Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
DISSERTATION
Dissertation Committee:
Approved by
________________________________
Adviser
Department of Mechanical Engineering
Copyright by
Suwat Jirathearanat
December, 2003
ABSTRACT
Procedure of automatic optimization of THF loading paths (i.e. pressure, axial feed
velocity, and counter punch force curve versus time) using PAM-STAMP and a
general optimization code, PAM-OPT, for typical THF complex part geometries
such as, simple bulges, Y-shapes, and automotive structural parts.
Adaptive Simulation (AS) program that works with a commercial code (PAMSTAMP) to automatically determine feasible loading paths of any given THF parts.
The current AS program can handle only simple part geometries such as
axisymmetric bulges.
iii
For my parents,
my sisters,
and my wife
iv
ACKNOWLEDGEMENTS
I express my sincere thank to my advisor, Dr. Taylan Altan for taking the time to
mentor and tutor me throughout the years of my graduate study program. His insight,
wisdom, support, and trust were indispensable. I also would like to give special thank
all of my friends, at the Engineering Research Center for Net Shape Manufacturing,
who have helped make this research effort possible. Their invaluable assistance in
technical areas and their uplifting emotional support will always be remembered.
Further, I would like to thank the following governmental agencies and industrial
companies for their generous financial and technical support:
In closing, I would like to express my gratitude to my entire family for their unyielding
support and love.
VITA
PUBLICATIONS
Peer Reviewed Journals:
M. Koc, T. Allen, S. Jirathearanat, and T. Altan, The Use of FEA and Design of
Experiments to Establish Design Guidelines for Simple Hydroformed
Parts, International Journal of Machine Tools & Manufacture 40 (2000)
2249-2266
vi
FIELDS OF STUDY
Major Field: Mechanical Engineering
Studies in:
vii
TABLE OF CONTENTS
Page
ABSTRACT
.............................................................................................................. ii
ACKNOWLEDGEMENTS ..............................................................................................v
VITA
............................................................................................................. vi
1.1
Introduction..........................................................................................1
1.2
1.3
CHAPTER 2.
2.1
Tube Hydroforming.............................................................................4
2.1.1
2.1.2
2.2
2.2.2
2.3
2.3.2
Numerical Methods..............................................................................12
2.3.2.1 Optimization Simulation Methods.......................................................12
2.3.2.2 Feedback Control Simulation Methods ...............................................15
2.3.2.3 Adaptive Simulation Methods .............................................................16
CHAPTER 3.
3.1
Hydroforming of Y-shape....................................................................19
3.1.1.1 Tube Hydroforming Process Procedure...............................................21
3.1.1.2 Determination of the Process Parameters ............................................23
3.1.2
3.1.3
3.2
3.3
3.3.1
3.3.2
CHAPTER 4.
4.1
4.2
4.1.1
4.1.2
4.2.1
4.2.2
CHAPTER 5.
5.1
Components of Optimization...............................................................81
5.1.2
5.2
Design Variables..................................................................................86
5.2.2
Objective Function...............................................................................88
5.2.3
5.3
5.4
Simple Bulge........................................................................................92
5.4.1.1 FE Model Descriptions ........................................................................92
5.4.1.2 Optimization Descriptions ...................................................................94
5.4.1.3 Optimization Results..........................................................................100
5.4.2
Y-shape ..............................................................................................103
5.4.2.1 FE Model Descriptions ......................................................................105
5.4.2.2 Optimization Descriptions .................................................................105
5.4.2.3 Optimization Results..........................................................................110
5.4.3
CHAPTER 6.
6.1
6.2
6.2.2
6.2.3
6.3
6.4
Calibration Stage................................................................................156
6.4.2
CHAPTER 7.
7.1
7.2
7.2.2
7.3
Conclusions.......................................................................................180
7.4
Future Work.....................................................................................184
LIST OF REFERENCES..............................................................................................185
APPENDIX A FLOW STRESS DETERMINATION................................................192
xi
xii
LIST OF FIGURES
Page
Figure
Figure 1.1:
Figure 2.1:
Figure 2.2:
Figure 2.3:
Figure 2.4:
Figure 2.5:
Figure 2.6:
Figure 2.7:
Figure 2.8:
Figure 3.1:
Figure 3.2:
Figure 3.3:
Figure 3.4:
Figure 3.5:
Figure 3.6:
Figure 3.7:
Figure 3.8:
Figure 3.9:
Figure 4.2:
Figure 4.3:
Figure 4.4:
Figure 4.5:
Figure 4.6:
Figure 4.7:
Figure 4.8:
Figure 4.9:
Figure 4.10: Summary of the axial feed curves from the simulations
conducted to optimize the loading paths through SF
simulation approach...............................................................................65
Figure 4.11: Optimized loading paths from SF: pressure, left axial
feed, right axial feed ...............................................................................65
Figure 4.12: Intermediate tube hydroforming steps: side view and
front view .................................................................................................68
Figure 4.13: Thinning distribution on the final simulated part and a
table comparing the simulation and experimental
results at some specific areas.................................................................69
Figure 4.14: FEA modeling of hydroforming crossmember [Schuler
Hydroforming] ........................................................................................71
Figure 4.15: Pressure curves and corresponding nodal velocity from
the right tube end, obtained from SF simulations..............................73
Figure 4.16: Plots of axial feeds (left and right tube ends) selected
through SF simulation approach ..........................................................73
xvi
Figure 4.17: Plots of pressure and right axial feed versus left axial
feed (case B, Figure 4.16)........................................................................74
Figure 4.18: Intermediate simulation results of crossmember
hydroforming ..........................................................................................75
Figure 4.19: Plots of pressure and right axial feed versus left axial
feed used in the experiments.................................................................76
Figure 4.20: Crossmember parts hydroformed with the loading
curves above [Schuler Hydroforming] ................................................76
Figure 4.21: Thinning measurements of the Cross member from
prototyping ..............................................................................................77
Figure 5.1:
Figure 5.2:
Figure 5.3:
Figure 5.4:
Figure 5.5:
Figure 5.6:
Figure 5.7:
Figure 5.8:
Figure 5.9:
Figure 5.11: Initial and optimized loading paths for simple bulging .................102
Figure 5.12: Part thinning distribution of optimized simple bulge.....................102
Figure 5.13: Optimized loading paths for simple bulging ...................................104
Figure 5.14: Part thinning distributions of the simple bulge ...............................104
Figure 5.15: Design variables: counter punch force versus time curve
and pressure versus time curve ..........................................................107
Figure 5.16: Left and right axial punch velocity versus time curves ..................107
Figure 5.17: Objective function: maximizing the protrusion height or
maximizing the part controlled volume ............................................109
Figure 5.18: Constraint functions: a) tube-to-die distance, b)
protrusion corner curvature, and c) part maximum
thinning ..................................................................................................109
Figure 5.19: Objective function: evolution of part controlled volume................111
Figure 5.20: Constraint functions: evolutions of a) tube-to-die
distance, b) corner curvature, and c) part maximum
thinning ..................................................................................................111
Figure 5.21: Optimized counter punch force curve and pressure
curve versus time ..................................................................................112
Figure 5.22: RSM Objective function: part controlled volume.............................114
Figure 5.23: RSM constraint functions: a) tube-to-die distance, b)
corner curvature, and c) maximum thinning....................................114
Figure 5.24: RSM optimized a) counter punch force curve and b)
pressure curve .......................................................................................115
Figure 5.25: Comparison of part qualities obtained from Gradientbased method and RSM method; a) part thinning
distributions, and b) protrusion profiles ...........................................116
Figure 5.26: FE model of structural part: part geometry and material
properties [Kawasaki Hydromechanics, Japan] ...............................119
xviii
Figure 6.2:
Figure 6.3:
Figure 6.4:
a) loading path in the THF forming window, and b) inplane strain plot.....................................................................................133
Figure 6.5:
Figure 6.6:
Figure 6.7:
Figure 6.8:
Figure 6.9:
xx
Figure 7.2:
Figure 7.3:
Figure 7.4:
Figure 7.5:
Figure 7.6:
Figure A.1:
Figure A.2:
Figure A.3:
Figure B.1:
Figure C.1:
Figure C.2:
Figure C.3:
Figure C.4:
Figure D.1:
Figure D.2:
Figure E.1:
Figure E.2:
xxii
LIST OF TABLES
Table
Page
Table 3.1:
Table 4.1:
Table 4.2:
Table 4.3:
Table 4.4:
Table 7.1:
Table 7.2:
Table 7.3:
Table 7.4:
Table A.1:
Mechanical properties of SS304 tubes used in the Yshape hydroforming experiments ..................................................... 194
Table B.1:
NOMENCLATURE
Glossary
Geometrical Parameters
side
right side
AS = Adaptive Simulation
Dp = prtrusion diameter
Process Parameters
Self-feeding Simulation
self
d left
= left axial feed from SF simulation
self
d right
= right axial feed from SF simulation
xxiv
Optimization-based Simulation
Adaptive Simulation
d i1 = tube-to-die distance
indicator
V = normalized part fluid cell volume
xxv
CHAPTER 1.
INTRODUCTION, PROBLEM STATEMENT, AND GOALS
1.1 Introduction
guarantees a successful THF process without any of the failures, see Figure 1.1.b. Both
parameters pi(t) and da(t) are dependent on part geometry, tube material, and lubrication
conditions.
Effective classifications of hydroformed tubular parts are necessary for development of
THF part design and process systematically.
simulations can be used as a tool to extensively analyze THF. Design of the process
parameters are normally selected through time-consuming, trial-and-error iterative FEA
simulations. FEA simulation enhanced with optimization schemes can greatly reduce
the lead-time spent in the process development.
Tube
Pure shear
Tube
Uniaxial tension
Plane strain
a)
Figure 1.1:
b)
The development of tube hydroforming processes is plagued with long lead times,
which result from much iteration of tryouts either on trial-and-error based FEA
simulations or on expensive prototype tooling. Hydroformability of tubular parts is
affected by a large number of parameters such as material properties, tube geometry,
complex die-tube interface lubrication, and process parameters, i.e. loading paths. More
powerful design tools are needed to help engineers design better products and processes
and to reduce lead times and cost. Therefore the goals of the proposed work are:
Develop part design guidelines for THF processes that facilitate engineers to bring
conceptual THF part designs to production more efficiently by early eliminating bad
designs considering manufacturability issues and arriving to successful part designs.
Develop methodologies for design and optimize loading paths in THF. The
methodologies will utilize systematic FEA simulations and FEA enhanced with
numerical optimization methods and Adaptive Simulation (AS) method. These
tools will enable the engineers to select loading paths (i.e. pressure curve, axial feed
curve, and counter punch force curve versus time) optimized for simple to complex
tube hydroforming processes such as T-shapes, Y-shapes, cross members, and
engine cradles.
1.3 Dissertation Organization
Chapter 2:
Literature Review
Chapter 3:
Chapter 4:
Chapter 5:
Chapter 6:
Chapter 7:
CHAPTER 2.
LITERATURE REVIEW
2.1 Tube Hydroforming
C
Tools /
Dies
B
Bending /
Preforming
D
Tool-Workpiece
Interface
E
Deformation
Mechanics
F
Equipment /
Environment / Press
Figure 2.1:
G
Hydroformed
part
facilitate the tube alignment into the die but also to accumulate the tube material locally
for the subsequence expansion process. Crushed geometries are found frequently in
automotive structural parts.
Bulge: bulges are typically tube expansions, mostly axisymmetric about the tube axis.
Protrusion: protrusions are local expansions, stemmed out from the tube axis. They are
(a)
Figure 2.2:
(b)
(c)
(d)
Tube hydroformed part features (a) bent feature, (b) crushed feature, (c)
bulge feature, (d) protrusion feature (referred as Y-shape), and (e)
automotive hydroformed structural part (SPS, Germany)
FEA for hydroforming process assists die designers and process engineers to (a) assess
the manufacturability of parts at the design stage, (b) explore alternative design
schemes, and eventually (c) arrive at an optimized design in a cost effective and timely
fashion. With the aid of FEA simulation, the part quality control, and the design of the
tube hydroforming process can be easily implemented and monitored. FEA simulations
provide insights on the necessary process parameters/ loading paths (i.e. internal
pressure and axial feed), part geometry, and part formability by analyzing the thinning,
thickening, and strain distribution in the deformed tube.
2.2.1 FEA Modeling
There are a few analytical equations to predict formability, i.e. thickness distribution, of
simple THF parts such as T-shapes [Shen-Zhang; 1999] and simple axisymmetric
bulges [Asnafi, 1999]. However, compared to FEA, the analytical equations have
limited applicability for THF of general part geometries [Lei, 2001a]. FEA assists the
die designers and process engineers to (a) assess manufacturability of a part during the
design stage, (b) explore alternative designs, and eventually (c) arrive at an optimized
design in a cost effective and timely fashion. FEA simulations provide insights about
the part formability by predicting its stress and strain distributions in the deformed tube.
This information facilitates selection/optimization of the process parameters (i.e.
internal pressure and axial feed curves versus time), as well as part geometry
modification if necessary.
Until now a number of researchers have applied three-dimensional FEA on several THF
processes: simple bulges [Donald, 2001], T-shapes and Y-shapes [Lei, 2001a]
[Jirathearanat, 2001a], and automotive structural parts [Yang, 2001a] [Kim, 2002]
[Jirathearanat, 2001a]. Most of the FEA simulations were conducted by Dynamic
Explicit FEA packages, e.g. PAM-STAMP or LS-DYNA, which have advantages in
fast changing boundary conditions necessary in forming with complex die surfaces and
6
capability to handle large deformation forming. Some researchers prefer applying the
Static Implicit FEA formulation for its more reliable and rigorous scheme in
determining equilibrium at each step of deformation. However, there exist intrinsic
problems associated with the Implicit FEA formulation such as convergence and long
computation time [Lei, 2001a]. Therefore, the implicit FEA packages are normally
limited to hydroforming of simple tubular part geometries.
2.2.2 Failure Analysis
Major failure modes in THF are buckling, wrinkling and fracture (bursting), Figure 2.3.
Robust methods of predicting and analyzing failures in stamping parts have been under
intensive investigation by a number of researchers. However, a reliable analysis method
for the failure problems in THF has not yet been established. Due to the lack of reliable
failure prediction methods, the methods used in sheet metal forming are inevitably
applied in tube hydroforming processes. Forming Limit Diagrams (FLD) are generated
experimentally based on assumption of proportional loading path. Levy (1999) applied
FLDs to predict formability in THF. He suggested that FLDs for THF should be above
the FLDs determined from a flat sheet of the tubular material. Another competing
method is ductile fracture criteria. Unlike the FLDs, these ductile fracture criteria are
not loading path dependent. Fracture is predicted when the ductile integral value
exceeds a critical value, which is determined experimentally through tensile tests. Filice
(2001) successfully implemented Crokroft and Latham ductile fracture criterion on
simple bulge simulations, and validated the criterion experimentally. Lei (2001b)
applied Oyane ductile fracture criterion on simulations of a bumper rail and a subframe.
However, no experimental data was used to validate the simulation studies. Maximum
thinning criterion is also considered another good way of predicting fracture. However,
maximum thinning criterion is not reliable when biaxial tensile state of stress is
dominant.
The analyses of onset and growth of wrinkles in the literature are found mostly on sheet
metal forming. The analyses are generally based on three main methods: a) plastic
7
bifurcation theory, b) energy method, and c) geometry method. The underlying idea of
the plastic bifurcation theory [Hill, 1958] [Hutchinson, 1974] is that, for unperturbed
(perfect) shell structures, wrinkles may take shape when the solution to an energy
equation describing the solid mechanic problem (elastic and plastic regions) is not
unique. After this bifurcation point, wrinkles may appear or the unwrinkled state may
hold until another bifurcation point. A drawback of the plastic bifurcation theory is that
it only deals with initially unperturbed structures. In THF processes, wrinkling may
appear at any stage of the process where the part is deforming.
Figure 2.3:
Common failure modes that limit THF process, winkling, buckling, and
bursting [Koc, 2002]
Figure 2.4:
Even though the wrinkle detection criteria discussed above were invented many years
ago, they have been implemented only on in-house FEA codes. Commercial FEA
codes, such as PAM-STAMP, still have not implemented those criteria. Nevertheless,
different types of wrinkles in sheet metal forming (flange wrinkles, sidewall wrinkles
and wrinkles beneath the punch) can be predicted by PAM-STAMP, where a wrinkled
part is depicted by its deformed mesh [Aita, 1992]. In PAM-STAMP, wrinkles are
predicted based on the energy minimization of plastic deformation combined with some
intrinsic numerical round off in its explicit FEA formulation. This method may not
predict onset of wrinkles as accurately as predicted by the previously mentioned
criteria. However, unlike in stamping, wrinkles in a hydroformed part may be
controlled/straightened by an appropriate increase of internal pressure. Therefore, it is
justified to detect wrinkles as they have formed to a relatively small size (noticeable
wrinkle amplitude). This simplifies the wrinkle detection in THF simulations. The
wrinkles can be identified based on simple geometrical considerations, rather than on
energy/stresses.
In using most of FEA commercial codes for THF, some measurements/quantities of the
visible wrinkles based on their geometric considerations need to be devised, in order to
enable adjustment/optimization of the part and process design. There are several
advantages of a geometry base approach. It is simpler mathematically than most of the
other criteria. A small amount of wrinkles in the THF part may be even helpful in
preventing excessive thinning in the bulging area, since it is a way to accumulate
9
Figure 2.5:
The main process parameters in THF are pressure, axial feeds, and counter punch force.
These are also often referred to as loading paths or part program when presented in
time domain. The success of THF processes is largely dependent on the choice of the
loading paths. Part geometry, tubular material, and lubrication conditions need to be
taken into account in designing of loading paths. The selection of proper loading paths
can be done using empirical methods, analytical methods, or numerical methods.
10
Empirical methods are most suitable to roughly estimate the process parameters for
simple to moderate complex THF part geometries. Usually these methods are quick but
not accurate. Analytic methods are developed based on plasticity theory. Most of the
analytical models available for THF are often not applicable for even part geometries
with moderate complexity. However, for simple part geometries the available analytical
models can predict proper process parameters rather accurately. For general cases,
numerical methods (FEA simulations) are very practical and widely applied in the
industry.
2.3.1 Empirical and Analytical Methods
Most empirical rules for THF part and process designs are developed through
prototyping. A number of THF guidelines can be found in handbooks from large
hydroforming companies such as Schuler Hydroforming and Nippon Steel, see Figure
2.6. These empirical rules, however, are always of adhoc nature. Therefore, they should
only be used to get some conceptual ideas during initial design stage.
(b)
(a)
Figure 2.6:
Analytical equations enable the engineer to estimate accurately the necessary process
parameters. Analytical models for THF are normally developed based on plasticity and
thin-walled or thick-walled theories. Koc (2002) and Asnafi (2000) developed equations
to determine process limits, such as yielding, bursting, and calibrating pressure levels
and necessary amount of axial feed, for simple bulging. Analytical equations to
determine process limits are difficult to be developed, particularly for complex THF
parts. From the literature reviewed, all the analytical models only calculate process
11
parameter limits, such as yielding and bursting pressures. None of the equations can
predict the necessary loading paths (i.e. evolution of process parameters in time) for
successful THF. Most of these analytical equations are developed for particular THF
features, i.e. simple bulges and T-shapes, see Figure 2.2. The analytical equations
become inefficient if not useless when designing a complex THF part, which consists of
many THF features, and its process. It is noted here that even though the empirical
rules and analytical equations provides guidelines on THF part and process design,
many more design iterations are often necessary. FEA simulations are normally used in
the design improvement stage. The process parameters are modified till successful THF
process is obtained.
2.3.2 Numerical Methods
Trial-and-error simulation method for the process design can be very time consuming,
i.e. pressure and axial feed curves versus time are selected to conduct a simulation. If
the results are not satisfactory, the input curves are modified by intuition and the
simulation is run again until satisfactory results are obtained. Fortunately, this iterative
FEA method can be done systematically and automatically with kinds of optimization.
For example, determination of the loading paths can be treated as a classical
optimization problem. By this way the resultant loading paths are optimized to
maximize the part formability. Alternative approaches, aimed at efficient process FEA
modeling are under development in several research institutes and companies. Three
main different strategies can be followed: a) Optimization Simulation Methods, b)
Feedback Control Simulation Methods, and c) Adaptive Simulation Methods
2.3.2.1 Optimization Simulation Methods
Optimization can be broadly divided into two main groups: a) static optimization and b)
dynamic optimization. In static optimization problems, design variables are time
invariant, such as optimizing dimensions of a mechanical component to minimize its
weight. There are two main methods to solve static optimization problems; gradient12
In dynamic
optimization problems, the design variables are time variant, such as an optimization of
flight trajectory control. One of the most powerful methods to solve the problems is
dynamic programming.
In metal forming, FEA simulations integrated with an optimization solver are used to
optimize either geometric parameters or process parameters in order to maximize
formability of that specific process. To understand the applications of optimization, the
literature review in various metal forming processes was conducted. In forging, the die
shapes are optimized to achieve the most uniform deformations (constant strain rates),
which improves metallurgy properties of the forged components [Fourment; 2001]. The
die profile was represented by a Bezier curve with a finite number of control points, see
Figure 2.7.a. Fourment (2001) applied Direct Differentiation Method (DDM) to
determine the objective function (i.e. strain rates) sensitivity to the change of design
variables (positions of the control points representing the die profile), and then BFGS
algorithm was used for the optimization through iterative FEA simulations. To avoid
complexity in calculating the derivatives of the objective functions, non-gradient
methods such as genetic algorithm were applied by Jo (2001), Kusiak (1996), and
Chung (1997) on the similar problem of forging die design.
13
Control points
a)
Figure 2.7:
b)
Design parameters are the control points (pi) describing a B-spline function of
loading path (Figure 2.7).
Objective function takes into account of element thickness variations after each
forming simulation run:
N h h
0
f ( p) = N1 i
i =0 h0
1
2
14
Constraint function represents the distance from the desired shape to the final part at
simulation end:
1
M 2 2
g ( p) = d i
i =0
Control theories have been applied in many industrial applications for many years, such
as control of temperatures in chemical processes. A controller regulates some quantities
to stabilize a process by automatically adjusting a variable(s) (controlled variable) in
real time. The simplest and most widely used control schemes are PID controllers. For
highly non-linear processes, non-mathematics based controllers, such as fuzzy logic
controller, and neural network controller are preferred. A few researchers have applied
feedback control schemes in conjunction with metal forming process FEA simulations
[Cao, 1994]. With the help of a feedback controller integrated into a process FEA code,
the process parameters can be adjusted at every simulation time step to achieve high
process formability predicted through the simulation.
The main difference in determination of process parameters through FEA using
optimization methods mentioned above and feedback controllers is apparent in the time
duration where corrective actions, i.e. adjusting process parameters, take place. The
optimization simulation method requires many simulation runs. After the end of a
simulation, a parameter correction is done and applied into the next simulation run with
the attempt to minimize the objective function value. A feedback controller adjusts
process parameters at every time step in one simulation run in order to maintain the
controlled quantities, i.e. formability, see Figure 2.8. The advantage of the feedback
15
control simulation method is that it requires less total computation time in predicting the
process parameters than the optimization simulation method does.
Cao (1994) controlled wrinkles and maximum strains in a conical cup drawing
simulation by automatically adjusting the binder force by using a PI controller. Thomas
(1998) further developed Caos work by introducing the control of stresses as well.
Grandhi (1993) and Feng (2000) implemented optimal feedback controllers in
simulation of forging processes. The controller tried to regulate the ram speed to track
the predefined strain rate of the part being forged. In tube hydroforming, Doege (2000)
applied fuzzy logic control theory to simulations of tube hydroforming. The controller
adjusted the internal pressure and axial feed curves in order to prevent wrinkles
throughout the process.
Process
Parameters
Parameter
Adjustment
Desired formability
(tend)
PID controller
Formability
Good part
(ti) to (t1+1)
Defect
Identification
Figure 2.8:
General flow chart of the feedback control simulation method for process
design in metal forming
This method makes use of both optimization method and feedback control method in
design of process parameters. The adjustments of process parameters are carried out at
each time step (or certain interval of simulation time step) during a single simulation
run similar to the feedback control method. However, the adjustments of process
parameters are done with the help of optimization methods. By this way, the automatic
design of the process parameters can be done quickly and in an optimized manner.
16
There have been two slightly different methods, proposed for THF process parameter
design, which fall into the adaptive simulation category:
ERC applied adaptive control theory combined with optimization in selecting of THF
loading paths [Strano, 2001a]. This adaptive simulation method uses a quadratic
objective function considering a wrinkle quantity (failure indicator) to be minimized at
each time step. To implement this method, a linear plant model is generated and
updated and each time step to describe evolution of the wrinkle quantity through the
forming time in a function of pressure and axial feed. The coefficients in this model are
evaluated at every time step.
Gelin (2002) devised a kind of adaptive simulation with function interpolation and
optimization techniques. In his paper, the thickness variation of the THF part is
minimized. To generate the objective function, a spline formulation was used to model
the evolution of the thickness variation in a function of pressure and axial feed. Unlike
the method mentioned above, this method requires many simulation runs in each time
step with perturbed process parameters to interpolate the thickness function. In respect
to the optimization simulation methods mentioned earlier, this method may be named
local-time optimization of process parameters.
17
CHAPTER 3.
TUBE HYDROFORMING PART AND PROCESS DESIGN
USING FEA MODELING
In all metal forming processes, part and process design is an essential step in successful
manufacturing of any products. Tube HydroForming (THF) process demands a lot of
engineering knowledge starting from the part design which is constrained by part
functionality and geometry, to the process design where appropriate combination of
internal pressure, axial feed, and counter punch pressure (if necessary) need to be
determined. It has always been of a primary concern in the industry to reduce the leadtime in part and process design developments and produce better parts with lower costs.
One of the most efficient ways to achieve this goal is utilize Finite Element Analysis
(FEA) during the part and process development stage. Specifically, due to the lack of
extensive knowledge in both analytical and experimental in THF, FE modeling of THF
processes is very useful in 1) reducing or even eliminating the need for trial-and-error in
the developing stage, and 2) optimizing the part and process to minimize the
manufacturing costs (i.e. increasing the robustness of the process thus reducing the
scrap rate).
This chapter discusses FE modeling of THF process and it applications. Considerations
of modeling any THF processes by FE simulation are given. Simulation and
experiments of hydroforming of a Y-shape is used as an example in this chapter. Some
simulation work was also conducted in an attempt to understand mechanics of the
deformation process. Through this study, effects of tubular part geometry (initial tube
length) and process parameters (pressure and axial feed) on hydroformability were also
explained. At last some THF part and process guidelines are given.
18
In FE modeling of any metal forming processes, a good understanding of the FEA code
is as important as an understanding of the process itself. Typically, A THF process
requires two motivational forces, i.e. axial force exerting on the tube ends and internal
pressure acting normally to the tube inner surface. These two forces (i.e. loading paths)
should be applied appropriately on the tube if a sound part is to be produced. In terms of
process design, FEA is used to verify and refine loading paths. This section first gives
an overview of the Y-shape hydroforming process. Then, FE modeling of the Y-shape
hydroforming is discussed. PAM-STAMP (explicit dynamic non-linear FEA code) is
used throughout this work. There are some considerations in using any dynamic codes
to simulate THF processes. These considerations are also discussed at the end of this
section.
3.1.1 Hydroforming of Y-shape
Y-shapes are (see Figure 3.1) commonly used as fitting parts in automotive exhaust
manifolds. The parts are usually made of stainless steels 304, which is rust resistant.
Typically, in hydroforming of these Y-shapes, a counter punch is usually used to
support the protrusion tip while it is growing. By this way, premature protrusion
bursting is delayed and thus increasing the useful height of the protrusion (useful
protrusion height is defined in Section 3.3). However, the use of a counter punch adds
one more process parameter to be controlled properly with the axial feeds and internal
pressure. The Y-shape studied in this work has a protrusion that is angled to the tube
axis by 60 degrees. The detailed dimensions of the part are given in Figure 3.1. The
load paths and tube material properties of this part will be discussed shortly in the next
section.
19
Counter Punch
Internal Pressure
a)
Axial Punches
b)
c)
Figure 3.1:
Figure 3.2 shows pictures of Y-shape hydroforming procedure. These pictures are taken
from Y-shape hydroforming experiments (conducted at research facility of SPS,
Germany) that are to be discussed later in this chapter. General specifications of the
SPS hydroforming press used are given in Figure 3.3. Please note that in these forming
experiments, from which the photos are taken, the counter punch was not used. The
process descriptions are given below.
1. Upper and lower die inserts were installed onto the press. The axial punches were
connected to the pressure intensifiers with high-pressure hoses. Figure 3.2.a shows
the lower die insert.
2. A rough drawing of the axial punch is shown in Figure 3.2.b. The punch has a
conical shape at the tube-punch contact area. In this case, the tube blank has OD =
50 mm, and to = 1.5 mm. For good sealing performance, the punches were designed
to have 5 mm of sealing distance (i.e. initial axial punch displacement of 5 mm
without internal pressure build-up).
3. The tubes were spray-lubed with a solid film lubricant (Gleitmo 965). The tubes
were allowed to air-dry for about 2-3 hours.
4. Figure 3.2.b shows how a tube blank was placed and positioned in the lower die
cavity. The axial punches were fully retracted to their home positions, while the
counter punch axis was not in operation and rested at its home position.
5. Figure 3.2.c shows a completely formed Y-shape (without any use of a counter
punch). It is shown that the length of the final part has been shortened due to the
axial feeds.
respectively. The maximum internal pressure was 600 bars. The protrusion height,
Hp, was measured and considered as formability index in this study.
21
Counter Punch
Axis (no
counter punch
attached)
Right Punch
Left Punch
Y-shape
a)
Sealing Distance
5 mm
Tube Blank
1.5 mm
LRO
LLO
b)
Y-shape
Hp
40 mm
c)
Figure 3.2:
80 mm
LLO
LRO
22
Figure 3.3:
Closing Force
24.7 kN
40 mm/s
A successful THF process largely depends on process parameters (loading paths), part
geometry, initial geometry of the tube, and interface friction conditions. In THF of a
given tubular part geometry, the main process parameters to be determined are the
following:
Axial feeds vs. time
Internal pressure vs. time
Counter punch force vs. time (for some part geometries with protrusions)
Initial estimates of these parameters can be obtained from simple metal forming
equations. Typically, these estimated parameters are not accurate depending on the
complexity of part geometry and non-linearity of material properties. The parameters
will then be tried out and tuned through iterative FEA simulations till satisfactory
results are obtained. It is certainly preferred that the initial process parameters be
reasonably well estimated. Thus, the number of iterative simulations, necessary to
obtain the best process conditions, can be reduced. Design of the process parameters for
Y-shape hydroforming is discussed here. Analytical equations were used to determine
23
(at the left, daL, and right, daR, tube ends) necessary to form a Y-shape with a desired
protrusion height (H), Figure 3.4. The original tube wall thickness is assumed to remain
unchanged. However, it is also possible to assume that the part final thickness is linearly
distributed (thickened at the tube ends and thinned at the protrusion tip), which is more
realistic. For simplicity, in this case study, the part final thickness distribution is
assumed to the same as the original tube wall thickness. The material volume at the
protrusion section of the Y-shape is converted to obtain the necessary axial feed. The Yshape geometry with a fixed angle is shown in Figure 3.1. The tube blank outside
diameter (D0) is 50.5 mm (1.988), tube initial thickness (t0) is 1.5 mm (0.059), the
protrusion diameter (Dp) is 50.5 mm (1.988).
The volume of material formed into the protrusion area was calculated. Each half of the
protrusion was assumed to have been contributed from the axial feed applied on the
corresponding side of the protrusion. For this Y-shape geometry with the specified
angles and Dp=D0, the relationship approximated between necessary axial feeds to the
protrusion height (H) indicated that the left axial feed (daL) and right axial feed (daR)
should be about H and 2H, respectively.
This procedure to estimate the axial feeds can be applied to hydroforming of any other
Y-shape geometries. Once the axial feeds have been estimated, the initial tube length
(the sum of LL0 and LR0) can be calculated by adding the approximated axial feeds (daL,
daR) to the designed final Y-shape lengths (LL1 and LR1). It should be noted that this
axial feed calculated is just an initial estimate. The necessary axial feed also depends
on the length of the Y-shape, tube material, and interface friction conditions. Therefore,
a few FEA simulations are usually necessary to optimize the initial tube length.
24
Dp
D
t0
LL0
LR0
H
da L
LL1
Geometry
D
Dp
t0
H
LR1
Description
Protrusion angle
Tube diameter (outside)
Protrusion diameter (outside)
Tube initial wall thickness
Protrusion height
Geometry
daL
daR
LL0
LR0
Figure 3.4:
da R
Value
60 degrees
50.5 mm (1.988)
50.5 mm (1.988)
1.5 mm (0.059)
to be designed
Estimation formulas
H
2H
LL1 + daL
LR1 + daR
Internal Pressure Limits: A yielding pressure (Pi)y is the minimum pressure required to
initiate deformation in hydroforming process. The level of this yielding pressure varies
depending on tube material and geometry. An equation to approximate this yielding
pressure is derived; based on a simple axisymmetric expansion of a tube with fixed
ends, see Equation (1). Although the calculated yielding pressure is accurate only for a
simple tube expansion with fixed ends, it is also a good initial guess for hydroforming
of more complex parts (i.e. Y-shapes) with axial feed applied.
(Pi)y
2t 0
( D0 t 0 )
Bursting pressure (Pi)b is the maximum pressure that expands a tube without bursting.
Equation (2) estimates the bursting pressure for a Y-shape hydroforming in which no
counter punch is applied. It is based on a balanced biaxial bulging of sheet metals. This
equation is used because balanced biaxial tensile state prevails, approximately, in the
top area of Y-shape protrusion with no counter punch applied [Jirathearanat, 2000b].
Clearly, the bursting pressure is expected to be larger than that calculated by equation
(2) when a counter punch is applied.
(Pi)b
4 t0
( D p t0 )
Eq. 2
Calibrating pressure (Pi)max is the internal pressure level required to form/coin a tube
wall into small die corners (coining). The calibrating pressure can be estimated by using
Equation (3) [Koc, 2002].
(Pi)max
2
f
3
rb
ln r t
b 0
With all the estimated pressure limits, i.e. yielding, bursting, and calibrating pressures,
an initial pressure curve for THF of the corresponding Y-shape can be constructed using
linear lines connecting these pressure limits. The optimal pressure curve will be
26
determined through iterative FEA simulations. In cases where initial estimates of the
process parameters are difficult to determine such as complex automotive tubular parts,
a kind of process optimization scheme integrated in FEA should be used to design the
process parameters. This will be discussed in Chapter 5 and 6.
Counter Punch Force: Due to complexity of the deformation, there is no simple formula
available to analytically determine appropriate counter punch force curve versus time.
However, the counter punch force profile can be estimated through FEA simulations.
The displacement curve governing the counter punch movement can be modified until a
good Y-shape with a designed protrusion height can be formed. Then, the necessary
counter punch force can be obtained from the contact force between the counter-punchtube-protrusion interface. Alternatively, one may also utilize the optimization based
simulation method to determine optimal counter punch force curve versus time. This
technique is demonstrated in Chapter 5.
The estimates of axial feed, pressure limits and counter punch force were used to
construct the linear loading paths. These linear paths were then improved through
conducting FE simulations. The process modeling of this process will be discussed
shortly after this section. The final loading paths used in the experiments are shown in
Figure 3.5. Based on the axial punch velocity of 4 and 8 mm/s for the left and right
punch, respectively, the total forming time was calculated to be 12 sec (i.e. 10 sec for
hydroforming and the last 2 sec for calibrating). It should be noted that, in a real
forming process, the total forming time can be sped up depending on the capability of
the hydroforming press. It is, however, important that the relationship of each process
parameters is held at all time (e.g. plot of pressure versus axial feed remains unchanged
regardless of total forming time applied). A small study [Jirathearanat 2000] showed
that forming speed might result in different part quality. Different interface friction
conditions as a result of different forming speeds were assumed to be the cause this
phenomenon.
27
23200
1400
20300
1200
17400
1000
14500
800
11600
600
8700
400
5800
200
2900
t = 0 1 sec: sealing
t = 1 11 sec: hydroforming
t = 11 13 sec: calibrating
(Psi)
Pressure (bars)
1600
0
0
a)
10
15
Time (sec)
100
60
Sealing displacement = 5 mm
3.1
2.4
40
1.6
20
0.8
(inch)
80
3.9
0.0
0
10
15
-20
-0.8
140
30
120
25
100
20
Displacement
80
15
60
10
40
Force
20
Time (sec)
b)
0
0
c)
Figure 3.5:
10
12
14
Time (sec)
28
Die mesh
Counter punch mesh
Axial punch mesh
Tube mesh
a)
b)
Figure 3.6:
= K ( 0 + ) n
50 mm (1.968)
1.5 mm (0.059)
0.584
Pre-strain ( 0 )
0.06
30
Constitutive modeling and flow stress equation: Hills 1948 constitutive model is used
to represent the elastic-plastic-planar-anisotropic behavior of the tubular material. This
constitutive model works well with most Low Carbon steels and Stainless steels. There
are two main methods of tube manufacturing: a) roll forming and b) extrusion. Roll
formed tubing seems to behave anisotropically, which is derived from the original sheet
material. Due to the complexity of the tube making process, the real anisotropy of the
tubular material is still not known. Typically, the anisotropy value of the original sheet
material, if available, is used in the THF simulation. In this case, the tube is assumed
isotropic.
Sliding friction contact: There are two main sliding friction contact algorithms available
in PAM-STAMP: a) penalty method and b) Lagrangian method. Lagrangian method is
normally chosen over penalty method in tube hydroforming simulations. Coulombs
friction of 0.05-0.06 is normally assumed for the tube-die interface friction.
Loading: The explicit time integration scheme of PAM-STAMP requires very small
time steps (about 10E-5 sec) to guarantee reliable and stable solutions. To reduce the
computational time, the work-piece mass is increased or forming time is sped up
artificially. Typically, in sheet forming simulations, the forming time is sped up by
approximately 1000 times without any dynamic effect (i.e. erroneous stresses due to
mass inertia). In this Y-shape simulation, the loading curves applied are sped up by
1000 times. Therefore, the total forming time is now 12 ms and the left and right punch
speeds are now 4 and 8 mm/ms.
Pressure is applied to the interior of the tube elements along their normal directions.
The fluid cell option in PAM-STAMP builds up internal pressure by an artificial fluid
medium flowing into the fluid cell (i.e. closed deforming tube). A limiting pressure
curve is input as to regulate the pressure build up inside the fluid cell through an
artificial relieve valve.
31
Axial feeds from both tube ends can be prescribed to the axial punches using a)
displacement boundary condition or b) velocity boundary condition. The velocity
boundary condition is safer to apply than using the displacement boundary condition in
terms of preventing erroneous inertial stresses from happening in the simulation. As a
rule of thumb, which is borrowed from sheet metal forming simulation, the speeds of
tooling components such as those of axial punches should be kept below 10-15 mm/s to
minimize such erroneous stresses (dynamic effect).
As mentioned earlier, accurate counter punch force curve versus time is difficult to
estimate a priori due to part complexity and material non-linearity. The most convenient
way to prescribe a boundary condition on the counter punch is by using velocity
boundary condition. However, this is an iterative procedure.
3.1.2.2 FE Simulation Results and Verification
The estimated process parameters (pressure, axial feeds, and counter punch force) were
refined through conducting FE simulations, i.e. the process parameters are adjusted in
the consecutive simulations based on the results of the previous simulations. Through
this exercise, a sound Y-shape was simulated with part maximum thinning of 23% and
without any wrinkles. These loading paths, determined from conducting FE simulations,
were used to hydroform the real Y-shapes. The procedure of Y-shape hydroforming is
discussed in the beginning of this chapter. The process parameters were measured from
the real process as shown in Figure 3.7.
From the simulation results, deformation of the Y-shape can be observed clearly. The
Y-shape hydroforming process can be roughly divided into three main stages, see
Figure 3.7, a) free expansion, b) expansion against a counter punch, and c) calibration.
The counter punch is positioned in the die just above the left die corner radius such that
it would not pinch the growing protrusion in the early hydroforming stage. After the
protrusion has come in contact with the counter punch, the counter punch will slide
32
slowly upwards as it is supporting the growing protrusion and come to a stop during the
calibration stage.
The internal pressure curve (input and measured from the press), shown in Figure 3.7.a,
consists of two main stages, i.e., forming stage (1-11 sec) (including free expansion and
expansion against the counter punch) and calibrating stage (11-15 sec). During the
forming stage, the pressure goes up from 0 bar to 800 bars, the left and right axial feeds
are 40 mm and 80 mm, respectively (see Figure 3.7.b).
calibrating stage because the calibrating pressure is usually very high, so that the tubedie interface friction force becomes too large for the tube material to be fed in. Figure
3.7.c shows the counter punch force curve and the displacement of the counter punch,
which determines the protrusion height of the Y-shape.
The Y-shapes hydroformed in the experiments were cut and measured for thickness
distributions along the tube axis direction. Figure 3.8 shows a comparison of thickness
distributions of the Y-shape simulated with FEA and measured from the experiments.
The comparison results indicate that FEA simulation accurately predicted the metal
flow in the Y-shapes when compared with the real part.
33
Time = 0
Protrusion comes in contact
with the counter punch
1600
23200
1400
20300
1200
17400
1000
14500
800
11600
600
8700
400
5800
200
2900
a)
(Psi)
Pressure (bars)
0
0
10
15
Time (sec)
100
80
3.1
60
2.4
40
1.6
20
0.8
0.0
0
b)
10
15
-20
-0.8
Time (sec)
Time = 11
140
25
Counter punch
displacement
100
20
80
15
60
10
40
Counter punch
force
20
c)
30
120
Calibrated
Time = 15
(inch)
Time = 5
3.9
0
0
10
12
14
Time (sec)
Figure 3.7:
FEA simulation demonstrates intermediate hydroforming steps of a Yshape, a) Pressure, b) axial feeds and c) counter punch force versus time
curves used to hydroform SS 304 Y-shapes
34
3.00
EXP # 9 ( 2 samples)
2.75
FEM # 1
Thickness (mm)
2.50
2.25
2.00
1.75
1.50
1.25
1.00
0
Figure 3.8:
50
100
150
Curvilinear Distance (mm)
200
250
35
36
Commercial
FE Codes
Advantages
PAM-STAMP, LS-DYNA
Table 3.1:
38
2.50
Thickness (mm)
2.25
2.00
1.75
1.50
1.25
FEM
EXP1-sample1
EXP1-sample2
EXP1-sample3
Initial Thickness
1.00
0.75
0.50
0
Figure 3.9:
50
150
200
250
Stage: Intermediate
FE: 1 layer - brick elements
Figure 3.10:
100
Stage: Final
FE: Multiple layers brick elements
39
25
20
mesh size = 2.5 x 2.5
15
0
0
20
40
60
80
100
120
140
Figure 3.11:
160
180
Part geometry design can greatly affect part formability. Therefore, hydroformability of
THF parts should be considered early on in the designing stage. As hydroforming is
gaining momentum in the industry to replace traditional stamping of automotive parts,
part design methodology has to be executed with added constraints imposed by the tube
hydroforming process itself. Small part corner radii demand very high hydraulic
pressures. Only certain sizes of part corner radii can be hydroformed depending on the
base material, wall thickness and press capacity. The length of the part measured along
the main axis of the tube (so called spline length) also plays an important role in
imparting sound parts. In this study, the effect of part spline length to hydroformability
is studied through hydroforming of the Y-shape; discussed earlier, see Figure 3.2.
3.2.1 Tube Spline Length Effect
The Y-shape is considered here again to investigate the effect of part spline length to
the part formability. This knowledge is particularly useful when working with long
automotive structural parts such as an engine cradle whose part geometrical features
such as bulges, protrusions, and bends are located along the part spline axis, Figure
3.12. Axial feed from both tube ends may only facilitate the forming of part geometrical
features located near the tube ends but may not benefit in forming the ones that are far
removed from the tube ends. Thus, those part geometrical features end up being formed
only by pure expansion, which are subject to premature bursting. Typically, in the
engine cradle case, tube performing/crushing is done such that extra material is
accumulated in the large expansion area to improve the formability, Figure 3.12.a.
It is clear now that the effectiveness of axial feeding depends on ability of the tube
material to flow from the guiding zones (i.e. tube end areas) into to the tube expansion
zone (i.e. part geometrical feature areas). This ability to flow of the metal is influenced
by many factors such as part geometrical features adjacent to the tube ends, tube-die
interface frictional force, tube material strain hardening, and part spline length.
41
Large
expansion
far away from
tube end
Axial feed
Axial feed
a)
3 protrusion
features
2 bend
features
b)
Figure 3.12:
42
Y-shapes with different spline lengths were studied in this work. The part spline lengths
are measured (LL1 and LR1) along the tube axis of the final part, see Figure 3.13. SS304
tubing with 1.5mm wall thickness and 50 mm in diameter was used. There were three
sets of experiments. In each experiment, the final part spline lengths on the left (LL1)
and right (LR1) ends are varied, see Figure 3.13. The pressure and axial feed amounts
were all the same for all the experiments, Figure 3.14. The pressure and axial feed
curves were determined from the metal forming equations as discussed earlier and
refined through iterative FE simulations. The axial feed from each end is 40 mm and 80
mm for the left and right end, respectively. The initial tube lengths on the left and right
sides (LL0 and LR0) were determined simply by adding LL1 and LR1 with the
corresponding axial feeds; LL0 = LL1 + da L and LR0 = LR1 + da R. The hydroforming
experiments were conducted at SPS research center, Aalen, Germany [Jirathearanat,
2001b].
Three tube samples were formed in each forming experiment. Figure 3.15 summarizes
and compares protrusion heights (Hp) measured from parts of experiments 1 to 3, refer
to Figure 3.13. It can be concluded from the figure that different part spline lengths
(LL1 and LR1) affect the obtainable protrusion height of the Y-shapes. It can be shown
that variations of the protrusion heights among the three part samples from the same
experiment are much less than those caused by changing the part spline lengths. It is
now obvious that reduction of the right spline length (LR1) from 80 mm to 45 mm
increased the protrusion height, Hp, by 5 mm; comparing experiments of LL1 = 120 and
LR1 = 80 mm to LL1 = 120 and LR1 = 45 mm. More improvement on the protrusion
height was achieved by reducing the left spline length; comparing experiments of LL1 =
120 and LR1 = 45 mm to LL1 = 85 and LR1 = 45 mm. The conclusion drawn from the
results, seen in Figure 3.15, is that, with this particular Y-shape geometry, a larger
protrusion height HP can be achieved by reducing LR1 and LL1. No significant variations
in thinning percentage were observed between different samples.
thinning was in the range of 19.5% to 21.5%.
43
The maximum
H
da L
da R
LL1
LR1
LL0
Y-shape
Spline Length
EXP. #
LL1 (mm)
LR1 (mm)
120
80
120
45
LR0
Axial Feed
da L and da R
(mm)
LRO
(mm)
160
160
160
125
Material
DIA x Thick
(mm)
SS304
40 - 80
50 x 1.5
3
Figure 3.13:
85
125
45
125
44
10000
700
9000
8000
500
7000
400
6000
5000
300
4000
200
3000
(Psi)
Internal Pressure
(bars)
600
2000
100
1000
0
0
10
Time (sec)
90
3.5
80
3.0
2.5
60
50
2.0
40
1.5
30
(Inch)
70
1.0
20
10
0.5
0.0
0
10
Time (sec)
Figure 3.14:
Internal pressure versus time curve and axial feed versus time curves
used in all the experiments, see Figure 3.13
45
avg. Hp
115.2 mm
120-80
120
80
avg. Hp
120.4 mm
120-45
120
45
avg. Hp
122.1 mm
85-45
85
45
110
112
114
116
118
120
122
Figure 3.15:
46
124
The use of proper internal pressure and axial feeding versus time curves is very crucial
in THF. Typically, yielding pressure and some axial feed (for sealing purposes) are
initially applied to start plastic deformation of the tube. Then, the pressure is raised to
expand the protrusion as well as prevent/straighten wrinkles, but controlled under the
bursting level. Axial feeding is applied simultaneously with the pressure forcing the
material to flow into the expansion zone.
This study focuses on hydroforming of a T-shape. Particularly, hydroforming of Tshapes that only one-sided axial feeding is effective due to the part geometry. This
forming difficulty arises often in hydroforming of automotive structural parts with large
ratios of part length to diameter. Figure 3.16 shows a simplified structural part that is
consisted of many part geometrical features (i.e. two bends, a bulge, and a T-shape).
Due to the bend on the left side, left axial feeding (daL) will not improve formability of
the part. The right axial feeding (daR) is possible but only beneficial to the T-shape
forming. A quick part geometry analysis would suggest that the bulge will be
hydroformed through pure expansion, and the T-shape will be hydroformed with only
one-sided axial feeding. Several simulations were conducted here to study the effect of
pressure, one-sided axial feed, counter punch force curves on the T-shape
hydroformability.
Figure 3.17 shows the geometry of the initial tube and the final T-Shape part. Four
different initial lengths of tubes for simulations with a set of four different axial feeds
(i.e. daR = 30, 50, 70, 90 mm) were chosen in such a way that the final part spline length
will be the same for all simulated parts regardless of the amount of axial feed, i.e. the
distance L1 marked in the final part will remain constant for all the simulations. Only
the protrusion height will vary from one simulation to the other because of variation in
axial feed (daR) and internal pressure. The useful portion of protrusion height (Huseful)
may be regarded as the total protrusion height (Htotal) minus the corner radius at top, see
Figure 3.17. All the simulations were modeled with a LCS material expressed by the
47
power law = 500(0.02 + ) 0.19 MPa ( = 72.5(0.02 + ) 0.19 ksi). The coefficient of
friction was kept as 0.05 in all the simulations.
Axial Feed and Internal Pressure
For the present study, simple linear loading paths were chosen for internal pressure and
axial feed, Figure 3.18, since such linear loading paths had been used successfully to
form T-shapes at the ERC [Jirathearanat, 1999]. According to the past experiments and
simulations for low carbon steels, thinning of 30% + 0.5% was selected as a failure
criterion to stop the simulations. Any value of the thinning above this limit is
considered as unacceptable, and the simulation is re-run with suitable changes (i.e.
decrease or increase the maximum pressure) to obtain the part maximum thinning
within the prescribed limits.
Axial feeds of 30 mm, 50 mm, 70 mm and 90 mm were applied in the all simulations.
The internal pressure was quickly increased to reach the yield pressure py and then
gradually increased linearly to reach a specified maximum value pm in the given process
time (Figure 3.18). The maximum pressure pm was to be such that the thinning did not
exceed the set limit of 30% + 0.5%. However, this value of pm can only be
approximated using the equations discussed earlier. In the actual process, the protrusion
portion will not be symmetrical (due to one-sided axial feeding) and the thickness will
not be uniform over the protrusion. Moreover the area of maximum thinning will have a
reduced radius of curvature that will increase the internal pressure required to cause
more thinning. Hence it is more appropriate to determine the pm through iterative FE
simulations.
48
Feature # 2: Bend
da R > 0
Feature # 3: Bulge
Feature # 1: T-shape
One sided axial feed (da R)
T-shape hydroforming
Figure 3.16:
Htotal
R = 10
Huseful
T-shape
da R > 0
da L = 0
t0 = 2
Initial Tube
D = 63.5
da R
L1
L0
Figure 3.17:
Geometry of the T-shape die cavity and part geometry with one-side
axial feeding (dimensions are in mm; 25.4 mm = 1 in.)
49
100
90
Feed 90 mm
80
Feed 70 mm
70
Feed 50 mm
60
Feed 30 mm
50
40
30
20
10
0
0
10
12
10
12
Tim e (m s)
0.0700
0.0600
High
0.0500
Low
Med
0.0400
0.0300
0.0200
0.0100
0.0000
0
Time (ms)
Figure 3.18:
a) axial feed versus time curves used in all simulation cases (25.4 mm =
1 in), and b) pressure versus time curves corresponding to the different
axial feeds (1 GPa = 145,038 psi)
50
51
70
0.042
H total
Max. Press
50
Height (mm)
0.041
H useful
0.04
Htotal
40
Huseful
30
0.039
da
0.038
20
10
0.037
0.036
30
40
50
60
70
80
90
Figure 3.19:
Effect of axial feed on protrusion height (all the simulated parts have
maximum thinning of 30%)
65
Feed 90 mm
60
Feed 70 mm
Feed 50 mm
55
Feed 30 mm
50
45
40
35
30
25
0.03200
0.03400
0.03600
0.03800
0.04000
0.04200
Figure 3.20:
60
53
30
Counter punch force (kN)
Trial #1
25
Trial #2
Trial #3
20
15
10
5
0
0
10
12
Time (ms)
60
30
H total
H useful
50
25
Height (mm)
40
20
30
15
Ht
Hu
Hu
20
10
10
Max. Thin%
0
0
10
15
20
25
30
Figure 3.21:
CHAPTER 4.
SYSTEMATIC APPROACH TO SELECT LOADING PATH
USING PROCESS FEA SIMULATION
4.1 Self-Feeding Simulation Approach
55
No interface friction
No end constraints
Pure expansion
No interface friction
Figure 4.1:
(Analytical models)
SF simulation
Normal simulation
dao L, dao R
Excessive thinning?
Yes
Scaled up by asf
Figure 4.2:
No
Good part
56
with a linear pressure curve from [0 to PMAX ] to expand the part to the die
cavity,
without any boundary conditions on the tube ends no external axial feeds, and
without tube-die interface friction,
self
self
to obtain the SF axial feeds: d left
(left feed) and d right
(right feed) from the SF
simulation
Step 3. Rerun self-feeding simulation:
feed amount) until a good part with an acceptable maximum thinning can be
formed:
with scaled-up axial feeds a sf (d left ) and a sf (d right ) , scaling factor a sf > 1
There exists an upper limit to this scaling factor, a sf . This can be estimated by applying
volume constancy onto the part with uniform thickness assumption to calculate the
max
maximum axial feed necessary. Then, the maximum axial feed at left end, d left
, and
max
, can be determined based on the SF axial feed ratio d left : d right found in
right end, d right
max
sf
max
d left
d left
max
d right
d right
Therefore, the possible range of the scaling factor runs between 1 < a sf < a sfmax , in which
the optimal scaling factor can be found. The optimal a sf should be such that it yields
the axial feeds (left and right tube ends) that form the part with an acceptable part
maximum thinning depending on the part materials (e.g. 25-30% for Low carbon steels
and 10-15% for Aluminum alloys), and without any wrinkles or minimum wrinkles
depending part functionality.
Additionally, some proper optimization schemes can be implemented here to automate
the search for the optimum scale factor (asf). For example, the scale factor (asf) can be
the design parameter describing the loading path to be determined through optimizing
hydroformability of the part (minimum thinning with no wrinkles on the final part).
However, this method limits the optimized load path to be only within a family of
curves imparted from the SF simulation approach.
58
data from Table 4.1. The bent tube dimensions are given by relative parameters, see
Figure 4.4. FEA simulations of all the three bends were conducted in a total of six
simulations in order; a bending simulation for the 1st bend followed by a springback
simulation then repeat the same simulation procedure for the other two bends. General
descriptions of tube bending operations and FEA modeling are discussed in a past ERC
report No. 99-R-01 [Shr, 1999].
Thinning distributions along the longitudinal top and bottom profiles of the simulated
bent tube (after spring back simulation) are shown in Figure 4.5. It can be seen that the
bends on the left and right sides (along profile A) have larger thinning values than that
in the middle of the tube. This is because of the bend radius of the middle bend is larger
than those of the bends on the sides. Therefore, one needs to pay attention to the part
formability of the part areas on the sides, particularly, during expansion in the
hydroforming process.
59
Left end
Right end
Largest expansion
Figure 4.3:
Lo
1033 mm
OD
63.5 mm
62.35 mm
to
2.3 mm
669 MPa
0.173
Initial strain
0.01
Table 4.1:
60
2 nd Bend
Radius: ~ 1.4R mm
Angle: X
OD
63.5 mm
Angle: X
Angle: 2X
st
1 Bend
Radius: R mm
Figure 4.4:
1000 mm
3 rd Bend
Radius: R mm
A
20.00
Profile A
Profile B
B
15.00
Thinning %
10.00
5.00
0.00
0
100
200
300
400
500
600
700
800
900
1000
1100
-5.00
-10.00
-15.00
Curvilinear Distance (mm)
Figure 4.5:
61
Figure 4.6 shows the pressure curve (input in the SF simulation, no tube ends
constrained, no tube-die interface friction applied), and the resultant velocity curves of
the left and right punches. From the resultant velocity curves, the following adjustments
were made to determine the optimized loading paths:
The slope of the pressure curve was reduced in the SF simulation #2; see Figure
4.7, at the portion where the resultant punch velocity curves (from SF simulation,
see Figure 4.6) seemed to be abruptly increasing (large accelerations).
With the smoothened axial feeds, simulation #3 can be run. In this simulation,
tube-die interface friction was applied. Coulomb friction coefficient of 0.06 was
used in all the bending-crushing-hydroforming simulations. The loading paths are
shown in Figure 4.9 (simulation #3)
From the simulation #3, the axial feed curves were to be scaled up depending on
the severity of the maximum part thinning that occurred in the resultant part. In
this case study, the best scale factor, asf, was 1.5.
With the axial feed curves applied in simulation #3 multiplied by 1.5, simulation
#4 was conducted. Figure 4.10 shows the development of the amount of the axial
feed curves necessary for successful hydroforming of the part. Figure 4.11 shows
the optimized loading paths applied in simulation #4. Table 4.2 summarizes all
the simulations conducted through SF simulation method to determine the proper
loading paths.
62
2250
10
2000
Pressue (bar)
1500
1250
0
1000
-2
750
-4
500
1750
-6
Input pressure
250
-8
-10
0
10
12
14
16
18
20
22
24
26
Time (sec)
Figure 4.6:
Simulation #1, SF: input pressure and output punch velocity curves
2250
6.0
2000
4.0
1750
2.0
1250
0.0
1000
-2.0
Velocity (mm/ms)
Pressure (bar)
500
-4.0
250
0
-6.0
0
Figure 4.7:
10
12 14 16
Time (sec)
18
20
22
24
26
6.0
5.0
4.0
Velocity (mm/ms)
3.0
2.0
1.0
0.0
-1.0
10
12
14
16
18
20
22
24
26
-2.0
-3.0
-4.0
Time (sec)
Figure 4.8:
Modified axial feed velocity curves (the right axial feed velocity is
represented in negative values, left axial feed is in positive values)
6
2000
1750
Pressure (bar)
1500
2
1250
1000
-2
Pressure
500
-4
250
0
-6
0
10
12
14
16
18
20
22
24
26
Time (sec)
Figure 4.9:
45.0
35.0
Left
25.0
15.0
5.0
-5.0 0
10
15
20
25
30
-15.0
-25.0
Right
-35.0
-45.0
Time (sec)
Figure 4.10:
2250
2000
50
41 mm
40
Pressure (bar)
20
1500
Pressure
1250
10
0
1000
-10
750
-20
500
-30
37 mm
250
-40
-50
0
10
12
14
16
18
20
22
24
30
1750
26
Time (sec)
Figure 4.11:
Optimized loading paths from SF: pressure, left axial feed, right axial
feed
65
Simulation
Type of Simulation
Description of Simulation
Max.
Thinning
%
Wrinkle
#1
Self feeding
32%
Wrinkle-free
#2
Self feeding
36%
Wrinkle-free
#3
43%
Small wrinkle,
but removed
later,
burst part
23%
Small wrinkle,
but removed
later, sound
part
asf = 1
#4
asf = 1.5
Table 4.2:
Figure 4.12 shows some of the important intermediate forming steps from the
simulation. The crushing operation was done during the first 2 sec. of the entire forming
time (25 sec). Pressure was not applied during these first 2 sec. Then, the pressure was
increased from 0 to 260 MPa (yielding pressure calculated based on the materials
yielding stress) during 2-3 sec. During this time the punches started to feed material.
The early application of the axial feeds (i.e. axial feed before the internal pressure
reaches the yielding pressure) was done in order to push more material into the die as
much as possible. Although wrinkles appeared as a result of that, they were straightened
out later in the calibration process. See Figure 4.12, a large wrinkle on the left side
occurred during time 2-5 sec. Then, it was removed towards the end of the process.
In Figure 4.11, a pressure curve of 260 to 755 MPa was applied during 3-17 sec. It
should be pointed out that the slope of the pressure curve during 3-17sec. was much
smaller than the other portions of the same curve. This was done in order to slow down
66
the rate of material necessary to be fed into the expansion zone. Consequently, the left
and right punches could be controlled to move at a reasonable speed to feed an adequate
amount of material into the expansion zone. The pressure of 755-2000 MPa was
applied during the rest of the forming time (17-25 sec) for calibration purpose.
It should be noted that the relation between the axial feeds and the pressure, see Figure
4.11, should be held at all time. During the real hydroforming, the forming time
duration might be changed depending on capability of the press (for example, forming
time can be shortened from 25 sec to 12.5 sec or it can be elongated to 50 sec).
However, if the punch velocities are to be input to the press, they have to be
recalculated based on the required axial feeds (for example, if the forming time is
shortened from 25 to 12.5 sec, the velocities have to be doubled).
Figure 4.13 shows the thinning contour plotted on the final simulated part, and also
compares the thickness distribution of the part from simulation and experiment at the
most critical areas. The most thinned areas according to the simulation results were in
the areas A (flat area) with thinning of 17% and B (corner area) with thinning of 22.5%.
A sample of the automotive part was hydroformed at KS, Japan, with the same pressure
and axial feed curves, see Figure 4.11, as used to conduct the simulation. The part was
successfully hydroformed without any wrinkles or fracture. The thickness
measurements of the real part showed that the most thinned area was at the area A with
the thinning of 23%, which was in accordance with the highly thinned area A from the
simulation results.
67
T = 0 sec
Wrinkle
T = 5 sec, wrinkles appear
Figure 4.12:
A
B
Most thinned
Thickness (mm)
Thinning (%)
Locations
Figure 4.13:
FEM
EXP.
FEM
EXP
1.91
1.76
17.0
23.0
1.78
1.89
22.5
18.0
69
Figure 4.15 and Figure 4.16 show the pressure curves and axial feed curves found
during the application of the SF simulation approach for hydroforming of the cross
member. Due to its confidentiality, the dimensions of the part cannot be disclosed. The
following are the steps conducted in determining the proper LP:
Run the 1st SF simulation (indicated as SF#1 in Figure 4.15) with a linear pressure
curve [0 calibrated pressure]. Run the 2nd SF simulation (indicated as SF#2 in
Figure 4.15) with a modified pressure curve in order to reduce the nodal velocities
at the tube end.
70
Left End
Upper Die
Lower Die
Right End
Side Segmented Dies
Figure 4.14:
Material
Lo
1250 mm
OD
101.6 mm
to
3.8 mm
590 MPa
0.223
Initial strain
0.0188
Table 4.3:
Smoothen the end nodal velocity curves obtained from SF#2, then used them as
initial baseline axial feed rate curves in the subsequent normal simulations (i.e.
with interface friction) where these curves are scaled up by different asf factors.
Determine the good axial feeds from normal simulations (friction applied): Run
normal simulations (with friction) with the minimum axial feeds (smoothened
axial feed rate curves from SF#2) scaled up by a factor asf until a good part is
formed, i.e. a part with an acceptable thinning (<30%). The determination of the
best factor asf can be done considering the resulting part maximum thinning, see
Figure 4.16.
72
12
10
2000
8
6
Pressure SF #1
1000
Pressure SF #2
0
0
-1000
10
12
14
16
18
20
24
-2
-4
Nodal Velocity SF # 2
-6
Nodal Velocity SF # 1
-2000
22
-8
-10
-3000
-12
Pressure (bar)
3000
Time (sec)
Figure 4.15:
Pressure curves and corresponding nodal velocity from the right tube
end, obtained from SF simulations
LEFT
Case C: asf = 5 (max. thinning = 20%)
100
80
60
40
20
0
-20
10
12
14
16
18
20
22
-40
SF#2:
-60
Case A: asf = 3
-80
Case B: asf = 4
-100
Case C: asf = 5
24
RIGHT
Time (sec)
Figure 4.16:
Plots of axial feeds (left and right tube ends) selected through SF
simulation approach
73
Simulation Name
Scaling
Factor asf
Left Axial
Feed (mm)
Right Axial
Feed (mm)
Maximum
thinning %
Wrinkles?
SF # 2
18.7
20.8
42
No
Case A
46.2
54.5
34
No
Case B
61.7
72.7
27
No
Case C
77.1
90.8
20
Yes (Small)
80
3000
left
Pressure (bar)
2500
70
right
60
2000
50
40
1500
30
1000
20
Pressure
500
Table 4.4:
10
0
0
10
20
30
40
50
60
70
Figure 4.17:
Plots of pressure and right axial feed versus left axial feed (case B,
Figure 4.16)
74
Crushed Tube
Hydroformed/Calibrated Crossmember
Figure 4.18:
Part #1, see Figure 4.20, was hydroformed with the pressure curve 1st trial and the axial
feed curve shown in Figure 4.19. As seen from the picture, part #1 has a severe wrinkle
occurring on the left side of the part. In an attempt to remove the wrinkle, part #2 was
hydroformed with a modified pressure (pressure 2nd trial, Figure 4.19), while the axial
feeds were kept the same. The pressure curve was shifted up by 100 bars before the
pressure curve starts to rapidly increase for calibrating. With just a single adjustment of
the pressure curve, the wrinkle was removed and a good crossmember could be formed,
i.e. part #2. Thinning measurements from the part #2, Figure 4.21, confirm that the part
#2 is a sound part, i.e. thinning <<25% and no wrinkles. The prototyping of the
crossmember was successful and efficient. The SF approach seems to result in good
initial loading curves. This leads to reduced lead-time and effort during the prototype
development stage for tube hydroforming processes.
75
3000
80
left
70
Pressure (bar)
right
60
2000
50
1500
40
30
1000
20
500
2500
10
0
0
10
20
30
40
50
60
70
Figure 4.19:
Plots of pressure and right axial feed versus left axial feed used in the
experiments
Wrinkle
PART #1
With pressure curve 1st trial
: wrinkled part
PART #2
With pressure curve 2nd trial
: sound part, no wrinkle
Wrinkle
disappears
Left end
Figure 4.20:
76
Right end
a
25.0
Left end
Thinning (%)
20.0
Left end
15.0
10.0
Right end
5.0
0.0
a
Figure 4.21:
77
CHAPTER 5.
AUTOMATIC APPROACH TO SELECT LOADING PATH
USING OPTIMIZATION BASED SIMULATION
78
Minimize:
Objective function
f ( x)
Subject to:
Inequality constraint functions
g i ( x) ci
i = 1,2,..., I
hl ( x) = 0
l = 1,2,..., L
a j x j bj
j = 1,2,..., J
x k = [ x1 , x 2 ,..., x k ]T
(5.1)
Equation
The optimization solution method involves searching for the optimum design variable
vector x that minimizes the objective function f ( x ) , while the optimum vector x is
Figure 5.1:
80
x = [ x1 , x 2 ,..., x N ]T
Typically, an initial set of the design variables x0 are specified, which are updated
iteratively. The common form of the iterative procedure to update the design variables
is:
x k +1 = x k + x = x k + d k
Where, k is the iteration number, d is the search direction vector in the design space,
and is a scalar indicating the distance to move the design variables in the direction
vector. By iteratively varying these variables (using certain techniques described later),
a set of optimum design variables that minimizes the objective function, and obeys all
the constraints may be found. It is well known that the optimization efficiency depends
on total number of the design variables. The total number of the design variables should
be kept as small as possible. More design variables can be added to the optimization if
necessary.
Constraint Functions
Constraints are normally expressed in functions of the design variables and other
relevant design parameters to satisfy certain physical limitations of the design problems.
This is to ensure that the optimization results are not just simply numerically possible
but also realistically applicable. Therefore, the design variables, while being varied
81
iteratively, should satisfy the constraint functions, which restrict the design variables to
a certain region of the design space. These constraint functions can be either:
a) Inequality functions:
g i ( x) ci , i = 1,2,..., I
b) Equality functions:
hl ( x) = 0 , l = 1,2,..., L
Objective Function
An objective function f ( x) expresses specific design intentions that need to be
minimized (or maximized). In other words, the objective function is a performance
measure of whatever the design problem intends to optimize. There are two main types
of objective functions: a) functions to be minimized, and b) functions to be maximized.
Fortunately, based on the duality principle, the maximization problems can be
converted to minimization problems by simply multiplying the objective functions by
1. Objective functions may be explicit or implicit in ( x ) and may be evaluated by
analytical or numerical techniques (i.e. FE simulation). Specific objective function,
design variables and the constraint functions used for the determination of pressure
curve, axial feed curves, and/or counter punch curve versus time for THF processes are
demonstrated in optimization of various THF processes in Section 5.4.
Variable Bounds
Upper and lower bounds limit the search algorithm of the design variables within a
smaller design space as opposed to all the entire design space of specific design
variables, These are sometimes considered as another specific type of constraint.
a j x j bj
j = 1,2,..., J
These bounds may not be necessary for some optimization problems. However, in some
cases where there are a large number of design variables and the problem is quite
complex (i.e. the objective and constraint functions are very non-linear functions of the
82
x ( k +1) x ( k )
x ( k ) x ( k 1)
In some cases, a hard stopping criterion is imposed such that if total number of the
iterations exceeds a certain number then the optimization process will be ended. This is
to save unnecessary computational time.
5.1.2 Optimization Algorithms
There exist many optimization algorithms. Proper choice of optimization algorithm
relies on the types of the problems. Most constrained non-linear problems are
traditionally solved by gradient-based methods (e.g. Augmented Lagrangian method
and BFGS). However, for some problems where a total number of design variables is
large, non-traditional optimization algorithms (e.g. the function surfaces are
approximated by Response Surface Method RSM, then gradient methods or other
evolutionary algorithms are applied to find the minimum) may be most efficient. In this
work, two types of optimization schemes are used; a) Augmented Lagrangian method
and b) RSM. The theoretical details of each of the methods are given in Appendix C.
83
Most complex metal forming processes (e.g. sheet metal forming and tube
hydroforming) can only be analyzed numerically, i.e. FE simulations. In formulating
optimization problems of these metal forming processes, proper optimization algorithms
have to be applied in conjunction with FE simulations. Generally, part characteristics
such as part/die dimensions, shape, and weight are to be optimized. For example,
optimization of forging performs in order to maximize/improve the part formability.
This numerical optimization can also be applied for selection of the loading profiles, i.e.
for optimization of time-dependent process parameters (nodal displacement, forces,
pressures, and etc.).
84
In contrast to typical metal forming design optimization where one searches for the
optimal values of part characteristics, process optimization addresses the problem of
determining the process variables that influence the form and quality or formability of
the final part.
Particularly, for the tube hydroforming process, the loading path that minimizes part
thickness variations, and maximizes part dimensional accuracy (eliminates or minimizes
wrinkles) is searched for. This process optimization is very challenging because it
involves a number of issues that are either not well understood or are computationally
complex [Gomes et al., 2001].
a) It is not clear that there is an obvious or universally applicable definition for the
objective function to be optimized, and thus one needs to develop an appropriate
metric to measure the quality of the part.
b) Given the large number of variables that could influence this measure of
performance, one must identify the (relatively few) variables that have a sufficiently
significant effect and separate these from the (relatively many) ones that have only a
marginal effect. Moreover, these variables may not be independent from each other
and there may be significant interactions between them.
c) There may not be closed-form analytical characterizations of the objective function
as a function of the parameter values, which rules out direct differentiation basedsearch methods. However, this problem of not being able to use gradient-based
optimization methods can be handled by using FE simulations to numerically
evaluate the gradients (which is already discussed earlier).
d) Objective function evaluation is very time consuming since each such evaluation
typically involves a call to the FE program that conducts a detailed finite element
analysis. The results of this analysis will then be used to arrive at a single number
that captures the objective function value.
85
(n+m) should be kept minimum because the efficiency of the optimization largely
depends on the size of the design variables.
An appropriate number of the design variables (control points) can vary depending on
how well these control points can represent the shape of the process parameter curve of
interest. For most tube hydroforming processes, typical good pressure versus time curve
and axial feed versus time curve seem to take certain shapes. The pressure curve usually
slightly increases during the tube expansion stage, and then rapidly ramps up to its
maximum value during the calibration stage, see Figure 5.2.a. As for the axial feed
curve, unlike the pressure curve, during the tube expansion stage, the axial feed
increases rapidly to feed material into the expansion zone as much as possible.
However, as soon as the calibration phase starts the axial feeding starts to slow down
and stop increasing soon after, see Figure 5.2.b. Five design variables (control points)
seem to be sufficient for capturing the shapes of these process parameter curves, see
Figure 5.2.
86
xp5
Pressure (GPa)
Design variables
xp4
xp2
xp1
xp3
Fixed point
a)
Time (ms)
xf3
xf4
xf5
xf2
xf1
b)
Time (ms)
Figure 5.2:
Typical shapes of (a) pressure versus time curve and (b) axial feed versus
time curve represented by piecewise linear curves
xf2
xf3
xf1
xf4
xf5
Time (ms)
Figure 5.3:
Normally, in conducting a THF simulation in PAM-STAMP, the axial feed (axial punch
displacement) at the tube ends is accomplished by imposing velocity boundary
conditions on the tube end nodes. Velocity boundary condition is normally used instead
of displacement boundary condition to prevent unrealistic dynamic effect. This
erroneous effect may occur if there exist large nodal accelerations due to sudden change
of nodal displacements, which is quite possible during the optimum variable search in
the application of optimization-based FE simulation. Typical shape of axial feed
velocity is the derivative curve of the axial feed displacement curve. The axial feed
velocity curve usually increase rapidly in the beginning and decrease toward the end,
see Figure 5.3. Five design variables (control points) also seem to be sufficient to
represent this curve by a piecewise linear curve.
5.2.2 Objective Function
In general, the main goal of optimization in any THF process is to determine the best
loading paths that would hydroform the given part with the most uniform part thickness
distribution. However, this is not so for all the THF parts. Common objective functions
for tube hydroforming are the following:
I
2
1
f ( x) = Min (hi )
I i =1
where hi = h1 , h2 ,..., hI are the part thickness at different locations. These thickness
values are implicit functions of x , the design variable vector. This is discussed
more in section 5.4.1.
88
Minimize the maximum part thinning: Parts with complex geometry tend to prevent
large axial feed to be applied, thus resulting in parts with excessive thinning. These
difficult-to-form parts are usually very challenging to find proper loading paths that
would form the part successfully. Proper forming goal in this case should be to find
the best loading paths that can form the part with minimum possible part thinning.
The objective function for this part quality can be expressed as follows:
f ( x) = Min[Max(hi )]
where hi = h1 , h2 ,..., hI are the part thickness at all the locations in the entire part.
These thickness values are implicit functions of x , the design variable vector. This
is discussed more in section 5.4.1.
Maximize the protrusion height: For parts with protrusion features such as T-shapes
and Y-shapes, the main objective in the optimization problem is to form the part
successfully without any fracture and with largest protrusion height obtainable. The
objective function of this type is discussed in detail in section 5.4.2
g ( x) = Max(d i )
Where, [d i = d1 , d 2 ,..., d I ] is the tube-to-die normal distance of all the tube elements.
And, d i is an implicit function of the design variable vector ( x) .
All other types of constraint functions for THF are formulated and explained in detail in
Section 5.4, to best express several critical part qualities of concern.
90
Optimization input
file.cds
Optimization code
PAM-OPT
Optimization output
File.curves
File.history
FEA
Iterations
Optimization descriptions
THF simulation
Input file.ps
FE simulation
PAM-STAMP
PAM-Solver
PAM-VIEW
Figure 5.4:
91
OD1/2
OD0/2
W/2
L0/2
Value / Units
128 (mm)
42 (mm)
1.98 (mm)
64 (mm)
54(mm)
0.567 (GPa)
0.264
0.007
Figure 5.5:
93
The main goal of this simple bulge hydroforming is to optimize the loading paths that
would improve the part quality in two different cases, a) minimize the part thickness
variations, and b) minimize the part maximum thinning. These two objectives suggest
two different tube hydroforming approaches.
5.4.1.2.2 Optimization Data Design Variables, Objective Function, Constraints
Design Variables
The design variables in this problem are the points controlling the piecewise linear
curves representing the pressure and axial feed velocity curves versus time, see Figure
5.6. The design variable vector is written as follows:
25
Upper bound
20
Design variables
15
Fixed points
10
xf1
xf2
xf3
xf4
xf5
Lower bound
0
0
0.2
0.4
0.6
0.8
Time (ms)
0.08
Upper bound
Pressure (GPa)
0.06
Design variables
0.04
xp3
Fixed point
0.02
xp5
xp4
Lower bound
xp2
xp1
0
0
0.2
0.4
0.6
0.8
Tim e (m s)
Figure 5.6:
95
In Figure 5.6, the design variable bounds limit the feasible region of the design
variables. Without these bounds the search for the optimal variables may not be
efficient. The axial feed curve is allowed to have more room to move around, whereas
the possible movement of the pressure curve is more restricted. This is done because of
the idea that the axial feed should be applied as much as possible and the pressure is just
applied enough to prevent wrinkles. The pressure curve is allowed to have more room
to move towards the process end such that an increased pressure for calibration would
be possible. Please notice that the bounds are governing only the design variables not
the fixed points.
The design variable bounds for axial feed velocity can be calculated (or estimated) from
a) maximum allowable axial feed of the initial tube relative to the die (i.e. tube cannot
be fed in so much such that the tube end is pushed beyond the guiding zone into the
expansion zone), and b) maximum allowable thickening of the final part; in this case
study, the first method is used to determine the maximum axial feed (25 mm), thus the
maximum velocity is 25 mm/ms over the forming time period, see the velocity upper
bound in Figure 5.6. Of curse, this velocity upper bound does not have to be such a
horizontal straight line. It can be of any sensible shapes with a total area underneath the
curve of 25 mm.
The pressure bounds are more difficult to estimate. In this work, usually, the pressure
lower bound is a straight line increasing from zero at the first control point to 80%-90%
of the initial pressure at the last control point. The pressure upper bound is constructed
such that it would make a narrow band bounding the initial pressure curve with a gap
between the lower and upper bound of about 0.02 GPa (200 bar). Finally, a pressure of
110%-120% of the calibration pressure calculated for the given part can be used as the
last point of the pressure upper bound.
These design variable bounds sometimes can be impossible to estimate. As a rule of
thumb, it is advised to always use the strictest bounds sensible first. Then, if the
96
optimized variables are all placed on the bounds imposed, the bounds may need to be
expanded in the next optimization run.
Objective Function
The objective function of minimizing the part thickness variations is applied here first.
Then, at the end of this section, the second objective function of minimizing part
(maximum) thinning is applied to the exact same problem. Finally, both of these two
results are then compared in terms of the part quality.
Figure 5.7 shows the tube mesh used in formulating the objective function. To consider
the part thickness variation along the tube axis, three chosen tube elements are spread
out from the tube center (which is subject to excessive thinning) to the tube end (which
is subject to excessive thickening). The thickness values of these chosen tube elements
at the final simulation step are extracted and then used to evaluate the objective function
of minimizing part thickness variation. This objective function can be expressed as
Where the h1, h2, and h3 are the tube elemental thicknesses at the final simulation step.
The chosen elements are on the same longitudinal line due to the fact that the part is
axisymmetric. More tube elements can be considered in the objective function if a
tighter tolerance of the part thickness variations is to be obtained.
Constraint functions
The final part should be formed completely and without any wrinkle at the final
simulation step. The constraint function in this case should reflect that goal. One
convenient way to check the final part dimensional accuracy and part wrinkle-free
condition is to consider the controlled volume of the final part (controlled part volume
is the part volume calculated within a fixed imaginary boundary over the expansion
zone, see Figure 5.8).
97
Centerline
Tube axis
h1
Figure 5.7:
h2
h3
Controlled volume
Process start
Controlled volume
Process end
Figure 5.8:
Voltube = Voldie
98
If the controlled volume of the final part ( Vol tube ) is the same as that of the die ( Vol die )
then the part is successfully formed. If the controlled part volume is less than that of
the die cavity, then that part is not completely formed or has some wrinkles. The detail
of this concept is elaborated more in Chapter 6. This constraint function can be
expressed as below:
*100 1
The alternative way of imposing the part dimensional accuracy constraint is to monitor
the final tube-to-die distance, as explained earlier. However, considering all the nodal
tube-to-die distance can become quite inefficient if the number of element is large. The
part volume is just one single scalar readily calculated. The use of the controlled part
volume to check the completeness of the part may not be as accurate as the part
becomes bigger, e.g. structural parts, due to larger numerical errors in calculating the
total part volume from the facetted tube mesh. In those cases, the tube-to-die distance of
some critical part locations (i.e. areas that are wrinkle prone) is a better option to
impose this constraint.
5.4.1.2.3 Optimization Algorithms
In this optimization of the simple bulge hydroforming, the problem is relatively simple.
The metal flow only experiences a simple biaxial state of stress without any shear
stresses (i.e. axisymmetric forming and thin shell assumption). Therefore, this
optimization problem is solved using the gradient-based method (i.e. Augmented
Lagrangian method). In general, gradient-based methods are known to converge the
fastest among all the optimization methods. Unfortunately, if the problem is too
complex (a large number of design variables or the problem is very non-linear)
sometimes the gradient-based methods do not converge at all. However, this is not the
case here.
99
100
16
Initial
14
10 iterations
12
15 iterations
10
25 iterations
8
6
4
2
0
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
0.080
Pressure (GPa)
Initial
10 iterations
0.060
15 iterations
25 iterations
0.040
0.020
0.000
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
Figure 5.9:
1.500
0.100
Constraint function
Objective function
0.120
0.080
0.060
0.040
0.020
0.000
upper limit
1.000
0.500
lower limit
0.000
0
10
15
20
25
Figure 5.10:
10
15
Iterations
Iterations
20
25
0.08
Pressure (GPa)
Optimal
Initial
0.06
0.04
0.02
0.00
0
10
Figure 5.11:
0.04
Part Thinning
0.02
0.00
-0.02 0
10
20
30
40
50
60
-0.04
-0.06
-0.08
-0.10
-0.12
-0.14
Longitudinal Curvilinear Distance (mm)
Figure 5.12:
102
Another run of the optimization is carried out on the same simple bulge using a
different objective function, i.e. minimizing the part maximum thinning. The objective
function for this problem can be expressed as follows:
h h0
f ( x) = Min Max i
h0
Where hi = all the tube elemental thickness at the final simulation step and h0 = initial
tube thickness.
The optimized loading paths and the part thinning distribution are shown in Figure 5.13
and Figure 5.14. It can be seen that in this case the total axial feed is 11 mm, which is
larger that from the first case. The maximum part thinning now is reduced to 2%.
However, due to the large amount of axial feed applied, the thickening on the tube end
is increased to 12%. Obviously, this part has unnecessary thickening. However, this
situation often happens in difficult-to-form part. For this simple bulge geometry, this
situation is done only for an example.
5.4.2 Y-shape
Hydroforming of parts with protrusions is usually more challenging than typical THF
due to the added process parameter, counter punch force versus time curve. Normally,
the counter punch force versus time curve cannot be calculated analytically. Trial-anderror FE method is usually used to figure out the proper values of the counter punch
force. In this optimization case study, the Y-shape, used in Chapter 3, is chosen again to
be the case study for demonstrating how to optimize the counter punch force versus
time curve.
103
0.08
Pressure (GPa)
Optimal
Initial
0.06
0.04
0.02
0.00
0
10
12
Figure 5.13:
0.04
0.02
Part Thinning
0.00
-0.02 0
20
40
-0.04
-0.06
-0.08
-0.10
-0.12
-0.14
Curvilinear distance (mm)
Figure 5.14:
60
Two optimization algorithms were used in this work. First, the gradient-based method
was used. Then, it was realized that the gradient method was not able to find any
converged solution. The RSM optimization method was later successfully applied to the
exact same optimization problem. However, the converged solution from the RSM
seemed to be inferior than that obtained from the gradient method, which was not a
converged solution.
5.4.2.1 FE Model Descriptions
Descriptions of the Y-shape FE modeling are explained in detail in section 3.1.2. All the
simulation setups are the same except that the counter punch force and pressure versus
time curves are to be optimized.
5.4.2.2 Optimization Descriptions
5.4.2.2.1 Optimization Problem
In hydroforming of the Y-shape, three process parameters are involved; a) left and right
axial feed curves, b) pressure curve, and c) counter punch force curve. Since the main
goal here was to optimize the counter punch force, only the counter punch force curve
and pressure curve were used as the design variables, while the left and right axial feed
curves were kept the same as applied in section 3.1.2. The pressure curve was still
chosen to be the design variables because it would allow the optimization to suppress
any part wrinkles that may happen during the adjustment of the counter punch force
curve.
5.4.2.2.2 Optimization Data Design Variables, Objective Function, Constraints
Design variables
The counter punch force curve and pressure curve were represented by piecewise linear
curves, of which the control point positions are the design variable in this problem, see
Figure 5.15. There are five control points for each of the curves. The initial positions of
105
the control points were taken from past experience on hydroforming simulation of this
part. The design variables are expressed as follows:
x = [ xp1 , xp 2 ,...xp n | xf1 , xf 2 ,...xf m ]
xp1 , xp 2 ,...xp5 = Design variables for pressure curve
xf1 , xf 2 ,...xf 5 = Design variables for counter punch force curve
The variable bounds on these two curves were estimated and imposed in the same way
as in the optimization of the simple bulge, i.e. the design space for the pressure curve
was small and more room was given only for the calibration stage. And, on the other
hand, the counter punch force curve was given a lot of design space. This was to give
more freedom to the search of optimal counter punch force. As a result of the axial
feeds used in this Y-shape, the growing protrusion did come in contact with the counter
punch until around the time of 0.4 ms. Therefore, the counter punch force curve was
fixed at zero till around the time of 0.4 ms. The left and right axial feed velocity curves
were the same as applied in chapter 3, see Figure 5.16.
Objective function.
The goal of this optimization was to determine the best combination of counter punch
force curve and pressure curve versus time such that the height of the protrusion was
maximum obtainable, which is, typically, a desirable feature for parts with protrusions.
In this work, the part controlled volume was used to express the protrusion height. It is
obvious that the protrusion height should be directly proportional to its part controlled
volume, see Figure 5.17. The objective of this problem can be expressed as follows:
f ( x) = Max(Vol tube )
Constraint functions
Three different constrains, see Figure 5.18 were necessary in this optimization problem.
These constrains were imposed onto the problem in order to guarantee that final part
obtained is successful. All the constrains are explained below:
106
160
Upper bound
140
120
xf5
100
xf4
80
60
xf3
Fixed control
point
40
20
xf2
xf1
0
0.0
0.2
Lower bound
0.4
0.6
Time (ms)
0.16
0.8
1.0
Upper bound
0.14
xp5
Pressure (GPa)
0.12
0.10
0.08
xp3
xp4
xp2
0.06
xp1
0.04
0.02
Lower bound
0.00
0.0
Figure 5.15:
0.2
0.4
0.6
Time (ms)
0.8
1.0
Design variables: counter punch force versus time curve and pressure
versus time curve
90
80
70
Right end
60
Left end
50
40
30
20
10
0
0.0
Figure 5.16:
0.2
0.4
0.6
Time (ms)
0.8
1.0
Tube-to-die distance: This constraint was imposed in order to prevent the final part
from wrinkling. The constraint considered distance from the part surface normal to the
die surface (so called tube-to-die distance) at the end of the forming process. If this
distance exceeded a certain value then part was not formed successfully, i.e. not
completely formed or wrinkled. As for the Y-shape forming, the part was prescribed
with profiles along which the tube-to-die distances at the process end were monitored.
These profiles were put on the parts most wrinkle-prone areas, as shown in Figure
g 2 ( x) = 1 * Max( i ) 2
Hp
Voltube
Hp
Controlled volume
Figure 5.17:
Tube-to-die
distance profiles
a)
b)
Max.
thinning
c)
Figure 5.18:
109
h h0
g 3 ( x) = Max i
h0
< 3
Where hi = all the part elemental thickness at the final simulation step and h0 = initial
tube thickness.
5.4.2.2.3 Optimization Algorithms
Both gradient and RSM method were used in this optimization problem. First the
optimization results from the gradient method are discussed. Then, the results from the
RSM method are given in comparisons with the previous results.
5.4.2.3 Optimization Results
The gradient-based method was not able to find a converged solution. The objective
curve fluctuates and shows no sign of converging, see Figure 5.19. The constraint
function curves are shown in Figure 5.20. The tube-to-die distance constraint curve
seemed to stay within the bound. This indicated that the final part completely took the
shape of the die and had no wrinkles. However, the other two constrains, i.e. maximum
thinning and corner curvature, went out of the upper bound many times during the
optimization. Since the optimization did not seem to be converging, it was terminated at
the iteration 16, where all the constraints were satisfied. The resultant counter punch
force curve and pressure curve taken from the optimization at iteration #3 and #16
(these curves satisfied all the constraints) are shown in Figure 5.21. Since the results
from iteration #16 are the best, its resultant counter punch force curve and pressure
curve are considered the best from this optimization using gradient method.
The optimized counter punch force increased up to 50 kN then decreased to 10 kN at
0.6-0.8 ms, and finally increased up to 90 kN at the end, see Figure 5.21.
110
Non-converging
160000
158000
156000
154000
152000
150000
148000
0
10
12
14
16
Interation
Figure 5.19:
a)
0.40
upper limit
1.00
0.80
0.60
0.40
0.20
lower limit
b)
0.00
0
10
12
14
Out of bound
0.35
upper limit
0.30
lower limit
0.25
0.20
16
Interation
10
12
14
Interation
0.35
Maximum thinning
Out of bound
c)
0.30
upper limit
0.25
lower limit
0.20
0.15
0
10
12
14
16
Interation
Figure 5.20:
16
120
Initial
5 Iterations
100
16 Iterations
80
60
40
20
0
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
0.16
Initial
0.14
5 Iterations
16 Iterations
Pressure (GPa)
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
Figure 5.21:
Optimized counter punch force curve and pressure curve versus time
112
From an observation of the simulation, the protrusion grew the fastest during the time of
0.6-0.8 ms (the counter punch force fluctuated between 10-50 kN), then its growth
slowed down and the protrusion corner formed into a tighter radius during the process
end (the counter punch force curve increased to the maximum value at the process end).
The final part from this gradient optimization has the maximum part thinning under the
critical thinning of 25%.
RSM optimization method was applied to solve this exact problem with an attempt to
find a converged solution to this problem. Figure 5.22, Figure 5.23, and Figure 5.24
show evolution of the objective function, constraint functions, and the optimal loading
paths, respectively. With the application of RSM method, a converged solution was
found and all the constraint functions were converging inside the lower and upper
bounds.
Though the RSM took much less number of iterations than the gradient-based method
did (3 iterations with 35 simulation runs in RSM method and 16 iteration with 176
simulation runs in gradient-based method) the RSM optimum solution seems to be just
a local optimum not a global one. This is clearer if one compares results from the RSM
method to that of the gradient method, see Figure 5.25. The part protrusion height from
the gradient-based method is lager than that from the RSM optimization method, see
Figure 5.25.b. The protrusion maximum thinning of the RSM part is smaller than that of
the gradient method simply because the RSM part has a smaller protrusion, see Figure
5.25.a.
The optimization of counter punch force and pressure curve for the hydroforming of Yshape has shown to be a rather difficult task. The gradient-based method did not seem to
be able to find any converged solution. This usually happens in cases where the
problems being optimized are very non-linear (i.e. the objective and constraints are
strong non-linear functions of the design variables), and have a large number of design
variables. RSM optimization method is known to be able to handle such problems.
However, in this case study, through RSM, only local optimum seems to be found.
113
154000
153000
152000
151000
150000
149000
0
Interation
Figure 5.22:
0.35
a)
upper limit
1.00
0.80
0.60
0.40
0.20
lower limit
b)
0.00
0
0.3
0.25
upper limit
0.2
lower limit
0.15
0
Interation
Interation
Maximum thinning
0.35
c)
0.30
0.25
upper limit
lower limit
0.20
0.15
0
Interation
Figure 5.23:
114
140
Initial
120
1 Iteration
3 Iteration
100
a)
80
60
40
20
0
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
0.14
Initial
Pressure (GPa)
0.12
1 Iteration
0.10
3 Iteration
0.08
0.06
0.04
0.02
b)
0.00
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
Figure 5.24:
115
Gradient-based method
RSM
0.40
Gradient
Part thinning
0.20
0.00
-0.20
50
100
-0.40
150
200
250
RSM
-0.60
-0.80
a)
-1.00
Curvilinear distance (mm)
80
b)
Gradient
70
60
50
RSM
40
30
20
10
0
50
75
100
125
150
175
200
225
250
Figure 5.25:
116
Design variables: There were a total of 15 design variables in this problem, the design
variables were used to represent the piecewise linear pressure and left and right axial
feed velocity versus time curves. All the design variables are shown in Figure 5.28. The
variable bounds were also given in the Figures. The initial design variables and bounds
117
of the left and right axial feed velocity were the same. The design variables can be
expressed as the following:
x = [ xp1 , xp 2 ,...xp n | xf1 , xf 2 ,...xf m ]
xp1 , xp 2 ,...xp5 = Design variables for pressure curve
xf1 , xf 2 ,...xf 5 = Design variables for left axial feed velocity curve
xf 6 , xf 7 ,...xf10 = Design variables for right axial feed velocity
curve
Objective function: The main goal of this forming was to successfully hydroform that
part with minimum part thinning possible. Therefore, the objective function can be
expressed as follows:
h h0
f ( x) = Min Max i
h0
Where, hi = all the tube elemental thickness and h0 = initial tube thickness.
Constraint functions: This part was a rather difficult to form due to its non-symmetry
feature from left side to right side of the part. A substantial amount of axial feed was
necessary for forming of this part, which in turn heightened the risk of part wrinkling.
Therefore, wrinkles in the final part were the main concern. The tube-to-die distance
constraint was imposed along the profile shown in Figure 5.27. This profile was placed
on the tube surface along the tube axis direction and close to the part corner. This area
was largely unsupported, thus most susceptible to wrinkles.
5.4.3.2.3 Optimization Algorithms
This current optimization problem had a large number of design variables (i.e. 15
variables compared to 10 variables of the previous case studies). Based on the
experience gained from the previous case study, the gradient-based method was not able
to find any converged solution for rather complex problems; it was decided to use the
RSM optimization method in this problem.
118
Left side
49.3
62.3
Right side
62.3
100
100
100
450
Material Properties
Value / Units
SS304
1.5 (mm)
49.3 (mm)
1.207 (GPa)
0.351
0.021
Material
Tube wall thickness (t0)
Initial tube outside diameter (D0)
Strength coefficient (K)
Strain hardening coefficient (n)
Pre strain (e0)
Figure 5.26:
Left side
1/8th FE model
Tube-to-die
distance profile
Part corner
Right side
Figure 5.27:
119
60
Upper bound
50
40
30
xf2
xf1
xf3
xf5
xf4
20
10
Lower bound
0
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
0.25
Pressure (GPa)
0.20
Upper bound
0.15
0.10
xp5
xp4
xp3
0.05
xp1
xp2
Lower bound
0.00
0.0
0.2
0.4
0.6
0.8
1.0
Time (ms)
Figure 5.28:
Initial design parameters: left and right axial feed velocity versus time
curve and pressure versus time curve
120
121
Left side
Right side
60.00
50.00
40.00
30.00
20.00
Initial
1 Iteration
6 Iteration
10.00
0.00
0.0
0.2
0.4
0.6
0.8
1.0
0.8
1.0
0.8
1.0
Time (ms)
60.00
50.00
40.00
30.00
20.00
Initial
1 Iteration
10.00
6 Iteration
0.00
0.0
0.2
0.4
0.6
Time (ms)
0.16
Initial
Pressure (GPa)
0.14
1 Iteration
0.12
6 Iteration
0.10
0.08
0.06
0.04
0.02
0.00
0.0
0.2
0.4
0.6
Time (ms)
Figure 5.29:
1.40
0.25
0.20
0.15
0.10
0.05
0.00
upper limit
1.20
1.00
0.80
0.60
0.40
0.20
lower limit
0.00
a)
Figure 5.30:
Right axis
(main axis)
Left axis
Pressure
0.100
35
Pressure (GPa)
25
0.060
20
Pressure
0.040
15
10
0.020
0.000
0
0
10
15
20
25
123
30
35
40
30
0.080
Figure 5.31:
b) Iteration
Iteration
0.10
Part thinning
0.05
0.00
0
50
100
150
200
250
300
350
400
450
-0.05
-0.10
-0.15
Curvilinear distance (mm)
Figure 5.32:
124
CHAPTER 6.
AUTOMATIC APPROACH TO SELECT LOADING PATH
USING ADAPTIVE SIMULATION
The ultimate goal of the adaptive simulation approach is to completely eliminate the
trial-and-error simulation approach and to generate feasible process parameter curves
(LP) within only a few or just one simulation run.
125
t0
Tube
t1
Piy
t2
Wrinkle
Da
Piy
t3
Da
Piy+Pi
t4
Wrinkle
2Da
Piy+Pi
Time (t0-t1): yielding pressure (Piy) is applied without any axial feed to initiate the
deformation.
Time (t1-t2): pressure is kept constant (or may be reduced) while an axial feed increment
(Da) is applied.
Time (t2-t3): a wrinkle is detected at t2; no axial feed is applied while a pressure increment
(Pi) is added.
Time (t3-t4): the wrinkle has been removed at t3; the pressure is kept constant (or may be
reduced) while another axial feed increment is applied.
Figure 6.1:
In adaptive simulation, the process parameters for the future simulation step are to be
predicted to proper values based on the forming part quality information collected from
the past and current simulation time steps. In other words, during a THF process
simulation run, the simulation intermediate results about forming part qualities (i.e. part
wrinkles, part thinning, and etc) up to the current time step is deduced through
knowledge of THF mechanics and then used it to calculate/project appropriate process
parameters for the next simulation time step. This is in contrast to the traditional
application of FE process simulations where only the simulation results at the final step
are considered and used to infer parameter adjustments for the next trial simulation run
in an attempt to improve/optimize the forming process.
The adaptive simulation approach relies on ability to detect/identify an existence of
defects (i.e. wrinkling and fracture) in the part being formed, and appropriate
adjustments of the relevant process parameters to correct these defects during the
following simulation time steps. The essential parameter adjustment strategy of the
proposed method is to maximize axial feed as it prevents the part from excessive
thinning and minimize pressure just to be sufficient to prevent the part from wrinkling.
Upon completion of adaptive simulation, the evolution of process parameters predicted
by the methodology is the resultant feasible loading paths.
6.2 Implementation of Adaptive Simulation Method
127
PAM-STAMP
THF SIMULATION
TIME STEP (N)
? Thinning >
YES
Max. Value?
Stop Simulation
Wrinkle Indicator
Calculation
NO
Wrinkling?
?
YES
Defect Detection
Move Punch
Stop Punch
Figure 6.2:
reasons but also for part rigidity reasons. Part fracture, obviously, should be avoided at
all costs. A less obvious defect is large part thickness variations. For structural parts,
uniform thickness variations are preferred due to part rigidity and weight reasons. Some
smaller parts also require tight part thickness variations as well such as musical wind
instruments as large part thickness variations would deviate the sound quality of the
instruments.
Monitoring of intermediate simulation results enables a realization of the part defects as
they first appear and grow. The defect detection module in AS has to be able to identify
both existence and severity of these defects as this information will be used as
performance indexes for decision-making in the process parameter adjustment module.
In THF process, sometimes, it is only possible to hydroform a difficult-to-form part
successfully if some tolerable wrinkles are allowed during the intermediate forming,
with the condition that these wrinkles can be straightened out at the end of the forming
process, see Figure 6.3. Two types of part wrinkles are defined in this work: a) dead
wrinkle and b) alive wrinkle. Dead wrinkles are defined as part wrinkles that exist in
the final part, see Figure 6.3. And, alive wrinkles exist in intermediate forming parts
whose surfaces are still far away from the die cavity wall. There are two types of alive
wrinkles: b.1) good alive wrinkles, which are defined as part wrinkles that can
potentially be strengthened out at the process end, and b.2) bad alive wrinkles, which
are defined as part wrinkles that cannot be strengthen out at the process end.
The following are requirements on calculations of the defect detection module:
Identify good alive wrinkle, bad alive wrinkle, and dead wrinkle
Quantify the above defect attributes into non-dimensional performance index value
so called wrinkle indicator ( Iw) and fracture indicator (If)
130
During forming
Alive wrinkle
C.L.
a)
Process end
Good part
Die
Tube
b)
Process end
Dead wrinkle
c)
Figure 6.3:
a) intermediate part with alive wrinkle, which, at the process end, can
turn into b) good final part, or c) bad final part with dead wrinkle
131
Bursting
Typical THF loading path
Pressure
Wrinkling
Yielding
(a)
Axial feed
Plane strain
FLD
Bursting
Balanced bi-axial
Pure shear
THF
Wrinkle
2
(b)
Figure 6.4:
a) loading path in the THF forming window, and b) in-plane strain plot
133
Figure 6.4 shows a process window of a typical THF process. The feasible process
window is bounded by the leaking, wrinkling, bursting as explained above. A successful
THF process requires a load path that lies inside this feasible window. However, the
boundary of this feasible window cannot (if possible at all) be derived easily prior to
selection of proper load paths for a given part. When considering this problem in the
framework of AS approach, one can see that the defect detection module can be used to
estimate the boundary of the feasible process window. And, the parameter adjustment
module has to be devised to navigate the loading path inside this feasible process
window to the final process end.
Ideally, all hydroformed parts demand uniform part thickness variation and wrinkle-free
quality. In practice, these stringent requirements are relaxed depending on the intended
part functionality and the hydroformability of part itself. The goal of the parameter
adjustment module is essentially to select loading path that would result in best part
quality possible. It should be noted that global optimum part quality cannot be obtained
through using AS approach as it only utilizes the simulation results on part formability
from past up until current simulation time step to project the best future loading path
in the following time step. No global optimization is attempted in this AS approach.
6.2.2 Integration of Adaptive Simulation Program to PAM-STAMP
The adaptive simulation program was written in FORTRAN and linked to the internal
solver of PAM-STAMP called PAMSOL. The schematic description of the adaptive
simulation program is shown in Figure 6.5. This section discusses only the general
description of the program. The details of the AS program are given in Appendix E.
134
PAM-STAMP
Subroutine: LDCMOD
PAM-solver: PAMSOL
Simulation
Time Step
ti
Finite Element
Calculation Results
- Nodal coordinates
- Strain data table
Adaptive
Simulation
Control
Time Step
Tj
Modifications of
Load Function Curves
- Pressure curve
Figure 6.5:
T j ti
135
136
137
138
dY i
Y y i 1
, Where i = 1n (n+1 = number of node number)
= i
dX i
X i X i 1
Iwd = 0 -> wrinkle free condition: when there is no change in sign of the slope
Iwd = 1 -> wrinkled condition: when there is a change in sign of the slope
This slope criterion though is very simple but it has some drawbacks. First, it only
works for simple bulge geometry. Second, this slope criterion will not distinguish part
wrinkles from part geometry with curvatures. Last, this criterion is not able to indicate
whether the wrinkles detected are alive wrinkles or dead wrinkles as defined in section
6.2.1.1.
139
Y
Symmetry
Line
Tube
Profile
Z
T ube
Axis
a)
Symmetry Line
Y
1
3
dY
dZ
Tube
b)
Center Line
Z
Symmetry Line
Y
1
3
4
dY
dZ
Tube
c)
Center Line
Z
Figure 6.6:
140
This criterion was developed to distinguish between alive wrinkles and dead wrinkles.
Let us consider hydroforming of the same bulge used in the previous section, see Figure
6.6.a. In this criterion, both geometrical information of tube and die are used. Within a
control window, the cutting plane makes a tube profile with an arch length of (Lt) and a
die profile with an arch length of (Ld), see Figure 6.7.a. It is noted here that the control
window used in this criterion normally has to be large enough to cover the entire
forming area (i.e. expansion zone, excluding guiding zone). But due to symmetry of the
bulge, only a half control window is applied in this specific case.
The basic idea is based an observation that when a wrinkle-free bulge is hydroformed
completely against the die cavity surfaces the tube profile arch length will be the same
as that of the die profile, see Figure 6.7.b (i.e. for a wrinkle-free part at process end: Lt
= Ld). If the final part is formed badly with some dead wrinkles then the tube profile
arch length is now longer that of the die profile, see Figure 6.7.c, (i.e. for a deadwrinkled part at process end: Lt > Ld).
This idea may be extended further to distinguish part with good alive wrinkles from part
with bad alive wrinkles (i.e. bad alive wrinkles are those that can potentially turn into
dead wrinkles at the final part). At any instance during the forming, if the tube profile
arch length, Lt(ti), becomes greater than the die profile arch length, Ld, then bad alive
wrinkles are indicated (i.e. Lt(ti) > Ld, where ti is intermediate simulation time step).
The concept discussed above suggests some idea about how much the process can
tolerate having part wrinkles during the forming (i.e. alive wrinkles) which can be
removed at the process end. The main assumption here is that the tube profile always
lies on the cutting plane throughout the hydroforming. In other words, there should be
no in-plane shear in the part during the hydroforming. With this constraint, therefore,
this criterion is valid only for axisymmetric parts.
141
a) During forming ( ti )
Control
Window
Symmetry
Line
Ld
Die
Tube
Lt
Centerline
Expansion zone
Guiding
zone
Control
Window
Guiding
zone
Symmetry
Line
Die
Tube
Lt = Ld
Centerline
Control
Window
Symmetry
Line
Die
Tube
Lt > Ld
Centerline
Figure 6.7:
142
In AS, it is also necessary that severity of part wrinkles be quantified. Basic knowledge
of geometry can be used to quantify the severity of part wrinkles. If one considers all
arbitrary 2D shapes that share a common straight portion and all have the same enclosed
area, as shown in Figure 6.8.a, an arch of a perfect circle must have the shortest length
compared to all the other shapes. Under the same geometrical constraints, it is also true
that a non-convex shape will have a longer arch length than that of any convex shapes.
By the same analogy, see Figure 6.8.b, considering the area on the cutting plane
bounded by the tube centerline, the tube profile and the control window, among all
possible shapes of the hydroformed bulge with the same bounded tube interior area the
following are true:
a. Wrinkled bulge has a longer tube profile length than that of a wrinkle free bulge
b. More severe wrinkled part (with larger wrinkle amplitudes) has a longer tube profile
length than that of a mildly wrinkled bulge (with smaller wrinkle amplitudes)
Formulation of the length-to-area wrinkle indicator (see the parameters in Figure 6.8.b)
Let
Ld:
Lt (ti): be an arch length of the enclosed forming tube profile and at any time ti
Ad:
be the enclosed area of the cutting plane bounded by the die profile, tube
center line, and control window
At (ti): be the enclosed area of the cutting plane bounded by the tube profile, tube
centerline, and control window at time ti
t i:
Then, the bounded arch length of the forming tube profile and the bounded tube interior
area at any time step ti can be normalized as the following:
L(t i ) =
Lt (t i ) Lt (t 0 )
: Normalized tube profile length
Ld Lt (t 0 )
A(t i ) =
At (t i ) At (t 0 )
: Normalized tube interior area
Ad At (t 0 )
143
Arch length
of wrinkled circle = L2
(Non-convex)
Arch length
of perfect circle = L1
(Convex)
Control
Window
Ld
b)
Symmetry
Line
Die
Ad
Centerline
Control
Window
Symmetry
Line
Lt ( ti )
Tube
At ( t i )
Centerline
Figure 6.8:
a) shortest arch length illustration and b) parameters used in length-toarea wrinkle criterion
144
The normalized tube interior area ( A ) can be used to indicate whether the part is
completely formed or not:
A = 0:
A = 1:
The normalized tube profile length ( L ) can be used to indicate the severity of wrinkles:
L = 0:
L = 1:
1.
L > 1:
To assist the understanding of this length-to-area wrinkle criterion, the simple bulge
hydroforming used in Chapter 5, Figure 5.5, was used again as an example. Three
different loading paths were applied to form the bulge: 1) Self-feeding LP, 2) optimal
LP (taken from Yang, 2002), and 3) bad LP with excessive axial feed, see Figure 6.9.
The self-feeding LP was applied here to form a wrinkle-free part to be used as a
reference part. However, as discussed before, self-feeding part would result in a
wrinkle-free part with excessive thinning. The optimal LP would form the final part
without any wrinkles and acceptable thinning. The bad LP would result in a final part
with dead wrinkles. From the simulation results, the normalized tube profile lengths
( L ) of the parts were plotted against their corresponding normalized tube interior areas
( A ), Figure 6.10.
From Figure 6.10, all the normalized length-to-area curves start from the same point
L = 0 with A = 0; however, the shape of the curves depends on the state of wrinkle
145
0.07
SF LP
Pressure (GPa)
0.06
Optimized LP
Bad LP
0.05
0.04
0.03
0.02
0.01
0
0
10
12
14
16
18
20
Figure 6.9:
1.6
Bad LP
Control window
Control window
1.4
1.2
Opt LP
1
0.8
SF LP
0.6
0.4
0.2
0
0
Figure 6.10:
0.2
0.4
0.6
0.8
146
The curve of the SF part represents a wrinkle-free state of the part throughout the
forming process (i.e. any normalized length-to-area curves laid above this curve of the
SF part would indicate existence of alive or dead wrinkles). It gradually increases from
the starting point to the process end at L = 1 with A = 1, which indicates a successful
forming. The curve of the optimal part ends at L = 1 with A = 1, which also indicates a
successful forming, see Figure 6.10. During the forming of the optimal part, the curve is
well above the SF part curve. This suggests that during the hydroforming process of the
optimal part there are some wrinkles (alive wrinkles) that are straightened out at the
process end. It should be noticed that toward the process end the curve goes L > 1,
which indicates the risk of having dead wrinkles in the final part. However, in this case,
the calibration pressure is large enough to suppress those wrinkles and bring the curve
down to L = 1. Unlike, the optimal part, the part formed with the bad LP (with
excessive axial feed) yields the curve that never comes down to the end point of L = 1
with A = 1. This indicates that the part is formed unsuccessfully with some dead
wrinkles. It can be seen that the part starts to have dead wrinkles ( L > 1) at the middle
of the process ( A = 0.4). The wrinkles are too severe such that the calibration pressure
cannot suppress those wrinkles by the process end.
Unfortunately, the length-to-area wrinkle criterion discussed above is good only for
hydroforming of axisymmetric bulges. When dealing with non-axisymmetric parts, the
length-to-area wrinkle criterion does not work any longer for the following reasons: a)
geometrical information from just a single cutting plane is simply inadequate for nonaxisymmetric parts and b) existence of in-plane shear stress state.
However, the length-to-area wrinkle criterion can still be useful if applied to many
different locations of the part, see. Especially, it should be applied on the hard-to-form
area such as corners and complex geometry. Based on this wrinkle criterion, a better
geometrical wrinkle criterion is developed in the next section.
147
this section. This geometric criterion considers the evolution of the forming part surface
area and volume enclosed by a control box, Figure 6.11.
Based on the geometry concept developed for the length-to-area criterion that a
wrinkled part has a longer profile length than that of a wrinkle-free part at the same
section area (i.e. tube interior area):
It is true that a wrinkled part has a larger surface area, St, than that of a wrinklefree part at the same part volume, Vt.
It is also true that at a given of part volume, Vt, the surface area, St, is larger for
the part with more severe wrinkles than that of part with mild wrinkled.
Moreover, at the process end, a wrinkled part will have surface area, St(tfinal),
larger than the die cavity surface area, Sd, (i.e. when St(tfinal) > Sd, the final part
has some dead wrinkles). On the other hand, if the part is successfully formed,
the part surface area will be equal to the die cavity surface area (i.e. St(tfinal) =
Sd).
The main assumptions underlying the statements above are a) the tube surface is a
monotonously increasing function of the fluid volume St=St(Vt) and b) the volume is a
monotonously increasing function of the maximum bulge height Vt=Vt (Hp). These two
assumptions can be considered correct only if St and Vt are calculated in a fixed control
box, centered on the bulge area, as in Figure 6.11. In fact, for many THF parts (as for Tshapes and Y-shapes) the total value of Vt, if calculated including the guiding zones,
can even decrease as the bulge height grows.
148
Tube axis
a)
Control box
Control
Box
Die surface area
Sd
Symmetry
Line
Die
Cavity volume
Vd
Tube axis
Control
Box
Tube surface area
St ( ti )
Symmetry
Line
Tube
b)
BH
Tube axis
Figure 6.11:
149
The evolution of part surface area, St, as the part volume, Vt, increases is not known a
priori. In order to use it for indication of existence of wrinkles, it is necessary to track and
compare its evolution with a known wrinkle-free state deformation history. Similar to the
length-to-area wrinkle criterion, an ideal simulation with the self-feeding condition is run
to obtain an evolution of a wrinkle-free state deformation history (i.e. part surface area
and part volume).
Formulation of the surface area-to-volume wrinkle indicator (refer to Figure 6.11)
Let
Sd
Then, the tubular part surface area and volume at any time step ti can be normalized as:
Let
S (t i ) =
St (t i ) St (t 0 )
: Normalized part surface area
Sd St (t 0 )
V (t i ) =
Vt (t i ) Vt (t 0 )
: Normalized part volume
Vd Vt (t 0 )
150
151
V =0.6
1.6
Bad LP
1.4
1.2
Self-feeding LP
1
Opt LP
0.8
SF LP
0.6
Optimal LP
0.4
0.2
0
0
0.2
0.4
0.6
0.8
Figure 6.12:
Bad LP
0.8
0.6
0.4
0.2
Opt LP
0
0
a)
Figure 6.13:
0.2
0.4
0.6
0.8
0.08
Wrinkle
control
limit
0.06
0.04
Opt LP
0.02
0
0
b)
0.1
0.2
0.4
0.6
0.8
It can be seen from Figure 6.12 that both SF LP and OPT LP formed the bulge
successfully as their surface-to-volume curves ended where both normalized surface area
and volume becomes one ( S = 1 and V = 1 ). Generally speaking, due to more axial feed
applied, the OPT part obviously has less maximum thinning than that of the SF part (i.e.
2% to 14%).
From the snap shots of the parts at V = 0.6 , the OPT intermediate part has some wrinkles
as compared to the SF intermediate part. The wrinkles in the OPT part are straightened
out at the process end. The S (V ) curves of the OPT and SF parts exhibit the same trends
as found in the length-to-area criterion, compare Figure 6.12 to Figure 6.10. Due to the
excessively large axial feed, the bad LP unsuccessfully formed the bulge. This is evident
from its S (V ) curve, Figure 6.12. The curve ends at V = 0.6 , which suggests that the
final part is not completely formed against the die surface. Also the S (V ) curve shoots
pass the S (V ) = 1 borderline, which indicates that the part will most likely to have dead
wrinkles, see snap shot of the final part at Figure 6.12.
The surface area-to-volume wrinkle indicators Iwsv (V ) = S (V ) S sf (V ) , of the parts
formed with Opt LP and Bad LP are plotted in Figure 6.13.a. From the concept that any
forming part will likely to develop dead wrinkles if its S (V ) curve exceeds the S (V ) = 1
borderline, on the plot of Iwsv , the dead wrinkle limit line is simply calculated by
1 S sf (V ) . The Iwsv plot of the bad LP clearly exceeds the dead wrinkle limit line. This
Iwsv plot is useful in adaptive simulation because it can be used to:
1. indicate wrinkle severity: the larger Iwsv is the more severe the wrinkle is,
2. indicate how far away the part from developing dead wrinkles: distance from the
dead wrinkle limit line, 1 S sf (V ) ,
3. indicate whether the part is completely formed or not: the part is completely
formed when V = 1.0 .
153
The area-to-volume criterion is considered to be a more global wrinkle criterion than the
first derivative and length-to-volume criteria. This is because the area-to-volume criterion
can indicate existence of part wrinkles but cannot pinpoint locations of the wrinkles, or
sometime it can fail to catch small part wrinkles. In most cases where the parts are simple
this area-to-volume criterion is informative enough for process parameter adjustment in
AS. However, when working with rather complex parts, some local geometric
information of the part is needed. The area-to-volume wrinkle indicator is normally
applied with some wrinkle tolerance. It is entirely possible that a small dead wrinkle may
be forming in the part and the apparent area-to-volume wrinkle indicator is still below the
tolerable value. In this case, the length-to-area wrinkle indicator can be used (i.e. placing
the cutting plane across the section that is wrinkle prone) to obtain local geometry
information thus enhancing the ability to monitor the wrinkle state of the part. Therefore,
it can be seen that both length-to-area and surface-to-volume criteria can be used together
to better monitor wrinkles in hydroforming of parts.
Calculation accuracy of the FE geometrical information used in area-to-volume criterion
can be influenced by certain numerical parameters. For this reason, the simulation should
be carried out with numerical parameters similar (or equal) to those used in the SF
simulation. The most important precautions to be taken are listed as follows:
The initial mesh and the mesh adaptivity factors should be equal, since the mesh size
Since the THF simulation is usually carried out with explicit codes, the range of
axial feed rates used should be similar to that used in the SF simulation.
According to the fracture defined above, the closer If th is to 1 the likelihood that the part
will fracture is greater. According to the proposed adaptive simulation strategy, an
indicator that quantifies forming window of the part in terms of fracture failure needs
to be devised. Besides the currently used thinning criterion, any of the prediction methods
mentioned above can be applied in the adaptive simulation as well if necessary in the
future. Particularly, ductile fracture criteria such as Oyanes criterion can easily replace
the thinning criterion. The difference between the calculated damage value to the critical
damage value can be used as a fracture indicator.
6.4 Process Parameter Adjustment Algorithms
This section mainly focuses on how the adaptive simulation program adjusts the pressure
curve and axial feed curve (or axial feed velocity curve) using the wrinkle and fracture
indicators developed in the previous section. The example application of the simple
155
bulging continues here from the last section. A few important process parameter
adjustment schemes are discussed. The details of each process adjustment schemes are
given in the Appendix E to keep the chapter concise.
The adaptive simulation program adjusts the pressure and axial feed curves at every
control time steps (Tj) based on the part formability (or part qualities) known at current
simulation time step (ti), see Figure 6.5. Figure 6.14 and Figure 6.15 show the part
qualities and the adjusted process parameters, respectively, from an adaptive simulation
run of the simple bulging. These curves will now be referred to repeatedly to explain how
process adjustments are carried out in the AS program.
The part qualities considered in the current program are part wrinkles, part thinning, and
part volume. Figure 6.14.a and Figure 6.14.b show the progress curves of the first two
part qualities (i.e. surface area-to-volume wrinkle indicator and maximum part thickness
strain) starting from the beginning to the end of the forming process. These curves are
plotted against the normalized part volume, as it is a convenient way to indicate progress
of the hydroforming process (V = 1 implies that the part is completely formed). However,
the pressure and axial feed (or axial feed velocity) curves must still be applied into FE
simulation in the time domain, see Figure 6.15. In this specific example of a simple
bulge, the part volume progresses through the simulation time as shown in Figure 6.14.c.
6.4.1 Calibration Stage
In order to properly adjust the process parameters in any THF processes, it is important to
first identify the two main forming stages: a) hydroforming and b) calibration. This can
be done by using the normalized part volume versus time curve, Figure 6.14.c, to
determine an appropriate time the start the calibration stage (i.e. stop the axial feeding
and rapidly increase the pressure to calibrate the part). Typically, the calibration stage
should begin when the part is almost fully formed against the die cavity surfaces. This is
to ensure that there is no large surface expansions of the part during the calibration,
which will result in excessive part thinning or fracture.
156
0.12
0.10
0.08
0.06
0.04
0.02
AS part wrinkle
0.00
0.0
a)
0.2
0.4
0.6
0.8
1.0
0.20
SF part thinning
0.10
AS part thinning
0.00
b)
0.0
0.2
0.4
0.6
0.8
1.0
c)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Hydroforming
20
40
60
Calibrating
80
100
120
Figure 6.14:
157
Pressure [GPa]
0.15
V = 0.9
Hydroforming
Calibrating
0.10
0.05
0.00
a)
20
40
60
80
100
120
b)
6
4
V = 0.9
2
Calibrating
Hydroforming
0
0
20
40
60
80
100
120
c)
Calibrating
Hydroforming
3
2
1
0
0
20
40
60
80
100
120
Figure 6.15:
158
In this example, the part volume of 90 % (i.e. normalized part volume V = 0.9) is chosen
to be the calibration part volume cutoff, see Figure 6.14.c. From Figure 6.15, it can be
seen that starting from the time when V = 0.9 the pressure is ramped up while the axial
feed is stopped (axial feed velocity is zero) until the process ends where the part is
completely formed, V = 1.
6.4.2 Hydroforming Stage
During the period where the part forms with 0 < V < 0.9, the process parameters should
be applied such that the tube material be fed in as much as possible to prevent fracture
without causing any part wrinkles. This has been the main concept in implementation of
this adaptive simulation approach, as discussed earlier. From the general flow chart of AS
procedure, Figure 6.2, one can see that the state of wrinkles demands changes of the
pressure and axial feed, while the state of part thinning only checks for fracture failure in
the part, but does not influence the process adjustment (i.e. the adaptive simulation is
aborted if the critical thinning is exceeded). Fortunately, applying maximized axial
feed following the concept, stated above, should also result in minimized part thinning.
Therefore, the part wrinkle is used as the main control state variable in this work.
Figure 6.14.a shows the winkle control trajectories (upper and lower limits) and an
example of a wrinkle state plot of the simple bulging. These wrinkle control trajectories
are derived from the optimum forming of the same part in Figure 6.13.b. The main goal
here is to develop process control strategies that would form the part with the wrinkle
state tracking closely along these wrinkle control limit trajectories. The physical meaning
of tracking this triangle wrinkle trajectory is to adjust both pressure and axial feed at the
control time step such that the part has some beneficial alive wrinkle during the forming
and has none at the end of the process. On the other hand, if the wrinkle trajectory is flat,
the tracking of this trajectory will result in a part that has no wrinkles at all time during
the forming, i.e. part formed by pure expansion or self-feeding part.
159
Of course, in practice these wrinkle control trajectories are not known a priory.
Experience gained from using this adaptive simulation on several different parts may be
useful to approximate proper trajectories for resembling part geometries. Nevertheless,
the shape of the trajectories should such that it allows some wrinkles (only alive wrinkles,
refer to Figure 6.13.a) during the forming process and allows no wrinkles at the process
end, e.g. the triangle shape. The amount of the alive winkles allowed depends on the part
formability. Some level of trial-and-error is, unfortunately, necessary here.
The current AS program has two main process adjustment strategies for the hydroforming
state: a) Wrinkle Control Strategy and b) Pure Shear Control Strategy. Figure 6.16 shows
the loading path predicted by AS with these two strategies for the simple bulging.
6.4.2.1 Wrinkle Control Strategy
The first process parameter adjustment scheme was first based only on the wrinkle
control strategy, where the pressure is increased while the axial feed is stopped when the
part wrinkle state exceeds the upper limit trajectory and the pressure is kept constant
while the axial feed is increased when the part wrinkle state goes below the lower limit
trajectory. This strategy results in the loading paths that are of a step-liked shape, see
Figure 6.16.
It was found that this strategy could not handle the tracking of part winkle state during
the first half of the process (0 < V < 0.5), where the trajectory demands the part to have
increasing (or more severe) amount of wrinkles. The part always fractured due to the
increased pressure during this period. This does not necessarily mean that the tracking of
wrinkle state in the first half of the process is impossible. A better and more sophisticated
control strategy is needed to achieve this task. As an alternative solution, the pure shear
control strategy, discussed next, was developed to handle the process adjustment during
the early forming stage, see Figure 6.16. The wrinkle control strategy was found to work
better in the later forming stage, especially (0.5 < V < 1), when the wrinkle limit
trajectory tapers down to zero, see Figure 6.14.a.
160
Pressure [GPa]
0.15
Calibrating
0.10
Hydroforming
Wrinkle control
(step-liked shape)
0.05
0.00
0
Figure 6.16:
4
6
Axial feed [mm]
Hoop
Hoop
Axial
Axial
0.5
Stress [GPa]
0.4
hoop
0.3
0.2
0.1
axial
0
-0.1 0.0
0.2
0.4
-0.2
0.6
0.8
-0.3
1.0
normal = 0
Figure 6.17:
Plot of hoop and axial stresses showing pure shear control strategy
161
It is well known, based on mechanics of sheet metal forming, that the pure shear state of
stress will deform the sheet metal without changing the sheet thickness. This strategy
attempts to regulate the pressure and axial feed such that the critical part area (i.e.
excessive-thinning-prone area), see Figure 6.17, deforms with an in-plane pure shear state
of stress at all time.
Theoretically, when applying this pure shear control strategy, the tubular part should
form with that critical part area having the same thickness throughout the forming
process. However, due to intrinsic sphere-liked shape of most THF parts while being
expanded, the tensile biaxial state of stress tends to eventually dominate the critical area
of the part. The pure shear state of stress will simply break down and be no longer
possible to enforce it later in the process, when the tube has become sphere-liked.
Therefore, this pure shear control strategy is only applied in the beginning of the process.
In this example, the pure shear control is active till the part wrinkle state exceeds the
upper wrinkle limit, after which the wrinkle control strategy becomes active instead; see
Figure 6.14.a and Figure 6.16. It should be pointed out, from Figure 6.14. a and b, that
during period where the part volume is 0 < V < 0.6, where the pure shear control is
active, the part maximum thinning is kept quite small. This is the direct result of the pure
shear control that tries to keep hoop = - axial, see Figure 6.17.
6.4.2.3 Modified Wrinkle Control Strategy
The wrinkle control strategy that gives a step-liked loading path, previously discussed,
actually does not track the wrinkle control trajectory so well, see Figure 6.14.a. The part
wrinkle state actually goes under the lower wrinkle limit trajectory (starting around V =
0.7) until almost at the end of the process. This is because of the use of constant pressure
while increasing axial scheme in and attempt to maximize the axial feed. This constant
pressure level induces increased tensile hoop stresses the instance when the part grows
larger in diameter that is caused by the pushing of the axial feed. This hidden
162
shortcoming of the step-liked process adjustment strategy actually thins out the part
unnecessarily.
The modified wrinkle control strategy, so called increased-decreased pressure process
adjustment strategy, is developed to better track the wrinkle control limit trajectories.
Figure 6.18 shows the part wrinkle state plot and the loading path of the same simple
bulging process predicted by the same pure shear control strategy and the modified
wrinkle control strategy. The comparison of maximum thinning curves of the parts,
Figure 6.19, clearly shows that the modified control strategy reduces the maximum
thinning consistently during 0.7 < V < 1.0. The only difference of the increaseddecreased pressure control strategy from the step-liked control strategy is that the
pressure is decreased while the axial feed is increased when the part wrinkle state goes
below the lower wrinkle limit trajectory (i.e. attempting to maximize the axial feed).
One may question the practicality of the rather zigzag loading path predicted. Figure
6.20 shows the smoothened loading path, which closely approximates the predicted one.
From FE simulation results, this smoothened loading path forms this simple bulge
successfully with the part maximum thinning of 8%.
So far, this newly developed adaptive simulation program has been applied successfully
with a very simple geometry such as a simple bulge. The applications of this program to
more complex parts have not yet been shown successful. The main problem lays on the
development of the process parameter adjustment strategies. It may be because of the fact
that the adaptive simulation only has the part quality information from the current and
part time steps to project the proper process parameters in the future. Unlike the adaptive
simulation approach, the optimization based simulation approach has access to entire
deformation history to use for generating the optimum process. Of course, the trade off is
the large computation time. Clearly, the adaptive simulation approach is still worth
further research, as it is an attractive approach for a rapid loading path determination.
163
0.12
Upper wrinkle limit
0.1
Lower wrinkle limit
0.08
0.06
0.04
0.02
Better wrinkle control
0
0.0
0.2
0.4
0.6
0.8
1.0
Pressure (GPa)
0.07
0.06
Wrinkle control
(increased-decreased pressure)
0.05
0.04
0.03
0.02
0.01
0
0
4
Axial feed (mm)
(b)
Figure 6.18:
164
0.16
SF part thinning
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0.00
0.0
0.2
0.4
0.6
0.8
1.0
Figure 6.19:
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0
Figure 6.20:
165
CHAPTER 7.
CONCLUSIONS AND FUTURE WORK
7.1 Performance Comparison of Different Loading Path Determination Methods
So far in this research, three main different methods for loading path determination
were developed: a) Self-Feeding simulation (SF), b) Optimization-based simulation
(OPT), and Adaptive simulation (AS). These methods have different advantages and
disadvantages depending mainly on the complexity of part geometries and lead-time
dedicated for designing of loading paths. In this section, all of these methods were
benchmarked through applying them in determining loading paths for the same simple
bulging, discussed in the earlier chapters. The loading path determination of this simple
bulge using OPT and AS is already discussed in chapter 5 and 6, respectively. The
detailed SF simulation work on the simple bulge is omitted in this report due to its
simplicity. The SF method is explained in details with application examples of more
complex part geometries in Chapter 4.
Figure 7.1.a compares all the different LPs determined by the methods. Figure 7.1.b
and Figure 7.1.c show the developments of part maximum thinning versus normalized
part volume, and the part thinning distributions, respectively. Then, Table 7.1 compares
the performance of these methods. The performance criteria are the computational time
spent in obtaining the final loading paths, the final part maximum thinning, and the total
axial feed amount (for one side). Simulation computational time varies from computer
to computer. Therefore, in this comparison, the number of simulation runs required in
each method is compared. For the specific HP workstation machine used in this work, it
took about 2 minutes for a single simulation run of this simple bulging.
168
0.08
AS
Pressure (GPa)
0.07
Optim ized
Best SF
0.06
0.05
0.04
0.03
0.02
0.01
0
0
10
12
14
(a)
0.1
0.08
0.06
Best SF
0.04
AS
0.02
Optim ized
0
0
0.2
0.4
0.6
0.8
(b)
0.10
Part thinning
0.05
0.00
-0.05
20
40
60
-0.10
-0.15
-0.20
Best SF
AS
Optimized
-0.25
(c)
Figure 7.1:
Table 7.1:
SF
AS
OPT (Gradient)
9.5
8.0
2.5
12.30
7.30
8.80
0.08
Optimized
0.07
SF#1
Pressure (GPa)
0.06
SF#3
Best SF
SF#2
0.05
axial feed scaling : LP stretching
0.04
0.03
0.02
0.01
0
0
10
12
14
Axial feed (m m )
Figure 7.2:
170
AS required the smallest number of simulation runs among all the other methods. Only
2 simulation runs were needed to obtain a feasible loading path using AS, while the SF
method required 4 runs and the OPT method spent over 275 runs. However, the OPT
yielded the optimum loading path that minimized the part maximum thinning, while the
other two methods resulted in only feasible (i.e. not optimum) loading paths that
hydroformed successfully without any wrinkles or fracture.
All the loading paths predicted by the three methods show similarity in the typical shape
of proper THF loading paths smaller pressure with large axial feed in the beginning
then high pressure with small axial feed towards the end of the process, see Figure
7.1.a. The AS loading path is most similar to the OPT loading path in both shape and
value. However, their part maximum thinning progress curves, Figure 7.1.b, are quite
different. The AS part has a much larger maximum thinning than that of the OPT part at
the end of the process (i.e. 8.0% compared to 2.5%). This shows that this THF part (i.e.
parts with large expansion in general) is very sensitive to the shape and value of loading
path. Nevertheless, it is encouraging that the AS method seemed to predict a feasible
loading path that is close to the optimum loading path. This implies that the AS method
can give a good initial loading path that may shorten lead-time in further numerical
optimization or physical process prototyping.
The maximum thinning of the AS part is just slightly smaller than that of the SF part
(i.e. 8.0% compared to 9.5%). The improvement in terms of maximum thinning from
the AS method may seem insignificant when compared to that of the SF part.
Interestingly, the SF part needs a much larger axial feed (40% more, 12.30 mm
compared to 7.30 mm) than that of the AS part. In other words, SF method predicts the
loading path with unnecessarily large axial feed. This large axial feed results in
excessive thickening of the part in the guiding zone area, see Figure 7.1.c, when
compared with the thinning distributions of the AS and OPT parts. This SF part can be
undesirable in cases where part weight is of a primary concern.
171
This problem of large axial feed in the SF part comes directly from the scaling-up
method of axial feed used in the current SF method - through the simulation sequence
of the SF method, while pressure curve is fixed, axial feed is continuously being scaled
up until part maximum thinning falls below the fracture level. This SF method basically
searches for the best loading path by stretching the first found loading path (i.e.
loading path obtained from pure expansion without any forced axial feed, refer to
Chapter 4) along the X-axis, see Figure 7.2. In the figure, it can be seen that the SF#2
loading path is the most similar to the optimized loading path. It may be speculated that
if the pressure level in the SF#2 loading path had been lowered the resultant part could
have been as of good thinning distribution as that of the OPT loading path. However,
since there is no mechanism to adjust the pressure level in the current SF method, this
method always tends to give unnecessary large axial feed. Unlike the SF methods, the
AS and OPT methods, both axial feed and pressure are free to be adjusted. This allows
more degree of freedom in the process parameter design thus parts can be hydroformed
to more strict part quality (i.e. small thickening in the guiding zone).
Table 7.2 lists the pros and cons of the common trail-and-error simulation method for
THF loading path determination. From the brief comparisons of all the loading path
determination methods and the earlier discussions in this report, the main advantages
and disadvantages of the three methods are summarized and compared in Table 7.3.
This summary is useful in selecting the proper method when designing process
parameters for any given parts.
Methods
Advantages
Requires no extra programs
Table 7.2:
Disadvantages
Very time consuming process
Requires experienced users
Sometimes, the process modification
can be non-systematic (no clear
directions)
Methods
Advantages
Disadvantages
(Self-Feeding
Simulation)
(normally < 5)
Requires no extra programs
Results in the optimum loading path
Once running, no human involvement is
Works for any given part geometries
Very versatile for the fact that any
Simulation)
pressure curve
Tends to predict unnecessarily large
Simulation sequence requires human
involvement
Very large computational time is
required (normally >>100 runs)
Computational time increases with
required
(Optimization-based
OPT
SF
variables
be designed a priory
Process parameter adjustment
2 runs)
(Adaptive Simulation)
AS
required
Allows some alive wrinkle during the
exist
Table 7.3:
Bulges. A bulge can be identified when the tube cross sectional perimeter rapidly
increases (with either a round, square or irregular shape) without any preferential
radial direction.
Bends. In this work, the part spline shapes (1-D, 2-D, and 3-D) are also considered
as bend features.
Most small THF parts have one of these geometrical features that come in different
sizes and shapes, see Figure 7.3. For larger THF parts, see Figure 7.4, the parts are
usually consisted of these common geometrical features located along the part axis
spline.
174
Categories
Materials
Part Examples
Piping
Stainless Steel
(AISI 304, AISI
309)
Aluminum
alloys
Structural
parts
Low to medium
carbon steels (A
570 Gr. 36, A
738, etc.)
Aluminum (AA
1050, AA 5015,
etc.)
Engine and
drive cases
Table 7.4:
Case hardening
steel (SAEM
1015, SAE 1045,
SAES 115)
(a)
Figure 7.3:
(b)
(c)
176
1D
3D
2D
(a)
Figure 7.4:
(b)
Axial feed
(material disp.)
d0ax
d1ax
Straight (guiding
zone)
d0ax
d2a
Bend
d3ax
Straight
d4ax
Simple Bulge
T-shape
d1ax
d2a
Material
displacement
Figure 7.5:
d3ax
d4ax
177
Pressure-dominant process: Parts that fall into this category normally have
geometrical features (such as sharp bends located near the tube ends) that really
limit effective axial feed into the center of the part or totally prevent possibility of
axial feeding. In this case, the process almost becomes like pure expansion without
any axial feed. In other words, there is no need to control the axial feed in this case.
o
Suggested Process Design Methods: The Adaptive Simulation (AS) and Self
Feeding (SF) approaches are inappropriate since there is no need to find the
axial feed curve. Moreover, the pressure curve can just be applied linearly and
can be analytically calculated using the calibrating pressure formulas already
available. An optimization-based simulation approach (OPT) could be applied
but not for process design; instead it can be used to determine the best value of
the initial tube diameter or the correct choice for the shape of the extrusion (i.e.
optimum part/preform geometry design).
have a protrusion with a straight and short spline belong to this category. In this
case, axial feed is not only possible, but also required in order to obtain a significant
178
useful protrusion height. The pressure is applied mostly to prevent wrinkle but does
not contribute to the useful height of the protrusion, as discussed in Chapter 3. In
other words, in this case, the working range of pressure is large such that control of
the pressure is relaxed.
o
categories above. Parts in this category are mostly bulged parts with medium to
large expansion ratios. In this case, axial feed and pressure (i.e. manipulation of
axial feed and pressure) are equally crucial to the success of the forming process.
o
depending on the amount of axial feed required by the part geometry one
method will be more appropriate than the other. The amount of axial feed
necessary can be estimated using the volume constancy. If the axial feed
necessary is small then SF should be used because the SF method is currently
more robust that the AS method. If large axial feed is required, the AS seems to
be a better choice since in those cases both fracture and wrinkling can easily
occur and the process window seems to be narrow. However, again, the AS
method needs more improvements to handle complex part. The SF approach
could be used also, but it could take to many iterations before obtaining an
acceptable solution. Furthermore, use of the SF method tends to result in
179
unnecessarily large axial feed, previously discussed. The question of the exact
amount of axial feed to be considered small or large is still difficult to answer.
Experience, unfortunately, plays an important role here.
Based on the discussions above, a flow chart is given; see Figure 7.6, to systematize the
procedure of how to select proper FE methods to determine loading paths for THF
parts.
7.3 Conclusions
This research work was intended to develop methodologies for design of part
geometries and process parameters in tube hydroforming processes. The specific goals
of this study were to develop a) part design guidelines for THF processes that facilitate
engineers to bring conceptual THF part designs to production more efficiently and b)
methodologies for design and optimize loading paths in THF using process FE
simulation.
It was realized during this study that THF part geometries could vary so much from
very simple to very complex. Thus, generating new THF part design guidelines (besides
the guidelines already available in literature) seemed to be a very backbreaking task and
may not be as useful. It was then realized that part geometry and process parameters
were very much interrelated (i.e. design for manufacturing). Thus, the main goal was to
focus only on developing systematic and time-efficient FE approaches to determine
proper process parameters (i.e. loading paths). This could be used to evaluate THF part
design for manufacturability, thus, in turn, fulfilling the part design guideline objective
as well.
180
Part Geometry
Dimensions of
Analysis
Database
Experience
Pressure-dominant process
Pressure-axial-feed-driven process
Axial-feed-dominant process
Analytical
SF
OPT
OPT
AS
AS
OPT
Figure 7.6:
Adaptive simulation approach (AS) for rapid THF loading path determination was
developed. An AS program (in FORTRAN) was coded and integrated into the
internal solver of PAM-STAMP. A couple of geometrical wrinkle indicators and
process parameter adjustment schemes were created and implemented as the main
working components in the AS program. The program was, so far, only applied
successfully on the simple bulging process. The current AS program is not
sophisticated and robust enough to handle more complex part geometries. The main
problem is to come up with a proper scheme of process parameter adjustment for
those parts. This approach clearly requires further research. Nevertheless, this study
provided a general framework of adaptive simulation method, which may be
adopted for process parameter design of other metal forming operations such as
sheet metal forming.
All these advanced FE methods developed in this study for loading pat
determination certainly have shortcomings as they always come with benefits. Pros
and cons of all the methods were compared quantitatively and qualitatively in terms
of lead-time, final part quality obtained, and practicality. Some practical guidelines
were also given in selecting proper methods for determination of loading paths for
most typical THF parts.
183
Develop combined applications of OPT and SF: a) using SF results as the initial
guess, b) using SF results to reduce the number of design variables of the loading
path. All of these is to mainly cut down the large computational time in the normal
OPT approach.
Apply the AS program to more THF part geometries as continuously improve the
process parameter adjustment schemes.
Collect the loading paths and experience gain from usages of these FE methods and
store them in THF-PAL* database for future use. (* THF-PAL: a THF database for
part and process design, on-going effort in the ERC-THF consortium)
184
LIST OF REFERENCES
[Aita, S., 1992]
[Asnafi, 1999]
N. Asnafi, Analytical Modeling of Tube Hydroforming, Thinwalled Structures 34 (1999), pp. 295-330
[Asnafi, 2000]
[Aue-U-Lan; 1999]
[Brewster, 1996]
[Cao, 1994]
185
[Chung, 1997]
[Deb; 1998]
[Dohmann, 1991]
[Dohmann, 1996]
[Doege, 2000]
[Donald, 2001]
[Feng, 2000]
J.P. Feng, and Z.J. Luo, A Method for the Optimal Control of
Forging Process Variables using the Finite Element Method
and Control Theory, J. Mater. Proc. Technol. 108, 2000,
pp.79-84
186
[Fourment, 2001]
[Filice, 2001]
[Gelin, 2002]
[Ghouati, 2000]
[Grandhi, 1993]
R. Grandhi, A. Kumar, A. Chaudhary, and J. C. Malas, Statespace Representation and Optimal Control of Non-linear
Material Deformation using the Finite Element Method,
International Journal for Numerical Methods in Engineering,
Vol. 36, 1967-1986 (1993)
[Haug, 1998]
[Hill, 1958]
[Hora, 1999]
[Hutchinson, 1974]
[Kim, 2002]
[Koc, 1998]
[Koc, 2002]
[Kusiak; 1996]
[Lei, 2001a]
L.P. Lei, D. H. Kim, S.J. Kang, S.M. Hwang and B.S. Kang,
Analysis and Design of Hydroforming Processes by the Rigid
Plastic Finite Element Method, J. Mater. Proc. Technol. 114,
2001, pp.201-206
[Lei, 2001b]
L.P. Lei, B.S. Kang and S.J. Kang, Prediction of the Forming
Limit in Hydroforming Process using the Fintie Element
Methods and a Ductile Fracture Criterion, J. Mater. Proc.
Technol. 113, 2001, pp.673-679
[Levy, 1999]
189
[Nordlund, 1997]
[Nordlund, 1998]
[Mosca, 1995]
[Roux, 1998]
[Schuler, 1998]
[Shah, 1997]
[Shr, 1999]
[Shen-Zhang; 1999]
190
[Strano, 2001a]
[Strano, 2001b]
[Thomas, 1998]
[Tibari; 2000]
[Vanderplaats, 1984]
[Whittle, 1982]
[Yang, 2001a]
J.B. Yang, B.H. Joen and S.I. Oh, The Tube Bending
Technology of a Hydroforming Process for an Automotive
Part, J. Mater. Proc. Technol. 111, 2001, pp.175-181
[Yang, 2001b]
J.B. Yang, B.H. Joen and S.I. Oh, Design Sensitivity Analysis
and Optimization of the Hydroforming Process, J. Mater. Proc.
Technol. 113, 2001, pp.666-672
191
APPENDIX A
FLOW STRESS DETERMINATION
The Hydraulic Bulge tooling was designed and manufactured by the ERC/NSM. This
tooling is used for determination of flow stress data of tubular materials. The flow stress
determination procedure is shown schematically in Figure A.1. The detailed procedure
can be found in the THF/ERC report, Prediction of Tubular Material Properties for
Aluminum Alloy 6260 T4, [Aue-U-Lan; 1999]. With online displacement
measurement in the testing tooling, point data of bulge height (hi) versus internal
pressure (Pi) are obtained, and then accurate flow stress can be determined.
The Hydraulic Bulge tooling was used to determine the flow stress of the SS304 tubing
used in the Y-shape hydroforming experiments, see Chapter 3. Figure A.2 shows the
pressure versus bulge height measured from bulging experiments of SS304 (to = 1.5
mm). Then, the flow stress curve, Figure A.3, was calculated from the pressure vs.
bulge height curve.
Data
Acquisition
Potentiometer
Tube Stress
(compute data)
Pi
Flow Stress Curve
hi
= K ( 0 + ) n
Pi = internal pressure
True Strain
hi = bulge height
Figure A.1:
9000
60
8000
50
40
Bursting point
6000
5000
30
4000
20
3000
10
Deformation starts
2000
Presssure (MPa)
Pressure (psi)
7000
1000
0
-10
0
10
12
Figure A.2:
800
700
600
500
400
300
200
100
0
0
0.1
0.2
0.3
0.4
0.5
Effective strain
Figure A.3:
= K ( 0 + ) n
Outside Diameter
50 mm (1.968)
Wall Thickness
1.5 mm (0.059)
0.584
Pre-strain ( 0 )
0.06
Tube Supplier
SPS, Germany
Table A.1:
Based on the flow stress curve determined, the mechanical properties of the SS304 were
calculated
through
fitting
the
flow
stress
data
to
the
flow
stress
194
APPENDIX B
DETERMINATION OF FRICTION COEFFICIENT AT GUIDING ZONE
The lubricant used in the Y-shape forming experiments at SPS, Germany, was Gleitmo
965 supplied by Fuchs Lubricants Co. Gleitmo 965 is water based solid film lubricant.
It is especially suitable for metal forming applications with high deformation rates such
as high-pressure hydroforming and cold extrusion.
Figure B.1 shows the ERC friction tooling. This tooling was designed for conducting
tests that determine interface friction coefficients at the guiding zone in tube
hydroforming processes, more details can be found in [Tibari; 2000]. The tooling
allows testing at various internal pressure levels and sliding speeds (i.e. axial feeding
punch speed).
Fa
Load Cell
Fa
sax
Rod
Tube
Sliding Direction
FFR
Ff
pi
FFR
Insert
Figure B.1:
In the Y-shape forming experiments, the applied pressure versus time curves were in the
range of 200 800 bars (this does not include high pressure level in the calibration
stage, during which there was no axial feeding applied), the right punch speed was 8
mm/s, the left punch speed was 4 mm/s, and the longest axial feed was 80 mm (on the
right punch). Based on capability of the friction tooling, three pressure levels were
chosen for the experiments, as shown in Table B.1.
interface friction coefficients of Gleitmo 965, which was applied on SS304 tubes,
determined at various pressure levels and sliding speeds.
Test #
Lubricant
Material
Pressure
(bar)
Sliding
Speed
(mm/s)
Sliding
Length
(mm)
Average
Friction
Coeff.
Gleitmo 965
SS 304
250
80
0.034
Gleitmo 965
SS 304
400
80
0.066
Gleitmo 965
SS 304
600
80
0.051
Gleitmo 965
SS 304
400
80
0.072
Gleitmo 965
SS 304
600
80
0.056
Table B.1:
196
APPENDIX C
OPTIMIZATION ALGORITHMS
The optimization algorithms applied in the Optimization-based simulation approach,
Chapter 5, are a) gradient approach using Augmented Lagrangian Method, and b)
Response Surface Method (RSM), which are both available in PAM-OPT. These two
different optimization algorithms have their applications different optimization
scenarios (i.e. generally, gradient methods are suitable for problems with a small
number of design variables; and RSM methods are suitable for problems with a large
number of design variables). This appendix gives a brief mathematical overview of
these two optimization algorithms.
Augmented Lagrangian Method (ALM)
In solving non-linear optimization problems with constraints, one of the most popular
set of techniques is Sequential Unconstrained Minimization Techniques (SUMT).
Generally, this approach is designed to minimize the objective function as
unconstrained function but to provide some penalty to limit constraint violations. The
classical approach to using SUMT is to create a pseudo-objective function of the form
[Vanderplaats, 1984]: ( x, rp ) = F ( x) + rp P( x) , where, F ( x) is the original objective
function and P( x) is an imposed penalty function, the form of which depends on the
SUMT being employed. The scalar rp is a multiplier which determines the magnitude
of the penalty, and rp is held constant for a complete unconstrained minimization; the
subscript p is the unconstrained minimization number. Figure C.1 shows how the
penalty function transforms a constrained optimization problem to an unconstrained
optimization problem using pseudo-objective function.
197
(a)
Figure C.1:
(b)
Augmented Lagrangian Method (ALM) provides a way to include the conditions for
optimality, which is derived from Kuhn-Tucker conditions, into the SUMT in order to
improve its efficiency and reliability. This method can reduce the dependency of the
algorithm on the choice of the penalty parameters and the way by which they are
updated during the optimization process [Vanderplaats, 1984]. The ALM function can
be written as follows:
r
L( x, , r ) = f ( x) + i i ( x) + i2 ( x)
2
i
With i ( x) = Max[ g i ( x), i
r
Where f(x) is the objective function, gi(x) are the constraint functions number i = 1,, I
with an upper bound equal to 0 (gi <= 0), i are the Lagrange multipliers of the
constrain functions number i = 1,, I, r is the penalty factor, and i is the augmented
constraint functions number i = 1,, I.
198
The augmented method consists in minimizing the function L(x,,r) on x with and r
being constants. Then, one modifies values of and r according to the iteration
strategies (i n +1) = (i n ) + r ( n ) i( n ) , which tend towards the classical Lagrange multipliers
and, r ( n +1) = r ( n ) , where the value of the penalty factor r is increasing if the constraint
violation doe not decrease, with being constant, and where n is the number of current
iteration [Haug, 1998]. Figure C.2 shows the iteration scheme of this method. The
minimization of the pseudo-objective function is solved by the quasi-Newton methods
such as Davidson-Fletcher-Powell (DFP) and Broyden-Fletcher-Goldfarb-Shannon
(BFGS). As for the unidirectional minimization (line search), an algorithm of quadratic
approximation is used.
Figure C.2:
Figure C.3:
i =1
i =1
f ( x) = 0 + i x x + i xi2 + ij xi x j ,i < j
i
Unlike the standard RSM, the PAM-OPT RSM algorithm uses approximated functions,
which have similar characteristics to the B-Splines in CAD applications. The main
characteristics are as follows [ESI Software, 2001]:
The number of points is free (in the standard RSM, the number of points is equal to
the number of coefficients of the polynomial: (n+1)(n+2)/2, n = design variables)
The approximated functions pass through all points, and the approximated function
values, calculated in a given position, only depend on points close to this position.
The shape of these approximated functions is always good. For example, these
approximated functions cannot create artificial local minima like polynomials can
do.
During optimization iterations of the PAM-OPT RSM algorithm, it is possible to
increase the response surface accuracy by locally reconstructing the response surface
bounding the solution point found from the previous iteration. This requires calculations
of a few new points. The algorithm then uses the full set of all calculated points to
reconstruct new response surfaces, which will be used to find a new solution. The
design parameter domain is kept the same throughout the optimization process. This
algorithm is called Adaptive Response Surface Method. Figure C.4 shows how this
algorithm works in a two design variable space, X1 and X2.
The first figure shows the design domain specified by the design variables.
The second figure shows the first iteration. Four initial points are calculated
(launching of solvers: PAM-STAMP) on the design domain. The calculated point
201
set is used to build response surfaces (for objective function and constraint
functions). These response surfaces (approximated function values) are then used to
find an approximated solution.
The third figure shows the second iteration. The real function values of the
approximated solution are then calculated (launching of solvers). Three other points
are also calculated close to the intermediate solution. The full calculated point set is
used to build new response surfaces. A new approximated solution is then
calculated.
The fourth figure shows the last iteration. The same method as in the previous
iterations is used in all intermediate iterations. It should be noticed that the final
solution is closely surrounded by points (showing that the response surfaces become
most accurate near the final solution).
Figure C.4:
Example of optimization (2 design variables) progression using PAMOPT adaptive response surface method [ESI Software, 2001]
202
APPENDIX D
INTERFACING BETWEEN PAM-OPT AND PAM-STAMP
This appendix gives a brief overview of the interaction (interfacing) between PAMOPT and PAM-STAMP. First, a general flow chart of the PAM-OPT algorithm is
given, and each important component in the algorithm is described. An example of THF
optimization input file is also given and explained as to demonstrate how PAM-OPT
formulates the problem and extracts the simulation results from PAM-STAMP. The
example given in this appendix is the loading path optimization of the simple bulge
hydroforming using the ALM gradient method, discussed in Chapter 5.
General PAM-OPT Algorithm flow chart
Figure D.1 shows the general flow chart of PAM-OPT in both user level and internal
algorithm level. The following files are in the user level [ESI Software, 2001]:
Algorithm Input Data Set File: To initialize an optimization, the user must prepare an
input file to the PAM-OPT Algorithm. This file set is actually consisted of two files:
a) Composite Data Set file (*.cds), and b) Auxiliary set file, see Figure D.2. The
composite data set file contains information that defines the problem of optimization:
- Definition of design variables,
- Definition of constraint functions and of the objective function,
- Design variable values which define the starting point (initialization),
- Optimization method used, and convergence parameters
The auxiliary set file contains executable UNIX commands that extract simulation
results from the solver results (PAM-STAMP) using PAM-VIEW script commands.
Algorithm Signal file: In addition, the user can make a signal file to change
203
creates a number of output files (history file, curve file, restart file).
The Algorithm reads design variable values from the Algorithm input data set file to
initialize the optimization. Then, it calls a process, which works out optimization
function (and gradient) values, as well as constraint set values. In this process level, the
FE process simulation PAM-STAMP solver is run according to the input data. The
called process reads design variable values from the process input data set file and it
will write function (gradient) and constraint set values onto the Algorithm process
output data set file. During an optimization run, the user can interact with the Algorithm
module via the Algorithm signal file. Finally, the Algorithm module will write the
results of the optimization on several Algorithm output files.
PAM-STAMP
Figure D.1:
Figure D.2:
Below is an example of a cds file (composite data input file), an input file for PAMOPT optimization. This file is an excerpt from the bulge.cds formulated for the loading
path optimization of the simple bulging (Section 5.4.1) with minimum thinning
distributions using gradient method (ALM). In the file, there are two main sections: a)
PAM-OPT keywords, and b) executable program baseline (PAM-STAMP input file of
the simple bulge simulation), see Figure D.2. The input file was written with the help
OPT-EDIT software [ESI Software, 2001]. From Section 5.4.1, there are 10 design
variables (5 for axial feed velocity curve and 5 for pressure curve). The objective
(minimizing thinning distribution) and constraint (filling up the control volume)
functions were formulated using some parameters (e.g. &thn1, &vol, &dx, and etc.)
extracted from the simulation results. The auxiliary set file (script file) PVcmdBulge2_2
was written using PAM-VIEW commands to extract the FE results. PAM-OPT modifies
the design variables through using stickers placed on specific locations in the curve
definitions defined towards the end of the executable program baseline file.
205
206
above
0
1
0
1
0
1
207
YEND
$
CALVAL/ &vol
&&vol_crv
0
1
YEND
0
$
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
CPFILE/ 2
AUXILIARYset
$---5---10----5---20----5---30----5---40----5---50----5---60----5---70----5---80
$
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
EXEINP/
$
$ This file is generated by PAM-GENERIS version 2000.1 on 2002/08/29 at 17:27:12
$ PAM-GENERIS Version 2000.1 - Compiled 2001/05/03
$
.
.
$
$ FUNCTIONS CARDS
$
$---5---10----5---20----5---30----5---40----5---50----5---60----5---70----5---80
#GPNAM Curve
1
FUNCT /
1
7
1
1
0
0
0
0
0.01
5
#STICKER/fd1,-1
#
??????
0.2
5
#STICKER/fd2,-1
#
??????
PAM-OPT changes axial feed velocity variables
0.4
5
here in the velocity curve function definition
#STICKER/fd3,-1
#
??????
0.6
5
#STICKER/fd4,-1
#
??????
0.8
5
#STICKER/fd5,-1
#
??????
1.05
5
#GPNAM Curve
2
FUNCT /
2
6
1
1
0
0
0
0
#STICKER/ps1,-1
#
??????
0.2
0.012
#STICKER/ps2,-1
PAM-OPT changes pressure variables here in the
#
??????
0.4
0.024
velocity pressure function definition
#STICKER/ps3,-1
#
??????
0.6
0.036
#STICKER/ps4,-1
#
??????
0.8
0.048
#STICKER/ps5,-1
#
??????
1.05
0.06
ENDDATA
208
APPENDIX E
ADAPTIVE SIMULATION PROGRAM
This chapter explains in detail the preparation of the inputs to the adaptive sub-routine.
Before going into the details of the input file preparation, an overall understanding of
how the sub-routine is implemented to determine a THF loading path is given. In this
appendix, adaptive simulation case of the simple bulge hydroforming is given. The
adaptive sub-routines consist of files as explained below:
1
ldcmod.f: This source code file is a PAM-STAMP user-subroutine that allows the
user to access all the metal forming simulated data and to adjust any loading curve
functions at any simulation time step while the simulation is running. The
abbreviation stands for LoaDing Curve MODification. The user can develop his
own control strategies (i.e. process parameter adjustment algorithms) and
implement them in into the user-subroutine using FORTRAN commands and some
key words provided by Engineering Systems International (ESI). The detailed
information of these commands is given later in this appendix.
make: This is a batch file which executes the commands to compile the core
program with the user subroutine to generate the adaptive simulation executable
file.
All these files are contained in the same directory, under which by running the batch
file, make, the source code ldcmod.f is compiled and linked with the library file
basically record the hydroforming history of the part in the self-feeding condition in
order to obtain the progression curve of part surface area and volume. This is to be used
as reference data in the second adaptive simulation run (i.e. the surface area-to-volume
wrinkle criterion). From Figure E.1, in the first simulation run, the AS program takes in
the control parameter inputs from the user through an input file called control.prm, and
the PAM-STAMP simulation input file, SF_bulge.ps, which is setup with the selffeeding condition. In this step, the control.prm provides the part geometric information
(cross section area, boundary of the control volume, and etc.) necessary for calculating
the surface area-volume progression of the part in self-feeding condition, which is
stored in the output file called SF_results.out. All the other output files are the usual
PAM-STAMP output files, see Figure E.1, in the 1st adaptive simulation run.
ldcmod.f
source code
v2kpa7_102.a
PAM-STAMP library file
SF_bulge.DSY
SF_bulge.THP
SF_bulge.OUT
SF_results.out
AS_bulge.ps
control.prm
SF_results.out
(from 1st run)
Figure E.1:
AS_results.out
In the second adaptive simulation run, the input files to the AS program are: a) PAMSTAMP simulation input file, AS_bulge.ps, b) control.prm, and c) the output file from
the 1st run, SF_results.out. The AS program extracts the results from the FE simulation,
monitors the simulation for any defects (i.e. wrinkling using the surface-volume
criterion and thinning), adjusts the pressure curve and axial feed velocity curve (i.e.
loading curves) if the wrinkle indicator deviates from wrinkle control trajectories, and
gives the adjusted loading curve values in an output file called AS_results.out, Figure
E.1.
Contents of the control parameter input file: control.prm
The user needs to enter some important parameters to the adaptive program through an
input file called control.prm. These parameters are mainly used in a) geometric defect
detection criteria, and b) process adjustment algorithms. Some of the important input
parameters to be given in this file are explained below in detail (some of the input
parameters are omitted here to keep this appendix readable and focused). The units used
here should be the same as the units that one wants to use in the *.ps file of PAMSTAMP, since the values given in control.prm are used for calculating the variation of
the THF loading paths in the simulation. For example, the pressure increment, DELUP,
is to be given in GPa; the axial feed velocity increment, DELUD, is to be given in
mm/ms. The first part of the control.prm collects all the geometric information of the
part necessary: (see Figure E.2)
XLT: X-coordinate on the left side of the control box (or volume)
XRT: X-coordinate on the right side of the control box. The control box should be
placed over all the expansion area.
ENOD: Number of a node located at the tube end. This is used to track the axial feed.
ENODXC: The X-coordinate of the ENOD.
CRSAREA: Cross sectional area of the blank tube.
ARCLEN: Arch length of the tube blank along the circumferential direction.
211
Controlled volume
ENODXC
ENOD
ELE
XRT
Controlled volume
ENOD
ELE
Process end
Figure E.2:
The second portion of the control.prm file requires all the necessary parameter for the
process parameter adjustment algorithms. These parameters indicate the user intention
of how best the loading paths should be adjusted depending on the part geometry. It is
noted here that the options available in the current AS program are suitable for simple
bulge hydroforming only. The following are the descriptions of the parameters:
IFLAG0: Types of simulation run <0> self-feeding simulation run (1st simulation
run, see Figure E.1), <1> adaptive simulation run (2nd simulation run, see
Figure E.1), <2> normal simulation run (this option is used to extract all the
adaptive simulation related results from a normal simulation with already
defined loading paths).
IFLAG2: Process parameter adjustment pattern <11> impulse loading path: in this option, the pressure is applied only when
this is wrinkling, and the axial feed is applied only when this is no wrinkling.
This impulse loading path pattern ((loading path plot of pressure versus
axial feed) is implemented here for the purpose of parameter adjustment
demonstration, but it is not practical for real hydroforming presses.
<12> step loading path: this loading path pattern is originally proposed in
the preliminary work on adaptive simulation concept for THF, see Figure
6.1. In this option, the axial feed is applied while the pressure is kept
constant when wrinkling is not present, and the axial feed is stop while the
pressure is increased when wrinkle is present. Thus the loading path is a step
function.
<13> step loading path with decreased pressure: this loading path pattern is
213
similar to that of the step loading path but the pressure is decreased while the
axial feed is applied when wrinkling is not present. This option is developed
to reduce the excessive thinning that normally occurs when using the step
loading path pattern, see Section 6.4.2.3.
CTST:
Control time step. This is defined in multiples of simulation time steps, e.g
CTST = 100 means that the process parameters will be adjusted every 100
simulation time steps.
CALIVOLN: Normalized part volume cutoff for starting the calibration, after which
the axial feed is stopped and the pressure is increased until VOLN becomes
214
ELE:
Number of tube mesh element located in the critical area of the part (subject
to most excessive thinning, see Figure E.2). This element will be used in the
pure shear control.
PSHTOL: Pure shear state tolerance. This value determines the stress state window of
the chosen part element to be consider pure shear or else. If
0<PSHI<PSHTOL then the pure shear state is present. If PSHI>PSHTOL
then only shear state (not pure shear) is present or PSHI<0 then tensile
biaxial state is present (see example of the output file for the definition of
PSHI).
SVTOL1 and SVTOL2: Lower and upper apex values of the triangle wrinkle control
trajectories, see Figure 6.15a.
DELUD: Axial feeding velocity increment for adjustments (both increasing and
deceasing) of the axial feed over one single time step in a ramping fashion.
Care must be taken in using this parameter. As discussed in Chapter 6, if this
DELUD value is too large dynamic effect error will result. The
determination of a proper value of this parameter relies on experience and
trial and error. For this simple bulge geometry and material, a value of
DELUD = 0.040 mm/ms seems to be appropriate.
DELUP: Pressure increment for adjustments (both increasing and deceasing) of the
pressure over one single time step in a ramping fashion. To avoid the
dynamic effect in the adaptive simulation, a value of DELUP = 0.000020
GPa is suitable for this simple bulge geometry.
REDUP: Multiply factor for the reduction in pressure increment. This parameter is
used along with the IFLAG2 = 13 (step loading path with decreased
pressure). This is to increase the rate of pressure reduction to prevent the
excessive thinning while applying axial feeding. A value between 1-2 should
be used to avoid the dynamic effect.
215
v2kpa7_102.x, see Figure E.1. An excerpt of the output file is given below. It records
relevant geometrical parameters of part as well as the shell element stresses. Each line
of the data is recorded at each simulation time step. In this file excerpt, the shown data
are from the beginning of the forming and towards the end of the forming. The example
output file below is taken from the simple bulge adaptive simulation case. The
following are some descriptions of the parameters recorded:
SURF: Part (FE mesh model) surface area bounded in the control box (unit = mm2)
VOL: Part (FE mesh model) enclosed volume or the PAM-STAMP fluid cell volume
bounded in the control box (unit: mm3). These two parameters (SURF and
VOL) are used for the surface area-volume wrinkle criterion in the 2nd run.
SFD: Axial feed or the tube end node displacement (unit: mm)
216
MSIG22: Stress in the longitudinal direction of the chosen element (unit: GPa). These
two stress parameters are used for the pure shear state control, will be discussed
later.
LEN1: Arch length of the part bounded by the control box (unit: mm)
ARE1: Die-tube profile enclosed area bounded by the control box (unit: mm2). These
two parameters (LEN1 and ARE1) are for the length-area wrinkle criterion, but
they are not used in the process adjustment in this case.
VOL
SMXTHN
SFD
SFP
MSIG11
MSIG22
LEN1
ARE1
1223.758 12181.94
0.00
0.00005
0.000095
0.000028
39.000
191.757
1223.763 12182.00
0.00
0.000001 0.000105
0.000723
0.000208
39.000
191.754
1223.768 12182.11
0.00
0.000013 0.000161
0.001765
0.000423
39.000
191.751
1223.771 12182.20
0.00
0.000053 0.000216
0.002543
0.000411
39.000
191.748
1498.325 17680.59
0.14
4.074892
0.04581
0.221191
-0.079302
39.879
36.884
1498.320 17680.08
0.14
4.077746 0.045865
0.216184
-0.097425
39.879
36.893
1498.329 17680.14
0.14
4.074459 0.045921
0.219557
-0.084367
39.879
36.887
1498.318 17680.30
0.14
4.066472 0.045976
0.232466
-0.048132
39.879
36.876
Please note that, the output parameters shown above are not the entire output parameters
from then 1st simulation run. However, they are the most important and relevant to the
user. It should also be noted that these results from the 1st simulation (self-feeding
simulation) serve mainly as reference data for the real adaptive simulation in the 2nd
simulation run, see Figure E.1. In the future, the AS program can be improved such that
the 1st simulation is conducted automatically without any human-machine interaction.
217
SURFNAS
IWSV
IWSVTOL1
IWSVTOL2
SMXTHN
AFEED
ASD
ASP
PSHI
0.000
-0.001
-0.000018
0.000031
0.000034
0.000
0.002
0.000
0.000
0.723
0.001
0.000
-0.000061
0.000201
0.000223
0.000
4.002
0.047
0.000
0.964
0.012
0.011
-0.000088
0.002136
0.002373
0.000
4.002
0.140
0.014
0.482
0.024
0.023
-0.000188
0.004371
0.004857
0.001
4.002
0.232
0.008
0.079
0.030
0.030
-0.000540
0.005483
0.006092
0.000
4.002
0.324
0.008
0.399
0.041
0.041
-0.000397
0.007429
0.008254
0.000
4.002
0.417
0.013
0.159
0.052
0.051
-0.000077
0.009306
0.010340
0.001
4.002
0.509
0.010
0.091
0.987
0.999
0.000857
0.002317
0.002574
0.090
0.042
7.374
0.185
-1
0.987
0.999
0.000872
0.002313
0.002570
0.090
0.042
7.375
0.187
-1
0.987
0.999
0.000892
0.002310
0.002567
0.090
0.042
7.375
0.189
-1
0.987
0.999
0.000909
0.002307
0.002563
0.090
0.042
7.376
0.191
-1
0.987
0.999
0.000926
0.002303
0.002559
0.090
0.042
7.377
0.193
-1
0.987
0.999
0.000946
0.002299
0.002555
0.090
0.042
7.378
0.195
-1
ASP:
PSHI:
MSIG11
when
MSIG 22
STRTAB(N): Return various strain and stress values from the index table.
NUMCUR: Number of velocity and forces curves
LABCUR(N): External curve number of internal curve number N
FUNVAL(N): Current value of internal curve N
MATTYP(N): Material type of internal material number N
CCM(N,M): For N=68,69,70, returns the G, F, and N values from Hill's Yield Function
for internal element M.
ESI has also provided a routine which iteratively returns various strain and stress values
for each and every element successively in a DO loop. The various values and their
names are listed below.