on the well-known equation by Lawn, Marshall and Evans (Lawn 1980),
which relates mode I fracture toughness Kc to maximum indentation load Pmax and surface crack
length c (Figure 1) measured from the loading axis in outward radial direction
Pmax
Kχ = χ . (1)
χ3/2

The constant χ can be understood as the relative strength of the residual field that drives the crack
and was reported to be a function of the ratio of Young’s modulus E to hardness H. Subsequent
studies yet revealed that Equation (1) may be inaccurate owing to an oversimplification of the
stress field (e.g. Ponton 1989a,b; Quin 2007; Lee 2012; Hyun 2014).
Knoop indentation differs from Vickers or Berkovich indentation insofar as (1) only one crack
forms (hence no interaction of radial-median cracks); (2) the plastic zone has approximately the
shape of a prolate spheroid, whose major axis is nearly as long as the major impression diagonal
2a. Based on experiments with Si3N4, Marshall (1983) found that the point-load assumption
inherent in Equation (1) holds also for Knoop indentation, provided a sufficiently high load. The
threshold for Equation (1) to be applicable was defined by a crack length-to-impression half-
diagonal ratio c/a ≥ ~ 1.3 (which however applies only to the tested material). Knoop indentation
cracks are much larger than Vickers indentation cracks at equal Pmax, as Petrovic (1983) concluded
in a similar study from Knoop indentation tests on glass-ceramic (C9606). Keer (1986) attempted
at a 3D fracture analysis taking into account the large aspect ratio of the Knoop indenter and the
resulting elongated plastic zone. Lube (2001) conducted Vickers and Knoop indentation tests on a
gas pressure sintered silicon nitride and analyzed the shape of resulting cracks and how it is
influenced by Pmax.
The single sharp and rather short crack can be advantageous because it is more amenable to the
analyses of fracture, surface flaws or residual stresses (Ahn 1996). Further, crack growth
experiments can be conducted on Knoop crack-induced specimens to determine the material’s R-
curve after removal of a thin surface layer containing the core zone. However, in order to use
Knoop indentation cracks as starter cracks, we need to be able to predict the size and shape of the
induced crack as a function of material properties. In this study, we first present an XFEM-based
model that allows the simulation of Knoop indentation cracking and discuss critical issues
associated with cohesive zone-based XFEM. We then study the crack evolution for a Si-like
reference material and discuss the influence of friction and material properties on parameter χ.
The results are to serve to better understand the crack evolution, and to explain phenomena
occurring in Knoop indentation cracking tests.
top view (x-y plane)
Knoop indenter
c
a
Knoop indenter
plastic zone
cz
crack (after
unloading)
section view (x-z plane)
Figure 1. Semi-elliptical configuration of plastic zone and Knoop indentation crack.

2. XFE model
The XFE model for Knoop indentation cracking is shown in Figure 2. Considering geometric and
loading symmetries, modeling one half of the specimen is sufficient. Symmetry constraints with
respect to the two indenter diagonal planes are imposed. The bottom nodes are fixed. The model
features a relatively fine mesh in the inner region, where all damage is expected to occur, and an
outer region, the mesh of which gradually coarsens in outward radial direction. As diamond or
WC indenters, whose Young’s moduli are much higher than that of the material to be indented, are
usually employed in indentation cracking tests, for simplicity and to keep computational costs low
the indenter is assumed rigid. In the contact between indenter and material, the node-to-surface
contact pair option (Abaqus 2013) gives a more realistic and accurate contact distribution and is
therefore preferred. Coulomb friction is considered by µ = 0.2, which is representative of the
contact of diamond with various types of materials. The whole FE model consists of ≈ 2×105 8-
node brick elements (C3D8).
y
θ
x
Knoop
<<
indenter
<<
symmetry
repetitive sector planes
(not modeled)
Figure 2. Half model for Knoop indentation cracking.

2.1 Mesh
In XFEM, it is generally problematic when a crack propagates along an element edge because this
may lead to instability (Mohammadi 2008). As major damage is supposed to occur in the planes
normal to the surface and through the indenter diagonals – indenter edges act as stress raisers –,
the mesh is structured so that no nodes are placed in the indenter diagonal planes. In connection
with XFEM, Abaqus allows the use of 8 node brick elements only. Although two nodes can be
placed at one position, damage cannot be initiated between these two nodes. A slender cylindrical
segment (with the cylinder axis aligned with the z-axis) of radius rc = 0.02e is therefore removed
from the center of the model. The innermost nodes, i.e. nodes with radial distance rc from the
loading axis, are constrained in circumferential (θ ) direction. FE results (enrichment deactivated)
obtained using two models – one with rc = 0, one with rc = 0.02e –yield a negligible difference in
the load-depth curves, and maximum principal stresses are observed in the same element at the
same depth. Further, variation of rc in the range rc /e = [0.005, 0.05] does not affect crack shape
and size.

Using only one enriched zone (EZ) for the inner region we observed that damage propagation
ceases during the analysis and that damage does not initiate in other regions although stresses are
well beyond the damage initiating threshold. However, by dividing the inner region into several
EZ, which are arranged such that elements of one and the same EZ do not share nodes (Figure 3),
we observe smooth damage propagation and even merging. Eight EZ are necessary here, four for a
median element section (EZ 1-4) and four for the neighboring median section (EZ 5-8), and then
again EZ 1-4 (not displayed in Figure 3).
Knoop indenter EZ 1 EZ 5
EZ 2 EZ 6
EZ 3 EZ 7
EZ 4 EZ 8
θ = 0°-plane
Figure 3. Arrangement of enriched zones (EZ) in the inner region.

2.2 Material model
The material is assumed to be isotropic and exhibit elastic-perfectly plastic behavior. Indeed, the
compressive behavior of many brittle materials can be accurately described by the elastic-perfectly
plastic material model (Francois 1988). Yielding occurs according to the associated flow rule
using the Mises potential, and plastic deformation is volume-conserving. Mode I fracture
toughness Kc remains constant (flat R curve), as is the case for an ideally brittle material
(Anderson 2004). A reference material is defined with E = 200 GPa, ν = 0.3 and σy = 5.0 GPa and
fracture energy Γ = 3.0 MPa µm, values that are close to Si and that were also used by Lee (2012).
Since all materials used in this study have a high hardness-to-Young’s modulus ratio (H/E),
fracture is assumed to occur in the framework of linear elastic fracture mechanics (LEFM).
Damage initiation is governed by the maximum normal stress (σ ) criterion which states that
damage initiates when the condition
σ = σˆ (2)
is met at the centroid of an element; the resulting damage plane is oriented normal to σ (Abaqus,
2013). σˆ is the damage-initiating stress threshold and 〈…〉 denotes the Macaulay brackets. To
prevent potential lateral cracking, which is often observed in indentation, only stresses normal to
median planes are considered damage-causing. It is to be noted that damage initiation is mesh-
dependent, and the exact time of initiation cannot be determined accurately. However, the eventual
goal here is to find the final crack shape, which is hardly influenced by the time of initiation.
Damage propagation occurs according to a traction-separation law, which governs the stiffness
degradation – represented by damage variable D – associated with an incremental increase in
crack opening displacement (Figure 4). Hutchinson (2000) showed that the shape of the curve is of

secondary importance, so that we use the simplest, bilinear, form. The fracture energy is equal to
the area below the traction-separation curve
1
Γ = σˆ δ c , (3)
2
where δc is the critical displacement, beyond which the stiffness is degraded to zero. The damage
zone is defined by the region where δˆ < δ < δc. δˆ is the displacement corresponding to σˆ .
Assuming plane strain conditions and linear softening, its length LDZ is (Rice 1968; Tomar 2004)
π ΓΕ ′
LDZ ≈ , (4)
8 σˆ 2
where E' = E / (1–ν2) for plane strain, and ν is Poisson’s ratio. Fracture energy Γ is related to Kc by
Γ = Kc / E'. Since material softening may lead to severe convergence problems, a small viscosity ζ
is introduced to the material model (Gao 2004). The meaning and significance of the parameters σˆ
, ζ and LDZ will be discussed below.
Figure 4. Traction-separation (σ - δ ) law constituting damage.
3. FE results for Knoop indentation cracking

3.1 General aspects of Knoop indentation crack evolution
Indentation is performed on the reference material (Γ = 3.0 MPa µm, σˆ = 0.8 GPa) applying Pmax =
480 mN. Cracking occurs exclusively in the long diagonal plane (θ = 0°); only a small damage at
the surface (but no crack) is observed in the θ = 90° median plane (Figure 5). The crack initiates
on the z-axis as a subsurface crack, grows in outward radial direction during the loading cycle
without penetrating the surface. Shortly after load reversal, the crack penetrates the surface.
Varying hmax = [0.4, 1.0] µm (Γ = 3.0 MPa µm) and Γ = [1.5, 7.0] MPa µm (hmax = 0.8 µm), we
find that for hmax ≤ 0.4 µm, the crack does not open up to the surface; it seems that in order for the
crack to open up its length inside the material at load reversal must be > ~ a (or the length of the
plastic zone). For hmax ≥ 0.8 µm or Γ ≤ 4.0 MPa µm, Pmax/c3/2 is constant above a threshold of c/a ≈

1.2 (Figure 6). This supports the experimental observation that the point-load assumption can be
used provided a sufficiently high Pmax. Marshall (1983) found from experiments with Si3N4 that
Equation (1) is valid for c/a ≥ ~ 1.3 (but where a was measured after unloading). However,
experiments reveal that in general the c/a threshold increases with E/H and Kc. If c/a is below the
threshold, application of Equation (1) results in an underestimation of Kc. In contrast to Vickers
indentation, once a surface crack occurs, we can be sure that the crack is of half-penny shape.
small damage in 2
StatusXFEM θ = 90° plane P = 360-240-0 mN
0
1.0
0.8 -2
0.6 60
-4
0.4 150
-6
z [m]
0.2 240
0.0 -8
360
-10 480 mN
-12
E = 200 GPa,  = 0.3
-14
y = 5.0 GPa,  = 3 MPa m
-16
0 2 4 6 8 10 12 14 16 18
r [m]
Figure 5. Crack evolution applying Pmax = 480 mN: loading cycle (red), unloading
cycle (blue) (left); final crack in Abaqus/Viewer (right).
0.14 1.6 0.14 1.6

E = 200 GPa, = 0.3, y = 5 GPa
0.12 1.5 0.12 1.5
K
K = Kc / (Pmax / c3/2 )
K = Kc / (Pmax / c3/2 )
0.10 K 0.10
1.4 1.4
0.08 0.08
c/a
c/a
1.3 1.3
0.06 c/a 0.06
1.2 c/a 1.2
0.04 0.04
196 265 346 434 528 1.1 1.1
0.02 0.02
Pmax [mN]
0.00 1.0 0.00 1.0
0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 0 2 4 6 8
hmax [m]  [MPa m]
Figure 6. Change of χK with maximum indentation depth hmax and fracture energy Γ.
Having obtained HK = Pmax/Ac = 11.2 GPa (Ac is contact area at Pmax) from a separate FE analysis
with a refined mesh (hmax/e = 12.5 µm) and making use of the relation χK = α K (E / HK )1/2, where
α K is a proportionality factor (subscript K denotes the reference to Knoop indentation), αK
approaches a constant value of 0.0251, which is higher than the value obtained by Marshall

(1983), αK = 0.0229. However our reference material differs in material properties such as ν,
which αK (or χK) is sensitive to, as will be shown later.
3.2 Load-displacement curve
FE results follow Kick’s law (P ∝ h2, h is indentation depth; Fischer-Cripps 2002) very accurately
(Figure 7). Cracking makes Pmax decrease only slightly– 2 % for the reference material –, which
can be explained by the very low damage energy Edmg relative to the energy dissipated by plastic
deformation, EPD (Edmg ≈ 0.01 EPD).
1.2
E = 200 GPa, = 0.3, y = 5 GPa
1.0 no cracking
cracking
P / Pmax, no cracking
0.8 theoretical
relation
0.6
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
2
( h / hmax )
Figure 7. Load-depth data confirm the theoretical relation P ∝ h ; cracking hardly

2
affects the load-depth curve (hmax = 1.0 µm).

3.3 Relation between damage-initiating stress threshold σˆ and damage zone
length LDZ
The (bi-) linear cohesive zone law of a material is constituted by two independent material
parameters: fracture energy Γ and the damage-initiating stress threshold σˆ , which indicates the
brittleness of the material, e.g. for an ideally brittle material σˆ → ∞ (no damage zone). On the one
hand LDZ must be large enough to reach convergence, and on the other hand small enough to
comply with LEFM. Considering computational costs, LDZ / e should be small. A small e may
further result in an extreme stress peak at first contact and premature damage, which does not
occur in real indentation due to tip rounding. Further, convergence necessitates LDZ to be minimum
2e (Moës 2002). We find that with LDZ/e values between 1 and 2, convergence can be reached, but
at the expense of a significant increase in computational costs, while even lower LDZ/e lead to an
underestimation of c. Consequently, a trade-off must be made between a sufficiently high σˆ and a
reasonable mesh density in the inner region. Since a lower σˆ gives a lower crack length but a
larger damage zone, the crack length for an ideally brittle material, c*, must be c < c* < c + LDZ
(where c is the crack length corresponding to a finite σˆ ). c* can thus be approached by lowering e

and at the same time increasing σˆ while keeping in mind that LDZ ≥ 2e. Note also that a too high
σˆ inhibits damage initiation since maximum tensile stresses are limited by 1/3σy.
Assuming σˆ = 0.5, 0.8, 1.1 and 1.2 GPa, XFE analysis results using the reference material are
summarized in Table 1. The crack and damage shapes as well as Edmg are hardly influenced by σˆ
(Figure 8). The damage zone size obtained by FE results is in agreement with the theoretical value
obtained by Equation (4). As expected from above discussion, the crack length c increases with σˆ
and approaches, while LDZ decreases; for σˆ = 1.2 GPa the damage zone is only 1.5 % of the crack
length; yet LDZ < 2e. A method for estimating c* is given in the Appendix.
In this study we choose σˆ = 0.8 GPa, which gives on the one hand a small damage zone and on
the other hand good convergence and relatively low computational costs. However the brittleness
σˆ is a material property and exerts some influence on χK.
2
E = 200 GPa, = 0.3, y = 5.0 GPa
0
-2
σˆ [GPa]
-4 0.5
z [m]
0.8
-6 1.1
1.2
-8
-10
 = 3.0 MPa m
-12
0 2 4 6 8 10 12 14
r [m]
Figure 8. Damage (dotted lines) and crack (solid lines) shapes for diverse σˆ at load
reversal (semi-transparent) and after full unloading.
Table 1. Damage parameters for diverse σˆ (E = 200 GPa, ν = 0.3, σy = 5.0 GPa, Γ =
3.0 MPa µm, hmax = 0.8 µm).
σˆ Pmax c LDZ LDZ (FE)
LDZ / e LDZ / c χK
[GPa] [mN] [µm] [µm] [µm]
0.5 339.3 10.84 1.04 1.1 5.2 0.096 0.0854
0.8 345.0 11.44 0.40 0.5 2.0 0.035 0.0911
1.1 345.2 11.62 0.21 0.2 1.1 0.018 0.0932
1.2 346.2 11.72 0.17 0.1 1.0 0.015 0.0936

3.4 Viscous regularization
Models exhibiting softening behavior are susceptible to severe convergence difficulties because
the element stiffness tensor may cease to be positive definite. Gao and Bower (2004) showed that
the problem can be mitigated by viscous regularization damping. For the reference material the
range of appropriate viscosity values ζ that yield identical crack sizes and shapes is investigated
(Table 2). Plotting the energy dissipated by damage, Edmg (normalized by Edmg for ζ = 10–6 after
unloading), over indenter position h/hmax, we find that even for relatively high values (ζ = 1×10–4),
accurate crack shapes are obtained (Figure 9). Higher ζ values however retard crack evolution and
give highly inaccurate crack areas. On the other hand, analysis time generally decreases with
increasing ζ; in this particular case it drops between 10–5 and 10–4. In this study we use ζ = 5×10–5
to provide accuracy and at the same time relatively fast convergence. However, it is expedient to
check if Edmg still rises shortly after detachment of the indenter from specimen.
Table 2. Damage parameters for diverse ζ (E = 200 GPa, ν = 0.35, σy = 5.0 GPa, Γ =
3.0 MPa µm, σˆ = 0.8 GPa, hmax = 0.7 µm).
c cz Edmg tot tot
ζ Edmg / Edmg ,ζ =10−6 χK
[µm] [µm] [nJ]
1e–6 9.64 8.91 0.391 1.00 0.0909
1e–5 9.64 8.91 0.391 1.00 0.0909
1e–4 9.63 8.86 0.390 0.99 0.0904
5e–4 9.50 7.92 0.351 0.90 0.0870
1.2
loading unloading
1.0
-6
Edmg / Edmg, tot =10
0.8
0.6

0.4 1e-6
1e-5
0.2 1e-4
5e-4
0.0
0.0 0.5 1.0 0.5
1.5 0.0
2.0
h / hmax
Figure 9. Influence of viscosity value ζ on energy dissipated by damage, Edmg.

3.5 Influence of friction between indenter and material
Lawn (1975) interpreted the influence of friction in indentation theory as an increase of the
indenter angle ϕ by a friction-dependent term ± arctan µ (+ for loading; – for unloading), which
means that friction has a crack driving force-decreasing effect during loading (lower wedging

forces) and a driving force-raising effect during unloading (higher wedging forces). XFE results
for µ = [0, 0.4] (hmax = 0.8 µm) confirm this trend partly (Figure 10). While during loading
frictionless contact does not lead to a larger crack, during unloading the increased wedging forces
due to friction make the final crack slightly larger than the one resulting from frictionless contact.
The friction effect on c is to some part balanced by a decrease in Pmax, so that the sensitivity of χK
to µ is rather low (Table 3). The crack shape is not affected by friction.
Further, the influence of friction has an upper limit. No further increase in the final crack size can
be observed when raising µ from 0.2 to 0.4, which is because for µ ≥ 0.2 the contact type of a
surface node changes from slipping to sticking immediately after first contact. Edmg and c do not
increase either. The frictional effect on c is however too small to be considered quantitatively.
Table 3. Influence of friction (E = 200 GPa, ν = 0.3, Γ = 3.0 MPa µm, hmax = 0.8 µm).
c Pmax Edmg
µ χK
[µm] [mN] [nJ]
0.0 10.6 336.1 0.508 0.0840
0.1 10.9 338.6 0.534 0.0860
0.2 11.0 339.4 0.536 0.0870
0.4 11.0 339.4 0.536 0.0871
2
E = 200 GPa, = 0.3, y = 5.0 GPa
0
-2

-4 0.0
z [m]
0.2
-6 0.4
-8
-10 hmax = 0.8 m

 = 3.0 MPa m
-12
0 2 4 6 8 10 12 14
r [m]
Figure 10. Damage (semi-transparent) and crack shapes at load reversal

(dotted/squares) and after full unloading (solid lines/circles).
4. Influence of Poisson’s ratio, Young’s modulus and yield strain
Based on analytical considerations, Lawn (1980) arrived at χ ∝ (E/H)1/2 cot2/3ϕ ; the derivation
however bases on an oversimplified stress field and the neglect of Poisson’s ratio, which affect the
plastic zone shape and the crack driving forces. For Vickers and Berkovich indentation, the
problem was investigated by Lee (2012) and Hyun (2014). In this final section, we investigate the

sensitivity of χK to E, ν and σy. The results can be used to establish functions that map material
properties to χK.
Parameter studies are performed for combinations of E = {100, 200, 300, 400, 600} GPa; ν = {0.0,
0.1, 0.2, 0.3, 0.35}; σy = {3, 5, 8} GPa. The change of χK with E and ν is plotted in Figure 11 for
the three σy values. For the silicon nitride used by Marshall (1983) (E = 300 GPa, ν = 0.24, σy =
7.8 GPa 1), we get by interpolation χK ≈ 0.089, which is close to the experimental value of 0.096.
y = 3.0 GPa y = 5.0 GPa
0.22 0.18
0.20 0.16
0.18
0.14
0.16
K
K
0.12
0.14
0.12 700 0.10 700
600 600
0.10 500 0.08 500
400 ] 400 ]
0.3 300 Pa 0.3 300 Pa
0.2 200
[G 0.2 200
[G
0.1 100
E 0.1 E
 0.0  0.0 100
y = 8.0 GPa
0.14
0.12
0.10
K
0.08
700
600
0.06 500
400 a]
0.3
0.2
300 [ GP
200 E
0.1 100
 0.0
Figure 11. Change of coefficient χK with E and ν for σy = 3.0, 5.0 and 8.0 GPa.
Generally, the crack lengths cA and cB from indentations applying equal Pmax on two materials (A
and B) with equal Γ are related by
1
σy is found by varying its value until the resulting hardness matches the given Vickers hardness of H = 18.5 GPa.

2/3 1/3
χA  χ KA   EB 1 −ν A2 
Pmax A= Pmax ↔ =    2 
(5)
B
χB  χ KB   EA 1 −ν B 
Based on FE results (Figure 11) and Equation (5), the following trends for Knoop indentation
cracking can be deduced. Comparing two materials that vary only in their
(1) ν values: νA > νB, χKA < χKB → cA < cB (smaller ν → larger crack)
(2) σy values: σyA > σyB (HKA > HKB), χKA < χKB → cA < cB (smaller σy → larger crack)
(3) E values: EA > EB, χKA > χKB, cA < cB
Whereas (1) and (2) hold for arbitrary values of ν and σy, respectively, this does not necessarily
apply to (3), but for the range of materials used in this study.
ν has a significant influence on the crack formation. The length-to-depth ratio of the crack at load
reversal increases with ν (while that of the plastic zone decreases). This has ramifications on the
propagation during unloading where only the upper part of the crack opens up so that with
decreasing ν the crack loses its characteristic half-penny shape (Figure 12). Analysis results reveal
that χK decreases approximately linearly with increasing ν (Figure 11).
2
pl. zone
E = 200 GPa, y = 5.0 GPa
0
crack at
-2 Pmax

-4 0.0 ν = 0.0
z [m]
0.35
-6 pl. zone
-8 crack at
 = 3.0 MPa m Pmax
-10
hmax = 0.7 m
-12 ν = 0.3
0 2 4 6 8 10 12 14
r [m]
Figure 12. Change of crack shape with Poisson’s ratio ν (note that load Pmax
increases with ν for equal depths hmax).

plastic zone
crack at
Pmax
E = 100 GPa E = 200 GPa
Figure 13. For low E, the crack does not grow beyond the plastic zone in radial
direction during loading and does not open up during unloading; E = 200 GPa for
comparison; here the crack will open up during unloading (ν = 0.1, σy = 5.0 GPa).
For the lowest E value (100 GPa), the crack does not propagate to the surface during unloading
because the crack length in radial direction is smaller than the length of the plastic zone, as shown
in Figure 13 for ν = 0.1 and σy = 5.0 GPa. This observation is independent of σy, ν, and increasing
Pmax did not change this either. This is in agreement with the study by Petrovic (1983), who
reported for glass ceramic (E = 108 GPa) that the length of the crack in radial direction was
always slightly below or above a, irrespective of Pmax. For the highest E (600 GPa), we observe
that although damage initiates first in the material, a second (radial) crack initiates at the surface at
a later stage of the loading cycle. The radial crack then merges with the median crack. Thus, with
increasing E the region of first initiation is expected to move to the surface.
An increase in σy is tantamount to an increase in hardness HK so that the decrease of χK with
increasing σy is in agreement with the proportionality χK ∝ (E/HK)1/2.
5. Summary
An XFE model was established that allows the simulation of Knoop indentation cracking. Critical
issues associated with cohesive material behavior were discussed and suggestions were made for
the viscosity parameter ζ , the damage-initiating threshold σˆ and the element size e. The crack
evolution was in accordance with experimental findings from the literature. Parameter studies
were performed to elucidate the influence of material properties on the crack size and shape.
Results serve to predict the fracture toughness of a material from Knoop indentation by taking into
account material properties that affect the plastic zone evolution and hence the final crack shape.
Knowledge of the final crack shape and size is further essential if the crack is used as a starter
crack for crack growth experiments or for the evaluation of in-plane residual stresses.
6. References
1. Abaqus Users’ Manual, Version 6.13, Dassault Systèmes Simulia Corp., Providence, RI, 2013.
2. Ahn, Y., Chandrasekar, S. and Farris, T.N., “Determination of Surface Residual Stresses in
Machined Ceramics Using Indentation Fracture,” Journal of Manufacturing Science and
Engineering, Vol. 118, 1996.

3. Anderson, T.L., “Fracture Mechanics: Fundamentals and Applications,” CRC Press, 2004.
4. Fischer-Cripps, A.C., “Nanoindentation,” Springer New York, 2002.
5. Francois, P., Lefebvre, A. and Vanderschaeve, G., “Low Temperature Plasticity of Brittle
Materials: A New Device for Compressive Testing Under Confining Pressure,” Physica
Status Solidi (A), Vol. 109, 1988.
6. Gao, Y.F. and Bower, A.F., “A Simple Techniques for Avoiding Convergence Problems in
Finite Element Simulations of Crack Nucleation and Growth on Cohesive Interfaces,”
Modelling and Simulation in Materials Science and Engineering, Vol. 12, 2004.
7. Hutchinson, J.W. and Evans, A.G., “Mechanics of Materials: Top-down Approaches to
Fracture,” Acta Materialia, Vol. 48, 2000.
8. Hyun, H.C., Rickhey, F., Lee, J.H., Hahn, J.H. and Lee, H., “Characteristics of Indentation
Cracking Using Cohesive Zone Finite Element Techniques for Pyramidal Indenters,”
International Journal of Solids and Structures, Vol. 51, 2014.
9. Keer, L.M., Farris, T.N. and Lee, J.-C., “Knoop and Vickers Indentation in Ceramics
Analyzed as a Three-Dimentional Fracture,” Journal of the American Ceramic Society, Vol.
69, 1986.
10. Lawn, B.R., Evans, A.G., and Marshall, D.B., “Elastic / Plasic Indentation Damage in
Ceramics: the Median / Radial Crack System,” Journal of the American Ceramic Society, Vol.
63, 1980.
11. Lawn, B.R. and Fuller, E.R., “Equilibrium Penny-like Cracks in Indentation Fracture,”
Journal of Materials Science, Vol. 10, 1975.
12. Lee, J.H., Gao, Y., Johanns, K.E. and Pharr, G.M., “Cohesive Interface Simulations of
Indentation Cracking as a Ftracture Toughness Measurement Method for Brittle Materials,”
Acta Materialia, Vol. 60, 2012.
13. Lube, T., “Indentation Crack Profiles in Silicon Nitride,” Journal of the European Ceramic
Society, Vol. 21, 2001.
14. Marshall, D.B., Evans, A.G., Khuri Yakub, B.T., Tien, J.W. and Kino, G.S., “The Nature of
Machining Damage in Brittle Materials,” Proceedings of the Royal Society of London A, Vol.
385, 1983.
15. Moës, N. and Belytschko, T., “Extended Finite Element Method for Cohesive Crack Growth,”
Engineering Fracture Mechanics, Vol. 69, 2002.
16. Mohammadi, S., “Extended Finite Element Method,” Blackwell Publishing, 2008.
17. Petrovic, J.J., “Effect of Indenter Geometry on Controlled-Surface-Flaw Fracture Toughness,”
Journal of the American Ceramic Society, Vol. 66, 1983.
18. Ponton, C.B. and Rawlings, R.D., “Vickers Indentation Fracture Toughness Test: Part 1.
Review of Literature and Formulation of Standardised Indentation Equations,” Materials
Science and Technology, Vol. 5, 1989a.
19. Ponton, C.B. and Rawlings, R.D., “Vickers Indentation Fracture Toughness Test: Part 2.
Application and Critical Evaluation of Standardised Indentation Toughness Equations,”
Materials Science and Technology, Vol. 5, 1989b.
20. Quin, G.D. and Bradt, R.C., “On the Vickers Indentation Fracture Toughness Test,” Journal
of the American Ceramic Society, Vol. 90, 2007.

21. Rice, J.R., “A Path Independent Integral and the Approximate Analysis of Strain
Concentration by Notches and Cracks,” Journal of Applied Mechanics, Vol. 35, 1968.
22. Tomar, V., Zhai, J. and Zhou, M., “Bounds for Element Size in a Variable Stiffness Cohesive
Finite Element Model,” International Journal for Numerical Methods in Engineering, Vol. 61,
2004.
Appendix
c* can be estimated based on the following notions: (1) Edmg is invariant of σˆ ; (2) the damage and
crack shapes are independent of σˆ . The sum of damages D over all elements n multiplied by the
corresponding element area, here e2, is
Edmg
∑ e2 Dn ≈ Γ
(6)
n
Invoking assumption (1), this must be equal to the ideally brittle case. Thus,
∑ e2 Dn = k c* 2 (7)
n
where k is a parameter describing the shape of the damage/crack area. Since k is independent of σˆ
(assumption 2), k can be obtained from the finite σˆ case, i.e.
1 for Di > 0
k ( c + LDZ ) ∑ e2 H (=
Di ) ; H ( Di )
2
=  (8)
i 0 for Di = 0
where LDZ is given in Equation (4). Solving for k and inserting the resulting expression into
Equation (7), we get
*
c= ( c + LDZ ) ∑ e2 Dn ∑ e2 H ( Di ) (9)
n i
Using the value of output StatusXFEM for D we get for our reference material c* ≈ 11.81 µm (hmax
= 0.8 µm).
Acknowledgement
This work was supported by the Basic Science Research Program through the National Research
Foundation of Korea (No. NRF-2012 R1A2A2A 01046480).

Evaluation of the Fatigue Strength of a Steel-
Aluminum Tapped Thread Joint with Local
Concepts
D. Olvedaa, R. Waterkotteb, M. Oechsnera
a
Technische Universität Darmstadt, Fachgebiet und Institut für Werkstoffkunde, Darmstadt
b
Operational Strength Competence Center, Schaeffler Technologies AG & Co. KG,
Herzogenaurach
Abstract: VDI (Association of German Engineers) guideline 2230 (VDI, 2013) provides a basis on
which the fatigue strength of bolt-nut joints can be estimated. However, it is not suitable for use in
reference to general threaded joints, as the VDI guideline is based on the nominal stress concept
and it is often not possible to define a nominal surface area for such general joint types.
Furthermore, regulations that do not explicitly deal with the design of threaded joints – such as
the FKM guideline "Analytical Strength Assessment of Components” (FKM, 2013) – also deliver
unsatisfactory estimates of the fatigue strength of threaded joints.
This paper deals with the design of a steel/aluminum tapped thread joint. The concept developed
by Schneider (Schneider,2011) is used as the basis for assessment up to the occurrence of
technical cracks. Crack propagation, up to component failure, is described with linear elastic
fracture mechanics and using the finite element method (FEM). For the purpose of validating the
concept, the calculated lives are compared with experimental tests.
Keywords: Fatigue, Fatigue Life, Fracture Mechanics, Threaded Connections
1. Introduction
Screw connections are designed according to the VDI guideline on the basis of the nominal stress
concept. The assumption is made here that the bolt will fail. Dimensioning of general threaded
joints for which a nominal surface area cannot be defined and/or in which the nut component fails
can therefore not take place according to the guideline. In such cases, the local concept is an
approach that allows the fatigue strength to be estimated.
A method for calculating the fatigue strength of threaded joints based on the local concept is
presented in (Schneider, 2011), refer to figure 1. The local concept is used to calculate the fatigue
life until the occurrence of technical cracking. Crack propagation until fracturing occurs is
measured by means of an additional crack propagation calculation using linear elastic fracture
mechanics. It is thus possible to make a direct comparison between the calculated fatigue life
values and the S-N curve produced for the components by the tests.

Damage parameter PJ curve Crack propagation calculation
a,c
fracture
PJ
crack
PJ(N)
Number of cycles N NB N
Stress amplitude Sa
fracture S-N curve
crack initiation
S-N curve
Number of cycles N
Figure 1. Procedure for assessing general threaded joints according to (Schneider, 2011)
In addition to the fatigue life calculation, the test procedure also presents a particular challenge
with regard to the assessment. Due to the complexity of the threaded joint being tested, the cracks
do not extend to the outer edge of the joint, which makes it more difficult to detect component
failure.
2. Tested Joint
The joint being tested is a steel bolt with a through hole that is screwed into an aluminum body
with a cut internal thread (M22 x 1), refer to figure 2. The steel bolt is fitted into the aluminum
body via the support surface, with a maximum of 4-5 thread turns engaged. The aluminum body is
made from the wrought alloy EN AW-6082-T6.
Figure 2. Steel/aluminum tapped thread joint: Left = top view, right = half-section

3. Fatigue Life Calculation
3.1 Geometry and meshing
The basis for the assessment according to (Schneider, 2011) is the calculation of the local load at
the base of the thread. An axially symmetrical 2D FE model with an elastic-plastic material law is
used for this purpose. Tests in (Schneider, 2011) and (Seybold, 2006) show that, at the point of
maximum stress at the base of the thread, the calculation results of 2D models are only minimally
different to those of 3D models.
Figure 3 shows the model used, together with its boundary conditions. The lower end face of the
nut is fixed (u1 = u2 = uR3 =0). Operating loads are applied via a reference point that is
kinematically coupled with the bolt support surface. The axial preloading force is specified using
the elongation of the elements within a partitioned range. The FE program Abaqus uses the load
command "Bolt load" for this purpose (Abaqus, 2012). In the contact areas of the support
surface and thread flanks, a Coulomb friction of μ = 0.1 was assumed.
Figure 3. Model connection left: with boundary conditions; right: with contact pairs
The meshing of the overall component and the base of the thread is shown in figure 4. It was
possible to achieve even meshing of the thread geometry of the nut and bolt using several
partitioned ranges. At least 30 elements were used in the base of the nut thread here, which
corresponds to the recommendations in (Seybold, 2006). An element length of 0.01 − 0.02 mm
was selected along the flanks of the nut and the bolt. The global element length was set at approx.
0.5 mm. Approximately 33 000 elements are therefore required for the nut and 7 300 elements for
the bolt when calculating the joint.

Figure 4. Meshing of the tested threaded joint
3.2 Material model and load-time function

The material behavior of the bolt was assumed to be linear-elastic, whereby an elasticity module
of 𝐸 = 210 GPa and a Poisson’s ratio of 𝜈 = 0.3 were used. In the case of the aluminum body,
elastic-plastic material behavior was used as the basis. In addition to the elasticity module of
𝐸 = 64.9 GPa that was determined through experiments and the Poisson’s ratio of 𝜈 = 0.34, static
and cyclical flow curves are used that are determined through experiments (refer to figure 5).
The hardening law used for the calculation depends on the load case. In the case of a static load,
an isotropic hardening is used as the basis, taking the static flow curve into consideration. The
cyclical load is calculated using the material model according to LEMAITRE and CHABOCHE and
the cyclically stabilized flow curve.
The load-time function is dependent on the load type. In the case of static loading, a calculation up
to a highest load of 𝐹𝑂 is sufficient. If a cyclical load is to be calculated, a further calculation stage
is used in addition to the highest load in order to apply the lowest load 𝐹𝑈 (figure 5, right).

300
250
true stress in MPa
200
150
100 static σ-ε curve

cyclic σ-ε curve
50 E = 64900 MPa
ν = 0,34
0
0 1 2 3 4 5
true plastic strain in %
Figure 5. Left: Static and cyclical flow curve EN AW-6082-T6, Right: Load-time function F(t)
3.3 Measuring the local stress-strain history

To determine the load, the tangential stresses and strains in the base of the thread are extracted. As
an example, figure 6 shows the local stress-strain histories at the point of highest load, i.e. in the
first load-bearing thread turn, for a force amplitude of 𝐹𝑎 = 10kN (𝑅 = 10, 𝐹𝑉 = 18kN). A
differentiation must be made between the first principle stress 𝜎1 , the tangential stress 𝜎𝑡 , and the
equivalent tensile stress according to von Mises 𝜎𝑉 . In figure 6, it can be seen that the values for
the first principle stress are virtually identical to the values for the tangential stress until the
maximum stress 𝜎𝑜 is reached, while the von Mises stress already deviates from the main stress in
this range. At high load amplitudes (as shown here) however, an incorrect evaluation of the actual
load occurs after the load reversal in the case of these specific load parameters. When the load is
reversed, the tangential proportion of the first principle stress declines to such an extent that,
according to the convention 𝜎1 > 𝜎2 > 𝜎3 , the highest stress value is read out but not the highest
stress value in terms of amount. It can be concluded from this that the tangential stress 𝜎𝑡 is best
suited to the evaluation of the local load.

350
300
250
Stress σ in MPa
200
150
100
50
0
Load:
-50 R=10, F = 10 kN
a
-100 σt
σ1
-150
σv
-200
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Strain ε in %
Figure 6. Various local stress-strain curves in the first load-bearing thread turn, 𝑭𝒂 = 𝟏𝟎kN, 𝑹 = 𝟏𝟎
3.4 Calculation of the incipient cracking S-N curve

The S-N curve for incipient cracking is calculated according to the method described in
(Schneider, 2011) on the basis of the local concept. This requires both the hysteresis and its exact
position to be estimated. To do this, the local maximum stress/strain that occurs when the screw is
tightened is calculated using the static material law. The hysteresis is then calculated with a
cyclical material law and subsequently applied to the static maximum stress/strain. The hysteresis
is then shifted using the stress relaxation value derived by means of experiments. The position and
hysteresis have thus been determined and can subsequently be assessed.
As a damage parameter, the parameter PJ is selected according to VORMWALD (Vormwald, 1989)
and the number of load cycles for incipient cracking calculated for the relevant load case, taking
the support effect into account. The S-N curves that is thus calculated is shown in figure 7.

12
11 S-N curve Steel-Aluminum Tapped Thread Joint
10
9
load amplitude Fa in kN
8
7
6
5
3 FE-analysis - R=10:
Evaluated with PJ
Crack initiation S-N excluding support effect nσ=1,0
Crack initiation S-N including support effect nσ=1,84
Crack initiation S-N including nσ und stress relaxation
2
103 104 105 106 107 108
load cycle N
Figure 7. Calculated S-N curves for incipient cracking in the joint according to PJ
3.5 Calculation of crack propagation

The joints are not tested until incipient cracking occurs, which cannot be determined by
measurement means, but until a specific crack size is reached. A comparison of the calculated and
experiment-based S-N curves can therefore only be made if the crack propagation is taken into
consideration. A number of simplified replacement models for calculating the number of stress
cycles for crack propagation is provided in regulations such as the FKM guideline "Fracture
Mechanics Proof of Strength for Engineering Components" (FKM, 2009). A further option for
determining crack propagation is to use the FE method. The advantages of this type of calculation
include the consideration of thread load distribution mapping and the contact between the thread
flanks and the support surfaces, with which the friction is also taken into consideration.
In the axially symmetrical model, a crack is applied at the point of highest load and the stress
intensity factor is determined. The crack length is varied in the direction of the highest load in
order to test the influence of the crack length on the stress intensity factor. A linear-elastic
calculation is carried out for this purpose; the mesh was refined in the vicinity of the crack edge
(refer to figure 8).

Figure 8. Stress graph for the crack model, crack length 𝒂 = 𝟎, 𝟕mm, 𝑭𝑶 = 𝟐𝟐, 𝟐kN
After evaluation of the contour integrals, a dimensionless shape function can be determined
through an interpolation of the results and taking the basic equation 𝐾 = 𝜎√𝜋𝑎 ∙ 𝑌(𝑎, 𝑑) into
consideration. A power function was used as the interpolation formula here. The resulting shape
function is as follows:
𝑎 −0,486
𝑌(𝑎, 𝑑) = 0.419 ∙ � � ,
𝑑
whereby 𝑎 is the crack length and 𝑑 is the bolt diameter. Figure 9 shows a comparison of the stress
intensity factors of the calculation with those of the interpolation formula. The hypothetical
nominal stress was defined as follows:
𝜎 = 𝐹 ⁄A
With 𝐴 as the hypothetical bolt cross-section (𝐴 = 𝑑 2 ∙ 𝜋⁄4) and 𝑑 as the outside bolt diameter
𝑑 = 22mm.
The crack propagation is calculated according to linear elastic fracture mechanics. The resulting
function is applied to the PARIS-ERDOGAN equation (𝑑𝑎�𝑑𝑛 = 𝐶 ∙ 𝐾 𝑚 ) and integrated.
𝑎End 𝑎End
1 1
𝑁= � 𝑑𝑎 = � 𝑑𝑎
𝐶∙ (∆𝐾)𝑚 𝐶∙ (𝑓(𝑎)) 𝑚
𝑎Begin 𝑎Begin
The number of stress cycles for crack propagation can be calculated with the help of this equation
by applying the initial and final crack length (𝑎Begin , 𝑎End ) and the material parameters 𝐶 and 𝑚.

6.5
stress intensity factor K in MPa⋅√m 6.0
5.5
5.0
4.5 K-FEM
K-interpolated
4.0
0.0 0.2 0.4 0.6 0.8 1.0
crack length a in mm
Figure 9. Interpolation 𝑲 − 𝒂 and calculated results with a load of 𝑭𝑶 = 𝟐𝟐, 𝟐kN
4. Component Tests
4.1 Test setup
Test setup for testing the threaded joint is shown in figure 10 (left). The assembled threaded joint
is placed on a shim, the purpose of which is to compensate for possible bending loads. Below the
shim is a supporting body that is connected with the load cell of the high-frequency pulser (HFP).
The force is applied via the end face of the bolt being tested using a pin, which exerts a pulsating
compressive load.
Cracks form in the uppermost thread turn of the load transfer of the aluminum housing (figure 10,
right). Crack propagation initially occurs in a radial direction and, if further propagation occurs,
continues axially in the direction of the support surface of the nut and bolt. Ultrasonic testing
equipment is used to measure the crack growth. Figure 10 shows a schematic representation of a
measurement of the test specimen.

Figure 10. Left: Test setup on the high-frequency pulser;
right: Schematic representation of the crack measurement using ultrasound
4.2 Test procedure - determination of a cut-off criterion

A fracture in the joint that is not due to the direction of crack propagation does not occur during
the course of the test and can therefore not be used as a cut-off criterion. Additionally, there is no
frequency dip in the crack propagation, which also eliminates the possibility of machine
monitoring. Iterative crack detection using ultrasound (US) was therefore carried out. Figure 11
shows a process diagram for crack detection. Here, the test specimen is measured before initial
fitting using the ultrasonic testing device and subsequently tested up to 𝑁1 load cycles. The S-N
curve for estimating the number of stress cycles is derived from the previous calculation. After the
number of stress cycles 𝑁1 has been reached, the test sample is tested again using ultrasound. If no
identifiable crack is present, the test is extended by ∆𝑁 stress cycles and the inspection is then
repeated. This procedure continues until the presence of a crack can be verified.

Initial measurement of
the specimen via US
Test up to
Nlimit
Check tested specimen Continue N1
with US testing until
Crack detected? ∆Ntest
Yes No
Remove reached?
specimen
Yes No 104 105 106 107
load cycle N
Figure 11. Process diagram for iterative crack detection
The test specimens are opened up after the test in order to inspect the cracks that have been
detected using ultrasound. This is done by creating a recess that is level with the defect and
statically pulling the test specimen apart. Test specimens in which no crack was identified were
not included in the evaluation. The S-N curve for the components that is derived after crack
detection is shown in figure 12.
4.3 Test evaluation

Due to the iterative process, the tested specimens have cracks of differing sizes. The reason for
this is that the detection of cracks between the iteration periods is not always clear. It is thus
possible that test specimens in which cracking has already occurred are not clearly detected until
several iteration periods have been completed and the crack is therefore larger. In order to define
uniform failure conditions, the crack initiation fatigue life of each individual test specimen was
calculated on the basis of the experimental fatigue life values and using the crack propagation
calculation described above. For this purpose, the fracture surfaces of the individual test
specimens were laid open using the procedure described above and the cracks measured.
Figure 12 shows the failure S-N curves and the S-N curves without (calculated) crack propagation.

11
S-N curves Steel-Aluminum Tapped Thread Joint
10
residual fracture surface

9
8 fatigue fracture surface

kcrack ini.=2,78
kfailure=2,74
7 Crack length:
thread flank a=0.79 mm
σa,D =5,94kN
6 Nk,D =4,9⋅105 → 2,2 x crack

→4
Test: R=10 Failure S-N curve P=50%
Specimen with propagated crack Crack initiation S-N curve P=50%
Specimen without (calculated) crack propagation Crack initiation S-N curve P=10% u. 90%
Run out
5
104 105 106 107
load cycle N
Figure 12. S-N curve defined using experiments after subtraction of the number of stress cycles for
crack propagation
5. Comparison of Results
Figure 13 shows the comparison of the S-N curves that were calculated and those that were
defined using experiments.

11
Tested vs. Calculated S-N curves
10
Test: R=10
Specimen with propagated crack
9 Specimen without (calculated) crack propagation
Run out
Failure S-N curve P=50%
Crack initiation S-N curve P=50%
8 FE-Analysis: nσ=1.84
Crack initiation S-N curve PJ with σRelax
6 → 2, 2 x crack
→4
5
104 105 106 107
load cycle N
Figure 13. Comparison of the S-N curves defined by calculation and by experiments
6. Summary
A calculation-based estimation of the rating life of a threaded joint, which cannot be calculated
using regulations (VDI, 2003 / FKM, 2013), was carried out during this project. The basis for this
calculation is the concept according to (Schneider, 2011). In addition, a stress intensity solution
that can be used to describe the crack propagation that occurs in the component being analyzed
here was calculated on the basis of the threaded joint's FE model.
Using ultrasonic testing, it was possible to carry out fatigue tests on the joint, which produced
different-sized cracks. An individual failure criterion was then determined by individual
calculation of the number of stress cycles for crack propagation for the individual tests, using the
numerical model. The experimental S-N curve for incipient cracking that was extrapolated from
this can be compared with the calculated S-N curve.
The comparison of the results showed that the fatigue life had been overestimated with regard to
fatigue resistance. Further tests on the model are needed here in order to improve the results. A
calculation using the 3D model may be necessary. This might, for example, allow a more reliable
mean load to be determined based on the tightening procedure and the influence of the thread pitch
to also be mapped.

7. References
1. Verein Deutscher Ingenieure (VDI): Systematic calculation of highly stressed bolted joints –
Joints with on cylindrical bolt 2230 Part 1, 2003.
2. Forschungskuratorium Maschinenbau (FKM): Analytical Strength Assessment of
Components, 6th revised ed. Frankfurt am Main, 2013.
3. Schneider, R.: Örtliche Bewertung der Schwingfestigkeit von Gewindeverbindungen,
Dissertation Darmstadt, 2011.
4. Seybold, R.: Finite-Elemente-Simulation örtlicher Beanspruchungen in Schraubengewinden,
Dissertation. Darmstadt 2006.
6. Vormwald, M.: Anrisslebensdauervorhersage auf der Basis der Schwingbruchmechanik für
kurze Risse, Dissertation, Darmstadt 1989.
7. Forschungskuratorium Maschinenbau (FKM): Fracture Mechanics Proof of Strength for
Engineering Components, 3rd ed. Frankfurt, M. 2009.
8. Acknowledgements
IGF project 16920 N/1 by the FKM - Forschungsvereinigung
Forschungskuratorium Maschinenbau e.V. - , Lyoner Straße 18,
60528 Frankfurt am Main, Germany, was subsidized by the
German Federal Ministry for Economic Affairs and Energy via the
AiF as part of its subsidization program for industrial research and
development partnerships (IGF) following a resolution by the
German Federal Parliament.

Strip Shape Produced by 4 High Cold Rolling
Process Simplified with a Static Model.
M. A. Bello-Gomez1, M. P. Guerrero-Mata1, L. A. Leduc Lezama1, T. P. Berber-
Solano1, F. Gonzalez2, H. R. Siller3
1
Facultad de Ingeniería Mecánica y Eléctrica, Universidad Autónoma de Nuevo León, Pedro de
Alba S/N, Ciudad Universitaria, C.P. 66450, San Nicolás de los Garza, México.
2
Ternium Mexico, Av. Los Angeles 325 Ote, San Nicolás de los Garza, 66452, N. L., México
3
Tecnológico de Monterrey, Ave. Eugenio Garza Sada #2501 Sur. C. P. 64849. Monterrey, NL.,
México.
Abstract: The cold rolling process is one of the most important in the metal industry. The strips
produced by this process are used in different applications. The quality and shape control of these
products are very important factors during the production process. It is well known that the strip
shape is influenced by: the rolling speed, strip width, forces, rolls deflection, etc. This work
considers modelling a 4 high cold rolling process and obtains the strip profile, with a static model
using an arc contact with springs representing the whole model. This idealized model was
developed to reduce the computational time of the dynamic model being used. An arc is defined
with the optimal dimensions to reproduce the process, taken into account concentrations of the
stresses, forces and displacements at the work roll/strip contact; the arc used in the static model
was meshed in accordance with the mesh of the work roll in the dynamic model. The arc in this
model is helped by springs placed at the arc inner surface; the stiffness of the springs was
calculated with previous measurements of the work roll deflection values. The results show that
the arc static model reproduces correctly the roll profile and strip shape with an important
difference in the computational time.
Keywords: Cold rolling, Flatness, Strip Crown
1. Introduction
The cold rolling is a process where a metal strip is plastically deformed to a certain thickness with
better surface finish and dimensional tolerance than the hot rolled strips. The process consists in a
metal strip (steel) going through a rolling mill at room temperature. The strips used in the cold
rolling must be previously hot rolled. Cold rolling decreases the thickness but also improves the
surface finish.
Some cold rolled strips are used at the car industry and the house hold appliances; this is the
reason why the surface finish and the shape control play are essential at this process.

1.1 Types of mills
There is a wide variety of mills used for the cold rolling process. The more common are the; two
high, four high, six high and the Sendzimir mill, these are classified by the number of rolls
composing the mill. It is important to mention that the shape control and surface finish can be
affected by the mill configuration.
This study considers a 4 high mill composed by for rolls: two work rolls and two back up rolls, as
it is showed in Figure 1. The work rolls are the small rolls which are in contact with the strip. The
backup rolls are those supporting the work rolls.
Figure 1. 4 High mill.
1.2 Shape control and flatness
The term shape applied to the rolled strips can be used to define the cross-sectional geometry of
the strip or the ability of the strip to stay flat on a horizontal planar surface. The crown is defined
as the difference in thickness between the center of a strip and its edges. (Roberts).
The flatness defects are common during the rolling process; they are principally grouped in four
cases: quarter buckle, wavy edges, center buckle and short quarter. The shape of the strip is caused
by the changes of the strip thickness profile during the rolling process, these thickness variations

occur due the roll force distribution (Dieter). The Figure 2 shows a scheme illustrating how the
flatness is produced by the profile evolution.
Figure 2. Strip flatness defects illustration (Abdelkhalek).
The strip flatness and the strip crown (Figure 3) are basic parameters during the flat strip
production. The strip crown is calculated using equation 1:
Figure 3. Crown definition (Miani).
(𝑥1 +𝑥2 )
𝐶 =𝑌− Equation 1
2
2. Modelling
This work is a simplification of a dynamic model created before; this model was elaborated with
the intention to decrease the computational time of a 4 high rolling mill dynamic model. Both
models were prepared in Abaqus/CAE; the dynamic model was solved using Abaqus/Explicit. The
characteristics of the dynamic model and the static simplification are detailed below.

2.1 Dynamic model
The dynamic model recreates the cold rolling process by a 4 high mill. But this model uses a
symmetry consideration in the horizontal and vertical axes to decrease the number of nodes,
elements and computational time. The Figure 4 illustrates the resulting geometry of this model.
1/4
Total 1/2
Figure 4. Symmetry conditions applied to the model
2.1.2 Dynamic model mesh
The mesh of this dynamic model is composed by 5,642,832 nodes and 5,248,108 elements (table
1); it is showed in Figure 5.
Table 1. Number and element type for Dynamic model.
Element type Number

Linear hexahedral C3D8R 5,214,372
Linear wedge C3D6 33,736
Figure 5. Mesh used for the dynamic model.
The elements in the strip use hourglass control: Stiffness.
2.1.3 Obtained profile

The work roll profile and the strip shape obtained by the dynamic model, replicated the real
response ofthe 4 high rolling mill. The Figure 6 shows the obtained roll profile by this model.
Obtained profile
0.12
0.1
Vertical displacement
0.08
0.06
0.04
0.02
0
0 200 400 600 800
Roll distance
Figure 6. Work roll obtained profile with the dynamic model.
The experimental validation of this model was analyzed using a comparison between the load
applied at the industrial process and the reaction force obtained with the dynamic model. The
Figure 7 shows how the reaction force is changing at different times, stabilizing in the same value
of the measured force.

Measured force:
6
10.9x10 N=1114.9 Ton
Figure 7. Comparison of the Reaction force (Dynamic model) and the Measured
force (Industrial process)
2.2 Static model simplification
With the purpose of decrease the computational time of the dynamic model. Some simplifications
were applied to the model. The model simplification is composed by an arc representing the outer
face of the work roll and springs simulating the roll deflection (Figure 8).
Figure 8. Arc representing the work roll outer face.
2.2.1 Simplification Model components

As it was mentioned before the simplification model the model has two components.

a) Static Arc: This part represents the outer part of the roll, were the roll profile is measured.
The thickness dimension, was obtained with the results of the dynamic model. It was
obtained with the effect of the stress in the work roll as is shown in Figure 9.
Figure 9. Stress distribution between the work roll and the strip.
The circumference of the arc represents 1/6 of the roll circumference. The longitude of
the arc is the same as the roll table.
b) Springs: this model reproduces the roll deflection during the rolling process with springs
located at the inner face of the arc (Figure 10).
Figure 10. Springs distributed over the arc inner face.
2.2.2 Mesh
The mesh applied to this model tries to replicate the mesh used at the dynamic model, with some
variations to decrease the number of elements (Figure 11 and 12).

Figure 11. Dynamic model mesh. Figure 12. Arc mesh.
The elements of the mesh at the center of the arc are smaller than those at the edges, because the
stress center, where the arc of contact between the strip and the roll is located. The mesh of the
static model is detailed in table 2.
Table 2. Arc mesh description
Total number of nodes 638, 664

Total number of elements 594, 396
Type of elements Linear hexahedral elements C3D8R
2.2.3 Boundary conditions and Load distribution
As a part of the simplification of the model, the strip is represented by a load applied as pressure at
the arc of contact (Figure 13).
Figure 13. Load distributed at the arc of contact.

The arc inner face is covered with springs at each node. To calculate the spring stiffness was
necessary to obtain the roll deflection in the dynamic model; using a node path at the center of the
roll. The position of the node path is illustrated in the Figures 14 and 15.
Node path
Node path
Figure 14. Node path position. Figure 15. Node path position.
The vertical displacements of each node of the node path were graphed versus the roll distance
from the edge to the center of the roll. This graph can be observed in Figure 16.
Figure 16. Roll deflection.
The “k” coefficients applied to the springs at the inner face of the arc, were calculated applying the
Hooke´s law (Equation 2).
𝐹
𝐾= Equation 2
𝑥

Where F is the reaction force obtained at the dynamic model and x is the maximum vertical
displacement obtained in the roll deflection graph. These values where multiplied by 0.5 due the
symmetry conditions.
This obtained coefficient “k” was also divided into the number of springs located at the arc. The
zones where the strip does not interact with the strip used a K 1000 times bigger than the one
located at the center.
3. Results and discussion.

The mesh used at the static arc is simpler than the one used at the dynamic, but trying to replicate
the structured applied for the dynamic model. Table 3 shows the differences between the two
meshes.
Table 3. Differences between the two models mesh densities.
Dynamic Arc Static
Number of nodes 5,642,832 638, 664
Number of elements 5,248,108 594, 396
Element type Linear hexahedral (C3D8R) and Linear hexahedral (C3D8R)

Linear wedge (C3D6)
Due the difference in the number of elements used at the meshes from the dynamic model and the
static model, the computational time of the analysis varies drastically, changing the time form
days to minutes.
The resultant profiles obtained by each model are very similar; the comparison of these profiles
can be seen in Figure 17. It can be observed how the arc static model can replicate the response of
the dynamic model and therefore the cold rolling process.

Profile comparison
0.14
0.12
Vertical displacement
0.1
0.08
0.06 Dynamic
Static Arc
0.04
0.02
0
0 200 400 600 800
Roll Distance
Figure 17. Comparison of the static and dynamic profile.
4. Conclusions
The static simplified model replies correctly the profile obtained at the cold rolling process. This
works probes a new way to reproduce the strip shape obtained at the cold rolling with a minimum
number of nodes and elements. Using a static arc to represent the roll flattening, as well as a
springs to recreate the work roll deflection and the backup roll support.
The computational time changes from days to minutes, making the arc simplification a convenient
tool to reproduce the cold rolling process. The table 4 shows the different CPU time and
Increments of the Dynamic model and the Arc Static model.
Table 4. Differences between the two models CPU time and Increments.
Dynamic Arc Static
CPU time 291, 851 4, 816.2
Increments 886, 952 17

Both models were running in a computer with two Intel® Xenon® CPU X5650 @ 2.67 GHz, and
64 GB of RAM.
The thermal part of the simulation will be added to this model, to recreate how the heat flow
interferes with the strip crown.
5. References
1. Abdelkhalek, S, Montmitonnet, P, Legran, N, Buessler, P, “Coupled approach for flatness
prediction in cold rolling of thin strip”, International Journal of Mechanical Sciences 53
661-675, 2011.
2. Dieter, G, “Mechanical metallurgy”, Mc Graw-Hill, 1988.
3. Miani, Fabio, “Definition of a New Parameter for Flatness Control in Hot Rolling Thin
metal Strip Production”, AISTech Conference Proceedings, 2007.
4. Roberts, W L, “Cold rolling of steel”, Marcel Dekker, 1978.
6. Acknowledgment
The authors acknowledge the support provided by CONACYT, and Ternium Mexico for the
facilities given during this work.

Validation of Flow Simulation on Aabqus/CEL™
Saucedo O. Alan R.*, Ramírez D. Edgar I., Ruiz C. Osvaldo, Ortiz P. Armando
Unidad de Investigación y Asistencia Técnica en Materiales (UDIATEM), Facultad de Ingeniería,
Universidad Nacional Autónoma de México (UNAM), Circuito exterior, Ciudad Universitaria,
Anexo de Ingeniería, Laboratorios de Ingeniería Mecánica Alberto Camacho Sánchez,
Delegación Coyoacán, C.P. 04510, México D. F.
alanr.saucedoo@gmail.com, edgarisaac80@gmail.com, oruiz@unam.mx, armandoo@unam.mx
Abstract: The main topic of the paper is to show the validation of the Coupled Eulerian-
Lagrangian (CEL) analysis technique applied to Newtonian fluid dynamics simulations described
by an equation of state (EOS) suggested by Mie-Grüneisen. The model is driven only by body
forces through cavity filling. This implementation allows to obtain free surfaces profile, splash
droplet formation, velocity field, instantaneous pressure at the contact points between solid and
fluid, as well as in solid parts, stresses and strains, all at the same time.
The free surface validation obtained in the fluid is made through comparisons with radiographs
obtained experimentally.
The filling is achieved through solid-fluid interaction (FSI) with "General contact (explicit)" (GC)
this is able to track the material distribution during the free surfaces formation and applies non-
slip boundary condition as interaction between fluid and solid in order to determine the effect of
this approach on the material flow. As a result the complete standardized cavity filling is achieved
satisfactorily using CEL analysis.
Keywords: Cavities Filling, Material Flow Analysis, CEL, Fluid-Solid Interaction (FSI).
1. Introduction
The cavity filling modeling research already has accurately dynamic fluid behavior results on the
entering into a specific geometry; but the approach accuracy depends on the geometry complexity,
material properties defined, the quality and experimental information repeatability of for
comparison.
Using the model proposed by (Campbell, 1996), several researchers have conducted their own
analysis to determine the values range expected under exposed conditions, for filling time is about
2 s (Domanus, 1996; Usmani, 1995; Rigaut, 1996; Xu, 1996; Barkhudarov, 1996; Ohnaka, 1996),
(Layton, 1996), that matches the experimental time recorded, on the other hand the interaction
with the air inside the mold is negligible due to the porous nature of the mold and is considered
that the porosity does not affect the material flow. Finally, a system using water is able to compare

with system using molten metal and determine the correct fluid pouring speed due to kinematic
viscosity (Stefanescu, 1992).
The simulation of this phenomenon arises through an alternative CEL analysis that allows to
define in easy way all restrictions and avoid programming several conditions like contact type by
considering the simultaneous application of Lagrangian and Eulerian coordinates interaction (Fan,
2009). The main advantage of this is that eulerian mesh integrity is preserved during all
calculation because never is distortioned (Tippmann, 2009). To enforce the right interaction
between coordinates a ratio of minimum 3 Eulerian elements and maximum 5 Lagrangian must be
asegurate (Latorre, 2012).
Hence this method is applied to take advantage of the possibility to restore calculation from a
specific increment, cycle or iteration defined, hence this allows to watch the calculation progress
at very short time intervals and if necessary stop and correct the model avoiding to lose the
progress already made. Finally if a numerical error has occurred or equipment failure happened,
also is possible to restore de calculation from the last correct result (Alameda, 2010).
2. Methodology
2.1. Geometry
The solids for the simulation are: the standard cavity for code and software validation, Figure 1 (a)
front view and (b) right view; this is the reference piece that defines the fluid within the Eulerian
domain and ensure that the model considers the equivalent volume, 2 kg of molten aluminum
(Campbell, 1996) Figure 1 (c); pouring basin was simplified into a cylindrical section and a
truncated cone that leads smoothly to filling system geometry, Figure 1 (d); the three solids were
defined as a shell constrained with rigid body condition.
Domain is a 3D Eulerian solid type with dimensions such that enclose all solids assembly, also
limits the region where the Eulerian fluid will be calculated, has a L shape with depth 150 mm
Figure 1(e).

Figure 1. Dimensions of the parts used in the model a) validation standard cavity,
b) validation standard cavity right view, c) auxiliary part, d) pouring basin, e)
Eulerian domain. Measurement units in mm.

2.2. Material Properties.
The molten metal and water flow inside a cavity can be compared reliably because the two fluids
behavior in the casting conditions is incompressible 0

  0
, Newtonian and the kinematic
viscosities are similar (Stefanescu, 1992). Therefore the water properties used in the simulation are
shown in Table 1, these properties are used to satisfy the Mie-Grüneisen EOS and Hugoniot
model.
Table 1. Eulerian material Properties (Simulia, 2012).

Property Magnitude
𝜌0 998.2 kg/m3
𝜐 0.001003 Ns/m2
c0 1500 m/s
s 0
Γ0 0
2.3. Analysis conditions.

CEL analysis needs dynamic explicit steps, in order to allow efficient solution for nonlinearities
that occur in the material flow during the filling, using known position and velocity values to
obtain the next state of the system, which prevents the stiffness matrix formation and converge to
a solution using less calculation time (Alameda, 2010).
Variable mass scaling option is currently not supported in the CEL method; hence, the stable time
increment is controlled by the Eulerian element size and the material properties of the material
defined through EOS definitions. A good option for optimizing computational requirements is
scaling c0, which causes an acceptable compressibility in the system, if a c0 is divided by a
hundred the maximum compressibility calculated is 1.5% that is acceptable in this type of analysis
but must be evaluate for a particular system (Tippmann, 2009).
The filling is driven only by gravity acceleration set at 9.81 m/s2.
On the visualization module the first result is shown at twentieth of the total step time, in order to
simulate 6 s were needed 60 steps with 0.1 s long, this generates a resolution of 5e-3 s, to ensure
that the intervals shown in radiographs (Campbell, 1996) were displayed on the simulation.
The only restriction on the fluid is the solid geometry, so it is only required to define the
interactions between them. The mechanical type "Rough" induces non-slip condition and
simulates the effect of fluid adherence, which is assigned to the entire model by GC algorithm
(Simulia, 2012).
Rigid body constrain was applied to each solid defining an inertial behavior (Simulia, 2011) and
this allows the surface properties calculation, such as pressure or reaction force (Saucedo, 2013).
This was applied at each part reference point.

Figure 2. Solid pieces assembled with rigid body condition (yellow circles) and
applied ENCASTRE (red circles).
The normal zero velocity boundary condition on the Eulerian surfaces prevents lack of material
(Tippmann, 2009), thus ensuring that energy remains constant.
The Eulerian material was applied as an initial condition using a predefined field generated by the
volume fraction tool (VFT) and the reference part located in the pouring basin. Is necessary that
the domain and the reference part were meshed with the same element size to calculate which
elements are interfered and assign volume correctly.
The mesh control used a predefined element type EC3C8R with 3 mm size .The reference part
also has 3 mm, while the pouring basin has a 2.5 mm mesh size.
The cavity uses a global element size of 19.8 mm, which is refined in the front and back faces to
achieve the minimum ratio between domains (Fan, 2009). In the solid parts, it was used linear
element R3D4.

Figure 3. Cavity Mesh.
3. Results and Discussion

Velocity profile shows that the contact condition "Rough" induces fluid adhesion properly, Figure
4(a), element vectors in contact with both domains show where is possible to see change direction
as a result of non-slip condition, while in Figure 4(b) is possible to observe some vectors flowing
to the center, which is the expected direction for the material that have no contact with solid yet.

a) b)
Figure 4. a) Fluid flow through the pouring basin, b) instant of material flow toward
center.
The volume results and velocity profile are consistent together, in Image 5(a) two regions are
identified, one at bottom in full contact with the pouring basin (black) conical section and second
at top which is still in free fall (red), this effect is appreciated in vectors show in Figure 5(b)
almost all vectors are changing direction from down to up where contact start between solid and
fluid, and the bottom of the pouring basin the velocity has the biggest magnitude because this
section of the fluid have no contact with any solid in Figure 5(c) the volume profiles shown that
the preferential flow appears at center where is the expected zone and in Figure 5(d) is shown that
the velocity profile is uniform when the entire volume has contact with the solid, resulting in the
lowest value compared with instant in Figure 5(b). The above behavior shows that "Rough"
condition stops the material flow, but guarantee the FSI from material runoff and made greater the
minimum ratio needed for Eulerian and Lagrangian elements, as well as the contact material
Eulerian stops moving and spreads the effect by viscosity , this could approximate the oxide layer
formed in a common gravity cast.

a) b) c) d)
Figure 5. a) Contact with solid, b) velocity vectors showing the interaction between
the fluid and solid, c) volume profile of central flow tendency d) velocity profile
when total contact is achieve.
The fluid profile at pouring cup center is shown in Figure 6(a) is the natural sloshing behavior
produced by a normal impact on the fluid free surface. In Figure 6(b) a meniscus forming is shown
as a result of fluid displacement toward the free surface, meanwhile in Figure 6(c) is observed
volume accumulation in a column shape precipitating and forming the shape in Figure 6(d). The
previous results demonstrate that calculated profiles are comparable to known fluid behavior,
suggesting that volume distribution results are suitable.

a) b)
c) d)
Figure 6. Behavior of the material at the pouring basin free surface, a) initial wave
profile, b) meniscus formation, c) meniscus falling by gravity d) profile formed by
fluid meniscus drop.
Determining flow validity along the cavity filling is made by comparison with radiographs taken
by (Campbell, 1996) at intervals of 0.25 s, we selected the most representative orientation and
filling percentage.
For volume distribution analysis, Figure 7(a) shows that filling system and the gate to the main
cavity are completely full, as expected, but the fluid entering to the cavity shows discontinuities on
the left side as a result of direction, speed and inertia calculated at this moment. On the other hand
Figure 7 (b) shows all surfaces in complete contact and properly confined fluid, the main
accumulation appears at center and free surface profile shows a wave type, in Figure 7(c) has
completely filled the cavity, no discontinuities are observed.

Figure 7. Eulerian volume profiles obtained in the simulation a) discontinuity at the
beginning of the standard filling cavity, b) accumulation waveform, c) complete
filling mould.
The velocity profile is presented at the same moment that the volume profile, except for the last
which was earlier, in order to show a relative maximum recirculation rate. In Figure 8(a) the
velocity vector direction are result of the collision with the right wall of cavity gate, Figure 8 b the
fluid has the highest density of velocity vectors at the center, suggesting that the material flow is
stable at this moment and go almost vertically, while left accumulation suggests that induced flow
by collision with the gate to the cavity is still present. Figure 8(c) is the instant before the cavity is
completely full, at this moment is shown a vortex induced by the collision is still present.
a) b) c)
Figure 8. Velocities calculated by the software, a) discontinuity at the beginning of

the standard cavity filling, b) waveform with center accumulation c) full cavity with
vortex.
The simulation predict that the velocity magnitude is between 0.636 m/s and 1.273 m/s. To
determine the accuracy of this result a basic calculus was done using Torricelli's theorem
considering the filling system high and comparing the value with the predicted value at the filling

system bottom, the calculation obtained is 2.97 m/s and in Figure 9 the displayed result by the
software is between 2.5 and 3.33 m/s this range is close enough to the analytical calculation, the
minimum speed is result of viscosity consideration and no-slip boundary condition on the wall
which retards the flow and the highest velocity is result of fluid natural oscillations and the
solution method.
t = 1.92 (s) t = 2.18 (s)
a) b)
Figure 9. Filling system velocity profile.
Figure 8(c) shows how the inlet flow changes the material state inside the cavity, inducing two
vortex with opposite direction, this is an expected result because it was filled from below and is
also clear that the system enters in a stationary state showing average profile at any moment.
Figure 10. Vorticity moments before the full filling, the average speed shown is
1.66 m/s
Determining the profile validity obtained by the simulation is comparing with radiographs
(Campbell, 1996) at the instant where an approximate profile appear, Table 2.

Figure 11. Radiographs obtained during Campbell test (Campbell, 1996)
Table 2. Comparison between simulation and radiographs
Figure Similarities Differences

Figure 8
11
In 8 the flow
Flow like goes to the
t = 1.9 t = 0.74
fountain left and 10 to
the right
Volume
Fluid
accumulation
t = 2.97 t=1 accumulation
on the right
at center
side
Filling system In 8 partial
t = 5.42 t=2 completely filling and in
full 10 full cavity
The most remarkable difference is a delay on filling time as a result of contact interaction "Rough"
application.
Flow direction during cavity filling indicates that the filling in the simulation runs differently than
shown on radiographs direction Figure 11. But both behaviors shows a wave profile, suggesting
filling is cyclical in time, the flow go to the right and in the next increment goes to the left as a
result of the vortex formation throughout the filling this is clearly seen in Figure 10, so the
moment they compare are not coincident but in the wave profile are.
4. Conclusions

Filling the standardized cavity given by Campbell, implementing CEL method is suitably achieved
by FSI without edges runoff, using contact type "Rough" causing Eulerian material flow delay.
Velocity vector field calculated on Eulerian material which is not in contact with solids is very
accurate to real flow because at this zone just viscous properties and EOS are solved.
The profiles obtained are accurate according with the comparison with the experimental ones
indicating that the fluid dynamics and free surfaces recognition are acceptable, however due to
random behavior in fluid waves, turbulence and cycle behavior cavity filling should not be
compared directly, but by observing trendings and characteristics of vortex areas is possible the
comparison.
The velocity vectors, filling profiles and recognition surfaces along the entire calculation defines
CEL as a useful for getting coupled systems analyzed, showing each domain response, both the
solid and the fluid.
5. References
1. Ahmadzadeh M., Saranjam B., Hoseini-Fard. A., Binesh A.R.,"Numerical Simulation of

Sphere Water Entry Problem Using Eulerian–Lagrangian Method", Applied Mathematical
Modelling, 38(5-6): 1673-1684, 2014.
2. Campbell, J., Holliday, M., Sirrel, B., "Benchmark Testing the flow and solidification
modeling of Al castings", JOM (Journal of Materials), 48, pp 20-23, 1996.
3. Domanus H.M., Schmitt R.C., Chuzhoy L., Nastac L., "Computer Simulation Using CAPS of
an Aluminum Plate Casting," JOM, 48, pp. 947-954, 1996.
4. Usmani A.S., Cross J.T., Lewis, R.W., Ravindran, K., and Tran, V., "Finite Element
Simulation of the Bench Mark Mold Filling Problem," JOM, 48, pp. 955-962.
5. Rigaut C., Meyer H., Charbonnier A., Bourg A., "Round Robin Bench Mark Exercise for
Mold Filling and Solidification," in JOM, 48, pp. 1007-1014, 1996.
6. Xu, Z.A. and Mampaey, F., "Mold Filling and Solidification Simulation of the Bench Mark
Casting," JOM, 48, pp. 963-970, 1996.
7. Barkhudarov, M.R. and Hirt, C.W., "Casting Simulation: Mold Filling and Solidification—
Benchmark Calculations Using FLOW-3D," JOM, 48, pp. 935-946, 1996.
8. Ohnaka, I. and Zhu, J.D., "Computer Simulation of Fluid Flow and Heat Transfer of the
Bench Mark Test by 'DFDM/3DFLOW,' " JOM, pp. 971-974, 1996.
9. Layton M.A., Wang C.M., Cheng C., Yu K.O., Paul A.J., "Modeling of Mold Filling and
Solidification" JOM, 48, pp. 975-982, 1996.
10. Stefanescu D.M. (Volume Chair), "ASM Metals Handbook: Casting". ASM International®
The Materials Information Company, Vol. 15, 9° edition, 1992.
11. Fan, L. Peiran D. Steve S., "Coupled Fluid/Structure Interaction Simulation Using Abaqus
CEL", Dassault Systemes Simulia Corp, 2009.
12. Tippmann, J., Prasad, S., Shah, P., "2-D Tank Sloshing Using the Coupled Eulerian-
Lagrangian (CEL) Capability of ABAQUS®/Explicit", Dassault Systemes Simulia Corp,
2009.

13. Latorre, X., Marijuan, P., Viñas J., "Road Restraint Systems in ABAQUS®", IDIADA
Automotive Technology S.A., 2012.
14. Alameda E., "Modelado del proceso de forjado orbital mediante paquetería de elemento
finito", Tesis para obtener el título de Ingeniero Mecánico, Facultad de Ingeniería,
Universidad Nacional Autónoma de México, 2010
15. Simulia ABAQUS®6.12, "Eulerian analysis of a collapsing water column", ABAQUS®
Benchmarks Manual, section 1.7.1, 2012."
16. Simulia ABAQUS®6.12, "Specifying frictional behavior for mechanical contact property
options", ABAQUS®/CAE User´s Manual, section 15.14.1, 2012.
17. Simulia Abaqus Technology Brief, "Water Landing of Space Flight Re-entry Vehicles Using
Abaqus/Explicit", TB-11-SD-1, 2011
18. Saucedo, A., "Simulación de la colada gravitacional mediante paquetería de elemento finito",
Tesis de Licenciatura, Facultad de Ingeniería, Universidad Nacional Autónoma de México,
2013.
6. Acknowledgment
The authors would like to thank V. H. Jacobo, G. Álvarez, I. Cueva and E. Ramos, for technical
assistance during the elaboration of this study.
This work was supported by DGAPA, UNAM, under grants IN116512(2).

On the Numerical Implementation of a 3D Fractional
Viscoelastic Constitutive Model
G. Alotta1†, O. Barrera2, A.C.F. Cocks2, M. Di Paola3
1
Dipartimento di Ingegneria Civile, Ambientale, Aerospaziale, dei Materiali (DICAM), Università
degli Studi di Palermo, Viale delle Scienze Ed. 8, 90128 Palermo, Italy, currently visiting student
at the Department of Engineering Science, University of Oxford
2
Department of Engineering Science, University of Oxford, Parks Road, OX1 3PJ, Oxford, United
Kingdom
3
Dipartimento di Ingegneria Civile, Ambientale, Aerospaziale, dei Materiali (DICAM), Università
degli Studi di Palermo, Viale delle Scienze Ed. 8, 90128 Palermo, Italy
†Email: gioacchino.alotta@unipa.it, gioacchino.alotta@eng.ox.ac.uk
Abstract: The aim of this paper is the implementation of a 3D fraction al viscoelastic constitutive
law in a user material subroutine (UMAT) in the finite element software Abaqus. Essential to the
implementation of the model is access to the strain history at each Gauss point of each element in
a constructive manner. Details of the UMAT and comparison with some analytical results are
presented in order to show that the fractional viscoelastic constitutive law has been successfully
implemented.
Keywords: Fractional viscoelasticity, Constitutive model, Abaqus/Standard, Creep, Relaxation
1. Introduction
Real viscoelastic materials like rubbers, polymers, biological tissues, asphalt mixtures, soils etc
exhibit power law creep and relaxation behaviour (Nutting, 1921; Di Paola, 2014; Deseri, 2013;
Di Mino, 2014; Bagley, 1984). Relaxation and creep of this type of material has been modelled in
the scientific literature, mainly, by means of single and/or linear combinations of exponential
functions, in an attempt to capture the contributions of both solid and fluid phases. This approach
does not allow for a correct fit of experimental results. Power law creep and relaxation leads to
fractional viscoelastic constitutive models that are characterized by the presence of so-called
fractional derivatives and integrals, namely derivatives and integrals of non-integer order; when
the order of derivation (or integration) is integer, the fractional operators restore the classical
differential operators. The most interesting aspect of fractional operators is that they have a long
“fading” memory. In this context the term “hereditariness” is usually used in the sense that the
actual response in terms of stress/displacement depends on the previous stress/strain history. If a
relaxation or creep test is well fitted by a power law decay then the fractional constitutive law can
be directly derived. Such a constitutive law is defined by a small number of parameters to avoid
the conventional use of combinations of simple models which can require a much larger number
of parameters to capture both the creep and relaxation behaviour. The aim of this paper is to
describe the implementation of a 3D fractional viscoelastic constitutive model in a user material

routine (UMAT). The use of other subroutines is necessary to access the strain history at each
gauss point to evaluate the stress response.
2. Preliminary concepts
In this section we introduce some preliminary concepts on fractional viscoelasticity and fractional
differentiation and integration.
It is well known that a viscoelastic material can be characterized, for one dimensional problems,
by its Relaxation and Creep functions, R ( t ) and C ( t ) respectively. These functions describe the
behaviour of the material when a constant strain and a constant stress are applied, respectively.
Classical models are characterized by exponential type relaxation and creep functions. This
happens when viscoelastic materials are modelled by different combinations of elastic elements
(springs) and viscous elements (dashpots); the simplest models of this kind are Maxwell and
Kelvin-Voigt models in which a spring and a dashpot are in series and in parallel, respectively.
Although these models are able to describe the time-dependent behaviour of viscoelastic materials,
they fail to capture both the relaxation and the creep behaviour; for this reason more complicated
models, with combinations of springs and dashpots are used, but this leads to more complex creep
and relaxation functions and governing equations; furthermore these classical models are not able
to describe the long-time memory of real viscoelastic materials.
Creep and relaxation tests on real viscoelastic materials, such as polymers, rubbers, asphalt
mixtures, biological tissues, have shown that creep and relation tests are well fitted by power laws
of real order (Nutting, 1921; Deseri, 2013; Di Mino, 2014; Di Paola, 2014; Bagley, 1984) rather
than exponential functions. These functions can be written as follows:
 
C t t
R (t )  ; C (t )  (1a,b)
 1    C   1   
where   is the Euler gamma function. For viscoelastic materials 0    1.
It is well known that, within the framework of linear viscoelasticity, the Boltzmann superposition
principle is valid; this principle allows us to obtain the response of a material when the imposed
stress or strain history is not constant and can be expressed in two forms:
t t
 (t )   R ( t   )  ( ) d  ;  (t )   C ( t   ) ( ) d  (2a,b)
0 0
These integrals are often labelled as “hereditary” integrals, because the actual value of  ( t ) (or
 ( t ) ) depends on the previous history of  ( t ) (or  ( t ) ). By taking the Laplace transform of
Equations 2, an interesting relationship between the relaxation and creep functions is obtained in
Laplace domain:
1
Rˆ ( s ) Cˆ ( s )  (3)
2
s

where the ‘hat’ means Laplace transform and s  is the variable in the Laplace domain. This
implies that it is sufficient to perform a single creep or relaxation test to determine all the relevant
parameters of the viscoelastic model.
Substitution of Equations 1 into 2 leads to constitutive laws that involve fractional operators,
namely derivatives and integrals of real order (Podlubny, 1999). This is straightforward for the
case in which we apply a strain history (Equation 2a) and we want to evaluate the corresponding
stress history:
t
C
 (t )   (t   )

 ( ) d   C   C
0
D t   (t )

(4)
 (1   ) 0
In Equation 4  C0 Dt 

is the so called Caputo fractional derivative (Podlubny, 1999), which is a
convolution integral with a power law kernel. If we consider the case in which we apply a stress
history (Equation 2b), performing an integration by parts and after some manipulations we obtain
the Riemann-Liouville (RL) fractional integral  0 I t   (Podlubny, 1999):
t t
1 1 1
 (t )   (t   )

 ( ) d    (t   )
 1
 ( ) d    0

It   (t ) (5)
C   (1  ) 0
C  ( ) 0
C
These constitutive laws do not correspond to a springs or dashpot or a simple combination of

springs and dashpots and is generally referred to as a springpot. in the scientific literature (Scott
Blair, 1949).
Caputo’s fractional derivative and the Riemann-Liouville fractional integral are considered
integro-differential operators because all rules of integer order derivatives and integrals are still
valid (Podlubny, 1999). Moreover, when the value of  reaches the limit values of 0 and 1,
derivatives of order 0 and 1 are obtained. This illustrates a very important feature of these
equations: when   0 the fractional viscoelastic constitutive law of Equations 4 and 5 reduces
to the purely elastic (one-dimensional) Hooke’s law, while for   1 the fractional constitutive
law becomes the constitutive law of a dashpot. For this reason, as the fractional operators are
generalization of integro-differential operators of integer order, the constitutive law of the
springpot can be seen as a generalization of the constitutive laws of springs and dashpots. This
concept is summarized in Figure 1.

Figure 1. Spring, springpot and dashpot and related constitutive law
For numerical purposes there is a discrete version of the fractional derivative, namely the
Grünwald-Letnikov (GL) fractional derivative; after some manipulation the GL fractional
derivative can be written in this form:
N
 GL
0

Dt f  (t )  lim  t
t  0

  k f (t  k  t ) (6)
k 0
where  k are coefficients evaluated in a recursive way in the form:
k 1 
k   k 1 ; 0  1 (7)
k
The fractional derivative of Equation 6 becomes an integral if   0 . The summation of the GL

fractional derivative is a discrete convolution and for sufficiently small  t Equation 6 gives the
same results of the Caputo fractional derivative.
An important feature of fractional operators is that the fractional derivative (or integral) depends
on the past history of the function, hence they are able to describe the “fading” memory of real
viscoelastic materials.
In some cases it is necessary to introduce a more complete model to better represent the behaviour
of real viscoelastic materials. This model is the so-called fractional Kelvin-Voigt model and it is
constituted by a springpot in parallel with a spring with elastic modulus E (see Figure 2).

Figure 2. Fractional Kelvin-Voigt model
This ensures that the relaxation and creep responses asymptote towards constant stress and strain,
respectively, and reflect a more realistic material behaviour, especially when the stress tensor has
a hydrostatic component. For this model, the relaxation and creep laws are:

C t 1   E  
R (t )  E  ; C (t )  1  E   t  (8)
 1    E   C 
   
where E    is the one parameter Mittag-Leffler function defined as

 j
z
E z   (9)
j0  (  j  1)
In the next section the extension to a 3D fractional constitutive law is presented.
3. 3D fractional constitutive law

The constitutive model is obtained by means of a generalization of the elastic constitutive law
(Hooke Law); in that case only two parameters are required to define the whole stiffness (or
compliance) matrix for an isotropic material and these two parameters can be chosen as Young’s
modulus and the shear modulus, Young’s modulus and Poisson’s ratio, the Lamé constants, or
shear modulus and Bulk modulus. In this case we choose to write the stiffness matrix in terms of
shear modulus and Bulk modulus, for their clear physical meaning (deviatoric and volumetric part
of the stiffness). The terms of the stiffness matrix D can be written as follows:
 2 
D ijk h   K  G   ij  k h  G   ik  jh
  ih  jk  (10)
 3 
E
where K  is the Bulk modulus, G is the shear modulus,  ij is the Kronecker delta, E
3 (1  2 )
is Young’s modulus and  is Poisson’s ratio.

To generalize these elastic laws and obtain a fractional viscoelastic constitutive model, it is
sufficient to substitute the shear modulus and Bulk modulus with appropriate relaxation functions;
we choose to consider both the behaviour of the deviatoric part (shear relaxation function) and the
behaviour of the volumetric part (Bulk relaxation function) as given by the fractional Kelvin-
Voigt model:

G t
G (t )  G  
 (1   )

(11a,b)
K t
K (t )  K  
 (1   )
where G  a n d K  are the elastic parts of the deviatoric and volumetric relaxation functions,
respectively, while G  , G  ,  a n d  are parameters of the time varying parts of the deviatoric
and volumetric relaxation functions, respectively.
By assuming the relaxation functions in Equation 11, a six parameter mechanical model is
obtained; the model can be particularized in many ways, simply by changing values of these
parameters. The strain-stress relationship can be obtained simply by applying the Boltzmann
superposition principle:
t
σ (t )   R ( t   ) ε ( ) d  (12)
0
where R (t ) is the relaxation matrix and σ (t ) and ε (t ) are the stress and strain tensor,
respectively. The relaxation matrix R (t ) can be written in the same way as the stiffness matrix of
Equation 10, in which G is substituted with G (t ) of Equation 11a and K is substituted with
K (t ) of Equation 11b.
Since relaxation functions contain power laws, Equation 12 contains relationships that involve
fractional derivatives of order  and  . It is obvious that this model is also able to reproduce
Hooke’s law (for   0 and   0 ) , a generalized 3D Kelvin-Voigt model (for   1 and
  1) or a generalized 3D viscous law (for   1 ,   1 , G  0 and K  0 ).
In order to obtain the inverse relationship of Equation 12 we need to obtain the creep matrix C (t )
by using Equation 3. C ( t ) is evaluated by performing a Laplace transformation of the relaxation
matrix (that is Laplace transforms element by element) and evaluating its inverse:
Rˆ
1
(s)
Cˆ ( s )  2
(13)
s
4. Numerical implementation of the 3D fractional constitutive law

The constitutive model in equation 12 has been implemented in a user material subroutine UMAT
in Abaqus/Standard. The subroutine calculates the increment of stress at the end of each increment
and the Jacobian.

When the UMAT is called, the following information is available as ‘Input’: the stress at the
beginning of the increment, the strain at the beginning of the increment and the increment of
strain. We also need to have access to the history of strain i.e. the values of strain in all the
increments of the analysis. We will give some details of how this is achieved in the next
paragraph.
We start evaluating the direct component of stress  11 at the end of increment k , which is the first
component of the stress vector in Equation 12:

 k 1
( )
 4   22 ,k 1   33,k 1 
 1 1, k  1  K   V , k  1  K   t    j  V ,k  j  2  
3
G    1 1, k  1 
2

 j 1   
(14)
4 
k 1
( )
  22 ,k  j  2   33,k  j  2 

3
G  t  j   1 1, k  j2

2

j 1  
where  V   1 1   2 2   3 3 is the volumetric strain. The stress at the beginning of the increment
(which is known when the UMAT is called) can be written as follow:

 k
( )
 4   22 ,k   33,k 
 1 1, k  K   V , k  K   t    j  V ,k  j 1  
3
G    1 1, k 
2

 j 1   
(15)
4 
k
( )
  22 ,k  j 1   33,k  j 1 

3
G t  j   1 1, k  j 1

2

j 1  
We obtain the increment of stress by evaluating the difference between Equations 14 and 15 that
will be useful to calculate the Jacobian:

 k ( ) ( )
 4    22 ,k 1    33,k 1 
  1 1, k  1  K    V , k  1  K   t    j   V , k  j  2   k  1  V ,1   G     1 1 , k  1  
 j 1  3  2 
4 
 k ( )
   22 ,k  j  2    33,k  j  2  ( )
  2 2 ,1   3 3 ,1  
 G  t    j    1 1, k  j  2     k  1   1 1 ,1  
3  j  1  2   2  
(16)
where   i ,k  j  2   i ,k  j  2   i ,k  j 1 , with i  1 1, 2 2, 3 3 , and   V ,k 1   V ,k 1   V ,k . In the UMAT,
we code Equation 14; the other direct components of stress can be obtained simply by rotating
indices of the strain components.
To evaluate all three direct components of the increment of stress, we define a scalar quantity
related to the volumetric deformation (TERMV) and a three components vector(TERMK) each
related to one direct component of strain:
 k

T E R M V  K    V ,k 1  K   t

  j
( )
  V ,k  j2
( )
  k  1  V ,1  (17)
 j 1 
4 4  k

TERM S j
 G jj , k  1
 G t

  j
( )
 jj , k  j  2
( )
  k 1  jj ,1  (18)
3 3  j 1 

With these two quantities we are able to evaluate the direct component of stress as follows:
1  
 ll
 TERM V  TERM Sl   TERM S j
 TERM Sl  (19)
2  j 
In an analogous way it is possible to compute of shear components of stress and their increments:
k 1
 lm , k  1  G   lm , k  1  G   t


( )
 j  lm , k  j  2 (20a)
j 1
 k

 lm , k  1
 G    lm , k  1  G   t

  j
( ) ( )
  lm , k  j  2   k  1  lm ,1  (20b)
 j 1 
Where   lm , k  j  2   lm , k  j  2   lm , k  j  1 and   lm , k  1   lm , k  1   lm , k , with lm  1 2,1 3, 2 3 . These

terms can be computed directly one by one.
At this point, from Equation 16 and 20b, we can evaluate the components of the Jacobian as:
  4 4
(21)
jj , k  1  ( )  ( ) ( ) ( )
 K   K  t 1  G  G  t 1 ; 1  1 1
  jj , k  1
3 3
  2 2
(22)
jj , k  1  ( )  ( ) ( ) ( )
 K   K  t 1  G  G t 1 ; 1  1 1
   ll , k  1 3 3
  jl , k  1
 G   G  t

1
( )
(23)
  jl , k  1
To code this terms we need two quantities; one related to the volumetric relaxation function
(TERM1) and the other related to the deviatoric function (TERM2)
TERM 1  K   K  t

(24)
T E R M 2  G   G  t

(25)
and then
  4
jj , k  1
 TERM 1  TERM 2 (26)
  jj , k  1
3
  2
(27)
jj , k  1
 TERM 1  TERM 2
   ll , k  1 3
  jl , k  1
 TERM 2 (28)
  jl , k  1
The main issue in the implementation of the fractional viscoelasticity law is that we need to have
access to the history of strains in order to obtain the increment of stress. To overcome this problem

we store the values of the components of strain at each increment in a COMMONBLOCK. that
keeps track of these values when the UMAT is called.
Storing the history of strain of all integration points of all elements can potentially lead to a huge
memory requirement when running large models.subjected to long histories of loads. In order to
tackle this problem we ran a model with 31000 linear 3D elements C3D8 for a load history of 10
sec, with  t  0 .1 sec . This analysis gave us some information about(1)the RAM memory usage
during the analysis which was of about 4 GBs; (2) the results file (.odb) containing all the stress
and strain histories for each integration point and all displacements at each node is of about 1.2
GBs; the analysis ran for about 12 minutes. Although these are not small numbers, in the author’s
opinion they are still acceptable for the capability of modern computers.
5. Testing of the UMAT

In order to validate the UMAT a number of simple tests have been carried out for which we are
able to evaluate the analytical solutions.
Here we show a viscoelastic cube (Figure 3) subjected to creep and relaxation tests. The
mechanical properties of the cube are: K   5 x 1 0 8 P a s e c  , K   1 0 9 P a , G   3 .7 5 x1 0 8 P a s e c  ,
8
G   7 .5 x 1 0 P a ,     0 .3 .
In the creep test of Figure 3a the cube has one of its faces normal to the direction x constrained
to prevent motion in the x direction. On the opposite face a uniform and constant tensile stress
 x x    1 0 M P a is applied in the x direction.
Figure 3. Viscoelastic cube for the creep test (a) and relaxation test (b).
The analytical solution is obtained by using Equation 13 and the convolution of Equation 5 as
follows:

   G     K 
 xx ( t )  1  E  

t

  1  E  

t

 (29)
3G   G  9K   K 
         
   G     K 
 yy ( t )   zz ( t )   1  E  

t

  1  E  

t

 (30)
6G   G  9K  K 
         
Figure 4 shows a comparison between the Abaqus/Standard result and the analytical solution of
Equations (29) and (30); red dashed lines are responses evaluated with Abaqus/Standard with a
constant time step of 0 .1 sec , while black continuous lines represent the solutions of Equations
(29) and (30). From this figure it is possible to appreciate that the numerical procedure reproduces
the analytical results.
In the relaxation test in Figure 3b all of the faces of the cube but one are fixed only in the normal
direction. We then apply a displacement of 1 mm to the free face (normal to the x direction),
which corresponds to a strain  x x    0 .0 1  1 % (see Figure 5); the displacement was applied
with a linear ramp of 1 sec and then held for another 9 secs, as shown in Figure 5. With the
Figure 4. Comparison between analytical and Abaqus/Standard responses for

creep test of the cube in Figure 3.
boundary conditions described above,  y y ( t )   zz ( t )  0 and all the direct components of stress are
different from zero. The history of the superimposed strain can be written as follows:
 x x ( t )    t  U ( t )  U ( t  1)   U ( t  1)  (31)
where U ( ) is the Unit-step function; then by inserting Equation (31) in Equation (12) we obtain:

 xx

3  2   
 t   4 G 
 3K  t 
  4G   3 K     2    

(32)
 t  1 

  t  1   4 G   3 K    t  1   4G   3K     2     U



 yy
  zz

3  2   
 t   2 G 
 3K  t 
  2G   3 K     2    

(33)
 t  1 

  t  1    2 G   3 K    t  1   2G   3K     2    U
 
Figure 5. Strain history during the relaxation test

The analytical solution given in Equations (32) and (33) have been compared with the solution
obtained by running the UMAT and it can be seen in Figure 6 that the analytical results are
reproduced by the numerical calculations .
Figure 6. Comparison between analytical and Abaqus/Standard responses for

relaxation test of the cube in Figure 5.
A 2D plain strain model of a viscoelastic Euler-Bernoulli beam (Figure 7) under a uniformly
distributed load and constant over the time has been analysed. The beam is 5 m long, has a
rectangular cross section with base 10 cm and height of 20 cm and the same material properties
used of the previous examples; the beam is simply supported. Figure 7 shows the Abaqus model
of the beam modelled with 100 4-noded plain strain element (CPE4) with dimension 10x10 cm.

Figure 7. 2D plain strain model of the viscoelastic beam
Displacements of two points of the beam were monitored, located at 1 m (A) and 2 m (B) from the
left end of the beam, respectively; Abaqus results have been compared with analytical results
evaluated with the same approach of (Di Paola, 2013) as shown in Figure 8.
Figure 8. Comparison between analytical and Abaqus results for displacements of

two points of the beam
The analytical and Abaqus results are in very good agreement; a contour plot of  11 (stress in x
direction) is also reported in the deformed configuration in Figure 9.
Figure 9. Contour plot of longitudinal normal stress in the deformed configuration

of the beam. Values of stress in Pa
A number of other tests, such as creep tests with initial ramps and relaxation tests with a range of
different boundary conditions have been carried out and they all give the same responses as the
analytical solutions. These results are discussed elsewhere.

6. Conclusions
In this paper the implementation of a fractional viscoelastic constitutive model in
Abaqus/Standard has been presented. This constitutive model is derived by fitting creep or
relaxation data of real materials like rubbers, polymers, asphalt mixtures, biological tissues, and
many others and the application of the Boltzmann superposition principle (linear viscoelasticity).
Moreover it has the feature that the actual strain (or stress) at a point of the solid depends on the
previous history of stress (or strain) and not only on the current stress (or strain). In order to access
the history of strain, the components of strain at each increment are stored in a
COMMONBLOCK. Comparison between analytical results and results obtained with the UMAT
shows the accuracy of the implementation of the fractional viscoelastic model presented here.
We believe that this novel fractional viscoelastic model simulates the viscoelastic behaviour of
real material in a more realistic way. Examples of the use of the UMAT to solve a range of
practical engineering problems are presented elsewhere.
7. References
1. Bagley, R.L., Torvik, P. J., “On the appearance of the fractional derivative in the behaviour of
real materials”, Journal of Applied Mechanics, no. 51, pp. 294-298, 1984.
2. Deseri, L., Di Paola, M., Pollaci, P., Zingales, M., “Power law hereditariness of hierarchical
fractal bones”, International Journal for Numerical Method in Biomedical Engineering, no 29
(12), pp. 1338-1360, 2013.
3. Di Mino, G., Airey, G., Di Paola, M., Pinnola, F. P., D’Angelo, G., Lo Presti, D., “Linear and
non linear fractional hereditary constitutive laws of asphalt mixture”, Journal of Civile
Engineering and Managements, doi:10.3846/13923730.2014.914104, 2014.
4. Di Paola, M., Fiore, V., Pinnola, F. P., Valenza, A., “On the influence of the initial ramp for a
correct definition of the parameters of fractional viscoelastic materials”, Mechanics of
Materials, no. 69 (1), pp. 63-70, 2014.
5. Di Paola, M., Heuer, R., Pirrotta, A., “Fractional visco-elastic Euler-Bernoulli beam”,
International Journal of Solids and Structures”, no. 50, pp. 3505-3510, 2013.
6. Nutting, P. G., “A new general law of deformation”, Journal of the Franklin Institute, no. 191,
pp.679-685, 1921.
7. Podlubny, I., Fractional Differential Equations, Academic Press, 1999.
8. Scott Blair, G. W., Caffyn, J. E., “An application of the theory of quasi-properties to the
treatment of anomalous stain-stress relations”, The Philosophical Magazine, no. 40 (300), pp.
679-685,1949.

Hydrogen Embrittlement Mechanisms in Metals: a
Modelling Approach
O. Barrera1†, E. Tarleton2 and A.C.F. Cocks1
1
Department of Engineering Science, University of Oxford, Parks Road, OX1 3PJ, Oxford, United
Kingdom
2
Department of Materials, University of Oxford, Parks Road, OX1 3PH, Oxford, United Kingdom
†Email: olga.barrera@eng.ox.ac.uk
Abstract: It is well known that high strength steels are tremendously affected by hydrogen.
The aim here is to provide a modelling of the HELP (Hydrogen Enhanced Local Plasticity)
mechanism fully coupled with the hydrogen transport equation. The hydrogen diffusion equation is
implemented in a UMATHT subroutine by considering similarities between the heat and mass
diffusion equations. A coupled temperature-displacement procedure has been adopted to allow the
coupling between hydrogen diffusion and the mechanical behaviour of the material.
Keywords: Hydrogen embrittlement, Constitutive models, hydrogen diffusion equation, UMATHT.
1. Introduction
Experimental studies and fractography analysis of high strength steels indicate a deleterious
influence of hydrogen on the constitutive response. The idea is to model combined HELP
(Hydrogen Enhanced Local Plasticity) and HID (Hydrogen Induced Decohesion) mechanisms
(Sofronis, 1989; Robertson, 1999; Robertson, 1986) in which the hydrogen content is evaluated by
solving the hydrogen diffusion equation (McNabb, 1963; Oriani, 1978). Previously we have
analysed a carbides-rich region in dissimilar weld (Barrera, 2013; Barrera, 2014). An ABAQUS
finite element model of the microstructure was created by converting a TEM image.
The presence of fine carbides plays an important role in the constitutive response of these
materials. Simulations of the response of similar materials show that in regions where the
hydrogen content is high the matrix surrounding the carbides softens and plastic flow is localized.
Moreover, the presence of hydrogen lowers the cohesive strength, giving rise to microcrack
formation at the carbide- matrix interface, leading to microvoid formation. As deformation
proceeds the pores enlarge and link to form cracks, which generates the failure surface. The
interface between the carbides and the surrounding metal matrix is represented by cohesive
elements. A user cohesive element (UEL) has been coded incorporating a traction separation law

which is a function of local hydrogen content and of plastic strain which also accounts for the
development of dislocation structures at the carbides. The behaviour of the matrix was modelled
by a hardening law that depends on the local hydrogen content via a UMAT, and a simple linear
distribution of hydrogen was used. Here we ignore the presence of the carbides. We then model
the constitutive behaviour of the matrix whose hardening law is function of the hydrogen content
evaluated by solving the diffusion equation. The hydrogen diffusion equation is implemented in a
UMATHT subroutine by considering similarities between heat and mass diffusion equations. A
coupled temperature-displacement procedure has been adopted to allow the coupling between
hydrogen diffusion and the mechanical behaviour of the material.
2. Hydrogen diffusion equation

It is well known that hydrogen being the smallest element diffuses more easily through metals
than any other element. Even at low concentration hydrogen often leads to the embrittlement of
metals for reasons that are not yet well understood but are certainly related to the speed at which it
can diffuse to highly stressed regions.
Hydrogen atoms move through the metals by normal interstitial site (NILS) diffusion or
dislocation transport. Hydrogen atoms reside either at NILS or at trapping sites such as:
dislocations, grain boundaries, carbide/matrix interface, microvoids and other defects (see Figure
1). The vast majority of sites are the normal sites NILS and the minor fraction of the sites are the
trapping sites. Here we consider a lattice consisting of two kinds of sites for occupancy by
hydrogen. The hydrogen atoms in the normal sites are denoted CL and the hydrogen atoms
trapped at the dislocations are denoted CT . Hydrogen trapping at defects has a large effect on its
diffusion in solids of solute. Hydrogen diffuses so easily that even shallow traps can produce a
significant effect on the diffusivity. One of the first theories on the mobility of dissolved hydrogen
in an iron lattice containing trapping sites was given by McNabb and Foster (McNabb, 1963).
They introduced a diffusion equation solved with terms for sources and sinks. Oriani (Oriani,
1978) reformulated the work by McNabb and Foster introducing the assumption of local
equilibrium between the mobile and the trapped populations for a restricted domain of degree of
trap coverage. Mass conservation states that the changing rate of the total hydrogen in the volume
Ω is equal to the flux through the surface Γ:
∂
( CL + CT ) d Ω + ∫ J ⋅ ndS = 0
∂t Ω∫
(1.1)
∂Ω
The driving force for diffusion is the chemical potential gradient:
DC
− L L ∇µ
J= (1.2)
RT

Where DL is the diffusion coefficient of hydrogen and μ is the chemical potential defined as
follows:
µ=
µ0 + RT ln CL + µσ (1.3)
µ0 denotes the chemical potential at standard condition. µσ Is the chemical potential function of
stress:
σ kk
µσ = − VH (1.4)
3
σ kk
and σH = the hydrostatic stress and VH the is the partial volume of hydrogen in solid
3
solution. Substituting eq. (1.3) into eq. (1.2) we can express the flux as follows:
DL CL V H
=J ∇σ H − DL ∇CL (1.5)
RT
Now substituting eq. (1.5) into eq. (1.1) we obtain the following equation:
∂  DL CL V H 
( )
∂t Ω∫ ∫∂Ω  RT
C L + CT d Ω +  ∇ σ H − DL ∇ C L  ⋅ ndS = 0 (1.6)

By applying the divergence theorem eq. (1.6) becomes:
∂CL ∂CT D C VH 
+ − ∇ ( DL ∇CL ) + ∇  L L ∇σ H  = 0 (1.7)
∂t ∂t  RT 
Now Oriani’s theory assumes that the trapped hydrogen population is in equilibrium with the
population upon normal lattice sites not only in the static case but also during diffusion. We
denote the fraction of the available sites occupied by the population in NILS as ϑL and ϑT the
site fraction occupied by the trapped population. The equilibrium between the two populations can

αL ϑL
be described by introducing the equilibrium constant K given by K= where αL =
αT 1 − ϑL
ϑT
and αT = are the activities of the hydrogen upon the normal lattice sites and the trapping
1 − ϑT
sites. Considering that the situation that Oriani considered was that ϑL  1 it is possible to
express the equilibrium constant K as follows:
1 ϑT
K= (1.8)
ϑL 1 − ϑT
Now CL ϑ=
= L N L , CT ϑT NT , where N L , NT are the numbers of atoms in NILS and in the
traps respectively.
It can be derived that:
∂CT CT (1 − ϑL )
= (1.9)
∂CL CL
Sofronis and McMeeking (Sofronis, 1989) formulated the hydrogen transport problem coupled
with large deformation elastic-plastic behaviour of a material based on Oriani's theory. In
(Sofronis, 1989) the authors incorporated the effect of hydrostatic stress and trapping site. In this
model they assume that hydrogen atoms diffuse through lattice sites and that trap sites are filled by
lattice diffusion. These trap sites are formed due to plastic deformation. Krom et al. (Krom, 1999)
demonstrated that the hydrogen transport model proposed in (Sofronis, 1989) does not provide a
correct hydrogen balance. Hence in (Krom, 1999) a modification on the hydrogen diffusion model
is introduced which includes a factor depending on the strain rate. The strain rate factor decreases
the hydrogen concentration in lattice sites due to the filling of trap sites. As results of this the
modified hydrogen transport model predicts a strong dependence of the hydrogen concentration in
lattice sites on the strain rate, instead the hydrogen concentration in the trap sites is not
significantly affected.
CT = ϑT NT ( ε p ) (1.10)
∂CT
Considering eqns. (1.9) and (1.10), the term in eq.(1.7) can be expressed as follows:
∂t

∂CT ∂CT ∂CL ∂CT ∂NT ∂ε P
= + (1.11)
∂t ∂CL ∂t ∂NT ∂ε P ∂t
Substituting eq.(1.9) into eq. (1.11) we obtain:
∂CT CT (1 − ϑL ) ∂CL ∂CT ∂NT ∂ε P

= + (1.12)
∂t CL ∂t ∂NT ∂ε P ∂t
Substituting eq. (1.12) into eq. (1.7) we obtain the hydrogen diffusion equation finalized by
(Krom, 1999) :
CL + CT (1 − ϑT ) ∂CL D C VH  ∂NT ∂ε P
− ∇ ⋅ ( DL ∇CL ) + ∇ ⋅  L L ∇σ H  + ϑT =0
CL ∂t  RT  ∂ε P ∂t
(1.13)
CL + CT (1 − ϑT )
Where Deff = represents the effective diffusivity.
CL
Eq.(1.13) has been implemented in a UMATHT by using the similarity between the diffusion
equation and the heat equation. This is shown in the next paragraph.
Figure.1. Hydrogen atoms reside either at NILS or at trapping sites such as: dislocations, grain
boundaries, carbide/matrix interface, microvoids and other defects

3. Analogy with the heat equation
In order to implement the coupled hydrogen transport equation given in equation (1.13) we use the
similarities with the heat equation as done by Chang-Sik OH et al. (OH, 2010). The coupled
thermo-mechanics procedure (available in Abaqus) is used to solve coupled diffusion-mechanics
problems. In Chang-Sik OH et al. (OH, 2010) the evaluation of the gradient of hydrostatic stress
inside the element was calculated by storing the component of hydrostatic stress (evaluated at each
gauss point) in an external file that was then read inside the UMAT and passed to the UMATHT.
Here we show that we can compute directly the gradient of hydrostatic stress inside the UMATHT
by using the subroutines USDFLD and GETVRM avoiding other cumbersome steps. Here we
report all the details necessary to fully understand the procedure. Table 1 summarizes the analogy
between the heat equation and the hydrogen diffusion equation. It can be seen that the form of the
equations are similar. The degree of freedom in the heat equation is temperature: T whereas in the
mass diffusion equation is the concentration of hydrogen at NILS CL . We can write the
expressions of the terms appearing in the mass diffusion equation (1.13) as shown in Table 1.
Table 1. Analogy between heat equation and diffusion equation.
HEAT EQUATION MASS DIFFUSION EQUATION

∂T  ∂C 
ρ cP + ∇ J q + rq = 0 + ∇ J m + rm = 0
∂t ∂t
Thermal energy per unit mass Chemical potential per unit mass
U q = cP
∂T
 ∂C ∂ ( CL + CT )
ρU= =
∂t m
∂t ∂t
Degree of freedom: temperature T
Degree of freedom: concentration CL

Heat flux: Jq  DL CL V H  
=
hydrogen flux: Jm ∇σ H − DL ∇CL
RT
Heat source: rq ( assume rq = 0) Hydrogen source: rm ( assume rm = 0 )
3.1 Implementation in UMATHT subroutine

Subroutine UMATHT is usually used to define the thermal constitutive behaviour of the material
and internal heat generation during heat transfer processes. In this paper a UMATHT is used to
implement the hydrogen diffusion equation. The quantities that are defined in a UMATHT are
summarized in Table 2. It is essential to note that the variation of internal energy with temperature

∂U q ∂U q
and with spatial temperature gradient i.e. ; respectively in the heat procedure
∂T  ∂T 
∂ 
 ∂x 
correspond to the variation of chemical potential with hydrogen concentration at NILS CL and
∂U m ∂U m
with trap density gradient function of plastic strain εP in equation (1.10) i.e. ;
∂CL  ∂N 
∂ T 
 ∂ε P 
respectively. Moreover the variation of heat flux with temperature and with spatial temperature
 
∂J q ∂J q
gradient i.e. ; respectively corresponds to the variation of internal energy with
∂T  ∂T 
∂ 
 ∂x 
hydrogen concentration at NILS CL and with the hydrogen concentration gradient i.e.
 
∂J m ∂J m
; . The hydrogen diffusion equations can be written using the formalism
∂CL ∂ ( ∇CL )
appropriate for the subroutine as follows:
 
∂ρU m ∂ρU m dNT ∂J m ∂J m
ρU m ( t=
+ ∆t ) ρU m ( t ) + dCL + dε p + dCL − d ∇C L
∂CL ∂NT d ε p ∂CL ∂∇CL
(1.14)
The terms of equation (1.14) can be obtained by manipulation equations (1.9-1.12) as follows:
∂ρU m NT KT N L
= Deff = 1 +
( KT C L + N L )
2
∂CL
∂ρU m KT C L
=
∂NT KT C L + N L
 (1.15)
∂J m DVH 
= ∇σ H
∂CL RT

∂J m
= − DI
∂ ( ∇C L )
Where I is the identity matrix.


It is essential to note the in order to calculate the gradient of hydrostatic stress ∇σ H in equation
(1.13) we need to use a USDFLD subroutine. USDFLD allows as defining field variables at a
material point as functions of time or of any of the available material point quantities in this case
the direct components of stress: σ 11 , σ 22 , σ 33 . USDFLD calls the utility routine i.e. GETVRM
to provide access to the values of the material point quantities at the start of the increment. In
order to evaluate the gradient of hydrostatic stress we interpolate the hydrostatic stress inside the
element at each gauss point. The interpolation is identical to the procedure used to calculate the
displacement field from the nodes of the element. We use a common block to store the nodal
values, which enables the gradient to be calculated without needing to resort to a UEL. However 4
– node coupled temperature-displacement plane strain elements CPE4T cannot be used for this
purpose because the strain operator implemented in Abaqus for linear plane strain elements
provides constant volumetric strain throughout the element, therefore the gradient of hydrostatic
stress within the element is zero. Although linear plane stress elements could be used we prefer to
use plane strain elements as the UMAT can more readily be extended to 3D using a plane strain
formulation. Therefore we use quadratic 8 – node coupled temperature-displacement plane strain
elements CPE8T.
Table 2. Quantities that need to be coded in the UMATHT.
HEAT EQUATION MASS DIFFUSION EQUATION

Quantity to be defined in the UMATHT Quantity to be defined in the UMATHT
   
∂U q ∂U q ∂J q∂J q ∂U m ∂U m ∂J m ∂J m
; , ; ; , ;
∂T  ∂T  ∂T  ∂T  ∂CL  ∂N  ∂CL ∂ ( ∇CL )
∂  ∂  ∂ T 
 ∂x   ∂x   ∂ε P 
4. Numerical results
The example in Figure 2 illustrates a plate with the hole subjected to a displacement field and
hydrogen concentration C = 1 at the edge of the hole. C is the hydrogen concentration in the
lattice calculated as number of hydrogen atoms per unit volume normalized with respect to the
maximum number of interstitial hydrogen atoms. C varies between 0 and 1. C = 1 means that all
the interstitial sites are occupied by hydrogen atoms. We consider a quarter of the plate with
symmetric boundary conditions as shown in figure 2. We also impose no-flux on the bottom side.
A description of the effect of the hydrogen content on the flow stress in the matrix is given as
follows [1-3]:

1
 ε n
σ Y σ 0H 1 + P 
= (1.16)
 ε0 
Where σ 0H = Ψ ( CL ) σ 0 is the initial yield strength in the presence of hydrogen that decreases
with increasing hydrogen concentration. Ψ ( CL ) is a monotonically decreasing function of
hydrogen concentration at NILS. ε0 is the initial yield strain in the absence of hydrogen, ε P is the
plastic strain and n is the hardening exponent which is considered not to be affected by hydrogen.
For simplicity, we will ignore swelling due to introduction of hydrogen into a material. Equation
(1.16) has been implemented in a UMAT subroutine within ABAQUS (ref). Equation (1.16)
models the effect of hydrogen-induced material softening, this is to be viewed as an attempt to
describe the experimental observations of the effect of hydrogen on dislocation mobility (Sofronis,
1989; Robertson, 1999; Robertson, 1986). Figure 3 shows the distribution of the concentration of
hydrogen in the plate. Hydrogen is localized in the area where the hydrostatic stress is high such at
the edge of the hole. Figure 4 illustrates how the initial yield strength is affected by the presence of
hydrogen σ 0 = Ψ ( CL ) σ 0 . In the regions where the hydrogen content is high (i.e. at the edge of
H
the hole) the value of the flow stress is considerably reduced by 20% .The presence of hydrogen
promotes softening of the material and localization of plastic flow (HELP). Figure 5 show the
distribution of plastic strain that is localized in the area where (a) the concentration of stress in
higher and (b) hydrogen concentration reach its maximum value.
Figure 2. Plate with the hole subjected to a displacement field and hydrogen
concentration at the edge of the hole

Figure 3. Distribution of the concentration of hydrogen inside the plate, max concentration of hydrogen
is localized at the edge of the hole.
σ 0H
Figure 4. Distribution of the function Ψ ( CL ) =inside the plate.
σ0

Figure 5. Distribution of plastic strain inside the plate.
5. Conclusions
This paper deals with coupling hydrogen diffusion equation with the plastic response of high
strength steel. We show a strategy in order to implement the hydrogen diffusion equation in a
UMATH in which we also calculate the component of the gradient of hydrostatic stress needed to
solve the hydrogen diffusion equation. This model can be used to study the HELP (Hydrogen
Enhanced Local Plasticity) mechanism.
Acknowledgment
This work was supported by the Engineering and Physical Sciences Research Council
[Programme grant number EP/L014742/1].
6. References
1. Barrera O., Cocks A.C.F., ”Computational modelling of hydrogen embrittlement in welded
structures”, Philosophical Magazine, DOI:10.1080/14786435.2013.785638.
2. Barrera O., Tarleton E., Cocks A.C.F.,”A micromechanical image based

model for the featureless zone of a Fe–Ni dissimilar weld”, Philosophical Magazine,
DOI:10.1080/14786435.2014.886023
3. Krom, A.H.M., Koers, R.W.J., Bakker, A., “Hydrogen transport near a blunting crack tip”,
Journal of the Mechanics and Physic of Solids, no. 47, pp. 971-992, 1999.
4. Liang, Y., Sofronis, P.,”Toward a Phenomenological Description of Hydrogen-Induced
Decohesion at Particle/Matrix Interfaces”, Journal of the Mechanics and Physic of Solids, no.
51, pp. 1509-1531, 2003.
5. McNabb A.,_Foster P.K., “A new analysis of the diffusion of hydrogen in iron and ferritic
steels, Journal of Transaction of the Metallurgical Society of AIME, no. 227, pp. 618-627,
1963.
6. OH C., Kim Y., Yoon K., “Coupled analysis of hydrogen transport using ABAQUS”, Journal
of solid mechanics and material engineering, no.7, pp,908-917, 2010.
7. Oriani, R.A.,”Hydrogen embrittlement of steels”, Annual reviews in Materials Science, no.8,
pp327-357, 1978.
8. Robertson, I.M., “The effect of hydrogen on dislocation dynamics”, Engineering fracture
mechanics, no. 64, pp. 649-673, 1999.
9. Robertson, I.M., Birnbaum, H.K., “An HVEM study of hydrogen effects on the deformation
and fracture of nickel”, Acta Metallurgica, no. 34, pp.353-366, 1986.
10. Sofronis P., McMeeking R.M., “Numerical analysis of hydrogen transport near a blunting
crack tip”, Journal of the Mechanics and Physic of Solids, no. 37, pp.317-350, 1989.

LIFE SCIENCES

Developing a Constitutive Model for a Highly Deformable Latex
Rubber with Implementations in Lagrangian Frameworks
Christopher A. Basciano1, Marcus Rademacher1, Patrick Downie1, Anita Bestelmeyer1,
Brian Croop2, Jorgen Bergstrom3
1
BD Corporate CAE, 21 Davis Drive, Research Triangle Park, NC 27709
2
DatapointLabs, 95 Brown Rd. #102, Ithaca, NY 14850
3
Veryst Engineering LLC, 47 Kearney Rd, Needham Heights, MA 02494
Abstract: To improve product design and enhance the realism of BD’s advanced finite element
models, the development of a constitutive model capable of predicting mechanical behavior at
high deformations was initiated for a commercially available latex rubber. A set of baseline
experiments was conducted to determine the isothermal mechanical response(s) of the latex
rubber. The experiments were designed to provide fundamental insight into any viscoelastic,
hyperelastic, and Mullins effects exhibited by the material. A constitutive model was constructed
for specific load cases from the aforementioned experimental data. Attention was given to the
model’s primary use case, numerical stability, and its ability to perform at high deformation
ranges. A second set of experiments was then conducted, aimed at characterizing the failure of
the rubber and providing a set of experimental data to validate the material’s constitutive model.
The first round of experiments revealed that the material’s stress vs. strain curves held limited
dependence on the applied loading rate, while the failure stress and strain had a moderate
dependence. Additionally, cyclic loading and unloading of the material was shown to have a
prominent effect on the failure stress and strain. Based on the model’s intended use case and
baseline mechanical response, a hyperelastic constitutive model was calibrated to the monotonic,
uniaxial tension data. The model was then implemented in Lagrangian and Eulerian geometries
and compared against the second set of experimental data. The constitutive model exhibits good
agreement to the calibration data when strain is less than 5.0, but begins to significantly
underpredict the stress values at strains above 6.0. Validation studies revealed good agreement
between the experimental tear test data and different implementations of the constitutive model
with different boundary conditions. Work is ongoing to extend the implementations of the
constitutive model to an Eulerian framework and refine the constitutive model calibration.
Keywords: Constitutive Models, Hyperelasticity, Hyperelastic Material, Tear Test, Rubber

1. Introduction
Highly-deformable rubbers are often used in medical devices and normally require
nonlinear constitutive models to accurately represent the material’s mechanical response.
Abaqus includes a wide range of such hyperelastic material models available for
implementation in finite element simulations, but care must be taken to ensure the
constitutive material model provides a numerically stable formulation and accurately
represents the material’s response for the desired context of use. Challenges often arise in
determining the appropriate constitutive model and its corresponding parameters that
enable a satisfactory fit to the experimental data used for model calibration. Comparing
the model against experimental datasets not used to calibrate the constitutive model is an
imperative step to define the model’s context of use and understand the model’s
predictive capabilities. In this paper, we present a study to fundamentally characterize the
material’s mechanical response, calibrate and select a constitutive material model,
characterize the tear strength of the material, and implement the constitutive model in
finite element models of the tear strength experiments.
This paper is organized into the following sections. In Section 2, we discuss the methods
and methodologies for each workstream: the selection of appropriate experiments and the
procedures followed for the material characterization (cf 2.1), the methodology followed
for the constitutive material model selection and calibration (cf 2.2), and the constitutive
model implementation in Abaqus/Standard and Abaqus/Explicit. Section 3 illustrates and
discusses the results from each workstream: the experimental tests (cf 3.1), the
constitutive material model calibration and selection (cf 3.2), and the constitutive model
implementation in Abaqus/Standard and Abaqus/Explicit (cf 3.3). The paper concludes
with comments of continuing work (Section 4) and overall conclusions of the
aforementioned studies (Section 5).
2. Methods
2.1 Experimental material characterization
The selected material for this work was commercially available latex rubber. The
material came as a single sheet 36 inches long, 6 inches wide, and 0.04 inches thick. Test

specimens were punched or cut into the desired forms as needed. A suite of tests were
conducted to broadly explore and characterize different aspects of the material response.
The tests included i) monotonic uniaxial tensile tests at different strain rates, ii) cyclic
uniaxial tensile tests at different strain rates, iii) Mullins effect uniaxial tensile tests, iv)
cyclic compression tests, and v) creep relaxation DMA testing. No pre-cycling was
administered to the material to represent the end use of the material. An Instron 5566
universal material testing system was used for all of the tension and compression tests.
Tensile experiments were conducted using standard ASTM D412 Type C tensile bars and
video extensometry was used for the strain measurements. Compression experiments
were conducted after ASTM D-695 2010 and used multiple 1 inch diameter disks that
were stacked on top of one another between lubricated plates 0.331 inches apart. DMA
testing was performed at two different strain rates and the strain of 100% was held
constant for 8 hours.
Tearing tests were also conducted to supplement the previously discussed experimental
dataset with material failure data. The ASTM D624 Bowtie/Type C geometry was used
for the tearing tests and experiments were conducted at two different strain rates. An
important note is that the tearing data was not used to determine the constitutive model
parameters and could be used as a validation dataset for the constitutive model parameter
calibration.
2.2 Material model calibration
After reviewing the material test data and the material model’s intended use case (i.e.,
loading conditions), the monotonic, tensile loading was deemed the most representative
for the model’s context of use and the compression, DMA, and cyclic tensile testing data
was not used to calibrate the material model. Specific focus was given to numerical
stability and achieving a good fit at multiple regions along the stress vs strain curve.
Multiple material models were considered including the Bergstrom-Boyce, Yeoh, Ogden,
and reduced polynomial strain energy functions. Details on the mathematical
formulations of hyperelasticity and strain energy functions each model can be found in
(Holzapfel, 2000), the PolyUMod User’s Manual, and the Abaqus analysis guide. The
calibration tools included Abaqus/CAE and MCalibration (Veryst Engineering, LLC,
Needham Heights, MA). The net error between the experimental data and constitutive
model predictions was calculated using the normalized mean absolute difference (NMAD)
between the experimental data. An iterative error minimization scheme semi-

automatically determined the model parameters that would minimize the net error
between the experimental data and model predictions.
2.3 Material model implementations
The material model was implemented in static Abaqus/Standard v14.1 and quasi-static
Abaqus/Explicit v14.1 simulations of the ASTM D624 Type C experimental tearing test,
which were not used to calibrate the material model. All simulations were displacement-
driven and the Abaqus/Explicit simulations had a load step time, displacement amplitude,
and target stable time increment to ensure minimal mass scaling and kinetic effects. The
element types used in the simulations were C3D8H elements for Abaqus/Standard
simulations and first-order C3D8R elements with enhanced hourglass control for
Abaqus/Explicit simulations. A mesh refinement study was conducted by comparing the
calculated force vs. displacement curves or different element densities across the Type C
geometry.
A variety of boundary conditions were tested to assess the influence of different gripping
conditions and the enforcement of symmetry. The gripping regions were kinematically
coupled to reference points outside the geometry to enable efficient load specification and
force vs. displacement post-processing. Figure 1 illustrates the different Lagrangian
formulations of the Type C tear test simulation.
Half Symmetry Type C Tear Test Geometry Full Type C Tear Test Geometry
with Maximum Grip
Half Symmetry Type C Tear Test Geometry Full Type C Tear Test Geometry with
with Minimum Grip Maximum Grip

Figure 1. Kinematic coupling and symmetry boundary conditions of the Type C tear
simulation for maximum and minimum gripping scenarios
3.1 Experimental material response
Figure 3 illustrates representative tensile data collected for the latex rubber for three
different strain rates on the orders of 0.001, 0.01, and 0.1 /s. The material’s mechanical
response exhibited a relatively low sensitivity to strain rate and a limited energy loss
during loading and unloading cycles. Ultimate strains at failure were around 1000% for
monotonic uniaxial tension. Whereas, the Mullins test data revealed a noticeable
sensitivity to cyclic load history with a failure strain around 750% and a reduction in the
ultimate stress when compared to the monotonic uniaxial tension data.

Cyclic Tension Monotonic Tension
Red:
Green:
Blue:
Mullins Test Cycles Mullins and Monotonic
Figure 3. Representative experimental tensile data for the latex rubber.
Figure 4 illustrates representative tearing test data for the ASTM D624 Type C geometry
for different extension rates an order of magnitude apart (0.315 and 0.0315 in/s.). An
important note is that the extension rate had a noticeable dependence on the displacement
and force at material tearing/failure.

Figure 4. Experimental tearing test data for the latex rubber.
3.2 Material model calibration
As previously mentioned, multiple models were considered for material calibration to the
monotonic uniaxial tension experimental data. Due to the material’s limited viscoelastic
effects and high levels of deformation, hyperelastic models were deemed suitable to
represent the primary characteristics of material’s mechanical response. The Yeoh model
was selected due to its simplicity, overall numerical stability, and enhanced convergence
properties at high levels of element distortion. Figure 5 illustrates the fit between the
Yeoh model and the experimental data.

Experimental Data
Yeoh
Model
Figure 5. Calibration of the Yeoh model parameters to the monotonic uniaxial tension
data
The model parameters were selected to ensure the existence of real roots for all possible
shear modulus values according to the Yeoh model parameter restrictions outlined in
(Holzpafel, 2000). An important note of the model’s deficiency is the material model
does not exhibit the same level of stress stiffening as the experimental data at engineering
strains above 5.0 (500%). Decreasing the discrepancy at higher strains is an ongoing
effort for material calibration refinement, but the model matches the data well for
engineering strains below 5.0.
3.3 Material model implementations in Abaqus
The mesh refinement analysis revealed force vs displacement curves that showed no
changes for the different element densities. The medium-grade mesh density was thus
selected as the base mesh density for all simulations. Whereas, a comparison of the
different grip geometries yielded significantly different force vs. displacement curves and
also local deformation gradients for both Abaqus/Standard and Abaqus/Explicit
simulations. Figure 6 illustrates the von Mises stress contours for the different grip-types
in Abaqus/Explicit.

Half Symmetry,
Maximum Grip
Abaqus/ Explicit Max stress at full
displacement
Half Symmetry, Max stress at full

Minimum Grip displacement
Abaqus/ Explicit
Figure 6. Stress contours for the maximum and minimum grips in Abaqus/Explicit.
An important note is the visibly different deformation gradients at the maximum stress
region for the two different simulations, where the only difference between the
simulations is the grip boundary conditions. For corresponding grip types, no visible
difference was observed between the calculated force vs displacement curves from
Abaqus/Standard and Abaqus/Explicit. Moreover, when compared with the experimental
data, the two different grip conditions straddle two sets of experimental data that were
collected at two different testing laboratories. Figure 7 depicts the different Abaqus
simulations’ force vs displacement to the experimental data. Note that Figure 7 contains
data from Abaqus/Standard and Abaqus/Explicit simulations but their resulting curves lie
on top of one another.

Figure 7. Comparison between the force vs displacement curves from
Abaqus/Standard and Abaqus/Explicit simulations with different grip conditions and
two different experimental datasets.
The results in Figures 6 and 7 are for half symmetry geometries. To ensure the
constitutive model would maintain stability in full geometry simulations, a full geometry
was created and run with the maximum grip condition in Abaqus/Explicit. Figure 8
illustrates the close match between the force vs displacement curves for the full and half
symmetry geometry.

Figure 8. Comparison between the force vs displacement curves from half-symmetry
and full-geometries with maximum grip conditions in Abaqus/Explicit and two
different experimental datasets.
4. Ongoing Work
Significant work has been completed on developing a constitutive model for highly
deformable latex rubber, but opportunities exist to further enhance the constitutive model
and its implementations. The first is to refine the model’s mechanical response at higher
strains. To accomplish this task, we are trying different calibration procedures such as
fitting the model parameters in a piecewise fashion. Another approach is to consider
models with higher order functions of the first invariant. Additional model enhancements
include developing material failure model(s) that have a dependence on loading rate as
shown in the Type C tearing test experimental data. The last piece of ongoing work is to
implement the constitutive model in a coupled Eulerian Lagrangian (CEL) simulation of
a standard experimental test and compare the model’s predictive capabilities in both
Lagrangian and Eulerian frameworks.

An important note is that only global mechanical response has been compared against
experimental data. However, the local deformation gradients (nodal displacements) have
not been compared to any experimental characterization. Thus, care must be taken when
interpreting stress magnitude values calculated from the different tear simulations due to
potential differences in the cross sectional areas created by nodal displacements at the
maximum stress locations. Such experimental characterization requires significant
additional work and the necessity of such data for constitutive model validation is
dependent on the constitutive model’s context of use.
5. Conclusions
The authors have discussed and illustrated efforts aimed at developing a material for
highly deformable latex rubber that can be used in a wide variety of simulations.
Implementations of the constitutive model in a tearing test not used to calibrate the model
parameters resulted in good agreement with the experimental data. Additionally, the
influence of boundary conditions on a specimen’s mechanical response was clearly
illustrated and reinforces the need to critically analyze simulation results for physical
relevance and overall insight to the simulation’s dominant physics.
6. References
1. Abaqus Analysis Manual, Version 6.14-1, Dassault Systémes Simulia Corp.,
Providence, RI.
2. Holzapfel, G.A., Nonlinear Solid Mechanics: A Continuum Approach For
Engineering, John Wiley and Sons, Chichester, England, 2007.
3. PolyUMod, User’s Manual, Version 2.6.2, Veryst Engineering, LLC,
Needham Heights, MA.

Developing a Finite Element Head Model for Impact
Simulation in Abaqus
P.G. Young1, S.M. Qidwai2, A. Bagchi2, N.
Kota2, C. Pearce3, R.T. Cotton1
1
Simpleware Ltd.
2
US Naval Research Laboratory
3
Atkins Ltd.
Abstract: A computational model of the human head has been developed by

Simpleware and the U.S. Naval Research Laboratory (NRL) to simulate head impact.
The head model was generated using high-resolution scan data, segmentation and
image-based meshing techniques, and has been used in Abaqus for a range of
simulations. The NRL/Simpleware head model addresses several key limitations of
existing techniques for head impact modelling, most notably mesh adaptability for
different applications. This paper will focus on the development of the modeling
techniques used to create the NRL/Simpleware head model, and its use in
Abaqus/Explicit to demonstrate accuracy and feasibility against a widely used
experimental benchmark in impact loading and through investigation of potential
brain injury under blast overpressure loading.
Keywords: Abaqus, Abaqus/Explicit, Biofidelic, Blast Impact, Blunt Impact, Brain

Injury, CAD, FEA, Finite Element Method, Head Impact, Head Model Mesh
Generation, Image Segmentation, Simpleware, Military. MRI Scan, Traumatic Brain
Injury, US Naval Research Laboratory, and Visualization.
1. Introduction
Accurate finite element (FE) head models generated from scan data can be used to
simulate a range of head and brain injuries in Abaqus software. Traumatic Brain Injury
(TBI) from head impact is a serious issue across areas such as sports, the military and
general accident treatment. For example, amongst surviving casualties of the Iraq and
Afghanistan wars, between 14% and 20% suffer from TBI’s that result in long-term
neurological and motor disorders (US Department of Veterans Affairs). While
significant experimental research has been carried out into the mechanisms that cause
TBI from impact or blast, these tend to be costly or limited by the availability of
cadavers, suitable animal samples and other in vivo cases.

Numerical modelling using the FE method and 3D image data (MRI, CT, micro-CT…)
is growing in popularity as a solution for researching TBI. However, to date methods
for generating realistic human head models have included some limitations that reduce
model flexibility and the speed by which they can be produced. This paper outlines a
new method and applications for a computational head and neck model that promises
new breakthroughs in numerical TBI-related research using solvers such as Abaqus.
2. Benefits of an Image-Based FE Method for Head Modelling

Advances in computational power and imaging techniques have contributed towards the
growing effectiveness of image-based FE simulation. In terms of head modelling, FE
simulation allows complex trauma mechanisms from impacts and blasts to be
researched from scanned images, and offers potential for running repeatable simulations
that can go hand-in-hand with other experimental tests. The major challenge of FE-
based head modelling, however, involves accurately reproducing the geometry of an
original scan.
Past approaches to building a computational head model typically involved manual

generation of structures using Computer-aided design (CAD) tools, followed by
meshing and export to a FE solver. However, CAD methods create problems in terms of
reconstructing the complex anatomy of the head and internal structures, which can
require a large amount of manual fixing. In this scenario, CAD-based head models are
prone to error, particularly when it comes to the ability and experience of the user
building the model. In addition, models built using CAD techniques often become
computationally intractable as new levels of geometric fidelity are targeted.
By comparison, image-based modelling techniques that work with image data provide a
faster and often more robust route to simulation. Image-based meshing techniques that
work with volume scan data, such as computed tomography (CT) or magnetic resonance
imaging (MRI), can generate FE meshes using full and semi-automated techniques. The
particular methods outlined in this paper reduce user input to a minimum and produces
meshes that are accurate, robust and include multiple complex geometries and
parameters for simulation.
This approach builds upon and improves previous efforts to generate FE meshes from
image data that, while successful in creating models, have experienced challenges in
adapting models to new applications. For example, work has been carried out into
generating a model from 14 cross-sectional MRI image slices. Image processing tools
highlighted the skull, with outlines read by a C++ code and converted into CAD
coordinate and spline data (Mehta et al, 1997). The resulting model provided a good
approximation of the complex anatomy of the skull. Subsequent research has focused on
taking advantage of developments in imaging and computational power to generate
high-quality head models.
Examples include the Simulated Injury Monitor (SIMon) FE model, which was
developed by the US National Highway Traffic Safety Administration. CT data
representing the average adult male was used to build the mesh, with the head model

having since been validated against different types of experimental data. The drawback
of this model is that it was designed for a specific set of inputs, namely crash impact
(Takhounts et al, 2008). Other head models include those created from T1-weighted
MRI data), with diffusion tensor imaging used to inform an anisotropic material model
including neural tissue (Kraft et al, 2012). A number of other models have been
generated using image-based and CAD techniques, with examples including work
carried out at the University of Illinois (Chen and Ostoja-Starzewski, 2010).
While these examples demonstrate advances in generating models from image data,
they share the problem of being ‘fixed’ meshes that cannot be easily adapted to new
applications. Users wanting to refine meshes or add in new structures such as helmets or
soft and hard tissue have to carry out work directly on meshes. As well as creating a
high risk of error, the resulting models are typically unsuitable for adaptation to a
specific application or as tools for different validation inputs.
The NRL-Simpleware head model described in this paper solves these problems by
using novel image-based meshing techniques to create a far more flexible head model
than previous efforts; meshes can be generated to suit both specific problems, and can
be adapted to changes in computational resources. Rather than working with pre-
defined meshes, this approach generates FE meshes from pre-segmented image data. As
a result, the NRL-Simpleware head model represents a cutting-edge solution for rapidly
and accurately generating meshes that can be tailored to particular simulation tasks in
Abaqus and other solvers. The rest of this paper outlines the development of the
techniques used to produce the pre-segmented data and the creation of a particular
NRL-Simpleware model for use in blast loading and blunt impact simulations in
Abaqus/Explicit.
3. NRL-Simpleware Head Model Development
Step 1: Data Acquisition: Image data for the head model was obtained from a whole
head in vivo MRI scan of a 25-year old male volunteer at the Exeter MR Centre, UK.
The T1-weighted scan resulted in coronal plane image slices with a resolution of
1.03516 mm x 1.03516 mm, with a slice-to-slice separation of 1.04001 mm. The
volunteer was 1.8 m tall and 81 kg, putting him close to the American 50th percentile
male.

Figure 1. MRI image data used to develop the NRL-
Simpleware model.
Step 2: Image segmentation: The segmentation process involved importing the

DICOM (Digital Imaging and Communications in Medicine) image series from the
MRI scanner into image processing software ScanIP (Simpleware Ltd., Exeter, UK).
Segmentation then entailed identifying and labelling regions of interest (ROI’s) within
the greyscale data, including brain, skull and muscles, as masks used for reconstructing
surfaces and generating volume meshes. A range of image processing techniques were
used within ScanIP to build the masks, including image threshold and flood-fill
algorithms to obtain initial geometries for major structures of the head. Manual
segmentation was also used to disconnect neighbouring regions with similar greyscale
values.
Segmented parts included the cerebellum (separate white and grey matter), cerebrum
(separate white and grey matter) and brain stem, while image filters were employed to
smooth regions within masks, to edit the morphology and to fill cavities. Boolean
operations were used to remove any overlapping masks. Tools for ‘confidence
connected region growing’ (CCRG) and ‘magnetic lasso’ were particularly suitable, in
this instance, for capturing regions including the skull, vertebrae and ventricles. The
CCRG tool functions by selecting a seed point before building a region of neighbouring
voxels based on their similarity to the original seed point. The magnetic lasso tool used
for segmentation also has a propagation option that allows the outline of a region to be
tracked to the next image slice. A technique that uses existing parts to help build
neighbouring parts was also applied to the model, which enabled the generation of the
tentorium cerebelli by creating a one voxel thick layer on the superior surface of the
cerebellum. The final bitmapped (voxel) images were anti-aliased and smoothed to
reproduce the smooth contours of the biological structures. Proprietary multi-part
algorithms in ScanIP were used for this purpose to ensure that no changes in
connectivity or the volume of the smoothed structures took place. This technique
reproduces the accuracy of the scan data.

Table 1. List of anatomical structures in
current head model.
Structures differentiated, defined as separate element sets.
1 Skin 12 Cerebrum – grey matter 23 Spinal cord

2 Muscles 13 Cerebellum – grey matter 24 Ventricles – lateral (right)
3 Soft tissues 14 Cerebrum – white matter 25 Ventricles – lateral (left)
4 Skull inner and outer table 15 Cerebellum – white matter 26 Ventricles – third
5 Skull dipole 16 Falx cerebri 27 Ventricles – fourth
6 Mandible 17 Tentorium cerebella 28 Ventricles – aqueduct of Sylvius
7 Cervical vertebrae 18 Eyes 29 Ventricles – foramen of Monro
8 Intervertebral discs 19 Optic nerves 30 Venous sinuses and bridging veins
9 CSF 20 Brain stem – medulla 31 Venous sinuses and bridging veins walls
10 Frontal sinus 21 Brain stem – midbrain (shell elements)
11 Maxillary sinus 22 Brain stem - pons 32 Dura mater (shell elements)
33 Sclera (eyes) (shell elements)
Step 3: Image enrichment and modification: Following segmentation, the image data
was enriched by incorporating CAD models of structures where the original greyscale
contrast made it impossible to distinguish between structures. Facial and neck muscles
were imported from a Standard Tessellation Language (STL) surface triangulation
library. ScanIP module +CAD was used to scale and position parts, and to convert them
into image masks using a process called ‘voxelisation’, whereby a distance function is
used to calculate representative greyscale information. This technique also means that it
is straightforward to modify original segmented data to incorporate new structures such
as helmets and probes, and to introduce pathologies or combine different structures into
one.
Figure 2. Example of image enrichment: (a) introduction of neck muscles

from external CAD source, (b) example of a modification incorporating a
helmet design.

Step 4: Image-based mesh generation: After segmentation, the image data can be
converted into a mesh using two routes. The first approach involves directly converting
ROIs into volumetric FE models by using a multi-part Extended Volumetric Marching
Cubes (EVOMAC) approach (Young et al, 2008). Alternatively, the boundaries of the
voxelised ROIs can be converted into surface representations (such as triangulated
surfaces (STL/NURBS) and used as the basis for mesh generation using a range of
automated mesh-generation algorithms. Both mesh generation techniques were
explored, incorporating the EVOMAC-based mesher and a multi-part surface
decimation algorithm followed by a mixed Delaunay advancing front approach (Young
et al, 2008).
The EVOMAC-based approach produces a structured mesh composed of both

hexahedral and tetrahedral elements, or purely of tetrahedral elements. The second
approach results in unstructured, fully tetrahedral meshes (see Figure 4), that can be
modified by controlling element sizes and the final element count. The NRL-
Simpleware head model used for simulation of Traumatic Brain Injury was generated as
an unstructured, all tetrahedral mesh consisting of 3.72 M volumetric elements, with
options remaining for new meshes to be created from the pre-segmented data.
Figure 3. Model generation procedure following image acquisition
Other mesh modification options that can be applied before simulation include:
specifying contact pairs to allow for sliding surfaces or separation; adding node and
surface sets based on mask geometry or in arbitrary user-defined regions; defining
material properties based on the underlying greyscale intensity of the image; and
applying Boolean operations directly to structures in image space to merge or remove
features or specific regions, reducing the element count if a distinction is not required.

Figure 4. Sagittal section of the unstructured, all tetrahedral mesh
with 3.72 M volumetric elements: (left) oblique view (right) normal
view.
4. Adaptation of prototype model for Abaqus simulation
The NRL-Simpleware model was developed using meshing techniques suitable for
converting pre-segmented data into FE models for different applications. In this context,
the main unstructured mesh used for the model was adapted into a mixed-element mesh
to simulate head injury under mid-to high-rate impact and blast loading conditions.
Four-node linear elements were initially used as a computationally inexpensive option,
but required modification as they are stiffer than hexagonal elements, and are liable to
experience volumetric locking, making them unsuitable for displacement-based FE
formulations and modelling near-incompressible materials such as the brain.
Hybrid elements capable of describing pressure as an independently interpolated

solution variable one degree lower than displacement degrees of freedom were
considered, but are not available in Abaqus/Explicit. By comparison, 10-node quadratic
tetrahedral elements (used in Abaqus/Explicit) avoid locking and allow high-fidelity
geometric representations, but are computationally expensive. To ensure high-rate
simulations, a compromise was made between computational efficiency and accuracy
by building the mesh with regions of near-incompressible materials, such as the
cerebrospinal fluid (CSF) and the brain, with quadratic elements. Changes in nodal
definitions were performed to describe the interface of the quadratic and linear parts of
the FE mesh, including splitting shared nodes between quadratic and linear elements to
prevent co-penetration of the additional nodes on the quadratic side into the linear
elements, and defining contact interactions at these interfaces.

Linear elements were converted into quadratic elements to build regions of near-
incompressible materials such as the brain, spinal cord, CSF, sinuses filled with air,
ventricles filled with CSF, veins filled with fluid, and eyes. Contact surfaces were
created between quadratic elements of veins and linear elements of skull, veins and soft
tissue outside the intracranial region, and at the eyes and soft tissue/skin, by duplicating
shared nodes at interfaces. Tied contact conditions were also imposed at these
interfaces, and the skull and mandible partitioned into multiple regions with similar
outward facing normal vectors, allowing transversely isotropic material properties to be
assigned to each region.
In addition, tied contact conditions were imposed between the sub-arachnoid region
filled with CSF, and the dura mater/pia mater. Conditions were also imposed for the
spinal cord, where appropriate, with material models for individual parts based on
experimental data whenever available. The functional forms of models included simple
linear elastic and equations of state, as well as more advanced hyper-viscoelastic
models. A modified Monte Carlo-based optimisation scheme was used for calibration of
these complex models. More details on constitutive models, calibration and verification
and validation are available (Kota et al, 2014).
5. Applications
The adapted NRL/Simpleware model was used for blunt impact (low-to-mid rate) and
blast overpressure (mid-to-high rate) simulations. Abaqus/Explicit was chosen for
running simulations using 200 Intel® cores (2GB RAM/core; 2.6 GHz core speed) on
an SGI Ice X system. A variable mass-scaling scheme was used, resulting in a
reasonable time increment of 50 ns or more during simulation; this allowed the
completion of around 20 ms of total simulation time in 24 hours. Change of total model
mass during simulations was checked and recorded as being lower than 0.09%
throughout, indicating that no significant effect on inertial characteristics took place.
The blunt impact loading application was validated against results from an experiment
performed on a post-mortem human subject (Nahum et al, 1977) to reproduce
automotive impact. The experiment (#37 in the original study) involved a frontal impact
to the head at 45° by a padded impactor of 5.6 kg mass travelling at 9.94 m/s (Figure 5).
Simulations were run using the experimental force recorded on the head as the input to
the model, with no other boundary conditions applied. The predicted intracranial
pressures were compared with experimental data for frontal and parietal regions (Figure
6 (a) and (b)). The simulated pressures in the model reasonably capture the pressure
histories recorded within the skull.
Using this computational modelling approach enables a build-up of detailed spatial data
that is otherwise typically not practical to collect from experiments. Post-processing of
the simulation data can, for example, be performed to quantify the injured brain volume
as a function of event time based on injury threshold measures described in the
literature for focal and diffuse injuries (Bešenski, 2002). For the current simulation,
Python scripting was used to calculate the evolution of these volumes from the large
output database. For example, the temporary history for focal injury volume was based

on a pressure threshold of 173 kPa (Zhang et al, 2004). Diffuse injury volume was
based on a shear strain threshold of 5% (Margulies and Thibault, 1992), as shown in
Figure 6 (c), with post-processing capabilities highlighting the utility of the
computational models.
Figure 5. (left) Schematic of a padded impactor hitting the

head model at 45° angle; (right) instead of the impactor,
the experimental force-time plot was used.

Figure 6. Comparison between experimental and predicted
pressures for blunt impact to the head in (a) frontal and (b)
parietal regions of the brain, respectively; (c) Temporal history
of injured brain volume based on maximum principal strain
threshold of 5% and pressure threshold of 173 kPa.
Blast overpressure loading was simulated by choosing the incident wave-loading option
in Abaqus/Explicit. The magnitude of the applied pressure pulse is specified on a plane
with a specific normal vector; as the wave progresses in the direction of the normal, it
encounters the target in a manner similar to a true overpressure impact. However, this
does not allow for the wraparound effect. A frontal loading with a Friedlander wave
profile (Dewey, 2010) was chosen for this study, with a peak pressure of approximately
430 kPa (Figure 7). The passage of the pressure wave passing across the face and close
to the mid-sagittal plane of the head is shown in Figure 8.
Figure 7. Friedlander wave profile

used in frontal blast loading
simulation.

Figure 8. Passage of peak pressure pulse across the face and
near mid-sagittal plane of the head during loading.
The injurious effect of this pressure pulse can be displayed in a temporal-spatial form
by collecting the temporal injury data for each individual material point and plotting it
over the target volume. These graphics can then be used to predict injury patterns or to
draw spatial comparisons amongst different metrics of injury to determine their relative
effectiveness in predicting actual injury. For example, brain regions based on thresholds
of 11 kPa effective stress (Kang et al, 1997) for focal injury, and 5% maximum
principal strain (Margulies and Thibault, 1992) for diffuse injury, are highlighted in
Figure 9 (a) and (b). The red-coloured regions represent cumulative injury in the first 5
ms based on each threshold.
Figure 9. Spatial identification of injured brain regions due to blast

overpressure loading based on (a) effective stress threshold of 11 kPa for
focal injury and (b) maximum principal strain threshold of 5% for diffuse
injury, up to 5 ms. Red-coloured regions have experienced values beyond
the respective thresholds.

6. Summary
The NRL-Simpleware head model was developed using novel image-based

segmentation and meshing techniques for simulation in Abaqus/Explicit. By generating
meshes from pre-segmented data, the approach represents a breakthrough in terms of
flexibility compared to previous FE modelling methods. Users are able to adapt models
to suit different simulation needs, with options available for incorporating new
structures of interest, cropping extents of interest and merging features together. In
addition, the ability to introduce robust contact surface definitions at the interface
between parts, and to locally refine or decimate meshes, makes it straightforward to
customise models for different applications.
Simulations using the NRL-Simpleware head model that recreated experimental data
(Nahum et al, 1977) showed excellent agreement with benchmark results for blunt
impact. Blast overpressure simulation results that reproduced conditions associated with
military and explosive weaponry also demonstrated the head models’ capabilities for
predicting brain injury as a function of both location and time. This success opens up
the use of biofidelic head models for future simulations in Abaqus.
The approach outlined in this paper is capable of generating more detailed structures
from the head and brain geometry, but is currently limited by the number of elements
that can be reasonably solved by solvers using existing, and cost-effective,
computational resources. As imaging and simulation techniques develop, the image-
based segmentation and mesh generation techniques this paper proposes will be easily
adaptable and scalable to new challenges. Future investigations might include
generating a population of head models from image data used to study variations in
anatomy according to impact response. Other challenges for future research include
more accurately representing the highly complex features of hard and soft tissue
structures.
7. References
1. Abaqus Documentation, 2012. Dassault Systémes Simulia Corp., Providence,

RI.
2. Bešenski N., “Traumatic injuries: imaging of head injuries,” Eur Radiol.,
12(6), 1237-1252, 2002.
3. Chen, Y., and M. Ostoja-Starzewski, “MRI-based finite element modeling of
head trauma: spherically focusing shear waves,’’ Acta Mech., 213, 155-167,
2010.
4. Dewey, J.M, “The shape of the blast wave: studies of the Friedlander
equation,” 21st international symposium on military aspects of blast and
shock, Israel, 2010.
5. Kang, H.S., R. Willinger, B.M. Diaw, and B. Chinn, “Validation of a 3D
anatomic human head model and replication of head impact in motorcycle
accident by finite element modeling,” SAE Trans., 106(6), 3849-3858, 1997.
6. Kota, N., A. Bagchi, and S.M. Qidwai, “On the challenges in verification and

validation of biomechanical models of humans: a case study on head model
development for impact loading,” J Biomech, 2014.
7. Kraft, R., P. Mckee, A. Dagro, and S. Grafton, “Combining the finite element
method with structural connectome-based analysis for modeling neurotrauma:
connectome neurotrauma mechanics,” PLoS Comput Biol. (Internet),
Available from: http://dx.plos.org/1-.1371/journal.pcbi.1002619, 2012.
8. Margulies, S.S., and L.E. Thibault, “A proposed tolerance criterion for diffuse
axonal injury in man,” J Biomech, 25(8), 917-923, 1992.
9. Mehta, B.V., S. Rajani, and G. Sinha, “Comparison of image processing
techniques (magnetic resonance imaging, computed tomography scan and
ultrasound) for 3D modeling and analysis of the human bones,” J Digit.
Imaging, 10, 203-206, 1997.
10. Nahum, A.M., R. Smith, and C.C. Ward, “Intracranial pressure dynamics
during head impact,” Proceedings of the 21st Stapp Car Crash Conference,
339-366, 1977.
11. Takhounts, E.G. S. Ridella, V. Hasija, R.E. Tannous, J.Q. Campbell, D.
Malone, et al., “Investigation of traumatic brain injuries using the next
generation of simulated injury monitor (SIMon) finite element head model,”
Stapp Car Crash J., 52, 1-31, 2008.
12. U.S. Department of Veterans Affairs, Traumatic brain injury: a guide for
patients (Internet), Available from:
http://www.mentalhealth.va.gov/docs.tbi.pdf
13. Young, P.G., T.B.H. Beresford-West, S.R.L. Coward, B. Notarberadino, B.,
and A. Abdul-Aziz, “An efficient approach to converting three-dimensional
image data into highly accurate computational models,” Philos Teans R Soc
A Math Phys Eng Sci, 366, 3155-3173, 2008.
14. Zhang, L., K.H. Yang, A.I. King, “A proposed injury threshold for mild
traumatic brain injury,” J Biomech Eng, 126(2), 226-236, 2004.
8. Acknowledgement
Research was carried out at Simpleware Ltd. and the US Naval Research Location.
Funding was supported by the Office of Naval Research (ONR) through the US Naval
Research Laboratory’s Basic Research Program, and the Department of Defense (DoD)
High Performance Computing Modernization Program (HPCMP) using the Air Force
Research Laboratory (AFRL) Major Shared Resource Center (MSRC) under project
416, subproject 231.

Impact Simulation of a Hearing Aid
Morten Birkmose Søndergaard
GN ReSound A/S, Ballerup 2750, Denmark. e-mail: morsoe@gnresound.dk
Abstract
In the development of hearing aids, many factors are important to make a good hearing aid. It
can be the sound quality, design and easiness of use. However, it is also important that the device
can withstand the environment that it is used in, both regarding moisture, sweat and to be dropped
on a hard surface. One of the most sensitive parts in the hearing aid regarding dropping the
device on a hard surface is the receiver (name for the loudspeaker in the hearing aid).
The development of hearing aids demands shorter time to market and higher rate of success in
achieving all the development requirements for the device. Therefore, it is important to be able to
predict and improve designs in development projects.
In this case, a FE model of the hearing aid is built to predict the g force of a receiver inside the
hearing aid, dropped from 1 m height. The simulated g force is compared to real drop test.
The FE simulations are performed in both Abaqus/Standard and Abaqus/Explicit in order to
pre-bend some parts before they are imported in the impact simulation in Abaqus/Explicit.
The g force of the receiver during an impact are simulated with different designs of the soft
suspension surrounding it.
Physical impact test with hearing aids containing the different suspension designs were
performed and compared to the results from the simulations with good correlation.
Keywords: Impact Simulation, Finite Element, Hearing Aid, Hearing Instrument, Industrial
Application, Microphones, Receiver, Loudspeaker, Ear, Hearing, Life Science, Medical.
1 Introduction
A hearing aid (in this case a BTE, Behind The Ear, instrument) comprises a microphone
which receives acoustic signals. The received acoustic signals are processed where the processing
may include amplification of the data. The processed signals are transmitted to a receiver
(loudspeaker) of the hearing aid which converts the processed signals into another acoustic signal
e.g. with a larger amplitude at certain frequencies. The receiver broadcasts this acoustic signal
towards the tympanic membrane of a user of the hearing aid.

The trend for designing hearing aids is that the size is getting smaller to meet a more cosmetic
appealing look and to reduce the stigma for being a hearing aid user. This leads to a smaller
distance between the receiver and the parts and components next to it and thereby decrease the
possibility of damping the impact force.
The most vulnerable component due to an impact is normally the receiver inside the hearing
aid. The receiver is softly suspended, to prevent its vibrations from being transferred through the
device and up to the microphones (in order to minimize feedback between microphones and
receiver). The suspension around the receiver is usually made of rubber. The rubber suspension
also helps to absorb an eventual impact when dropping the hearing aid on a hard surface.
The purpose of this simulation is to calculate the g force in the receiver when the hearing aid is
dropped from 1 meter and landing on a hard surface.
The supplier of the receiver specifies the g force that the receiver under investigation can
handle as 14.000 g’s.
Figure 1.X-ray of good receiver to the left and receiver with bended armature to the right.
As seen in in the right picture in figure 1, the armature /pin is bended due to a high impact. If the
armature is bended, the receiver can still generate sound but it will be distorted. The impact
damage of the receiver is detected by an electro acoustical measurement of distortion from the
receiver.

Figure 2. Assembly of hearing aid impact model.
Rubber
Suspensio
Receive
1.1 Advantages and benefits

The advantages from the simulations are that we can gain knowledge just from visualizing the
impact and see where parts are moving and deflecting.
This increases the understanding of what happens during the impact, which we can use to improve
the design and thereby durability of the device, which will benefit the user at the end.
Other advantages are that we can easily try out new designs virtually and save development time,
because we exclude the time for drawing, molding, assembling and testing the next designs.
The physical impact test is a destructive test that only can give a pass/fail result and not provide
the resulting g force. The simulation gives more explaining and understanding of what happens in
the impact.
2 Description of finite element model
The simulation is divided in two simulations. First simulation will deform the receiver suspension
to fit around the receiver. The last simulation is the actual impact simulation where the deformed
suspension is imported.
The geometry for the parts are created in CREO and then imported as STEP files into Abaqus.

The material parameters for the plastic parts are from datasheet. The material parameters for
the rubber is from mechanical testing of static stress-strain measurements as well as dynamic shear
modulus (G’ and G’’).
The viscoelastic behavior for the rubber is changed from frequency domain (used in the pre
deforming simulation in Abaqus/Standard) into time domain (using the frequency data) for the
impact simulation in Abaqus/Explicit, (Abaqus Analysis User's Guide, V6.14, “22.7.1 Time
domain viscoelasticity”).
2.1 Pre-deforming rubber suspension

The rubber suspension around the receiver is made smaller than the receiver in order to stretch
the rubber, so it will hold the receiver inside it.
The rubber suspension needs to be in the deformed state for the impact simulation, else it will
have overlapping geometry with the receiver.
This is done in a model run in Abaqus Standard in a Static General step with Surface-to-
Surface Contact using small sliding and automatic shrink fit.
Figure 3. Pre-deforming of rubber suspension.
The undeformed suspension is shown to the right in figure 3. The deformed suspension is
shown to the left, where the color map is indicating U (displacement).
2.2 Materials
The only deforming part is the suspension made of rubber. All other parts are made as rigid
parts to make the model as simple as possible. In the future more parts will be changed to non-
rigid, when we know more about their mechanical behavior regarding impact. A stepwise
approach is proposed, where more and more parts are changed from rigid to linear elastic parts,
and then later on changed to elastic-plastic behavior once we have the right material properties for
this (requires material testing at the strain rate and corresponding strain). The contacts between the
parts will also be changed from Tie to more appropriate contacts, which need to be investigated
and stepwise applied.

3 Results and discussion
Several designs of the rubber suspension were simulated before actual design were manufactured
and new hearing aid prototypes were built and tested.
Once the model is working, it can be used to simulate new and improved designs, much faster than
producing new parts, building new devices, test them and compare the result.
Figure 4. Simulated g forces on the receiver with different suspensions absorbing the impact
The results from the impact simulations made in Abaqus/Explicit are shown in figure 4. The
original design, build 4, failed in the physical impact test when hitting from the bottom, Y
direction. The simulation shows a maximum g force peak around 15.000 g’s which is above the
specified limit for the receiver which was 14.000 g’s. The two other designs shown in figure 4
have added rubber taps in the suspension that are caught by corresponding plastic taps added
inside the receiver chamber. These designs prevented the receiver in having a “free fall” and gave
more impact absorption during the impact. The difference between the two last designs is that the

final design, “Max G 10812” is not having the extra tab on the left side, because it did not give
further impact absorption (as can be seen in figure 4).
The final design shown in figure 4 resulted in a maximum g force on the receiver around
11.000 g’s which is below the specified limit for the receiver.
Subsequent physical impact test also showed that 10 devices built with the final design of the
suspension passed.
The g force on the receiver was reduced from 15.000 g’s of the original design to 11.000 g’s on
the final design.
Every time a new design of the rubber suspension was made and tested for impact, it was also
necessary to make vibro-acoustic test on the devices to test for feedback between microphone and
receiver. A good impact absorbing suspension will normally also have increased contact between
the receiver and the surrounding of the receiver. The receiver vibrations will be transferred more
strongly to the microphones, which will pick up the vibrations and send an electrical signal to the
amplifier, and a feedback loop is created, (Søndergaard, 2009).
It is a balance to create the right receiver suspension for a hearing aid, because the impact criteria
demands impact absorption material with contact to reduce the g force and at the same time the
vibrating receiver must not have strong contact to the rest of the device due to transfer of vibration
to the microphones. At the same time, the hearing aid should be as small as possible with small
distance between all parts inside.
The final design of the rubber suspension was also tested for vibro-acoustic stability and passed.
Many other designs were also simulated, and in some cases it was possible to reduce the g force to
5.000 g’s, but unfortunately they either failed the vibro-acoustic test or were not accepted by the
project group due to mechanical constraints or time schedule (it would take too long time to
implement).
3.1 Validation
The simulation will show the g force on the receiver. If the simulated g force is higher than
specified by the supplier, we will assume it will not pass the physical impact test.
The supplier specifies a g force where their receivers in 90% of the cases will have less acoustic
distortion (THD, Total Harmonic Distortion) than 10%.
The physical test at GN are performed by hitting 10 devices from all 6 sides as shown in figure 5.
Each side is hit 5 times. They are hit by a pendulum that simulates a drop from 1 meters height.
After each hit the THD of the devices are measured. If 9 out of 10 devices have less than 10%
THD after the 6 x 5 impacts, then the test has passed.

Impact data from supplier (g force, 90% survival rate) are compared with physical impact test of
devices and correlated with the simulated g force.
Figure 5. Impact directions on device of

physical impact test.
Experiments with high-speed video recording (10.000 frames per second) of the physical
impact have also been performed as seen in figure 6.
Figure 6. High speed video recording of impact. Left picture is 0.9 ms after impact. Middle
picture is 0.1 ms after impact. Right picture is at impact.

The video recordings are difficult to interpret due to low image quality. It was however possible to
determine that the receiver hits the inner wall of the hearing aid after 0.1 ms (middle picture in
figure 6) which corresponds well with the simulations shown in figure 4. A picture of the hearing
aid 0.9 ms after the impact is shown to the left in figure 6.
3.2 Assumptions
Other parts are made rigid to isolate the problem to the receiver suspension. We know that the
plastic parts surrounding the hearing aid are providing damping for the impact.
Rubber material properties are changed from frequency domain to Time domain and we have not
tested the rubber material for Mullins effect (permanent set when exposed for several impacts)
We also assume the device is hit exactly the same direction for the simulation as for the real
impact test.
The simulation does not take into account if the friction force between the receiver and the
suspension is strong enough to keep the receiver in place. We have seen some earlier dropt test
cases where the receiver was moved from its original position after an impact. In the simulation,
we have made a Tie between the suspension and the receiver, because we don’t know the correct
friction between them. At the same time we also made test where the suspension was glued to the
receiver to make a better comparison to the simulations with the Tie.
4 Conclusions
A FE model representing a receiver (loudspeaker) inside a rigid hearing aid was built in
Abaqus/Explicit to simulate a drop from 1 meters height. The resulting g force on the receiver was
derived from the simulations for different rubber suspensions surrounding the receiver. These
results were both compared to physical pass/fail impact tests and to the manufacturer’s data for
acceptable g force for the receiver. Good correlation was found between simulation results and
tests. By using the approach described in this paper it was possible to design an impact protection
that did not damage the receiver after impact and at the same time could fulfill the vibro-acoustic
stability criterion.
5 Further work
We need to know more about the material properties in order to make more precise impact
simulations, as explained in the section of assumptions.
The other parts that were made rigid should also be included to absorb the impact in order to
give a more precise result.

6 Acknowledgement
Anton Jurinic (Senior Technical Sales Specialist SIMULIA Nordics) have been very helpful to
support in setting up the FE model.
References
1. Abaqus Analysis User's Guide, V6.14, “22.7.1 Time domain viscoelasticity”
2. Søndergaard, M.B, “Design optimization in vibro-acoustic design of hearing instruments”, SIMULIA
Customer Conference, June 1 - 5, London, United Kingdom, 2009.

FE Modeling of a Flatfoot Deformity for Improving
Surgical Planning
Zhongkui Wang1, Kan Imai2, Masamitsu Kido2, Kazuya Ikoma2,
and Shinichi Hirai1
1
Department of Robotics, Ritsumeikan University, Kusatsu, Japan
2
Department of Orthopaedics, Kyoto Prefectural University of Medicine, Kyoto, Japan
Abstract: This work presents a finite element (FE) model of a flatfoot deformity to study its
biomechanics and improve the surgical plan for individual patients. Despite being a common foot
deformity, the biomechanics of flatfoot is not fully understood. Regarding the optimal surgery, foot
surgeons have different opinions for individual patients. Studies using cadaver feet are difficult to
manage due to individual differences in foot geometry; therefore, these studies have made limited
contributions to improving surgical planning. One option for solving this problem is to use a
computer-based model. Due to the essence of continuum mechanics, the FE method has been used
to model the human foot for various applications. In this work, the software package Abaqus was
used to model and simulate flatfoot. The geometries of bones and soft tissue of a flatfoot were
generated from CT images and imported into Abaqus as separate parts, which were then
assembled and meshed with linear tetrahedral elements. Typical ligaments, tendons, and plantar
fascia were modeled manually and the approximate positions of origin and insertion were
referenced from an anatomy book together with the instruction of foot surgeons. The contact
interactions among bones, soft tissue, and the ground were modeled using the penalty method with
friction. A total of 17 bones, 62 ligaments, 9 tendons, 6 plantar fascia segments, and 1
encapsulated tissue part were modeled. Balanced standing simulations were performed, and the
stress distributions were compared with experimental measurements to validate the model. The
validated model will be used to study flatfoot biomechanics and investigate the optimal surgery for
individual patients.
Keywords: Flatfoot Deformity, Surgical Simulation, Biomechanics, Elasticity, Hyperelasticity
1. Introduction
Flatfoot is a common foot deformity in which the arch of the foot collapses, with the entire sole of
the foot coming into complete or near-complete contact with the ground. Most people with flatfeet
experience no obvious differences compared to people with normal feet. No medical treatments
are needed for them. In mild cases, orthoses such as an arch support can be used to lessen
discomfort. In severe cases, which mostly happen in adult acquired flatfoot deformity (AAFD),
surgery must be performed. According to different conditions, the appropriate surgeries include:
(1) medializing calcaneal osteotomy (MCO), in which the back part of the calcaneus is cut and
shifted internally, (2) lateral column lengthening (LCL), in which an open wedge osteotomy is

performed between the calcaneus and cuboid and this “lengthens” the foot and rotates the foot
back to its correct position, (3) tendon and ligament procedures, in which certain tendons (e.g., the
posterior tibial tendon) or ligaments (e.g., spring ligament or deltoid ligament) need to be repaired,
and (4) double or triple arthrodesis, in which the fusion of two or more foot joints is performed.
For an individual patient, the foot surgeons tend to have different opinions regarding the optimal
surgery. Although agreement could be achieved on a certain surgery, finding the optimal surgical
procedures or parameters (e.g., the incision position in the case of MCO or the thickness of the
wedge in the case of LCL) is still a challenging task faced by the surgeons. Unfortunately, the trial
and error method is not feasible in medical cases.
To improve this situation and also study the flatfoot biomechanics, physical models, such as a
cadaveric foot, have been used by researchers. Due to the lack of donors, healthy foot samples
were often used to manually generate a flatfoot deformity by releasing or sectioning specific
ligaments and tendons (Kitaoka, 1998; Niu, 2008). However, these artificially generated flatfoot
samples do not fully reproduce the functions of a real flatfoot deformity. In addition, flatfoot
patients usually have different foot deformities. Therefore, research results on artificial flatfoot
samples do not provide much information to foot surgeons.
Computer-based models, meanwhile, can cope with the individual differences of flatfoot
deformities and can easily simulate different surgical situations. The finite element (FE) method
has been used frequently to model the human foot because of its essence of continuum mechanics
and its capability of modeling irregular geometry and complex material properties. Cheung et al.
presented an FE foot model to study the effects of soft tissue stiffening on stress distribution of the
plantar surface during balanced standing (Cheung, 2005). The same model was also used to study
the biomechanical effects of different types of foot orthosis for improving footwear design
(Cheung, 2006). A foot-boot model was built for the future study of footwear design and
parachute landing impact for the military (Qiu, 2011). FE foot models were also proposed for
studying a clawed hallux deformity (Isvilanonda, 2012) and for developing an ankle prosthesis
(Ozen, 2013), respectively. Unfortunately, only one FE model has been found to work for flatfoot,
which was presented by Lewis in his PhD dissertation (Lewis, 2008). In his model, 14 bone
segments, 65 ligaments, and a part of the plantar soft tissue were included, but the toes and dorsal
soft tissue were missing. Regarding the development of FE models, commercial FE software
packages are frequently used by researchers, such as Abaqus (Cheung, 2005; Cheung, 2006;
Lewis, 2008; Ozen, 2013), ANSYSⓇ (Qiu, 2011), and LS-DYNAⓇ (Isvilanonda, 2012). In our
previous work, we have developed a dynamic FE model of flatfoot without using commercial FE
software. The model was used to simulate balanced standing, and simulated stress distribution was
compared with measurements (Wang, 2014). Unfortunately, the contact and friction among
neighboring bones were not formulated in the model, and neighboring bones were connected by
soft tissue instead. This results in relatively stiff joints and further affects the model performance.
In this paper, the flatfoot was modeled using the commercial FE package Abaqus. The geometries
of bones and tissue were generated from CT images. Ligaments, plantar fascia, and tendons were
modeled manually as either 3D solid parts or tension-only wire parts. The approximate positions
of origin and insertion of these tissues were referenced from an anatomy book together with the
instruction of foot surgeons. Contact interactions were modeled with the penalty method. The tibia
and fibula were fixed in space, and a ground plate was loaded and pushed upward with half of the
body weight to simulate a balanced standing situation. The simulation results with different

material properties of encapsulated tissue and different boundary conditions of the ground plate
were presented and compared with each other and also with experimental measurements.
2. Methods
2.1 Geometry Generation of Bones and Encapsulated Tissue
The flatfoot (right foot) geometries were generated from a series of CT images obtained from a
male volunteer, 38 years old, 168 cm in height, and 62 kg in weight. During the CT scan, a
custom-made foot loading device was used to fix the subject’s lower limb (Kido, 2011). The
boundaries of the bones and encapsulated tissue were segmented from CT images using MimicsⓇ
(Materialise Inc., Leuven, Belgium). The resulting boundary meshes were imported into a free
mesh editor called Meshlab (http:meshlab.sourceforge.net/). The meshes were smoothed, and node
densities were reduced to save computation time in the FE simulation. The simplified surface
meshes (Fig. 1) were saved as “.stl” files. A free MATLABⓇ file converter called
“CONVERT_stl_to_sat.m” was then used to convert the “.stl” files into “.sat” files, which can be
imported into Abaqus. The bone structure consists of 17 segments: tibia, fibula, talus, calcaneus,
cuboid, navicular, cuneiform (3 merged into 1), 5 metatarsals, and 5 phalanges (proximal, middle,
and distal phalanges were merged together). These bone instances were cut out from the tissue
instance to generate an encapsulated tissue instance surrounding the bones.
(a) (b)
Figure 1. Surface geometrical meshes of (a) the bone structure and (b) the
encapsulated soft tissue.
2.2 Geometry Generation of Ligaments, Plantar Fascia, and Tendons

To connect the bones to each other, and apply tendon forces to certain bones, the modeling of
ligaments, plantar fascia, and tendons is necessary. However, from CT images, it is difficult to
discriminate ligaments, plantar fascia, and tendons from the surrounding tissue. To this end, the
geometries of these tissues were generated manually by referring to an anatomy book (Netter,
2011) and also by following the suggestions of foot surgeons. Two different methods were used to

model these tissues. One method models these tissues with simple wire parts (Fig. 2), and the
second method uses 3D solid parts (Fig. 3). All of the foot models found in the literature use the
first method, in which the end nodes of the wire parts were manually selected to approximately
represent the origin and insertion positions of the ligaments and plantar fascia. The joint capsules
connecting metatarsals and phalanges were modeled by four wire parts surrounding the two bones.
Plantar fascia was divided into 6 separate wire parts connecting the calcaneus and the metatarsals,
in which 2 wire parts were used to connect the calcaneus and two sesamoid bones on the 1st
metatarsal bone. Tendons were not explicitly modeled in this method. Instead, tendon forces were
applied directly on the appropriate nodes on corresponding bones. A total of 82 wire parts were
generated in this model to represent the ligaments and plantar fascia.
(a) (b)
Figure 2. (a) Tarsal and (b) plantar views of the generated wire ligaments (red) and
plantar fascia (yellow)
(a) (b)
Figure 3. (a) Tarsal and (b) plantar views of the generated 3D solid ligaments (red),
plantar fascia (yellow), and tendons (cyan).
The first method has the advantage of easy construction, but simplifying ligaments and plantar
fascia into wires might lead to unrealistic results. Therefore, the second method was proposed, in
which these tissues were constructed as 3D solid parts. First, several nodes were manually selected
on the neighboring bones to approximate the origin and insertion positions of the ligament or

fascia. The selected nodes were then connected with each other to construct a 3D surface mesh to
form the ligament (red in Fig. 2) or plantar fascia (yellow in Fig. 2). Tendons (cyan in Fig. 2) were
modeled in the same way, and the insertion end was represented by several nodes selected from
the corresponding bone surface. The other end, which represents the cut position of the tendon,
was manually constructed using a surface with the normal direction aligning with the loading
direction of the tendon force. The joint capsules were also divided into 4 solid parts, and they
could cover almost 360 degrees around the joint. The plantar fascia was also divided into 6 parts,
and these parts could cover a large area on the foot bottom. Modeling with 3D solid parts gives us
a more realistic representation of the ligaments, plantar fascia, and tendons, but the modeling
process is more complicated than using wire parts. The performance of both methods will be
demonstrated in this paper. In addition, to simulate balanced standing, a horizontal ground plate
was created directly under the flatfoot model to support it.
2.3 Meshing the Parts

All of the solid parts, except the ground plate, were modeled with a 4-node linear tetrahedron
“C3D4” element, and the wire parts were modeled with a 2-node linear truss “T3D2” element. The
ground plate was modeled with an 8-node linear brick “C3D8R” element. The total number of
nodes for the first and second models is 33,259 and 40,011, respectively. The total number of
elements for both models is 131,309 and 148,108, respectively.
2.4 Material Properties

All of the materials were considered isotropic and linearly elastic except for the encapsulated
tissue, which was modeled with both linear elastic material and nonlinear elastic material
(hyperelastic) properties. Linearly elastic properties were listed in Table 1, and the hyperelastic
properties (the second order polynomial model) of the encapsulated tissue were listed in Table 2,
where the referenced literature was also indicated.
Table 1. Linearly elastic material properties
Component Young’s Modulus E (MPa) Poisson’s Ratio 𝜸
Bone 7,300 (Cheung, 2006, Qiu, 2011) 0.3 (Cheung, 2006, Qiu, 2011)
Ligament 260 (Cheung, 2006, Qiu, 2011) 0.4 (Cheung, 2006, Qiu, 2011)
Plantar Fascia 350 (Cheung, 2006, Qiu, 2011) 0.4 (Cheung, 2006, Qiu, 2011)
Tendon 1,500 (Peltonen, 2010) 0.4 (Peltonen, 2010)
Encapsulated Tissue 2.49 (Isvilanonda, 2012) 0.49 (Isvilanonda, 2012)
Ground 17,000 (Cheung, 2006, Qiu, 2011) 0.1 (Cheung, 2006, Qiu, 2011)
Table 2. Hyperelastic properties (nonlinear material)

C10 (MPa) C01 (MPa) C20 (MPa) C11 (MPa) C02 (MPa) D1 (MPa-1) D2 (MPa-1)
0.08556 -0.05841 0.03900 -0.02319 0.00851 3.65273 0.00000
The data were referenced from the literature (Cheung, 2005; Cheung, 2006; Qiu, 2011).
2.5 Contact Interactions and Constraints

Contact interactions among neighboring bones and between the encapsulated tissue and the
ground plate were modeled using penalty tangential behavior with a friction coefficient of 0.6 and
the discretization method is defined as “surface to surface”. A total of 21 contact pairs were
defined. Tie constraints were created to connect ligaments and plantar fascia to corresponding

bones and also to connect bones and the encapsulated tissue. A total of 183 tie constraints were
defined in the models.
2.6 Loading and Boundary Conditions

As shown in Fig. 4, to simulate balanced standing, the top surfaces of the tibia, fibula, and
encapsulated tissue were fixed in space. The ground reaction force and the Achilles tendon force
were applied in the model. The ground reaction force was applied as the pressure load with half of
the body weight (32 kg) acting on the ground surface underneath. Only the Achilles tendon force
was considered, and the forces on the other tendons were neglected (Cheung, 2006). The Achilles
force was assumed to be 50% of the ground reaction force (Lewis, 2008) which is half of the body
weight. Therefore, the Achilles force is calculated as 25% of the body weight, which is 151.9 N.
Both loads were applied as a ramp function of time.
fixed surfaces
Achilles force
loading pressure
(a) (b)
Figure 4. (a) Loading and boundary conditions defined in the model and (b)
measured plantar stress distribution.
2.7 The Measurement of Plantar Stress during Balanced Standing

To validate the FE model, experiments of balanced standing were performed with the same subject
for the CT scan. The measurements were obtained using an Emed-MⓇ pressure platform system.
Three measurements were taken on both feet, and the average stress of the right foot was
calculated (Fig. 4b) and compared with the simulation results.
2.8 Sensitivity Studies of Linear Material Property and Boundary Condition

The material properties of the encapsulated tissue significantly affects the stress distribution.
Therefore, sensitivity analysis was conducted to evaluate the sensitivity of the Young’s Modulus

of the encapulated tissue. Different Young’s Moduli, as shown in Table 3, were evaluated, in
which E1 = 2.49 MPa is used from the literature (Isvilanonda, 2012), and E2, E3, E4 were selected
as 1/2, 1/5, and 1/10 of E1. The reason why we only focus on the values smaller than the original
value (E1) is because we found that E1 is a very large value and deformation on the foot bottom is
very small during balance standing. The deformation results will be shown in the next section.
In the initial state of the model, the flatfoot bottom is parallel to the surface of the ground plate and
the loading direction of the ground plate is perpendicular to this surface. However, this loading
direction may change during the simulation of balanced standing, similar to the way human beings
may adjust their longitudinal axis to achieve a stable standing position. Therefore, in the
simulation, three different boundary conditions of the ground plate, as shown in Table 4, were
simulated to evaluate the differences. In the first case (BC 1), the ground plate was loaded along
the longitudinal axis (z-axis) and no other motions were allowed. In the second case (BC 2), the
ground plate was loaded longitudinally while it was also allowed to rotate around the coronal axis
(x-axis). In the third case (BC 3), the ground plate was allowed to move in all 6 degrees of
freedom.
Table 3. Different Young’s Modulus of the encapsulated tissue

Case E1 E2 E3 E4
Young’s Modulus E (MPa) 2.49 1.245 0.498 0.249
Table 4. Different boundary conditions of the ground plate

Case BC 1 BC 2 BC 3
Boundary U1=0, U2=0, U3=Free, U1=0, U2=Free, U3=Free, U1=Free, U2=Free, U3=Free,
Conditions UR1=0, UR2=0, UR3=0 UR1=Free, UR2=0, UR3=0 UR1=Free, UR2=Free, UR3=Free
U1, U2, and U3 indicate translation motion along the x-, y-, and z-axis. UR1, UR2, and UR3 indicate the rotation
around the x-, y-, and z-axis. The x-, y-, and z-axis are aligned with the coronal, the sagittal, and the longitudinal
axis, respectively.
2.9 Solver Information and Performance Metrics

Three steps were defined in Abaqus/Standard for the balanced standing simulation. They are:
initial, contact (1sec), and load (1sec) steps. For simulations with nonlinear properties of soft
tissue, geometric nonlinearity flag “Nlgeom” was turned on, but not for simulations with linear
properties. The total number of the variables in the model is 127380. The total CPU time is
2516.8sec and 2848.4sec, respectively, for models with linear and nonlinear properties of the
encapulated tissue.

The simulation results of plantar stress in different models were shown in Fig. 5. We found that
modeling the ligaments and plantar fascia with either wire or solid parts does not yield an
appreciable difference in the stress distribution. This finding gives us more confidence in using the
wire parts to model these tissues because the construction of the wire parts is less time-consuming.

Comparing the results between the linear and nonlinear models, we found that the stress was
distributed more evenly in the nonlinear model but was more concentrated in the middle and rear
of the foot in the linear model. We also found that the linear elastic material is much harder than
the nonlinear hyperelastic material. The maximal displacements of the reference node on the
different models are: 0.1525 mm in Fig. 5a, 0.1339 mm in Fig. 5b, and 1.3422 mm in Fig. 5c. The
reference
node
(a) (b) (c)

Figure 5. The simulation results of plantar stress from different models: (a) linear
elastic tissue with wire ligaments and plantar fascia, (b) linear elastic tissue with
3D solid ligaments and plantar fascia, and (c) nonlinear hyperelastic tissue with 3D
solid ligaments and plantar fascia.
displacement is almost 10 times larger in the nonlinear model compared with the linear model.
The results of the parametrical studies will be discussed later in the paper. Furthermore, we found
that the simulation results (Fig. 5) were not consistent with the experimental measurements (Fig.
4b). In our experiments, plantar stress was distributed mostly in the forefoot and heel areas, and
some stress also appeared in the big-toe area. However, in the simulation results with linear
models, higher plantar stress was distributed mostly in the middle and heel areas, and slightly in
the forefoot area under the sesamoid bones. Two reasons may cause these differences. One is the
geometry of the mesh. To obtain a relatively stable initial state, the CT images were taken while
the subject was lying on the CT table with his lower limb fixed by a custom-made device. While
the measurements were taken, both feet were subjected to 5.7±2.6% of the subject’s body weight
(Kido, 2011). This initial loading makes the initial geometry of the plantar surface relatively flat,
rather than its natural curved surface in which the forefoot and heel areas usually have a lower
profile compared with the middle foot area. This fact explains why a greater amount of stress
occurs at the forefoot area in the experiments but the middle foot area in simulation. Another
reason concerns the stress on the big toe. Stress on big toe appears in the experiment but not in the
simulation (linear model). This finding may be caused by the ground loading direction or the

ground boundary condition, which will be discussed later. If we compare Fig. 5c to Fig. 4b, we
found that the nonlinear model results in some stress on the big toe as observed in the experiment,
but there was also very high stress at the fourth and fifth toes, which is not a normal occurrence.
This observation was also caused by the initial mesh. There are very thin tissue layers in between
the phalange distal bones and the ground plate, especially in the fourth and fifth toe areas. When
these thin layers are in contact with the ground plate, the stress increases more quickly than in
other parts of the tissue. Figure 6 shows a comparison of simulated plantar stress using different
material properties of the encapsulated tissue and with different boundary conditions of the ground
plate. We found that if the encapsulated tissue became softer, the plantar stress would distribute
more evenly, the contact area would slightly increase, and the maximum stress would decrease for
all three cases of boundary conditions. The maximal displacements of the reference node in the
boundary condition BC 1 are 0.1525 mm, 0.2574 mm, 0.5249 mm, and 0.8848 mm along with a
decrease of Young’s Modulus. Even the softest tissue yielded a smaller displacement than the
model with nonlinear hyperelastic tissue. Therefore, we believe that the original linear elasticity of
2.49 MPa for the encapsulated tissue is a relatively large and inappropriate value. From Fig. 6, we
also found that the boundary condition of the ground plate affects the plantar stress distribution.
Allowing for the rotation motion of the ground plate shifts the plantar stress towards the forefoot
and distal areas and a greater consistency with the experimental measurements. With a softer
encapsulated tissue, boundary condition case 2 and case 3 do not result in an appreciable
difference in plantar stress.
Figure 7 shows the simulated plantar stress with different boundary conditions for the model with
a hyperelastic encapsulated tissue. We again found that the rotation of the ground plate affects the
plantar stress, and the boundary condition BC 3 shows the best performance. If we compare Fig. 7
with Fig. 6, we found that the nonlinear model yields better results compared with the linear
model. In addition, we found that the peak stress in our simulations is quite small compared with
the experimental measurements and published data. Peak stresses in our simulations are 28.34
KPa, 27.85 KPa, and 20.22 KPa for the results shown in Fig. 5a, 5b, and 5c, respectively.
However, the peak stress in the experimental measurements is 165 KPa. Published data on
simulated peak stress in a healthy foot were found to be 131 KPa (Antunes, 2010), 168 KPa (Chen,
2001), 198 KPa (Qiu, 2011), and 230 KPa (Cheung, 2005), respectively. We believe that this
difference is also caused by the initial mesh geometry of the flat plantar surface, which increases
the contact area of the foot when it comes in contact with the ground. As a result, the stress
amplitude on the plantar surface and the peak stress are decreased. This discrepancy could be
improved by using foot geometry with a natural curved plantar surface. However, it is difficult to
ensure the stability of the foot during a CT scan. Such stability will be one of our future tasks.
E1 = 2.49 MPa E2 = 1.245 MPa E3 = 0.498 MPa E4 = 0.249 MPa

BC
1
BC
2
BC
3
Figure 6. Simulated plantar stress distribution with different material properties (E)
of encapsulated tissue and different boundary conditions (BC) of the ground plate.

(a) (b) (c)
Figure 7. Simulated plantar stress with different boundary conditions: (a) BC 1, (b)
BC 2, and (c) BC 3 for the model with nonlinear hyperelastic tissue.
4. Conclusion
To better study flatfoot biomechanics and improve the decision-making for flatfoot surgery, a
computer model of flatfoot with detailed anatomical structures is one of the options. In this study,
a 3D FE model of a flatfoot was developed using the 3D reconstruction of CT images. Two
methods for manually modeling ligaments, plantar fascia, and tendons were introduced. Balanced
standing simulations were performed using different tissue properties and different boundary
conditions. The simulation results of plantar stresses were compared with experimental
measurements to validate the proposed models. Through the comparisons, three conclusions were
made. First, modeling ligaments and plantar fascia with wire parts (truss element) has almost the
same performance compared with 3D solid parts. Therefore, modeling with wire parts will be used
in our future work because it is less time-consuming for model construction. Second, the material
properties of the encapsulated tissue and the boundary condition of the ground plate affect the
plantar stress distribution. The boundary condition BC 3 yields the best performance in both the
linear and nonlinear models. Compared with the linear model, results of the nonlinear model are
more consistent with the experimental measurements. Therefore, the nonlinear model will be used
in our future work. Third, simulation results using current geometry with a flat plantar surface are
not consistent with the experimental measurements. A relatively curved plantar surface will
improve the consistency to some extent and will be investigated in our future work. Modeling and
simulation of a healthy foot will also be conducted to compare the difference with a flatfoot. After
validating our flatfoot model, surgical simulation, such as MCO and LCL, will be performed and
we hope that these simulations will contribute to the development of an optimal surgery plan and a
tailor-made surgery for individual patients.

5. References
1. Antunes, P. J., Dias, G. R., Coelho, A. T., Rebelo, F., Pereira, T., “Non-Linear Finite Element
Modeling of Anatomically Detailed 3D Foot Model,” Online Available.
2. Chen, W., Tang, F., and Ju, C., “Stress Distribution of the Foot during Mide-Stance to Push-
Off in Barefoot Gait: A 3-D Finite Element Analysis,” Clinical Biomechanics, vol. 16, no. 7,
pp. 614-620, 2001.
3. Cheung, J. T. M., Zhang, M., Leung, A. K. L., and Fan, Y., “Three-Dimensional Finite
Element Analysis of the Foot During Standing-A Material Sensitivity Study,” Journal of
Biomechanics, vol. 38, no. 5, pp. 1045-1054, May, 2005.
4. Cheung, J. T. M., Zhang, M., “Finite Element Modeling of the Human Foot and Footwear,”
ABAQUS Users’ Conference, pp. 145-159, 2006.
5. Isvilanonda, V., Dengler, E., Iaquinto, M., Sangeorzan, B. J., and Ledoux, W. R., “Finite
Element Analysis of the Foot: Model Validation and Comparison between Two Common
Treatments, of the Clawed Hallux Deformity,” Clinical Biomechanics, vol. 27, no. 8, pp. 837-
844, 2012.
6. Kido, M., Ikoma, K., Imai, K., Maki, M., Takatori, R., Tokunaga, D., Inoue, N., and Kubo, T.,
“Load Response of the Tarsal Bones in Patients with Flatfoot Deformity: In Vivo 3D Study,”
Foot Ankle Int., vol. 32, no. 11, pp. 1017-1022, Nov., 2011.
7. Kitaoka, H. B., Luo, Z., and An, K., “Three-Dimensional Analysis of Flatfoot Defomity:
Cadaver Study,” Foot & Ankle International, vol. 19, no. 7, pp. 447-451, 1998.
8. Lewis, G. S., “Computational Modeling of the Mechanics of Flatfoot Deformity and Its
Surgical Corrections,” Ph.D. Dissertation, Pennsylvania State University, 2008.
9. Netter, F. H., “Atlas of Human Anatomy,” Fifth Edition, Elsevier, pp. 51-525, 2011.
10. Niu, W., Yang, Y., Fan, Y., Ding, Z., and Yu, G., “Experimental Modeling and
Biomechanical Measurement of Flatfoot Deformity” in Proc. 7th Asian-Pacific Conference on
Medical and Biological Engineering (IFMBE), vol. 19, pp. 133-138, 2008.
11. Ozen, M., Sayman, O., and Havitcioglu, H., “Modeling and Stress Analyses of A Normal
Foot-Ankle and A Prosthetic Foot-Ankle Complex,” Acta of Bioengineering and
Biomechanics, vol. 15, no. 3, pp. 19-27, 2013.
12. Peltonen, J., Cronin, N. J., Avela, J., and Finni, T., “In Vivo Mechanical Response of Human
Achilles Tendon to A Single Bout of Hopping Exercise,” The Journal of Experimental
Biology, vol. 213, pp. 1259-1265, 2010.
13. Qiu, T., Teo, E., Yan, Y., and Lei, W., “Finite Element Modeling of a 3D Coupled Foot-Boot
Model,” Medical Engineering & Physics, vol. 33, no. 10, pp. 1228-1233, Dec., 2011.
14. Wang, Z., Imai, K., Kido, M., Ikoma, K., and Hirai, S., “A Finite Element Model of Flatfoot
(Pes Planus) for Improving Surgical Plan,” Proc. In 36th Annual International Conference of
the IEEE Engineering in Medicine & Biology Society, pp. 844-847, Chicago, Aug., 2014.

TRANSPORTATION
& MOBILTITY

IDIADA’s Virtual Proving Ground (VPG) for
durability analysis
Jordi Arbiol, Xavier Armengol, Jose Antonio Muñoz, Inés Lama, and Enric
Aramburu
Applus IDIADA
Abstract: It is usual that the loads applied on structural elements are in fluctuating situations.
Sometimes the component can break under its ultimate or its yield strength. This breakage
situation is due to the fluctuating loads that have been applied over long periods. For this purpose
this breakage type is called “fatigue breaking”. Fatigue breaking is a consequence of crack
growth. When no prototypes are available, it is necessary to calculate the life prediction as soon
as possible. Simulation through the VPG is increasingly being used for life prediction. The present
work shows a methodology that uses a finite element model (FEM) that runs on a proving ground.
In this case, no MBS (multibody software) model is used to obtain the loads on the structure.
Because the whole simulation of the finite element model needs a lot of time, a co-simulation is
carried out. An implicit FE solver is used for problems in which the response is moderately
nonlinear. This implicit method is computationally expensive. On the other hand, an explicit FE
solver is ideal for modelling highly nonlinear problems such as changes in contact surfaces. This
explicit method is relatively inexpensive. The ideal situation is to use the correct solver where it
will be needed. Components with moderate nonlinear behaviour (Body, Suspension) will be solved
using the implicit method. Explicit method will be used to model the impacts of the tyre against the
Virtual Proving Ground. Fundamental parts of this methodology are the digitalization of
IDIADA’s proving ground, its implementation as a rigid body and co-simulation.
Keywords: VPG (Virtual Proving Ground), Fatigue, Durability, Connectors, Co-simulation.
1. Introduction
The Virtual Proving Ground becomes the most accurate methodology to estimate suspension
component durability and is, also, useful to know the component loads on the finite element
model.
Several processes are used for fatigue life prediction. Some of them come from tests and others
come from virtual inputs. Some of these methods are:
• Equivalent quasi-static
• Unitary responses
• VPG using MBS
• VPG using FEM

1.1. Equivalent quasi-static
This methodology is a simple one, but it is effective in the early stages of the vehicle’s
development phase. It permits several loops with less CPU time consuming. The base of this
process consists of applying quasi-static inputs.
These inputs are very easy ones. It is necessary to define different fatigue cycles. Each of them
comes from two different states:
• turn left & turn right
• braking & accelerating
• vertical oscillations
Once they are defined, a fatigue postprocess is performed. The quasi-static method does not have
to take into account the dynamic effects. For this reason it is required to perform a previous modal
analysis to assure that the frequency of the components will not be excited by the inputs. Figure 1
shows the procedure to be followed.
FEM model
1st Loop optimization
Material test Counter

measures
no
Modal analysis Target
2nd Loop optimization
yes Counter
measures
Quasi-Static no
Target
(Fatigue loadcases)
yes
Proposed
geometry
Figure 1. Procedure in quasi-static load cases
1.2. Unitary responses

If there is the information of the forces acting over time on a component, it is possible to obtain
the time history of stresses by a linear superposition of the base unit response.
The base of unitary response is obtained applying unitary forces (1N) and moments (1Nmm) on
each direction separately. Figure 2 shows the inputs for the unitary responses on a connection
point for x, y and z directions.

Figure 2. Unitary loads on a connection point
To know stress levels on the component, it is necessary to know the stresses for a unitary load. By
multiplying these values by the force of the test at each time step, a stress state can be obtained. A
linear superposition of the unitary responses is performed:
𝑟 6
[𝜎]𝑡 = � �� 𝜎𝑖𝑗𝑢 �. �𝐹𝑖𝑗 � ��

𝑡
𝑖=1 𝑗=1
where,
𝜎𝑖𝑗𝑢 is the stress unitary response on the i hardpoint due to the unitary force that
acts on the j direction
�𝐹𝑖𝑗 �𝑡 is the j component force of the force acting on the i hardpoint at instant t
r is the total number of hardpoints
1.3. VPG using MBS

This method has less limitation than the previous one, because all of the modes of the components
are taken into account.
Running an MBS vehicle model on a digitalized track, it is possible to obtain the contribution of
each mode at each time step on the flexile bodies and a modal superposition can be performed:

q7(t) · +q8(t) · +q9(t) · +q10(t) · +···
Mode 7 Mode 8 Mode 9 Mode 10
Figure 3. Superposition of modes
The stresses come from the simultaneous application of all of the modes, at time step t, and it is
calculated (supposing elastic behaviour) as follows:
𝑟
[𝜎]𝑡 = ��[𝜎𝑖𝑢 ]. 𝑞𝑖𝑡 �

𝑖=1
where,
𝜎𝑖𝑢 is the stress unitary response for the i mode
𝑞𝑖𝑡 is the modal contribution of mode i at instant t
r is the total number of modes considered in the base
1.4. VPG using FEM

For simulations of the complete vehicle, the length of mesh elements may have an important
variation range. Currently, the tendency is to develop models with a very large number of
elements with a short length.
For the explicit method, the time step will be limited for the small element length of mesh and we
would need a large number of increments to calculate response of the structure for the desired time
interval. For the implicit method, a big time increment would not be able to accurately capture the
response in mesh model. The use of a single implicit or explicit method with a uniform time step
will be computationally inefficient.
Because explicit and implicit methods have advantages and disadvantages, a co-simulation process
between them has been used in this work.
2. Co-simulation
The best solution for this problem is to use a methodology that divides the model in two different
parts: the first will be solved with implicit methodology and the second with the explicit method.

Decomposition is given in the way that each region contains elements with similar requirements of
stability and precision. Both parts are solved independently and individual solutions are coupled
with each other to ensure continuity of global solution through the interface between the parts
(figure 1). Decomposition is performed to divide the study area into subdomains and to process in
parallel. A subdomain will be calculated with implicit methodology and another with an explicit
one.
This concept allows the use of different time steps for each region of the mesh, preserving the
accuracy and stability from individual parts. The advantage of this system is the use of big time
steps for implicit dynamic analysis. It is desirable that the total number of interface nodes be
small.
The implicit scheme is very inefficient when there are several contact areas that change from open
to closed or vice versa. For this reason, tyre and road contacts are simulated using the explicit
scheme. The vehicle body and suspension are solved using the implicit dynamic method and they
are modelled as substructures. Wheel centres are the co-simulation interface nodes.
Implicit Static (initial state)
Co-Simulation
Implicit (dynamic) Explicit (dynamic)
Figure 4. Implicit & Explicit subdomains
3. Description of the model

The main fields of study for fatigue prediction analysis are:

• Tyre modelling
• Modelling of body
• Modelling of suspension and components
• Assembling the implicit model
• Extraction of calculation inputs
3.1. Tyre modelling
As the minimum time increment of the explicit solver is governed by tyre, the model used to
simulate the response of the tyre represents a significant challenge, and the main issue in such an
approach is to simulate at the best level the tyre/road interaction.
Two versions of the tyre model were developed, as the front and rear tyres are different.
The front tyre is a 245/35R20. Its maximum load is 690 kg with a pressure of 0.29MPa.
The rear tyre is a 295/30R20. Its maximum load is 825 kg with a pressure of 0.29MPa.
The different components of the tyre are shown in figure 5.
Figure 5. Part description for tyre model

The front tyre and the rear tyre have different geometries, but both models have the same materials
and properties. They have about 7200 elements. A radial stiffness of 400N/mm is considered.
The upper tread and lower tread are modelled using solid elements, the chafer is modelled using
shell elements, and they are made of hyperelastic material. The inner tread, sidewall, wheel and
wheel flange are modelled using shell elements, and they are made of elastic material.
The solid elements used in the model are of the type C3D8R and C3D6.

• C3D8R: 8-node linear brick, reduced integration with hourglass control.
• C3D6: 6-node linear triangular prism, reduced integration with hourglass control. The
shell elements used in the model are of the type S4R.
• S4R: 4-node general-purpose shell, reduced integration with hourglass control, finite
membrane strains.
The enhanced hourglass control formulation was used for solid and shell. This formulation
provides improved coarse mesh accuracy with slightly higher computational cost. The enhanced
hourglass control approach represents a refinement of the pure stiffness method in which the
stiffness coefficients are based on the enhanced assumed strain method.
Reinforcements are included on membrane elements inside the rubber material as embedded.
Figure 6 shows a reinforce definition.
Figure 6. Plies of tyre

An initial static analysis for tyres, inflating, gravity load and vehicle preloads will be simulated.
With the *IMPORT capability, we can import the new tyre position for nodes and elements in the
dynamic explicit model.
3.2. Modelling of body
To connect all of the suspension component models to the chassis, some retained nodes are
defined.

Figure 7. Body substructure and retained nodes
3.3. Modelling of suspension components
It is important to optimize the size of shell elements: in areas which are outside of fatigue study
the nominal length of elements can be increased.
There are a number of good reasons to use substructures:
• Computational advantages: Efficiency is improved when the same substructure is used

multiple times. The stiffness calculation and substructure reduction are done only once;
however, the substructure itself can be used many times, resulting in significant savings
in computational effort. Substructuring can isolate possible changes outside substructures
to save time during reanalysis. During the design process large portions of the structure
will often remain unchanged; these portions can be isolated in a substructure to save the
computational effort involved in forming the stiffness of that part of the structure.
• Organizational advantages: Substructure libraries allow analysts to share substructures. In
large design projects large groups of engineers must often conduct analyses using the
same structures. Substructure libraries provide a clean and simple way of sharing
structural information. Figures 8 and 9 show front and rear suspension models.
The front and rear suspension components defined as substructures are:
• upper control arm
• lower control arm
• support damper bracket
• spindle

Figure 8. Front suspension description
Figure 9. Rear suspension description

3.4. Assembling the implicit model
The Abaqus Standard model of the co-simulation analysis consists of the front suspension
components and body; both are modelled using substructure techniques and will be assembled by
connector elements.
The connections between parts are modelled using connector elements (figure 10). Join/Revolute
connectors are used to model the hinges between the upper and lower control arms with chassis.
Struts are modelled using an axial connection specifying nonlinear elasticity and damping. Link
connectors are used to join steering and spindle. Coil springs have initial preload as given by their
respective spring load deflection curves.
Figure 10. Front suspension to body assy
3.5. Extraction of calculation inputs
A virtual model (VPG) from Pave fatigue road (figure 11), scanned with digital technologies was
developed. As a result of this technique, a high resolution surface model of this road has been
obtained. This will allow us to have a realistic 3D model for automate fatigue life prediction on
vehicle components.

a) Track b) Digitalized track
Figure 11. Pavé fatigue track
Modelling of the road surface is important. The simulation time depends on contact definition
between road surface and a car. In order to greatly reduce the number of elements on our road
model, only the part of the track that is in contact with tyres has been taken into account during the
simulation (figure 12).
Figure 12. FEM for Pave Fatigue track

Rigid elements R3D3 (3-node, triangular facet) and R3D4 (4-node, bilinear quadrilateral) have
been used to mesh road model (figure 12). Previous STL elements from 3D sampling road will be
remeshed in order to decrease the density of mesh elements.

4. Analysis procedure
Several steps have been performed for the simulation model:
• Tyre inflation
• Footprint analysis
• Assembly the Implicit/Explicit domains
• Applying velocity
A previous static analysis for tyre inflating and preloading will be done, in order to know initial
position for the preloaded complete vehicle model. Co-simulation regions across which data will
be exchanged during the co-simulation analysis are identified on each model at the location of the
wheel centre as shown in figure 13.
Figure 13. Assembling the model
5. Boundary conditions
The Pave fatigue road is commonly used from car manufacturers for the structural evaluation of
suspension components and car body. The simulation test is carried out with a constant velocity of
35 km/h. The Volare, FEM model car has been considered for the present study.
From the simulation point of view, during explicit-implicit analysis, the following initial
conditions have been imposed on the model:
• Initial velocity v=12500 mm/s , in X car direction applied to car.

• An angular initial velocity field ω = 36.4 rad/s for front tyres and ω = 36.3 rad/s for rear
tyres.
The following loads and boundary conditions are applied:
• Gravity load applied to the entire vehicle model, in order to guarantee the continuity with
the static equilibrium.
• Nodes of the fatigue road are totally constrained on all their 6 d.o.f with a rigid body.
• Tyres, front and rear, will be initially preloaded in order to support the total mass of
vehicle. An initial step for tyre inflation ( performed for the four wheels) at 0.29 MPa , a
second step to define the contact enforcement between the wheels and the road surface
and finally a third step with the preload force applied to front tyres Fz= 2930 N and
3914N for rear tyres.
6. Results
Due to the large dimensions, the complete model analysis is limited to a 100-metre fatigue road.
The mises stress history distribution of the finite element results, obtained from flexible bodies for
front and rear suspension and internal car structure, will be reported.
For front suspension components, the maximum Von Misses stress level is located on lower
control arm link support (figure 10).
Figure 10. Front suspension mises stresses at 0.2, 0.3 and 0.4 seconds
For rear suspension components, the maximum Von Misses stress level is located on upper control
arm support (figure 11).

Figure 11. Rear suspension mises stresses at 0.2, 0.3 and 0.4 seconds
Finally, through Virtual Lab software the fatigue life of the components will be estimated by
finding the life in cycles that corresponds to calculated stress amplitude by the intersection on the
SN curve.
References
1. Dixon, J.C.,”Tires, suspensions and handling”. Cambridge University Press, England, 1991.
2. D.F. Socie, M.R. Mitchell, and E.M. Caulfield, "Fundamentals Of Modern Fatigue Analysis";
College of Engineering, University of Illinois, 1977.
3. E. Duni, G. Toniato, “Vehicle Fatigue Load Prediction based on Finite Element TIRE/ROAD
Interaction implemented in an Integrated Implicit - Explicit Approach”. FIAT Group
Automobilies , 2008.
4. E. Duni, G. Monfrino, R. Saponaro, M. Caudano and F. Urbinati. “Numerical Simulation of
Full Vehicle Dynamic Behaviour Based on the Interaction Between Abaqus/Standard an
Explicit Codes”. FIAT Research Center, Orbassano (TO), Italy.
5. H.O. Fuchs; R.I. Stephens, "Metal Fatigue in Engineering"; Wiley,1980.
6. Reimpell, J., and H. Stoll, “Automotive Chassis-Engineering Principles”, Arnold,
England,1996.
7. Abaqus User´s Manual, Version 6.12.Dassault Systèmes Simulia Corp., Providence, RI,
USA.
Acknowledgments
We would like to express our gratitude to the employees of Abaqus Technical support from
Principia Spain for their professional assistance and patience in helping us to solve several
technical questions.

Thermal Management of Electronic Devices
used in Automotive Safety – A DoE approach
Vinod Kumar, Vinay Somashekhar and Srivathsa Jagalur
Autoliv India Private Limited, Bangalore, India
Abstract: Electronic devices used in automotive safety has a trend to reduce size in
packaging and increase performance of the product, both of which contribute to
exponential increase in power consumption of the system consequently increasing the
heat generation. Designers have very less choice on PCB board components but have
flexibility in designing the enclosures over the PCBs. This calls the need for an optimal
enclosure design to maximize the heat dissipation and to reduce the temperature well
below the allowable operational limit. Majority of heat generated by these devices get
dissipated by conduction and natural convection modes. Heat generated by the chip on
PCB due to power consumption is transferred to the cover and housing via heat sink and
thermal pads respectively by conduction. The surface film coefficients estimated from
the bench tests were used to simulate convection heat transfer from housing and cover to
ambient using Abaqus.
To arrive at an optimal house and cover design, a DoE approach is used with
geometrical, material and positioning as parameters. Efficient cover and housing design
ensuring better functionality with maximum heat dissipation capability in terms of
conduction and convection is achieved with minimum number of physical tests by DoE
approach of thermal analysis using Abaqus.
Keywords: Thermal management, Steady State Heat Transfer, Automotive safety, DoE
1. Introduction
Electronic equipment has made its way into practically every aspect of modern life,
from toys and appliances to high-power computers. The reliability of the electronics of
a system is a major factor in the overall reliability of the system. Electronic components
depend on the passage of electric current to perform their duties, and they become
potential sites for excessive heating, since the current flow through a resistance is
accompanied by heat generation. Continued miniaturization of electronic systems has
resulted in a dramatic increase in the amount of heat generated per unit volume,
comparable in magnitude to those encountered at nuclear reactors and the surface of the
sun. Unless properly designed and controlled, high rates of heat generation result in
high operating temperatures for electronic equipment, which jeopardizes its safety and
reliability. The failure rate of electronic equipment increases exponentially with

temperature. Also, the high thermal stresses in the solder joints of electronic
components mounted on circuit boards resulting from temperature variations are major
causes of failure. Therefore, thermal control has become increasingly important in the
design and operation of electronic equipment.
In the past, thermal analysis of electronic systems relied upon a combination of the
empirical data and analytical techniques which were based on regular geometries and
simple boundary conditions. For the analysis of more complex systems, electro-thermal
analogue representations of the heat transfer, based on thermal resistance network
methods were employed. Numerical schemes were also employed for the analysis of
conductive heat transfer problems, but these depended on the prescription of the
illusive convective heat transfer coefficient on the convective surfaces. Realistically the
conjugate problem can only be accurately represented by Computational Fluid
Dynamic (CFD) analysis that simultaneously solves the appropriate governing
equations for the solid and fluid domains, and couples both at the solid-fluid interface.
However to date, the predictive accuracy of CFD codes dedicated to the thermal
analysis of electronic equipment has not been comprehensively validated.
Consequently, CFD based thermal designs still require lot of experimental verification
during the design phase, thereby diminishing the potential of this design method to
reduce product development cycle times.
With ever-reducing product development cycle times preventing extensive prototyping,
increasing reliance is being placed on numerical predictive techniques, a need therefore
exists to comprehensively evaluate the predictive accuracy of these codes, so as to
anticipate what extent of concurrent experimentation would be required during the
design cycle. Although a thermal management system may make use of all modes of
heat transfer to maintain temperatures within their appropriate limits, the simplest form
is by conduction and convection heat transfer.
In this work authors wants to demonstrate the thermal management using steady state
heat transfer analysis involving conduction and convection modes in a simplified
manner. For thermal management FE simulation challenges lies in chip modelling,
PCB modelling, design of enclosures, thermal boundary conditions and positioning of
these electronic devices.
With increase in occupant safety consciousness in an automotive, the electronic control
of Air bags, breaking and seatbelt pre-tensioners have become standard equipment in
almost all the vehicles. The Figure 1 shows a typical safety control systems where such
electronic circuitry is used and the structure of one such circuit board shown in Figure 2.

Figure 1. Safety Controller
Figure 2. Circuit Board
To arrive at an optimal house and cover design, a DoE approach is used with
geometrical, material and positioning as parameters. Efficient cover and housing design
ensuring better functionality with maximum heat dissipation capability in terms of
conduction and convection is achieved with minimum number of physical tests by DoE
approach of thermal analysis using Abaqus.
2. Numerical Model
In this study, the prediction of a component’s thermal interaction is investigated on the
PCB. The current design is referred base model. Component is centrally located on the
PCB. The PCB is housed inside enclosures. The base model of the enclosure consists
four components outer fin, House, Inner heat sink and Case which is made of
aluminum material. The Figure 3 shows the base model of the Safety Control Device.

Figure 3. Safety Controller Base Model
2.1 Problem Description
The heat generated by electronic circuit in Safety Control Device must be dissipated
effectively to improve reliability and to keep the operating temperature of PCB
component well below the desired limit. Here chip (resistor) is the heat generating
component with operational limit of -35°C to 150°C. To serve this purpose of heat
dissipation, aluminum fins are mounted at different locations in order to minimize the
temperature of PCB at Chip PCB interface.
Heat generated by the chip is transferred to thermal gel which are cured in place to give
a permanent and durable thermal interface. In the uncured state, these materials are soft
enough to assemble components under low force and then cure to a harder state.
Thermal gel carries the heat to Inner heat sink, House and Outer fin in top side. Heat
flow path for current design is as shown in the Figure 4.

Figure 4. Heat Flow Path
2.2 Objective and Performance Parameter

Based on these house topologies, a methodical approach is employed to permit both the
modeling methodology applied and solver capability to be carefully evaluated. Different
configuration of the housing and case, poses the greatest challenge for numerical
prediction due to it generating a high degree of component thermal interaction.
Measured component junction temperature is used as the primary benchmark criterion,
while component and PCB surface temperature measurements serve to validate the
component-PCB modeling methodology. All component dimensions and constituent
material thermo-physical properties correspond to nominal vendor specifications.
Objective of this work is to reduce this component junction temperature by finding an
optimal house and cover design.
2.3 Modeling Approach

Keeping in mind the physical structure of Component shown in Figure 5, where the
core resisting layer which is sandwiched between a plastic or ceramic layer coating.
The Component is modelled as multi-layered with DC3D8 (Continuum Thermal
Bricks) elements in which the heat is generated at mid layer of resistor and is conducted
further. The Heat generation is modeled using surface heat flux.

Figure 5. Physical Structure of a Component
PCB is modelled as general purpose shell elements (DS3 or DS4) with thermal degrees
of freedom and orthotropic material property with higher thermal conductivity along in-
plane directions and least thermal conductivity along thickness direction. Thermal gel
and other components of the assembly is modelled with solid elements (DC3D8) with
appropriate connections assumed.
Conduction heat transfer happens due to contact between two components. Transfer of
heat depends on the surface finish of surfaces in contact. In simulation it is assumed to
have ideal contact between two components with no gaps in between and the two
components are tied to each other. Convective heat transfer coefficient for natural
convection is assumed 5-10 W/m2K
The thermal material properties of the components used are as listed in the Table 1.
Conductivity
Components Material
(W/mK)
House, Case, Inner
Aluminum 237
Heat sink, Outer fin
K11=32.7, K22 =32.7,
PCB Substrate (Ortho)
K33=0.28
IC Component Composite 0.16
Thermal Gel Thermal Gel 6
Table 1. Material Properties
3. Experimentation
A prototype of current baseline model was tested and temperatures on PCB, Case, Outer Fin
and house is measured using a thermocouple. The Figure 6 shows the measured temperatures
on different components of the assembly. Conditions assumed for the testing: Room temp =
25°C, Voltage =9 V, Current = 1.8 A, Safety Control Device position – vertical

Figure 6. Physical Test Results
4. Numerical Model Validation

The present simulation model is correlated with a bench test by tuning parameters like
Convective film coefficient and Convection surface.
4.1 Simulation at Room Temperature
The Figure 7 shows the thermal profile and maximum measured temperature on
different components of the system in room temperature when it reached steady state.
In Room temperature case initial temperature of the assembly is considered to be at
25°C

Figure 7. Temperature plot for Room Temperature case
The Table 2 shows the comparison of measured temperatures on different components

in simulation and physical test. The simulation results correlated well with the physical
test results. Thus providing a benchmark model to go for further experiments with
numerical model.
Component Test (max °C) Simulation (max °C)
House 54 57
Case 55 57
PCB (Junction
76 76
Temperature)
Table 2. Results for Room Temperature case
4.2 Simulation at High Temperature
The benchmark model is simulated in an ambience of higher temperature to see the

effect on the temperature rise at the junction. The Figure 8 shows the thermal profile

and maximum measured temperature on different components of the system at high
ambient temperature when it reached steady state. In high temperature case initial
temperature of the assembly is considered to be at 85°C
Figure 8. Temperature plot for High Temperature case
4.3 Limitations of Base model
Having multiple components for conduction of heat increases the mass of the enclosure
assembly, and makes the design a complex assembly. Increase in number of
components also increases the effort of material handling and lead time in assembly
process. When two or more mating components are assembled by means of screws, the
ability of transfer of heat over their contacting surfaces determined by certain
parameters of contacting solids such as surface roughness, flatness deviation, gap
thickness etc. which is difficult to avoid while manufacturing. And is also complicated
to model such losses in simulation.
This calls for need of reduction of number of components in enclosure design with
optimal conductive material and surface area for natural convection through DoE
approach.
With the current validated Simulation model, Design of Experiments were performed
in order the make the Baseline design more robust. The important criteria considered
are reduction of number of components, reduction of Assembly weight and increase of
Surface area in order to achieve minimum possible temperature on PCB and maximum
heat dissipation.

5. DoE results
The list of design iterations performed are as tabulated in the Table 3.
The Table 3 shows the brief structure of modified design highlighting key parameters
like Number of components, final mass of enclosure assembly and component junction
temperature on PCB at chip PCB interface. Design no. 4 gives an optimal solution with
lower junction temperature and with minimum number of components.
Component Component
No. of Assy. Junction Junction
Design
Model Assy. Enclosure mass Temp. [°C] Temp. [°C]
No.
Comps [gms] with ambient with ambient
at 25°C at 85°C
Base
4 285 76.17 132.02
Model
1 3 210 94.66 156.62
2 3 254 76.16 138.17
3 3 265 76.38 138.38
4 2 220 67.7 129.75
Table 3. DoE Results
6. Conclusion
With Abaqus authors were able to build a Simulation model to replicate the actual
physical model, which was further used to modify the design and arrive at final design
with lesser components, better functionality and with minimum design cost. Further the
modified model can be analyzed for CFD to see the effects of forced convection and
Radiation when number of components on the PCB is increased.

References
1. Abaqus User’s Manual, Version 6.13-1, Dassault Systémes Simulia Corp., Providence, RI.
2. Giuseppe Petrone & Giuliano Cammarata – Numerical Modelling for Thermal Design of
Electronic Equipments
3. Hiroyuki Komaki & et al., Development of the electronic "Safing" system for airbag ECUs
4. Mangesh D Shende (1) & Dr. Ashish Mahalle (2) - Cooling Of Electronic Equipments
with Heat Sink: A Review of Literature
5. McGraw Hill, Heat and Mass Transfer: Fundamentals and Applications - Chapter 15-6,
15-7, Cooling Of Electronic Equipment
6. Wikipedia -
http://en.wikipedia.org/wiki/Thermal_management_of_electronic_devices_and_systems
7. Z. J. Zuo, L. R. Hoover & A. L. Phillips – An integrated thermal architecture for thermal
management of high power electronics
Acknowledgments
The authors gratefully acknowledge Autoliv Electronics France and Autoliv Electronics
Germany for their technical support. The base model was built and tested at Autoliv
Electronics Germany. Authors would like to express their gratitude towards Autoliv
India, who funded this work and helped them to use the computational facilities to
undertake the numerical analyses.

Development of a Separator Plate for Valve Bodies
in a Vehicle Automatic Transmission
Harald Rebien, Tobias Tauscher, Rohit Ramkumar

DANA® Holding Corp., Power Technologies Group
Abstract
Object:
Development of a Separator Plate for Valve Bodies in a Vehicle Automatic Transmission
Automatic transmissions in light vehicles are complex hydraulic systems. The hydraulic control
unit (HCU) distributes oil to shift valves and sliders. The separator plate as a part of it needs to
seal all fluid channels and provides fluidic connections between the flanges.
Problem:
Nowadays the automatic transmissions need to be an economically optimized system. To meet CO2
goals, the transmission has to be more efficient. One possibility is to raise the operating pressure
in the hydraulic fluid. But this leads to a higher leakage rate and thus larger pressure pumps are
required. To overcome this dilemma several innovative solutions were needed to develop a new
generation of separator plates. These replace the conventional separator plates in new high
efficient automatic transmissions.
Lower leakage rates at higher pressure demand a more effective sealing solution. If this solution
works particularly efficient it could provide an opportunity to reduce the amount of bolts required
for the joint. This opens the possibility of more flexible design of the hydraulic channels.
Results:
With the help of Abaqus® it was possible to create a FEA model which described the real world
good enough to predict the sealing performance, flange and valve seat distortions at different
operating conditions. Confidence in the model brought us in the position to develop a new high
performance gasket. The sealing performance is much higher and can be adapted to the
customers’ design needs like hydraulic pressure, flange stiffness and bolt positions. Comparisons
with traditional separator plates confirmed that it is possible to reduce the number of bolts in the
joint. This leads to a better hydraulic channel design at lower cost.

Conclusion:
Using CAE technology helped us to create a simulation driven design. New innovative sealing
technology could be developed and applied to individual customer demands.
Value:
With the new innovative separator plates DANA® could convince customers to apply the new
technology in their products. Additionally this new innovative separator plate is nominated for the
PACE-Award.
Keywords:
automatic transmission, powertrain, drivetrain, automotive, bending, bolt load, CAE, deformation,
design optimization, FEA, forming, gasket, sealing pressure, transportation,

Body of Paper
a) Introduction
As a global technology leader in efficient power conveyance and management solution DANA®
develops products that enable their customers to achieve their sustainable objectives. In this case
DANA® was focused on improving the efficiency of automatic transmissions in light vehicles.
As many governments tighten environmental regulations, the automotive industry has to find
fields to reduce their CO2 output.
Figure 1. Global legislations permanently push down the CO2 limits. CO2 limit
[g/km]
The powertrain, including engine and output side, offers many options to bring down the amount
of CO2. On the one hand is the reduction of the fuel needed, on the other hand the better usage of
the power provided by the engine. This puts the focus on the gear box. While mechanical gears
have in general a high efficiency, automatic transmissions have more potential. The HCU needs
energy to actuate valves and sliders.
b) Functioning of Separator Plates

Automatic transmissions in light vehicles are complex hydraulic systems. To change gears, oil in
hydraulic channels needs to activate the right valves. These hydraulic channels are arranged
mainly in two valve body flanges. The separator plate separates the channels within each flange
and from one body flange to the other. Additionally the separator plate needs to provide at some
locations a fluidic connection between two channels. Beside this, the separator plate has to manage
low bolt forces, avoid distortion in the valve body flanges and has to be durable under all
conditions.

c) Benchmark
At the beginning of this innovation a benchmark showed different solutions and their pros and
cons. The conventional separator plates had
Low sealing potential due to low sealing features and recovery
High bolt load loss due to plastic deformations
Cross-channel leakage due to low sealing potential
Layout for low pressure due to used design and materials
Erosion and delaminating potential due to used materials
Contamination Risk (cleanliness) due to used materials and process
Examples:
Figure 2. Steel plate -> low sealing potential
Figure 3. Soft gasket plate -> low robustness

d) Development targets
The efficiency of transmissions depends on their torque. High efficiency transmissions need to
have the advantage at their most frequent driving conditions. This is at low torque.
Figure 4. Transmission efficiency

®
DANA ’s MLS transmission separator plates enable OEMs to have the following design features.
smaller oil pump, through lower leakage rate
higher pressure, through higher sealing performance
Lightweight design, through used layout and material
pressure storage, through very low leakage rate

This is enabled through
Improved transmission efficiency
o Supports smaller oil pumps
o Enables pressure storage
o Reduces package size
Lightweight design
o Fewer bolts
o Reduced component size and weight
Enhanced cleanliness
o Increased reliability
o Decreased warranty cost
e) Coating improvement
A drastic improvement in the coating development enables a softer sealing design. A soft and
more flexible coating is able to adapt to the surface roughness. Less pressure is required to press
the coating into the uneven surface.

steel plate soft plate partially coated MLS plate
exhaust gasket cylinder head gasket O-ring
Figure 5. Partially coated multi-layer steel transmission separator plates show best
sealing potential

Figure 6. Comparison of leakage rate
With the reduced pressure it is possible to use less bolts and apply less bending at the flanges. Low
bending is required to guarantee a reliable function of the valves and sliders in the flanges. The
goal is now to find a design which on one hand is able to seal all channels and on the other head
bends the flanges as little as possible. Only the custom-tailored sealing beads at the right place will
lead to an optimal gasket layout.
Figure 7. Load deflection curves of two different beads

f) Simulation
Abaqus®, as a strong CAE tool, helps us to achieve this goal. With Abaqus® we can do a lot of
upfront simulations to examine the parts that should be sealed. In doing so, we observe the
responses of the valve and actuator body and the separator plate of the HCU to different loads and
boundaries. These loads and boundaries can be bolt forces, plate thickness, material and operating
conditions like fluid pressures and temperatures. The responses are bending of the flanges,
cylindricity of the valve bores, sealing gaps and sealing pressures.
Figure 8. FEA-Model of a HCU of an automatic transmission
The separator plate is modeled with solid elements. In the sealing area they represent the loading-
and unloading characteristics of the beads. These loading- and unloading curves are measured at
our physical laboratory.
Figure 9. FEA-Model of the separator plate

When assembling the HCU it is important to know the bolt forces at the end of the bolt tightening
sequence. This way, we can use Abaqus® to simulate the assembling and examine the bolt load
loss during the tightening sequence. Besides the flanges and the tightening sequence, the gasket
also has an influence on the bolt forces after assembly. In the case presented in the following
chart, the bolts were tightened two times. The chart below shows the bolt load history and also
indicates that at the end of the tightening process, all bolts are tightened properly.
Figure 10. Example bolt load history
The following table shows the dynamic sealing gap for some plate thicknesses and materials. The
dynamic sealing gap is the change in sealing gap due to different operating conditions caused by
shifting different gears. Now, the challenge is to find a gasket layout that can follow the
deformations due to assembly and operating conditions. Further, the gasket has to provide enough
sealing pressure in all working conditions to avoid leakage.
Example results matrix of different distance layer plates
STEEL ALUMINUM
PLATE 1,0 mm 2,0 mm 3,0 mm 1,0 mm 2,0 mm 3,0 mm
THICKNESS
ASSEMBLY 0 0 0 0 0 0
GEAR_1 14,2 5,5 4,4 30,1 8,4 5,2
GEAR_2 1,8 1,9 1,9 2,8 1,7 1,7
GEAR_3 5,5 2,3 2,4 14,8 2,4 2,0
GEAR_4 3,1 1,5 1,5 9,0 1,5 1,4
BURST 17,8 8,2 7,2 35,6 11,0 7,7
PRESSURE
Figure 11. Maximum change in sealing gap in µm for different plate thicknesses
and materials

The application of a gasket to a new system is a process that has to deal with many issues like bolt
load, deformations and sealing pressure. Abaqus® is supporting us to find the best solution for our
customers in consensus with functionality, weight and cost.
Comparison of the DANA® MLS separator plate with a conventional separator plate:
The DANA® MLS (multi-layer steel) separator plate has two outer layers of steel (a top layer and
a bottom layer on each side of the distance layer) with stamped beads to seal the hydraulic
channels. The distance layer adds stiffness to support the sealing function of the outer layers.
Compared to conventional separator plates which use screen printed or molded rubber, or paper
gasket, the DANA® MLS provides a much higher sealing performance and durability.
®
Conventional plate with pressure holes DANA MLS plate with sufficient sealing pressure
Figure 12. Sealing pressure at working condition GEAR_1_3
®
Conventional plate with higher deformation DANA MLS plate with lower deformation
Figure 13. Deformations at working condition GEAR_1_3
A typical 3D simulation runs on 12-48 cores and takes 1-2 days on a HPC cluster.

Sealing bead optimization:
The primary function of the bead in a gasket is to provide adequate sealing pressure while also
providing adequate recovery due to the change in pressure. This can impact the durability of the
bead long term. To achieve optimum bead stiffness and recovery, DOE studies were completed
using ISight & Abaqus®. The objective of the study was to maximize the bead recovery based on
various tool parameters used to create the bead shape. The bead was formed, subjected to load-
unload cycles to measure recovery. Other factors studied include forming and assembly stresses in
the bead for the range of tooling parameters.
A simple 2D axis-symmetric model was used in Abaqus®, with MS Excel® component

used to change the tool design parameters. Output results were read from results file (.odb) by the
Abaqus® component in ISight.
Figure 14. Optimization loop
Results
Figure 15. Contact pressure at full compression
Figure 16. Contact pressure at maximum unloading

Figure 17. Comparison of load – unload curves between FEA and laboratory
Figure 18. Results of bead recovery vs. tool height

Results indicate that tool height is the factor that contributes most to the variance of the recovery
of the sealing bead. The results from ISight study were checked with physical laboratory data for
comparison. There seemed to be good correlation in the results between the Abaqus® predicted
recovery and lab tested samples. The variation in the results were attributed to variation of the
coating thickness which may have influence on friction between the sealing bead and the test
plates.
g) Summary and conclusion

The teamwork of different development disciplines like CAD, FEA, testing, material research,
brought up a new high performance sealing solution. As a part of the development Abaqus®
enabled the team to understand the nature of transmission plates and thus brought the team in a
position to improve and optimize the performance of the gasket.
The new DANA® partially coated multi-layer steel transmission plate meets the highest customer
sealing requirements and is accepted by our key customers.

System Modeling and Simulation at Dana Holding
Frank Popielas
Dana Holding Corporation, Lisle, IL, USA
Abstract: System Modeling and Simulation (SMS) is the next logical step within Dana’s
engineering community to achieve a fully integrated, front-end driven, optimized engineering
ecosystem. We believe this evolution will enable Dana to maintain continued leadership in the
industry for delivering product innovation, quality and customer responsiveness. In order to
achieve this vision in an efficient and coordinated way, we at Dana identified the basic steps we
have to take, determine which engineering areas are critical participants, and, decide on which
technology pillars we have to focus our attention. At the very beginning of this journey we
understood that this means a step change for a new way of thinking as well as a cultural change
within our engineering community, company and extended enterprise.
In this paper we explain what SMS means to Dana, how we approach this topic, and what
technologies we are investigating in order to achieve this new level of engineering. Since the
process is very complex, we decided to start with a business value assessment (BVA). This
engineering assessment was undertaken to understand the current status of our engineering
organizations and existing technology, and then identify the gaps to where we want to be with our
business. This helped to formulate a foundational strategy and long-term roadmap. With an
existing strong focus on the CAE environment this also meant we needed, as previously
mentioned, enhance the mindset and our global culture to make the next step in our systems
evolution. This only can be achieved through a dedicated effort to educate and engage the people
throughout the process Therefore, as a very important topic we will also discuss the meaning of
simulation, CAE in general, system engineering (SE), model-based system engineering (MBSE) in
the context of SMS for this new environment.
Introduction
Innovation and the development of new technologies are no longer ethereal marketing promises,
but have evolved into critical requirements to effectively compete in the automotive industry. Of-
course, it’s not just something special to this industry but across all the different areas in general.
But, the competitive pressure in the automotive industry in particular pushed engineering into new
thinking, applying new techniques and approaches. While all industries are feeling the competitive
pressure, the automotive market is driven heavily by individual consumers in mass to produce an
environment where a technical shortcoming can result in massive changes in market share for
original equipment manufacturers (OEM’s). The highly competitive nature of the market has
forced the individual suppliers and OEM’s to invest heavily in new thinking, new techniques, and
new approaches to try to achieve and sustain a market lead. Presented recently ([1], [4]) that in

order to be considered a key supplier one need not just to be a full service supplier but a
development partner. (Figure 1).
From Supplier to Development Partner
Development Partner
SLM
CAE Development in
Partnership with Customer Analysis
Driven Design
CAE for Mfg
Full Service System CAE
Supplier System Testing
Value
System Design
Supplier Dyno Testing
Dedicated Program Managers
Product FEA
On Site Engineers
Product Testing
Product Design
© Dana 2013
Figure 1: From Supplier to Development Partner
Initially, we approached this concept from a computer aided engineering (CAE) point of view.
We found that companies that did not incorporate CAE could not stay competitive in the long run.
CAE, however, is not the complete story because this point of view focuses almost entirely on the
structural design and performance of individual components. In order to go beyond structural
simulation we had to expand our focus across additional engineering disciplines, such as
computational fluid dynamics (CFD). As our use of CFD matured we began to relate this
discipline back to structural mechanics through the use of fluid structural interaction and
multiphysics in general to execute cosimulation. Cosimulation was the essential basis to be able
to contribute as a development partner and add value to the final OEM. (Figure 2; [2] [4])

Changes in the Automotive Supplier Market
OEM Consolidation Supplier Consolidation

Number of OEM’s Number of Suppliers Number of Suppliers / Module
16 30000
7-8
174 System / Modul 4-6

10 Suppliers with
2000 Sub Suppliers
8 3-4
8000
2175
2000 2005 2010 1988 1998 2008 2000 2005 2010
Shift related to Value Add and Competency

Value Add Development Competency
30% 25% 20% OEM’s

49%
57%
67%
75% 80%
70%
51%
43%
33%
Supplier
Source: Roland Berger & Edscha
2000 2005 2010 2000 2005 2010
© Dana 2013
Figure 2: Changes in the Automotive Supplier Market (Source: Roland Berger and Edscha)
Current Market
The competitiveness of the market, the push by the developed countries in particular into greener
product technologies, and the end-consumer requirements with ever more sharpened focus on
comfort, created the need to introduce electronics on a broad scale. (Figure 3, [2])

Automotive Landscape
> 70% of new

features are
Software enabled
EE systems.
> 60% of new
systems are
distributed in nature
Sample Vehicle has approx.

 40-60 Computers on board
Space Shuttle  > 8 MByte of Software
 5 Computers on board  >10 Million LOC/Instructions
 700 kByte Software  > 5,000 Software Parameters
 500.000 LOC/Instructions  > 30,000 functional requirements
 > 600,000 pages of E/E specifications
 > 10,000 buildable Vehicle-Series-
Variants (based on ECU component
permutations)
© Dana 2014
Figure 3: Automotive Landscape [2]
The starting point where we had to apply principles of system engineering (SE) was developed in
the aerospace industry and automotive industries. Simple application of existing SE as we knew it
was not sufficient. A more modern approach was needed, which went beyond just the application
of the SE principles. The automotive industry is a much faster evolving industry where new
models and technologies need to be introduced into the market every two to three years in order to
stay competitive. The need for agility to adapt to a rapidly changing market was the principle
driver for the implementation of CAE – simulation, or even simpler – virtual engineering. The
international council for system engineering, INCOSE, formulated the Model-based System
Engineering (MBSE) initiative to address this and it became a major pillar within the INCOSE
vision 2025 ([3]).
Earlier it was mentioned that CAE was applied by leading companies as a systems approach, but
the traditional CAE is focused on detailed physical models. The market demand for a two to three
year model cycle provides for just enough time to optimize systems with all their variations based
on highly complicated models, though. NAFEMS, the international association for engineering
modeling and simulation understood this new requirement early on and started to reach into new
upfront simulation techniques. The logical result was that NAFEMS and INCOSE got together
formulating a new strategy by combining forces in creating the joint System Modeling and
Simulation Working Group (SMSWG). (Figure 4)

To develop a vendor-neutral, end-user driven consortium that
not only promotes the advancement of the technology and
practices associated with integration of engineering analysis and
systems engineering, but also acts as the advisory body to drive
strategic direction for technology development and
standards in the space of complex engineering.
This group will support activities that bridge engineering
analysis and systems engineering to provide digital solutions to
represent real life experiences; and optimize the integration of
systems engineering and simulation solutions for both OEM and
supplier.
This includes education, communication, promotion of standards,
and development of requirements that will have general benefits
to the simulation and analysis community with the identification of
benchmarks and major strategic issues (grand challenges).
Figure 4: Mission Statement of the joint NAFEMS / INCOSE System Modeling and Simulation Working
Group
Market drivers capture this very clearly [4]:

• “First time right”
• Reducing time-to-market
• Front-end optimization
• Reduced development time – reduced development costs
• Improved product robustness
• Quality guarantee
• Flexibility
• Globally one voice to customer
• Global regional availability and support
• Partnership across companies
This results in a change the way we approach engineering development. The previous upfront
spending needed to be turned around and, at the same time total spending needed to be reduced
(Figure 5; [5]).

Front-end spending Traditional
Development costs
Early design commitment
Development cycle
Reversed spending Modern

Development costs
Virtual front-end optimization

Physical testing only shifted to back end
Development cycle
Future State
Development costs
Virtual front-end optimization

Reduced costs due to reduced physical testing
for final verification
Development cycle
Figure 5: Development Costs over the Develoment Cycle
System Modeling and Simulation at Dana
It became clear that a completely new thinking was needed. In Dana we started to call this iCAE™
([4]). This idea was further developed within Dana over the last few years and resulted from an
organization perspective in merging SE and CAE. Coordinating isolated activities wasn't sufficient
anymore. This made Dana a very active contributor to the SMSWG activities, as well as NAFEMS
and INCOSE.
At the same time, we looked beyond the initial understanding of SMS focusing on the product
related system. For Dana we define it the following way [5]:
► System Modeling: represent / link all engineering domains as a system (might it be from
a product or ecosystem perspective) within Dana and across companies (customer and
suppliers)

► Simulation: being able to predict behavior, efficiency and performance upfront and in
real time using only virtual methods
This provides an immediate and different dimension to the term “system”. It isn’t just simply one
definition. System has a different meaning for each of us based on the environment, industry and
product line. In order to bring this together in a relatively simple way we identified the main areas
and disciplines throughout the engineering life cycle (Figure 6; [5]).
Engineering Areas / Disciplines throughout the Life Cycle
Maturity
Assessments
► R&D
► Market
► Design
► Requirements
► Release
► Quality
► Production
► Performance
► Retire Enabling foundation
► Organization
► Culture
► Collaboration
► Communication
► Process ► Technology
► Integration ► Infrastructure
► Traceability ► Tools
► Impact Analysis ► Common
► Validation ► Flexible Simulations / Analysis
► Verification ► Advanced ► Virtual design
► Unification
In-the-Loop Management ► Virtual Testing
► Requirements ► Virtual manufacturing
► Quality ► Risk assessment /
► Change management certification of
correctness
► Supply chain
► End-of-life
► Program management
► Costing
► Trade-off studies
© Dana 2014
Figure 6: Engineering Areas / Disciplines throughout the Life Cycle
A few things need to be highlighted here:

• The life cycle spans over the different maturity stages of the product. Those can be easily
captured with stage gate processes. The way we approach this means that we assigned
over those different levels of maturity a general engineering approach, which can be
captured by an overall engineering Vee. At the same time, within each of the stages there
exist separate Vee’s (approaches) which evolve from a flexible to a more stringent
environment (or from an ad-hoc to a standard (best practice) -driven flow. This process is
an iterative one. (Figure 7)
• The way one goes through the process was traditionally a sequential one. But, in order to
make a truly front-end optimization driven development process it needs to be able to
loop back at any time during the process for any of the engineering and supporting
functions, i.e. in-the-loop processes.
• A very important feature becomes optimization with its statistical approach. The nature
always comes with a certain uncertainty. The understanding of the risk associated with
each level of uncertainty is a must in this new approach.

• The new principle of looking at the system from an ecosystem perspective (and not just
linked to the product) means we need to incorporate manufacturing, costing, and other
enabling and supporting functions.
• The middle of the circle represents the foundation of this approach. Way too often, the
traditional focus was just on what is around it. But, without the enabling foundation this
approach will fail. Each organization needs to identify for itself what this means for them.
The system thinking spans across the entire life cycle, ……
Figure 7: System thinking across the entire Life Cycle
This complexity as mentioned here on a high level clearly shows that business-as-usual doesn’t
work anymore. A step change is needed in order to achieve this. In order to do this properly Dana
decided to assess first where we stand today and, where we need to be a few years from now in
order to be competitive. The business value assessment performed covered all diversity within the
company from a cultural, regional and product perspective. This was the basis for our System
Modeling and Simulation roadmap, and helped to define the areas we need to focus on.
This requires a very flexible environment consisting of a wide basis of vendors and different
linked technologies. It can be characterized in a simple way the following:
• The overall umbrella is intuitive, model-based with, where possible 3D to enable a
visually intuitive system, and an open environment in order to be able to have a vendor-
independent system. This is the foundation of a model-based enterprise;
• Product Lifecycle Management (PLM) and Simulation Lifecycle Management (SLM)
systems complement each other and link to all other engineering systems seamlessly in
order to achieve a “single source of truth” environment;
• An easy linking between the different tools is not just done by using SLM but also
through the utilization of the Functional Mockup Interface (FMI) standard. This allows
simultaneous time and cost-optimized deployment of the various tools, and very
importantly, democratization of virtual tools within the entire engineering organization
(Figure 8, [6]).

Figure 8: Collaboration Model [6]
Furthermore the following areas play a major role:

• Organization with its focus on deploying and teaching the idea of SMS and SE. This is a
process which takes time and commitment on several levels of the organization. If
organizational support is missing any attempt of deploying new tools and approaches will
fail.
• Requirements need to be managed properly and interactively. We can never assume that:
o The received or defined set of requirements is accurate or complete. A quality
check is required.
o Requirements are changing and accumulating constantly. The established
system needs to be flexible and agile enough to adjust to this on-the-fly.
• The supporting infrastructure must be able to provide fast access to a large amount of
data on a global scale. This means technologies, like:
o Compute cloud
o Virtual desktop infrastructure.
By discussing the above mentioned focus areas one can immediately see the link to two of the
modern discussed topics:
• Big Data, and
• Internet-of-Things.
The major challenge from here on is not necessarily resources and funding, but speed of execution
(deployment). In order to achieve here the optimum performance, a modern engagement approach
is needed, which, at the same time requires a modern communication and collaboration
infrastructure. Of-course, coordinating this is challenging due to the number of activities that must

occur simultaneously. The right roadmap is achievable through a proper engagement and
awareness so people start to “live” it.
Summary
System Modeling and Simulation is the next logical step for engineering-driven companies in
competitive markets. It involves many modern technologies such as big-data, internet-of-things,
model-based-enterprise, cloud computing, and much more. But most of all, in order to achieve
this, the whole organization needs to be in agreement and alignment. This can only be achieved
through engagement of the people. Every single person needs to be committed in order to be able
to execute this. Dana has reached the momentum where the whole organization is becoming part
of this and makes this a “living” environment.
References
[1] Frank Popielas: “Enabling Simulation as an Enterprise Engineering Practice”; Dassault

3DEXPERIENCE Conference 2013, Las Vegas, NV, USA
[2] Richard Riff, Technical Fellow; Ford Motor Company; September 2010
[3] INCOSE: System Engineering Vision 2025 - SE2025
[4] Frank Popielas: “Simulation and its new Role as the Development Engine in the virtual
Engineering Environment”; NAFEMS World Conference 2013, Salzburg, Austria
[5] Frank Popielas: “System Modeling and Simulation at Dana”; Dassault 3DEXPERIENCE
Conference 2014, Las Vegas, NV, USA
[6] Hubertus Tummescheit / John Batteh, Modelon Inc. & INCOSE/NAFEMS SMSWG, INCOSE
IW, January 25th, 2015, Torrance / CA

New Constitutive Model for Woven Thermoplastic
Composite Materials.
Authors: Pieter Volgers1, H. Kuhlmann2, Z. Zhang2
1
DuPont Performance Polymers, European Technical Center, Geneva, Switzerland
2
DuPont Engineering Technologies, Chestnut Run Plaza, Wilmington, Delaware, U.S.A
Abstract: This paper describes the behavior of the DuPont thermoplastic composite sheet offering,
under the name DuPont™ Vizilon™ and matches this to composite models as implemented in
Abaqus. Subsequently, to obtain a more accurate description of the actual behavior of these
materials, DuPont has developed a new material model by combining both plastic deformation of
the thermoplastic matrix material with a more classical composite damage model for the
description of the non-linear permanent deformation of the woven continuous fiber reinforcement
and subsequent failure. The (current) results obtained with this model, implemented in Abaqus,
are presented, demonstrating the potential of this model for thermoplastic composites.
Keywords: Composites, Constitutive Model, Crashworthiness, Damage, Fabrics, Failure,

Plasticity, Polymer.
1. Introduction
In order to meet the ever more stringent requirements on fuel consumption and CO2 emissions,
the automotive industry pursues various solutions, of which the reduction of mass is an important
aspect, which creates a snow-ball effect to emission reduction, such as further engine downsizing.
Thermoplastic composites (TPCs) are being developed to compete with metals for the use in
structural design and light-weighting. To that goal, the end-customer needs to be able to accurately
predict its performance to a level comparable to that of the non-in-kind competition. As part of its
offering, DuPont has been improving the state-of-the-art of thermoplastic composite analysis for
its Vizilon™ thermoplastic composite materials with the models available in various software
packages.
Figure 1: 'Classical' UD composite vs Vizilon™ TPC

Vizilon™ thermoplastic composite differentiates itself from a ‘classical’ fiber-reinforced
composite material, such as a carbon-fiber epoxy as found in many applications in a few distinct
areas, as shown in Figure 1. Whereas a ‘classical’ composite uses uni-directional fibers set in a
thermoset matrix, such as epoxy, on a layer-by-layer basis, Vizilon™ thermoplastic composite is
based on a woven fabric fully impregnated and consolidated in a thermoplastic polymer (such as a
polyamide) as a sheet of multiple layers, which each layer placed in the same direction. This is the
result of the continuous manufacturing process to produce the sheet, non unlike the rolling of sheet
metal plates. As a result, where a ‘classical’ UD composite layer has a main direction with
significantly more dominant properties in the fiber direction than the properties perpendicular to
the fibers (denoted as the 1- and 2-direction, respectively), the woven sheet has two main
directions with identical or similar properties and behavior. In addition to that, the thermoplastic
polymer matrix made of Vizilon™ product shows a much more ductile behavior than a much
stiffer and more brittle thermoset matrix. As will be shown in this paper, these differences are
fundamental when trying to provide a mathematical description of the material behavior.
An extensive material testing program allowed the full characterization of the material behavior,
the selection of suitable material models and their corresponding material model parameters. This
paper describes the material testing program and the evaluation of different material models
available in Abaqus. Based on a study of the theoretical models in Abaqus and their corresponding
performance in the correlation to the actual material behavior, DuPont implemented an updated
material model to more accurately describe the actual observed material behavior.
2. Material testing program

The first step for the modeling of any material is the generation of material property data through
standard material tests. For a continuous fiber reinforced composite, it is natural to assume that the
material behavior is anisotropic. As the thermoplastic composite sheet is made from woven fabric,
whereby all the layers are oriented in the same direction, as a first approximation we assume the
material behavior to be globally orthotropic, despite the multiple layers, with the warp and weft
directions corresponding to the principal orthotropic axes. The assumption of material orthotropy
reduces the linear elastic in-plane material parameters to 2 elastic moduli, one shear modulus and
one Poisson ratio. For reasons of simplicity, the out-of-plane properties will not be considered in
this paper. Basic material testing then consists of measuring the behavior of the tensile and
compressive behavior in the main orthotropic (fiber) directions and the shear. The off-axis tensile
properties should then follow automatically and can be used to verify the material model.
Following initial material characterization and evaluation, a second round of material testing was
added to further understand the material behavior with the objective of more accurately predicting
the performance of structural parts, particularly under complex and/or crash loading.
DuPont’s Vizilon™ offering consists of different types of materials. In this paper we will
concentrate on one single grade, Vizilon™ SB63G1, a balanced, 2/2 twill glass fabric with 63%
(mass) fiber content, with a PA66-based matrix resin, which corresponds to a 44% fiber volume
fraction (to remain consistent with traditional short fiber polymer nomenclature DuPont refers to
the fiber content in terms of mass). The balanced weave allows us to assume that the properties in
the warp and weft direction are (in theory) identical. Mechanical testing has shown a slight

difference in performance between the two directions, but this is sufficiently small to be neglected
for the purpose of this paper when determining the corresponding material model parameters.
For legal and confidentiality reasons, the stresses are presented on a relative scale, whereby the
failure stress in the main (warp) direction at room temperature is taken as 100%.
2.1 Testing conditions
Standard material testing of tension and compression in the main fiber direction has been
performed following ISO standards for composite testing (ISO 527-4, 199; ASTM D7078, 2005).
With the matrix material being a thermoplastic polymer, a certain dependency on the temperature
is to be expected. Hence, material tests have been performed at a range of temperatures from -40
°C to +120 °C. This allows us to get a good indication of the relative influence of the matrix
material, the mechanical properties of the glass fibers not being subject to change in this
temperature range. As the polymer is PA66-based, the behavior is also dependent on the moisture
content. Following traditional practice of the polymer resin business, the material tested has been
conditioned to moisture content corresponding to the equilibrium found at 50% relative humidity
(RH) at 23°C. The specimen testing sample schematic is shown in Figure 2.
Figure 2: Schematic of tensile testing in the warp/weft direction, shear and 45-degree biased
tensile

2.2 Tensile and compression in main fiber direction
Figure 3: Tensile and compressive stress-strain curves in warp direction

Figure 3 shows the tensile and compressive stress-strain curves for the composite material in the
warp (main fiber) direction. As the material considered uses a symmetric weave for the continuous
fiber reinforcement, the properties in the waft direction are very similar. We notice an almost
identical modulus in both tension and compression in the temperature range from 23°C to 120°C,
but with a reduced failure stress with increasing temperature. The property change at -40°C is
more pronounced, as this takes the polymer matrix below the glass-transition temperature (Tg).
Furthermore, the compressive failure strength is well below the tensile strength. A closer study of
the test specimen shows that, while failure in tension is abrupt, in compression the sample
maintains a limited cohesion, allowing a continued (but strongly reduced) structural resistance
which allows for the continued absorption of energy. This behavior is well known for composites,
but is much more pronounced when using a woven fabric and a thermoplastic polymer matrix.
However, this behavior is hard to measure using traditional equipment, as the standard
programming of a MTS machine stops the equipment and the recording when a strong abrupt
reduction of resistance is reached.
2.3 Shear testing
The shear testing uses a V-notched Iosipescu sample and set-up (ASTM D3410, 2008) which
provides a direct measurement of the shear properties. Aramis digital image measurements are
used instead of strain gauges as described in the norms to better interpret the test data and avoid
the averaging of the strain in the zone of the gauge. This is all the more important given the large
strain gradient observed in the center of the sample where the shear deformation is measured.
Contrary to tensile testing in the main fiber direction, a small linear elastic region is soon
transformed into a large non-linear plastic zone with failure strains up to 50%. At this point the
woven fabric is severely deformed and fibers oriented towards the direction of the loading. Also, a
strong dependency on the temperature is observed. This shows that the shear properties are heavily

dependent on the performance of the matrix material. Tests on sheets with higher glass content
show a lower shear performance, demonstrating that the shear behavior comes almost completely
from the matrix material.
2.4 45-degree off-axis tensile loading
To verify the orthotropic behavior and the shear properties, particularly the shear modulus, tensile
tests at specimens cut at 45 degree angle have been performed. Results are shown in Figure 4. One
immediately notices that the stress-strain curves resemble much closer the shear data than the
tensile testing in fiber direction, which a significant plastic region. While the failure stress is
significantly lower when tested in the 45-degree direction compared to the main fiber directions,
the failure strain is much larger.
Figure 4: Shear and 45-degree off-axis tensile testing
3. Evaluation of existing material models

The ultimate objective for the material modeling by DuPont is to provide customers with the
necessary know-how and capabilities to analyze and design structural components using the
Vizilon™ thermoplastic composite sheet offering. It was therefore decided to concentrate the
development effort towards existing material models. Abaqus is the global standard in DuPont for
structural analysis and has excellent material model capabilities and the possibility to change from
static implicit to dynamic explicit analysis within the same environment using the same material
model. A study in the various orthotropic material models available in Abaqus was done to
evaluate their performance.
To evaluate the behavior and performance of the various material models considered, the tensile
test in warp direction, the V-notched shear test and the tensile test at 45-degree bias were
simulated and the resulting nominal stress-strain curves compared.
3.1 Linear orthotropic model with classical failure criteria
As a first step, a linear orthotropic material model was used excluding any permanent deformation
due to plasticity or damage. Providing the failure stress and strain in the different orthotropic
directions allows the post-processor to evaluate the different classical failure criteria (Abaqus,

Section 22.2.3), such as maximum stress, maximum strain, Tsai-Hill or Tsai-Wu. The material is
assumed broken when the failure criterion is larger than 1.
The maximum stress or –strain criterion simply considers the stress or strain in the main
orthotropic directions and shear, with possible different values for tension and compression.
𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙
𝑋 = 𝜎𝑡𝑒𝑛𝑠𝑖𝑜𝑛 𝑖𝑓 𝜎 > 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 𝑋 = 𝜎𝑐𝑜𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛
𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙 𝑓𝑎𝑖𝑙
𝑋 = 𝜎11 ; 𝑌 = 𝜎22 ; 𝑆 = 𝜎12
𝜎11 𝜎_22 𝜎12
FCmax = 𝑚𝑎𝑥 � ,
𝑋 𝑌
,�
𝑆
��
For the maximum strain criterion replace the stresses by the strains.
The Tsai-Hill criterion provides an orthotropic failure measure similar to Hill’s orthotropic yield
criterion for plasticity and is given by:
2 𝜎 𝜎
𝜎11 2 2
11 22 𝜎22 𝜎12
𝐹𝐶𝑇𝑠𝐻 = 𝑋2
−
𝑋2
+ 2+ 2
𝑌 𝑆
The material specimen test set-ups were modelled and simulated. Obtained reaction forces and
displacements between the extensometer grips are converted into nominal stress and strain and
compared to the average stress-strain curves obtained from testing. (We will use the stress as a
comparative measure and not the reaction force to eliminate the measured differences in
thickness.) Figure 5 shows the resulting curves from the test data (continuous lines) and the
simulated test (dashed lines) for the 0- and 45-degree tensile test as well as the shear test.
Figure 5: Predicted failure using different criteria for 3 test conditions

Using the different failure criteria available, we can determine their predicted failure point for the
different test conditions. For the material data available, the Tsai-Hill and Tsai-Wu gave very

similar results, and hence only the former will be shown here. As expected, the failure of the
material in the main fiber direction is predicted properly using all conditions. The problem occurs
for the off-axis and shear tests, where the orthotropic material model predicts a linear response
while the test data shows a large non-linear region. For the calculated stress-strain response not to
deviate too much from the measured data beyond the elastic region, the initial shear modulus was
reduced by 50%. The maximum strain criterion is the best to predict the maximum allowable
deformation, but even this criterion strongly under-predicts the deformation capabilities of the
material in off-axis testing. The Tsai-Hill criterion predicts the maximum stresses in both off-axis
and shear remarkably well, but is also hampered by the linear response of the orthotropic elastic
model.
3.2 Orthotropic elastic model with Hashin-type damage

In order to account for the non-linear response of a continuous fiber composite material, a Hashin-
type damage model is available in Abaqus (Abaqus, Section 24.3). The theory developed by
Hashin (Hashin, 1980, Hashin and Rotem, 1973) forms the basis for most composite damage
models found in the various commercial finite element codes.
The damage model consists of two components: the damage initiation criteria to determine the
start of the damage, and the damage evolution equations. The model is clearly developed for uni-
directional fiber reinforced layers, as described in the beginning of this paper. Damage initiation is
based on 4 criteria:
�
𝜎 2 𝜏� 2
• Fiber tension failure 𝐹𝑓𝑡 = � 𝑋11 � + 𝛼 � 𝑆12 �
𝑡 𝑙
�
𝜎 2
• Fiber compression failure 𝐹𝑓𝑐 = � 𝑋11 �
𝑐
𝜎� 2 𝜏� 2
• Matrix tension 𝐹𝑚𝑡 = � 𝑌22 � + � 𝑆12 �
𝑡 𝑙
�
𝜎 2 𝑌 2 �22
𝜎 𝜏� 2
• Matrix compression 𝐹𝑚𝑐 = �2𝑆22 � + ��2𝑆𝑐 � − 1� ⋅ 𝑌𝑐
+ � 𝑆12 �
𝑡 𝑡 𝑙
Where Xt, Xc, Yt, Yc tensile and compressive strength in longitudinal and transverse direction,
respectively, as before, and Sl and St the longitudinal and transverse shear stress, and ^ indicates
equivalent stress.
One immediately notices the very different expressions for the compression failure in the fiber and
matrix direction. While for a ‘classical’ UD composite layer the failure in the main fiber direction
is determined by the performance of the fiber, and the failure perpendicular to this is determined
by the properties of the matrix material, this assumption is not valid for a woven thermoplastic
composite, and determining material parameters which fit the observed behavior is therefore
complicated.
The damage evolution has been implemented by reducing the stress according to:

1
⎡1−𝑑𝑓 0 0 ⎤
⎢ 1 ⎥
𝜎� = ⎢ 0 1−𝑑𝑚
0 ⎥⋅𝜎
⎢ 1 ⎥
⎣ 0 0 1−𝑑𝑠 ⎦
With the following evolution relations:

𝑓𝑡 2
• Fiber tension (𝜎�11 ≥ 0) 𝛿𝑒𝑞 = 𝐶�〈𝜖11 〉2 + 𝛼𝜖12
𝑓𝑐
• Fiber compression (𝜎�11 < 0) 𝛿𝑒𝑞 = 𝐶〈𝜖11 〉
𝑚𝑡 2
• Matrix tension (𝜎�22 ≥ 0) 𝛿𝑒𝑞 = 𝐶�〈𝜖22 〉2 + 𝜖12
𝑚𝑐 2
• Matrix compression (𝜎�22 < 0) 𝛿𝑒𝑞 = 𝐶�〈−𝜖22 〉2 + 𝜖12
The damage evolution for each mode is then given by:

𝑓𝑎𝑖𝑙
𝛿𝑒𝑞 (𝛿𝑒𝑞 − 𝛿 0 )
𝑑= ; 𝑑 = 1 − (1 − 𝑑𝑡𝑡 )(1 − 𝑑𝑓𝑐 )(1 − 𝑑𝑚
𝑡 𝑐
)(1 − 𝑑𝑚 )
𝛿𝑒𝑞 (𝛿 𝑓𝑎𝑖𝑙 − 𝛿 0 ) 𝑠
Notice that the basic Hashin damage does not include an independent evolution of the shear
damage.
Figure 6: Predicted failure behavior using Hashin-type damage

The numerical results after longtime optimizing the material damage parameters are shown in
Figure 6. As one would expect, the tensile behavior in the main fiber direction is reproduced
properly. Having adapted the material model parameters to the best of our capabilities, the shear

behavior could be matched to a satisfactory level, but the behavior in the weft (perpendicular to
the main fiber, or warp, direction – not shown) and the off-axis direction is far from satisfactory.
The initial response of the model of the 45-degree off-axis tension test is too stiff, in order to
approach the curve until about 4% strain, after which the stiffness softens and generates a
downward curve not representative of reality.
The conclusion has to be that the material model does not describe the actual physical behavior.
However, for components which are principally loaded in the main fiber direction and with
possibly a significant shear contribution (such as a general beam under 3-point loading) the model
can reasonably predict the structural performance. But for structures loaded off-axis to the warp
direction, results do not correspond to observed physical behavior.
3.3 Orthotropic elastic-plastic model with Hill’s yield criterion
Given the large plastic deformation in the off-axis tensile and shear test, the use of an elastic-
plastic model should be considered. Abaqus allows this to be combined with an orthotropic elastic
model and an anisotropic yield criterion for the plastic deformation according to Hill (Hill, 1979):
2 2 2
𝑓(𝜎) = �𝐹(𝜎22 − 𝜎33 )2 + 𝐺(𝜎33 − 𝜎11 )2 + 𝐻(𝜎11 − 𝜎22 )2 + 2𝐿𝜎23 + 2𝑀𝜎31 + 2𝑁𝜎12
Where F, G, H, L, M and N are constants (Abaqus, Section 23.2.6) related to the yield stress ratios
Rij.
The plastic response is provided in terms of a true-stress vs plastic strain curve based on the shear
test data and the yield stress ratios are calculated accordingly. We assume an isotropic plastic flow
function.
Figure 7: Predicted behavior (without failure) using elastic-plastic model

Simulating the three standard coupon tests as before, we obtain the response curves as shown in
Figure 7. A perfect match is found for the 0-degree tensile test as well as the shear test, although
the material will continue to deform due to the absence of any failure mechanism in the model.
But also the 45-degree bias tensile test provides an accurate result, with the measured response
being slightly stiffer after about 6% strain. This effect is due to the fact that the woven fabric
fibers start to shear whereby the fibers are being oriented in the direction of the load, increasing
the effective stiffness. The material model itself maintains its assumption of orthotropy of the
woven fabric.
4. Development of new material model

Looking at the simulation results obtained with the different material models as described above,
we can conclude that classical anisotropic Hill plasticity allows for an accurate description of the
shear deformation, including off-axis tension loading, while the Hashin damage modelling can
predict the loss of stiffness and failure in the fiber direction.
An improved material model which accurately describes the material behavior as observed in the
material tests above should therefore consist of:
• A general plasticity model whereby the main plastic deformation occurs through shearing
of the matrix material;
• A (Hashin-type) damage model superimposed on top of this plasticity model to account
for loss of stiffness and failure;
• A damage model with identical damage mechanism (but not parameter values) for both
orthotropic fiber directions (warp and weft);
• An independent damage factor and failure mechanism for shear.

Figure 8: Proposed superposition of plastic and damage initiation
A schematic of the mapping of plasticity and damage is shown in Figure 8. The red/dash-dot line
indicates the initial plastic yield surface similar to the yield function used in this paper. The
yellow/shaded region shows the potential damage initiation surface, which partially overlaps. For
loading in the main fiber directions the stress-state will reach the damage initiation surface first,
causing the material to rupture or fail before the potential plastic yield surface has been reached.
For stronger off-axis loading, the plastic yield will be reached first. Pure elastic-plastic
deformation will occur, expanding the yield surface, until the damage initiation surface is reached.
At that point, both permanent deformations will occur simultaneously until total failure has been
reached.
This principal should be relatively easy to implement in Abaqus starting from the existing elastic-
plastic material model. This differs from the Johnson-Pickett material model for woven fabrics
(Johnson et.al., 2001) as implemented in PAM-Crash, which only considers plasticity for the shear
deformation in the constitutive equations. The resulting model creates not only complex equations,
but also requires non-physical material model parameters for calibration.
4.1 Theory and implementation
A macro mechanical approach was used to predict damage progression and failure. This approach
represents the damage onset and damage progression whose properties are degraded prior to
ultimate failure. A continuum damage mechanics formulation is used following the work by Pinho
(Pinho et.al., 2006) in which material degradation parameters are internal state variables which are
governed by damage evolution equations. It is computationally more efficient, especially for
simulating damage in a large scale structure.

4.1.1 Damage initiation
Damage initiation refers to the onset of degradation at a material point. Once damage initiation
occurs, the material starts to be degraded. Damage initiation is defined as follows (t indicating
tension, c indicating compression):
2
𝜎�11
𝑂𝐷1 𝑡,𝑐 = √� �
𝑋𝑡,𝑐
2
𝜎�22
𝑂𝐷2 𝑡,𝑐 = √� �
𝑌𝑡,𝑐
Onset of fiber damage takes place when 𝑂𝐷𝑖 𝑡,𝑐 = 1 ; 𝑖 = 1,2. Onset of shear damage takes place
when 𝑂𝐷12 > 1 and is defined as:
2 2 2 2
𝑂𝐷12 = � 𝑚𝑎𝑥{�� 𝜎�11 � + �𝜎�12 � , �� 𝜎�22 � + �𝜎�12 � }
𝐴𝑆1 𝑆12 𝐴𝑆2 𝑆12
Where
𝑖𝑓 𝜎11 > 0 𝐴𝑆1 = 𝑋𝑡
𝑖𝑓 𝜎11 < 0 𝐴𝑆1 = 𝑋𝑐
𝑖𝑓 𝜎22 > 0 𝐴𝑆2 = 𝑌𝑡
𝑖𝑓 𝜎22 < 0 𝐴𝑆2 = 𝑌𝑐
4.1.2 Damage Evolution
The damage evolution is implemented by reducing the stress as described in (Abaqus, Section
24.3). The compliance matrix has three scalar damage parameters df, dm, ds, which have values 0
≤ di ≤1 and represent modulus reductions under different loading conditions due to damage in the
material. For laminate plies, df and dm are associated with damage and failure in the principal
fiber directions, and ds with in-plane shear failure.
The damage evolution is based on an equivalent displacement and stress for each of the damage
modes that are defined as follows:
• Tension in principal direction 1 when 𝑂𝐷1 𝑡 = 1 𝑑𝑓𝑡𝑒𝑞1 = 𝐿𝑐 ∗ 𝑒𝑓𝑡
𝑐
• Compression in principal direction 1 when 𝑂𝐷1 = 1 𝑑𝑓𝑐𝑒𝑞1 = 𝐿𝑐 ∗ 𝑒𝑓𝑐
• Tension in principal direction 2 when 𝑂𝐷2 𝑡 = 1 𝑑𝑓𝑡𝑒𝑞2 = 𝐿𝑐 ∗ 𝑒𝑓𝑡
𝑐
• Compression in principal direction 2 when 𝑂𝐷2 = 1 𝑑𝑓𝑐𝑒𝑞2 = 𝐿𝑐 ∗ 𝑒𝑓𝑐
• Tension in principal directions and shear when 𝑂𝐷12 = 1
𝑑𝑓𝑡𝑒𝑞12 = 𝐿𝑐 �𝑒𝑓𝑡 2 + 𝑒𝑓𝑠 2

• Compression in principal directions and shear when 𝑂𝐷12 = 1 𝑑𝑓𝑐𝑒𝑞12 =
𝐿𝑐 �𝑒𝑓𝑐 2 + 𝑒𝑓𝑠 2
The damage evolution is based on an equivalent displacement and stress for each of the damage
modes that are defined in a similar manner as the Hashin-type damage as described in the previous
section, but with an independent damage variable for shear:
𝑑𝑠 = 1 − (1 − 𝑑𝑠1 )(1 − 𝑑𝑠2 )
Where 𝑑𝑠1 , 𝑑𝑠2 are defined as above, and:
𝑂𝐷
(𝛿𝑒𝑞 − 𝛿𝑒𝑞 )
𝑑= 𝑓 𝑂𝐷 )
(𝛿𝑒𝑞 − 𝛿𝑒𝑞
𝑂𝐷 𝑓
Where 𝛿𝑒𝑞 is the equivalent displacement at the onset of damage, 𝛿𝑒𝑞 is the equivalent
𝑓
displacement at which the material is completely damage and 𝛿𝑒𝑞 , which is always 𝛿𝑒𝑞 ≥ 𝛿𝑒𝑞 ≥
𝑂𝐷
𝛿𝑒𝑞 , is the equivalent displacement at time i. The characteristic length 𝐿𝑐 is based on the element
geometry and formulation. For a shell it is a characteristic length in the reference surface,
computed as the square root of the area.
4.1.3 Implementation in Abaqus
The damage and failure modelling has been implemented in Abaqus through the use of user-
defined fields (UFIELD) and user dependent state variables (DEPVAR). This method was
preferred over the writing of a separate independent user-defined material model to allow the use
of the existing material model in Abaqus/Explicit and only modify the behavior in the damage and
failure domain.
The 4 state variables are related to the damage variables and allow Abaqus/Explicit to determine
the effective stiffness of the material. These state variables are calculated in a user subroutine
VUSDFLD implemented in Abaqus/Explicit. This routine takes the 8 user defined fields as input,
which determine the failure stress in warp direction in tension and compression, the failure stress
in weft direction in tension and compression, failure strain in tension and compression, the
damage initiation stress in shear and a controlling parameter, respectively. Current implementation
does not include strain-rate dependency for the failure stress, but different values can be specified
depending on the actual strain-rate in the model to obtain a more accurate prediction for dynamic
loading.
The nonlinear material behavior of a composite laminate can be specified as a function of
solution-dependent variables. The user subroutine VUSDFLD in Abaqus/Explicit was used to
modify the behavior of the nonlinear material and to include the effects of damage. The material
model presented in this paper, includes damage, and various modes of failure. After onset of
damage, the stress-carrying capacity degrades gradually with increasing strain. The values of the
failure indices are not assigned directly to the field variables: instead, they are stored as solution-
dependent state variables.

The user defined variables are defined for the input of the elastic engineering constants. In
addition to this, the plasticity definition as used in the elastic-plastic model described above is
added. This method allows for the inclusion of strain-rate dependency for the plastic behavior.
4.2 Evaluation of initial performance
Evaluation of the performance and potential of the material model is done through the specimen
tests as before and results are shown in Figure 9. This shows the capabilities of the newly
implemented material model.
Figure 9: Predicted behavior with new material model

The damage parameters were systematically obtained by correlating simulations with standard
quasi-static and high-rate tensile coupon, compression and shear test. The parameters are adjusted
over several simulation runs until satisfactory correlation is obtained. The linear behavior of the
tensile test in main fiber direction is preserved, but now allows failure to occur. The 45-degree
bias tensile response follows both the elastic and the non-linear part of the curve until failure
occurs, after which the material is damaged and fails.
The material model was then further validated and refined with simulations on dynamic impact
plaque tests and three-point bending tests, which induce more complex loading conditions.
5. Summary and Conclusions

This paper describes the theoretical aspects of material models available in Abaqus applied to the
Vizilon™ offering, a composite sheet based on woven continuous fiber reinforcements and a
thermoplastic polymer matrix. The material is characterized for these models and their
performance evaluated by simulating the standard material tests of traction in the main fiber

direction, V-notched shear and traction at 45-degrees angle to the main fiber direction, of which
the results are summarized in Figure 10. A symmetric (balanced) woven fabric was used to
eliminate the additional effort of different material parameters in warp and weft direction.
It has been shown that advanced material modeling for composite materials is currently aimed
principally at high-performance use of these materials, which use a uni-directional (UD) fiber
placement per layer with a thermoset matrix. Simulation results with the aim to reproduce the
behavior of a woven thermoplastic composite sheet demonstrate that these models do not address
the specific characteristics of such a material, principally:
• Similar damage and failure mechanism in both axes of orthotropy (fiber warp and weft
direction);
• Large plastic deformation in shear and in off-axis loading.
Further literature study revealed that very little work has been done on the modelling of woven
thermoplastic composite materials. This motivated DuPont to develop and implement a new
material model in Abaqus by superimposing a damage and failure model on top of an elastic-
plastic model using Hill’s anisotropic yield criterion. This approach should (theoretically) have the
advantage of a methodology approaching micro-mechanical modelling and allow for a simple
implementation and determination of the corresponding material parameters.
Figure 10: Summary numerical analysis results for 45-degree off-axis tensile test
The material model implemented by DuPont in Abaqus has been described and initial results
presented. This paper shows the promising performance of this material model for describing more
accurately the behavior of a thermoplastic composite woven sheet.

6. References
1. Abaqus Analysis User’s Guide, Version 6.14, Dassault Systèmes Simulia Corp., Providence,
RI.
2. ASTM D7078/7078M-05, ASTM International, West Conhohocken, PA, USA, 2005
3. ASTM D3410/3410M-03, ASTM International, West Conhohocken, PA, USA, 2008
4. Hashin, Z., “Failure Criteria for Unidirectional Fiber Composites,” Journal of Applied
Mechanics, vol. 47, pp. 329–334, 1980.
5. Hashin, Z., and A. Rotem, “A Fatigue Criterion for Fiber-Reinforced Materials,” Journal of
Composite Materials, vol. 7, pp. 448–464, 1973.
6. Hill, R., “Theoretical plasticity of textured aggregates,” Mathematical Proceedings of the
Cambridge Philosophical Society, 85(1), pp 179–191, 1979
7. ISO 527-4, International Organization for Standardization, Geneva, Switzerland, 1997
8. Johnson, A.K., Pickett, A.K., and Rozycki, P., “Computational methods for predicting impact
damage in composite structures,” Composite Science and Technology, Vol 61 (15), pp 2183-
2192, 2001.
9. Pinho, S.T., Iannucci, L., and Robinson, P., “Physically-based failure models and criteria for
laminated fibre-reinforced composites with emphasis on fibre kinking: Part I: Development,”
Composites Part A: Applied Science and Manufacturing, Vol. 37 (1), pp 63-73, 2006.
10. Pinho, S.T., Iannucci, L., and Robinson, P., “Physically-based failure models and criteria for
laminated fibre-reinforced composites with emphasis on fibre kinking: Part I: FE
implementation,” Composites Part A: Applied Science and Manufacturing, Vol. 37 (5), pp
766-777, 2006.
Copyright © 2015 DuPont. All rights reserved. The DuPont Oval Logo, DuPont™,and Vizilon™ are
trademarks or registered trademarks of E. I. du Pont de Nemours and Company or its affiliates.

Durability Assessment of Heavy Duty Diesel
Engine Cylinder Head
A.H. Guzel, S.Erpolat, O. Cicekdag
Ford Otosan A.Ş, Turkey
Abstract: The recent heavy duty truck market towards higher power output, lower fuel
consumption and reduced emissions. This trend leads challenges in heavy duty diesel engine
design due to increasing thermal loads for hot components. One of the most important design
feature for heavy duty diesel engines is cylinder head fatigue durability and gasket sealing
performance. Cylinder head is subjected to low cycle thermo-mechanical fatigue (TMF) and high-
cycle fatigue (HCF) which are driven by combustion process. Gasket sealing performance also
has to be considered for avoiding any performance loss resulting from gas leakage. Fatigue life
prediction robustness of cylinder heads has reached a high level of confidence past few years;
there is still work to increase the reliability of these analyses by improving modeling
methodologies and material models. Current study will focus on heavy duty engine cylinder head
made of compacted graphite iron development methodology using coupled 3D combustion
simulation, conjugate heat transfer and both high cycle fatigue (HCF) and thermo-mechanical
fatigue (TMF) analysis methodologies. Also gasket sealing performance is calculated with a
proper methodology and it is verified by durability tests.
Keywords: Cylinder Head, High Cycle Fatigue, Thermo-Mechanical Fatigue, Engine Gasket
Failure
1. Introduction
Heavy duty engine cylinder head is probably the most complex engine component to design.
Several conflicting design aspects should be taken into consideration. At an early stage of
development of cylinder head port development to meet combustion performance and strict
emission legislations is important. Water jacket is designed to meet adequate cooling of the engine
structure to stay well below critical metal temperature limits and to achieve optimum thermal
efficiency. Gasket selection is crucial, since sealing and distortion requirements against assembly,
thermal, and firing loads. Other important aspects are manufacturing constraints, such as minimum
wall thickness, porosity, and residual stresses thermal growth has to be taken into consideration.
In addition to that, one of the most important design aspect is both low cycle and low cycle fatigue
durability A typical cast iron cylinder head reaches a maximum temperature level up to 380°C and
sometimes more whereas the minimum temperature could be below –30°C during cold winter
climates. The thermo-mechanical fatigue (TMF) of a cylinder head during such a start-stop cycle
needs to be considered in engine design. Temperature fluctuations lead to low-cycle fatigue of
cylinder head particularly around valve bridge areas, injector holes and glow plugs.

At peak temperature, the cylinder head is also subjected to mechanical loading due to firing
pressures. In a modern heavy duty diesel engine, peak firing pressures may reach up to 230 – 240
bar. Therefore, high cycle fatigue (HCF) should be considered in engine design as well. Typical
HCF critical regions are fillets of water jacket, particularly those around main bolts and injectors.
HCF is a very different phenomenon to TMF and they are usually treated independently. A typical
workflow of Ford Otosan (FO) for CAE assessment of cylinder head is shown fig. 1.
Figure 1. Typical workflow for cylinder head durability analysis in FO.
2. Thermo-Mechanical Fatigue Calculation Approach

TMF is a very complex phenomenon that is a combination of a number of different mechanisms,
such as mechanical fatigue, oxidation, and creep. There are different TMF models currently used
at Ford Otosan to evaluate fatigue life of cylinder head based on material. In this paper GJV-450
type compacted graphite iron TMF analysis methodology will be discussed.
In general thermo-mechanical fatigue life prediction models consider the interaction between
fatigue and creep at varying temperatures. The difficulty in understanding and predicting the
interaction behavior has stimulated significant research effort in finding reliable life prediction
methodologies. The models can be distinguished between damage based, stress based, strain
based, energy (work) based criterion or combined approaches. Physically they are based on e.g.
fracture mechanics, microstructural theories or empirical observation. In this paper Z-Set Zebulon
Abaqus coupled methodology is given to calculate TMF life by using modified Chaboche
approach. This methodology is used at Ford Otosan for heavy duty cylinder head made of GJV-
450. Z-Set Zebulon Abaqus coupled approach can also utilize dissipated energy based on
hysteresis loop by using stress and strain tensor calculation scripts.
The basic idea of classical Chaboche [1-2] approach is the summation of the damage caused by
different failure mechanisms. Chaboche approach introduces a term called brittle to consider cast
iron material behavior. Table 1 illustrates how this general approach is utilized by Chaboche.

Table 1. Basic equations Standart Chaboche damage model.
Damage Summation
Creep + Fatigue Damage
Brittle Damage
Under cyclic loading compacted graphite cast iron’s behavior is specified by a closure region
under compressive loading direction with previous tensile loading direction. This has led model to
formulate a damage based deformation behavior model for GJV-450 with a core elasto-
viscoplastic Chaboche type model. This model includes Von Mises type yield criteria, isotropic
and kinematic hardening and a Norton flow law. Standard Chaboche model is improved by a
damage state variable depending on the maximum principal stresses [3-4].
Figure 2. Damage energy (red area) during cyclic loading
It should be noted that current life estimation using Zebulon is actually made by post processing
inelastic dissipated energy criterion :
ΔWine = 𝐴(𝑇𝑚𝑥)(𝑁𝑓)𝑎
Where ΔWine is the inelastic dissipated energy density per cycle determined at stabilized
hysteresis loops; A is damage parameter, which is a function of maximum temperature; and α is a
material constant. In general energy based methods result in specific damage parameters as well.
In terms of physics they rely on empirical observations. Although there are many different

approaches the basic concept is the same. It is assumed that the work done upon the material by
the external forces causes fatigue damage. It includes elastic as well as plastic energy. Graphically
the energy represents the area inside stress-strain hysteresis loop (fig. 2).
3. Computational Fluid Dynamics Analysis

In Ford Otosan, for the thermal analyses of heavy duty cylinder head structure, conjugate CFD
analysis is used. Fig. 3 shows the CFD methodology for cylinder head metal temperature
calculation.
Figure 3. Conjugate Heat Transfer Modeling Approach
CFD analyses are started from 1D engine model calculations and combustion analysis. The gas
dynamics and combustion is solved by Converge CFD code which is a three-dimensional CFD
program to simulate moving surfaces of components such as piston and valves at internal
combustion engines (ICE) and is capable of solving compressible, chemically reacting, transient,
multiphase flows.
In conjugate analysis, the model in CFD environment involves both fluid and solid domains and
the temperature field of the structure is simultaneously solved with the coolant flow, and hence,
eliminates the need to define boundary conditions on the structure-coolant interface at an expense
of running times. Star CCM+ is used to perform conjugate CFD analysis, where the solid and fluid
domains are meshed with polyhedral elements and conformal mesh is produced on the interface
which minimizes the data exchange errors on the interface. Particular attention is paid to evaluate
the boiling phenomenon on water jacket interface of flame deck. In order to incorporate boiling
effect, Star CCM+’ built-in modules are used. Heat transfer coefficient (HTC) distribution of
water jacket and flame deck temperature distribution heavy duty cylinder head’s at peak power
condition is shown fig. 4.In order to maximize the cooling efficiency, water jacket is designed
top-down cooling type which has two different stratified cooling channel in cylinder head.

Figure 4. HTC distribution at water jacket & Temperature distribution on a
flame deck
5. Finite Element Modeling of Structural Analyses

Commercial finite element package Abaqus is used for cylinder head structural analyses. A typical
finite element model includes cylinder head, block, main bolts, gasket, valve seats and guides.
Valves are usually included in heat transfer analyses. A number of assembly loadings are defined
including valve seat, valve guide interferences, cylinder head bolt clamping. Thermal expansion
step following assembly loading is based on the temperature distribution obtained from conjugate
CFD analysis performed under rated power operating conditions. If it is HCF and gasket sealing
analyses, for the final set of loading steps, peak firing pressures are applied into cylinder head,
block and valve seats. Global model for cylinder head is shown in fig 5.
Figure 5. Global model for heavy duty cylinder head durability analysis
Since TMF material models are computationally expensive for Z-set Zebulon, submodels are used
for the assessment of flame deck area. Global cylinder head and submodel is modeled in Abaqus

and Z-set Zebulon only uses submodel to calculation of TMF life. For HCF and gasket failure
analysis ,no need to create submodel. Typical submodel for a heavy duty engine is shown in fig. 6
Figure 6. Submodel of TMF assessment and driven nodes

In Abaqus models nonlinear material properties for each component, gasket loading/unloading
curve and nonlinear contact definitions-interference between each component are defined.
6. Fatigue Life Post-Processing

6.1 TMF Postprocessing
Heat up and cool down steps are defined as one TMF cycle and a cylinder head must meet a pre-
defined minimum number of TMF cycle as a design criteria. Abaqus –Zebulon coupled approach
for TMF life calculation contains five cycles. In heavy duty diesel cylinder heads, critical regions
for TMF are valve bridges and injector hole’s flame face region. These regions are subjected to
out-of-phase TMF loading (fig.7). During OP-TMF loading, the oxidation mechanism becomes
significant, whereas creep damage is negligible, because void growth is suppressed in
compression [5]
.
Figure 7. Out-of-phase and In-phase loading in TMF

Fig. 8 shows TMF life assessment of heavy duty cylinder head at rated power condition. TMF
failure is expected around injector hole around 10000 cycles.
Figure 8. Thermo-Mechanical Fatigue Life Result of the most critical

cylinder
6.2 HCF Postprocessing

As next study to investigate HCF assessment of a heavy duty cylinder head, same FE model used
in the TMF analysis, however HCF analysis procedure is different .Model runs with different
loading steps while cylinder pressure is dominant loading condition in HCF failure. Different from
TMF analyses, the engine components are heated up to rated power temperatures at steady state
condition after assembly loads applied. Peak firing pressure is applied on both cylinder head,
cylinder block and valve seat’s gas surfaces as separate steps following engine firing order.
Resulted Abaqus stress file is used in FEMFAT Transmax and stress amplitudes, mean stress and
constant stress are calculated at each node using scaled normal stress in critical plane method.
Surface roughness, range of dispersion and isothermal influences are taken into account
calculating local S-N curve at each point. The resulting stress amplitudes are used together with a
tension/compression S-N curve for the calculation and linear summation of partial damages
according to Miner rule. The plane with maximum damage is assumed to be most critical.
Fig. 9 and fig. 10 shows minimum safety factor calculated at different cross section in a heavy
duty cylinder head. According to results, the hottest spot is on lower water jacket bottom around
intake ports (fig. 10).

Figure 9. HCF results for top water jacket
Min. Fatigue safety factor (FSF) reported at these locations is around 1.2. Hot spots also
investigated around exhaust ports shown at fig 7. Minimum FSF calculated at these locations are
around 1.18. FSF around bolt columns are also shown.
Figure 10. HCF Results bottom water jacket
7. Gasket Failure Analysis

Ford Otosan has developed a methodology that predicts gasket sealing performance and gasket
failures between cylinder head and engine block. Methodology is very useful at early selection
stage of gaskets.
As a final study to investigate the gasket failure assessment of heavy duty cylinder head, again
same FE model that is used in TMF and HCF analysis is handled. Number of steps and procedure
is same with HCF analysis. However, post-processing methodology is different. Ford Otosan has
in-house developed post-processor is used to perform gasket failure analysis .After completed
post-processing stage results are compared with test results.
Nonlinear gasket behavior curves and special Abaqus gasket elements are used to simulate gasket
sealing performance during rated power condition. Both thermal and peak fire pressure applied to
the model. Gasket FE model of one of the cylinder is shown in fig. 9.

Fig. 9: Gasket FE model of one of the cylinder
This type of a heavy duty engine generally includes liners for each cylinder. This makes the
problem a challenging task.
Fig 9 shows a typical heavy duty engine liner, engine block and cylinder head assembly
representation. It is important that values of protrusion and recess should be input correctly to the
model.
Fig. 9: Representative Liner Positions
Abaqus offers gasket modeling approach to overcome modeling issues about assembly of gasket
between block and head. Both recess and protrusion can be modeled by defining initial gap
selection of Abaqus gasket behavior [6].
Correlation between Abaqus results and test results are shown in fig 10. Maximum dynamic gap
between gasket beads and cylinder head is occurred and reaches the critical value 1.1 times. It can
be seen that very good correlation is observed.

Figure 10. Correlation CAE results with test
8. Conclusions
Failure of a cylinder head is mainly caused by both high cycle fatigue due to high firing loads and
thermo-mechanical failure due to thermal loading cycle during start-stop of an engine. A
methodology is developed to predict both of these fatigue failure and another failure mechanism is
gasket failure in an engine head based on combination of combustion model, CHT model with
both TMF and HCF calculation tools. Future strategies like integration of material inhomogeneity
and residual stresses will be implemented into durability analyses. Also component level rig tests
to improve both CAE studies and Design Verification (DV) efficiency will be studied.
9. Acknowledgement
Authors would like to thank Cengizhan Cengiz and Serdar Guryuva for leading CFD work and Dr.
Ronald Foerch for their support in Z-set Zebulon software. The authors wish to express their
gratitude Yigit Yazicioglu ,Demirhan Manav and Aidin Dairo to help in developing the TMF and
HCF and test methodology.
10. References
1. Chaboche, J.L.: Time-independent constitutive theories for cyclic plasticity. Int. J. Plast. 2,
149-188, 1986
2. Chaboche, J.L.: Constitutive equations for cyclic plasticity and cyclic viscoplasticity. Int. J.
Plast. 5, 247-302, 1989
3. F. Zieher, F. Langmayr, A. Ennemoser, A. Jelatancev, G. Hager, K. Wieser : Advanced
Thermal Mechanical Fatigue Life Simulation of Cylinder Heads” , 2004 ABAQUS
Users’Conference

4. R. Foerch, F. Zieher, and F. Langmayer “TMF Applications Using a Damage-Based Cast Iron
Model”, 2002 ABAQUS Users’ Conference
5. Minichmayr, R.; Riedler, M.; Winter, G.; Leitner, H.; Eichlseder, W.: Thermo-mechanical
fatigue life assessment of aluminium components using the damage rate model of Sehitoglu.
International Journal of Fatigue, Volume 30, Issue 2, February 2008, Pages 298-304

Prediction of Compressive Characteristic of Hat
Section High-strength Steel Frame Taking the
Bauschinger Effect into Account
Kenji Takada a, Masaki Matsuo a, Kenichiro Fukagawa a, Kentaro Sato b,
Pierre Farahat c , Lamine Diallo d , Yangwook Choi e , Anwer Yasin c
a
Honda R&D Co., Ltd. 4630 Shimotakanezawa, Haga-machi, Haga-gun, Tochigi, 321-3393 Japan
b
JFE Steel Corporation, 1 Kawasaki-cho, Chuo-ku,Chiba 260-0835, Japan
c
Dassault Systèmes K.K., ThinkPark Tower, 2-1-1 Osaki, Shinagawa-ku, 141-6020 Tokyo, Japan
d
Dassault Systemes Canada Inc. 393 St-Jacques Street West, Suite 300 Montreal, QC H2Y 1N9 | Canada
e
Dassault Systemes Simulia Corp, 1301 Atwood Ave., Suite 101W, Johnston, RI 02919, United States
Abstract: In recent years, the usage of high-strength steel is increasing in order to improve the
rate of body weight reduction and the safety performance of a vehicle from crash. The
Bauschinger effect is generally strong in high-strength steel sheets. Material properties used in
crash analysis is conventionally treated as isotropic hardening material. However, the analysis
results such as average loading and buckling wavelength from analysis deviates with those from
tests, which is observed in some axial collapse problems of hat section frame that is composed of
high-strength steel sheet. When axial collapse occurs in the hat cross section frame, buckling
occurs along with a frame ridge rolling. The Bauschinger effect will occur in the steel sheet at
both front and back surfaces as the rolling progresses. In order to perform the axial collapse
analysis considering the Bauschinger effect, performing high-strength steel sheet tension-
compression tests and calibrating the parameters of the Bauschinger effect are necessary. In this
study, a correlation tool has been developed to perform the correlation process by using Isight
and Abaqus. It is confirmed that the analysis results fit test results in good precision in both
Chaboche and modified ANK models using the parameters generated by the tool.
Keywords: Material models, the Bauschinger effect, Chaboche, Hardening, Crash analysis.
1. Introduction
The use of high-strength steels in automobile body structures has been considered an efficient
solution because of the lower cost of volume production in the automotive industry. With the
improvement of the formability of high-strength steels by optimizing the metallic microstructure,
the range of application of high-strength steels has been expanding in automobile body structures.
Recently, advanced high-strength steels with strengths from 980MPa to 1500MPa used in
automotive bodies have played a crucial role in crash safety performance. To design and
manufacture automotive parts from high-strength steels, simulation has become an indispensable
tool for the prediction of energy absorption during crash of a vehicle.
The Bauschinger effect has been studied in plastic forming field, and there are many models such
as Armstrong and Frederick model, Chaboche model, ANK model, ZANG model and etc. have
been proposed for representing the Bauschinger effect. On the other hand, the Bauschinger effect

is not considered in crash analysis fields. In many recent study of static crash analysis of hat-
section problems, the amount of energy absorption was found higher than the ones from the tests.
For studying the Bauschinger effect in crash analysis, author (1) has performed many tension-
compression tests since 1999 but the author did not always satisfied with the results because of
occurrence of the buckling in the compression period. After adding buckling prevention back
plates to the tesreg in 2012, additional tests were performed by JFE. As the result, we are
succeeded to obtain test results with high accuracy for tension-compression test using high-
strength steel materials. We studied on FEA analysis to reproduce the characteristics of the
tension-compression model. For the first, using the kinematic hardening component of Abaqus, we
performed calibration of material parameters for the first back-stress of Chaboche model. And
then we performed static crash analysis of both hat section frames and hexagonal section frames
using those parameters. It was found that the analysis results were reproducing test results with
good accuracy. During the study, we developed a tool for calibrating the material properties of
Chaboche and modified ANK models up to the second back-stress quickly. From analysis results,
it was found that modified ANK model was more effective for obtaining results with high
accuracy in tension-compression crash analysis.
2. Axial crush tests and calculations

The axial crush tests and simulations with hat section and hexagonal section frame were
performed to investigate the energy absorptions of each frame. The steel grades used in this test
were 590MPa and 780MPa grade. Figure 1 shows the specification of the hat section and
hexagonal section frames. The hat section is 60mm x 60mm cross section, 20mm flange width,
and thickness t=1.6mm while the hexagonal section is 34.64mm sides, 20mm flange width, and
thickness t=1.6mm. The width-to-thickness ratios (C/t) of the hat section and hexagonal section
were 37.5 and 21.7, respectively. Therefore, it was supposed that these frames had plastic buckling.
The amsler tests were performed with a constant velocity of 1mm/sec. The Table 1 shows the
material properties for two classes of steels.
Figure 1. Dimension of the hat section and hexagonal section

Table 1. Material properties for two classes of steels
2.1 FEA models

The hat section frame of 60mm x 60mm cross section and length L=300mm with thickness
t=1.6mm was used. Furthermore, the hexagonal section frame of one side a=34.64mm and length
L=300mm with thickness t=1.6mm was also considered in this study shown in Figure 2. The fixed
boundary condition was applied at the bottom end of both frames. For the top end of both frames
five degrees of freedom ( 𝑢𝑦 , 𝑢𝑧 , 𝜃𝑥 , 𝜃𝑦 , 𝜃𝑧 ) were fixed to generate the plastic buckling in the
middle of frames. The enforced velocity was applied at the top end of the frames. FEM models
were modeled with 4-node shell element using Abaqus element type S4.
b b a
Enforced velocity
a
110 111
300 ㎜
111 111 x
Figure 2. Dimension of the hat section and hexagonal section of FEA
2.2 Determination of mesh pitch

In this study, the element length (mesh pitch) of a 4-node shell element was decided based on the
length of the plastic buckling folding wave. There are several equations to calculate the plastic
folding wave length of thin-walled columns. In this study, Abramowicz and Wierzbicki equation
was applied as shown in Equation 1 (Wierzbicki, Recke, Abramowicz and Gholami, 1994).
H = t1 / 3b 2 / 3 (1)
1
C= (a + b ) (2)
2

where H is a quarter folding length, t is the thickness, C is the mean value of two sides (a,b) of
the cross section.
Figure 3 shows the plastic folding wave of thin-walled prismatic frame. It is supposed that plastic
folding wave shows a cubic curve in the cut section (Wierzbicki and Abramowicz, 1989).
Therefore, it is assumed that the total number of the divisions to predict the folding wave correctly
is 7 or 8. According the Equation 1, the length of plastic folding wave of hat section frame and
hexagonal section frame are 35.2mm and 24.3 mm respectively. Consequently, in the FEA, it was
defined that the mesh pitch of hat section was 5mm and hexagonal section was 2.88mm as shown
in Table 2.
8 divisions 7 divisions
F 𝛿𝛿
b
1 1
a 2 2
3 3
2H 4
H 4
5
H
t
5
6
6
7
8 7
Figure 3. Relationship between plastic folding wave and mesh pitch
Table 2. Element length of FEA models

Length of a plastic
Element length
folding wave, 2H
Hat section 35.2mm 5mm
Hexagonal section 24.3mm 2.88mm
2.3 Stress-strain curves

In FEA, the swift equation shown in Equation (3) is applied to express the stress-strain relation
after uniform elongation (n-value) (Datsko, 1966).
(
σ =σ0 ε0 + ε )
pl n
(3)
where 𝜎0 , 𝜀0 and 𝑛 are the strength coefficient, yield strain and strain hardening exponent
respectively. Figure 4 shows the true stress-strain curves for two classes of steels used in FEA.

1400
n-value
1200
1000 Swift equation

true stress (MPa)
780MPa
800 Swift equation
590MPa
600
400
200
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
true strain
Figure 4. Dimension of the hat section and hexagonal section of FEA
2.4 Axial crush calculation

Figure 5 shows the force-stroke and energy-stroke curves of the hat section frames and the
hexagonal section frame obtained by the experiment and Abaqus. In Figure 5 (a), (b), the energy
absorption of 590MPa and 780MPa grade from FEA were larger than test results. Meanwhile, in
the case of hexagonal section frames shown in Figure 5 (c), (d), the energy absorption of 590MPa
and 780MPa grade calculated by FEA were much larger than test results. In the FEA, the isotropic
hardening material model of Abaqus was applied. The 590MPa and 780MPa grade are high-
strength steels. It was supposed that both steels had to consider the Bauschinger effect.

Calculation error
Calculation error
Calculation error
Calculation error
Figure 5. Force-stroke and energy-stroke curves of the hat section frames

3. Modeling of the Bauschinger effect
3.1 Moving hinge lines in deformation mechanism

In this study, it was considered the reason why energy absorptions of 590MPa and 780MPa grade
frame had calculation errors. Figure 6 shows the deformation mechanism of thin-walled prismatic
frame. The deformation of thin-walled prismatic frame is progressed by the folding of the wall.
Initial hinge line (1-C-2) moves to point B (1-B-2) by the folding of wall, which is named as
moving hinge line. In the initial condition, the point B is on the middle of the wall. However,
when the initial hinge line (1-C-2) moves to point B, the tensile stress occurs at the surface of the
point B (reverse face of point B becomes compression). In the next step, when the moving hinge
line (1-B-2) moves to point A (1-A-2), the compression stress occurs at the surface of point B (on
the surface of the other side point B becomes tensile). In other words, in the case of axial crush
problem of thin-walled frame, along with the moving hinge line the material experiences cyclic
loading. Therefore, the Bauschinger effect should be considered.
1 1
1
2 2 2
Figure 6. Moving hinge line in deformation mechanism
Figure 7 shows the energy absorption of point B. If the material model is for isotropic hardening,
the energy absorption of point B is expressed in Figure 7 (a). In contrast to that if the material
model is for complex hardening (the Bauschinger effect), the energy absorption is expressed in
Figure 7 (b). Hence, the energy absorption of isotropic hardening is larger than complex hardening.
It is supposed that the Bauschinger effect is the reason why energy absorptions of 590MPa and
780MPa grade frame had calculation errors.

(a) Isotropic hardening (b) Complex hardening (Bauschnger effect)
Figure 7. Energy absorption of isotropic and complex hardening
3.2 Tensile and compression tests

Tensile and compression test was performed to measure the Bauschinger effect of standard high-
strength steels, which was performed by JFE steel. Figure 8 shows the measuring system for the
Bauschinger effect. Basically, in the tension and compression test, buckling of the test specimen
occurs in the compression state. In this system, the buckling prevention back plate was used to
prevent the buckling. High quality oil was added between buckling prevention back plate and test
specimen to decrease the friction.
Figure 8. Tensile and compression test with buckling prevention back plate
Figure 9 shows the tensile and compression curves of 590MPa and 780MPa grades obtained by
tests. In the case that FEA supposes the stress-strain curve as isotropic hardening; the dotted lines

are stress-strain curves of FEA in compression state. Its flow stress is larger than test and Young’s
modulus is also larger than test.
Stress-strain curves of CAE supposing the isotropic hardening
(a) 590MPa grade (b) 780MPa grade
Figure 9. Tensile and compression curves of 590MPa and 780MPa grade
3.3 Numerical models for the Bauschinger effect

Some materials show that the yield stress obtained in the reloading or reversed directional loading
is substantially less than the yield stress obtained in the original loading direction. It has been
observed in polycrystalline metals as well as single crystal metals. The behavior is called as the
Bauschinger effect which can be described by the kinematic hardening model. Some materials
show that the yield stress for the cyclic loading does not reach the first yield which we call
permanent softening. These behaviors can be explained by the dislocation pile up and tangles due
to the previous loading cycles and can also be described by Armstrong-Frederick type backstress
evolution model for the transition of the yield surface (Armstrong and Frederick, 1966). The
understanding of the Bauschinger effect and implementation of the effect in the plastic
deformation are important to predict the material deformation correctly. There are many models
that can describe the Bauschinger effect: Chaboche (Chaboche, 1986), modified Chaboche
(Chaboche, 1989), ANK (Chun et al, 2002), Frederic-Armstrong (Armstrong and Frederick, 1966),
etc. These models are believed to be the most suitable algorithms for describing the behaviors of
high-tension steels.
The underlined material behavior cannot be explained by isotropic hardening model; hence,
kinematic hardening model was introduced
F = f (σ − α ) + σ 0 = f (σ − α ) + σ 0 + R (4)

where σ is stress tensor, α is backstress tensor, and R is the isotropic hardening part. The
backstress α can be considered to be the sum of multiple backstress component tensors
α= ∑ αi . (5)
i
Each component of the backstress is assumed to evolve independently as described by Armstrong-
Frederick type backstress evolution (Ref):
Ci
α = 0
(σ − α )ε pl − γ iα iε pl (6)
σ
where Ci and γ i are material parameters that should be defined through cyclic loading tests. The
parameter γ i determines how fast the i th backstress saturates to a certain value. If γ i is equal to
zero, then the backstress linearly increases. For the multiple backstress evolution, Armstrong-
Frederick describes each component as evolving independently; therefore, different kinematic
shifts can be considered for the initial and reversal loadings. Chun (Chun et al, 2002) proposed the
ANK model which captures the permanent softening in the cyclic loading by evolving the second
backstress as
C1
α 1 = (σ − α )ε pl − γ 1α1ε pl (7)
σ0
 C2
 (σ − α )e pl − g 2α 2e pl for initial loading
α 2 = s 0 (8)
 0 for reversal loading
The second backstress is evolving with respect to the loading path history. The evolution of the
second backstress is then considered as a function of the angle between the previous and current
loading paths. In order to determine the loading direction, the strain increment direction is
monitored by measuring the angle between the previous loading and current loading (Choi et al,
2006)
 ε ⋅ ε* 
 p p
ϕ = cos −1   (9)
 ε p ε *p 
 
*
where ε p is for the current loading and ε p is for the previous loading. If ϕ > 90 o , then it is
considered to be in the reverse direction.
The isotropic expansion of yield surface in this Chaboche model is described by
σ t = σ 0 + RANK + α . (10)
where σ 0 is the yield stress when ε pl (the equivalent plastic strain) is zero.
The isotropic hardening part of ANK model can be written as,
(
R = RANK = K 1 − e − Nε
pl
)− Cγ (1 − e )− C ε
1 −γε pl
2
pl
with
C1
γ
− α1 ≥ 0 (11)

R describes the change of yield surface size from the initial yield and K, N are curve fitting
parameters and α1 is a scalar version of the backstress for a uniaxial loading condition. Therefore,
the material parameters to be determined are C1 and γ 1 . In this study, we modified the ANK
model to turn the second backstress to zero for the reverse loading, and it fits better to the high-
strength steel. These models have been tested by using VUSDFLD and corresponding field
variable that is monitoring the loading direction to turn off the second backstress when the reversal
loading is detected.
3.4 Calibration of material model parameters using Isight
In this study, a program for calibrating Chaboche material model parameters was developed,
which can be used i) to extract plastic strain-stress data from test results and graphically compare
with results generated by selected Chaboche model using material parameters inputted into GUI
directly; ii) to calibrate material parameters of Chaboche model by using an predefined Isight
workflow for Half cycle model and full cycle models (ZANG and modified ANK) respectively;
iii) and to perform confirmation analysis by updating material parameters in a template Abaqus
model and submit it with an user-subroutine to reproduce the Bauschinger effect in analysis.
Figure 10. A Java program for fitting material parameters of Chaboche models
When button [Fitting (Isight)] on the program is pressed, a workflow shown in Figure 11 will be
performed and the values of the material parameter fields will be updated with the fitted values
automatically. The workflow consists of two branch processes for fitting parameters for half cycle

and full cycle models separately. To improve the performance of the calibration processes, each
process is separated into two phases for fitting isotropic and kinematic hardening components
separately. The material parameters fitted will be written into a text file using Data Exchange
component for later use.
Figure 11. Workflow for fitting material model parameters
4. FEA results taking the Bauschinger effect into account

4.1 Axial crush calculation with the Bauschinger effect
Figure 12 shows the force-stroke curves and energy absorption curves of hat section frame and
hexagonal section frame with 590MPa and 780MPa grades. In this simulation, Chaboche one back
stress parameters obtained by Isight-Abaqus system were applied. When considering the
Bauschinger effect, it was found that the accuracy of the energy absorption was greatly improved.

Figure 12. Force-stroke and energy-stroke curves with the Bauschinger effect

4.2 Three-point bending simulation with the Bauschinger effect
To investigate the Bauschinger effect, three-point bending tests were performed as shown in
Figure 13 (Takada, Sato and Ma, 2015). The hat-section of a 70mm x 78mm, with 20mm in the
flange width and 1.6mm in the thickness was used. The steel grade of hat-section was 590MPa.
The angles of the left corner and the right corner were 98 degrees and 102 degrees, respectively.
The hat section was assembled with a back plate with a 980MPa grade and a thickness t=1.6mm.
The radius of the punch was 112mm and the radius of the support cylinders was 12.5mm. The
friction factor between the back plate and the support cylinders was about 0.15. The span of the
beam between the two supported points was 700mm. The punch was lowered with a constant
velocity of 100mm/min. Figure 14 shows the FEA model of a three-point bending frame. In this
three-point bending simulation, Chaboche one back stress parameters obtained by Isight-Abaqus
system were applied.
P 100 mm/min
(b)
R=112
70
L=700mm
100 100 back plate t=1.6mm

980MPa grade
Figure 13. Dimensions of three-point bending model
Figure 14. FEA model of three-point bending model
4.2.1 Determination of mesh pitch

In this three-point bending simulation, the mesh pitch of 4-node shell element was decided based
on the Equation (12) (Wierzbicki, Recke, Abramowicz and Gholami, 1994).
H = 1.276 ⋅ t1 / 3b 2 / 3 (12)
In this case, it is also assumed that the total number of divisions to predict the plastic folding wave
accurately is 7 or 8. According the equation (14), the length of plastic wave 2H of this frame is
54.5mm. Consequently, the mesh pitch of this frame use the 7.76mm.
4.2.2 FEA results

Figure 15 shows the force-stroke curve of the experiment result, the FEA results with and without
the Bauschinger effect, respectively. When considering the Bauschinger effect, it was found that
the accuracy of force-stroke curve was greatly improved.
20 1000
900
Experiment 800
15 実機
FEA isotropic 700
CAE Abaqus(isotropic)
FEAAbaqus(combined)
CAE with Bauschinger effect
600
Force (kN)
E（J）
10 実機 500
CAE Abaqus(isotropic)
CAE Abaqus(combined) 400
300
5
200
100
0 0
0 5 10 15 20 25 30
Stroke (mm)
Figure 15. Comparison of force-stroke curves between the experiment

and FEA results
5. Modified ANK model

In this study, it was found that Chaboche one back stress model was efficient to improve the
energy absorption of hat section and hexagonal section frames. In the crash problem, most
important point is to identify the compression state accurately. In other words, stress-strain curves
of FEA should to reproduce the first tensile curve and next compression curve only with high
accuracy, which is half cycle of tensile and compression curve. However, in the case of stamping
simulation, FEA have to reproduce the full cycle curve of tensile and compression curve. In this
case, modified ANK model will be very efficient to produce the full cycle curve. Figure 16 shows
the identified results for standard high- strength steels.

Figure 16. The identified results for standard high- strength steels with
modified ANK model.

6. Conclusions
From this study, the following conclusions can be drawn:
The Bauschinger effect parameters of Chaboche one back stress model and modified ANK model
were identified efficiently by using the Isight-Abaqus calibration program.
In the case of energy absorption of axial crash problem, it was found that by using Chaboche one
back stress model, the energy absorptions of hat section frame and hexagonal section frame with
high-tension steels were improved accurately.
Three-point bending test and simulation supposing an actual automobile structure were performed.
By using Chaboche one back stress model, it was found that the energy absorption was accurately
improved.
The tensile and compression tests were performed to measure the Bauschinger effect parameters
for standard high-strength steels. By using the modified ANK model based on Isight-Abaqus
calibration program, it was found that program results correlated well with the tensile and
compression tests. It will be supposed that this system is useful for future stamping simulation.
7. Acknowledgments
The authors would like to thank the late Dr. W. Abramowicz of the Impact Design Europe for his
valuable comments.
8. Reference
1. Armstrong, P.,J., Frederick, C.O., “A Mathematical Representation of the Multiaxial
Bauschinger Effect,” G.E.G.B. Report RD/B/N 731, 1966
2. Chaboche, J.L., “Time independent constitutive theories for cyclic plasticity,”
International Journal of Plasticity 2 (2), 149, 1986
3. Chaboche, J.L., “Constitutive equations for cyclic plasticity and cyclic viscoplasticity,”
International Journal of Plasticity 5, pp.247-302, 1989
4. Choi, Y., Han, C.-S., Lee, J.K., Wagoner, R.H., “Modeling multi-axial deformation of
planar anisotropic elasto-plastic materials, part I: Theory,” International Journal of
Plasticity 22, pp. 1745-1764, 2006

5. Chun, B.K., Jinn, J.T., Lee, J.K., “Modeling the Bauschinger effect for sheet metals, part
I: theory,” International Journal of Plasticity 18, 571-595, 2002
6. Datsko, J., “Material Properties and Manufacturing Processes,” John Wiley & Songs,
New York, pp. 13-40, 1966.
7. Takada, K., Sato, K. and Ma, N., “Fracture Prediction for Automotive Bodies using a
Ductile Fracture Criterion and a Strain-Dependent Anisotropy Model,” SAE Paper 2015-
01-0567, 2015.
8. Wierzbicki, T., Recke, L., Abramowicz, W., and Gholami, T., “Stress profiles in thin-
walled prismatic columns subjected to crush loading-П. Bending,” Computer &
Structures, 51(6):pp. 624-640, 1994.
9. Wierzbicki, T., Recke, L., Abramowicz, W., and Gholami, T., “Stress profiles in thin-
walled prismatic columns subjected to crush loading-І. Compresion,” Computer &
Structures, 51(6):pp. 611-623, 1994.
10. Wierzbicki, T. and Abramowicz, W., “The Mechanics of Deep Plastic Collapse of Thin
Walled Structures Chapter 9” In Structure Failure. Wierzbicki and Jones (Editors). Wiley,
New York,1989.
11. Zang, S.-L., Lee, M.-G., Kim, J.H., “Evaluating the significance of hardening behavior
and unloading modulus under strain reversal in sheet springback prediction,”
International Journal of Plasticity 77, 194-204, 2013
Contact Information
Kenji Takada, Chief Engineer
Honda R&D Co., Ltd. Automobile R&D Center 4630 Shimotakanezawa, Haga-machi, Haga-gun,
Tochigi, Japan, 321-3393 Kenji_Takada@n.t.rd.honda.co.jp +81-80-9150-1451

The Effect of Rubber in Strength Analysis for Strut
Insulator Assembly
Hoo-Gwang Lee, Ji-Hun Kim, and Sung-Jin Yoon
Chassis CAE Team, Hyundai Motor Company
Abstract: The strut insulator assembly in the suspension system which supports shock absorber,
spring and bump stopper takes large upward load when vehicle crosses pot hole or speed bump
with high speed. As the permanent deformation of insulator bracket could cause noise problem,
the development of insulator bracket strength is required. Though the main bumper stopper load
applied to steel bracket directly in the inner core type insulator, there is rubber between steel
bracket and bumper stopper in the cup molded type insulator. Because there could be some
convergence problem caused by large deformation of rubber, the modeling and strength analysis
procedure for the cup molded type insulator is set up in this study. A series of step-by-step trials
would give an insight in setting up an analysis procedure of similar problem. The Yeoh hyper-
elastic model for rubber material property is used when the large load is applied to rubber and
the strain is high. However rubber increases the strength of permanent deformation, the change of
rubber stiffness does not have a great influence on the strength result. Although the total
displacement and initial stiffness of system is quite different, the comparison between test and
analysis with different bush stiffness gives same strength level.
Keywords: Assembly Deformation, Hyperelasticity, Rubber, Strut insulator, Suspension
1. Introduction
The strut assembly is the main part which sustains body of vehicle against various types of load. It
is composed of spring, shock absorber and bumper stopper. Also there is an insulator assembly at
the top of this system to link suspension and body of vehicle. It is composed of core, bracket and
rubber (Fig.1). Rubber which is attached to both sides of core and bracket isolates small vibration
from road. Large upward force caused by pot-hole or speed bump on the road is transferred to
insulator assembly through bumper stopper. If the strength of insulator assembly is not enough,
the permanent deformation or break of bracket could happen. Because the permanent deformation
of bracket can make some squeak noise during running, the strength development of bracket is
required. Inner core type strut insulator assembly in which core and rubber are enclosed by bracket
is widely used so far. However it is simple and robust, there could be some noise at the contact
region between strut rod and bumper stopper. This noise caused from the independent motion of
strut rod and bumper stopper. In this point of view, cup molded type strut insulator in which strut
rod and bumper stopper moves as one part is better choice. As the emotional quality including
noise is going to be critical sales point, the cup molded type insulator is adopted more than ever.
In the strength analysis of insulator assembly, these two types of insulator systems have quite
different load path. The upward load from the bumper stopper in the inner core type strut insulator

is applied only to the bracket and rubber could be ignored during strength analysis. But there is
rubber in the load path of the cup molded type strut insulator and the large load transfer through
rubber gives some troubles in the implicit analysis. Some results for case studies during
establishment of strength analysis process for the cup molded type will be shared in this work.
Strut insulator assembly

Upper bracket Upper cup
Spring Bearing
Spring pad Rubber Bracket

Shock absorber Core
Lower bracket Lower cup Bearing plate
[Inner core type] [Cup molded type]
Figure 1. Two types of strut insulator assembly and their structure
2. Strength analysis for inner core type insulator

A plunger gives load to the bracket which is put on the jig in the strength test of the insulator
assembly. The bottom area of the plunger is same with that of bumper stopper. The change of
inclination in the force-displacement curve means the start of permanent deformation for the inner
core type insulator assembly and that force level is checked as the strength. The strength analysis
for this type is simple and the analysis result shows quite good correlation between analysis and
test (Fig.2).
Plunger reaction force
Plunger displacement
Figure 2. Strength analysis and test result for inner core type

3. Strength analysis for cup molded type insulator
3.1 Composition of analysis model
Rubber is molded with lower cup and bracket at the same time and then upper cup is inserted to
the top of the assembly. A bearing plate which sustains bearing for spring is also molded at the
same time. Sometimes stopper plate could be added at the lower cup to restrict upward
displacement of strut rod. The materials for cup, bracket and bearing plate are some sorts of steels
as shown in the Table 1. The material property for rubber is selected from our suspension rubber
material database which is made up with SIMULIA and AXEL PRODUCT according to the Shore
A hardness scale. However the selected rubber is not exactly same with that of insulator, the effect
of rubber property to the permanent deformation of metal parts might be in relative small. To see
the effect of rubber material, another rubber with different hardness is tested in this analysis. The
Yeoh hyper-elastic material model is used as rubber material property because stability at large
strain is required. The loading and boundary condition for the strength analysis are same with
those for the inner core type test.
Table 1. Material list used in this analysis.
Part Material name Rubber Shore A hardness scale
Cup SAPH440 Original Hs55
Bracket SAPH440 Compared Hs70
Bearing plate SPCC
3.2 Rubber model with fine C3D10H element

The second order tetrahedral element is used in finite element modeling, C3D10 for steel plates
and C3D10H for rubber with less than 2mm element length (Fig.3). Nodes on bolted to jig are
fixed and contact pairs are assigned between jig and bracket interfaces. Rigid elements are used
for jig and plunger. Jig is fixed at the reference node and the plunger is able to move freely only in
the stroke direction. As plunger presses rubber, the distortion of element happens easily. At last
the analysis diverges with only little stroke. Convergence might be impossible with C3D10H in
this case.
Rubber
Section view of FE model Distortion of rubber element during analysis
Figure 3. Strength analysis with fine C3D10H element in rubber

To avoid distortion problem of rubber element, first order tetrahedral element, C3D4H, is used for
rubber. The stroke of plunger is larger than before and the distortion of element is removed.
However the contact between rubber and lower cup happens normally, some rubber elements curl
up in a ball (Fig.4). Nodal residual forces at those nodes are too large to converge. This means that
curling up of rubber element is the main problem to resolve.
3.4 Rubber model with coarse C3D4H element

To remove curling up problem of rubber, rubber elements are re-meshed using larger element
length (Fig.5). The stroke of plunger is long enough to see the permanent deformation of steel
plates in this case. The slope change happens twice in the force-displacement curve of plunger.
First change happens when contact between lower cup and rubber starts and the second change
happens when permanent deformation of steel bracket starts. So the force at the start point of
second slope change is the strength value for this bracket. However it is easy to generate
tetrahedral element using automatic mesh generator in pre-processor, it is difficult to control mesh
shape of rubber inside. It can give rise to the curling up of elements from time to time. Besides,
first order tetrahedral element is well known for stiff behavior. For these reasons, another stable
method of modeling is required.
FE model Plunger displacement Strain distribution (LE)
Figure 5. Strength analysis with coarse C3D4H element in rubber

It is not easy to make hexagonal element in rubber part because the outside region of rubber part is
not the cylindrical shape. To save modeling time, reducing modeling region which does not affect
largely to the strength analysis in rubber part is required. Strain distribution result in the previous
analysis (Fig.5) shows no contribution of rubber elements outside from bearing plate to the result.
However the ignoring outside part of rubber in meshing makes modeling much easier, the curling
up problem of rubber elements (Fig.6) happens again. The comparison of force-displacement of
plunger shows that the stiffness of first order tetrahedral element is too large, also. The solution
for removing curling up problem is now adopted.

FE model
Curling up of rubber element
3.6 Rubber model with coarse C3D8H element

It is simple to compose coarse rubber model from finer one and the strength analysis ended up
successfully. The force-displacement curve (Fig.7) shows that the stiffness difference between fine
and coarse element of rubber is ignorable. However the second change of the slope from the
coarse first order tetrahedral element seems to be same with this result, it is necessary to check up
the effect of rubber stiffness to the start of permanent deformation of steel bracket.

Plunger reaction force Strain distribution (LE)
FE model Plunger displacement
Figure 7. Strength analysis with coarse C3D8H element in rubber
3.7 Rubber hardness effect to the permanent deformation of bracket

Two different rubber materials in Shore hardness scale, Hs55 and Hs70, are used in the same FE
model (Fig.8). The harder rubber increases the initial slope of force-displacement curve of plunger
as expected. But, the second and third slopes are almost same with the softer one. Moreover, the
reaction force when the second change of slope starts is exactly same. If the element type is same,
the strength of bracket does not differ according to the rubber stiffness.
Figure 8. Rubber hardness influence to the bracket strength

3.8 Comparison with the test
The test and analysis results of a cup molded type strut insulator for mid-size sedan are compared
(Fig.9). However the rubber stiffness in analysis differs from test specimen, the strengths for
permanent deformation are almost same. As the convergence with softer rubber material tends to
be bad, it is helpful to use harder rubber for improving convergence. So, the standard material
property for insulator rubber is selected as somewhat harder one according to this result.
Deformed shape after tes

Figure 9. Analysis and test result comparison
4. Conclusion
The modeling and strength analysis procedure for the cup molded type insulator are set up in this
study. The Yeoh hyper-elastic material model gives successful convergence during large load
application. However the difference of rubber hardness makes the difference of initial slope in
force-displacement diagram, it does not have a great influence on the permanent deformation of
steel bracket. It is no need to have all rubber tested to get precise material properties when the
analysis is focused on the strength of steel parts. But precise rubber material properties should be
required when the global displacement and the strain of rubber are important such as controlling
gap problem or durability of rubber.
5. References
1. Abaqus Users Manual, Version 6.12, Dassault Systémes Simulia Corp., Porvidence, RI.
2. Robert H. Finney,“Engineering with Rubber”, 1992.

Study on Body Rigidity, Strength and Fatigue
Performance for Carbon Fiber Reinforced Plastics
Vehicles
Sehee Oh, Yuna Son
Hyundai Motor Company
Abstract: Weight reduction of car is now main issue in order to increase vehicle fuel efficiency.
Among various lightweight materials, CFRP (Carbon Fiber Reinforced Plastics) is at the head
because of superior stiffness and strength to weight ratio.
A durability performance development process of CFRP vehicle was studied, and applied to fuel
cell concept car Intrado (code name HED9) in order to check compatibility of the process.
CFRP material was identified as a solution for satisfying weight lightening and performance of
vehicle concurrently by comparing with general steel structure vehicle.
Keywords: Composite, Vehicle, Stiffness, Strength, Fatigue
1. Introduction
Figure 1. Characteristics of specific stiffness and strength
Weight reduction of car, related with increase of fuel efficiency and decrease of CO2 emission gas,
is main issues for vehicle industry. Use of light materials can be a good solution and CFRP
(Carbon Fiber Reinforced Plastic) leads weight lightening trends in vehicle development. Figure 1
shows high specific stiffness and strength characteristics of CFRP comparing other materials.

2. Intrado (HED9) : CFRP Concept Car
Figure 2. Intrado Full car & BIW with chassis
Figure 3. Intrado BIW (skin & space frame)
Figure 4. Making process of AXONTEX beam

Intrado(code name HED9), HMC fuel cell concept car, is debuted at 2014 Geneva motor show.
Figure 2 and 3 show full car and BIW shape. CFRP space frame production process was used to
make BIW of Intrado. Figure 4 shows making process of a base beam named AXONTEX BEAM.
This process can make a closed box beam has multiple closed box beam at inner section.

3. Analytical study of body performances of CFRP vehicle
3.1 Modeling
Space frame of HED9 is made of braided carbon fiber. In order to express fiber orientation such as
Figure 5, material properties from specimen tests are need. However, specimen extraction from
beam structure is almost impossible. So unidirectional fiber specimens were made and material
properties were generated such as Table.1. At this study, lay-up approach using unidirectional
properties was used to express braided composite as Figure 6.
Figure 5. Fiber orientation of braided composite pipe
Table 1. Mechanical properties of unidirectional FRP

Stiffness Strength (MPa)
vf 0.56 XT 1963
3
ρ 1.5 g/mm XC 689
E1 117 GPa YT 22
E2 9.5 GPa YC 110
ν12 0.3 SL 54
G12 4.7 GPa ST 56.7
Th 0.35 mm
Figure 6. Analytical expression of fabric composite : Lamination

3.2 Static stiffness : Static torsional BIW stiffness
Torsional stiffness was checked for HED9 vehicle as Figure 7. BIW and trimmed body status
models were used. Figure 8 shows the stiffness results. Stiffness is lower than HMC’s Soul
vehicles slightly, however specific stiffness is higher. Characteristics of stiffness and weight of
Intrado were compared with EURO CARBODY entry vehicles. Body efficiency is almost same
level, but stiffness is lower than with other composite vehicles. Because of concept car, detail
design concept of high stiffness was not applied to Intrado. However weight reduction possibility
was checked.
Figure 7. Torsional stiffness FEA model : BIW & trimmed body
Figure 8. Torsional stiffness result : BIW & trimmed body
Weight
Body efficiency =
Torsional stiffness × Projection area of BIW
3.3 Strength : Roof strength

Roof strength is main performance to protect passenger from rollover of vehicle and that is one of
IIHS (Insurance Institute for Highway Safety) TSP (Top Safety Pick) items. Figure 9 shows roof
strength model and results. Similar with stiffness results, strength is less than Soul vehicles, but
SWR (Strength to Weight Ratio : specific strength) is same level with Soul 2008.

Figure 9. Roof strength FEA model & result
In order to express composite material failure during analysis, Hashin failure theory was used.
Fiber/matrix and tension/compression failure modes can be shown at Figure 10.
Figure 10. Composite failure modes
3.4 Durability : Belgian road fatigue

Figure 11 shows fatigue analysis model of Intrado. Because of lack of trim models, almost BIW
status model was used. Figure 12 shows fatigue analysis process. In this study, ADMAS for load
calculation, Abaqus for simulation of vehicle behavior, and fe-safe/Composite for life analysis
were used. A fatigue characteristic of materials was not secured, so properties of similar material
were used in fe-safe/Composite database. Figure 13 shows life contours at front strut mounting
parts per each plies. Lamination angle difference at ②-③ and ⑥-⑦ is large, so delamination is
predicted. However fatigue material properties is not compatible, this study is focused at
understanding composite fatigue characteristics and analyzing methods.

Figure 11. Belgian road fatigue FEA model
Figure 12. Belgian road fatigue FEA model
Figure 13. Fatigue result : Life contour at each ply

4. Conclusion
Analysis procedure for composite vehicle performance was considered such as stiffness, strength
and fatigue. These processes were applied to INTRADO.
1) Effective design is possible using CFRP in the point of stiffness and strength
2) Composite fatigue failure characteristics were studied.
5. References
2. Mechanics of Fibrous Composites, C.T. Herakovich, 1997.
3. fe-safe/Composite Theory Manual.

Simulation of Pipe Bending Process with Abaqus
Satoshi ISHIKAWA1, Yoshihiro ISHIKAWA2
[1]IDAJ Co., Ltd. and [2]Sango Co., Ltd.
Abstract: Exhaust manifolds are generally simple cast iron or stainless steel units which collect
engine exhaust from multiple cylinders and deliver it to the exhaust pipe. The exhaust manifolds of
steel pipe are formed with multiple bending operations. The pipe bending process is extremely
complicated and the forming conditions such as controls of pressure are significant.
This paper presents several Finite elements modeling techniques for simulating and optimizing
pipe bending process using Abaqus/Explicit. Designing and tuning pipe-metal forming tools are
quite complicated and time consuming tasks. These tasks must take into consideration a number of
potential issues, such as the rate of forming, pipe shape smoothness, reduction ratio of metal sheet
thickness, and pipe metal wrinkling. To shorten the design cycle and reduce the number of forming
tool prototypes for the multi-process bending pipe, simulations with Abaqus served as virtual test
grounds to provide valuable insight into the pipe’s complex forming processes. Because of large
deformation and contact interactions, Abaqus nonlinear capabilities were well suited for these
tasks. This paper demonstrates Abaqus forming applications that helped resolve matters arising
from realistic industrial forming design and production processes. Furthermore this paper argues
the difference of three element types which are shell, continuum shell and solid. The analysis
results were compared with the experimental thickness and the performance and accuracy were
examined.
Keywords: Plastic forming, Element formulation, Friction, Explicit.
1. Introduction
The exhaust manifolds of steel pipe are formed with multiple bending operations as shown in
Figure 1. The pipe bending process is complicated and forming conditions such as the control of
load conditions are extremely important matter. For example, a lower axial load induced a
reduction of thickness, while a lower pressure die load caused a wrinkle phenomenon at the
bending corner as shown in Figure 2. To achieve a higher level of design, the practical use of CAE
method is crucial. Therefore, we elaborated an analysis model for simple bending process.
In this paper, single bending process was examined. The analysis model consisted of one steel
pipe deformable body and seven rigid bodies which were machine components. Abaqus/Explicit
was used on this plastic deformation analysis which had complicated loading conditions as same
as reality. In addition element technology was investigated. Several element types were applied on
the same analysis condition, also machining conditions were studied.

All results were compared with the experimental data, and performance and accuracy were
examined. Shell element showed good computation time but the reduction ratio of thickness was
not provided, while the result of Incompatible solid element was as same as the experiment.
It is concluded that our proposal analysis methods can help us design more robust and provide the
Machining condition determining method.
Figure 1. Exhaust manifolds.
Figure 2. 90 degree test.

2. Model Description
2.1 Parts and Assembly
Figure 3 and 4 showed an assemble FEA model which consisted of one deformable pipe and seven
rigid parts. The model was analyzed by half model with symmetry boundary condition. The
Deformable Pipe was simple circular shape and the thickness was 2[mm].
A real mandrel shape was shown in Figure 5. In this study, Mandrel model was made up of one
long cylinder part and two bowls which were shown as red, green and white parts in Figure 3.
These three parts were connected by the Join connecter condition between each reference points.
Contact pair was applied at each area. The coefficient of friction between Roll, Pressure, Wiper
and outside pipe was 0.3, while the coefficient between Mandrel and inside Pipe was 0.1, because,
lubricating oil was set in the pipe inside for smooth deformation.
First step, a rotation velocity condition was applied on the Bending Roll Die and the Clamping Die
parts for bend forming. An axial certain load was applied on the left end of pipe to push the whole
pipe toward X direction, while a historical pressure was put on the part of Pressure Die. These
load conditions were argued below section 2.3 in detail. Also all degrees of freedom of the part of
Wiper were fixed during the analysis. The X degree of freedom of the mandrel was fixed during
first step in bending process.
Second step, after the bending process, the mandrel was withdrawn to minus X direction with a
certain velocity.
Mandrel
r0 r45
r90
Deformable Pipe
Figure 3. Assembly model.

Pressure Die
Clamping Die
Wiper
Bending Roll Die
Figure 4. Assembly model.
Figure 5. Mandrel.
2.2 Material model

Material model of pipe was an elastic plastic behavior with anisotropic yield. The Young’s
modulus was 180[GPa], the Poisson’s ratio was 0.3[-], and the density was 7.699E-6[Kg/mm3].
Since the steel pipe was formed by rolling at first process, we should take into account anisotropic
plastic flow. The r-value characterizes the anisotropic plastic behavior. For the planar anisotropic
plastic, three parameters (r0, r90 and r45) which express the thickness reduction ratio of each
direction are necessary. In our model, r0 meant axial direction and r90 meant circumference, those
were depicted in left side of Figure 3. These three parameters of r-value were experimented by
Sango and the parameters of Potential option were calculated from Equation 1.
Figure 6 showed the stress strain curve, but these real values were confidential. Because of cyclic
loading observed around bending area, nonlinear kinematic hardening was used. This hardening
rule is well known method for sheet forming analysis.

r90 ( r0 + 1) r90 ( r0 + 1) 3 ( r0 + 1) r90
R22 = R33 = R12 =
r0 ( r90 + 1) ( r0 + r90 ) ( 2r45 + 1)( r0 + r90 )
, , (1)
R=
11 R=
13 R=
23 1.0
True Stress
0 0.2 0.4 0.6 0.8 1

Plastic Strain
Figure 6. Stress-strain relation (Half cycle).
2.3 Load conditions

The load conditions are very important for bend forming. First condition, the Bending Roll Die
and Clamping Die were applied an angle of 90 degrees rotation displacement in 0.4 seconds with
certain velocity. Second condition, the axial load for Deformable pipe was applied as shown in
Figure 7(a) toward global X direction. Third condition, the pressure die load for Pressure Die was
applied as shown in Figure 7(b) toward global minus Z direction. The black solid line in Figure
7(a) and (b) showed typical nominal condition, the pipe was formed without any fault in those
conditions. While, the breakage of thickness was observed in red dashed line in Figure 7(a), also,
the wrinkle shown in Figure 2 appeared with the condition of red dashed line in Figure 7(b). To
keep confidentiality, real values of vertical axis were inaccurate also.
Hence this study had three cases regarding load condition. The load cases were summarized in
Table 1.

Table 1. Load case.
Load Axial load Pressure Result of Experiment
Case die load
a Nominal Nominal Good shape
b Lower Nominal Thickness reduction
c Nominal Lower Wrinkle occurrence
Nominal
Fault
Axial Load
Pressure Load
Nominal
Fault
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Process time[sec] Process time[sec]
(a) axial load of Deformable pipe (b) Pressure die load

Figure 7. Load controls.
2.4 Element types

Needless to say, the character of bending behavior of element formulation is important in the
process simulation, and our primary purpose was establishment of excellent estimation of
thickness reduction and criterion of wrinkle occurrence. In this study, therefore, we examined
various element types for the Deformable Pipe as shown in Figure 8. Also Table 2 summarized the
element type, the number of layers of thickness, the number of nodes and elements.
As a consequence, we performed 12 cases of analyses of three load conditions and four element
types.
Table 2. Element type.

Case Element layers Number of Number of Note
type nodes elements
1 S4R 5* 5829 5600 Shell element reduced integration
2 SC8R 1 11256 5400 Continuum shell element
3 C3D8R 4 28140 21600 Solid element reduced integration
4 C3D8I 2 16884 10800 Incompatible solid element
* The layers of shell element meant the number of integration points through the thickness.

(1) Shell Element(S4R) (2) Continuum Shell(SC8R)
(3) Reduced Integration Solid(C3D8R) (4) Incompatible Solid(C3D8I)

Figure 8. Element types and mesh controls of Pipe.
3. Results
3.1 Deformed shape
Figure 9 to 12 show the deformation shape and contour plot. Since element case 1 and 2 were
structural shell element, the contour component was element thickness. While element case 3 and
4 were solid element, therefore, the contour plot showed the strain component of thickness
direction.
Regarding to wrinkle occurrence which was appeared in load case (c), two types of S4R and
C3D8R could not show the virtual wrinkle shape, while another types of SC8R and C3D8I
revealed the wrinkle shape.

All element types could get the thickness reduction on the outer surface at bending area in the load
case (a) and (b). Since the ratio of reduction is very important matter, Table 3 shows the minimum
thickness at outer area on each analysis cases and experiment results on the last line. In case of
Shell element, the ratio of S4R was as same as experiment in load case (a) within 0.14% relative
ratio. But the ratio in load case (b) was 8.41%; it was not useful to estimate about the breakage.
Also it could not reveal the wrinkle shape in load case (c). In the next case of Continuum shell
element, SC8R showed larger thickness than experiment in both load cases (a) and (b), but it
showed excellent wrinkle shape in load case (c). Third case of the reduced integration solid
element, the relative ratio was 6.19% in load case (a) and 4.73% in load case (b). Apparently the
element of C3D8R looked good result, however, it could not get the wrinkle shape in load case (c).
The last case of the incompatible solid element C3D8I, the ratio in load case (a) was not good, but
the ratio 1.46% in load case (b) was the best of all element types. Additionally it could show the
good wrinkle shape in load case (c).
Table 3. Minimum thickness.

Load case
Element type
(a) (b)
S4R 1.426(-0.14%) 1.340(+8.41%)
SC8R 1.486(+4.06%) 1.450(+17.31%)
C3D8R 1.340(-6.19%) 1.295(+4.73%)
C3D8I 1.327(-7.77%) 1.254(+1.46%)
Experiment 1.428 1.236
load case (a) load case(b) load case(c)

Figure 9. Thickness of Element type S4R.

Figure 10. Thickness of Element type SC8R.

Figure 11. Strain of thickness of Element type C3D8R.

Figure 12. Strain of thickness of Element type C3D8I.
3.2 Analysis time

Table 4 shows the analysis time (wallclock time in the Abaqus log file) with parallel processing
under Windows 7 64 bit operating system and a computer having two Intel Xeon E5-2620
(2.10GHz processor six cores). All jobs were analyzed under 12 cores parallel processing with
Abaqus version 6.14-1.

Table 4. Analysis time.
Element type Load case Wall time
[hh:mm]
a 3:53
S4R b 3:51
c 4:15
a 4:52
SC8R b 5:00
c 4:53
a 21:56
C3D8R b 22:48
c 19:37
a 15:57
C3D8I b 16:34
c 14:52
4. Discussion
In general, the shell element is traditional formulation for sheet metal forming, but this element
could not show the large reduction of thickness in load case (b). The percentage relative to
experiment was 8.41% which was insufficient to make a decision over breakage. Besides it could
not reveal the wrinkle shape in load case (c). We assumed that these faults were caused from the
difference of coefficient ratio of friction. Since pipe bending area had double side contact between
the roll die and the mandrel, normal shell element could not handle shear deformation at cross
section.
In the second approach, we tried using the continuum shell element which discretizes an entire
three-dimensional body, unlike conventional shells which discretize a reference surface. This
element is often used in case of double side contact. But this element could not treat the thickness
reduction over 10%. We refer to one paragraph of Abaqus Analysis User’s Guide 29.6.2 below
and underline the important sentence.
“Although continuum shell elements discretize a three-dimensional body, care should be taken to
verify whether the overall deformation sustained by these elements is consistent with their layer-
wise plane stress assumption; that is, the response is bending dominated and no significant
thickness change is observed (i.e., approximately less than 10% thickness change).”
As a result, the model of SC8R could not get enough the ratio of reduction in case (a) and (b),
however, it showed superb wrinkle shape in case (c) for the sake of the advantage of double side
contact.
In the third approach, the normal solid element with reduced integration was examined. In case of
(b), the relative thickness was 4.73% which was not enough to decide the thickness failure in
design process. The decision criteria is the most important in design process, therefore, this
element was not adequate. Furthermore it could not show the wrinkle shape in case (c).
In the last approach, we have investigated the incompatible solid element which can handle the
good bending behavior with few numbers of layers. Although the percentage of case (a) was
7.77%, the most important reduction ratio of case (b) was in 1.46% relative, this precision was the

best of all cases. Moreover the wrinkle shape was arisen from the double side contact with
different friction ratios on both sides of the pipe. But the large computing cost was shortcoming
even in using only two layers as shown in Table 4.
In conclusion, we studied various element types in pipe bending simulation. To acquire the
thickness reduction and the occurrence of wrinkle, we found that the incompatible solid element is
the best way for the pipe bending formulation. However this element cost large amount of
computing resources. Therefore Sango usually analyze with the shell element in normal design,
however, they use the incompatible solid element in case of detail and precise design.
5. References
1. Abaqus Analysis User’s Guide, Version 6.14, Dassault Systèmes Simulia Corp., Providence,
RI.

Automated Weight Optimization Process for
Automotive Wheel Hub Bearing
Jungho AN*, Inha LEE*, Youngmin CHO**, Sanggyu PARK**,
Cheonsoo JANG**, Seungpyo LEE*
* R&D Center, ILJIN Global, 128-5 Samsung-Dong, Kangnam-Gu, Seoul, 135-875, Korea
** SIMULIA of Dassault Systèmes Korea Corp., ASEM Tower, 9F, 517 Yeongdong-Daero,
Gangnam-Gu, Seoul, 135-798, Korea
Abstract: Automotive wheel bearing is an essential component of the vehicle. It transmits engine
power into wheels and supports vehicle weight. In recent years, the demand worldwide for weight
reduction in automotive part in order to improve fuel efficiency has increased. Despite of it,
automakers are requiring that wheel bearing performance in the areas of stiffness, strength, and
life be maintained or improved. In this study, weight optimization for automotive wheel bearing
was performed. Design variables were the bearing geometry and bearing internal specification.
Cost function was bearing weight and constraint conditions were bearing life and stiffness. CATIA
was used to model the geometry and Abaqus was used to carry out the stiffness analysis. To
construct the optimization formulation, the commercial optimization software, Isight was used. As
a result, the optimal model satisfied all of constraint conditions and reduces the weight by
comparing with that of the initial geometry. In conclusion, by using the developed weight
optimization process, the test and development costs of an automotive wheel bearing can be
reduced.

1. Introduction
Steering characteristic is one of the important performances of the vehicle which it relates
comfortable ride during the driving. In response to recently increasing needs for ride comfort and
steering feeling, it has been emerged as design requirements from automobile manufacturers.
Automaker has to provide the light weight vehicle for satisfying customer’s request and
government’s regulation about the fuel consumption. The weight of wheel hub bearing for the
light weight vehicle should be reduced which sustains fatigue life and stiffness of that. The trial
and error approach to generate design model and evaluate their fundamental performances
individually is very time consuming. It’s not easy to generate the optimized design model. In
order to develop the light weight wheel hub bearing, the automated weight optimization process
should be built for evaluating fatigue life and stiffness as well as various design model creation
simultaneously. To verify the reliability and adaptability of the automated weight optimization
process onto the real product development process, the various design models were generated and
evaluate their performance through that process.
2. Procedure of Weight Optimization

In this research, the automated procedure of weight optimization consists of CAD model
creation automation, calculation of fatigue life, FEA(Finite Element Analysis) automation and
weight optimization. ILJIN Global Corporation’s Bearing Wizard Program and CATIA are used
for building up 3D Geometry. The fatigue life was calculated MS Excel based the in-house
program. To develop FEA automation, Abaqus for CATIA is adopted for CAE software because it
gives the same user interface environment as CATIA so that designer can use easily and run
quickly. In the process of weight optimization, Isight controls full process on the 3D Geometry
generation, calculation of fatigue life, FEA for stiffness and measurement of weight by batch
operation. Because Design or CAE engineer just define the design parameter and maximum
allowable job time for the optimization, it very comfortable and intuitive for engineer to generate
the weight optimized design model. Design variables were the bearing geometry and bearing
internal specification. Objective function was bearing weight and constraint conditions were
bearing life and stiffness.
2.1 Automated System for Wheel Bearing Design

Before this research, Design engineer build up wheel hub bearing 3D model manually. Because
of the time consuming and the difference of design model generation by design engineers in
manual, it’s difficult to standardize master drawing. To enhance the conventional design process,
the automated design system was setup in this research which can generate various design model
based on the vehicle specification and wheel bearing internal specification. The developed
automated system will be operated on the MS Windows which consists of the bearing wizard
program (Figure 1) and CATIA program (Figure 2). When users input the vehicle specification and
wheel bearing internal specification onto the bearing wizard program, basic bearing part model

will be generated based on the CATIA environment automatically. The detail model generation
and modification will be available on the CATIA program. Design parameters of generated design
model will be automatically linked into the in-house program for the fatigue life and the
automated FEA system. For the users convenience, users fill in the necessary information step by
step on the automated design system. Also the GUI of this system provides the preview wizard to
users.
Figure 1. GUI of bearing wizard program
Figure 2. GUI of CATIA program

2.2 Automated FEA System for the Wheel Bearing Stiffness
To develop FEA automation, Abaqus for CATIA is adopted for FEA automation. The design

model which was generated from the automated design program will be linked into AFC
environment automatically. The FE model, Load and Boundary Conditions were generated and
defined automatically without the additional operation. The stiffness analysis based on Abaqus
will be ran by batch mode automatically. So design engineer and CAE engineer can easily check
the FEA results. Also variety of Automaker’s FEA conditions will be reflected on the design model
automatically. The minimized FEA running time is necessary to review lot of design cases in
limited time. The hexa elements for critical area and tetra elements for the remains of wheel hub
bearing will be used to considering the accuracy and minimum run time simultaneously.
2.3 Calculation of Fatigue Life based on the in-house code

There are contact forces between ball and rail of wheel hub bearing during the operation. This
contact behavior can cause trouble in indention or lamination at the contact areas. It’s a criterion of
the fatigue failure of wheel hub bearing. In this process, the fatigue life of wheel hub bearing was
calculated by ILJIN Global’s in-house program as shown on Figure 3. Because of all of defined
design information are available on the in-house program automatically in this process, the in-
house program for checking the fatigue life of wheel hub bearing will be ran batch mode
automatically. So design or CAE engineer can check the fatigue life of wheel hub bearing without
any additional operation.
Figure 3. GUI of life calculation program

2.4 Weight Optimization System
In this research, the weight optimization process which are available on the parametric CAD
model creation, calculation of fatigue life, FEA(Finite Element Analysis) automation and weight
optimization simultaneously. This process was composed and controlled on the Isight GUI

environment as in Figure 5. The importing MS Excel file which was generated by the automated
wheel hub bearing design system is just necessary to execute this process full automatically. There
are 3 junctions in this weight optimization process to consider various users need. First process for
weight optimization can evaluate both stiffness and fatigue life simultaneously. Second one for
weight optimization can to generate various design model based on the fatigue life. Last one can
take care of only the automated FEA of the generated FE model without weight optimization. So
this process will provide the minimized user intervention and automatic operation to users.
Figure 4. Isight workflow
Figure 5. GUI of weight optimization program
3. Adoption of weight optimization Process

To verify the adaptability and robustness of the automated weight optimization program, this
process were reviewed with real design model and on the practical wheel hub bearing

development process. 9 design variables were reviewed which are related to the stiffness and
fatigue life. Objective function was bearing weight and constraint conditions were bearing life and
stiffness. As the results of weight optimization process, the Figure 6 shows the reduction of wheel
hub bearing by 3.4%. To check the validation of this automated process, it’s necessary to take side
view of accuracy FEA results in the developed weight optimization process. The stiffness analysis
result with FE model which was generated automatically in the weight optimization program was
good correlated with the stiffness analysis result with FE model in manual.
Figure 6. Optimized shape result
Figure 7. Comparison of analysis result
4. Conclusions
In this research, the automated weight optimization process was developed which is considering
weight reduction, fatigue life and stiffness of wheel hub bearing simultaneously. An optimized

wheel bearing model was generated through this weight optimization process. Weight reduction
rate of the generated weight optimization model is about 3.4%. Design engineer can build-up a
light weight wheel hub bearing based on this weight optimization process automatically. It’s very
useful for reducing development time and cost.
5. Reference
1. Numata, T., “Latest Technical Trends regarding Hub Unit Bearings,” Koyo Engineering
Journal English Edition No. 168E, pp.8~13, 2005.
2. Lee, S.P., Lee, I.H., Kim, Y.C., Kim, H.W., Bae, H.U., and Park, J.Y., “Structural Design and
Analysis of Automotive Wheel Bearing using Parametric Method,” Korean Society of
Mechanical Engineers 2010 Fall conference, pp.574~577, 2010.
3. Lee, I.H., Cho, Y.G., An, J.H., Cho, Y.M., Kim, M.S., Jang, C.S., Lee, Y.H., Lee, S.P.,
“Development of Stiffness Analysis Program for Automotive Wheel Bearing,” KSAE 2012
Annual Conference, pp.533~537, 2012
4. Lee, S.P., Lee, N.Y., Cha, C.Y., Lee, I.H., Lee, S.R., Kim, W.I., and Kim, T.W., “Bearing
Life Optimization for Automotive Wheel Bearing using Design of Experiments ,” KSAE
2012 Fall conference, pp.281~284, 2012

Automated Analysis in CATIA V5
Stephen Wallis
Jaguar Land Rover
The traditional process for assessing the strength, stiffness and modal performance of a chassis
component such as a suspension link, and iterating the design to meet targets in the optimum way,
is time consuming. Some design engineers experimented with using integrated analysis in CATIA
V5 to assess and iterate the design more efficiently. Success was limited due to the complexity of
setup and poor correlation with the results obtained by CAE specialists using the standardised
process.
This paper outlines how a combination of CATIA V5 analysis and scripting was used to solve
these problems for a number of component types. The success of the tools released to date and the
advantages of using them will be discussed. Guidelines for identifying processes which are
suitable for automation will be presented
1. Introduction
The evolution of virtual engineering has led to the design engineer and the analyst being separate,
specialist roles. This means that each time an analysis task is required, a CAD model is passed
from the designer to the analyst, whose work priorities may mean the task might be delayed in
starting. The processing time required is also significant. The impact of the high total time taken to
perform one analysis is that the design and analysis iterative loops may be stopped as soon as the
component meets the minimum requirements, rather than continuing until it is optimised with all
unnecessary mass removed. This is what is meant by the “traditional process” for analysis of a
chassis component.
1.1 Problem Summary

The problems experienced through using the “traditional process” for analysis of Chassis
components can be summarised as follows:
1. Designs may be completed late
2. Designs may not be fully optimised
3. Analysis expert’s time is limited
4. When design engineers tried to use CATIA V5 Analysis, their results might not
correlate with those of experts.
5. Standard analysis processes are repetitive: not interesting for experts to complete!

2. Solution
Design engineers experimented with using integrated analysis in CATIA V5 to assess and iterate
the design more efficiently. However success was limited due to the complexity of setup and poor
correlation with the results obtained by CAE specialists using the standardised process. Analysis
experts were able to produce analysis templates that gave results that did correlate to those from
their standardised process. Hence a proposal was made to automate the application of these
templates to the designer’s components, along with the reporting of results, to allow the designers
to complete the analysis process themselves. A tool was created to fulfill the proposal using
scripting. An overview of the process is shown in figure 1, with more detail given in the rest of
this section.
Figure 1. Process Summary Diagram

2.1 Template Selection
A graphical user interface (GUI) was created to guide the workflow. With the component design
in-session the designer launches the GUI though a toolbar button. Through making selections on
the first page of the GUI the user selects the correct template for their component (see figure 2).
2.2 Geometry Selection

The user makes selections of geometry on their component, to facilitate linking the analysis
template to their component. The geometry is then published. If the component is saved with the
created publications next time the user does not have to re-select them.
2.3 Link User Component to Analysis Template

The user links their component to the analysis template
2.4 Run Analysis

The user hits the run button and an estimated run time is given. The CATIA V5 session cannot be
used while the analysis is in progress. On completion the results are given in a quick results
window (see figure 3).
2.5 Create Report

Should the user desire a comprehensive report this can be created (see figure 4). Targets can be
input for automatic comparison with the results.
2.6 Re-Analysis
The user has the option to keep the analysis tool open while they change their design. After the
changes are made the links between the model and template are updated and the component can be
re-analysed without setting up the template again.

Figure 2. Template Selection in GUI

Figure 3: Quick Results Window
Figure 4. Typical Report

3. Reflecting on the Success of the Tools
To date five automated analysis tools have been created in the CATIA V5 environment, two of
those being extensions of previous tools. Good feedback was consistently received from the users,
who typically asked for the tool to be extended to contain other analysis processes or onto other
components! The benefits of having these tools have far outweighed what was previously
envisaged, and has in fact changed the way of working in Jaguar Land Rover Chassis. With the
analysis templates linked to the CAD model the efficiency of the analysis-design change loops has
been maximised.
3.1 Benefits Summary

1. Efficiency: The reduction in processing time for one analysis iteration of comparable
level is over 90%. This has supported an approximate 80% reduction in the mechanical
design lead time for a front knuckle, when developed by an experienced CAD designer.
See Appendices 6.1 and 6.2 for usage and time saving statistics.
2. Greater Optimisation: Because of the processing time reduction a greater number of
design/analysis iterations can be completed within the same overall time period. Hence in
a limited time period a greater level of optimisation might be reached than before.
3. Repeatability: Analysis in the design environment will now give the same result
regardless of the designer.
4. Bring Analysis Back In-House: Because of the efficiency gains it has become possible
to refine wheel designs to a greater level in house, reducing the number of analysis
iterations done by the supplier. This has saved money and taken 2-3 weeks out the wheel
development process.
3.2 Benefits in Terms of the “7 Wastes of Lean”

The 7 wastes are at the root of all unprofitable activity within your organisation.
1. Defects: Eliminated through repeatability
2. Overproduction: Not applicable.
3. Transportation: Not applicable
4. Waiting: Eliminated through analysis on demand
5. Inventory: Not applicable
6. Motion: Time spent making analysis requests and moving data is eliminated
7. Processing: Processing time significantly reduced

4. The Future
In the near future Jaguar Land Rover will be migrating to CATIA V6 in the 3D Experience
platform. This will open up more possibilities for analysis automation with increased capability,
for example non-linear analysis, and further process time reduction, perhaps through batch
submission of analysis tasks. We aim to continue to identify processes suitable for automation and
create automated tools using a standardised, modular approach which re-uses the experience
gained in producing the tools completed to date.
4.1 What Makes a Good Process for Automation?

A summary of the main points to consider when deciding whether a process is suitable for
automation is as follows:
1. Is the process standardised?

2. Are the designs to be analysed consistent enough such that they could fit a standardised
template?
3. Is the analysis process possible in the CATIA V5 toolset?
4. Is there a business case? I.e. how many times will each template be used?
5. How time consuming is the existing process?
6. How much potential is there to remove mass from the component through manual
iteration, which could be enabled through reducing the processing time?
7. Would an automated tool allow analysis to be done in-house rather than outsourced?
8. Is resource available to create the tool?

5. References
1. http://www.isixsigma.com/dictionary/7-wastes-of-lean/
6. Appendix
6.1 Usage Statistics
Tool Details Usage Statistics
Tool Level of Number of Release Date Total Usage Per

Analysis Templates Usage Month
Engine and Stiffness and 14 18/08/2014 16 16

Gearbox strength: not
bracket sign off
Wheel 2 Stiffness: sign 1 17/01/2014 197 25

off, strength:
not sign off
Suspension Stiffness and 22 25/01/2013 1502 75

link and modal: sign
knuckle 2 off
Wheel 1 Stiffness: sign 1 01/11/2011 Not tracked Not tracked

off
Note: Usage stats as of 22/9/14

6.2 Time Savings
Before: Approximate After: Approximate Processing time

Tool Details Processing Time Processing Time reduction for one
(hours) (hours) iteration (%)
Engine and
16 0.5 97%
Gearbox bracket
Wheel 2 16 1 94%
Suspension link
8 0.5 94%
and knuckle
Wheel 1 12 0.5 96%
Note: Timings quoted are approximate

Simulation of Fully Coupled Thermo-Mechanical
Effects in a Disc Brake Rotor
G.KARUNAKAR *, J. OSCROFT *, R.TAULBUT* AND D.STEWARD *
Email Id:
Ganesh Karunakar,gkarunak@jaguarlandrover.com, Julian Oscroft,joscroft@jaguarlandrover.com

Richard Taulbut, rtaulbut@jaguarlandrover.com, Daren Steward, dstewa24@jaguarlandrover.com
* Braking Development Team, Jaguar Land Rover Limited, United Kingdom.
Abstract: The frictional heat generated during braking of a vehicle can cause numerous negative
effects on the brake system, such as brake fade, premature wear, and thermal cracks causing the
brake to have inherent noise and vibration issues under braking, called rumble. An opportunity
therefore exists to eliminate brake rumble from the vehicle, thus providing a quality improvement
opportunity.
In this paper, a fully- coupled, temperature- displacement analysis using Abaqus/Standard was
carried out to take into the account both thermal and structural effects on the brake rotor. The
effect on thermal performance of rotational speed (corresponding to 𝑉𝑚𝑎𝑥 braking conditions) and
rotor temperature, due to the rotating heat source, has been evaluated. Disc thickness variations
are shown to arise from the temperature profile across the disc cheek surface.
The scenario simulated considers the thermal effect of braking a vehicle which was prone to
generating an audible rumble noise. The simulation measured the thermal and geometric changes
in the disc, demonstrating the disc thickness variation and thermal banding of the rotor, and the
location of these on the disc surface due to the effect of the thermo-mechanical loading.
Using this simulation technique, design changes to the pads to eliminate the rumble by moving the
hottest band of the disc to a more central location, were assessed and verified on a vehicle.
Future developments will include further investigation of the sensitivity of the rumble noise to pad
friction material compressibility and profile and to the internal vane pattern of the disc.
Keywords: Fully-coupled simulation, Brake System, Brake disc design, Heat power distribution,
Braking, Pad/disc system, Frictional heating, Moving heat source, Heat conduction, Finite
element method, DTV( disc thickness variation).

1. Introduction
In the current automotive world there is a high standard of customer expectation regarding brake
noise. This is having a large impact on the technical design and virtual development of braking
components and systems. This leads to engineers having a greater focus on the product
development in terms of testing, design and analysis to satisfy these customer requirements.
The main benefit in using simulation is to be able to assess the large numbers of design iterations
often required to achieve a good design in a cost effective manner. This helps design teams to
perform the design changes and analyses efficiently by changing the designs and meeting the
structural targets early during the design phase minimizing the effects later in the delivery of a
vehicle program.
These restrictive demands, in terms of the NVH and thermal performance, push the vehicle
manufacturers to find ever more innovative technologies and hence new design options to increase
the refinement of the car whilst working towards cost and weight reductions of those systems and
in overall reduction in the weight of the car.
During the virtual product development process the key technical targets are fatigue, thermal,
brake pedal feel, service life, brake cooling, NVH e.g. brake squeal, rumble etc.
The critical aspects of the brake design are in meeting all these performance requirements and
often there is conflict in meeting them.
Currently dynamometer and vehicle tests are carried to evaluate the NVH and thermal
performance physically. This will determine the Disc Thickness Variations (DTV) and
temperature of the disc surface under the varying contact pressures between the pad and disc. The
non-uniform contact pressure also leads to non-symmetric temperature profile on the disc cheek
surface.
Virtual simulation can play a significant role in the identification of the non- symmetric
temperature profile on the disc cheek surface leading to development of efficient brakes that meet
the performance and NVH targets, whilst reducing the need for testing of physical prototypes.
This paper mainly focuses on the noise resulting from the thermal effects on a disc during braking.
2. Thermo-mechanical simulations.
Thermal simulation of the problem defined above can be managed by uncoupled and coupled
simulations in Abaqus, where separate and simultaneous calculations are made for the
temperatures and mechanical loading respectively.
In order to consider the simulation of the above case, Abaqus/Standard Coupled Temperature-
Displacement Analysis is used to capture the effect of the structural performance under the

thermal and mechanical loadings in braking simulation. The advantage of the above simulation
procedure is that the thermal contours developed take into consideration the caliper effects, pad
geometry effects, and contact pressure effects that will vary during the braking. See figure 1.
Figure 1. Thermal Contour of a disc surface.
3. Thermo-mechanical CAE process.

A detailed CAE process has been developed to capture the effects due to thermal and mechanical
loading due to the high-speed braking. This process helps in predicting temperatures, thermal hot
band location, disc thickness variations ( DTV ), brake torque variations ( BTV ) , etc. which are
contributing factors in the generation of unwanted brake noise and vibrations.
Why we need to use this “Analysis” procedure?
• Because the mechanical and thermal solutions affect each other strongly and, therefore,
must be obtained simultaneously.
• For prediction of the temperature profile of the disc surface due to the varying contact
pressure between the disc and pad during braking.
• Taking into account both thermal and structural effects leads to a more representative
finite element model in which the geometric distortion of the brake system, as a
consequence of the heat generation, can be predicted throughout the full braking event.
• Capture the effect of brake caliper stiffness on the brake disc and pad positions or
movements during the high-speed braking.
• To study the effect of the “consequence of friction and imposed rotational motion”.
• To study the effect of ‘rotating heat source’.
• To explore the behavior of the temperature dependence of material properties.
• To predict the transient thermal and structural behavior of the brake system.

4. Brake System Finite Element Model.
A braking event under a particular deceleration is simulated by rotating the disc at an appropriate
angular velocity and applying the braking pressure through the brake fluid. This brake simulation
is done by simulating the brake dynamometer boundary conditions e.g. pressure, speed and
deceleration. The model includes all contacts between the relevant components e.g. between the
pad and disc, caliper and piston, anchor and pad abutment.
The pad and disc are defined as two deformable bodies, and sliding friction between them is
proportional to the normal force on each element. The friction coefficient is made further
dependent upon the surface temperature.
The brake disc, pad and pistons are modeled with C3D8T, C3D6T brick elements, caliper and
anchor with higher order C3D10MT tetra elements.
Figure 2. Finite element model of a brake assembly used for thermal simulation.
Anchor
Caliper
Disc
Piston
Figure 3. Finite element model of a brake assembly showing the cross section.

5. Results discussion.
A finite element model was used to conduct a study to minimize the brake DTV by simulating the
braking event under deceleration with two differing designs. The results of the braking simulation
are presented below showing the disc temperatures and the disc thickness variations (DTV).
It was observed that good combinations of the disc and pad design have a great influence in
minimizing the temperatures in the hot band and at the same time the disc distortions.
DESIGN A DESIGN B
Figure 4. Contour showing the disc temperature ( Outboard face.)
DESIGN B
DESIGN A
Figure 5. Contour showing the disc temperature ( Inboard face.)

DESIGN A DESIGN B
DESIGN B DESIGN A
Figure.6 Temperature profile across the cheek surface. Figure.7 Contour showing the temperature’s across the disc.
DESIGN A
Disc thickness variation
DESIGN B
Perimeter length of a disc.

Figure 8. Contour showing the disc thickness variation.
Vibration measurements made in the vehicle at the seat rail and at the steering wheel under the
two designs highlighted showed that Design B made contributions in reducing the vibration
experienced due to the thermal distortion of the disc.

6. Conclusion and future work.
In this paper, the braking event is simulated by using the Abaqus/Standard Coupled temperature-
displacement analysis. This methodology captures the mechanical and thermal behavior of a disc
during braking under varying contact pressures. Caliper stiffness affects the disc temperatures and
distortion; however, effects due to geometrical change in the disc and pad have a greater influence.
This process will give a better understanding of the disc and pad behavior under different braking
events during the design process. This method has great potential for disc design in terms of
temperatures and thermal fatigue.
Further study of disc thermal fatigue and distortion of different brake assemblies in various
braking scenarios and under variation of pad compressibility with respect to temperature, using
multi-physics simulation analysis incorporating brake cooling using CFD is recommended.

Simulation of Combined Forming and
Injection Molding Processes
Dipl.-Ing. Timo Mueller, Prof. Dr.-Ing. Frank Henning
Karlsruhe Institute of Technology, Lightweight Technology
Abstract: Injection and compression molding of fiber reinforced thermoplastic materials are
proven and efficient processes to produce lightweight parts for automotive applications. By
adding continuous fiber reinforced composites or sheet metal, the possible range of applications
can be greatly increased while retaining the advantages of the molding processes, cost efficiency
and design flexibility. So called hybrids parts are used for more and more applications, especially
in the automotive industry. The reinforcements are formed by the closing motion of the mold and
the flow of the molding material. A novel approach is suggested using Abaqus/Explicit coupled
eulerian lagrangian simulation to model combined forming and molding processes. All existing
material models and material data for forming simulation available in Abaqus/Explicit can be
used. The molding material is simulated using VUEOS und VUVISCOSITY, enabling the use of
material data available for commercial mold filling simulation tools. All thermoplastic molding
processes, including injection molding, injection compression and compression molding can be
simulated. Three examples are used to demonstrate the application to simple injection molding, a
combination of injection molding and metal forming and a combination of compression molding
and organic sheet forming.
Keywords: Injection Molding, Compression Molding, In-Mold Forming, Overmolding, Metal

Forming, Preimpregnated Sheet Forming, Organic Sheet Forming, Coupled Eulerian Lagrangian
Simulation, ABAQUS CEL, Folgar-Tucker, Fiber Orientation, Tait Equation, pvT model,
ABAQUS VUEOS, ABAQUS VUVISCOSITY, Hybrid Parts, Automotive Lightweight Technology,
Long Fiber Reinforced Thermoplastics

1. Introduction
Injection and compression molding of fiber reinforced thermoplastic materials are proven and
efficient processes to produce lightweight parts for automotive applications. By adding continuous
fiber reinforced composites or sheet metal, the possible range of applications can be greatly
increased while retaining the advantages of the molding processes, cost efficiency and design
flexibility. So called hybrids parts are used for more and more applications, especially in the
automotive industry. The reinforcements are formed by the closing motion of the mold and the
flow of the molding material.
Figure 1: Overmolded hybrid Part. Source: Engel Austria
Specialized software is used to simulate injection molding, but to date there is no solution to
simulate the combination of forming and molding. ABAQUS Explicit is often used for sheet metal
and composites forming simulation, but not for injection molding simulation.
For practical purposes, the forming and mold filling stages are simulated separate from each other.
While injection molding simulation software allows simulating anisotropic or deformable inserts,
the available material models are very limited and not suited for a complete forming simulation.
A Solution which combines the best of both worlds would be a great benefit for the development
of hybrid fiber reinforced parts.
2. Abaqus CEL
For most applications ABAQUS\Explicit is used with a Lagrangian formulation. In a traditional

Lagrangian analysis nodes are fixed within the material, and elements deform as the material
deforms. Lagrangian elements are always 100% full of a single material, so the material boundary
coincides with an element boundary. Many material models for different applications exist in
Abaqus/Explicit, or have been implemented using VUMAT.

But simulating process like injection or compression molding, where very large deformations
occur, is not possible, because the elements would deform to a point where they would not be able
to function properly. Injection molding software packages use an Eulerian Formulation, where
material flows through elements which are fixed in space. This is better suited for injection
molding simulation, although problems arise if the mesh needs to change to account for a
deforming insert or a changing mold cavity, for example in compression molding. The problem is
to combine both kinds of simulation using different software packages.
Abaqus/Explicit offers what is called coupled eulerian lagrangian (CEL) analysis, which has been
used for a wide range of specialized applications, but is less well known than the conventional
functions of Abaqus/Explicit. CEL offers the capability to combine materials in conventional
lagrangian formulation and eulerian formulation in one simulation, interacting trough
Abaqus/Explicit contact. The lagrangian part of the simulation is set up in the same way as a
normal Abaqus/Explicit simulation.
The eulerian part consists of a mesh defining the control volume in which the material can move
and deform. Only 8-node hexahedra are available for eulerain meshes, one normal and one
thermally coupled element. The eulerian mesh and lagrangian mesh can overlap, they will not
interact with each other. Each eulerian element can either be empty, completely filled with
material or anyway in between. This is tracked by its eulerian volume fraction (EVF). If a material
is completely filled by one or more materials, its EVF is 1.0, if it is completely empty, its EVF is
0.0. Empty space is filled with what is called “void” material, which has neither mass not strength.
An element with EVF 0.0 is completely filled with void material. To define the position of
material inside the eulerian mesh at the beginning of the simulation, the EVF of the elements
containing material is set to the appropriate value. Empty elements do not need to be defined, as
they are by default filled with void material. Abaqus/CAE offers the volume fraction tool to
compute the EVF of the eulerian mesh based on a geometric model of the material distribution.
Based on the EVF of each element, Abaqus computes the boundary of the fluid in each time
increment. The surface reconstruction algorithm approximates the material boundaries within an
element as simple planar faces, taking the EVF of neighboring elements into account. The surface
may be discontinuous, and a fine mesh is required to get an accurate description of the materials
position. Also, displaying the results in Abaqus/CAE can be a challenge.
Boundary Conditions for eulerian meshes are the same as in a conventional analysis, but some
differences exist. Displacement boundary conditions cannot be used on eulerian meshes. Material
that flows over the mesh boundary is lost. To stop material from flowing out of the mesh, the
velocity boundary condition normal to the mesh surface is set to zero. This type of boundary
definition can also be used to define a forced in- or outflow of material by setting a non-zero
value. If material flows into the control volume, it will have the same state and material content as
the material that exist within the element. If the velocity boundary conditions in-plane to the mesh
boundary are set to zero, the fluid will stick to the boundary. There are some specialized eulerian
boundary conditions to control in- and outflow of void and material.
Multiple eulerian materials interact with each other with a sticky behavior. It occurs by default.
Much more possibilities exist for the contact between an eulerian and a lagrangian material. By
default, the contact between these two is frictionless and cannot transmit tensile stresses. The
eulerian surface can be used to define specific interactions with lagrangian surfaces.
Because the contact is based on the eulerian surface, the eulerian mesh does not need to conform
to the shape of the lagrangian surface. It is always possible to use a simple, rectangular mesh for

the eulerian domain. Thermal interactions are also possible, although gap radiation and gap
conductance as a function of clearance cannot be used.
All Abaqus/Explicit material models can be used for the eulerian domain, but for mold filling
simulation, the equation of state materials in combination with a viscosity model are the best
suited ones. Abaqus also offers the possibility for user defined VUEOS and VUVISCOSITY.
3. Injection Molding Simulation
Injection molding simulation for thermoplastics has been used for long time, well proven models
for injection molding materials exist, and a large amount of test data for numerous materials is
available. For thermoplastic materials, the cross viscosity model is often used in a combination
with a time-temperature shift function based on the work of Williams, Landel and Ferry
(Williams, 1955).
1
𝜂 = 𝜂𝑖𝑛𝑓 + �𝜂0 − 𝜂𝑖𝑛𝑓 � 𝜂 ∗𝛾̇ 1−𝑛 (1)
1+� 0∗ �
𝜏
The melt viscosity 𝜂 is defined as a function of the infinite shear viscosity 𝜂𝑖𝑛𝑓 which defines a
lower limit for the viscosity and the zero shear viscosity 𝜂0 , which defines the upper limit the
viscosity approaches for low shear rates. The stress level at which the transition between
Newtonian and shear thinning behavior occurs is 𝜏 ∗ . The behavior in the shear thinning area is
defined by the shear rate 𝛾̇ and the power law index 𝑛. The Williams-Landel-Ferry equation
changes the value for the zero shear viscosity depending on the temperature of the material. The
formula uses three fitting coefficients 𝐴1 , 𝐴2 and 𝐷1 , the melt temperature 𝑇 and a reference
temperature 𝑇 ∗ , which is typically set to the glass transition temperature of the material.
𝐴1 ∗(𝑇−𝑇 ∗ )
𝜂0 = 𝐷1 ∗ exp �− � (2)
𝐴2 +(𝑇−𝑇 ∗ )
𝐴2 and 𝑇 ∗ can be further described as being dependent on pressure 𝑝 and additional fitting
coefficients 𝐴3 , 𝐷2 and 𝐷3 .
𝐴2 = 𝐴3 + 𝐷3 ∗ 𝑝 (3)
𝑇 ∗ = 𝐷2 + 𝐷3 ∗ 𝑝 (4)
The Cross-WLF viscosity model is available in Abaqus/Explicit. To incorporate the pressure

influence on the glass transition temperature, it was implemented using VUVISCOSITY. The
model and parameter sets typically available only describe the viscosity above the glass transition
temperature of the model. The viscosity at glass transition temperature was therefore set as a limit.
In conventional mold filling simulations using finite volume approaches, the infinite shear
viscosity is often set to zero, but setting it to low, non-zero values often helps in improving
simulation performance. Figure 2 shows the viscosity of a typical thermoplastic molding material.

0.1
0.01
viscosity (MPa*s)
0.001 T=500 K
0.0001 T=450 K
T=400 K
0.00001
T=350 K
0.000001
1.00E+00 1.00E+01 1.00E+02 1.00E+03 1.00E+04 1.00E+05 1.00E+06 1.00E+07 1.00E+08
shear rate (1/s)
Figure 2: Viscosity for different temperatures at reference pressure
Different Equations of state have been proposed for thermoplastic materials. For semi-crystalline
polymers, a two domain formulation of the Tait-Equation is often used. The model is based on the
works of Tait (Tait, 1888), although not in the form originally proposed (Hayward, 1967). The
basic equation calculates the specific volume of the material for a given temperature and pressure
using the specific volume at reference pressure, 𝑣0 , pressure 𝑝, the constant 𝐶 with a value of
0.0894 and the function 𝐵(𝑇) describing the pressure sensitivity of the material.
𝑝
𝑣(𝑇, 𝑝) = 𝑣0 (𝑇) �1 − 𝐶𝑙𝑛 �1 + �� (5)
𝐵(𝑇)
For injection molding, two different sets of parameters are used above and below the
crystallization temperature.. An additional term 𝑣1 (𝑇, 𝑝) is added to adapt the formula to volume
changes due to crystallization, some authors attribute this term to Schmidt (Schmidt, 1986).
𝑝
𝑣(𝑇, 𝑝) = 𝑣0 (𝑇) �1 − 𝐶𝑙𝑛 �1 + �� + 𝑣1 (𝑇, 𝑝) (6)
𝐵(𝑇)
𝑇 ∗ = 𝑏5 + 𝑏6 ∗ 𝑝 (7)
The glass transition temperature 𝑇 ∗ is defined by a linear equation similar to one used for the WLF
equation, depending on the pressure 𝑝 and two fitting parameter 𝑏5 and 𝑏6 . For temperatures
below the glass transition temperature the subscript s for solid is used, above the subscript m for
melt is used.
𝑇 < 𝑇∗
𝑣0 = 𝑏1𝑠 + 𝑏2𝑠 ∗ (𝑇 − 𝑏5 ) (8)
𝐵(𝑇) = 𝑏3𝑠 ∗ exp(−𝑏4𝑠 ∗ (𝑇 − 𝑏5 )) (9)
𝑉1𝑠 = 𝑏7 ∗ exp(𝑏8 ∗ (𝑇 − 𝑏5 ) − 𝑏9 ∗ 𝑝) (10)
𝑇 ≥ 𝑇∗

𝑣0 = 𝑏1𝑚 + 𝑏2𝑚 ∗ (𝑇 − 𝑏5 ) (11)
𝐵(𝑇) = 𝑏3𝑚 ∗ exp(−𝑏4𝑚 ∗ (𝑇 − 𝑏5 )) (12)
𝑣1𝑚 = 0 (13)
The coefficients 𝑏1 , 𝑏2 , 𝑏3 , 𝑏4 , 𝑏5 , 𝑏6 , 𝑏7 , 𝑏8 and 𝑏9 are used to fit the model to the material
behavior.
ABAQUS/Explicit does not offer this model, although for many applications a tabular form can be
used. A different approach is to use the built-in functionality for user defined equations of state,
VUEOS. VUEOS calculates the pressure and internal energy based on volumetric strain or density
and temperature, making a rearrangement of formula 5 necessary. The equation with the
crystallization term cannot be inverted, so the added term will be neglected.
𝜌
𝑝(𝜌, 𝑇) = 𝑏3 ∗ exp(𝑏4 ∗ (𝑇 − 𝑏5 )) ∗ �exp( (𝑏1 + 𝑏2 ∗ (𝑇 − 𝑏5 )) − 1� (14)
𝐶
𝑑𝑝 (𝑏1 +𝑏2 ∗(𝑇−𝑏5 ))

= 𝑏3 ∗ exp(𝑏4 ∗ (𝑇 − 𝑏5 )) � � (15)
𝑑𝜌 𝐶
𝑑𝑝 −𝑏4 𝜌
= ∗ 𝑏3 ∗ exp�𝑏4 ∗ (𝑇 − 𝑏5 )� (exp( �𝑏1 + 𝑏2 ∗ (𝑇 − 𝑏5 )� − 1) (16)
𝑑𝐸 𝑐𝑤 𝐶
Specific heat and thermal conductivity of thermoplastics are usually temperature dependent. This
can be simulated using available ABAQUS/Explicit functionality.
4. Injection molding example

A simple model based on DIN EN ISO 294 will serve as a test case for mold filling simulation.
Figure 3: Mold geometry based on DIN EN ISO 294
The part was meshed using C3D8RT elements. Only 8-node hexahedra are available for the
eulerian mesh, imposing restrictions on the geometries that can be meshed.

This is generally not a problem, because it is advantageous to model the mold as lagrangian or
rigid body and use a primitive geometry overlapping the mold boundaries for the eulerian mesh.
This approach offers more options for mechanical contact and heat transfer. For this model
though, just the eulerian mesh was created, and boundary conditions were used to prevent material
from leaving the mold, and to simulate the sticking of the molten material to the cavity walls. Heat
transfer from the hot polymer melt to the colder mold walls is not simulated, because of the short
duration of the actual filling process.
A constant flow rate was prescribed as boundary condition for the process, to achieve a filling
time of 0.5 s. A velocity boundary condition using a smooth step function was defined on the gate
cross section, leading to a constant inflow of material. If material flows into the eulerian mesh
crossing the mesh boundary, ABAQUS/Explicit assumes that the material flowing into the
elements will have the same state and the volume content as the material that is already present in
the elements, so an initial charge was defined at the gate. A temperature boundary condition was
also specified on the gate cross section, so all material enters with melt temperature.
Figure 4: Flow front during filling.

5. Metal Forming
Metallic inserts in injection molded parts are widely used for different applications. In most cases,
no large deformations of the metal occur, and the process can be simulated using conventional
injection molding software. Using the injection molding pressure to form metallic inserts offers
advantages for connecting the metallic and plastic parts (Al-Sheyyab, 2008), or to create
geometries that cannot be produced using conventional metal stamping tools.
The forming of a cup from stainless steel using a deep drawing process is used as an example. A
circular steel blank is formed into a cup similar to experiments used to characterize the
deformation behavior of steel. Instead of a deep drawing press, an injection molding machine is
used to clamp the steel blank, and instead of a solid punch material is injected onto the blank to
deform it.
Figure 5: Stainless steel cup formed using injection molding
The material data for the stainless steel, type 316, were taken from Hussaini (Hussaini, 2014).
Temperature dependent plasticity up to 300° C was modeled using available ABAQUS/Explicit
models. The model consists of four parts, the eulerian mesh, the lower part of the mold called the
die, the blank holder and the steel blank.
eulerian mesh
blank holder
blank
die
Figure 6: Components of the model.

The blank holder clamps the sheet metal to prevent it from buckling while allowing it to move in
plane during the drawing process. The clamping force is applied prior to injection. Different
interaction properties are used for the contact pairs. A velocity boundary condition is specified on
the gate surface to make material flow into the mold cavity. As soon as the free volume inside the
mold is filled with material, pressure builds up and the material starts to deform the steel blank.
Figure 7: Eulerian material and deformed steel blank.

After the injection, the steel blank is formed into a cup similar to the experiments. Mass scaling
can be used to speed up the simulation, because the high stiffness of the steel tends to define the
minimal time increment.
Figure 8: Deformed steel blank after injection molding.

6. Preimpregnated Sheet Forming and Compression Molding
Preimpregnated sheets are made of glass or carbon fiber fabrics impregnated with a thermoplastic
matrix. Forming and overmolding organic sheets in allows to produce light yet economic parts.
The continuous fiber reinforcements provide the load carrying ability of a such a part, while
overmolded features like ribs, bosses or gussets add additional functionality. By using Abaqus
CEL in combination with the Abaqus fabric material model, forming and overmolding of organic
sheets can be simulated. The modeling of overmolding processes is not limited to injection
molding, all kinds of processes using thermoplastic materials can be simulated, including injection
compression molding and compression molding. As an example, the model of a beam is
simulated. The beam consists of a single layer of organic sheet, 1 mm thick, made of a balanced
plain weave fabric and a polypropylene matrix, and a layer of polypropylene with long glass
fibers. A rib reinforces the beam in the middle. Compression molding will be used for
overmolding. Only half the beam is simulated to reduce the computational cost.
The model consists of five parts. Upper and lower mold surfaces are modeled as rigid bodies, the
lower one is fixed, and the upper one is moving. The organic sheet is modeled using 3D-shell
elements and the Abaqus material model for fabrics. The modeling details of the organic sheet are
not part of this work, but the main characteristics, large shear deformation with locking angle,
rotation of warp and weft during shear, strong temperature dependence, and different behavior for
tension, compression and bending are simulated. Multiple sheets would also be possible. The
fourth part defines the initial charge, where the overmolding material is positioned at the
beginning of the compression molding process. The fifth part is defining the eulerian domain.
Upper mold surface
organic sheet
Initial charge
lower mold surface
Figure 9: Constituents of the model. The eulerian domain is not depicted.

The organic sheet is pretensioned in the weft direction to keep it from wrinkling during the
process. The volume fraction tool is used to define the initial volume fraction of the eulerain
elements in the area of the initial charge. The initial charge is at melt temperature at the beginning
of the process, and the organic sheet is also heated to forming temperature. Both mold surfaces are
set to a constant temperature. A boundary condition is applied to the sides of the eulerian mesh to
prevent material from flowing out of the tool. The speed of the upper mold is slowing down the
closer it comes to the final part thickness. The contact between organic sheet and eulerian material
is established before the upper mold surface comes into contact with the initial charge.
Figure 10: Organic sheet with material orientations and Isosurface representation of the eulerian
material in the earlier stages of the compression.
The eulerian material sticks to the organic sheet. As soon as the upper mold surface starts to
compress the eulerian material directly, the rate of flow increases because more material is
displaced. The material flows along the cavity faster than into the rib, as can be seen in Figure 11.
This behavior is often observed in experiments. Conventional mold filling software tends to
predict a much faster filling of ribs.

Figure 11: Organic sheet with material orientations and Isosurface representation of the eulerian
material near the end of the compression.
7. Conclusion & Outlook
Abaqus/Explicit offers a unique possibility for the process simulation of hybrid parts made from
thermoplastic molding material and steel or continuous fiber reinforcements. The main advantages
are that available material models for forming simulations can be used, and the flexibility to model
many different processes like injection or compression molding. A disadvantage is the high
computational cost of the simulation.
To improve the performance of mold filling simulation, two features available in Abaqus could be
used. Depending on the process, large areas of the eulerian mesh might not be used the whole
time. Especially for combined compression molding and forming processes, the eulerian mesh
needs to be large enough to incorporate the initial charge and one or more organic sheets, while
the elements must small enough to allow for a correct simulation when the tool is close to its final
position. A much more efficient solution would be to use eulerian mesh motion, where the mesh
moves and deforms to follow the motion of the material inside.

Figure 12: Eulerian mesh motion. The mesh is deformed to enclose the eulerian material (blue).
The other feature is adaptive mesh refinement for the eulerian domain. Abaqus refines elements
with sharp gradients of density or pressure, or elements in contact with lagrangian bodys, also
leading to a more efficient solution, because the initial element size can be defined coarser without
losing accuracy.
The most important area for improvement is the implementation of a solution so simulate the
orientation of fibers in the thermoplastic material, for example the Folgar-Tucker model or the
ARD-RSC model for long fiber reinforced materials.
8. References
Abaqus Users Manual, Version 6.14-2, Dassault Systémes Simulia Corp., Providence, RI.
Al-Sheyyab: “Light-Weight Hybrid Structures - Process Integration and Optimized Performance”,
Institute of Polymer Technology (LKT), 2008
Hayward, A. T. J., “Compressibility equations for liquids: a comparative study “, British Journal
of Applied Physics, Volume 18, Number 7, 1967
Hussaini, S. M., Singh, S. K., Gupta, A. K., “Experimental and numerical Investigation of
Formability for Austenitic Stainless Steel 316 at elevated Temperatures”, Journal of Materials
Research and Technology, Volume 3, Issue 1, January–March 2014
Schmidt, T.W., Menges, G., “Calculation of the Packing Phase in Injection Mmolding with a two-
layer segment Model”, Proceedings of the 44th ANTEC, SPE, 1986
Tait, P. G., “The Voyage of H.M.S. Challenger”, Collected Scientific Papers, Vol. 2 (Cambridge:
Cambridge University Press), 1888
Williams, M. L., Landel, R. F., Ferry, J. D., "The Temperature Dependence of Relaxation
Mechanisms in Amorphous Polymers and Other Glass-forming Liquids". Journal of the American
Chemical Society 77, 1955

9. Acknowledgement
These investigations are carried out through the R&D activities of the project “KITe hyLITE Plus:
Technologies and network development for hybrid lightweight construction solutions in the
automotive industry”. This project is funded by the European Union through the program
“European Funds for Regional Development” as well as state government of Baden-Wuerttemberg
in Germany. Administrative agency of this program is the Ministry of Rural Development, Food
and Consumer Protection. For more information visit www.rwb-efre.baden-wuerttemberg.de

Application of Predictive Engineering Tool
(Abaqus) to Determine Optimize Rubber Door
Harness Grommet Design
Praveen Mishra, Dayananda Gowda
Mercedes Benz R & D India, Bangalore, Karnataka, India
Abstract: A cable grommet is a tube ring through which an electric cable passes. They are usually
made of rubber or metal. The cable grommet is used to protect, improve friction or seal cables
passing through it, from a possible mechanical or chemical attack and for aesthetic look. It
becomes very important to determine the optimum length of the grommet used for particular
application. Bigger length will have results into folding of grommet and lesser length will yield
into stretching of rubber grommet. Finite element simulation technique presented in this paper
can be used to determine the optimum geometry of the grommet. Component used for this study is
an automotive door harness rubber grommet.
Using Abaqus/CAE hyper elastic material models are evaluated by creating response curve using
selected energy potentials. Based on the stability for that particular application hyper elastic
model is finalized. Finite element simulation using Abaqus/Explicit helps the designer to optimize
the length or material of grommet for that particular application. This gives engineering direction
to designer to do necessary changes in the beginning of the design phase freeze which leads into
save of time and money.
Keywords: Grommet, Hyper elasticity, Optimization, Abaqus CAE, Abaqus /Explicit.
1. Introduction
In Automotive cars, it requires a lot of wiring harness system to distribute power from one system
to other. Any wiring harness should have sufficient strength to withstand any abrupt situations
without affecting the performance of the total system. An opening and closing member is attached
to a vehicle by a hinge enabling easy opening and closing of different moving member. Such
members include side door, rear door, trunk lid etc. A guide member called Rubber grommet made
of elastomer has an accurate portion which can be extend and contract while twisting with
A cable grommet is a tube ring through which an electric cable passes. They are usually made of
rubber or metal. The cable grommet is used to protect, improve friction or seal cables passing
through it, from a possible mechanical or chemical attack and for aesthetic look. It becomes very
important to determine the optimum length of the grommet used for particular application. Bigger

grommet length will have results into folding of grommet and lesser length will yield into
stretching of rubber grommet
The objective of this paper is to suggest a methodology to model the behavior of rubber grommet
under specified dynamic loading conditions. For this purpose rubber grommet has been modeled
with hyperelastic material properties and simulation is conducted with a non-linear commercially
available FE code Abaqus. First a stable hyperelastic material model is selected using Abaqus
/CAE material evaluation option and later an explicit simulation is performed using
Abaqus/explicit.
1.1 Defining Hyper Elastic Behavior using Test data
Rubber grommet is modelled with three dimensional solid elements. Fig 1 shows the rubber
grommet in folded conditions. Main challenge for this component is to have the stable material
law which can withstand for the given boundary and loading conditions. To have stable model a
stress –strain data is obtained for
• Uniaxial tension data

• Uniaxial compression data
• Biaxial data
• Plane shear
1.2 Defining hyperelastic behavior using test data

The convenient way to defining a hyperelastic material is to give experimental data to Abaqus.
Abaqus then calculate the constants based on the least squares method. Abaqus can fit the data for
the following experimental tests.
• Uniaxial tension and compression
• Equibiaxial tension and compression
• Planar tension and compression (pure shear)
• Volumetric tension and compression
Unlike plasticity data, test data for hyperelastic material must be given as nominal stress and
nominal strain values.For time-independent nonlinear elasticity, the fitting procedure may be

carried out for polynomial representations of incompressible materials. The significance of (non-
equivalent) multiple tests for material modeling cannot be overemphasized. In general, a
combination of uniaxial tension/compression and simple shear is required in the very least. Data
from biaxial tension or planar shear may also be needed depending on the deformation modes of
the structure. Volumetric data must be included for materials undergoing large compressible
deformations.
The quality of the results from a simulation using hyperelastic materials strongly depends on the
material test data provided to Abaqus to generate the material coefficient. It is always
recommended to have more than one deformation state to form accurate and stable material
model. For our grommet simulation uniaxial tension and compression data is enough to capture
the physics of the grommet behavior. It is common for the material model determined from the
test data to be unstable at certain strain magnitude. Abaqus performs a stability check to determine
the strain magnitude where unstable behavior will occur. So it is possible for the material to be
unstable at the strain level indicated in the deformation is more complex and it is also possible for
the material to become unstable at lower strain levels if the deformation is more complex.
Following Figure 1 & 2 shows the compression and tension data for the elastomer used for rubber
grommet simulation.
Stress (MPa)
Strain (mm/mm)
0
Figure 1. Uniaxial compression material data at 23 C.

Stress (MPa)
Strain (mm/mm)
0
Figure 2. Uniaxial compression material data at 23 C.
Both tension and compression data are allowed, with compressive stresses and strains entered as
negative values. During dynamic simulation some part of the grommet will be under compression
and some part will be under tension hence the both uniaxial tension as well as compression data is
used to generate the hyperelastic material coefficients.
Using Abaqus /CAE material evaluate option various hyper elastic models are evaluated. Fig 3
shows the correlation of the various models with the test data. For this simulation second-order,
polynomial strain energy function is used to model the rubber material. In input file these choices
are invoked by using the N= 2 and POLYNOMIAL parameters on the *HYPERELASTIC
keyword option. TEST DATA INPUT parameter is used to indicate that Abaqus should find the
material constants from the test data given in the input file. The test data are given on options that
immediately follow the *HYPERELASTIC option. The data is entered as nominal stress and the
corresponding nominal strain, with negative values indicating compression. Second order
polynomial strain energy function shows a better fit with the test data and material is also stable at
higher strain values. Hence for grommet simulation polynomial N=2 material model is used.

Figure 3. Least square fit for the various Hyperelastic models
2. FE Simulation
Rubber grommet has been modeled with very fine solid second order tetrahedral element (Figure
4). One end of the grommet is connected to the plastic connector and then attached to the door
and other end of the rubber grommet is connected to the door. Node of the rubber are rigidly ties
to the plastic connector .To simplify the problem and reduce the computational time door is model
as a rigid.
Figure 4. Grommet finite element model

Hinge location of the door is fixed in all degrees of freedom. A local coordinate system is defined
(whose one axis is made parallel to the hinge axis) and the door is rotated 70 degrees along the
hinge axis. Explicit simulation is performed and automatic general contact definition is used to
define the contact.
Figure 5. Door cut model with hinge rotation axis
3. Simulation Results
Figure 6 shows the stress and strain in the grommet which gives us an idea of the compression and
tension occurring in the grommet during closing and opening of the door. For this simulation 70 %
nominal strain is observed. Various other profiles are created based on the simulation results. By
performing the simulation with different designs with varying convolutes of the grommet it
becomes easy to find out the optimum length and design of the grommet.

Figure 6. Stress & Strain contour in grommet finite element model
4. Conclusion
Rubber grommet profile is not uniform. With the accurate hyperelastic materials model it become
easy to have the precise results. With the CAE simulation it is easy for designer to design an
optimized shape and size of the rubber grommet. In the development phase of the automobile
vehicle this simulation will help to avoid any last moment surprise at the end of the vehicle
development cycle. This gives engineering direction to designer to do necessary changes in the
beginning of the design phase freeze which leads into save of time and money.
5. Acknowledgement
I would like to thank Mercedes Benz research & development India management for giving me an
opportunity to work on this topic and extending their support whenever required.
6. References
1. Byeong-Sam Kim, Kyoungwoo Park, Young-Woo Kim, “Endurance Analysis of Automotive
Vehicle’s Door W/H System Using Finite Element Analysis,2009”
2. Abaqus Users Manual, Version 6.13-1, Dassault System Simulia Corp., Providence, RI.
3. Getting Started with Abaqus–Keyword Edition Version 6.13-1, Dassault System Simulia
Corp., Providence, RI.

Final Element Simulation of Blankholder's Lift-off in
a Deep Drawing Tool Using Abaqus/Standard
Krzysztof Swidergal, Marcus Wagner
OTH Regensburg, Laboratory for Finite Element Analysis and Structural Dynamics, Germany
Christian Lubeseder, Ingo von Wurmb, Josef Meinhardt
BMW Group, Construction methods and Standards Department, Munich, Germany
Steffen Marburg
Universität der Bundeswehr München, LRT4 – Institute of Mechanics, Neubiberg, Germany
Abstract:
In the deep drawing tools for forming car body parts, heavy blankholders are used to prevent
buckling and wrinkling of the blank. During each press cycle, those large masses need to be lifted,
raising thereby the structural dynamic load on the deep drawing tool and on the press. Therefore
a detailed knowledge about the blankholder’s lift-off event is essential for an accurate and robust
design of forming tools. In this paper, a dynamic finite element method (FEM) simulation of a
blankholder’s lift-off in a selected automotive deep drawing tool is presented enabling
identification of regions of critical stresses. The FEM model is built within the Abaqus/CAE
environment and solved with Abaqus/Standard. Each dynamic analysis is preceded by a static
analysis where the gravity load is applied and the lifting bolts are pre-stressed. A special
emphasis is put on modeling the elastomer dampers, which are installed between lifting bolts and
the blankholder to avoid hard impacts during the lift-off event. Those dampers are modelled using
a hyperelastic material with hysteresis. In addition, an experimental validation of a blankholder’s
vibration under operating loading was carried out. The simulation results are in good agreement
with the measurements.
Keywords: Blankholder, Deep Drawing Tool, Elastomer Damper, Vibration, Mechanical Press
1. Introduction
With new press systems with higher strokes rates (Osakada, 2011) emerging, further savings in
cost and time in the production of sheet blank parts are possible. One drawback of the stroke rate
increase is that also the loading on the deep drawing tool and its components rises. Especially, the
dynamic loads on the blankholder - used to prevent buckling and wrinkling of the sheet blank -
can get considerably higher, as during each press cycle, its large mass needs to be lifted. This
could in the worst case not only lead to the damage of the blankholder structure, but also affect the
forming tool and press. Therefore, a detailed knowledge about the blankholder’s lift-off event is
essential for an accurate and robust design of deep drawing tools. Such a dynamic structural
problem can, for example, be investigated by means of numerical simulation. In (Swidergal,
2014), the coupled multibody finite element method (MBS-FEM) is used to analyze the vibration

of the blankholder. Another well-established simulation method is the implicit FEM. In this work
the detailed implicit FEM analysis of a blankholder’s lift-off event will be conducted. To the
author’s best knowledge, no paper is published in which such investigation has been presented.
2. Developing the FE model of the blankholder

The FEM model of the blankholder is built within the Abaqus/CAE environment. From the CAD
geometry of the complete deep drawing tool the subassembly of the blankholder is extracted and
imported into Abaqus/CAE. Using the neutral Parasolid format, an automatic positioning of all
components is established. Sharp edges and sliver faces are defeatured to avoid bad shaped
elements during meshing. The blankholder is meshed with quadratic tetrahedrons of type C3D10
and for all other metal components the linear hexahedral elements of type C3D8R are used. For
elastomers also a hybrid formulation is activated. The final meshed assembly of the blankholder
can be seen in Figure 1.
Figure 1. FEM model of the blankholder assembly.

In the model all parts are defined deformable. Their material parameters are assigned using
homogeneous section definition. For example, for the blankholder, which is made of cast iron EN-
GJS-600-3, an elasto-plastic material definition, with properties listed in Table 1, is used.
Table 1. Material properties of EN-GJS-600-3.
Property Unit Value
Young’s modulus MPa 174000
Mass density MPa 7200
Poisson’s ratio - 0.275
Yield strength MPa 370
Ultimate strength MPa 600
In the investigated deep drawing tool, eight so called lifting bolts are used to lift the blankholder.
In addition, several sliding pads and guide pillars are installed to ensure that the blankholder can
only translate vertically. Therefore, as a first approximation, modeling the lift-oft event can be
reduced to one lifting bolt only.

2.1 Modeling the lifting bolt
The model of the lifting bolt assembly is shown in Figure 2. Here, the upper die and the
blankholder are simplified by a solid cylinder. To account for the proper weight, an extra
nonstructural mass is added and distributed on underlying mesh nodes. The matrix, which supports
the blankholder and prevents it from falling when the gravity is activated, is modeled as analytical
rigid shell part. The lifting bolt is tied to the upper die by the means of a tie constraint. For all
other interactions a surface-to-surface contact formulation is used. In addition, a frictional
behavior is included in the interaction properties.
upper die (reduced)
distance sleeve
blankholder (reduced)
matrix
elastomer damper
lifting bolt
Figure 2. Schematic view of the blankholder’s lifting bolt model.
The movement of the upper die is defined in an implicit dynamic analysis step and is realized by a
displacement boundary condition with a tabular amplitude, which describes the motion of the
press slide at specific stroke rate. The dynamic analysis is preceded by a static analysis where the
gravity load is applied and the lifting bolt is pre-stressed.
2.2 Modeling the elastomer damper

Between the lifting bolt and blankholder, elastomer dampers are installed to avoid hard impacts
during the lift-off event. Those dampers are made of carbon filled elastomer rubber which
possesses highly non-linear stiffness characteristics. To obtain that characteristic uniaxial
compression tests were carried out. The result of this experiment for strain rate 𝜀̇~0.03 s-1 can be
seen in Figure 4.

In FEM, the elastomers are typically modeled using a hyperelastic material formulation, which
utilizes the strain potential energy for computing the stresses. In Abaqus, there are several forms
of strain potentials available to model approximately incompressible isotropic elastomers (Abaqus,
2015). If only uniaxial material test data are available, the Marlow model (Marlow, 2003) is
recommended (Abaqus, 2015). This is an isotropic, incompressible hyperelastic model, with the
strain energy potential defined as
𝑈 = 𝑈𝑑𝑒𝑣 (𝐼1̅ ) + 𝑈𝑣𝑜𝑙 (𝐽𝑒𝑙 ),
where 𝑈 is the strain energy per unit of reference volume, with 𝑈𝑑𝑒𝑣 as its deviatoric part and 𝑈𝑣𝑜𝑙
as its volumetric part; 𝐼1̅ is the first deviatoric strain invariant defined as
𝐼1̅ = 𝜆21̅ + 𝜆2̅2 + 𝜆23̅ ,
1
where the deviatoric stretches 𝜆̅𝑖 = 𝐽−3 𝜆𝑖 ; 𝐽 is the total volume ratio; 𝐽𝑒𝑙 is the elastic volume ratio
and 𝜆𝑖 are the principal stretches (Abaqus, 2015). Hereafter, the Marlow model is chosen.
To represent the energy dissipation in the elastomer damper, a hysteresis model (Bergström, 2000)
is used. This model is controlled by four parameters, which were further investigated. An
optimization study with curve fitting is carried out to obtain their values.
Table 2. Parameters of the hysteresis model.
Property Unit Value
Stress Scaling Factor S - 1.6
Creep Parameter A s–1 MPa–m 0.7
Eff. Stress Exponent m - 2
Creep Strain Exponent C MPa -0.5
Finally, with the hysteresis parameters shown in Table 2 a good agreement between measured and
simulated damper response curve is found, as can be seen in Figure 4.
1
0.8
force / max. force
0.6
experiment
0.4 simulation
0.2
0
0 1 2 3 4 5
displacement [mm]
Figure 4. Uniaxial response of the elastomer damper.

In the end, the FEM model is solved in parallel with Abaqus/Standard.

3. Results and discussion
In Figure 5, the von Mises stress of the lifting bolt assembly for the highest stroke rate is shown.
(a) after pre-stressing the lifting bolt (b) during blankholder lift-off, t = 0.024s
Figure 5. Von Mises stress in the lifting bolt assembly.

In Figure 5a stress state after pre-stressing the lifting bolt can be seen, whereas in Figure 5b stress
state during blankholder lift-off at time t = 0.024 s is presented. Comparing both states, an increase
in stress during the lift-off event can be observed. In the reduced blankholder structure, for the
assumed stroke rate, the highest stresses of about 12 MPa occurs at time t = 0.024 s. Similarly, the
stress in the lifting bolt increases, reaching 206 MPa, which is about 23% more compared to the
pre-stressed state. For validation, the numerically obtained velocity response of the complete (non-
reduced) blankholder assembly was compared to the signal gained in experiments, which were
carried out under operational loading. The both velocity responses are shown in Figure 6.

velocity* / max. velocity 1.0 experiment
0.6 simulation
0.2
-0.2
-0.6
-1.0
2.20 2.25 2.30 2.35 2.40 2.45 2.50 2.55 2.60 2.65 2.70 2.75
time [s]
Figure 6. Vertical velocity of the blankholder (*highpass, cutoff freq.: 5Hz).

It can be seen, that the simulation results are in good agreement with the measurements. Moreover,
during the lift-off event, the blankholder vibrates three times before it comes to a rest.
4. Summary and conclusion

In this work, a finite element analysis of the blankholder lift-off event in Abaqus/Standard was
successfully conducted. Based on an experimental validation, the simulated resulting dynamic
loading on the blankholder showed good agreement with measurements. In addition, the regions of
critical stresses in the blankholder structure could be identified. Thanks to this knowledge, a
deeper understanding about the blankholder’s lift-off event was gained. Therefore, the structural
dynamic loading on the deep drawing tool and indirectly on the press can now be predicted.
Hence, with this approach, more accurate and robust design of forming tools in the future is
possible. Further work on modeling the complete deep drawing tool in Abaqus/Standard is in
progress.
5. References
1. Osakada, K., Mori, K., Altan, T., Groche, P., “Mechanical Servo Press Technology for Metal
Forming”, CIRP Annals - Manufacturing Technology, vol. 60, pp. 651-672, 2011.
2. Swidergal, K., Lubeseder, C., von Wurmb, I., Meinhardt, J., Wagner, M., Marburg, S.,
“Vibration Analysis of an Automotive Forming Tool Using Coupled MBS-FEM Simulation
and Experimental Validation”, Proceedings of ISMA 2014, pp. 2931 – 2942, Leuven,
Belgium, 15-17 September, 2014
3. Abaqus User’s Manual, Version 6.14-3, Dassault Systémes Simulia Corp., Providence, RI.
4. Marlow, R., “A General First-Invariant Hyperelastic Constitutive Model”, Proceedings of the
Third European Conference on Constitutive Models for Rubber, pp. 157-160, London, UK,
15-17 September, 2003
5. Bergström, J., Boyce, M., “Constitutive Modeling of the Large Strain Time-Dependent
Behavior of Elastomers”, Journal of the Mechanics and Physics of Solids, vol. 46, no. 5, pp.
931–954, 1998

Strength Assessment of Injection Molded Short-
Fiber-Reinforced Plastic Components
Wolfgang Korte, Marcus Stojek, Sascha Pazour
PART Engineering GmbH, Germany
Abstract: Components made of short-fiber-reinforced plastics (SFRP) are stressed highly both
mechanically and thermally. Therefore an intelligent component design is required in order to
fully exploit the potential of these materials. Hence, the design of such components must be based
on a reliable strength assessment. For this purpose models for the description of the anisotropic
and elasto-plastic failure behavior of SFRP are required. In contrast to the widespread use of
SFRP, methods for a reliable strength assessment based on FE analyses for components made of
these materials have not been sufficiently developed yet. This paper presents an approach for the
strength assessment of SFRP components based on FE analyses. In the scope of this appropriate
failure limits and failure criteria for these materials are presented.
Keywords: Anisotropy, Failure Criterion, Injection Molding, Polymer, Plastics, Short-Fiber-

Reinforced Composites, Tsai-Hill, elasto-plastic
1. Introduction
The use of injection molded SFRP parts take place in many different industries. In particular, in
the automotive industry SFRP are increasingly being used as the preferred engineering plastic.
This is due to the excellent mechanical and thermal properties of these materials compared to non-
reinforced grades. The use of these materials enable the manufacturer of automotive components
to a significant weight reduction compared to metallic materials and lower manufacturing costs as
well due to the fact that a plastic part is typically a finished part without any further processing
necessary.
The molding process is the cause for the formation of a specific microstructure within the molded
part, which in turn is the root cause for the mechanical properties of the part. Especially for
plastics this microstructure can be molecular orientations or in the case of SFRP fiber orientations,
leading to an anisotropic material behavior. In this sense it can be said that the material is
composed during the molding process. In figure 2 the effect of fiber orientation onto the
mechanical behavior of a plastic is shown.

Figure 1: Influence of Fiber Orientation on Material Properties
It is evident that due to the described strong impact of the injection molding process on the
properties of the manufactured part itself the structural analyst is interested in considering the
effect of anisotropy on part stiffness and strength. Such a consideration may result in a more
accurate prediction of the mechanical behavior of the investigated component.
Whereas the proper description of the stress-strain behavior of the part is a question of a suitable
material model that in the case of SFRP is e.g. capable to consider anisotropic elasto-plastic
behavior the proper assessment of the part failure is a question of a suitable failure limit and
failure criterion. In this paper is the focus on the failure assessment, assuming that a suitable
material model was already chosen in order to calculate the local stresses and strain in the part
properly.
2. Local Component Properties

In technical parts there is not a uniform component strength rather than a local strength
distribution depending on manufacturing and design. Particularly for injection molded SFRP parts
the material is generated during the molding, depending on the position in the part the material
possess different properties dependent on the local fiber orientation. Additional factors that
influence the local component strength are the degree of multiaxiality and to which amount the

local stress on the component surface propagates into the cross section. In the following the
influence of these different factors will be described.
Influence of fiber orientation
As already outlined SFRP have significant different mechanical properties parallel and transverse
to the fiber direction. Opposed to unidirectional endless fiber reinforced composites where the part
is composed of several unidirectional layers for injection molded SFRP parts there exist a fiber
orientation distribution instead of ideally aligned fibers.
The orientation state of these fibers at every discrete location in the part can be described more in
a probabilistic rather than in a deterministic manner. It is evident that for nearly ideal aligned
fibers (narrow fiber distribution curve) different mechanical properties result than for nearly
randomly oriented fibers (broad fiber distribution curve). This has to be considered in the
assessment of the local strength of the part (Figure 2).
Figure 2: Stress-/Strain-Curves for Different Fiber Orientation States (PBT+GF20)
For a narrow distribution the degree of orientation is close to unity (nearly uniaxial) and for a very
broad distribution close to zero (nearly randomly respectively quasi-isotropic). The information
about the local degree of orientation and the principal orientation direction can be extracted from

the orientation tensor which is provided for every element as standard output by most injection
molding solvers.
Influence of triaxiality
In real parts most probable multiaxial stresses occur in contrast to specimen e.g. for the
determination of the materials’ mechanical properties. Hence there is a need to choose a failure
criterion that is capable on the one hand to take into account multiaxial stresses and on the other
hand considers the anisotropy of the material strength with regard to the local fiber orientation.
Another effect to be taken into account is the increasing brittleness of the material due to
triaxiality which in turns leads to decreasing allowable strains and a change in the fracture mode
from e.g. ductile fracture due to deviatoric stresses to brittle failure due to normal stresses. This
effect is obvious in notches where a brittle failure occurs frequently though the material itself
behaves ductile under unidirectional loading. This effect can be explained mechanically with the
increasing constraint of shear deformations, that are typical for a ductile fracture, if a multiaxial
stress state is present. This leads to the need to evaluate the specific allowable material strength or
strain respectively with regard to the local degree of triaxiality in the part.
Influence of stress propagation into local cross section

If the stress hot spot comprises only small areas of the cross section e.g. typical for sharp notches
these stresses do not have to be evaluated as critical for the part failure as stresses that exist
uniformly across the local wall thickness of the part. Therefore there is a need to evaluate what can
be considered as failure with regard to the part not to be mixed up with failure due to the ultimate
exceeding of the material strength.
3. Proposed Approach for Strength Assessment

In Figure 3 an approach is outlined that we propose in order to make a strength assessment of
injection molded SFRP parts.

Figure 3: Proposed Approach for Strength Assessment of SFRP Parts
Basis of the assessment are the loading, the material chosen and the component design. In order to
determine the stresses and strains in the component properly, in this paper it is assumed that a
suitable material model for SFRP was used for the FE analysis, e.g. an anisotropic elasto-plastic
model. Which implies that a suitable software e.g. Converse (Converse, 2015) was used in order to
provide the local fiber orientation in the part element-wise as well as the material properties
needed for such an anisotropic elasto-plastic material model.
Second the strength properties of the material are needed in form of appropriate allowable stress or
strain limits. As mentioned earlier in the case of injection molded SFRP parts the material does not
exist by itself it is generated during manufacturing. However accessible are just material properties
that are determined with e.g. ISO tensile specimen either molded directly or cutted from test
plaques to which it is referred in the following as tabulated values. These tabulated values do not
necessarily reflect the values that exist in the component. The manufacturing conditions in the
specimen are in general different than the conditions in the component this is particularly true for
the local fiber orientation. Hence an approach is needed that is capable to determine the strength
properties of the material for different degrees of orientation based on the tabulated values.
Third design values are defined that consider the effects of the local degree of orientation, the
local degree of triaxiality and the propagation of stress from the component surface into the local
cross section. The design variables together with the orientation dependent material strength
determine the local component strength at each position in the part.
Eventually the ratio of the computed local stress and local component strength provides the
utilization ratio which is the inverse of the safety factor. Utilization ratios larger than unity
indicate a component failure.

4. Orientation-Dependent Material Failure Limits
In order to determine a functional dependency of the material failure limit on the degree of
orientation which is a characteristic number for a particular orientation distribution it is obvious
that for sufficient discrete points the related stress respectively strain limits are required. Each
orientation distribution is mathematically represented by the orientation tensor which in turn can
be obtained as output from an injection molding simulation.
The eigenvalues of the orientation tensor - the so-called a-values - which can be computed by a
principal axis transformation are a representation of the particular shape of the fiber orientation
distribution. For instance a1 = 1; a2 = 0; a3 = 0 represents a unidirectional distribution and a1 =
0.33; a2 = 0.33; a3 = 0.33 represents a quasi-isotropic (random) distribution without any pre-
dominant fiber direction. The principal axis system gives the directions in which the fibers are
aligned with the 1-direction always the direction in which most of the fibers are aligned and the 2-
and 3-direction always perpendicular to that direction. Assuming tranversely isotropic behavior for
SFRP which is a plausible assumption the mechanical properties in 2- and 3-direction are equal for
unidirectional conditions where a1 = 1, hence all fibers are aligned in the same direction. For such
conditions the strength limits in fiber and transverse to the fiber direction could be determined
directly if ideally unidirectional oriented specimen could be manufactured. Unfortunately this is
practically not possible. There are always deviation to an ideal orientation at each position of the
specimen and also there are different orientation across the wall thickness of the molded part due
to specific flow effects (Figure 4). Even if the specimen are cutted from plaques that are molded
with a film gate there is not a 100% orientation of the fibers.

Figure 4: Fiber Orientation in Tensile Specimen
The proposed procedure in order to attain the orientation-dependent strength limits of the material
is to conduct short term tensile tests with specimen cutted out of test plaques with an angle of 0°,
30° and 90° to the flow direction. In a re-engineering approach with an anisotropic elasto-plastic
material model (Hill yield criterion) the specimen are simulated with consideration of the fiber
orientation that are provided from an injection molding simulation. This approach then delivers
the Hill yield ratios Ri as a function of the eigenvalues of the orientation tensor ai. With that also
the strength respectively strain limits of the material are defined, by scaling of the complete stress-
/strain-curve with the yield ratios for different fiber orientation.
5. Component Strength
In order to determine the local component strength the design values are needed. The effect of
local fiber orientation is considered element-wise with the a-values of the orientation tensor and
the principal orientation direction. In Abaqus this is realized with local element orientations for the
principal orientation direction with the 1-direction always aligned in fiber axis and with field
variables for the eigenvalues ai of the orientation tensor. In this manner for each element a
belonging orientation-dependent stress-/strain-curve can be assigned.
The influence of the triaxiality on the material strength respectively strain is considered with the
stress triaxiality number η = -p/q, p is the hydrostatic pressure and q is the Mises equivalent stress.
The triaxiality number can be requested in Abaqus as output variable TRIAX. With this number
the allowed strain limit of the material is scaled from its maximum value breaking strain for
unidirectional loading to a minimum allowed strain at the transition point between linear-
viscoelastic and nonlinear viscoelastic behavior for high triaxiality.
The influence of the triaxiality on the failure value itself is considered with a suitable failure
criterion for which the Tsai-Hill criterion was chosen. Since in its original formulation the Tsai-
Hill criterion is only valid for unidirectional behavior here it is defined as a function of the
eigenvalues of the orientation tensor ai, which are accessible for each element as field variables
(Korte, Stojek, 2015).
In fact by applying this procedure for each element a different value of the Tsai-Hill value follows.
The square root of the Tsai-Hill value can be interpreted as utilization ratio since the allowed limit
stresses are in the denominator and the actual stresses are in the nominator. For values of the Tsai-
Hill criterion larger than unity failure has to be assumed. As allowed limit stresses the ultimate
breaking stresses in the different directions were used.
6. Application Example
As an application example in Figure 5 a bicycle brake lever is shown. The lever is made out of a
PBT with 20 percent glass fiber content. In order to test the components stiffness and strength the
test set-up also shown in Figure 5 was used (Stommel, 2015).

Figure 5: Test Set-up and Fiber Orientation in Bicycle Brake Lever
The component was simulated with Abaqus with an anisotropic elasto-plastic material model with
consideration of the fiber orientation distribution. The fiber orientation were obtained from an
injection molding simulation and then were transferred in the mechanical FE model by applying
Converse that provided the appropriate material parameters for the anisotropic elasto-plastic
material card in Abaqus as well (Pazour, 2015). The assessment of the component strength was
conducted in post processing with a software prototype (S-Life/Plastics, 2015) that is capable to
read an Abaqus odb file. As already described in this manner it is possible to access element-wise
the a-values of the orientation tensor that are stored in the odb file as field variables. Subsequently
within the software for each element the orientation-dependent value of the utilization ratio
(square root of the Tsai-Hill value) was computed. A contour plot of the computed utilization
ratios is shown in Figure 6.

Figure 6: Utilization Ratio and Failed Part
For values larger than unity failure has to be assumed. The computed utilization ratio indicates a
good correlation to the failed tested part with regard to the failure location, also shown in Figure 6.
As well a good correlation with the failure displacement respectively force was predicted with the
applied utilization ratio as failure indicator.
7. Summary
In the scope of this paper an approach was presented that is capable to give good failure
predictions for injection molded SFRP under mechanical loading. In the computed utilization
ratios as failure criterion an orientation-dependent formulation of the Tsai-Hill equation was used.
Not yet implemented in the computation of the utilization ratio is the proposed consideration of
the influence of the triaxiality on the failure limits as well as the effect of stress propagation from
the component surface into the component.
8. References
1. Korte, W., Stojek, M., “Entwicklung einer Software zur orientierungsgradabhängigen
Kalibrierung eines anisotrop elasto-plastischen Standard-Materialmodells zur FEM-
Simulation von kurzfaserverstärkten Kunststoffbauteilen” unpublished report ZIM Koop
research and development project No. KF3056701RR2, Bergisch Gladbach, Germany, 2015.

2. Korte, W., Stojek, M., “Entwicklung eines Verfahrens zur Festigkeitsbewertung von
kurzfaserverstärkten Kunststoffbauteilen” unpublished report ZIM Solo research and
development project No. EP110100, Bergisch Gladbach, Germany, 2013.
3. Stommel, M., “Entwicklung und Umsetzung einer hybriden Simulationsmethodik zur
effizienten Berechnung und Auslegung von kurzfaserverstärkten Kunststoffbauteilen”
unpublished report ZIM Koop research and development project No. KF2214510RR2,
Saarbrücken, Germany, 2015.
4. Pazour, S., “An Easy Procedure for Anisotropic Non-Linear Behavior of Short-Fiber-
Reinforced Plastics” Proceedings Simulia Community Conference 2015, Berlin, Germany
5. Converse, Software, PART Engineering GmbH, Bergisch Gladbach, Germany, 2015
6. S-Life/Plastics, unreleased software, PART Engineering GmbH, Bergisch Gladbach,
Germany, 2015
9. Acknowledgment
Part of the investigations in this paper are sponsored by Federal Ministry of Economics and
Technology due to a resolution of German Federal Parliament in the scope of the funding program
„ZIM - Zentrales Innovationsprogramm Mittelstand “
Teile der Untersuchungen in dieser Veröffentlichung sind gefördert durch das Bundesministerium
für Wirtschaft und Technologie aufgrund eines Beschlusses des Deutschen Bundestages im
Rahmen des Förderprogramms „ZIM - Zentrales Innovationsprogramm Mittelstand“

Simulation Driven Design for Manufacture of a
Pressed Composite Automotive Tailgate
Andy Ngai, Mark Arnold
Penso, UK
Abstract: A current key automotive industry initiative is delivery of mass savings through the use
of lightweight materials, to reduce vehicle C02 emissions. Penso were commissioned by a
European automotive OEM to design and manufacture a continuous fibre reinforced composite
tailgate prototype. The new carbon fibre design had to meet all existing strength and stiffness
targets, whilst maintaining existing A-surface geometry and all hardware. The three main parts of
the tailgate considered for light weighting were the tailgate inner panel, tailgate outer panel and
spoiler mechanism carrier. An optimised laminate design was developed by the engineering team
using a non-linear static FE analysis performed with Abaqus/Standard. The Tsai-Wu failure
measure and a user defined inter-laminar failure criterion were used as design metrics in post
processing. The design was then evaluated with Penso’s composite manufacturing team to develop
preliminary ply shapes and overlap joint locations suitable for manual draping and Penso’s
pressed composite technology. These ply boundaries and material directions were updated in the
FE model for further refinement of ply joint locations, prediction of finalised mass/performance
and generation of layup manuals. The A and B-surfaces of the parts were then updated within
CAD and used to create tool geometry for pressed composite. The final design of parts gave a
combined mass save of over 65% whilst having comparable structural performance to the
production design.
Keywords: Carbon Fibre, Composites, Design Optimization, Failure, Finite Element Analysis,
Laminate, Manufacture
1. Introduction
Penso were commissioned to design and manufacture a continuous fibre reinforced composite
tailgate. This would be a lightweight alternative to the current production tailgate which is
moulded from Sheet Moulding Compound (SMC), a discontinuous-fibre reinforced composite
material.

2. Tailgate design
The production tailgate structure comprised of a bonded sub-assembly of inner and outer SMC
panels, with a steel spoiler mechanism carrier bolted to the outer panel, as illustrated in Figure 1.
Metallic reinforcing mounting plates for the hinges, gas struts and latch were bonded to the inside
of the inner panel, with the rear screen glass bonded to the outer panel. The mass of the three main
panels were 5.67 kg, 3.25 kg and 1.13 kg for the tailgate SMC inner panel, tailgate SMC outer
panel and steel spoiler mechanism carrier respectively, giving a total mass of 10.05kg for the three
panels.
Figure 1. Tailgate parts considered for light weighting.

Three main parts of the tailgate were considered for light weighting; these were the tailgate inner
panel, tailgate outer panel and spoiler mechanism carrier. The new carbon fibre design had to meet
all existing strength and stiffness targets, whilst maintaining existing A-surface geometry,
hardware and functional requirements. The structural performance targets against which the
tailgate would be evaluated were torsion, cantilevered bending, latch load and margin & flushness.
The composite parts comprised a continuous fibre reinforced polymer material, using pre-
impregnated (prepreg) epoxy resin carbon fibre reinforcement. The composite parts were
laminated from a single prepreg composite sheet material supplied by Cytec Industries. The
reinforcement material was a Toray T300 carbon fibre in a 6k tow size, a standard
strength/modulus fibre used in a number of industry sectors. The matrix material was Cycom 2020

epoxy, a traditional autoclave cure resin developed by Cytec for structural motorsport
applications, with a nominal resin weight fraction of 40%. This carbon fibre was woven into a 2x2
Twill fabric having a nominal areal weight of 400g/m2. The nominal fibre volume fraction and
cured ply thickness for this prepreg material were 52% and 0.44mm, respectively. Mechanical
properties for this prepreg were obtained from static coupon tests performed by the material
supplier in longitudinal (warp) and lateral (weft) fibre directions, with an average tensile modulus
of 58 GPa, tensile strength of 550 MPa and compressive strength of 650 MPa. These orthotropic
elastic material models were validated through simulation of coupon testing performed according
to various ASTM test methods (Arnold et al, 2014).
The prototype composite parts of the tailgate were manufactured using Penso’s pressed composite
technology. Pressed composite technology is a process in which the prepreg composite material is
placed into an open matched metal mould cavity. The tool is then closed under pressure to
encapsulate the fibre and resin compound within the mould cavity. Heat and pressure are
maintained until the composite material has cured.
3. Laminate optimisation
3.1. Finite element model

The CAD geometry was imported into the ANSA pre-processing software and converted into a
surface suitable for shell meshing. The visible surface was chosen as the A-surface of each panel.
The mesh was created on the A-surface which corresponded to the preform tool surface, such that
the laminate plies stacked from A to B surface. Each panel was meshed with first order
conventional S4 and S3R shell elements, having a nominal element edge length of 5mm.
Abaqus/Standard was used for the non-linear static finite element analysis for the tailgate load
cases. The composite panels were modelled with laminate properties defined using *SHELL
SECTION, COMPOSITE, where the Material identifier (MID), cured ply thickness (T) and
orientation angle in longitudinal direction (PHI) of each ply are defined within the stacking
sequence. Material orientation for each element was defined using a combination of
*ORIENTATION, *DISTRIBUTION TABLE and *DISTRIBUTION such that the material X
axis was orientated predominately in the fore/aft direction of the panel. Orthotropic elastic
material models were defined for the macroscopic behaviour of the woven fabric prepreg using
*ELASTIC, TYPE=LAMINA with *FAIL STRESS. The modulus of elasticity for warp and weft
directions were both set to the minimum of these two values obtained from tensile test data,
however it was noted that these values differed from those derived by compression tests. The
tensile strengths in warp and weft directions were both reduced to the open hole tensile strength.
The compressive strengths in warp and weft directions were similarly reduced to the open hole
compressive strength. This was a method proposed by Niu (1992) to incorporate hole and defect
tolerance. The resulting knock down factors for tensile and compressive strengths were 0.75 and
0.45, respectively.
To complete the tailgate model, the metallic mounting plates were modelled as shell elements
referencing non-linear elastic-plastic material properties, whilst the rear screen glass was modelled
with shell elements referencing isotropic elastic material properties. The adhesive was modelled

using solid elements referencing isotropic elastic materials properties. The adhesive elements
connected to neighbouring panels using the distributing coupling constraints.
3.2. Optimisation
Two methods of optimisation techniques were carried out on the tailgate, a manual method and a
design optimisation software were utilised. Strain energy output was plotted to assist with manual
optimisation of the layup. Figure 2 illustrates a combined strain energy envelope plot for all load
cases of the tailgate; the areas of high strain energy indicate the areas where the part is working
hard. To improve the global stiffness of the part, either thickening the laminate by adding extra
reinforcement plies or else altering the fibre orientation is necessary.
Figure 2. Strain energy plot for torsion load case.

A design optimisation software, VR&D GENSIS, was also used for topometry optimisation of the
tailgate composite laminate. This is a method which indicates the layup required for each element
to satisfy the defined objectives. The component is automatically split into patches having
approximate dimensions of 50x50mm, to reduce the number of design variables and to simplify
rationalisation. A candidate laminate was assigned to the global laminate and propagated to the
separate patches. The candidate laminate had a layup of [45/0/45]s, so that the thickness could
vary between a specified minimum and maximum thickness for each angle. The results gained
from the topometry optimisation are continuous, not discrete ply thicknesses. Results are shown in
Figure 3. These continuous topometry results were used to create rationalised patches which group
similar thicknesses and orientations together to define the size and shape of each patch. Once the

rationalised patches have been defined a sizing optimisation can be carried out on the patches, the
results gained are now discrete rather than continuous.
Figure 3. Topometry optimisation results for tailgate inner panel.

An optimised layup for the composite panels, which disregarded joints between all-over plies, was
derived to comply with the structural performance targets, whilst also minimising mass. The
optimised layup was derived by using knowledge gained from both the manual and design
optimisation method. Each panel was constrained to have a minimum thickness of two all-over
plies and a symmetric balanced laminate, to reduce the possibility of distortion during the
manufacturing process. A symmetric laminate is a laminate where all plies are symmetrical about
the mid-plane with regards both material and fibre orientation. A balanced laminate is a laminate
where all the fibres that are oriented at angles other than 0° or 90° occur in pairs of +𝜃 and −𝜃,
but are not necessarily adjacent. Although 2x2 Twill fabrics have slightly different mechanical
properties in warp and weft fibre directions, the use of one or more of such plies orientated at 45°
was considered balanced.
The laminate for the outer panel comprised two all-over plies oriented at 45° relative to the
material orientation axes, with an additional local reinforcement ply oriented at 0° and positioned
local to where the spoiler mechanism carrier is mounted. The laminate for the inner panel
comprised three all-over plies oriented at 45°, 0° and 45°, with three additional local
reinforcement plies oriented at 45°, 0° and 45° enclosed within the main laminate. These
reinforcement plies were positioned at the four corners of the panel to improve global stiffness,
and also at the hinge, latch and gas strut mounting face to improve the strength locally within
these areas. Figure 4 illustrates a thickness plot of the optimised layup for the inner and outer
tailgate panels. The laminate for the spoiler mechanism carrier simply comprised three all-over
plies oriented at 0°, 45° and 0.

Figure 4. Optimised laminate design for tailgate inner and outer panel.
3.3. Finite element results for optimised laminate design
The Tsai-Wu Failure Measure and a user defined Inter-Laminar Failure Criterion were used as
composite design metrics in post processing, to determine if ply failure or inter-laminar shear
failure would occur between the laminate. Failure measure results were plotted using the μETA
post-processor. The in-plane integration points at the middle section point of each ply were used
to output values of the Tsai-Wu Failure Measure. Integration point results were used since these
are the locations where results are typically more accurate (Cook et al, 1989). It was also thought
that extrapolation of the results at the integration points to the corners may introduce additional
uncertainties in failure measures. For the user-defined Inter-Laminar Failure Index, the output
values were taken at surface section points of each ply. Nodal displacements were taken at certain
pre-defined nodes, depending on the load case, to ensure compliance with stiffness targets.
3.3.1. Failure index
For the post processing of the laminate, the Tsai-Wu Failure Criterion was used. The Tsai-Wu
Failure Index (FI) is used with composite materials under loading, with a FI ≥ 1 indicating failure.
Failure index is not proportional to applied load for more complex, interactive failure criteria (i.e.
Tsai-Wu), so is of limited use in design optimisation. The Tsai-Wu Strength Ratio (SR) is a
related failure criterion for composite materials under loading, with a SR ≤ 1 indicating failure.
Strength ratio is inversely proportional to applied load, regardless of failure criterion used, and
therefore preferred for design optimisation. The disadvantage of using strength ratio in post
processing is that the minimum value is of most interest to identify high stress regions, which is
less intuitive for the user. Therefore, an alternative derivation is provided in Abaqus/Standard
called Tsai-Wu Failure Measure, which is proportional to applied load since it is derived from the
reciprocal of Tsai-Wu Strength Ratio (Dassault Systémes Simulia Corp, 2012). To account for
load variability, a safety factor of 1.5 was considered for strength assessment (Niu, 1992). Rather

than incorporate this safety factor into the material strength terms, as with the hole and defect
tolerance, this was instead incorporated during post processing by ensuring the maximum Tsai-
Wu Failure Measure value was <0.67. This was also believed to provide some robustness against
variation in material allowables and uncertainty in FEA methods. Figure 5 shows a Tsai-Wu
Failure Measure plot for the torsion load case. For this load case, a vertical Z-direction load of
267N is applied at the right hand bump stop, while the left hand bump stop is constrained in z-
direction and the hinges constrained in all DOF. The max Tsai-Wu Failure Measure found is 0.36,
indicating ply failure will not occur.
Figure 5. Tsai-Wu failure measure plot for torsion load case.

3.3.2. Interlaminar shear
Interlaminar shear stress, an indication if delamination would occur, is analysed through the use of
the output variable transverse shear stress. Transverse shear stress (TSHR) was requested in
Abaqus/Standard to request results for transverse shear stress in the 13-component, τ13, and
transverse shear stress in the 23-component, τ23, for each ply boundary for every element. These
are defined in the ply local coordinate system. To convert the transverse shear stress from the local
material coordinate system to the global coordinate system the resultant of transverse shear stress
was calculated. This can then be used in a user defined equation shown in Equation 1 to calculate
the Inter-Laminar Failure Index (Bednarcyk et al, 2008). The allowable Inter-laminar shear
strength Fsu13 and Fsu23 was assumed to be the same for the woven fabric.

(1)
Figure 6 shows a resultant transverse shear stress plot for the torsion load case. The max resultant
transverse shear stress was found to be 9.79 MPa, which gave an Inter-Laminar Failure Index of
0.20, indicating delamination would not occur.
Figure 6. Resultant transverse shear stress plot for torsion load case.
3.3.3. Adhesive
Major Principal Stress and Max Shear Stress were evaluated for the failure of the adhesive. These
stresses along with the corresponding allowable stresses for tensile and shear, were used to
calculate a failure index to determine if tensile and shear failure would occur. This method, when
combined with the linear elastic material model, gives a conservative result. As the stresses found
within the analysis for the adhesive were very low, a more rigorous assessment was deemed
unnecessary.
3.3.4. Displacement
Figure 7 shows a displacement plot for the torsion load case. Using the z-deflection at the
unconstrained right hand bump stop and the distance between the two bump stops, the twist angle
can be calculated by trigonometry. The calculated twist angle was 1.63 degrees.

Figure 7. Displacement plot for torsion load case.
3.3.5. Results summary
Table 1 shows the results of the optimised laminate design for torsion load case discussed in
sections 3.3.1, 3.3.2 and 3.3.4, along with results for the remaining three load cases that the
tailgate was analysed against. It can be observed that the stiffness results comply with the targets
set for their production tailgates, apart from the torsion load case where the twist angle marginally
exceeded the target of 1.6 degrees by 2%. This was considered to be acceptable in the knowledge
that adding overlaps into the final model will increase the torsional performance. The strength
results also comply with Penso’s target of <0.67 for Tsai-Wu Failure Measure and Inter-laminar
shear failure index. It was found that the design was driven mainly by the stiffness requirements,
particularly for the torsion load case, which is a key reason the tailgate outer and inner panel
laminates are biased to plies oriented at 45 degree.
Table 1. Optimised tailgate load case results.
Deflection Deflection Max Tsai-Wu Max Interlaminar
Load case
Target Result Failure Measure Shear FI
Latch Load < 1 mm 0.11 mm 0.14 0.05
Torsion < 1.6 degrees 1.63 degrees 0.36 0.20
Margin & Flushness < 2 mm 1.29 mm 0.53 0.28
Cantilevered Bending < 0.85 mm 0.02 mm 0.24 0.12

4. Design for manufacture
The optimised laminate design was evaluated with Penso’s composite manufacturing team to
develop preliminary ply shapes and overlap joint locations, suitable for manual draping and
pressed composite. A preform tool was manufactured to the preliminary A-surface geometry and
used for layup trials. Due to the complex geometry of the inner and outer panels, each of the all-
over plies was subsequently divided into approximately twelve separate overlapped pieces to
enable draping of the prepreg. The overlap distance between any two plies was typically 10mm,
with the step in laminate thickness located on the B-Surface to ensure a smooth A-Surface. Figure
8 shows an example of a staggered overlap joint used, where the lower ply of an overlap joint is
butt jointed to the upper ply of the previous overlap joint. Consideration of the detailed joint
architecture was vital as the use of butt joints alone does not allow the transfer of loads from one
ply to another. Overlap joints were deemed to be more effective, as this method reduces the
thickness variation across the joint and will simplify the layup during the manufacturing stages.
Figure 8. Staggered overlap joint between three plies.

Once the ply layup and overlap locations had been developed for manufacture, the ply boundaries
documented by the laminator in their draped configuration were sent to Penso’s analysis team
along with information regarding fibre distortion. These ply boundaries, overlap locations and
material directions were updated and modelled in the FE model for further refinement, for
prediction of finalised mass/performance and generation of layup manuals. Material directions
gained from layup trials were compared to that in the analysis model and corrections made where
large inconsistences existed between the two sets of data. Since the parts were to be manufactured
using layup books generated directly from the FE models the laminates needed to be accurately
defined.
The laminate tool within ANSA was utilised for the ease of building, handling and modifying of
all the layers of the composite model, which exploits a special ANSA laminate property. This
ANSA laminate property simplifies the number of layers the user has to manage during pre and
post-processing e.g. the outer panel comprised 1 laminate having 3 layers compared to the 2
composite shell section properties having a total of 5 plies when written out to the
Abaqus/Standard FE solver. As draping simulation was not performed during the development of
the prototype tailgate, the material orientation was defined within ANSA by projecting a vector of
the fore/aft direction onto every element.
Figures 9 and 10 show thickness plots for the updated FE model of the tailgate outer and inner
panels respectively, incorporating the ply boundaries and overlap locations determined by the
manufacturing team.

Figure 9. Thickness plot of final tailgate outer panel including overlap joints.
Figure 10. Thickness plot of final tailgate inner panel including overlap joints.

The previous four load cases were rerun with the tailgate model including overlap joints. The
performance of the structure increased marginally, as shown in Table 2. This is due to the thicker
regions added for the overlaps; though this also made the mass of the parts increase. As classical
laminate shell theory is being used to represent the laminate, the performance is predicted to
increase due to the increase in thickness at these overlap regions. However this does not
accurately represent the joint as, under certain load conditions, it doesn’t factor in the reduction in
strength due to the discontinuity of the fibres at the joints.
Table 2. Final tailgate load case results.
Deflection Deflection Max Tsai-Wu Max Interlaminar
Load case
Target Result Failure Measure Shear FI
Latch Load < 1 mm 0.10 mm 0.12 0.04
Torsion < 1.6 degrees 1.30 degrees 0.25 0.14
Margin & Flushness < 2 mm 1.83 mm 0.48 0.20
Cantilevered Bending < 0.85 mm 0.02 mm 0.21 0.10
Ply shapes of the final tailgate design were transferred to Penso’s design team by outputting ply
boundaries as curves within an IGES file. By taking into account the ply shapes and ply thickness
the B-Surface was generated for the composite parts of the tailgate, by modelling the layup from
the A-Surface. The A and B-surfaces of the composite parts can then be used to create tool
geometry for manufacturing.
4. Conclusion
The main purpose of this project was to show that weight reduction can be achieved by
considering the production tailgate (Tailgate inner, tailgate outer and spoiler mechanism carrier)
manufactured from the existing SMC material into carbon fibre, whilst maintaining similar results
to the targets/baseline. Results gained from FE analysis of the tailgate showed that all strength and
stiffness targets were achieved, whilst having equivalent or better structural performance to the
production design. The total mass of the prototype carbon fibre tailgate panels was 3.04kg, giving
a mass saving of 6.65kg, compared to the equivalent SMC/steel panels. The overall mass saving
of the tailgate assembly was greater than this, due to the use of a new lightweight rear screen glass
and removal of components such as the anti-pinch strips and interior trim.
Further improvements could be made by redesigning the geometry for carbon fibre specific
application and reducing the section depth of the assembly, rather than use carry over geometry
designed for materials with lower strength and stiffness. This would improve optical quality of the
finished composite parts due to less possibility of the material stretching during draping and
reducing the complexity during the pre-forming phase. Also the locations of the ply joints were
positioned to optimise the stiffness, whereas ideally they should be located in areas where they are
less visible to give the best aesthetics. Carry over metallic mounting plates, such as at the hinge,
latch and gas strut, could be replaced by metallic inserts to reduce the assembly part count and
total mass.

Further work has been carried out since; firstly robustness studies have been carried out to produce
a practical method for quantifying the variability of continuous fibre-reinforced composite
structures to uncertainty in ply orientation. Draping simulation software has been evaluated to see
how accurately they predict the ply orientation due to draping. Their use would ensure analysis
models reflect the design intent data and would evaluate ply producibility during the design phase
to avoid any potential manufacturing difficulties downstream. Also on-going work is being carried
out with Penso’s laminators to optimise the tailgate layup to reduce laminating time, yet still
achieve equivalent structure performance.
5. Acknowledgements
This project was co-funded by Innovate UK as part of the IDP7 Low Carbon Vehicles programme,
in which Penso were a lead manufacturing partner.
6. References
1. Arnold, M., Kilby, C. & Ngai, A. (2014). Validation of Laminated Composite Shell Elements
and Material Models within Various FEA Solvers. Proceedings of NAFEMS UK 2014
Conference, Oxford, June 2014.
2. Bednarcyk, B.A., Aboudi, J., Yarrington, P.W., Collier, C.S., (2008). Simplified Shear
Solution for Determination of the Shear Stress Distribution in a Composite Panel from the
Applied Shear Resultant. Proceedings of 49th AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics, and Materials Conference, Schaumburg, IL, April 2008.
3. Cook R.D., Malkus D.S. & Plesha M.E., (1989). Concepts and Applications of Finite Element
Analysis, Third Edition. Chichester: John Wiley & Sons.
4. Niu, M. (1992). Composite Airframe Structures: Hong Kong Conmilit Press Ltd.

Brake System Model Reduction for
Squeal Noise Study
S. Nacivet1, O. Fazio1,2, and J-.J. Sinou2

1 - PSA Peugeot Citroën, Centre technique de La Garenne Colombes, 18 rue des Fauvelles, 92250
La Garenne Colombes, France
2 - Laboratoire de Tribologie et Dynamique des Systèmes, UMR CNRS 5513, Ecole Centrale de
Lyon, 36 avenue Guy de Collongue 69134 Ecully Cedex, France
samuel.nacivet@mpsa.com, olivier.fazio@mpsa.com, jean-jacques.sinou@ec-lyon.fr
Abstract: Although the problem of friction-induced vibration has been the subject of many
investigations over recent decades, it is still responsible for a large number of nuisances in the
field of automotive. Using Abaqus For CATIA V5 (AFC) and Abaqus, this study presents a
numerical process based on modal reduction to generate a Super-Element for the whole brake
system. The size of the model is minimized using a specific non-linear modeling at the frictional
interface. Then, the Super-Element is used in Matlab to check is ability to predict the stability
analysis of large Finite Element Models that correspond to real automotive braking systems
subjected to friction-induced vibrations. The effect on squeal instabilities (i.e the frequencies and
the associated positive real parts) of the number of contact nodes at the frictional interface is
investigated in order to reach a high-quality estimation with the smallest number of DOF possible.
What could be improved in AFC and Abaqus to make the Super-Element more representative and
easier to create is discussed. To finish, the limitation of a determinist point of view is illustrated.
Keywords: Vibration, Friction, Squeal, Reduction, Super-Element.
1. Introduction
The problem of brake squeal in automotive was a subject of great interest for many researchers.
Despite its limitations, a well appreciated method in industry consists to perform a Complex
Eigenvalue Analysis (CEA) on the linearized system in order to predict the squeal propensity of
the brake in a given frequency range (“Sinou, 2010”). For others methods in time (“Vermot des
Roches, 2011” & “Loyer, 2012”) or in frequency domain (“Coudeyras, 2009”), taken

nonlinearities into account, the system size becomes a huge problem since computation time and
data storage can become very expensive.
Therefore, some researchers proposed to develop reduced models when Finite Element Models are
considered in order to reduce the computational times. The most widespread reduction method in
industry is the Craig-Bampton method (“Craig, 1987”). One of the major limitations for brake
system and the squeal prediction is associated with the size of interface matrices due to the explicit
use of the degrees of freedom at interface. Recently, performances of some others reduced bases
built from the component modes or the real coupled modes have been tested by “Brizard, 2011”
for the stability analysis of a disc/pads system in sliding contact. They proposed some enriched
bases to improve the precision on the calculated complex modes and eigenvalues. “Loyer, 2012”
developed also spatial model reduction with different kinds of reduction bases to approximate the
non-linear vibrations of a TGV brake system. In these studies, all the physical contact degrees of
freedom are kept. “Vermot des Roches, 2011” proposed parametrized reduction called the
Component Mode Tuning method (CMT) that use the components free/free modes as explicit
degrees of freedom in order to perform non-linear time simulations. Finally, some researchers
discussed on the possibility to reduce the size of the frictional interface that is one of the main
drawback of the classical use of the Craig-Bampton approach (“Villard, 2012”, “Fazio, 2015”).
In the present paper, the main results of “Fazio, 2015” are shown and current limitations are
discussed. The Super-Element of brake system is evaluated according to CEA results. Notice that
the interest of using CEA is only to validate the Super-Element according to it. Future works will
attach to use the Super-Element with others kind of analyses to take advantage of the Super-
Element size on computation time.
The paper is organized as follows: firstly, the brake system under study is presented. Secondly, the
proposed reduction method and the global strategy for the generation of the reduced non-linear
interface are discussed. More precisely the Super-Element creation and assembly are detailed.
Thirdly, application for an industrial representative Finite Element Model is proposed. The
efficiency of the reduction strategy is undertaken by performing a stability analysis around a non-
linear equilibrium point for two level of reduction at the disc/pads contact interfaces. Finally, the
limitation of a determinist point of view is illustrated by sensitivity observations for another
industrial model.
2. The brake model under study

For this study a Finite Element Model of an industrial brake system that has been developed in
Abaqus For CATIA V5 is considered. It can be decomposed in several parts as detailed in Figure
1-a. This system uses a floating caliper technology. The caliper that holds the two brake’s pads can
move with respect to the disc, along a line parallel to the axis of rotation of the disc. During a
brake operation, hydraulic pressure is applied on the piston, which pushes the inner brake pad until
it makes contact with the disc. Then, reaction force pulls the caliper body with the outer brake pad
against the other side of the disc. Considering the Finite Element Model, a volume mesh is
realized using 10-nodes quadratic tetrahedron (see Figure 1-b). It represents a total of 176956
nodes for 100710 elements.

(a) Exploded view (b) Finite Element Model
Figure 1: Brake system details.

Contact and friction modeling are key parameters for squeal simulation. Here, *CONTACT PAIR
card is used for node to surface contact modeling. The principal components of the brake system
in contact with the pads are the disc, the piston, the caliper and the bracket. For all these contacts,
pads are always defined as slave parts.
A linear penalty method is used for the contact constraint enforcement and the classic Coulomb's
law is used for all the frictional interfaces. Also depending on the contact considered, separation
can be taken into account or not. All these parameters are summed up in Table 1.
Contact Friction coefficient Separation

Disc/Pads 0.5 Yes
Piston/Pad 0.15 No
Bracket/Pads 0.15 Yes
Caliper/Pad 0.15 No
Table 1: Contact modeling
3. Model reduction method

Previous research works were carried out to define a model reduction method to apply stability
analysis on industrial representative models (“Brizard, 2011” & “Vermot des Roches, 2011”).
Alternatively, in continuity of “Coudeyras, 2009” and “Villard, 2012” works, “Fazio, 2015” used
a Super-Element (SE) with a reduced contact interface. Various improvements have been made on
the contact interface reduction step to be more representative.
First of all, it must be noted that the proposed strategy and reduction formulation is not compatible
with the *CONTACT PAIR card used for contact in Abaqus due to the fact that the methodology
involves node-based surface creation which cannot be defined as the master surface of a contact
pair in Abaqus. One solution could be to directly define node to node contact between reduced
nodes of both side directly in Abaqus with *GAP elements with friction. However, this strategy

does not allow us to generate unsymmetrical terms in matrices using *MOTION which is the key
point for stability analysis. So we decided to create reduced contact nodes only for the pads:
coincident nodes on the disc volume mesh are imposed and a node to surface formulation with
*CONTACT PAIR card is kept. The reduced contact nodes and their coincident nodes on disc
mesh will then be used for condensation and node to node contact will be defined later in Matlab
between them.
This section will describe the main steps in detail, but it is interesting for the reader
comprehension to first sum up the whole process. Firstly, reduction points are defined in CATIA
V5 and are used as constraints on disc only to have nodes at these locations. Abaqus for CATIA
V5 is then used to generate the Abaqus input file. Using reduction points information, the Abaqus
input file is modified to introduce reduced contact interfaces at the disc/pads interfaces with
reduced number of nodes on both contacts sides of pads and disc that will be later used for
condensation and node to node contacts. Secondly, the static equilibrium is computed on this
Abaqus model with reduced contact interfaces. Then, deactivating the disc/pads contacts while
fixing the displacement of the nodes of these reduced contact interfaces observed at the static
equilibrium allows generating a Super-Element which is a linearization of this equilibrium
configuration.
Node to node contacts are then used in Matlab at the disc/pads interfaces when the Super-Element
(SE) is used to compute stability analysis with a new cubic contact law for the disc/pads
interfaces. The global strategy is resumed on Figure 2.
Figure 2 : Global strategy

The Abaqus for CATIA V5 (AFC) part is not detailed in this paper, this section is organized as
follows. First the simplified contact modeling will be explained. Then, the Super-Element creation
will be detailed to give some information concerning the Super-Element assembly and more
particularly the method used to define contact force in the node to node contact.
3.1 Simplified contact modeling
As proposed by “Fazio, 2015”, the idea is to reduce the disc/pads contacts defining node to node
contacts with a limited number of nodes which will be later used as reduction nodes for the Super-
Element creation. All the other interfaces are then linearized. This choice avoids adding errors
from the reduction of multiple contact interfaces. In so far as the reduction of the number of nodes
involved in contact must not affect the static solution “Villard, 2012” suggests to distribute the
contact forces around these reduced contact nodes. The *DISTRIBUTING and *COUPLING
options are therefore used in Abaqus. Figure 3 illustrates the contact modeling.

Figure 3: The reduced contact model.
An additional spring with arbitrary stiffness (i.e. the value of this spring is assumed to be very
superior to the local stiffness) allows to link the reference node (which is slave in the distributing
coupling) to the reduction node used for Super-Element creation. For the reader comprehension, it
can be noticed that these nodes have the same geometric position in the model but they are
dissociated on Figure 3 for a better illustration of the reduced contact modeling.
“Fazio, 2015” determined the best set of *DISTRIBUTING and *COUPLING cards parameters to
limit the error induced by the reduction of disc/pads interfaces in the case of an isometric
repartition of reduced contact nodes.
By default, using *DISTRIBUTING and *COUPLING cards to couple a single node to a meshed
surface transfers the forces computed at the reference node to all the nodes of the meshed surface.
An interesting option is to limit the distribution in a spherical region centered on the reference
node by defining a radius of influence. In this case, referring to “Abaqus Users Manuel” the forces
computed at the reference node are distributed only to the nodes of the surface that fall inside this
spherical region. Another interesting parameter is the control over transmission through weight
factors specified at each coupling nodes which allows the forces transferred to the coupling nodes
to vary inversely with the radial distance from the reference node.
For the proposed contact reduction, the radius of influence and weighting method are the main
parameters to be fitted in order to have a smooth variation of contact force over the coupling
nodes.
In Abaqus 6.10 the influence of the mesh size is also taken into account in the weighting factor
computation but it was not considered in this study. To facilitate the understanding of the present
study, the classic polynomials defined below as weighting method will be used
( )
( ) ( )
where is the distance between the reference node and the coupling node, the radius of
influence and the weight factor computed.
As said before, reducing the number of contact nodes and distributing force must not affect the
contact force mapping on the interface. Indeed it is not recommended to concentrate the forces at
the vicinity of the new reduced contact nodes and thus to modify the static equilibrium.
Considering these three weighting methods it is interesting to find for each of them the best radius

of influence in order to ensure an uniform contact force distribution that matches the most with the
initial non-reduced contact. As distribution is made in a circular region centered on the reference
node, it is interesting to define the reduced contact nodes group as an isometric grid to reduce
voids in distribution Figure 4. This isometric grid has been oriented using surface inertia.
Figure 4: Reduced contact nodes layout.

Thus optimal radius will depend on the triangle size of the isometric grid. So the optimum was
calculated for an academic example of a simple triangle of three nodes representing the elementary
pattern of our isometric grid. Weighting coefficient mapping inside this triangle of reduced nodes
is computed in Matlab for a triangle length of . As an equal force on each node of the
triangle is applied, the optimal radius that offers an homogeneous force distribution has to be
found for each weighting method, which means a computed weight factor centered on 1 inside this
triangle. Results are given on Figure 5 and Table 2.
It seems interesting to use a cubic distribution that offers the smallest standard deviation.
Whatever the number of reduced contact nodes and so the size of the triangle chosen later in our
reduction it will be easy to find, proportionally to these previous results, the radius of distribution
that needs to be used. However the edge of contact surface might be subjected to boundary effect
of the distribution.
(a) 1st degree (b) 2nd degree

(c) 3rd degree
Figure 5: Weight factor distribution inside a triangle of three reduced contact nodes for
several interpolation degrees. An equal force is applied on each node.
Interpolation Radius ̅
degree (mm)
1 0.9501 1 0.0574
2 0.7425 0.9999 0.0835
3 0.9597 1 0.0234
Table 2: Weight factor computation results for several interpolation degrees: Optimal
Radius, mean value ( ̅ ) and standard deviation ( )
In the future, it could be interesting to have new capacities on the *DISTRIBUTING and
*COUPLING cards to better distribute the forces on a given surface without forcing a reduction
on an isometric grid.
3.2 Super-Element creation
The Super-Element creation can be summarized as follows: first of all, the reduced disc/pads
contact interfaces (i.e contacts with a reduced number of nodes using *DISTRIBUTING and
*COUPLING cards) are created on the Abaqus model. Then, a non-linear static analysis is
performed on the Abaqus model with reduced contact interface and the displacements at the nodes
retained for Super-Element creation are stored as reduced contact variables (contact force and
gap). In a second step, contacts at disc/pads interfaces are deactivated but the displacements
measured are still applied to the future reduction nodes with the "*Boundary, FIXED" parameter.
In a third step, a frequency analysis is performed on 0-12kHz range. In the last step, a Craig &
Bampton reduction is computed using the *SUBSTRUCTURE GENERATE. The *RETAINED
NODAL DOFS card allows to specify the nodes used in the disc/pads contact interfaces to keep as
reduction nodes. Except disc/pads contacts, all others contacts are linearized in this substructure
generation. The mass and stiffness matrix of the Super-Element are exported by the
*SUBSTRUCTURE MATRIX OUTPUT card. For this card, we regret that options
“STRUCTURAL DAMPING MATRIX” and “VISCOUS DAMPING MATRIX” are not
available. We would be interested by them.

3.3 Super-Element assembly
Concerning the Super-Element assembly, it may be noticed that Abaqus in no more used and all
the computations are done thanks to Matlab software, as illustrated in Figure 2. This subsection
details the method to compute the non-linear static step of the SE assembly at the exact static
equilibrium.
The non-linear static system solved by Abaqus is described as follows:
Equation 1: ( ) ( )
Where corresponds to the static equilibrium position. represents the external force due to
the external pressure applied to the piston and the caliper representing a brake operation. is the
stiffness matrix due to the structural components and the Piston/Pad, Bracket/Pads and
Caliper/Pad contact interfaces. contains the non-linear forces at the frictional interfaces
between the disc and the pads.
As the Super-Element is created in a perturbation step, the system linearized around the static
equilibrium can be written as follows:
Equation 2: ( )
where defines the linearized stiffness matrix at the vicinity of the static equilibrium . All
the contacts, except for the disc/pads interfaces, are linearized as a constant value in the stiffness
matrix.
The local stability is studied introducing a perturbation around the static equilibrium:
Equation 3: ( ) ( )
Considering the Matlab implementation, the linearized system at the static non-linear equilibrium
point is described in the following equation:
Equation 4: ( )
where corresponds to the non-linear contact as the disc/pads interfaces that will be
described in the next section. is obtained thanks to the Super-Element creation in Abaqus.
The only difference with Equation 2 is the way to compute the non-linear force involved in
disc/pads contacts. According to the strategy defined above this contact is now a node to node
contact between condensation nodes. As the calculated linearized equilibrium points for Abaqus
software and Matlab software are the same, an equivalence of the contact force at the disc/pads
interfaces for the non-linear equilibrium point has to be performed:
Equation 5: ( ) ( )
Considering experimental data, “Coudeyras, 2009” chose a cubic stiffness for the disc/pads non-
linear contacts. For each contact node, its formulation is given in Equation 6 and Equation 7:
Equation 6: {
Equation 7:

where and are respectively the linear and the non-linear stiffnesses and and the
displacements of the coincident nodes and from the master and slave sides of the contact
respectively. The subscript indicates that the non-linear forces computation is performed
on the Matlab environment.
The main advantage of this choice is that this formulation is easy to linearize for further stability
analysis. As said earlier we need to ensure the continuity in the contact force estimation between
Abaqus and Matlab software. With two different contact formulations the only way to respect this
constraint is to find an equivalent contact penetration for each node to node contact defined in
Matlab in order to match the normal contact force value obtained with Abaqus software. Knowing
the value of Abaqus normal contact force for each reduced node and the cubic stiffness parameters
and , it is easy to solve the following relation:
( ) ( )
where are the contact forces computed earlier with Abaqus software and the contact
penetration associated to the static equilibrium point that has been previously estimated using
Abaqus software. As said earlier this contact is computed using a linear penalty method in Abaqus
software. Then the corrective term in node to node contact penetration can be easily found.
4. Application for the reduced industrial model

This section presents the application of the reduction strategy defined above to the industrial brake
model. The main objective of this section is to illustrate the efficiency of the proposed
methodology to build a reduced Finite Element Model while keeping a good prediction on the
three main instabilities observed on the reference Abaqus model.
4.1 Reduced models characteristics
To illustrate the effectiveness of the proposed strategy and discuss the impact of the size of the
proposed reduction, several Abaqus models with several triangle size for disc/pads contact
interfaces reduction will be investigated. Table 3 sums up the characteristics of the two models
under consideration.
Abaqus models Abaqus 104 Abaqus 212
Reduced contact nodes per pad 52 106
Total coincident nodes on disc volume mesh 104 212
Total references nodes for SE 210 526
Number of node to node contact elements 104 212
Table 3: Reduced models characteristics.
Figure 6 gives an insight of the reduced contact nodes layout over the pad surface. As point of
comparison, the industrial Abaqus model contains 1370 nodes on the surface of each pad.

(a) 104 nodes (b) 212 nodes
Figure 6: Reduced contact nodes layout over the pad surface for several reduced model.
In the following of the study, both Abaqus models with reduced contact interface will be called
Abaqus104 and Abaqus212 in reference to the number of reduced contact elements they contain.
4.2 Contact state resulting from a non-linear static analysis
As previously explained in “Sinou, 2004”, the complex eigenvalue analysis is based on the
determination of eigenvalues for the linearized system around the non-linear static equilibrium.
The first step to evaluate the performance of the reduced contact interface is to compare some
static quantities between the Abaqus reference and both reduced models (Abaqus104 and
Abaqus212). In this paper, we just report the comparison between both reduced models and the
reference model for the contact state when establishing the non-linear static equilibrium. So, a
non-linear static analysis for a friction coefficient value of (at the disc/pads interfaces) is
computed. Figure 7 shows the results for both reduced models (i.e. Abaqus104 and Abaqus212,
respectively). Contact state at each reduced node is represented by a circle (respectively by a plus
sign) if the contact is open (respectively closed). Moreover, the contact map of the reference
model is superimposed for each reduced models: the green and blue colors represent the closed
and opened contact
As illustrated in Figure 7 (a-b), despite one or two nodes at the boundary between opened and
closed contact zone the numerical, result for Abqus104 is consistent with those of the reference
model. On Figure 7 (c-d), it clearly appears that the reduced model Abaqus212 is in perfect
agreement with the reference model. This illustrates the fact that the number of nodes at the
frictional contact interface can be drastically reduced in order to estimate the contact states by
applying an appropriate strategy.

(a) (b)
(c) (d)
Figure 7: Contact state comparison for each reduced model. The left column is for the inner
pad and the right one is for the outer pad. (a)(b) Abaqus104 (c)(d) Abaqus212.
4.3 Stability analysis on the Super-Element assembly
Now we propose to illustrate the efficiency of the second part of the global strategy based on the
Super-Element assembly.
As shown by “Fazio, 2015”, a Super-Element remains valid only at the vicinity of the non-linear
static equilibrium point. More particularly, varying the friction coefficient at the disc/pads
interfaces changes the static equilibrium on the Abaqus reference model. Then, a Super-Element is
generated for each step of the friction coefficient (i.e. from to with a step size of
). Real modes over 0-12kHz are kept for the Super-Element creation so that its validity can be
assumed over 0-6kHz, our frequency range of interest. It represents a total of 135 modes. Next, a
complex eigenvalue analysis is performed for each friction coefficient by using the associated
Super-Element (SE) assembly. As illustrated in Figure 8, results for SE assemblies of both
frequencies and real parts of the three unstable modes are in agreement with the reference (for
Finite Element Model) even if a sudden rise in frequency of the third unstable mode starting at
can be noticed.
(a) (b) (c)

(d) (e) (f)
Figure 8: Evolution of the frequencies (a,b,c) and the real part (d,e,f) of the eigenvalues
against the friction coefficient for the (a,d) 1.9kHz, (b,e) 3.8kHz and (c,f) 4.7kHz
instabilities. A new Super-Element is generated at each for complex eigenvalue analysis on
SE assembly.
In order to have a more detailed comparison of these SE assemblies (Matlab104 and Matlab212
which are respectively associated to Abaqus104 and Abaqus212), the mean error on the three
instabilities of interest is estimated for all the values of the friction coefficient in the range of
interest (i.e. from to with a step size of ). The mean error is defined by:
| |
Equation 8: ∑ ∑
where corresponds to the value of the friction coefficient given by .

Considering one instability, (respectively is the frequency of the of the two modes
that couple at the friction value for the Abaqus reference model (respectively reduced model).
stands for the number of friction coefficient for which a complex eigenvalue analysis is computed
for the stability analysis. The minimum and maximum errors are also defined by
| |
Equation 9: ∑
| |
Equation 10: ∑
Finally, the mean error (given in Equation 8), the minimum and maximum errors (given in
Equation 9 and Equation 10, respectively) are calculated for the SE assembly. We recall that this
reduced model Abaqus104 was used for the SE generation. Results are given in Table 4 and Table
5. With this strategy the error never exceeds two percent whatever the unstable mode of interest.

In conclusion, these results illustrate the efficiency of the global strategy based on both nodal
interface reduction and Super-Element generation.
Instability number 1 2 3
Matlab 104 0.19% 0.19% 0.88%
Matlab 212 0.19% 0.11% 0.52%
Table 4: Mean error on frequency for each pair of mode on the whole friction range between
SE assembly and the reference model.
Instability number 1 2 3
Matlab 104 0%/1.77% 0.07%/0.36% 0%/1.89%
Matlab 212 0%/1.81% 0%/0.27% 0.21%/1.2%
Table 5: Min/Max error on frequency for each pair of mode between SE assembly and the
reference model.
4.4 Limitation of a determinist point of view
For another model which is not detailed here, we notice that results are not all the time as good as
those obtained previously.
2070 200
2060 150
2050 100
2040 50
Frequency (Hz)
Real part
2030 0
2020 -50
2010 -100
2000 -150
1990 -200
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
 
Figure 9 : Comparison of frequencies and reals parts for an instability and his conjugate
according to the disc/pads contacts modeling. In the case of gaps, the Super-Element is used.
On Figure 9, the evolution of the frequencies and real parts are drawn for different approximations
of disc/pads contacts. The first one, our reference, is a Finite Element Model (FEM) with a contact
pair of type “node to surface” using surfaces of elements to define the slave surfaces. The second
one consists simply to define the slave surfaces by groups of nodes extracted from the surfaces of
elements. It is interesting to notice for this small modification that the effect on ,the

coefficient of friction at Hopf bifurcation, is sensibly affected. The lower frequency of the
conjugates modes is more affected than the other one. It’s only an error of ~0.25%, but its leads to
a delay of about 0.09 on compared to the reference at ~0.51 (~17.5%). This result shows
that some instabilities are sometimes very sensitive to the behavior at the contact interface. Then,
approximations (interpolation and load transfer strategy) can lead to significant differences on
in some situations.
This sensitivity is still more important for the third approximation which is a Finite Element
Model with disc/pads reduced contacts using 172 reduced nodes for pads. A softening effect of
~0.75% on the lower frequency leads to a difference of about 0.37 on compared to the
reference (~72.3%).
Then, for the Super Element assembly with 172 gap elements, which is the fourth approximation,
a hardening effect of ~0.35% on the lower frequency is observed compared to the precedent case.
The softening effect of reduced contact was already observed by “Fazio, 2015”, but not with such
effects on . The same for the hardening effect of Super Element Assembly. Perhaps the
reduced contact could be improved using another strategy to distribute loads on the slave surface,
to decrease edge effects for example. Nevertheless, real surface topology is not as perfect as the
studied model. Then, another way could be to realize sensitivity analysis to compare models
according to parametric uncertainties.
5. Conclusion
The global reduction strategy proposed by “Fazio, 2015”, which is based on a nodal reduction, is
proposed for an industrial non-linear brake system model. One of the advantages of the proposed
strategy is based on an efficient contact reduction method for the disc/pads interfaces that allows a
Craig & Bampton reduction to be used for further stability analysis. But as presented at the end,
the sensitivity on the Hopf bifurcation can be very important in some cases. For these cases, it
could be more appropriate to take uncertainties into account to compare results issue from reduced
models with those from the Finite Element reference.
Based on these results, a future work could be to investigate effects of uncertainties. Another one
will be to use reduced models to compute the self-excited vibrations.
Final number of variables on this reduced model is about 1500 for case with 212 reduced nodes
for pads.
To finish, some evolutions are wished in Abaqus for future investigations: output on damping
matrix for *SUBSTRUCTURE MATRIX OUTPUT and compatibility between *GAP and
*MOTION cards. Moreover, dependent on agreement about PSA Peugeot Citroën patent, it could
be interesting to have tools to build reduced contacts.
6. Acknowledgment
This work was achieved within PSA Peugeot Citroën Stellab program which allows to work with
academics in openlabs, the VAT@Lyon openlab for NVH.

7. References
1. Abaqus Analysis User's Manual, Version 6.10, chapter 31 Constraints, Dassault Systèmes
Simulia Corp., Providence, RI.
2. Brizard, D., Chiello, O., Sinou, J.-J., and Lorang, X., “Performances of some reduced
bases for the stability analysis of a disc/pads system in sliding contact”, Journal of Sound
and Vibration, 330(4):703--720, 2011.
3. Coudeyras, N., Sinou, J.-J., and Nacivet, S., “A new treatment for predicting the self-
excited vibrations of nonlinear systems with frictional interfaces: The constrained
harmonic balance method, with application to disc brake squeal”, Journal of Sound and
Vibration, 319(3):1175--1199, 2009.
4. Craig, R. J., A review of time-domain and frequency domain component mode synthesis
methods, International Journal of Analytical and Experimental Modal Analysis}, 2(2):59-
-72, 1987.
5. Fazio, O., Nacivet, S., Sinou, J.-J., « Reduction strategy for a brake system with local
frictional non-linearities - Application for the prediction of unstable vibration modes”,
Journal of Applied Acoustics, 91:12--24, 2015.
6. Loyer, A., Sinou, J.-J., Chiello, O., and Lorang, X., “Study of nonlinear behaviors and
modal reductions for friction destabilized systems. application to an elastic layer”,
Journal of Sound and Vibration, 331(5):1011--1041, 2012.
7. Sinou, J.-J., Thouverez, F., and Jezequel, L., “Methods to reduce non-linear mechanical
systems for instability computation”. Archives of Computational Methods in
Engineering: State of the Art of Reviews, 11(3):257-344, 2004.
8. Sinou, J.-J., “Transient non-linear dynamic analysis of automotive disc brake squeal - on
the need to consider both stability and non-linear analysis”, Mechanics Research
Communications, 37(1):96--105, 2010.
9. Vermot des Roches, G.,” Frequency and time simulation of squeal instabilities -
Application to the design of industrial automotive brakes”, PhD thesis, Ecole Centrale
Paris, 2011.
10. Villard, P., Nacivet, S., and Sinou, J.-J., “Superelement reduction of industrial finite
element brake system for a constrained harmonic balance method”, INTER-NOISE and
NOISE-CON Congress and Conference Proceedings, pages 3142--3151, 2012.

Application of Coupled Eulerian Lagrangian
Approach in Finite Element Simulation of Friction
Stir Welding
Sanjeev N K*, Ravikiran B P

Robert Bosch Engineering and Business Solutions Private Limited, Bangalore, India
*Sanjeev.NambihallyKumaraswamy@in.bosch.com
Abstract— Finite Element Simulation of Friction Stir Welding (FSW) is a problem involving large
deformations and is often difficult to solve using the classical Finite Element Method (FEM).
Large mesh distortions and contact problems can occur due to the large deformations such that a
convergent solution cannot be achieved. Since in ABAQUS, a Coupled Eulerian Lagrangian
(CEL) approach has been developed to overcome the difficulties with regard to FEM and large
deformation analyses. In this article, this method is investigated regarding its capabilities in
simulating FSW process. FSW is a mechanical process whereby solid-state welding is performed
using heat generated from the friction of a rotating tool and plastic deformation of weld material.
FSW is a modification of the traditional Friction Welding, which was invented at the Welding
Institute in Cambridge, England in 1991. Since then, it is gaining significance in many joining
applications, by overcoming the limitations of other fusion welding processes. For successful
incorporation of its potential during industrial applications, mechanism of joining needs to be
properly comprehended. Various experimenting and study techniques are used in widening of its
applicability in joining process. FE Simulation is one such technique which would help in getting
an insight of the process phenomena during the process and the overall result could be used to
observe the effect of process parameters on weld quality. Here, an attempt is made to develop a
FE model using ABAQUS/Explicit with help of CEL formulation, Johnson-Cook material law and
Coulomb’s Law of friction. The model has been found to be capable of obtaining certain insight
on right processing conditions prior to performing welding process, by predicting the effect of
process parameters on outputs like temperature, force on tool and formation of defects. Finally
the FE Simulation results are used to obtain the optimized process parameters.
Keywords— Friction Stir Welding, Finite Element Simulation, Coupled Eulerian

Lagrangian, Abaqus.

1. Introduction
Figure 1. Schematic representation of friction stir welding process [1, 2]
Friction Stir Welding (FSW) is a solid-state joining process that uses a third body tool to join two
facing surfaces. The welding of surfaces is achieved with the heat generated from the friction of a
rotating tool and plastic deformation of weld material [3, 4]. Two metals that are to be welded
together are held in place against a backing plate using a clamping system. The rotating tool is
then slowly plunged with a downward force into the weld joint. It dwells for a few seconds while
enough heat is generated due to friction that the material to be welded begins to flow around the
tool. Once this point is reached, the tool is traversed along the joint forming the weld behind the
tool as it moves along [5, 6]. The schematic representation of FSW is depicted in Figure 1. The
main benefit of friction stir welding is that the pieces to be welded would not be reaching their
melting points [7, 8].
The FSW process offers significant advantages as compared with fusion joining processes for
aluminum, several of which are particularly important to the aerospace and automotive industry.
These advantages include improved joint efficiency (tensile strength), improved fatigue life, no
need for consumables and improved process robustness. Since from its invention, it has rapidly
evolved and has opened up a variety of research channels. It is being touted as the most significant
development in metal joining in the last decade [3, 4]. Many alloys, including most aerospace Al
alloys (e.g., Al 7xxx) and those regarded as difficult to weld by fusion processes (e.g., Al 2xxx),
may be welded by FSW [9].
2. Objective of the work

FSW presents a multiphysics modeling challenge, because it combines closely coupled heat flow,
plastic deformation at high temperature and microstructure and property evolution. All three
contribute to the process-ability of a material by FSW and to the subsequent properties of the
weld.Figure2 illustrates the key physical interactions involved in FSW process and to the outputs
needed by designers. But the bottle neck experienced in FSW is fixing the right processing
parameters when a certain job arrives on the shop floor, parameters change with thickness of job

and material of job. Here the main object is to get the parameters at which the weld can be
performed for selected material and thickness with in the stipulated time.
Figure2. Summary of interactions in FSW process
3. Approach and Actual Work Done

In literature, the existing method i.e., Experimental-based approach used for obtaining optimal
process parameters leads to wastage of high resources, generation of scrap and off high time
consuming. In order to overcome the problem faced, a methodology is developed and validated.
The methodology is essentially a model-based approach for optimization of FSW process. The
Figure3 presents a schematic representation of steps followed in overall methodology of model-
based optimization of FSW process. In going forward with the objective, the first step was to
develop FE model that can simulate FSW process and validate FE results with referring to various
published research papers and results obtained from experimentation. The model was developed
using ABAQUS-6.12 [10], that has a CEL formulation to simulate problems involving large
deformations and is often difficult to solve using the classical FE Method. In the next step, simple
optimization methods were used with goal of minimizing manufacturing costs and maximizing
weld quality.

Figure3. Methodology of model-based optimization of FSW process
4. Model Description
FE model is developed in ABAQUS/Explicit using the Coupled Eulerian-Lagrangian Formulation,
Johnson-Cook material law and Coulomb’s Law of friction. The Coupled Eulerian-Lagrangian
(CEL) method efforts to capture the strengths of Lagrangian and Eulerian methods. In general, a
Lagrangian reference frame is used to discretize the Tool while an Eulerian frame is used to
discretize the Work-piece [11-14]. Here the workpiece of 200 X 100 mm area and thickness of 5
mm is considered. The Eulerian domain is meshed with multi-material thermally coupled 8-node
(EC3D8RT) Eulerian elements and the void region thickness is taken as 1 mm. The Johnson-Cook
Johnson and Cook [15] equation (1) describes the flow stress as a product of the equivalent strain,
strain rate, temperature dependent terms and several parameters to adequate the real behavior of
the materials.
 m

 A + B ( ε )n  1 + C  ε p   1 −  T − Troom  
σy = (1)
 p
     
 ε o     Tmelt − Troom  
where Tmelt is the melting point or solidus temperature, Troom the ambient temperature, T the
effective temperature, A the yield stress, B the strain factor, n the strain exponent, m the
temperature exponent, εp /ε0 the plastic strain and C the strain rate factor. A, B, C, n, and m are
material/test constants for the Johnson-Cook strain rate dependent yield stress. The material
properties of AA2024-T3, considered for simulations are as per the values taken by Veljic, et al.
[16] and material constants for Johnson-cook are as per Mandal, et al. [17] and Hamilton, et al.
[18].
The tool with shoulder and pin made of High density steel (HDS) material used in experiment is
considered as rigid part and meshed with Lagrangian elements. The Figure4 shows schematic
representation of dimensions of tool geometry.

Figure4. Circular pin shaped frustum tool geometry
In ABAQUS, friction law used in solid mechanics and that suite for FSW modeling is modified
Coulomb friction law [19, 20]. However for FSW the conventional Coulomb friction law will be
only applied at the very beginning of welding when interface temperature is relatively low. As the
interface plasticized material is formed in larger volumes at elevated temperatures, the friction
behavior will be dominated by viscoplastic friction. Therefore, heat generation is dependent on
intense plastic deformation of the thin shear layer at the interface [21]. A modified Coulomb
friction law is then applied. Different values of Coefficient of Friction (COF) have been used in
literature. Tutunchilar, et al. [22] used COF values of 0.4, 0.5 and 0.6, under 100 mm/min
transverse speed and 900 rpm rotational speed. According to investigations made by Kumar, et al.
[23] the COF and temperatures do have a synergic influence on each other. The COF in FSW
condition was found to be as high as 1.2 to 1.4 in temperature range of 400-450°C. Therefore, in
present simulations a friction coefficient of 1.0 is considered for all simulation conditions.
5. Model validation and parametric study

The model is developed referring to aluminium 2024 alloy experimental results. The validation is
done using experimentally measured temperature, force on tool, torque, shape and size of weld
zones and capability of predicting generation of defects. Also further validation of model was
carried out using temperature results and macrographs published by Merzoug, et al. [24] and
Hirasawa, et al. [25]. Here the results are categorized and discussed under sub sections. The
welding parameters that are kept constant for different weld conditions during simulation and
experiment are; Plunge velocity of 10 mm/min, Dwell Time of 10 sec, Plunge depth is 0.2 mm.

5.1 Effect of Coefficient of friction
Figure 5. Effect of coefficient of friction on void size (Top view); (a) At µ= 0.2, (b) At
µ = 0.4, (c) At µ = 0.6, (d) At µ = 0.8, (e) At µ = 1
The simulation results show that the coefficient of friction has a major effect on void formation,
the lower the friction coefficient is applied, larger the void is formed. The Figure 5 shows the
effect of coefficient of friction on void size at rotational speed of 1000 rpm. As the friction
between tool and the workpiece increased the formation of void and moment of material was
closer to that of experimental conditions. It can be seen that any value of µ < 1 resulted in
unrealistic prediction of results. Also considering µ > 1.2 lead to over softening of material,
which intern showed the defect as shown in Figure 6. For a sound weld, it is found from literature
that the working temperature in FSW should be around 80% to 90% of melting temperature
(Tmelt) of the welding material [26, 27]. The Figure 7 indicate that with µ = 1, the maximum
temperature predicted in simulation is in the intended range. In Figure 7 the percentage of error is
calculated by considering the maximum temperature of 404.360C, recorded by thermocouple
during the experiment. The resulted simulation temperature at µ = 1 is in close agreement with
thermocouple reading.

Figure 6. Effect of high coefficient of friction (Top view); (a) At µ = 1.4, (b) At µ = 1.6
Figure 7. Simulation temperature with respect to friction coefficient

5.2 Temperature response
Plunge trials for temperature measurements were carried on AA2024-T3 (5 mm thick plate) with
plunge velocity of 10 mm/min, 00 tilt and 0.2 mm plunge depth. Three trials were carried for tool
rotational speeds of 350 rpm, 950 rpm and 1550 rpm each. The plunge and dwell phase was
approximately of 30 and 10 seconds respectively. To measure the temperature change during
experiments, thermocouples were employed. The mounting of thermocouples and the region in
which the nodes are selected in simulations correspond to similar locations of thermocouples as
illustrated in Figure 8. During plunge, gradual increase in temperature was observed for 350 rpm
unlike 950 and 1550 rpm (see Table 1). It is observed that temperature reaches a steady state in
dwell phase of the weld as shown in graphs (see Figure9). This points to the fact that temperature
has linear relation with plunge depth; it increases with increase in plunge depth and vice-versa. For

a given plunge depth there is a certain maximum temperature that can be obtained and once it is
achieved it remains the same in dwell phase too. The Figure 8(a) shows the close-up view of
temperature profile after plunge and dwell phase on the top sheet. Left column shows the images
from experiments and right column exhibits simulation images with predicted temperature.
Figure 8. Location of thermocouples in experiment and simulation.

Table 1. Comparison between numerically estimated and experimentally measured
temperature
Temperature (0C)
Tool rotation (rpm) Experimental FEM Error %
350 326.64 349.98 7.15
650 369.14 392.56 6.34
950 404.36 432.14 6.87
1250 435.43 454.79 4.45
1550 475.13 489.63 3.05
Average 5.57
Figure9. (a) Temperature measured during plunge, dwell and traverse in

experiment at 1550rpm,
(b) Comparison simulation readings for plunge and dwell phase

5.3 Defect prediction
The developed model is capable of predicting the formation of defects for a set of process
parameters. It gives the close representation of external defects as shown in Figure10 and also
internal defects as shown in Figure11 when compared with experimental images. The results show
that a defect is generated for lower rotational speed and it reduces as the rotational speed
increases. The same effects were observed in the study conducted on Al 6061 sheets by Ramulu, et
al. [28].
0
Figure10. Formation of external defect at 200 rpm with tool tilt angle of 0
0
Figure11. (a) Void Formation at 600 rpm with tool tilt angle of 0 (b) Defect free
0
weld at 1000 rpm with tool tilt angle of 0
The results showed that the increase of the rotational speed and the decrease of the welding speed
can improve the friction stir weld quality, but flash formation would be more obvious when
rotational speed is increased more and decreased welding speed leads to higher weld time. The

model capability of predicting the generation of various defects that accrue in FSW due to various
parameters can also be predicted. The Figure 12 shows comparison of various defect accrued in
physical welding to simulation prediction. The actual size of defect, material and parameters are
not exactly matches with the simulation conditions. Here the images are only compared to show
the capability of model in predicting the generation of various defects by modeling.
Figure 12. Comparison of various defect accrued in physical welding

to simulation prediction
5.4 Simulation time
The time taken for simulation as stated by Lasley [29] is 42 days to run 5.44 seconds of simulation
using supercomputing facility and as stated from Awang [30] it is 14 days and 12 hours on a 3.60
GHz Intel Pentium 4 processor for the simulation time of 1.505 seconds. With further
advancements in FE packages there was as a little reduction in computational time. As stated from
Assidi, et al. [31] it is 12 days 20 hours with ALE formulation and 6 days 14 hours with Eulerian
formulation. But in present work, with the use of CEL the time taken for simulation is greatly
reduced. It is 9.5 hours with core 2 duo processor and 4.5 hours with i7 processor. Also the high
distortion and convergence problems with Lagrangian and ALE formulations are overcome.

6. Process Parameters Optimization
FSW process operation in general can also be optimized by obtaining optimal values for
parameters such as tool rotational speed, axial force, traverse speed, tool dimension, temperature,
stress-strain, strain rate, defects and other parameters such as force on tool and stress in tool as
well as in fixtures. Several optimization techniques could be applied to optimize FSW models.
However, as the FSW is relatively new technology, there have been only a few attempts to use
mathematical optimization techniques to optimize the process. Here, a basic level of optimization
is done to show the capability of the results obtained from FE model for a given application.
(a) (b)
Figure13. (a) Main Effect Plot for means and (b) Main effect plots for S/N ratios

Figure14. Appropriate value of each factor which effect to temperature
(optimization plot)

Figure15. Process parameters optimization procedures: (a) Experimental bases
procedure, (b) FE modeling based procedure
With the data obtained from the simulated model, Taguchi and the Response surface optimization
techniques were carried out using MINITAB [32]. The optimization process is based on defects
observed, welding time taken and optimal temperature range, with tool rotation and weld velocity
as variables. Here, with the results of Taguchi and Response surface optimization techniques
obtained from MINITAB (Figure13 and Figure14), optimized process parameters predicted were
950 rpm (tool rotation) with 60 mm/min (weld velocity). There was a good correlation with
experiment and simulation at these parameters. This shows in a border sense, the capability of
results obtained from FE model in performing optimization. The Figure15 (a) shows the schematic
representation of experimental based procedure of obtaining optimized process parameters and
Figure15 (b) shows the FE modeling based procedure of obtaining optimized process parameters
in FSW. It can be visualized that FE modeling based technique takes lesser time and eliminates
experimentation, material and tooling costs.
7. Challenges
Four major challenges had been addressed while solving the present problem. The first challenge
was to develop a FE model using ABAQUS which closely simulates the actual process. This
involved the selection of proper formulation, material model and boundary conditions. The second
one was to predict the parameters leading to defect and defect free welds. The third one was to
accomplish the objective in minimum possible simulation time to make this approach effective
and efficient and the fourth, to obtain optimal process parameters with simple optimization
method and validate with experimentation.
8. Conclusions
Based on the investigations carried out in the present research and the results obtained, following
conclusions can been made:

(i) Based on the comparison of the simulation and experimental results, under the no slip
condition (µ=1) and Johnson-Cook material model in ABAQUS/Explicit environment, the
proposed model is capable of predicting right processing parameters.
(ii) CEL method exhibited the greater potential of defect prediction in FSW for given process
conditions which ultimately can help in selecting the appropriate process parameters prior
to performing welding process. It also helps in reducing simulation time to a larger extent
thereby making FE simulations of FSW economical.
(iii) Prediction of defects, temperature profile, stress strain profile and mainly optimizing the
process parameters using FE modeling of FSW would greatly help in widening the
applicability of FSW practice.
(iv) The proposed FE modeling method comprehended the possibility of obtaining the
optimized process parameters and presented the strong possibility of widening the effective
optimization study on FSW and its variants.
9. References
[1] ir. Kevin Deplus. (2014). ALUWELD : Innovative welding of aluminium alloys – Hybrid
Laser Welding and Friction Stir Welding.
[2] Sanjeev N.K, Vinayak Malik, and H. Suresh Hebbar, "Effect of Coefficient of Friction in
Finite Element Modeling of Friction Stir Welding and its Importance in Manufacturing
Process Modeling Applications," International Journal of Applied Sciences and
Engineering Research, Vol. 3, No. 4, 2014 vol. 3, pp. 755-762, 2014.
[3] R. S. Mishra and Z. Y. Ma, "Friction stir welding and processing," Materials Science and
Engineering: R: Reports, vol. 50, pp. 1-78, 2005.
[4] Rajiv S. Mishra and Murray W. Mahoney, Friction Stir Welding and Processing: ASM
International, 2007.
[5] K. Kumar and Satish V. Kailas, "On the role of axial load and the effect of interface
position on the tensile strength of a friction stir welded aluminium alloy," Materials &
Design, vol. 29, pp. 791-797, 2008.
[6] K. Kumar, Satish V. Kailas, and T. S. Srivatsan, "Influence of Tool Geometry in Friction
Stir Welding," Materials and Manufacturing Processes, vol. 23, pp. 188-194, 2008.
[7] Vinayak Malik, Sanjeev N K, H. Suresh Hebbar, and Satish V. Kailas, "Time Efficient
Simulations of Plunge and Dwell Phase of FSW and its Significance in FSSW " in
International Conference on Advances in Manufacturing and Materials Engineering,
NITK, Surathkal, 2014.
[8] Sanjeev N K, Vinayak Malik, and H. Suresh Hebbar, "Verification of Johnson-Cook
Material Model Constants 0f AA2024-T3 for use in Finite Element Simulation of Friction
Stir Welding and its Utilization in Severe Plastic Deformation Process Modelling,"
International Journal of Research in Engineering and Technology, vol. 3, pp. 98-102,
2014.
[9] R. K. Uyyuru and Satish V. Kailas, "Numerical Analysis of Friction Stir Welding
Process," Journal of Materials Engineering and Performance, vol. 15, pp. 505-518,
2006.
[10] "ABAQUS Version 6.12, User Documentation, Dassault Systems," 2012.

[11] Kevin H. Brown, Shawn P. Burns, and Mark A. Christon, "Coupled Eulerian-Lagrangian
Methods for Earth Penetrating Weapon Applications," Sand Report, pp. Sand2002-1014,
2002.
[12] Benson D.J, "Computational Methods in Lagrangian and Eulerian Hydrocodes," vol. 99,
pp. 235-394, 1992.
[13] Karim H. Muci-Küchler, Sindhura Kalagara, and William J. Arbegast, "Simulation of a
Refill Friction Stir Spot Welding Process Using a Fully Coupled Thermo-Mechanical
FEM Model," Journal of Manufacturing Science and Engineering, vol. 132, p. 014503,
2010.
[14] User Documentation, "ABAQUS Version 6.12," Dassault Systems, 2012.
[15] Gordon R. Johnson and William H. Cook, "A Constitutive Model and Data for Metals
Subjected to Large Strains, High Strain Rates and High Temperatures, ," Proceedings,
7th International Symposium on Ballistics, Hague, The Netherlands, pp. 541-547, 1983.
[16] Darko M. Veljic, Marko P. Rakin, Milenko M. Perovic, Bojan I. Medjo, Zoran J.
Radakovic, Petar M. Todorovic, et al., "Heat Generation During Plunge Stage In Friction
Stir Welding," Thermal Science, vol. 17, pp. 489-496, 2013.
[17] S. Mandal, J. Rice, and A. A. Elmustafa, "Experimental and numerical investigation of
the plunge stage in friction stir welding," Journal of Materials Processing Technology,
vol. 203, pp. 411-419, 2008.
[18] Robert Hamilton, Donald MacKenzie, and Hongjun Li, "Multi-physics simulation of
friction stir welding process," Engineering Computations, vol. 27, pp. 967-985, 2010.
[19] Olivier Lorrain, Jérôme Serri, Véronique Favier, Hamid Zahrouni, and Mourad El
Hadrouz, "A Contribution To A Critical Review Of Friction Stir Welding Numerical
Simulation," Journal Of Mechanics Of Materials And Structures, vol. 4, pp. 351-370,
2009.
[20] H. Schmidt, J. Hattel, and J. Wert, "An analytical model for the heat generation in friction
stir welding," Modelling and Simulation in Materials Science and Engineering, vol. 12,
pp. 143-157, 2004.
[21] Wenya Li, Zhihan Zhang, Jinglong Li, and Y. J. Chao, "Numerical Analysis of Joint
Temperature Evolution During Friction Stir Welding Based on Sticking Contact,"
Journal of Materials Engineering and Performance, vol. 21, pp. 1849-1856, 2011.
[22] S. Tutunchilar, M. Haghpanahi, M. K. Besharati Givi, P. Asadi, and P. Bahemmat,
"Simulation of material flow in friction stir processing of a cast Al–Si alloy," Materials
& Design, vol. 40, pp. 415-426, 2012.
[23] K. Kumar, C. Kalyan, Satish V. Kailas, and T. S. Srivatsan, "An Investigation of Friction
during Friction Stir Welding of Metallic Materials," Materials and Manufacturing
Processes, vol. 24:4, pp. 438-445, 2009.
[24] Mohamed Merzoug, Mohamed Mazari, Lahcene Berrahal, and Abdellatif Imad,
"Parametric studies of the process of friction spot stir welding of aluminium 6060-T5
alloys," Materials & Design, vol. 31, pp. 3023-3028, 2010.
[25] Shigeki Hirasawa, Harsha Badarinarayan, Kazutaka Okamoto, Toshio Tomimura, and
Tsuyoshi Kawanami, "Analysis of effect of tool geometry on plastic flow during friction
stir spot welding using particle method," Journal of Materials Processing Technology,
vol. 210, pp. 1455-1463, 2010.

[26] Jinwen Qian, Jinglong Li, Fu Sun, Jiangtao Xiong, Fusheng Zhang, and Xin Lin, "An
analytical model to optimize rotation speed and travel speed of friction stir welding for
defect-free joints," Scripta Materialia, vol. 68, pp. 175-178, 2013.
[27] Yuh J. Chao, X. Qi, and W. Tang, "Heat Transfer in Friction Stir Welding—
Experimental and Numerical Studies," Journal of Manufacturing Science and
Engineering, vol. 125, p. 138, 2003.
[28] P. Janaki Ramulu, R. Ganesh Narayanan, Satish V. Kailas, and Jayachandra Reddy,
"Internal defect and process parameter analysis during friction stir welding of Al 6061
sheets," The International Journal of Advanced Manufacturing Technology, vol. 65, pp.
1515-1528, 2012.
[29] Mark Jason Lasley, "A finite element simulation of temperature and material flow in
fricton stir welding," Master of Science, School of Technology, Brigham Young
University, United States, 2005.
[30] Mokhtar Awang, "Simulation of Friction Stir Spot Welding (FSSW) Process: Study of
Friction Phenomena," 2007.
[31] Mohamed Assidi, Lionel Fourment, Simon Guerdoux, and Tracy Nelson, "Friction model
for friction stir welding process simulation: Calibrations from welding experiments,"
International Journal of Machine Tools and Manufacture, vol. 50, pp. 143-155, 2010.
[32] "Minitab 17 Documentation," 2014.

APPENDIX
Johnson-Cook constants and material properties of AA2024-T3
Material constants for Johnson-cook

A (MPa) B (MPa) n C m Tmelt (°C) Tref
369 684 0.73 0.0083 1.7 502 25
Material properties
Temp Density Young’s Yield Thermal Specific heat Coefficient of
(°C) Kgm–3 modulus stress conductivity capacity thermal expansion
(GPa) (MPa) (Wm–1K–1) (Jk–1g°C–1) (10–6 °C–1)
25 2785 73.1 345 175 875 24.7
100 2770 71.5 335 185 897 25.2
200 2750 68.4 315 194 922 26.4
300 2730 64.7 295 191 952 28.7
400 2707 60.2 265 190 987 30.9
500 2683 55.8 125 188 1002 31.6
538 2674 49.3 75 184 1023 32.1
638 2500 42.3 25 85 1054 33.5
FSW Modeling Steps in ABAQUS Environment

FSW Phases as Visualized in Developed FE Model
Simulation results with respect tool rotation and weld speed
Inputs Outputs
Simulation results with
respect tool rotation and Weld speed Welding time Observation
Temperature (°C)
weld speed Tool rotation (mm/min) (sec) (Defect types)
(rpm)
350 20 251.82 180 Visible defect
1250 20 339.66 180 Visible defect
1550 20 356.87 180 Visible defect
950 40 390.56 90 Internal defect
1250 40 411.52 90 Defect Free
1550 40 451.96 90 Defect Free
950 60 425.23 60 Defect Free
1250 60 454.79 60 Defect Free
1550 60 475.41 60 Defect Free
650 80 409.76 45 Defect Free
950 80 452.87 45 Low flash
1250 80 475.41 45 Low flash
1550 80 503.43 45 High flash
650 100 430.21 36 Low flash
950 100 465.19 36 Low flash
1250 100 502.11 36 High flash
1550 100 512.64 36 High flash

Innovative Anisotropic Material Modelling Approach
for Fiber Reinforced Thermoplastics
Recep Yaldiz1 and Gertjan Kloosterman2
1
SABIC Innovative Plastics, 2Dassault Systemes B.V
Abstract: Current industrial state of the art for predictive engineering of fiber reinforced
thermoplastic materials utilizes nonlinear isotropic material modelling and/or anisotropic linear
modelling. Both approaches give inaccurate predictions, since the effect of the orientation and
length of the fibers is not taken into account in the former. Whereas, the latter omits the plastic
behavior typically observed in thermoplastics. An innovative approach is developed in the last
years which takes into consideration the anisotropic nonlinear behavior of the material. This
approach which is based on micromechanical homogenization theories that considers both correct
processing and fiber orientation, can significantly improve the accuracy of mechanical
predictions, but is unfortunately very costly in terms of analysis time. The authors follow the
innovative approach of combining the processing conditions of the fiber reinforced materials with
obtained fiber length and orientation in the end applications, but couple this with a novel material
model which incorporates anisotropic non-linear elasticity, anisotropic hardening plasticity, and
multi-layer anisotropic non-linear visco-elastic behavior which can support anisotropy ratios that
would cause a numerical breakdown when the Hill yield criterion would be employed. Since
thermoplastics have complex fracturing behavior, the material model also incorporates a tailor
made damage initiation/damage evolution process. The influence of the processing conditions
and fiber orientation on mechanical properties will be elaborated, as well as how the model was
calibrated to the large amount of test data available within SABIC.
Keywords: Thermoplastic Materials, Anisotropy.
1. Introduction
The capabilities in terms of material modelling thermoplastic materials have seen a considerable
improvement in recent years with the addition of the Parallel Rheological Framework in Abaqus
(Abaqus User’s Manual, 2014). It contains a plethora of options in terms of capturing the
complex behavior that thermoplastic materials exhibit. It includes the possibility to model non-
linear elastic behavior mix that with several different large strain plasticity algorithms. To address
the viscous behavior of thermoplastics it includes the possibility to add multilayer viscosity, where
each of the layers can exhibit non-linear viscosity. Apart from the complexity which has moved
into the calibration domain it can be considered the state of the art model for the isotropic
modeling of (thermo)plastics.
The limitation on isotropic behavior is the cause of this work, because fiber reinforced
thermoplastic material does not behave in an isotropic fashion necessitating the creation of an
anisotropic variant with equal capabilities.

Short fiber reinforced thermoplastic material has anisotropic properties which are a result of the
injection molding process. The short fibers align themselves with the melt flow into the mold and
the resulting distribution strongly influences the overall properties of the part thus created. This
fact has been long recognized within the plastics industry and has led to the development of
software like Moldflow, Moldex and Simpoe. These numerical solvers simulate the injection
molding process and deliver as a result the fiber orientation tensor at each point in the part. To
perform an accurate FE analysis of the result, this distribution needs to again be represented in the
part to be analyzed. Also this has been long recognized, and has led to the implementation of what
is the Abaqus from Moldflow translator.
However, so far the only material model in Abaqus, which is capable of using different orthotropic
material properties for every integration point is the linear elastic short fiber model, or
alternatively using a distribution table for orthotropic elasticity. This is quite far off from the
capabilities of the Parallel Rheological Framework. So currently a user that wants to model short
fiber reinforced material has two choices: either assume isotropy, and be able to use all the
complex material behaviors in the Parallel Rheological Framework or realize that the orthotropic
behavior is paramount (why else bother with using injection molding software in the first place)
and assume linear elastic behavior.
Now the situation for the linear elastic behavior can be mollified a bit, we could add an anisotropic
plastic behavior model like Hill‘s plasticity. However, what is not often stated is that Hill’s
plasticity is limited in the anisotropy ratio which it allows. If the anisotropy ratios start to deviate
significantly from the isotropic condition, the flow potential is no longer a positive definite
function, with non-convergence as a result.
So again we are limited, we may introduce anisotropic plasticity, but only up to a certain amount
and not in what is typically observed from experiments. And all this is still ignoring the also very
important viscous behavior of the thermoplastic.
Another approach has also come out onto the market which is the field of multi-scale
homogenization methods. These look very promising, particularly the Mori-Tanaka based
schemes. However, this methodology was developed for linear elasticity, and once we introduce
additional mechanical behavior the homogenization schemes become distinctly more involved.
In this paper we set out a new material model that has been added to Abaqus/Standard and
Abaqus/Explicit which tries to address the requirements of modeling orthotropic elastic-plastic-
viscous materials such as short fiber reinforced plastics. First we illustrate the shortcomings of Hill
plasticity in case of large anisotropy ratios. Then we introduce the concept of strain based yield
and equivalent uniaxial strain. We then outline the idea of the material model. The material model
as shown has been conceived and implemented by Dassault Systemes B.V. and calibrated using
material data from SABIC.
To make the model usable, a separate translator was made to interpret Moldflow data for the
material model, and to facilitate ease of use and an input file parser was written so that we can add
user defined keywords.

2. Fiber Reinforced Plastics and Common Current Practice
The introduction notwithstanding, it is still common practice in the automotive industry to use
isotropic material modeling when analyzing the performance of injection molded parts. Although
the assumption of isotropy is already a huge simplification, one still needs material data to
perform the analysis.
It may come as a surprise, that despite this widespread usage of isotropic modeling, that there is no
standardized way on how the isotropic material data is to be obtained for reinforced plastics. Thus,
depending on the measurement of the data, either obtained from a datasheet, or testing from
directly molded specimens, or specimens cut from parts, one obtains different values for stiffness
and strength. One of the prime reasons for this are the differences in fiber orientation.
The European Alliance for Thermoplastic Composites (EATC) has devised a method to properly
measure isotropic properties from specimen cut from injection molded plaques (Schijve, W and
Rüegg, 2008). Samples are cut in three directions (0°, 45°and 90°) with respect to flow front
direction. Subsequently classical laminate theory is used to calculate isotropic properties based on
a 0°, 45°, 90°, -45° stacking. For an illustration see Figure 1
Figure 1. Schematic overview of plaque and samples used in EATC method for
determining isotropic material properties
It is common practice to use “datasheet” values of materials for isotropic modeling. These values
are usually generated with directly molded tensile specimens, which exhibit high fiber alignment.
By consequence, the strength and stiffness of the material are significantly higher compared to
what is typically observed in actual applications. Designs based on these values typically
encounter failure below the expected values.
Doing isotropic analysis with the EATC values are on the other hand usually conservative
compared to reality, which in turn yields parts that are heavier than they need to be.
It should then be clear, that going the isotropic route is not the most ideal approach, and we should
consider attempting to do an anisotropic analysis of injection molded parts. To see an illustration
of typical value ranges consult Figure 2.

Figure 2. Mechanical performance of front-end part molded in 40% PP-LGF
TM
(SABIC®STAMAX 40YM240)
3. Strain Based Yield

The go to model once orthotropic plasticity is shown to be a factor is Hill’s plasticity or one of its
derivations. Before we can discuss the concept of Hill’s plasticity we need some terminology in
place. We denote by 𝝈 the stress, and 𝜺 the strain as usual. We assume that an orthonormal basis
for our space of interest is in place, so that vectors can be written as 𝒙 = 𝑥1 𝒆1 + 𝑥2 𝒆2 + 𝑥3 𝒆3 . In
that case 𝑬1 = 𝒆1 ⊗ 𝒆1 , 𝑬2 = 𝒆2 ⊗ 𝒆2 , 𝑬3 = 𝒆3 ⊗ 𝒆3 , 𝑬4 = 𝒆1 ⊗ 𝒆2 + 𝒆𝟐 ⊗ 𝒆1 , 𝑬5 = 𝒆1 ⊗
𝒆3 + 𝒆3 ⊗ 𝒆1 and 𝑬6 = 𝒆2 ⊗ 𝒆3 + 𝒆3 ⊗ 𝒆2 form a basis for the symmetric tensors, so that we
can for example write 𝝈 = ∑6𝑖=1 𝜎𝑖 𝑬𝑖 . Writing then 𝝈 = �𝜎𝑥𝑥 , 𝜎𝑦𝑦 , 𝜎𝑧𝑧 , 𝜎𝑥𝑦 , 𝜎𝑥𝑧 , 𝜎𝑦𝑧 � is then termed
Voigt notation, which is merely interpreting the symmetric tensors as a vector space and choosing
a particular basis. In Abaqus the basis for the strains differs from the basis for the stress in that the
last 3 base tensors are halved. In this way we can think about the stress as a ‘vector’ as being a
member of a vector space.
Fourth order tensors can be thought of as linear operators mapping a second order tensor into
another second order tensor. Once we have selected a basis for the second order tensors, every
fourth order tensor thus takes a matrix representation We will write fourth order tensors with
calligraphic letters, for example 𝓟.
3.1 Hill’s plasticity
Hill’s yield criterion (Hill, 1948) can then be written as
√𝝈: 𝓟: 𝝈 − 𝜎𝑦 ≤ 0,
Where 𝜎𝑦 is the equivalent uniaxial yield stress, and 𝓟 is the fourth order projection tensor. Now
𝜎
�
take 𝐴𝑖𝑗 = , which are the inverses of the relative yield ratios for each direction. In the basis
𝜎𝑦,𝑖𝑗
given as above, we can write this fourth order tensor as

2
2𝐴11 𝐴233 − 𝐴11
2
− 𝐴222 𝐴222 − 𝐴11
2
− 𝐴233 0 0 0
⎡ 2 2 2
⎤
⎢𝐴33 − 𝐴11 − 𝐴22 2𝐴222 2
𝐴11 − 𝐴222 − 𝐴233 0 0 0 ⎥
1 ⎢𝐴 − 𝐴11 − 𝐴233
2 2
𝐴11 − 𝐴222 − 𝐴233
2
2𝐴233 0 0 0 ⎥
𝓟 = ⎢ 22
3 0 0 0 2𝐴244 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 0 2𝐴255 0 ⎥
⎣ 0 0 0 0 0 2𝐴266 ⎦
Now assume we have large anisotropy, in which case we could have relative yield ratios of 0.1, 1
and 10. If you fill in these numbers in the above, we find a negative eigenvalue in the matrix
representation. This means that there is a stress 𝝈 so that 𝝈: 𝓟: 𝝈 is negative, no matter how small
we take that stress.
We can thus conclude that for large anisotropy ratios, the Hill criterion breaks down.
Unfortunately for short fiber reinforced plastics, we usually are dealing with large anisotropy
ratios. From this we can conclude that yet again we are left with either choosing a clearly incorrect
isotropic approximation or limiting the anisotropy ratios.
It is clear that neither is desirable. We are now going to try and move around this limitation from
moving our yield potential from stress space to strain space.
3.2 Poisson Free Strain
To see the equivalence, let us start with the Mises yield criterion. In this case we have similarly to
the Hill criterion that
√𝝈: 𝓦: 𝝈 − 𝜎𝑦 ≤ 0,
but here we have
2 −1 −1 0 0 0
⎡−1 2 −1 0 0 0⎤
1 ⎢−1 −1 −1 0 0 0⎥
𝓦= ⎢ ⎥,
2⎢ 0 0 0 3 0 0⎥
⎢0 0 0 0 3 0⎥
⎣0 0 0 0 0 3⎦
now to move from a stress based yield law to a strain based yield law, we simply multiply our
1
flow potential by , which gives us
𝐸
�𝝈/𝐸: 𝓦: 𝝈/𝐸 − 𝜎𝑦 /𝐸 ≤ 0,
where we will coin the term yield strain as 𝜖𝑦 = 𝜎𝑦 /𝐸. For the isotropic case let us investigate
what 𝝈/𝐸 brings us. Now 𝝈 = 𝓓: 𝜺. Thus 𝝈/𝐸 = (𝓓/𝐸): 𝜺. So if we now introduce 𝓤 = 𝓓/𝐸,
then we can introduce the Poisson free strain as 𝜺� = 𝓤: 𝜺. In which case our yield function
becomes
√𝜺: 𝓡: 𝜺 − 𝜖𝑦 ≤ 0,
𝑇
Where 𝓡 = 𝓤 : 𝓦: 𝓤 which is now fully posed in strain notation. We also note that this
formulation is identical to the isotropic stress based one.

For an orthotropic material we typically have 𝐸1 , 𝐸2 , 𝐸3 , 𝐺12 , 𝐺13 , 𝐺23 , 𝜈12 , 𝜈13 and 𝜈23 . We are
going to first construct an 𝐸4 , 𝐸5 and 𝐸6 . The construction is based on taking an average Poisson
ratio. We give the result for 𝐸4 and assume the rest follows similarly. For this stiffness we have.
𝐸4 = 𝐺12 �2 + (𝜈12 + 𝜈21 )�.
Note that the definition is given such that in the isotropic case all the stiffnesses are identical. We
now use the exact same logic as before, where we introduce an Poisson free strain by 𝜺� = 𝓤: 𝜺,
but now we define 𝓤 = 𝓔−1 : 𝓓, where 𝓔 is basically the fourth order tensor where in matrix
representation the diagonal consists of the stiffness terms 𝐸𝑖 . Note that for the isotropic case this
just reverts to the definition we gave before. We will now not change the yield strain, but rather
leave it at the form that we found, thus our strain flow potential looks like:
Φ(𝜺) = √𝜺: 𝓡: 𝜺 − 𝜖𝑦 ≤ 0.
Note that this has some very interesting consequences, because we now have that the yield point in
the first direction in terms of stress is 𝐸1 𝜖𝑦 . And similarly the yield points in terms of stress in the
second and third directions are 𝐸2 𝜖𝑦 and 𝐸3 𝜖𝑦 . The strain yield potential is also positive semi-
definite in terms of the strain tensor, so we have now circumvented the Hill degeneration problem.
We term √𝜺: 𝓡: 𝜺 the equivalent uniaxial Poisson free strain.
However, we now have that the yield points are directly related to the stress. We could again work
around this by making a similar assumption as Hill did, by allowing the weighting tensor to take
Hill’s form, but it turns out that isotropic strain as a predictor in yield point turns out to be a very
good predictor, so for the materials at hand we have chosen not yet to do this.
There is a simple motivation for this, and that is that it is the matrix which is responsible for the
yielding behavior. If we assume homogeneity of the continuum, and make the assumption that the
matrix and the fiber see approximately the same strain, then it is the matrix that deforms in a
plastic and/or viscous manner and not the fiber, which only deforms elastically. So we observe
yield at just about the same level of strain all the time. The contribution to the stress however is
wildly different, which is mostly caused by the orientation of the reinforcements. This observation
is given credence by looking at experimental results.
3.3 Strain Based Generalized Maxwell Model
At this point, we can take the idea of Poisson free strain and equivalent uniaxial Poisson free strain
and unite it into a nonlinear version of the generalized Maxwell model. The idea of which is
illustrated in Figure 3.

Figure 3. Generalized Maxwell Model (picture courtesy of Abaqus User’s manual)
This is conceptually the same concept as was employed for the PRF, which is the reason we can
use the same figure again. We are going to assume an arbitrary amount of layers, where the zeroth
layer is called the equilibrium layer. It consists of a spring in series with a friction element which
denotes the plasticity behavior. On top of that we can have several different layers which model
the viscous behavior. The current version of our implementation of the model several viscosity
laws (one of them which degenerates to linear viscoelasticity when the coefficients are chosen
correctly) All these laws are implemented in a strain based manner, so they all behave in an
orthotropic fashion. On top of this we also allow non-linear elasticity.
The complete material model thus described is implemented in a UMAT and a VUMAT for both
shells and solid elements.
4. Setting up the Material Model with Injection Molded Data

We have now seen what the material model entails, but we still need to feed the results of the
injection molding analysis into it. We have only written a utility that uses the data from a
Moldflow analysis on a shell mesh into Abaqus, but other formats could also easily be
incorporated.
So for simplicity we can assume that abaqus moldflow was run as a tool. When the translator
was run on a shell mesh, this will present us with two files:
1. The shf file, which contains the orthotropic data for a certain number of section points
through the thickness of a shell element. Each section point may have a distinct set of
orthotropic material coefficients, and have a distinct material orientation.
2. The str file, which contains the initial stresses due to thermal shrinkage.
Both these files are read by a small tool, which reads the data and writes it into a binary file, which
takes the extension sdv. The tool is written in C++. Essentially the sdv file contains the data which
is given in both files, but takes a more convenient directly digestible format to be used by the
subroutine.

The idea is to put the orthotropic material parameters, not as material constants, but as state
variables for the material. Even though they are not really state variables in the sense that they
would change as a function of deformation, they are still local to each integration point. We cannot
use a distribution with a user defined material for properties, and a field would be not ideal either,
since we would need to give nodal properties. We could of course circumvent this again by
implementing a USDFLD, but that would still entail the same amount of memory as is required for
defining them as state variables.
The initialization is then performed through implementation of the subroutines SIGINI and
SDVINI. We also use an ORIENT subroutine to set up the integration point orientations from 6.14
onwards.
The reason we do not also use a *Initial Condition, type=solution is because these values
can only be given on an element basis, and are then uniformly applied throughout the thickness.
Where it should be clear that the distribution through the thickness for a fiber distribution is
typically not constant.
Apart from these procedures we also implement the user subroutine UEXTERNALDB to control
the initialization process. The total analysis goes as follows.
• At the start of the Abaqus run, UEXTERNALDB is called, which opens the sdv file that
was created and reads the contents into memory. If there is no sdv file present, the routine
assumes that the data is given through initial conditions statements in the input file.
• At the beginning of the first increment SIGINI and SDVINI as well as ORIENT read the
data which was read into memory and places those in the appropriate state variables in
the Abaqus data structure.
• At the end of the first increment UEXTERNALDB is called again, which deallocates the
allocated memory. At this point initialization is complete, and the UMAT has access to
all the data.
This implementation was made prior to availability of VUEXTERNALDB, so initialization is
done through Abaqus/Standard, after which an import needs to be made.
The next issue is dealing with specifying all the constants that need to go into the material model.
Abaqus has a strict rule of specifying the material constants with 8 variables per data line. If we
want to change the material data slightly, this would mean all the data would need to be
rearranged.
To this end, we created a small script which can parse arbitrary input files, and write them out
again. The utility will create a hierarchical dictionary of all the keywords encountered in the input
file. Once an input file is in the internal data structure, it can also write it out again.
We will now introduce fictitious keywords. These fictitious keywords can then be removed from
the dictionary through a small add-on function, which will then add a *User Material card and
also a *Depvar card with the correct data lines. Subsequently calling the write methodology will
then create a valid input file for execution.

5. Results
5.1 Uniaxial tensile behavior
The model has been validated using simple 1-element tests for the material fitting, and dog bone
specimens to see its behavior in multiple element tests.
SABIC has provided all the required material data for performing a rate dependent fit. The result
of which is illustrated in Figure 4.
Figure 4. Results of Fitting Material Model
The fit for the rate dependency is done on the equivalent isotropic data. It is then assumed that the
anisotropic results follow the same behavior, but the actual scale of the curve will scale with the
stiffness in the respective direction.
Required Fiber orientation as input for anisotropic simulations has been be obtained from
commercially available injection molding simulation software Autodesk Moldflow Insight.

5.2 Real application example
The accuracy of isotropic, linear anisotropic and non-linear anisotropic simulation methods (see
Figure 5) have been evaluated by simulating a lock stiffness experiment on an existing frond-end
module carrier made of 40% long-glass fiber filled Polypropylene.
Fig. 5: Two main anisotropic modelling methods.
The carrier was subjected to a 2 kN load at room temperature and the resultant displacement was
measured, Figure 6. It was observed that the lock stiffness was under predicted by ~20% when
EATC isotropic values were used. When linear elastic anisotropic material data from Molflow
were directly applied, predicted stiffness was ~30% too high, while the results, obtained with
developed non-linear elastic plastic anisotropic model, were very close to the experiments (-4%)
(See Figure 6). It should be noted however, that for direct Moldflow-Abaqus simulation only
linear integrated triangular elements could be used. Other predictions could be done using
quadratic triangular elements. This will influence the results to some extent.
Figure 6. Simulated versus measured lock stiffness on an existing front-end

module made of 40% PP-LGF (SABIC®STAMAX™ 34YM240). Left: schematic setup
of lock stiffness test. Triangles indicate constraints, a 2 kN force was applied in

lock direction (arrow). Right: Predicted stiffness. Isotropic properties according to
EATC method.
6. Conclusions
Accurate prediction of mechanical performance and dimensional stability of the part is the key for
reaching the desired light weight design and minimized development time.
Current industrial state of the art for predictive engineering of fiber reinforced thermoplastic
materials relies in most cases on isotropic material modeling, where fiber orientation is not taken
into account. This approach gives contradictive predictions.
Anisotropic performance prediction, which considers fiber orientation in the end applications
delivers the best accuracy in performance prediction, which is only a few percent off from reality.
However, the anisotropic predictive engineering approach using multi scale homogenization
methods require significantly more computational resources compared to the current state of the
art approach which is isotropic modelling. Analyses take longer time and in general more memory
is required.
Sitting inbetween the simplified isotropic modeling, and the homogenization methodology is the
phenomenological framework we have presented in this paper that yields significantly improved
results compared to the isotropic modeling, but does not have the high analysis cost required for
the full scale homogenization approach.
7. References
1. Hill, R. “A Theory of the Yielding and Plastic Flow of Anisotropic Materials” Proc. Roy.
Soc. London A193 1948.
2. Schijve, W and Rüegg. “EATC Properties and Test Methods for LFT Materials”, Report of
European Alliance for Thermoplastic Composites, http://eatc-
online.org/media/EATC_Properties_and_test_methods_for_LFT_version_30jan2008.pdf,
2008.

Topology Optimization of a Motorcycle Swing
Arm Under Service Loads using Abaqus and
Tosca
Roberto Saponelli, Massimo Damasio

Protesa Spa (SACMI Group)&UNIMORE, Exemplar srl
In research and development environment and concept design, people involved in new projects
often need to design a completely new shape for the structure target of the analysis. The loading
conditions and constraints are usually known but the designer hardly knows how to create the
geometry of the structure that meets the requirements of the project and that can be manufactured
at the same time respecting the target costs.
The weight is known to be one of the main factors that impacts on the performance and on the
costs, and for this reason its reduction often becomes the main objective to achieve. Often
geometries are realized based on experience or on similarity to previous projects, but this
approach can be long and expansive, without the certainty of being able to reach the optimum.
The use of the CAE (Computer Aided Engineering) is strategic in this environment, in order to
verify and virtually simulate different types of geometries avoiding the cost and time of
prototyping and testing. A FEM model is the starting point for the optimization process and
provides, where appropriate, the subsequent integration of the results with different physical
disciplines, such as CFD or Heat Transfer.
In this paper is presented the topology optimization of a motorcycle swing arm, starting from a
design space which takes into account the overall dimensions, and trying to minimize the mass
while maximizing stiffness. The use of Tosca Structure, that drives the FEM simulation of the
swing arm performed with Abaqus, allowed to obtain a shape that can be manufactured in
compliance with the requirements of the project.
Keywords: Topology optimization, Nonlinear Analysis, Motorcycle, Swing arm
1. Introduction
SACMI IMOLA is the Italian parent Company of an international Group that includes over 80
Companies. Over the years, the core business of designing and building machines and complete
plants for the ceramic industry has been flanked by other businesses, to form an efficient
organization that joins Companies and technologies in the following sectors: Ceramics, Beverage,
Food e Plastics.
PROTESA is a Company of SACMI Group and offers technological and organizational services in
support of business processes. Its dynamic workforce of 100 consultants and technicians combines

top-level professionalism with experience in the intensive use of modern technologies, offering
full support for the entire manufacturing process to its customers.
In research and development environment and concept design, people involved in new projects
often need to design a completely new shape for the structure target of the analysis. The loading
conditions and constraints are usually known but the designer hardly knows how to create the
geometry of the structure that meets the requirements of the project and that can be manufactured
at the same time respecting the target costs. In this scenario CAE and Topology Optimization
fulfill all the requirements described. In particular, if used before a real prototype, CAE
technologies help to:
• analyse the full design cycle with fast steps
• understand the dominant constraints of the project
• keep all disciplines connected, multi physically, as the real-world complexity requires
• get a preliminary solution that will help technicians to understand in which way they have
to realize the engineering shape of the component
This paper describes the simulation steps involved in the CAE process, focused on increasing the
performance of a racing motorcycle swing arm taking into account different crucial aspects
(structural, weight, casting) and making them converge together. At first, a topological
optimization has been performed using Tosca Structure with Abaqus. Then, the optimized shape
was submitted for a quick casting analysis.
2. Motorcycle swing arm component
The motorcycle swing arm is the back of the same, which connects the rear wheel to the frame,
and that through the use of one or more suspensions controls the movement of the frame relative
to the tire and via a damper damps and slows the oscillation.
It is known that under racing conditions the motorcycle swing arm is one of the most important
components for the configuration of the bike and its performance. In the following figure a sample
of real motorcycle swing arm is shown.
Figure 1. Sample of motorcycle swing arm in production

3. The Topology Optimization Process
Topology optimization is an automated methodology that uses software able to generate a design
proposal and it is often used for finding new concept shapes.
In addition, a number of manufacturing constraints can be applied so that the design proposal can
be produced with casting technology, molding technology, etc. Concerning constraints, Tosca is
able to take into account several boundaries and in particular in this work, casting constraints,
member size constraints, freezing, symmetry and coupling constraints have been defined through
Tosca Strucutre using Abaqus as FEA solver. The process that has been followed for the activity
described in this paper, is resumed in figure 2.
Figure 2. Topology Optimization Process using Tosca Structure and Abaqus

(picture based on Tosca Structure documentation[1])

Concerning the topology optimization of the motorcycle swing arm, first of all a finite element
model of the design space has been created inside Abaqus/CAE, and element sets to be passed to
Tosca has been defined inside the graphical interface. Figure 3 shows the model representing the
initial design space in which the swing arm geometry has to be identified by Tosca Structure and
the element sets created.
Figure 3. Abaqus Finite Element Model of the Optimization Design space used by
Tosca
In Table 1 the material used in the simulation and the dimension of the design space in terms of
elements and nodes are listed.
Table 1. Finite element model characteristics.

Material Al6000
Number of
1169311
elements
Number of
1640840
nodes
Volume 0.03 m3
To get a suitable shape of the swing arm, it is necessary to define constraints those take into
account forging restriction (material pull direction), frozen area of the design space domain that

cannot be modified and the service loads to which the component is subjected. In Figure 4 the
defined frozen areas (red) and the material pull direction are shown.
Figure 4. Frozen areas definition (red zones) and material pull direction definition
in Tosca Analysis
Figure 5, 6, 7 show the three load cases used in the analysis. Those load cases must be fulfilled by
the topology optimization. The final shape of the swing arm in fact, must have the maximum
stiffness with the minimum mass, in compliance with all load cases corresponding to service
loads.
Figure 5. Load Case 1: Service loads acting on swing arm due to passenger load.

Figure 6. Load Case 2: Service loads acting on swing arm – torsional load
Figure 7. Load Case 3: Service loads acting on swing arm – lateral bending load
The Tosca optimization task set up is to find a shape of the swing arm with the maximum stiffness
for the component with a volume or weight restriction. This represents the most common standard
optimization task for the topology optimization. The value to be optimized is the compliance
which is used as a measure of the stiffness.
The compliance is represented as the sum of the strain energies if the different load scenarios of
the complete model. Here, this value has to be minimized. The constraint is the weight or volume
constraint which is defined to be 30% of the initial volume/weight of the structure (the available
design space).
As manufacturing constraint a casting/forging constraint has to be defined. The idea of the
constraint is to ensure that the created structure of the topology optimization has no undercuts and
can be remodeled (or removed from the forging die).
In Figure 8 the Tosca Structure panel on which the Casting Design Variable Constraint has been
set is shown. Topology Link to control symmetry of the final shape has been added as
optimization constraint.
Moreover, to control the maximum and minimum thickness of the final shape, a constraint of
maximum member size (16 mm) and minimum thickness (10 mm) has also been defined as shown
in Figure 9.

Figure 8. Casting Design Variable Constraint in Tosca Structure
Figure 9. Maximum Member Size Constraint in Tosca Structure

4. Optimization Results
Considering the load cases and the topology constraints defined, after 15 iterations Tosca
Structure obtains an optimized shape that is in compliance with the optimization task. The final
weight of the optimized swing arm is of 4.7 kg and the shape is shown in Figure 10.
Figure 10. Swing arm optimized shape – Tosca Report View
As result of topology optimization, Tosca shows an “eroded” shape coming from the deactivation
of elements that, during the optimization, have assigned a “zero stiffness”.
This kind of shape can also be smoothed with Tosca Structure.smooth to get a shape suitable for
detailed Finite Element analysis or for the final detailed design.
The smoothed geometry of the optimized shape is shown in Figure 11.
A final Abaqus stress analysis has been performed on the smoothed structure to verify the stress
level on the optimized swing arm. The stress contour map is shown in Figure 12.

Figure 11. Swing arm optimized smoothed shape – Tosca Smooth View
Figure 12. Swing arm optimized smoothed shape – Stress Contour Map

5. Conclusion
The proposed research aims to highlight the great potential that provides the topology optimization
for the development of systems to be used in high performance mechanical components.
The benefits using Tosca Structure consists in saving about 30% of the time to market, obtaining
shape that meets the requirements of the project and that can be manufactured at the same time
respecting the target costs.
Moreover, the technology used in this case skips the CAD creation, using a technology, called
Sensable Virtual Clay, able to get a polygonal geometric feasible part directly from STL coming
from FEM, using a sort of “virtual sculptor technique” as shown in the following figure.
Figure 13. Polygonal clay part obtained from FEM
The polygonal part can be used directly to virtually evaluate casting and to produce the
component. The virtual evaluation of casting has already be performed, as shown in figure 14. The
core is realized in polystyrene with addictive manufacturing. The component is then produced
with quick casting (lost foam technique). In the future the production time will be further reduced
using rapid prototyping on aluminum (aluminum sintering procedure), skipping the addictive
manufacturing phase to create the core and the lost foam casting.
Figure 14. Casting virtual evaluation

Future development: after the realization of the prototype with the shape obtained with Tosca
Structure based on static loads, we planned to go on with experimental experience of vibrational
dynamic monitoring on track of a rear swing arm with FBG fiber optical sensors. With dynamic
load monitoring we will evaluate if the prototype will need a further loop of optimization from the
dynamic point of view. In concept phase dynamic loads were not available, and we planned to
measure them on the real prototype.
6. References
1. Tosca Structure Documentation 8.0, Dassault Systémes Simulia Corp., Providence, RI.

Consideration of Manufacturing Pre-history Influence of
Fatigue Life of Sheet Metal Formed Automotive Parts
Dmitry V. Buzlaev1, Viktor V.Kalaev2, Dmitry Y. Zheleznyakov2, Dr. A.Winkler3
1
TESIS, Moscow, Russia
2
GAZ Group United Engineering Center, Nizhny Novgorod, Russia
3
Dassault Systemes
Abstract: This contribution considers the effects of sheet metal forming processes on the fatigue
performance of LCV wheel discs, manufactured from microalloyed steel. Traditionally, such parts
had to pass standard cyclic load test in the process of experimental validation, but road tests in
real life conditions sometimes show fatigue cracks in highly stressed areas, predominantly in
bending loading mode.
To estimate fatigue life and safety margins standard fatigue tests should be accompanied by
numerical simulations based on FEA results. Unfortunately, in spite of the long history of fatigue
analysis codes, based on FEA results, the industry still relies heavily on tests because of frequent
disagreements between simulation and test results.
In most cases stress analyst does not take into account the manufacturing pre-history of the part,
its thinning, plastic strains and residual stresses accumulated in a part as a result of sheet metal
forming process used for its manufacturing. This study tries to fill the gap between sheet metal
forming simulation tools, like Stampack, SIMULIA Abaqus as stress analysis tool and SIMULIA
fe-safe as a fatigue analysis tool. It show significant difference between the original simplified
model of wheel center disk, which used virgin material and did not take into account real disk
geometry and residual stresses and hardening, and the model which used results from the sheet
metal forming simulation as a starting point.
Keywords: Bolt Loading, Wheel Disk, Fatigue Life, Deep Drawing, Metal Forming
1. Introduction
Early implementation of automotive components simulation in the development cycle requires
tight integration of various departments and organizations, including design, manufacturing,
analysis and test departments. Reliability of results and reduction of development time can be
achieved either through multi-disciplinary simulations based on unique model or the possibility of
utilizing seamless model conversion between different codes. Unfortunately, at the moment there
is a certain gap between metal forming simulation and strength and fatigue analysis. Meanwhile,
the distinct influence of sheet metal forming processes on the fatigue cracks development at the
areas experienced significant thinning and stretching during plastic deformation (typical for deep

drawing processes) has been observed and documented among GAZ Group subsidiaries and in
open sources [4,5]. This influence can often be rather negative, while in some cases positive or no
effect is observed. As a positive influence, hardening caused by plastic work should be mentioned,
which increases component durability in the case of low cycle fatigue. One particular case where
the positive effect is obvious is shot peen forming, but this process is used in aerospace and not in
the automotive industry. For the typical deep drawing processes used in production of car
components, a stretching stress state is more characteristic which usually decreases fatigue
strength of a part. Also, for thick sheet parts bending can also be a problem, causing early
appearance of fatigue surface cracks in the stretched layers of components. In this work we
evaluate the influence of a manufacturing pre-history on the durability of the central portion of
light commercial vehicle wheel disk.
2. Simulation of fatigue testing

The United Engineering Center of GAZ Group set up a task of simulating the fatigue strength of a
wheel disk in standard bending with a rotation test according to Russian Standard GOST R50511.
The standard requires that the number of cycles before fatigue cracks should not be less than
600 000. The disk is fixed on the test rig using hub part and 6 studs. A rotating load equal to 2744
N is applied through a 1 m long console beam, attached to the hub (fig.1). The disk was made
from the Russian steel alloy 07 GBU – a microalloyed hot rolled high strength steel produced by
Severstal. Its closest European equivalent is S315MC. The sheet had a thickness of 6.5 mm and a
yield strength of 360 MPa.
2744 N
Figure 1. Model of wheel disk fatigue testing

The model was created in Abaqus/CAE with simulation execution using the Abaqus/Standard
solver. General Static was used as a solution procedure. Initially a simplified FE-model was used,
which included only the central flat section of the central disk (fig.2). The outer contour of the
disk was fixed, the load applied through a console connector beam and its rotation was simulated
by X and Y load components with associated amplitude curves. The disk itself was modelled with

5 layers of solid elements C3D8R. The material definition included elastic and plastic properties.
The plastic properties included the hardening curve, which was defined up to 0.8 equivalent plastic
strain. Studs, nuts and hub were also represented by C3D8R. The entire testing process was split
up into three load steps: studs tightening, console load application, console load rotation. At the
beginning of the studs tightening, only the disk component was fixed in space, the other
components were loose, as the contact conditions had not yet been established. This caused some
convergence problems in the first variants of the model. These issues were overcome by applying
general contact with contact stabilization (using default parameters). Studs tightening was
simulated using the standard procedure for bolt loads controlled by force magnitude (fig.3).
Figure 3. Studs vertical displacements after tightening

The use of solid elements for all components in the assembly was chosen to properly represent the
stress field in the disk caused by bending with rotation taking the studs tightening into account.
The stress field at each increment of rotation of the console load was used later for fatigue
analysis. The load rotation step considered a 360 degrees full rotation cycle and was subdivided by
5 degree increments.
Abaqus results are presented in fig.4: maximum displacement in the disk is equal to 0.29 mm, the
maximum principal stress over time is equal to 359 MPa. The area with the highest stress level is
concentrated around the stud holes. It correlates well with real fatigue cracks in the disk observed
in testing (fig.5). Fig. 6 represents the history of the maximum principal stress at the critical
location.

Figure 4. Displacements and Max. Principal Stresses at the end of
the cycle
Figure 5. Fatigue crack after testing near stud hole

Studs tightening Console loading Rotational loading
Figure 6. History of maximum principal stress in the critical point

Then, durability of the disk was analyzed in SIMULIA fe-safe. This system can directly use odb-
file from Abaqus/Standard containing the stress field at each increment of the rotating load for a
full cycle. This cycle was defined in fe-safe as a load block. In addition, a new material file for
07GBU was created, containing elastic-plastic properties and and endurance curves (fig.7).
Figure 7. E-N and S-N curves

For durability assessment of this HSLA steel combined strain Brown-Miller criterion with
Morrow mean stress correction was used. The analysis identified an element with a lowest life
equal to 141 905 cycles, close to the stud hole (fig.8). The factor of strength (FOS) for 600 000
cycles was equal to 0.8. FOS is the factor which, when applied to either the loading, or to the
elastic stresses in the finite element model, will produce the required target life at the node [1].

Figure 8. Factor of strength for 600000 cycles (left) and durability
(right)
The critical location corresponds well to the area of typical fatigue cracks seen on this disk. The
disks made from Severstal steel usually passed standard fatigue test, with cracks appearing only
after road tests (fig.5 shows failed disk made from steel from another supplier). Therefore,
simulation results based on the simplified disk FE-model were deemed to be unrealistic as they
showed too low fatigue strength of the part. A decision was taken to simulate the complete central
disk portion and take into account its manufacturing pre-history, including the thinning
distribution, hardening and residual stresses, estimating their influence on durability. In spite of
the generally negative effect of cold sheet metal forming on durability, some redistribution of
stresses over the part was expected due to lower stiffness of the whole component in comparison
with merely its central flat area.
3. Simulation of stamping process

The GAZ Group uses AutoForm software for sheet metal forming simulation. Nevertheless, it was
not directly used in this work. AutoForm uses shell element for the simulation, while Abaqus
model used hexahedral solid element. This element was used, in particular, to adequately take into
account through-thickness stresses, caused by studs tightening and further rotating bending load.
Unfortunately, the stamping process could not be directly simulated in Abaqus/Explicit due to
trimming and piercing operations between two forming operations. Such operation types requires
not only deletion of elements in scrap and holes areas, but also FE-mesh remeshing to guarantee
smooth hole and blank boundaries. As Abaqus does not have this functionality, Stampack (by
Quantech ATZ, Spain) was used (fig.9).

Figure 9. Disk mesh in Stampack before trimming/piercing (left) and
after (right).
Stampack is a specialized software for simulation of cold sheet metal forming, using an explicit
method of solving. The newest Stampack 7.0.1 has a possibility to use hexahedral solid element
for thick sheets and simulate trimming and piercing operations with particular remeshing of
hexahedral mesh in the areas of cut lines. The disk stamping process can be divided into three
main steps: drawing, trimming/piercing and restrike/forming. In the second step, the central hole
and stud holes are pierced, and scrap is trimmed out. The third step is needed to deform the part to
its final shape. After that step a springback of the part should be simulated to obtain residual
stresses and the true final geometry of the central disk, with consideration of all local thinning and
thickening of areas. The initial blank and the process sequence are shown in fig.10.

Figure 10. Process sequence in Stampack
The final mesh has 4 element layers through the thickness and reasonably good elements shape
quality which can be used in Abaqus simulations. Very simple text editing of the Stampack export
file with results, including coordinates and topology of the deformed mesh helped to create an
input file for the disk part to be instantiated in the Abaqus assembly (fig.11).
Figure 11. Export of deformed disk mesh from Stampack to Abaqus

model
Export of the hardening and residual stress tensor required some programming efforts, because
Stampack can export only stresses and strains as nodal results, while Abaqus allows definition of
initial state variables at centroids or integration points. An extra program, created as a result,

converted the element by element nodal results to elemental values using averaging at centroids.
The program generated input files for Abaqus with equivalent plastic strains and residual stress
fields, which were used as initial states of HARDENING and STRESS types (fig.12).
Figure 12. Mapping of hardening from Stampack on Abaqus model

Such conversion and averaging of stresses can potentially cause unbalanced stress condition in the
Abaqus model. To resolve this issue an additional balancing process step had to be added.
Introducing this step lead to another problem: the step had to be the first in the model, as it deals
with the initial model state. On the other hand, the studs tightening step would also need to come
first, at least when the model is defined in Abaqus/CAE. An attempt to combine residual stresses
with stud tightening in a single first step resulted in unpredictable behavior of the model.
Therefore, the problem had to be resolved on the level of manual modification of the input file in
the following way: the first load step was balancing and the second was stud tightening (it’s
possible to do it in an input file, but not in Abaqus/CAE). The third step was the console load
application and the fourth – console load rotation, at which stresses were exported in the odb-file
for the later fe-safe usage.
3 different cases were simulated:
1. The disk was defined by the deformed mesh from Stampack and considered true part
geometry after metal forming process with thinning and thickening. No INITIAL STATE
condition was used.
2. The disk was defined by the deformed mesh from Stampack. Hardening was used as
INITIAL STATE condition.
3. The disk was defined by the deformed mesh from Stampack. Hardening and residual
stresses were used as INITIAL STATE conditions.
The results from the Abaqus simulations were processed in fe-safe in the same way like it was
done for the simplified model. All three cases are summarized in Table 1.

Table 1. Results of fatigue analysis considering manufacturing pre-history.
Case Max value of Maximum Principal Cycles to failure Factor of strength

Stress at the end of rotation cycle for 600000 cycles
(MPa)
1 255 1 359 000 1.08
2 313 1 264 000 1.07
3 362 1 021 000 1.06
The results show the expected tendency of reduction in the number of cycles to failure with
consideration of the hardening and also, residual stresses. On the other hand, the influence of
residual stresses is relatively low. Evidently, this can be related to the fact that residual stresses
and plastic strains in critical areas, caused by metal forming process, are relatively low. The center
of the disk is flat and during the drawing operation makes contact with the corresponding flat
surface of the punch from the beginning. Frictional forces between the blank and the punch
restricted stretching of the central flat portion. It also did not experience any bending. Further, one
can observe significant increase of the number of cycles to failure in comparison with the initial
simplified model. This was caused by the redistribution of stresses from the center of the part, as a
result of increased flexibility.
4. Conclusions
A multidisciplinary simulation of an automotive stamped part was considered in this work from its
stamping simulation through to fatigue test simulation. Results demonstrate the importance of
detailed modeling and consideration of all factors for the reliability of the results. An integrated
environment for such a simulation chain is important and unfortunately not yet implemented well
in commercial codes. In spite of the fact that manufacturing pre-history analyzed in this work was
not of major significance for the fatigue life of the wheel disk, the GAZ Group has other parts,
where such analysis is both important and crucial. This work constitutes a good start for the
development of standardized methods and their implementation into company processes. In
addition, a further potential continuation of this project is a joint effort with a steel maker to
identify an optimal alloy composition for the steel, combining extended durability and good
formability, which can be contradictory requirements. Shell models should be regarded as a
perspective which allows improved utilization of AutoForm results, as a standard code in the
company.
5. References
1. fe-safe 6 User Manual, Dassault Systemes Simulia Corp., Providence, RI

2. Abaqus 6.14 Analysis User’s Guide, Dassault Systemes Simulia Corp., Providence, RI
3. Simulation Guide Stampack_Incremental 7.0.1, Quantech ATZ, Barcelona, Spain

4. P.J. Mc Grath, D.G. Hattingh, M.N. James and A. Els-Botes, “Effects of forming process
on fatigue performance of wheel centre disks”, 2013
5. Guschin A.N., Pachurin G.V., “Performance properties of deformed mild steel sheet”,
Fundamental Research, Vol.11, 2013
6. P.J. Mc Grath, “An investigation of residual stresses induced by forming process on the
fatigue resistance of automotive wheels”, 2001

Development of a Multifunctional Truck
Cross Member Made of Continuous Fiber
Reinforced Plastics
H. Kirchberger*, S. Fischmeister**, H. Dannbauer**, K. Puchner**
*TEUFELBERGER Composite Ges.m.b.H., Austria; **Magna Powertrain, Engineering Center

Steyr, Austria
Address for Correspondence:

Engineering Center Steyr GmbH & Co KG,
Steyrer Strasse 32,
4300 St. Valentin, Lower Austria, Austria
E-mail: stefan.fischmeister@ecs.steyr.com
Abstract: Lightweight design of trucks enables to increase the payload and represents a way to
reduce the fuel consumption per transported weight. Many engineers around the world are
focusing on this topic. One idea to handle this issue is to combine several vehicle parts or
functions in an integrated module.
This contribution shows the development of a cross member made of a combination of continuous
fiber reinforced plastic and steel. Beside the function as an important structure of the truck main
frame, the part represents a liquid/CNG reservoir at the same time. The management of design
issues, the assessment regarding stiffness and strength was the major focus of the project.
By means of FE analysis, performed with Abaqus the structure was optimized regarding
manufacturing possibilities in combination with the other essential properties like stiffness and
strength. Many activities were started to find an appropriate combination of the plastic part and
the steel brackets, which enable the fixing within the truck main frame structure. Due to the fact
that the function integrated structure is much stiffer (resulting from the bigger size and cylindrical
shape) than the original cross member, the specific decoupling of the truck main frame and the
new developed module was challenging. After solving the main design problems the development
team focused on the optimized layout of the plies with respect to stiffness and durability. Finally
the verification of the analysis results will be done regarding stiffness and fatigue results at a
frame test rig.
Keywords: Composites, Stiffness & Strength, Optimization, Commercial Vehicles

1. Introduction
Composites are already used in a wide range of technical fields: aviation, wind turbines, passenger
cars …. The composite material is used there already for system relevant parts which are high
loaded. With the increasing importance of composite parts driven by light weight ideas and
statutory fuel consumptions, new production methods and improved matrix systems were
innovated in the last years. Also the production capacity, units per year, increased during new
developments. Commercial vehicle developers follow the trend and use continuous fiber
reinforced plastics for chassis parts.
This paper will give a practical overview about the development process which was undertaken for
a commercial vehicle cross member (CM). Starting with a method for assessing how suitable the
chosen part is for a continuous fiber reinforced plastic (CFRP) design. Followed by the modelling
and assessing of the results using Abaqus as FE solver.
As production method for the CM composite braiding was chosen. The individual plies generated
by this production method include carbon and/or glass fiber yarns, which are interlaced with each
other, comparable to woven lamina. In the analyses strength and stiffness reduction due to the
waviness of interlacing fibers, which depends on the braiding angle, should be considered. Finally
the optimum braiding angle, depending on the loading of the structure and on the material strength
and stiffness reduction due to waviness, should be determined.
2. CFRP Cross Member structure
Reinforcement
Calotte
Liner
Braiding
90° Winding
Figure 1: Cross member cross section

The composite CM has the following structure.
• Liner: This is the most inner part. His function is to make the vessel airtight. The
material is high-density polyethylene (HDPE). During the production the outline of this
volume has an additional function. It is used as core for the 90° winding and for the
braiding process.
• 90° Winding: This layer carries the main pressure load which acts at the inside. The
operation pressure for a compressed natural gas (CNG) tank is 200 bar. The burst
pressure is 450 bar. The material is a CFRP.
• Steel Calotte: At both ends a steel spherical cap takes over the operation pressure.
• Braiding: The longitudinal force from the two calottes and the forces from the cross
member function are taken over by these layers. The material is CFRP.
• Reinforcement: At both ends of the CM the braiding is reinforced with a steel plate. The
reinforcements are form-closed connected with the braiding using metal pins (T-IGEL®
concept from company Teufelberger Composite), see Figure 6. As for conventional steel
cross members, the construction is mounted to the side rails via gussets. During the
production the outline of the reinforcement is the core for the braiding process
3. Layout
Whether a part is suitable for a CFRP design and which fibers and matrix material should be
chosen depends on several conditions. Some of the criteria are:
• Environmental conditions: area of operations, temperature range, weight
• Manufacturer influence: production method, costs, availability
• Mechanical condition: stiffness, type of loading
For an analyst, the first focus should be the mechanical condition depending on the peculiarity of
the composite material. The mechanical properties of FRP are anisotropic. The stiffness and the
strength of a unidirectional (UD) layer depend on the orientation of the fibers which are embedded
in the matrix. Depending on the type of material the fibers exhibit strength and stiffness values
which are certainly higher than for conventional steel, see Table 1.

Material Strength [N/mm²] Young Modulus [N/mm²] Density [g/cm³]
Steel 400 – 1000 210 000 7.85
Aluminum 100 – 450 70 000 2.7
CFRP R‖+: 1450 E‖: 139 360 1.7

(HT fiber, Epoxide, ϕ=0.6) R⊥+: 32 E⊥: 8 800
Carbon fiber R‖+: 4530 – 2450 E‖: 230 000 – 400 000 1.74
(HT fiber – HM fiber)
+
Table 1: Material Data, HT fiber … High Tenacity Fiber, HM fiber … High Modulus fiber, R‖ … strength
parallel to fiber tension, R⊥+ … strength normal to fiber tension
The composite material is optimal used when the direction of the maximum principal stress is
equivalent to the fiber direction. In the work of K. Durst 2008, a method is explained how parts
can systematically assessed weather they are suitable for a CFRP design or not. Based on this
work an article in the NAFEMS Magazine with further extensions is issued (Klein 2014). The
base is that in an early phase of the project the potential of a part, of different designs, should be
assessed considering the specialty of anisotropy of composite material. Therefor a simple method
is to generate a FE- model with an isotropic material. Looking at the maximum and minimum
principal stresses for each load case gives not only a first impression about the main orientation of
the fibers (maximum principle direction) it also show whether the part is suitable for a composite
material design or not (variation of principal stress direction).
At the chassis of a commercial vehicle the usage of CFRP material isn’t very common till now.
Nevertheless for commercial vehicle cabs, short fiber reinforced plastic is already used by MAN
Truck & Bus AG (H. Häberle 2013). The article contains further examples where composite
materials are used. Successful examples for a rigid beam axle or a cross members (trailer world
2012) exist also on the market. That was also a stimulus for us to start this investigation on a
braided cross member with an integrated CNG vessel function.
4. FE- Modelling
At first generating a mesh of a composite part is as easy as generating a shell mesh for a structural
steel sheet part. Furthermore the element orientation needs to be uniform and the laminate
property is filled with the composition of the plies.

Abaqus offers the possibility to use different modelling technics.
1. A shell element plus a composite property
2. A continuum shell element with a composite property
3. A continuum shell element (stacked) with a usual shell property (anisotropic material is
oriented via element coordination system)
4. A solid element with a composite property
5. A solid element (stacked) with a usual solid property (anisotropic material is oriented via
element coordination system)
If the classical laminate theory (CLT) should be used and only the in plane stresses of each ply
should be evaluated the first option fits well. If delamination or transverse shear stresses should be
assessed, a more detailed FE- model with stacked elements is necessary.
For the CM a shell FE- model with composite property is used. But deviant from the conventional
used reference surface the mesh isn’t placed at the mid surface. The shell mesh reference surface
is defined at the bottom side with an offset, see Figure 2. This type of modelling has an advantage
when the laminate thickness isn’t constant due to local reinforcements. The reference surface is
flat and the mid surface in the natural axis is handled by Abaqus.
Single Ply layer Surface normal direction Mid surface reference
Bottom surface reference
Property 1 | Property 2 | Property 1

Figure 2: Shell mesh mid surface vs. bottom surface modelling
The following option is used to define an offset:
**
*SHELL SECTION, ELSET=set_comp, COMPOSITE, ORIENTATION=rect_lam,
OFFSET=SNEG
0.30, , composit_material, 0., layer_01
0.30, , composit_material, 45., layer_02

5. Material Properties Definition
The result of the braiding process is a fabric that follows exactly the shape of the mandrel. The
yarn orientation is determined by braiding machine parameters and by the mandrel shape. Two
different types of fabrics are possible, a tri-axial fabric which is used for the CM or a bi-axial
fabric. The tri-axial fabric consists of braided yarns and yarns in CM longitudinal axis.
Material data for 3 different braided tri-axial fabrics and for the processed components (yarn and
matrix) are available. The angle between the braided yarns in this case is 30°, 45° and 60°. The
data include stiffness and strength values in axial yarn direction and perpendicular to the axial
yarn.
Modelling the laminate with a composite shell section the tri-axial fabric is split in its
unidirectional (UD) yarn components instead of using the averaged engineering characteristics.
E.g. the symmetric structure for the 30° yarn angle fabric is (30, 0, -30, 0, 30). The center layer is
twice as thick as the other layers. The reason for this proceeding is that here a more detailed
assessment of the fiber and the matrix is possible. For optimization purpose material data for in-
between braiding angles are also required. With the available data an approximation of the
stiffness should be done, considering the different yarn content of the three yarn directions and of
the waviness depending on the yarn angle. Depending on the used braiding machine, the bobbin
number is fixed and the packing density depends on the yarn angle and the fiber deposition width.
The following procedure was developed to fit the analyzed materials to the test data. From the
yarn and matrix data the UD stiffness is calculated (H. Schürmann 2007). At the test specimen all
three yarns have the same raw material (=UD material). But depending on the braiding angle the
braiding yarns are longer than the axial yarn, see Figure 3. This circumstance is considered with an
adjustment of the UD ply thickness.
1 1
t a : t −Θ : t + Θ = 1 : : Equation 1
cos(Θ ) cos(Θ )
ta ... UD axial yarn ply thickness

t − Θ , t + Θ ... UD braider yarn ply thickness
Θ ... yarn angle
Figure 3: Length difference between axial and braider yarn and relation of UD thickness
The braiding yarn is wounded around the axial yarn. The waviness influence is considered by the
waviness factor which reduces the original UD stiffness. It is assumed that the axial yarn has no
waviness. The braiding yarn waviness factor is optimized in such a way that the engineering
characteristic of the composite fits the test values. The process is displayed in a flow chart, see
Figure 4.

Figure 4: Process for adjusting the FE model stiffness with test data
Figure 5 shows the normalized stiffness’s for the 3 test data in comparison with the analyzed one.
With the described process and only two parameters it was possible to get a good match. The
gained waviness factor must not be seen as an absolute figure. Because also the 45° the braider
yarns are interlaced too. But the fabric is less distorted at 45° than for 30° or 60°. This is cause by
the symmetrical decrease of the waviness factor for angles differing form 45°. The waviness factor
for reducing the braider yarns is nearly 1 for 45° and around 0.75 for 15° degree deviation from
the 45°. The adjusted composite shell property has the following appearance. The thickness of the
UD plies is weighted with the UD thickness influence. The axial yarn UD ply material is the
original material and the braided yarn UD plies material have the original material with a modified
E‖ stiffness.
Figure 5: The diagram shows a comparison between test and analysis standardized young modulus
values for the three braiding yarn angles. Additional the waviness factor for the braiding yarn is shown.
Ex … Young modulus x direction, Ey … Young modulus y direction

6. Optimization of the Pined Surface
For the connection between metal reinforcement and braided composite a tie contact is chosen.
But for a detail analysis the interaction there is investigated more detailed. The tie contact is
converted into a contact pair. Additional the pins are realized with beams which are connected to
the shell mesh on both sides with distributing couplings.
area with
removed pins
Figure 6: T-IGEL® pins at the FE- model and at the prototype parts
The detail analyses shows that the interrupted pin rows close to the calotte are higher loaded than
the rest of the pins. When these pins are removed the local stresses at the reinforcement decrease
significantly and at the braided composite slightly.
a b c
Figure 7: Results with and without optimized pin area, (a) pin shear force distribution for frame torsion
load, Von Misses stress distribution without (b) and with (c) pin area optimization

7. Stiffness Analysis
The stiffness analysis was first hand performed with an idealized setup where three loads are
applied to determine the stiffness. A bending, a shear and a torsion load are investigated.
Reference values from standard CM profiles are also determined, see Table 2.
a b c
Figure 8: Conventional steel reference cross member, (a) light truck alligator CM,
(b) double-C CM, (c) tube CM
Bending [Nm/°] Shear [N/mm] Torsion [Nm/°]

Light truck alligator CM 4500 240 920
Double-C CM 36000 4300 635
Tube CM 7300 1280 4930
CFRP CM (incl. gussets) 17300 2070 4300
Table 2: Stiffness values for reference steel and CFRP CM
The torsional stiffness of the CFRP CM is very high. The large diameter plays a main role in this
case. A soft gusset to achieve a smooth transition between tank and side rail is necessary.
This configuration is used to analyze different versions of the CFRP CM and the results are
compared. Two previous versions are displayed in Figure 9. In design (a) the mounting to the side
rail is similar to the steel tube CM version. Basically the size of a premounted gusset is limited by
the braiding machine. On the other side welding after the braiding process is challenging. Because
the heat from the welding process will damages the composite material. The advantage of this
solution is that a conventional bolt hole pattern can be used. The second shown design (b) has no
diameter reduction across the CM length. Therefore the calotte has to be larger, which increases
the total weight. The bolts are not accessible when the CM is mounted and the bolt hole pattern
isn’t conform with any usual side rail hole pattern.

a b
Figure 9: Previous CM versions, (a) CM with a round reinforcement which is welded to a mounting plate,
(b) CM with large diameter reinforcement equal to the mandrel and bolts inside the circumference
8. Strength Analysis
For the strength analysis the CFRP CM is integrated in a commercial truck frame. To guarantee
that the CM is high loaded, the add on parts: fuel tank, battery box and muffler are arranged very
close to the CM, see Figure 10. The FE- model shown is a 4x2 tractor truck. The suspension is
modeled with an auxiliary beam structure representing the correct stiffness.
Fuel Tank
Muffler Battery Box

Figure 10: Truck frame for strength analysis. The add on parts surrounding the CM are named.
The analyzed load cases are misuse load cases (MLC):
• MLC 1, vertical bending: The add on parts mounted at the main rails burden the CM
with a bending moment. The frame itself is symmetrical and the global bending has no
influence on the CM.
• MLC 2, lateral force: A curve driving maneuver is simulated with this load case. The
side bending of the main rails causes a shear loading of the CM.
• MLC 3, twist: Driving over an obstacle or over a slope, the frame is burden with a twist.
This maneuver is most serious for the frame and CM. The CM has a combined load of
bending and of a torsion moment. The stiffness of the frame and of the truck suspension

cooperates. A torsion stiffness change of the frame will lead to a higher deflection of the
suspension.
• MLC 4, internal pressure 450 bar: The operating pressure of the CNG tank is 200bar.
For safety reason the tank has to resist a burst pressure of 450 bar.
The assessment was done according to the VDI regulation (VDI- Regulation Part 3, 2014). The
regulation favored the Puck criterion which is mainly used for UD plies. The main focus is the
fiber fracture (FF) stretch factor. The inter fiber fracture (IFF) stretch factor is secondary, because
of the used braiding process no distinct matrix direction exist in one ply. Some post processor
tools have implemented subroutines which determine the stretch factors or the stress exposure
values (the stretch factor is the reciprocal value of the stress exposure). For Abaqus is a subroutine
available called VDI2014_v2, which is a free software from Tobias Kremer (Kremer, 2009).
9. Results
The critical point of the CM is the transition from the round cross section to the quadratic. At first
hand the cross section is here smaller and for the torsion load the stresses are increased.
Furthermore the connection to the gussets on only two sides concentrates the load flow further.
The minimum fiber fracture stretch factor is above 3.5, see Figure 11. The braiding yarn angle has
only a minimal influence at the fiber fracture stretch factor.
Figure 11: Minimum fiber fracture stretch factor for the torsion load case.
The IFF stretch factor is below 1. But for a fabric the Puck criterion isn’t ideal. The following tests
will show where the static and the fatigue limits are. A comparison between analysis and test will
be the base for second optimisation loop.

a b c
Figure 12: Minimum inter fiber fracture stretch factor for the torsion load case for (a) 30°, (b) 45° and (c)
60° braiding yarn angle.
Further the exact orientation of the yarn especailly at the transitons to the metal reinforcement
wasn’t clear till the first prototyp parts were produced.
Figure 13: One of the first prototype without gussets.
The burst preassure misuse load case has a uniform stretch factor distribution. The inter fiber
fracture stretch factor at the braiding part has a minium where the calotte ends and the 90° winding
starts, see Figure 11.
braiding braiding 90° winding

part: FF part: IFF part: IFF
Figure 14: Minimum stretch factor for the misuse load case burst pressure 450 bar at braiding and 90°
winding part. FF … fiber fracture, IFF inter fiber fracture

10. Conclusions
The development of a CFRP product with software tools available at the market is possible.
During the project runtime, software implementation for assessing and displaying composite
results were steadily improved and developed. The software companies recognized the trend, to
light parts made out CFRP material and supporting the engineers with new implemented features
for this topic.
6 CFRP CM were produced and 5 of them will be tested in the servo hydraulic laboratory of the
Engineering Center Steyr. The stiffness and the strength of the CM should be compared with the
analyzed results.
11. References
1. H. Häberle, “Plastics-Based lightweight construction in commercial vehicles”, VDI

Wissensforum 2013
2. trailer world, “Revolution am Rohr”, trailer world, Ausgabe Zwei 2012, p24-27
3. trailer world, “Schlanke Linie”, trailer world, Ausgabe Zwei 2012, p32-35
4. K. Durst, “Beitrag zur Systematischen Bewertung der Eignung anisotroper
Faserverbundwerkstoffe im Fahrzeugbau”, CUVILLIER Verlag, Göttingen 2008
5. D. Klein, “Bewertung und Optimierung der Faserverbundeingung von Leichtbaustrukturen in
den frühen Phasen der Produktentwicklung”, NAFEMS Magazin 4/214, p55-66
6. H. Schürmann, “Konstruieren mit Faser-Kunststoff-Verbunden”, Springer Verlag, 2007
7. VDI-Richtlinien, “Development of FRP componets (fibre-reinforced plastics) – Analysis”,
VDI 2014 Part 3
8. ABAQUS 6.13 Online Documentation, Dassault Systèmes, 2013.
9. T. Kremer, KLuB-VDI 2014 v2.0, Technische Universität Darmstadt, 2007
12. Acknowledgements
The authors would like to express their gratitude to Mr. Aigner who worked for his master thesis
on this project and delivered the main work. This was a success full cooperation between ECS and
Teufelberger.

Modeling the Draping of NCF Composite Preforms
Muhammad-Ali Siddiqui, Colmar Wocke
The Dow Chemical Company, Bachtobelstrasse 3, CH-8810 Horgen, Switzerland
Abstract: Non-Crimp Fabrics (NCF) are widely used textile reinforcements in fibre reinforced
composite structures. The most important prediction from a draping analysis featuring preforms
from these materials is their resulting fibre orientations; the orientations influence several other
properties including fabric permeability and final component strength. The analysis is
complicated by the influence of localized stitching patterns and their constraint on relative sliding
of individual layers within a single ply during the draping process. The modeling procedure
involves a process of material characterization valid for fabric preform materials and has been
carried out using the Abaqus/Explicit finite element analysis code. The work is concerned with the
development of such a predictive model using a rebar-approach in Abaqus. Biaxial and triaxial
NCF are modeled using multiple shells. The anisotropic properties of the preform are naturally
represented by the model. The details and capabilities of the model are discussed and qualitative
comparisons are made with real preform shapes taken from NCF reinforcement on hemispherical
dome forming experiments. The model has been used to compare predictions to the force vs.
deflection results from single dome forming tests as well as the comparative ability to predict the
location and severity of wrinkles. The current model demonstrates a high level of functionality
allowing accurate prediction of yarn and stitch pattern movement. The adopted modeling
technique, using separate shells works reasonably well, although it brings with it an increased
requirement on computer resources.
Keywords: Abaqus, Preforms, Composite, Anisotropic.
1. Introduction
Non-crimp fabrics (NCF) are aligned fabrics made from plies that are stitched together within a
single layer with the stitching also looped as inter-ply stitching, to make up the full fabric. The
reinforcement, say carbon fibre, is generally aligned within a plane, giving such NCF layers in
principle better mechanical properties (Khan, 2010) than their woven counterparts provided “all
other things are kept equal”. The stitching binds individual plies into a whole-layer assembly, so
that the integrity of the fabric is fairly-well conserved; they do not fall apart so easily before they
are placed into a mould and then infused with resin (Siddiqui, 2012). The same advantages at the
layup stage, brings with it potential wrinkling issues when one wants to form or press such NCF
into moulds when there is significant deformation required. International light weighting
initiatives in the automotive industry have spurred a need for modeling the draping and
deformation of NCF materials for complex geometries. Earlier work has been very successful in
modelling this behaviour for woven-style fabrics (Wocke, 2014) based on the rebar approach in
Abaqus.

Extending our existing capabilities for weaves to NCF-style fabrics is the focus of this paper. The
process for arriving at a tuned finite element computational model, able to accurately describe
local deformations for such NCF materials is discussed. Carrying the successful modelling
technique for weaves, over to the NCF style of fabrics presented some initial difficulties. Since
standard shell theory was not applicable, a modification of the earlier modelling approach was
required. In the case of NCF, the individual plies needed to be modelled within a single shell in
order to first capture the internal sliding degree-of-freedoms and then the additional stitching
restraint needed to be modelled between the shells. A cursory literature survey (Mitchell, 2014),
(Ji Seok, 2007) shows that the approach of modelling NCF using separate shells is common in
purpose-written software (Willems, 2006). The process of going from a well-working single-shell
modelling technique for the woven fabrics to that of modelling the NCF materials was not as
smooth as initially thought and presented many challenges.
2. Experimental testing of NCF materials

The naming convention related to NCF material, that we have adopted for analysis purposes, is:
“In general, an NCF layer is composed of individual plies of
fibre rovings stitched together. A layer of NCF is an assembly of stitched plies.”
For example, a single biaxial +/-45 NCF is an entity named “layer” composed of two
unidirectional “plies” with fibre rovings at +45 and -45 respectively. These plies are held together
to act as an individual layer via stitches; a laminate of fabrics is composed of a number of layers.
2.1 Layer characterization
In developing testing protocols, we relied on a combination of brainstorming/technical evaluation
sessions as well as a literature study (Kreissl, 2013), (Khan, 2010) to identify the test cases. In
order to provide basic characterization, we carried-out two bias-extension tests for biaxial +/-45
NCF fabrics. The dimensions of the strip for the samples are 240 mm long by 120 mm wide; this
ratio of length to width results in three characteristic deformation zones during the test, most
easily seen for cases where there is appreciable internal fibre rearrangement. In the first test, the
stitches were aligned to the loading direction, while in the second test the stitching ran
perpendicular to the pull as shown in Figure 1.
Stitching
(a) Stitching perpendicular (b) Stitching along the loading
Figure 1. Unit cell of sample of +/- 45 NCF for experimental bias extension tests.

Results from the experimental tests are shown in Figures 2 and 3. For stitches along the loading
direction, the load is able to be carried primarily by the carbon fibres with some small non-linear
tightening, due to internal friction, of the stitches during deformation; cf. (Wocke, 2014) for an
exposition of how the frictional resistance is generated inside such a layer.
(a) Stitching running along the pull (b) Stitching running perpendicular.
Figure 2. Graphic of experimental deformation of bias test for NCF.
With the stitching running perpendicular to the pull direction, there was relative opening of the
stitching and reorientation of the fibre tows i.e. the resistance is almost all due to internal frictional
restraint between tows. Hence, the response on the force-displacement curve yielded very low
values, by comparison to the case in Figure2a – it is essentially all frictional.
(a) Stitching running along the pull (b) Stitching running perpendicular.
Figure 3. Force-displacement response of the fabric.

The distinct deformation zones in the fabric which is a feature related to bias-extension tests can
also be discerned with the naked eye during testing. Particularly, for the case when the stitches are
running perpendicular to the pull direction, a central diamond and triangular areas at the top and
bottom of the fabric sample developed, in agreement with the expected, clear delineation of zones.
This can be quite clearly seen in Figure 2b.
3. Modeling the behaviour and calibrating

In order to model the in-plane behaviour of an NCF layer, we need at a minimum, a description of
a ply response. The stitching within the ply modifies this behaviour in the sense that it introduces
an additional orthotropy in the response, which also needs to be accommodated. To model a
fabric, we need to include a representation of the response/constraints provided by the stitching
between layers. The calibration has been performed on an FEM model representative of the bias
test; basically a strip held at one end and pulled at the other. To obtain the analysis data for
comparison to the test crosshead data easily from the Abaqus model, two single nodes were tied
via respective *MPC keywords to the edges of the shell; the one node was held fixed; the other was
given a prescribed displacement.
3.1 Single-shell approach

The single-shell approach is similar to in our previous work on weaves already presented (Wocke,
2014) where the response of the fibres is modeled using a shell section reinforced with internal
rebars aligned to represent the fibre tows.
In Table 1 details regarding the architecture and material properties of the fabric are included, in
order to be able to specify the rebar properties.
Table 1. Summary of architecture & material properties of biaxial +/-45 NCF.
Dry Areal Weight (gsm) 600 (300/300)

Strength 4200 MPa
Elastic Modulus 240 GPa
Fibre count 24k
Fibre diameter 7e-6 m
As an illustration, using the single-shell approach, consider the case of the stitches running
perpendicular to the loading direction. The simulation results from the model for this case are
shown in Figure 4. When compared to the experimental tests, the FE results over-estimated the
inward movement of the fabric. As shown in Figure 4, the minimum width at the centre from the
experimental results was approximately 60 mm while the FEM predicted resulted were ca. 20 mm.
Hence, a much narrower hour-glass shape is predicted. This is attributed the kinematics of
deformation of a single-shell where the effects of the very high Poisson’s ratio, as a consequence
of the +/-45 rebar, are predominant since the shell is so well-connected internally.

~ 60 mm
Figure 4. Comparison of the FEM results from bias test simulation for Biaxial +/-45
NCF for stitching running perpendicular to the pull with experiment.
The comparison of the predicted reaction force vs. displacement results to the experimental ones is
presented in Figure 5.
Figure 5. Comparison of force vs. displacement results for single shell bias test.
3.2 Kinematics of deformation of NCF
The deformation mechanism of a single layer of NCF material is governed by the relative motion
of the fabric plies. These plies are stitched together, yet still can slide relative to one another

during deformation. Nevertheless, the deformation mechanism of NCF is not based on
interlocking of warp and weft tows as is the case in the woven fabrics (Wocke, 2014). From the
general shell theory, independent relative sliding of rebars cannot be captured while defining the
set of rebars layers with different orientations within a single-shell as shown in Figure 6:
Figure 6. Kinematics of deformation of single shell with rebars.
3.3 Characterization of base-shell using pseudo UD response

In principle, a single layer of NCF consists of two plies (or two half layers) of fabric, suitably-
oriented. Within each ply, there is interply friction as the fibre tows slide against/past one another
as has been illustrated in Figure 6 and between these plies there is more friction, making up the
full internal frictional/resistance of the whole layer. In addition to the friction, there’s the internal
restraining effect of the stitching which almost invariably increases as a result of severe
deformation. This restraining effect acts within a ply as well as between plies when these are
assembled into a layer and eventually into a fabric.
A workable technique we’ve adopted, after some numerical experimentation, is to tune out the
base shell w.r.t. a half shell or one ply. This allows us to define the interply behaviour, using the
general procedure for tuning out the bias-extension test behaviour; the details of this are in our
earlier paper (Wocke, 2014). This tuned model then serves as the basis for describing the +/-45
NCF biaxial and a generic triaxial fabrics. When this is extended to the full layer, then we only
need to define the separate shells with restraint between them. The last value of restraint needs to
be tuned out from the experimental feedback from the single dome forming experiments. In
essence, our approach separates the contribution due to interply effects from those between plies –
the interply response has the orthotropic restraint effects captured by the ghost layer description
defined within the shell section.
The predicted reaction force vs. displacement results for the pseudo UD response for the two
experimental cases are presented in Figure 7.

--- FE aligned --- FE perpendicular
--- Experimental --- Experimental
(a) stitching along the loading (b) stitching running perpendicular
Figure 7. Comparison of Force vs. displacement-Predicted results for bias test for
pseudo UD shell.
3.4 Multiple-shell approach for NCF

The details of the computational model development for multiple shell approach is not disclosed
in this paper– this is proprietary to DOW, but rather we focused to illustrate some of the analysis
possibilities that are enabled by such a model.
In the multiple shell approach, each shell section represents an individual ply of the layer and their
directions. For the multiple-shell approach, the summary of steps taken for material
parameterization is as follows:
1. Model each ply’s interaction behaviour of the NCF layer as a separate shell mesh.
2. In each shell in the *SHELL SECTION, COMPOSITE definition, define rebar orientation
according to fibre tow directions of the ply.
3. Use *POTENTIAL to account for the reduction/alteration of stitching contribution and
orientation within the ghost layer.
4. Define individual thicknesses to each ply i.e. layer thickness ÷ no. of shell meshes.
5. Define relative sliding definitions between shells using:
i. Use *FRICTION between each ply of the layer to control relative sliding and/or
ii. Use *BOND to control relative sliding and/or
iii. Define trusses to connect nodes of individual shell sections to control relative
sliding and/or
iv. Use COHESIVE CONTACT INTERACTION BEHAVIOUR to control sliding of plies.
6. Run analyses for pulling experiment & correlate results.

When constraining the shells together, we investigated the use of different options available in
Abaqus. A summary of our findings is given below:
1. *BOND: this option is intended for modeling local spot welds between adjacent nodes.
However, this can only be used in single cpu analyses. Using *BOND will not be viable due
to extremely long computation times.
2. COHESIVE CONTACT BEHAVIOR: this option is based on modeling surface-to-surface
contact between adjacent faces.
3. *FRICTION used to define the restraint force between the shell surfaces of the plies, can
be used in multiple cpus analyses.
Defining coefficient of friction and cohesive contact interaction definition are found to be the best
options as far as computational resource cost are concerned.
The deformed shape comparison results for the bias extension test for biaxial +/- 45 NCF using
multiple shells approach, with stitching running along the loading direction is shown in Figure 8.
The visual agreement shows very good correlation, as is the case for the force-deflection results,
presented in Figure 9.
Figure 8. Comparison of deformed shape & rotation of fabric for biaxial NCF +/- 45
along the stitching bias extension tests.

--- FE_aligned
--- Experimental
Figure 9. Comparison of predicted force displacement with experimental results for

biaxial NCF +/- 45 along the stitching bias extension tests.
Carrying out the simulation for the bias extension test case, where the stitching runs perpendicular
to the loading only requires an orientation change of the ghost material in the input file, since the
plies lie at +/- 45. In our case, the ghost is hence rotated by 90o, making an equivalent change in
the *POTENTIAL keyword; for the general case, one would perform this via an *ORIENTATION
keyword. The experimental test suggests that the stitching may open up, which is not possible to
capture with a smeared shell model. Very little locking of the fabric is observed during the
loading. The results obtained in the multiple-shell bias extension test, for biaxial +/- 45 NCF,
when the stitching is running perpendicular to the loading direction are shown in Figure 10, with
the force-deflection results presented in Figure 11:
Figure 10. Comparison of deformed shape & rotation of fabric for biaxial NCF +/- 45
across the stitching bias extension tests.

Figure 11. Comparison of predicted force displacement with experimental results
for biaxial NCF +/- 45 across the stitching bias extension tests.
An over-estimation of the force-displacement response is observed from the predicted results; fine
tuning of the data on the *POTENTIAL keyword did not provide enough possibility (Simulia, 2013)
to reduce the orthogonal response contribution of the ghost layer representing the stitching. The
magnitude of force along this direction is very small and global errors due to the overall
distribution of the force would be small in comparison to the forces introduced from the other
direction. Further fine tuning of the process parameters may help reduce this error and additional
work in this regard is still under progress.
4. Implementation of NCF computational model to hemisphere

draping
4.1 FEM description of inverse draping test
The model represents a 350mm square sheet of preform fabric, being pushed into a die using a
round-nosed punch. The fabric is laid flat onto the flat peripheral surface of the die (base) and held
in place by a blank-holder (binder). The pertinent geometric dimensions for the analytical rigid
surfaces in the problem are shown in Figure 12 (Jauffrès, 2010). The binding force is held
constant, once contact has been established between preform and the die periphery. Analytical
rigid surfaces represent the rigid surfaces, with the preform modelled using S4R elements. Friction
is constant, using 0.3 as the friction coefficient (static friction only).

The FEM analysis was carried out using the multiple-shell approach and the material parameters
as discussed in the previous section; the binding force is 650N in the case of biaxial NCF +/- 45
and 3100N, in the case of triaxial NCF.
The analysis is performed in 3 steps, similar to what was done in Problem 1.3.3 in the Abaqus
Example Problems Manual (Simulia, 2013) and in our previous work on weaves (Wocke, 2014):
Step 1 – Establish contact, using softened exponential contact to avoid “numerical chattering”.
Step 2 – Apply the preload onto the binder, via its reference node.
Step 3 – Move the punch reference node by the required displacement.
The punch is moved 60mm downwards in a *DYNAMIC, EXPLICIT step in Abaqus/Explicit.
Figure 12. Geometric details of the surfaces for hemisphere draping.
4.2 Biaxial +/- 45 NCF test
In this experiment, the NCF layer of two plies has fibres of the top ply aligned with the global x-
and bottom ply with y-directions, respectively. The stitching is aligned -45 degree to the
horizontal plane which is modelled using a ghost layer within the composite shell section. The
predicted deformed shape is shown in Figure 13. The predicted results show excellent visual
agreement with experimental results; the wrinkling patterns and their general locations are well
captured, with well-controlled sliding between layers.

Figure 13. Predicted deformed shape of biaxial +/-45 NCF (stitching reorientation
is displayed with white-lines in FEM results).
4.3 Triaxial NCF test

The triaxial fabric used in this experiment consists of 3 plies, with the stitching along the
horizontal x-axis of the tool surface. The deformed shape, together with the location of the
dominant wrinkles, is shown in Figure 14, shows good visual agreement with experimental results.
As was the case with the biaxial fabric, the relative sliding of the individual plies is well-
controlled, without any disturbing relative movement of the three shells over each other.
Figure 14. Predicted deformed shape with experimental results for triaxial NCF
(stitching reorientation is displayed with white-lines in FEM results).

Even though the detailed internal deformation mechanism of triaxial NCF is different from the
woven fabric, the general shape of the force-displacement response of the triaxial NCF fabric
during hemisphere test is similar to that of woven fabrics. The predicted reaction force of the
punch is overly flexible at the beginning and then becomes overly stiff as the punch continues past
60 mm of displacement, as shown in Figure 15. The stiffer response at ca. 60 mm is attributed to
the interlocking of the stitching and the pressure dependent increasing friction restraint between
the plies; the resolution of this effect is left for future work.
--- FE_results
--- Experimental
Figure 15. Comparison of predicted force-displacement with experimental results

for triaxial NCF.
5. Conclusions & Recommendations

Our approach for modelling the NCF as suitably-oriented rebars within separate shells with
constraints between them, using the existing modelling capabilities within Abaqus has been
demonstrated. The approach we have used for the material parameterization is to tune the response
of a single ply through its rebar contribution and a suitable material model for the internal friction
between tows, supplemented by a ghost stitching layer and adjustments for orthogonality using the
*POTENTIAL keyword. Once acceptable numerical agreement is obtained for a single ply, it may
be combined with others into a single shell. Single shells, in turn are constrained to slide/move
over one another in a fabric, using cohesive contact behaviour. The predictive modeling approach
was demonstrated in the hemisphere draping of NCF type materials; we were able to predict fibre
rearrangement and the general occurrence of wrinkling. The relative sliding of the individual plies
was controlled and we did not notice any large relative movement of the two shells over each

other; the cohesive contact behaviour defined in the analyses was able to control the movement of
the elements with acceptable agreement to the experimental results.
Our future work will focus on implementation of this approach to more complex shapes and also
carry out experimental correlation for predicted fibre rotations with experiments.
Modeling the NCF fabric using separate shells brings with it an increased requirement on
computer resources. Currently, we have no alternative way for modelling these, except by writing
custom routines to better describe the NCF properties.
6. Acknowledgements
Acknowledgements go to Dr. Liangkai Ma for analyses on hemisphere draping & Mr. Jay Tudor
from The Dow Chemicals Co. for supplying the required material and the experimental test data
during the computational material model development.
7. References
1. Muhammad Aurangzeb Khan, “Numerical & experimental forming analyses of textile
composite reinforcements based on a hypoelastic behaviour”, PhD thesis for Ecole
Doctorale des Sciences de l’ingénieur de Lyon, France 2009-2010.
2. Siddiqui, M. Ali, “High Pressure RTM Process modeling of Carbon Fibre Composites with
Ultra-Fast-Cure resins”, Kunststoffe + SIMULATION 2012, 23 & 24 April 2012,
München, Germany.
3. Wocke, C. & Siddiqui M.A. “Using the existing capability in Abaqus to model the draping
& consolidation of composite preforms”, Proceedings of 2014 SIMULIA Community
Conference.
4. Mitchell Cynthia J., Sherwood James A., Dangora Lisa M.,and Gorczyca Jennifer L. ,
“Using Abaqus/CAE and User-Defined Material Subroutines to Predict the Deformations of
a Stitched Triaxial Fabric during Forming Processes”, Proceedings of 2014 SIMULIA
Community Conference.
5. Ji Seok Lee, Seok Jin Hong, Woong-Ryeol Yu, Tae Jin Kang, “The effect of blank holder
force on the stamp forming behavior of non-crimp fabric with a chain stitch”, Composites
Science and Technology 67 (2007) 357–366.
6. Willems A., Lomov S. V., Vandepitte D. and Verpoest I., “Double dome forming
simulation of woven textile composites”, 7th International Conference ESAFORM 2006,
Glasgow, UK, April 26-28,2006, S747-S750.
7. Kreissl S., Senner T., Lipp A., Meinhardt J., “Umformsimulation von Faserverstarkten
Kunststoffen”, LS-DYNA Anwenderforum, Filderstadt, 25th September 2013.
8. Khan M.A., Mabrouki T., Gauthier S., Vidal-Salle E., Boisse P., “Preforming simulation of
reinforcements of woven composites approach within a commercial code”, LaMCoS,
INSA-Lyon, CNRS UMR5259, F69621, France, URL: www.insa-lyon.fr

9. Abaqus 6.13, Documentation Collection, Abaqus Example Problems, Keywords & Theory
Manuals, Simulia.
10. Jauffrès, David, Sherwood, James A., Morris, Corey D., Chen Julie, “Discrete mesoscopic
modeling for the simulation of woven-fabric reinforcement forming”, International Journal
of Material Forming (2010) 3 (Suppl. 2):S1205–S1216.

Virtual Simulation of the Engine Hood Misuse Test,
Development Practices and Correlation Activities
Tamer Aydıner, Halil Bilal
Tofaş Türk Otomobil Fabrikası A.Ş.
Abstract: Misuse analysis of the engine hood is a critical evaluation to verify robustness of
vehicles. Major studies are carried out for the recent light commercial vehicle in Tofaş. Virtual
simulation of the engine hood misuse test; development practices to improve performance and
correlation activities with physical tests are stated on this study. Abaqus/Explicit solver is used for
virtual simulations.
Keywords: Engine hood, Misuse, Abaqus/Explicit, Explicit Dynamic Analysis, Topography

Optimization, Correlation
1. Introduction
It is very important that no damage occurs on vehicle body or engine hood after misuse by the
vehicle user. Misuse test of the engine hood can be defined as releasing the hood free from its full
open position.
During the test, the engine hood closes with high speed and results excessive forces on body and
lock zones. Performance goals of the test are;
• No collision between engine hood and the components around
• No damage on engine hood and vehicle body that impedes function of the engine hood
The FEM model of the test is composed according to misuse slam physical test norm of the FIAT.
Analyses are performed with Abaqus/Explicit.
Some collisions are observed between the engine hood and headlight on initial steps of the project.
Following studies are performed to eliminate these negativities.
• Topography optimization has been performed to improve bending behavior of the engine
hood.
• Elastic bumpers are added to eliminate collisions between headlight-engine hood and
bumper-engine hood.
• The geometry and stiffness of the current snow sealing is improved to dampen collision
energy efficiently.

2. Engine Hood Misuse Slam Test
The engine hood is released from full wide-open position to assure drastic conditions for misuse
slam test. Performance evaluation criteria of misuse test are stated on Figure 1.
Figure 1: Test Conditions
Figure 2: Measuring Points

Minimum distances and contacts are checked on the zones indicated in Figure 2. After the test,
there shouldn’t be any damage that affects performance of the engine hood and the components
around. Also the engine hood lock and striker should function properly.
Additionally while the test is conducted, motions are recorded with high speed camera. The
motion behavior of the engine hood is evaluated in detail.

3. Misuse Slam Simulation
3.1 Finite Elements Details
The finite element model used for misuse analysis is shown in Figure-3. Sectional body model is
used to decrease calculation time.
Figure 3: Finite Element Model

The finite element model consists of 830.000 element and 815.000 nodes.
As shown in Figure 4, all sheet parts in the model are modelled with S3, S4 shell elements.
Components like sealing, adhesives are modelled with C3D8 solid elements. Spot welds are
modelled with fastener elements.
Figure 4: Shell and Solid Elements

3.2 Finite Elements Details
*GENERAL CONTACT is used to model contact behavior between engine hood and body
components.
*TIE Contact is used between adhesives and sheet parts as shown in Figure 5.
Figure 5: Tie Contact

*TIE, NAME = Tie:S:Motore_Cofano:19 S:COLLANTE_OSSATURA:35, ADJUST=NO,
TYPE=SURFACE TO SURFACE
S:Motore_Cofano:19, S:COLLANTE_OSSATURA:35
Elastic bumpers on body are modelled with connector elements as shown in Figure 6.
Figure 6: Elastic Bumpers Modeling

Also engine hood lock system is modeled with connector elements as shown in Figure 7.
*ELEMENT,TYPE=CONN3D2,ELSET=CON_LOCK
*ELEMENT,TYPE=CONN3D2,ELSET=CON_STOP
*CONNECTOR SECTION, ELSET = CON_LOCK, BEHAVIOR = LEGGE_LOCK
*CONNECTOR SECTION, ELSET = CON_STOP, BEHAVIOR = LEGGE_STOP
*CONNECTOR BEHAVIOR, NAME = LEGGE_LOCK

*CONNECTOR LOCK, COMPONENT = 1, LOCK = ALL
*CONNECTOR BEHAVIOR, NAME = LEGGE_STOP
*CONNECTOR ELASTICITY, COMPONENT = 1, NONLINEAR
*CONNECTOR STOP, COMPONENT = 1
Figure 7: Lock Modeling

3.3 Boundary Conditions
Analysis begins from the moment that engine hood is approximately 6 degrees above the lock
instead of full open position to decrease explicit calculation time. The angular velocity of the
engine hood on that moment is defined as initial condition of the analysis which is derived by the
simulation with MBD model. (Figure 8)
Sectional body model is fixed on all DOFs at the zones as shown in Figure 3. Analyses are
performed with these boundary conditions.
Figure 8: Initial Velocity

3.4 Explicit Dynamic Analysis
Engine hood misuse slam test is a dynamic test; therefore finite element analysis is performed with
Abaqus/Explicit solver as explicit dynamic.
Explicit dynamic analysis procedure executes large number of small time increments efficiently.
An explicit central-difference time integration rule is used; each increment is relatively
inexpensive (compared to the direct-integration dynamic analysis procedure available in
Abaqus/Standard) because there is no solution for a set of simultaneous equations. The explicit
central-difference operator satisfies the dynamic equilibrium equations at the beginning of the
increment, t; the accelerations calculated at time t are used to advance the velocity solution to time
t + ∆t / 2 and the displacement solution to time t + ∆t .
4. Results
Highest stresses occur when the engine hood hits the lock and locking takes place during misuse
analysis.
The step which highest stresses occur on engine hood and environmental components is shown in
Figure 9.
Figure 9: Stress Distribution on the Hood

The velocity curves, energy curves and the forces on lock system and elastic bumpers are shown
in Figure 10-11-12.
Figure 1: Velocity Curves Figure 11: Energy Balances
Figure 12: Reaction Forces of Lock System and Elastic Bumpers

On the analyses performed initially, penetrations were seen between engine hood, bumper and
headlight zones as shown in Figure 13 and Figure 14. (The general contact interaction between the
engine hood and the bumper is cancelled due to monitor the scale of penetration. Other contact
relations are still intact in the model.)
Figure 13: Section A-A
Figure 14: Section B-B / Section C-C

5. Improvement Studies
5.1 Topography Optimization
After misuse analyses are performed, too much deflection is observed on engine hood during
locking stage. Therefore a linear-static analysis is performed with loads on engine hood similar the
loads occur on misuse analysis. An optimization problem was built upon that linear analysis which
aims to minimize displacements at the ends of the engine hood. The stiffness of the engine hood is
increased with topography optimization. The definition of optimization problem and design
variables are shown in Figure 15.
Figure 15: Topography Optimization
5.2 Adding Elastic Bumper

2 new elastic bumpers are added to the zones which shown in Figure 16.
Figure 16: Elastic Bumpers

5.3 New Snow Sealing Design
Snow sealing is used to prevent leaking snow from the gap between bumper and engine hood as
the name suggests.
While the engine hood is on closed position, snow sealing is somewhat compressed. But the force
between the engine hood and sealing should be as small as possible. If this force is high when the
engine hood is on closed position, the gap between engine hood and bumper will increase.
The snow sealing design is modified as shown in Figure 17 to dampen the energy of collision
when the hood contacts to bumper surface on misuse analysis.
Figure 17: New Snow sealing Design

5.4 Improved Design Results
After all these improvements, penetration problems between engine hood, bumper and headlights
are eliminated. The analysis results before and after the improvements are shown in Figure 18 and
19. (The general contact interaction between the engine hood and the bumper is cancelled due to
monitor the scale of penetration. Other contact relations are still intact in the model.)
Figure 18: Before-After Section A-A
Figure 19: Before-After Section B-B and C-C

6. Comparison between Physical Test and Simulation
No negative results are observed on physical tests with the engine hood developed with mentioned
improvements.
To inquire the correlation between physical tests and virtual analysis, some physical tests are also
performed with the engine hood on the initial levels of the project. Similar negative results
acquired on virtual analyses are also observed on physical tests. The comparison is shown in
Figure 20 and 21.
Figure 20: Contact between engine hood and bumper
Figure 21: Contacts between engine hood and headlight

7. Conclusion
Collisions between engine hood, bumper and headlights are noticed on virtual misuse analysis at
the initial steps of the project. These collisions are eliminated by the development activities
carried out with virtual analysis methods. Likewise, physical misuse tests are carried out with the
improved design and the results are affirmed as positive. To inquire the correlation between
physical tests and virtual analysis; physical tests are also performed with the engine hood on the
initial levels of the project. Similar results are monitored on physical and virtual tests.
The correlation between the physical and virtual tests is fulfilled with a good margin. Henceforth,
misuse of the engine hood is evaluated only with virtual tests. The costs of physical tests and
prototyping are reduced substantially.
8. References
1. Abaqus Version 6.10 Documentation
2. Related Fiat Specifications.

INDEX
A Dasari, G. 495 Karunakar, G. 990
ABB Turbo Systems AG 674 Dekker, A. 401 Kilic, H. 116
Abedrabbo, N. 621 Dong, Y. 327 Kirchberger, H. 1113
Aernnova Engineering Division Dow Chemical 375 Kobayashi, T. 390
S.A. 12 DTU Wind Energy 509 Korte, W. 1024
Airbus Defence and Space 22 DuPont 904 Kostov, N. 546
Airbus Group 37 E Kujawski, D. 254
Airbus Operations Ltd. 48 Essa, Y. 12 Kumar, V. 869
Al-Khalil, M. 48 ESTECH Corporation 267 L
Alotta, G. 782 ExxonMobil Upstream Research Lee, H. 949
ALPLA Werke Alwin Lehner Company 495, 522, 536, Leibniz University Hanover 555
GmbH & Co KG 360 546 Liu, S. 636
Alvares de Castro e Sousa 566 Lui, P. 660
An, J. 974
F
Ansaldo Energia 417
Fedulov, B. 198 M
Fihri Fassi, H. 281 Maas, J. 405
APPLUS IDIADA 855
Fokker Landing Gear B.V. 102 Mechanical Design & Analysis
Arbiol, J. 855
Ford Otosan 920 Corporation 390
Arndt, S. 431
FSTS 281 Mercedes Benz India R & D
ATA Engineering, Inc. 63
G Pvt Ltd 1011
Autoliv India Pvt. Ltd. 869
Garzon, J. 522 Mercedes-Benz Turk A.S. 702
Aydiner, T. 1141
Gerlach, T. 555 Mishra, P. 1011
B Momii, H. 267
Goller, B. 129
Baker Hughes 431, 449 Moscow State University 160
Greene, Tweed & Co. 116
Bäker, M. 207 Mueller, T. 997
Güzel, A. 920
Bakir, M. 702
Barrera, O. 795 H N
Heimbs, S. 37 Nacivet, S. 1047
Basciano, C. 808
Hitachi, LTd. 691 Ngai, A. 1034
BD 808
Honda R&D 931 Nikaido, Y. 291
Belles Meseguer, A. 176
HPC Advisory Council 660 Ning, N. 536
Bello-Gomez, M. 756
Hyundai Motor Company 949, N. K., S. 1062
Birkmose, M. 833
Brusa, E. 717 956 O
BTU Cottbus–Senftenberg 75 I Olveda, D. 742
Buzlaev, D. 1102 IDAJ Co., Ltd. 963 OTH Regensburg 1018
C ILJIN Global Corporation 974 P
CEA/DAM/DIF 90 INTALES GmbH 129 PART Engineering 303, 1024
C-FER Technologies 463, 478 Ishikawa, S. 963 Pavlov, M. 219
Charnaux, C. 601 ITAINNOVA 144 Pazour, S. 303
Chiminelli, A. 144 J Penso Consulting 1034
Coskun, V. 188 Jaguar Land Rover 981, 990 Pilgaard Mikkelsen, L. 509
Cousin, L. 90 Pinna, D. 417
K Politecnico Di Torino 717
Crespo, M. J. 313
Kajima Corporation 291
Popielas, F. 894
D Karlsruhe Institute of
Prime Aerostructures 176
Dana Holding Corp 880, 894 Technology 997
Principia 313

PSA Peugeot Citroën 1047 Teufel, P. 674
R The Dow Chemical Company
Ramirez, E. 768 1126
Rebien, H. 880 Tofaş Türk Otomobil Fabrikası
Reveles, N. 63 A.Ş. 1141
Rickhey, F. 727 Trek Bicycle Company 405
Ritsumeikan University 842 TsAGI 219, 235, 242
Rivero Arevalo, I. 22 U
RMC Consultoria 566 UANL 756
Robert Bosch Engineering 1062 Universidad Nacional
Roketsan Missiles Inc. 188 Autónoma de México 768
Rozin, N. 235 University of Oxford 795
S University of Palermo 782
SABIC 1080 V
Sacmi/Protesa 1091 Valtiner, O. 360
Saponelli, R. 1091 van der Zee, W. 449
Se-Hee, O. 956 Vershinin, V. 160
Shanghai Turbine Works Volgers, P. 904
Company Ltd 582 W
Siddiqui, M. 1126 Wallis, S. 981
Simpleware 820 Wang, Z. 842
Singapore-MIT Alliance for Weatherford 621
Research and Technology Western Michigan University
327 254
Skolkovo Institute of Science West Virginia University 344
and Technology 198 Wilson, W. 102
Smurfit Kappa 401 Wocke, C. 375
Soendergaard GN Resound A/S Wood Group Kenny 636
833
Sogang University 727
X
Xie, J. 463, 478
Sosa, E. 344
Stelldinger, E. 75 Y
Subsea 7 601 Yaldiz, R. 1080
Swidergal, K. 1018 Young, P. 820
T Z
Takada, K. 931 Zharenov, I. 242
Tan, S. 582
Technische Universität
Braunschweig 207
Technische Universität
Darmstadt 742
Terasaki, T. 691
TESIS 1102
TEUFELBERGER Composite
Ges.m.b.H. 1113

trademarks or registered trademarks of Dassault Systèmes or its subsidiaries in the U.S. and/or other countries. All other trademarks are owned by their respective owners. Use of any Dassault Systèmes or its subsidiaries trademarks is subject to their
©2015 Dassault Systèmes. All rights reserved. 3DEXPERIENCE®, the Compass icon and the 3DS logo, CATIA, SOLIDWORKS, ENOVIA, DELMIA, SIMULIA, GEOVIA, EXALEAD, 3D VIA, 3DSWYM, BIOVIA, NETVIBES, and 3DEXCITE are commercial
Our 3DEXPERIENCE® platform powers our brand applications, serving 12 industries,
and provides a rich portfolio of industry solution experiences.
Dassault Systèmes, the 3DEXPERIENCE® Company, provides business and people with virtual universes to imagine sustainable
innovations. Its world-leading solutions transform the way products are designed, produced, and supported. Dassault Systèmes’
collaborative solutions foster social innovation, expanding possibilities for the virtual world to improve the real world. The group brings value
to over 190,000 customers of all sizes in all industries in more than 140 countries. For more information, visit www.3ds.com.
Europe/Middle East/Africa Asia-Pacific Americas
express written approval.

Dassault Systèmes Dassault Systèmes K.K. Dassault Systèmes
10, rue Marcel Dassault ThinkPark Tower 175 Wyman Street
CS 40501 2-1-1 Osaki, Shinagawa-ku, Waltham, Massachusetts
78946 Vélizy-Villacoublay Cedex Tokyo 141-6020 02451-1223
France Japan USA

2015 SCC Proceedings-Reduced PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

2015 SCC Proceedings-Reduced PDF

Caricato da

Copyright:

Formati disponibili

SIMULIA COMMUNITY CONFERENCE

MAY 18–21, 2015 | BERLIN, GERMANY

2015 SIMULIA Community Conference www.3ds.com/simulia 2

22 Evaluation of Abaqus XFEM Capabilities for Crack Growth Analysis in

48 Use of Abaqus Explicit for Composite Sandwich Impact Damage Prediction

63 Time Accurate Simulation of Aeroelastic Flap Deployment with Free Play

75 Failure Analysis of CFRP Tubes with Integrated Rubber Layers Subjected to

90 New Design of a Pressure Vessel Subjected to Blast Loads

102 Predicting In- and Out-of-plane Damage Evolution in Fiber-reinforced Composites

129 An Approach Towards an Optimal Design of Composite Structures

2015 SIMULIA Community Conference www.3ds.com/simulia 3

176 Simulation Lifecycle Management in the Real World: Practical Application

188 Modeling of Heat Transfer across Bolted Joints in Abaqus/CAE

198 Modeling of Forming Process of Composite Materials Based on

207 Finite Element Simulation of Thermal Barrier Coatings in Rocket Engines

219 Residual Strength of the Carbon Fiber Panel with Delamination

235 Shrinkage Modeling of Thermoplastic Wing Rib

254 A Simulation Tool for Fatigue Analysis and Lifecycle Prediction

ARCHITECTURE, ENGINEERING & CONSTRUCTION

281 Study of Mechanical Compatibility at the Mortar-block Interface in a Heritage Building

291 Improvement and Enhancement of Concrete Damage Plasticity Model

303 An Easy Procedure for Anisotropic Non-Linear Behavior of

2015 SIMULIA Community Conference www.3ds.com/simulia 4

327 Advanced Finite Element Analysis of Deep Excavations using Abaqus

344 Modeling of Confined Inflatable Structures

CONSUMER PACKAGED GOODS & RETAIL

390 Analysis of Material-Related Instability Problems for Inelastic Thin Shells

ENERGY, PROCESS & UTILITIES

431 Hydraulic Fracturing Simulation for Fracture Networks

449 Fracture Stability Assessment during SAGD operations using

495 Simulation of Hydraulic Fracturing of Unconsolidated Sands using Fully Coupled

2015 SIMULIA Community Conference www.3ds.com/simulia 5

520 Advanced Fracture Modeling for Cuttings Re-injection

536 Experimental Validation of Simulation Capabilities for Hydraulic Fractures Propagating

546 Dynamic Hydraulic Fracture Modeling for Wellbore Integrity Prediction in

555 A Numerical Investigation on the Bedding Resistance of Laterally

636 Abaqus Analysis Methods on Highly Restrained Pipeline with

2015 SIMULIA Community Conference www.3ds.com/simulia 6

691 Prediction of Tin-Whiskers Generation Using Stress and Mass-Diffusion

727 XFEM Simulation of Radial-Median Crack Evolution in Knoop Indentation of

742 Evaluation of the Fatigue Strength of a Steel-Aluminum Tapped Thread Joint

756 Strip Shape Prediction using a Simplified Static Model of a 4-High

768 Validation of Flow Simulation on Abaqus/CEL™

782 On the Numerical Implementation of a 3D Fractional Viscoelastic

795 Hydrogen Embrittlement Mechanisms in Metals: a Modelling Approach

2015 SIMULIA Community Conference www.3ds.com/simulia 7

833 Impact Simulation of a Hearing Aid

842 FE Modeling of a Flatfoot Deformity for Improving Surgical Planning

TRANSPORTATION & MOBILITY

869 Thermal Management of Electronic Devices used in Automotive Safety –

880 Development of a Separator Plate for Valve Bodies in a Vehicle

894 System Modeling and Simulation at Dana Holding

904 New Constitutive Model for Woven Thermoplastic Composite Materials

920 Durability Assessment of Heavy Duty Diesel Engine Cylinder Head

931 Prediction of Compressive Characteristic of Hat Section High-strength Steel Frame

2015 SIMULIA Community Conference www.3ds.com/simulia 8

963 Simulation of Pipe Bending Process with Abaqus

981 Automated Analysis in CATIA V5

990 Simulation of Fully Coupled Thermo-Mechanical Effects in a Disc Brake Rotor

997 Simulation of Combined Forming and Injection Molding Processes

1011 Application of Predictive Engineering Tool (Abaqus) to Determine Optimized

1018 Final Element Simulation of Blankholder’s Lift-off in a Deep Drawing Tool

1024 Strength Assessment of Injection Molded Short-fiber-reinforced