Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Crashworthiness Assessment
of Automobile Front Ends
using Explicit Finite Element
Formulations
Master Thesis
submitted by
Annika Sorg
in
May 2008
Crashworthiness Assessment
of Automobile Front Ends
using Explicit Finite Element
Formulations
November 23rd , 2007 May 23rd , 2008
Author
Annika Sorg
Matr.-Nr.: 2380825
1st Supervisor
2nd Supervisor
iv
Abstract
The primary topic of this dissertation is the general safety of road traffic
and especially the safety of cars; with emphasis on the assessment of crash
simulations modelled using finite element software.
The main aspects leading to safer road traffic are presented, focussing on
high speed frontal impacts and on the contribution to car safety afforded by
the design of the cars structure - particularly the design of the two engine
mounts towards the front end of a car (the main components to consider
during a crash). These are designed to deform, such that they absorb the
largest part of the released crash energy without penetrating the passenger
compartment. An example of an aluminium extrusion frontal side member
was chosen in order to evaluate the capability of crash simulations with finite
elements to predict real crash test results. Analysis was performed to determine the sensitivity of the model to changes made to individual modelling
parameters, allowing statements on the reliability of the computational results to be proposed.
Parameters altered during these assessments were the element size, the number of integration points along the thickness of the shell elements, the type
of numerical integration used, and the manner of meshing. Computations
that included three different types of failure modelling, and computations
that didnt, were both considered. In addition the strain rate dependency
was analysed. While the work focused more heavily on dynamic computations, quasistatic computations were also performed, and results from each
compared.
The final outcome of the study are recommendations where to pay special
attention when FE software is used for crash simulations; limitations and
possibilities are identified and critically summarized.
vi
Declaration
I, Ms. Annika Sorg, declare that this masters thesis is written independently
and no sources have been used other than the stated references.
..................................................
P lace/Date
..................................................
Signature
vii
viii
Contents
Abstract
Declaration
vii
List of figures
xiii
List of tables
xvii
Nomenclature
xix
Indices
xxi
Abbreviations
xxiii
1 Introduction
1.1
1.2
Structural Outline . . . . . . . . . . . . . . . . . . . . . . . .
2.1
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2
2.2.1
Passive Safety . . . . . . . . . . . . . . . . . . . . . . .
2.3
2.4
Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5
Deformation Zones . . . . . . . . . . . . . . . . . . . . . . . .
ix
CONTENTS
2.6
2.5.1
. . . . . . . . . .
2.5.2
. . . . . . . . . . . . . . . . . . .
11
2.6.1
11
2.6.2
12
2.6.3
Consumer Tests . . . . . . . . . . . . . . . . . . . . . .
16
2.6.4
Other Tests . . . . . . . . . . . . . . . . . . . . . . . .
21
23
23
3.1.1
Occupant Cell
. . . . . . . . . . . . . . . . . . . . . .
23
3.1.2
Front End . . . . . . . . . . . . . . . . . . . . . . . . .
24
3.1.3
Rear End . . . . . . . . . . . . . . . . . . . . . . . . .
25
3.2
25
3.3
26
3.3.1
28
3.3.2
29
4.2
4.3
31
Time Integration . . . . . . . . . . . . . . . . . . . . . . . . .
32
4.1.1
32
4.1.2
33
Contact Definition . . . . . . . . . . . . . . . . . . . . . . . .
37
4.2.1
38
Adaptive Meshing . . . . . . . . . . . . . . . . . . . . . . . .
39
41
5.1
Model Description . . . . . . . . . . . . . . . . . . . . . . . .
42
5.2
Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
5.3
Element Model . . . . . . . . . . . . . . . . . . . . . . . . . .
44
5.3.1
44
CONTENTS
5.3.2
xi
Impactor and Base Support . . . . . . . . . . . . . . .
44
Material Model . . . . . . . . . . . . . . . . . . . . . . . . . .
45
5.4.1
Elastoplasticity . . . . . . . . . . . . . . . . . . . . . .
45
5.4.2
46
5.5
Contact Definition . . . . . . . . . . . . . . . . . . . . . . . .
51
5.6
52
5.7
52
5.4
53
6.1
Procedural Method . . . . . . . . . . . . . . . . . . . . . . . .
53
6.2
. . . . . . . . . . . . . . . . . .
56
6.2.1
Reference model . . . . . . . . . . . . . . . . . . . . .
56
6.2.2
65
6.2.3
71
73
75
Including Failure . . . . . . . . . . . . . . . . . . . . . . . . .
79
6.3.1
Reference model . . . . . . . . . . . . . . . . . . . . .
79
6.3.2
86
6.3.3
86
87
87
6.2.4
6.2.5
6.3
6.3.4
6.3.5
93
xii
CONTENTS
A Input File
97
. . . . . . . . . . . . . . . . . . . . .
97
97
98
99
. . . . . . . . . . . . . . . . . . . . . . . . 104
109
113
117
121
F Tracking Approach
123
125
Bibliography
131
List of Figures
2.1
2.2
2.3
2.4
2.5
2.6
2.7
10
12
13
14
15
17
18
18
19
20
2.8
2.9
3.1
25
xiv
LIST OF FIGURES
3.2
3.3
27
28
3.4
4.1
32
4.2
36
5.1
42
5.2
Cross section. . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
5.3
Side view. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
5.4
46
5.5
48
5.6
49
5.7
50
5.8
51
6.1
54
6.2
54
6.3
55
6.4
55
6.5
58
6.6
59
6.7
60
61
6.8
6.9
Filtered reaction force over time curves for different mesh sizes. 61
62
62
LIST OF FIGURES
xv
63
63
63
64
66
6.17 Reaction force over time curves for 3, 5 and 9 SIMP (5.00 mm
mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
68
68
69
6.21 Reaction force over time curves for 3, 5 and 9 SIMP (2.50 mm
mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
71
71
73
74
6.27 Reaction force over time curves - Quad vs. mixed (5.00 mm
mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
75
6.29 Deformation over time curves from a dynamic and a quasistatic computation (no failure). . . . . . . . . . . . . . . . .
76
77
78
78
xvi
LIST OF FIGURES
6.33 Result of a computation on 1 CPU (2.50 mm mesh). . . . . .
81
6.34 Result of a computation on 2 CPUs with domain level parallelisation (2.50 mm mesh). . . . . . . . . . . . . . . . . . . . .
81
82
6.36 Reaction force over time curves for 1 and 2 CPU (2.50 mm
mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
82
83
85
6.39 Reaction force over time curves for different mesh sizes. . . .
85
87
88
88
6.43 Deformation over time curve from a dynamic and a quasistatic computation (failure included). . . . . . . . . . . . . .
89
6.44 Fracture with respect to the individual failure criteria for quasistatic and dynamic compression. . . . . . . . . . . . . . . .
91
92
. . 127
List of Tables
6.1
65
6.2
70
73
84
6.3
6.4
B.1 Data for the stress-strain-curves for the respective plasic strain
rate pl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
C.1 Data for the ductile damage initiation criteria curves. . . . . 115
D.1 Data for the shear damage initiation criteria curves. . . . . . 119
E.1 Data for the MSFLD damage initiation criteria curve. . . . . 121
xvii
xviii
LIST OF TABLES
Nomenclature
a
B
C
cd
l
d
E
E
pl
pl
F
F
F
fi
H
I
K
k
ks
l
m
M
p
R
acceleration
ratio of principal strain rates
inclination of the barrier
width
damping matrix
wave speed
increment
strain increment
Energy absorption
elastic modulus
strain
equivalent plastic strain at fracture
equivalent plastic strain rate
strain rate
stress triaxiality
force vector
force
vehicle
field variables
height
vector of internal element forces
stiffness matrix
stiffness
material parameter
length
mass
mass matrix
friction coefficient
Poissons ratio
frequency
integral criterion for damage initiation
pressure
rounding off
xix
xx
1
2
3
eq
m
y
t
s
u
u
u
u
u
U
u
v
W
NOMENCLATURE
density
deformation path
stress
principal stress in 1-direction
principal stress in 2-direction
principal stress in 3-direction
von Mises equivalent stress
hydrostatic stress
yield stress
time
shear stress
temperature
shear stress ratio
displacement
velocity
acceleration
acceleration vector
velocity vector
offset
displacement vector
velocity
width
fraction of critical damping
Indices
crit
D
e
major
max
minor
MSFLD
n
n+1
S
stable
t
critical
damage criterion ductile
element
major
maximum
minor
damage criterion MSFLD
at time n
at time n + 1
damage criterion shear
stable
at time t
xxi
xxii
INDICES
Abbreviations
AZT
CPU
DMP
DOF
ECE
EU
FEM
FFC
FLD
FMVSS
GP
HIC
HPC
IIHS
MSFLD
NCAP
NHTSA
NIC
SIMP
SMP
TCC
TCFC
TI
US
USA
V*C
Allianz Zentrum f
ur Technik
Central Processing Unit
Distributed Memory Parallel Processing
Degrees Of Freedom
Economic Commission for Europe
European Union
Finite Element Method
Femur Force Criterion
Forming Limited Diagram
Federal Motor Vehicle Safety Standards
Gauss Point(s)
Head Injury Criteria
Head Protection Criterion
Insurance Institute for Highway Safety
M
uschenborn and Sonne Forming Limited Diagram
New Car Assessment Program
National Highway Traffic Safety Administration
Neck Injury Criterion
Integration points with Simpsons rule
Shared Memory Parallel Processing
Thoracic Compression Criterion
Tibia Compression Force Criterion
Tibia Index
United States
United States of America
Viscous Tolerance Criterion
xxiii
xxiv
ABBREVIATIONS
Chapter 1
Introduction
The design and development of automobiles involves various aspects combining the expertise of engineers, designers, test centres and production
facilities. Strict legislations and standards, especially the safety standards,
dictate the design of the vehicles. This results in the design and design
methodologies being regularly updated. Aspects related to better design
which improve passenger safety, especially the crashworthiness of an automobile, have been of intense scrutinity in the recent years. This has led to
the design departments of automobile industries to employ novel methodologies which help in cutting costs and more importantly time. Highly expensive and time intensive experimental tests have given way to computer
simulations using algorithms which would accurately represent the dynamics involved in such tests. Algorithms using finite elements are commonly
used in these commercial codes. It is very important therefore to understand the intricacies of the way these codes work. This would help us
design algorithms which are better suited for such commercial purposes that
are accurate (in representation of the dynamics) and also not time intensive, as time is money for large commercial establishments like automobile
industries. The present work is a small step forward in this direction.
1.1
Commercial finite element programs offer an almost infinite number of variations to settings designed to model realistic environments and situations,
yet not all options are suited for the same things. In large companies, they
usually have standard protocols that describe how they are to setup programs for any given simulation, for instance in terms of the kind of elements
and the size of the elements required for a specific task etc. Clearly, however, these protocols are only guidelines for specific crash conditions, and
1
CHAPTER 1. INTRODUCTION
not valid for all crash computations. Generally, the user does not know the
means by which a particular set of recommendations has been generated they are unaware of the analyses that lead to the respective protocols, and
typically wont have the time to find out. The aim of this work was to find
out how sensitive present crash simulations are to changes made to single
computation parameters, like the element size, the meshing technique, the
integration method etc. This sensitivity analysis was performed on a very
simple model, of a so-called front side member which was crashed into a wall.
This model represents the main component in terms of energy absorption
during a crash, and is therefore very important to model correctly.
1.2
Structural Outline
Chapter 2
History
The word automobile is derived from the Greek word for self, auto, and
the Latin word for moving, mobilis. An automobile, therefore, is a selfmoving vehicle - rather than one that is moved by another vehicle or an
animal. The alternative title motor car (or just car) is also widely used,
and implies a wheeled passenger vehicle that carries its own motor.
The question of who invented the first automobiles does not have a straightforward answer, and is more a matter of opinion. The first self-propelled
automobiles were invented at some point during the 18th and early 19th centuries by various people, and had different kinds of engines, such as steam
or internal combustion engines, the early forms of which were fuelled by a
mixture of hydrogen and oxygen. The invention of our present-day cars,
however, began in 1885/86 with Karl Benzs patent for a three-wheeled vehicle with a gasoline engine. One year later, independently of Benz, Gottlieb
Daimler and Wilhelm Maybach constructed a motor-driven vehicle, based
on a carriage, with a maximum speed of 16 km/h. This was followed in
1888/89 by Siegfried Marcus, again independently, with the release of the
first petrol driven four-stroke engine, displaying all of the key features of our
present-day cars [Wika].
The spread of cars on the roads was accompanied by increased traffic accidents, especially with the rise in vehicle speeds. Bridget Driscoll is known as
the first person to die in a traffic accident involving an automobile when, on
August 17th 1896, she and her daughter crossed the grounds of the Crystal
Palace in London [Wikb].
Already by the 1930s, death rates due to automobile crashes in the United
States had surpassed 15.6 fatalities per 100 million vehicle miles, and contin3
ued to climb [BIR01]. It was time for the designers to find ways of making
their products safer.
Greater discussion of these safety precautions can be found in the subsequent section, but its worth noting just how effective these safety measures
were. As displayed in Fig. 2.1, the casualty rate (number of accidents involving casualties) continuously decreased over the last 50 years despite the
number of registered cars increasing over the same period (corresponding to
an increase in the number of kilometres driven on the road).
Figure 2.1: Comparison of driven vehicle kilometres and casualty rate in the
United Kingdom, after [DOT07].
2.2
There are different ways of making road traffic safer, cf. Fig. 2.2. One can
generally distinguish two main categories which are active (primary) and
passive (secondary) safety. Active safety is concerned with accident prevention measures, i.e. safety instructions, traffic steering and driving assistance
systems, whereas passive safety deals with the reduction of accident severity and the consequences of accidents. Passive safety systems include the
development of safety belts, air bags, and the design of a cars structure to
be able to plastically deform, thus allowing the absorption of crash energy
while ensuring a sufficient survival space for the occupants.
Figure 2.2: Different approaches for improving road safety, after [Kra98].
2.2.1
Passive Safety
Passive safety can be split into self and partner protection (Fig. 2.3), if one
considers that not only a cars own occupants may be injured, but also those
in another vehicle. Elements of passive self protection include, for example,
the stiffness of the passenger compartment, a good energy-absorbing deformation zone, and the compliance of biomechanical limit values. Partner
protection is mainly afforded through adjustment of the deformation characteristics of a vehicle to the safety requirements of other participants in the
accident, e.g. occupants of the other car, pedestrians or cyclists.
2.3
To mitigate injuries in car accidents, the structure of the car body has to be
designed according to rules for crashworthiness. One of the most important
requirements is that the car should have different deformation zones in the
front. The basic idea of this comes from Bela Barenyi who was awarded a
patent in 1952 for his description of the decisive features of passive safety.
He divided the car body into three sections, a rigid non-deforming passenger
compartment in the middle and a front and rear crumple zone that are able
to absorb the impact energy by deformation during collision. Newer cars
are also designed for side impacts, enabled by smaller deformation zones on
both sides as well.
2.4. COMPATIBILITY
Generally one can say that there are three basic requirements on the structure of the car body:
sufficient rigidity of the passenger compartment,
sufficient deformation zone to transform the kinetic energy,
compatibility for other road users.
In the early stages of developing a new car, structural values, such as a limit
for the deformation of the engine mounts, are taken to evaluate the safety
quality of the new car. These values are used instead of biomechanical values
measured on dummies, when simulations involving crash test dummies are
not yet possible. However, the structural values are generally chosen such
that subsequent biomechanical values are kept within their limits.
2.4
Compatibility
2.5
Deformation Zones
In current concepts for car bodies, the front longitudinal chassis beams are
the primary energy absorbing structures of a car, accounting for 60 70 %
of the total energy [Ada95]. Therefore engineers focus on these beams with
respect to safety when designing new car bodies.
To optimise the front structure of the car (including with regard to compatibility), the part of the car in front of the passenger compartment (the
survival space) is divided into 4 deformation zones (Fig. 2.6) depending on
the different crash-conditions, and the respective average deformations for
each situation (and therefore also the velocities that usually lead to the
crash). The four deformation zones are for:
Pedestrian protection: the car should be constructed in such a way
that pedestrians remain as unscathed as possible.
Protection at low velocities / urban crash: to maintain a structure
such that the repair costs are kept low.
Partner protection / compatibility: to observe the protection criteria
for the occupants of the other car.
Self protection: to observe the protection criteria for the occupants of
the principal car.
2.5.1
To reduce the accident severity for pedestrians and cyclists, the geometry
of the vehicles front structure comes into play. The effects on pedestrians
during accidents can be kept low by using broad bumpers, filled with energy
absorbing foam material where the loads are not concentrated on just a single
point on the leg. Nevertheless, the deformations should not be excessive in
order to keep the cost of repair as low as possible. This is a difficult situation,
where the goals are conflicting and compromises must be made. Pedestrian
safety is therefore tested in various ways. To name a few, there is
a lower leg test,
a test for pelvis and thigh (upper leg test),
a test for head impact of a normal adult and a child.
Regarding the last test, it is recommended not to fix anything rigid just
beneath the bonnet and to keep at least 6 7 cm space between bonnet and
engine components to make sure the bonnet can deform, and thus absorb the
impact energy. For further information on pedestrian safety refer to [MK06].
Urban crash testing is conducted according to the AZT-test (or Danner test)
where the focus is on the repair costs. It is a test the insurance companies in
particular are interested in. Levels for physical damage insurance are based
on this test. For more on this see Section 2.6.3.3.
2.5.2
If a car crashes against a barrier, or even worse if two cars crash head-on,
the possible relative velocities are extremely high, as are the requirements to
10
save the occupants. While the front comes to an abrupt halt during crash,
the remainder of the vehicle is still undergoing high deceleration. Substantial
compression forces develop between the front and rear of the vehicle that
have to be absorbed by the structure.
Generally, hollow sections are used for the front longitudinal chassis beams.
For those kinds of beams, two types of failure for axial loading exist: The
bending of the beam or the formation of a crumple zone.
In bending, the beam buckles just to one side without much deformation
to the other parts. The energy absorption is only high at the beginning,
until failure, at which point the load decreases considerably. So the energy
absorption is a mere E 2 kJ/kg for steel, refer Fig. 2.7 and [Ada95].
Figure 2.7: Failure performance of a beam under axial loading, after [Wal05].
11
aim is to achieve the crumple mode in the beginning and allow the beam to
bend for larger deformations only.
The force level required for deformation is strongly influenced by the material of the beam, the thickness, the shape of the hollow section and hence
the cross-sectional area.
2.6
Generally, all tests a car has to pass until it can be released are representations of real accident situations in which the interaction of the occupants
and the car is measured. In addition to these interactions, some technical
measurements like the impermeability (and thus fire-resistance) of the tank
after a crash, or the displacement of the steering column after the crash are
considered. There are quite a lot of combinations one can imagine. Hence,
it is probably easy to see that there are many different tests even for the
assessment of frontal impacts alone. Since the focus of this work is on the
front side members, only those tests relating frontal impact will be considered further.
Two main categories of tests are differentiated. The ones that are obligatory
under the law and the those that are optional, but more oriented to the users
interests. See the explanations and differences in the example of the frontal
impact test in the following sections.
2.6.1
12
In the context of the EU, the standards of all member countries were merged
into a new standard, valid throughout Europe. Further regulations exist for
the USA, Australia, Canada and Japan, where the last three are geared
to those of Europe and the USA. To be able to sell a car in a country,
the respective standards have to be fulfilled. It is advisable to look to the
European and US regulations for guidance, since they are the leading ones.
The big advantage of these standardised tests is that all vehicles are tested
under the same conditions and thus test give comparative information.
2.6.2
In the following the main aspects of the legal standards for Europe and
for the USA are named. That they are legal standards means they are
13
mandatory under the law. To sell a car, or any other vehicle, these tests
at least have to be passed, and the respective requirements fulfilled. Later,
in Chapter 2.6.3, it will be shown that there are further voluntary tests,
apart from the legal requirements and based on the users interests, where
the requirements are a little bit higher.
2.6.2.1
Europe: ECE-R 94
The decisive test in the European norm, released by the Economic Commission for Europe [UNEa], is the 40 % offset impact test, named ECE-R 94
[UNEb], where the car crashes against a deformable barrier which covers
40 % of the cars front side (Fig. 2.10). The barrier face is deformable to represent the deformable nature of the cars. The main requirements of this test
are related to values obtained from crash dummies. The values measured on
the dummies are compared to the biomechanical limit values. All settings
are exactly regulated to make the test as reproducible as possible. Specified
in the regulation one can find forms for communication, the arrangements
of the approval mark, an instruction on how to determine the performance
criteria, everything that has to be looked for regarding the dummies (e.g.
how to install them and how to adjust the restraint system), where to place
the points for the measurements, what has to be measured and how etc.
The test procedure with all its settings is described in annex 3 of regulation no. 94 and the properties and conditions of the deformable barrier in
annex 9. In the following, just the main characteristics are mentioned.
The test vehicle shall be a representative of its series production and in
normal running order. Some components may be replaced by equivalent
14
masses, as for example the tank is not filled with fuel but with water. The
test is carried out with two Hybrid III-dummies, each of them on one of the
front seats. The doors shall be closed, not locked, and should remain closed
during the test. For a successful test, one criterion is that they have to be
able to be opened afterwards to assist quick rescue. The vehicle shall overlap
the barrier face by 40 % 20 mm. The speed of the vehicle at the moment of
impact shall be 56 km/h with no tolerance downwards and 1 km/h upwards.
The barrier is divided into two parts, the main honeycomb block and the
bumper element. The geometry can be determined from Fig. 2.11. Both
parts are made of the same kind of aluminium, but with a different structure,
such that the density of the bumper element is higher than the density of
the main honeycomb block, yielding a different crush strength. The whole
block is fixed to a rigid wall, 200 mm above the ground.
The front face of the deformable structure is perpendicular within 1 to
the direction of travel of the test vehicle (Fig. 2.10). [...] The orientation of
the barrier is such that the first contact of the vehicle with the barrier is on
the steering-column side. Where there is a choice between carrying out the
test with a right-hand or left-hand drive vehicle, the test shall be carried
out with the less favourable hand of drive as determined by the Technical
Service responsible for the tests. ([UNEb] Annex 3, Sections 1.2 and 1.3)
The whole barrier should not deform, such that the car is able to crash
against the wall behind it. The wall does not represent anything in this test
and would distort the results. But since the barriers were designed in the
mid 1990s and meanwhile the vehicles became much more rigid, this in not
fulfilled anymore in most of the cases. Hence, there are activities to improve
the barrier design.
15
Figure 2.11: Deformable barrier for frontal impact testing (ECE-R 94).
[UNEb]
2.6.2.2
In the USA the regulations are formulated in the Federal Motor Vehicle
Safety Standards (FMVSS), provided by the National Highway Traffic Safety
Administration (NHTSA) [NHT]. The decisive tests for frontal impact are
given in the FMVSS 208. The tests in there differ considerably from the
ones in the European norm, cf. [NHT97]. The car here crashes into a rigid
barrier at any speed up to 48 km/h (30 mph) and at any angle between perpendicular and 30 to either side. But it is planned to increase the maximum
speed up to 56 km/h, similar to the European norm, but only for crashes
perpendicular to the barrier, cf. [NHT]. As in the ECE-R 94, the main requirements of this test are related to values obtained from crash dummies
two 50th percentile male Hybrid III-dummies, seated at the front of the car,
and then compared to biomechanical limit values. But, while in the EU, the
dummies must be belt restrained, in the USA the test has to also be passed
with unrestrained dummies. Thus the requirements for the occupant protection systems, like air bags and restraint systems, are higher in the USA,
since there they also have to protect the not-belted passenger. In the EU
one concentrates more on the structural safety of the car. The FMVSS 208
includes threshold criteria for the head (1,000 HIC), chest deceleration (60g),
chest deflection (76 mm) and femur (proximal) lower leg (10 kN), while the
ECE-R 94 includes additionally viscous criteria (V*C) and threshold criteria
for the neck, the knee, lower leg bending (tibia index), foot/ankle compression and compartmental intrusion. In the European 40 % offset impact test
16
a smaller area of the structure must manage the crash energy, while the
demand on the structure in the American full width frontal impact test is
less, but in return the restraint system is put under particular strain (while
the reverse is true in offsets).
Since this thesis is about crashworthiness and in particular about the energy
absorption of the front longitudinal chassis beams and their design, the
decisive test here will be the test after the European norm with 40 % offset.
Apart from the aforementioned test at 48 km/h and the 50th percentile
male Hybrid II-dummies, the US-norm also foresees the same test with 5th
percentile adult female dummies at a speed between 32 and 40 km/h, and
one at 48 km/h with the same 5th percentile adult female dummies but
only perpendicular to the barrier. A 40 % offset test perpendicular to the
barrier is also part of the US standard, but with 5th percentile adult female
dummies and at a reduced speed of 40 km/h, cf. [NHT] and [CRA].
2.6.3
Consumer Tests
Apart from the legal standards, some other testing programmes have been
established. These tests, designated as the New Car Assessment Program
(NCAP), are largely based on the users interests and should help the customer at making her/his decision regarding buying a new car. In countries
where they are used, they are based on the legal standards, but are even
more stringent. If a vehicle wants to get a good evaluation, it has to fulfil
(at least partially) these considerably higher requirements. To assert itself
on the world market, it is therefore important to fulfil these criteria. In the
USA they are called US-NCAP and in Europe, EuroNCAP. The European
NCAP [ENC] was founded by a consortium of ten members (governmental and non-governmental like motoring organizations). In both, US- and
EuroNCAP, the tests are evaluated in stars (max. 5, which means safe).
2.6.3.1
EuroNCAP
As in the legal standards, there is more than one test to be passed for a
vehicle. In the EuroNCAP one can find (amongst other tests for frontal
impact) side impact, pole impact and pedestrian tests. In the following only
the frontal impact test is described, which is again a test with 40 % offset
(Fig. 2.12).
In this test, the car crashes with a velocity of 64 km/h against a deformable
barrier; this is hence an even more severe test of the cars ability to ensure sufficient survival space during an impact (without suffering passenger
compartment intrusion) compared to the ECE-R 94. Contact between the
17
18
Smart ForTwo This car belongs to the category supermini and it might
be surprising that it received four out of five stars in EuroNCAPs Adult
Occupant protection rating [NCAb]. But first of all, the Smart ForTwo has
no rear seats and consequently has not been assessed for child protection
[NCAb]. And secondly, it is essential that no attempt is made to compare
the ratings between cars in different segments or mass groups. The frontal
crash test aims to measure the performance of the car impacting another
car of similar mass. There is no capability to determine what would happen
if cars of widely different masses impact each other, cf. Section 2.4. It is
not primarily the mass difference that has the effect, but the effect that
mass has on the structural stiffness combined with the relative height of the
structures from the ground.
19
20
2.6.3.2
In the US, they also have an optional test based on the users interests. It
is released by the Insurance Institute for Highway Safety (IIHS) [IIH] and
is comparable to the EuroNCAP, explained in the previous chapter.
2.6.3.3
The AZT test, also called the Danner test, is not an obligation according
to the law but, in addition to the legal regulations for low and high velocity tests, it is a supplementary test of a cars performance during accidents
that cause average damage. The test was invented by the technical centre
of the German insurance company Allianz and evaluates whether a car is
repair friendly, (i.e. consideration given to repair costs) while it does not
evaluate values from dummies. The car should be easy to recondition after
an accident, for example via beating out or through partial replacements.
The results are used by insurance companies to classify the car type regarding the levels of physical damage insurance for new car models. During the
test, the principal structures should remain undeformed, that means in return that there are no plastic strains allowed. Particular elements, like the
engine mounts already mentioned, are designed to reduce the deformations.
The settings of the test are illustrated in Fig. 2.16 and the main testing
conditions listed below.
The vehicle crashes with a velocity of 15 1 km/h onto a rigid barrier with
40 % offset (generally) on the drivers side. The barrier has inclination of
= 10 and has to be considerably higher than the front of the car. The
car should not be able to hit against the wall behind the barrier. There
has to be a belted 50 % male dummy in the drivers seat, the tank must be
21
fully filled, the vehicle has to be ready to drive and the battery connected.
Additionally some measurements have to be carried out before and after the
crash, for example the length of the axis and the car body itself. Currently,
there are activities to design additional repair tests to gather information
on the bumper matching problem.
2.6.4
Other Tests
Aside from the tests for frontal impact described above, there are various
further tests. Some for side impact, for rear impact, as well as so-called pole
impact tests for the side and front of a car. They simulate, for example,
the collision of a vehicle against a bollard, a street light or a pillar in a car
park. Also, special rollover tests are developed and real car-to-car crashes
for the sake of compatibility checks. But there are still a lot more tests not
described here.
22
Chapter 3
3.1
3.1.1
24
3.1.2
Front End
The front end is the most complicated part of a car with respect to the structural elements and their respective functionalities. The different automotive
manufacturers follow various concepts for the structure of their cars and it
is therefore hard to describe the general front end. However, one can say
that the front end is essentially built of two wheel-houses, the front top and
bottom cross members, sometimes a motor cross member, sometimes upper
side members, the fenders, the front hood, and the two front side members.
The wheel houses are the most complex elements of a cars front end, yet
despite their size they are very stiff due to their curved shape.
If the car has a cross member design and no other appropriate connections
between the body structure, adjacent assemblies and engine, both sides of
the upper cross members are welded to the wheel houses and the lower
cross members are welded to the side members. The function of the cross
members is to transfer the forces from one side of the vehicle to the other,
such that the loads are about the same level on either side. This is especially
important in a one sided crash impact, as tested in the 40 % offset test, cf.
Section 2.6.
The motor cross member/chassis underframe is a highly profiled sheet-metal
section that carries the engine and serves to make the structure of the front
end stronger, with more twisting stiffness, but is not part of every car. Some
automobile manufacturers prefer to mount the engine on an underframe.
Also the upper side members are not included in the structure of all manufacturers cars. Cars produced in the Far East are especially likely to have
them. They are located below the upper fender strip and provide additional
bending strength, and absorb additional energy in a severe frontal collision,
but they also increase the repair costs of a car.
The fenders play just a subordinate roll in the overall stability of the front
end. Their shape and mounting have to be designed such that the space
between them and the door is not immediately eliminated in a slight crash.
The bumpers are attached to the lower front part of the fenders.
The largest pressing of a car is the front hood. The hood itself is very
unstable and has to be supported on all four sides by the hood frame, a
specially formed strengthening section. The main requirement of the hood
is that it has to yield during a collision with pedestrians, to keep the damage
as low as possible, cf. Section 2.5.1.
The most important structural elements of a car body when speaking about
a crash, are definitely the front side members. Since they are the central
part of this dissertation, their function is described separately in Section 3.3.
3.1.3
25
Rear End
The rear of a car usually has a very simple design, compared to the other
parts. The rear ends of notchback cars comprise two side sections behind
the C-pillars, the trunk floor, the rear apron and the trunk lid. Hatchback
vehicles are generally built the same. The main difference is the trunk lid,
which is a door design, and hinged on the roof. The two side members are
often conjoint with the trunk floor for mounting and supporting the rear
bumper system. According to Anselm [Ans00] energy can be strategically
dissipated by strong profiled interior panels below the edge of the window
from the tail lamp to the C- or B-pillar. Panels around the rear wheel house
provide mutual support to the front in a structure where longitudinal and
cross members are combined. Strengtheners are often welded to the rear
side member to ensure good deformation behaviour during a rear impact.
3.2
As this work concentrates on a front crash, the forces one is mainly interested
in lead longitudinally through the car. During a front crash, a huge mass
pushes from the back of the car. The released energy has to be absorbed
by the two main energy-absorbing structures in the front of a car, the two
front side members.
Figure 3.1: Design junction between front side member and occupant cell
involving three-axis forked beams. [Ans00]
To transmit the forces from the rear to the front of the car and vice versa,
there has to be a continuous load path through the whole car. Usually there
are two paths, one on either side. All the load paths have to come together
in the two front side members. Therefore they have to continue under the
26
floor of the vehicle with several transverse supports to the centre tunnel,
the A-pillar and especially to the longitudinally stable structural zone of
the sills. This can be achieved, for example, by a three-axis forked beam
as depicted in Fig. 3.1. The solutions again differ between the automobile
manufacturers.
To have a continuous flow of forces, all junctions must be linked harmoniously and any inconsistencies, like steps, bends, taperings, reinforcements,
changes in thickness, cut-outs, drill holes and notches, are to be avoided
as far as possible, especially the ones introduced subsequently and in highstress regions. They disrupt the moment of resistance and thus weaken the
structure. The incorporation of holes into simulation programmes is thus
very important. Introducing only one hole into a primary component like a
front side member can completely change the energy dissipation behaviour.
To join the members, most commonly spot welding is used. Although in
some cases there are only a few welds needed, it is better to have more to
get a smoother transition.
In an optimally designed vehicle, the impact forces introduced at single
points are transferred to other areas of the front structure, such that locally
severe deformations are prevented. This is very important in overlap tests,
where the velocity is reduced only by a small part of the front structure.
To protect the passengers, an intrusion of the front side members into the
passenger compartment has to be avoided by all means. The front side
members are thus strongly connected to the floor assembly. In such a way
penetration of the bulkhead can be avoided as far as possible.
3.3
The front side members are the most important part of a car in terms of
energy absorption in a front crash. In Anselm [Ans00] one can find several
different references of how much energy is absorbed by the front side members, compared to the total crash energy. After Anselm, Betz/Laschet
state that they absorb at least 45 % of the impact energy in the entire deformation zone by transformation to deformation energy, under FMVSS 208.
On the other hand, Rauser/Grossmann are mentioned, who used a balance of energy determined from a dual-mass model to reach the conclusion
that 79 % of the kinetic energy in a 50 km/h frontal impact test is absorbed
by the vehicle front end structure, 12 % by the powertrain and 9 % by the
bulkhead. Within the front end structure, the energy absorption is further
distributed; 72 % to the front side members, 22 % to the wheelhouses and
6 % to the fenders. Finally Anselm refers to Piech/Behles, who describe
that in a frontal collision, about 70 % of the transformation of energy is as-
27
Figure 3.2: Bow-shaped design of the front side member and flow of forces.
[Ans00]
Optimising the front side members means first of all optimising the shape of
the structure, and secondly the properties of the material used. To be able to
best convert the kinetic energy, they are usually straight with a continuously
increasing cross section, and thus moment of resistance, from the bumper to
the bulkhead. Tailored increasing material thickness additionally supports
this effect as well as strengtheners and beads. Instead of straight front side
members, their external shape can also be single-offset, double-offset or bowshaped (see Fig. 3.2), which is usually worse, since they favour bending. For
the design of the front side members several shapes of cross sections are
used. Some are displayed in Fig. 3.3.
In terms of material, a standard high strength steel of deep-drawing quality
is mostly used. Steel has a high level of plastic absorption of energy and for
an acceptable price a high quality finish can be achieved. High strength steel
has higher energy absorption properties compared to normal steel and, due
28
Figure 3.3: Possible profile geometries for the front side members, after
[Ada95].
3.3.1
29
Force-Travel Curve The integral value of force and travel represents the
energy absorption properties of the structure. The aim is to get a structure that attains the highest possible value which remains constant for an
average folding force. The basic presupposition for optimal consumption of
energy would be a right-angled force-travel curve, which theoretically could
be achieved by a structure that transforms energy constantly during the
whole deformation process.
3.3.2
30
side member contacts the opposing structure. To support the specific deformation behaviour of building regular folds, the front side member must
be given a design element such that the folded buckle really commences at
its beginning, a so-called crush initiator. This could be a preformed fold
or crease, or can be achieved by means of adding a bolted or welded rail
extension with reduced panel thickness, see Fig. 3.4.
Figure 3.4: Front side member deformation aids (crush initiators). [Ans00]
Chapter 4
31
32
4.1
Time Integration
Especially when solving nonlinear problems, the solution cannot be calculated by solving a single system of equations, as in linear problems. Usually
several iterations are needed to determine an acceptable solution for the
nonlinear analysis. There are generally two ways of doing computations,
the implicit way, which ABAQUS/standard uses to solve, and the explicit
way, used in the ABAQUS/explicit solver. The main principles of both
strategies will be explained in the following.
4.1.1
(4.1)
is integrated at the end of the time step (tn +tn ). By applying a forward
difference scheme (implicit time integration), the following relations are obtained and visualised in Fig. 4.1.
u
(t)
u
n+1
u
n
u(t)
tn
u n
u n+1
u(t)
un
tn1
tn
un+1
tn+1
33
u n+1 u n
,
t
un+1 un
=
,
t
m
Fn+1 + t
2 (2un un1 )
=
.
m
+k
t2
u
n+1 =
(4.2)
u n+1
(4.3)
un+1
(4.4)
4.1.2
4.1.2.1
(4.5)
34
the time increments are smaller than the time a material wave needs to cross
the smallest element in the finite element mesh. The increment size depends
solely on the highest natural frequency of the model and is independent of
the type and duration of loading, see later on in Section 4.1.2.3.
Thus, one can say that the smaller the elements used, the smaller the time
increments have to be and the more time is needed for the whole computation. The computational cost is proportional to the number of elements
and roughly inversely proportional to the smallest element dimension. By
refining the mesh in all three directions (in 3D) by a factor of two, the computational costs thus increase by a factor of 2 2 2 due to the increase of
the number of elements and by a factor of 2 as a result of the decrease in
the smallest element dimension, which increases the stable time increment
(cf. Section 4.1.2.3). This means an increase of the computational costs by a
factor of 24 = 16. Disk space and memory requirements are proportional to
the number of elements with no dependence on element dimensions. Therefore, by refining the mesh by a factor of two in 3D, disk space and memory
requirements increase by a factor of 23 = 8.
But even though the time increments in the explicit method are smaller and
more timesteps are needed, the computation is more efficient as the standard
implicit solution method. Due to the immensely increasing stiffness matrices
of large models (about 1.5 million elements in current models), the explicit
method shows great cost savings as the model size increases, compared to
the implicit one, as long as the mesh is relatively uniform. It uses less disk
space and memory [ABAa].
The explicit dynamics method was originally developed to analyse highspeed dynamic events, such as the crash problem at hand, where many small
increments are required to obtain a high-resolution solution. As rapidly
as the load is applied, the structure has to deform. Accurate tracking of
stress waves through the metal sheet is important for capturing the dynamic
response. Since stress waves are related to the highest frequencies of the
system, obtaining an accurate solution requires many small time increments.
By using the explicit method, also contact conditions are formulated more
easily because this method can readily analyse problems involving complex
contact interaction between many independent bodies. Using an implicit
scheme, contact cannot be easily controlled, cf. Section 4.2.
4.1.2.2
In the explicit solver, the central difference method is used to integrate the
equations of motion explicitly through time, using the kinematic conditions
at one increment to calculate the kinematic conditions at the next increment.
35
That means the accelerations calculated at time t are used to advance the
velocity solution to time t + t/2 and the displacement solution to time
t + t, as described in the following, cf. Fig. 4.2.
At the beginning of the increment the programme solves for dynamic equilibrium, which means solving
M
u=FI
(4.6)
(4.10)
36
tt
u
u(t)
u tt/2
u t+t/2
u(t)
ut
t t
ut+t
t + t
4.1.2.3
The stability limit dictates the maximum time increment. For computational efficiency it is worth to choose a time increment as close as possible
to the stability limit, but without exceeding it. It is defined in terms of the
highest frequency in the system, max , by the expression [ABAa]
2 p
tstable =
1 + 2
(4.11)
max
where is the fraction of critical damping in the mode with the highest
frequency. According to [ABAa], it can be shown that the highest element
frequency determined on an element-by-element basis is always higher than
the frequency in the assembled finite element model. Due to the minor complexibility, it is thus more computationally feasible to calculate the stability
limit based on an element-by-element estimate, where it can be defined using the element length (Le ) and the wave speed (cd ) as a property of a linear
elastic material with a Poissons ratio of zero:
Le
tstable =
,
(4.12)
c
sd
E
.
(4.13)
cd =
Thus, if one knows the length of the shortest element and the wave speed
of the material, one can estimate the stability limit, which is smaller the
shorter the element and the stiffer the material, and larger the higher the
mass density.
In nonlinear problems (with large deformations and/or nonlinear material
response), the highest frequency of the model changes continuously and consequently also the stability limit. ABAQUS/explicit has two strategies for
37
4.2
Contact Definition
In a computation where the deformations are large and two parts are likely
or even meant to hit each other, contact has to be defined wherever this can
happen during the process of deformation. The programme therefore has to
be able to detect automatically if two surfaces are contacting each other, and
whenever they do it has to apply the previously defined contact constraints
accordingly. Similarly, it has to detect when two surfaces separate and
remove the contact constraints.
First of all, contact surfaces have to be created from the faces of the underlying elements. They can either be defined single-sided or double-sided,
depending on which side of the element forms the contact or if both sides
do. In ABAQUS, the general contact algorithm (Section 4.2.1) and selfcontact in the contact pair algorithm enforce contact on both sides of all
shell, membrane, surface and rigid surface facets, even if they are defined
single-sided.
38
(4.14)
Tie constraints can be used to tie two surfaces together such that the nodes
on the surfaces have the same motion on either side. This means that
the translational and optionally the rotational degrees of freedom are constrained. For general contact information see e.g. [Wri06].
4.2.1
39
Equal and opposite contact forces with magnitudes equal to the penalty
stiffness times the penetration distance are applied to the master and slave
nodes at the penetration points. The penalty stiffness is chosen automatically by the programme and is dependent on the stiffness of the underlying
element.
The kinematic contact formulation searches for surface-to-surface contact.
It uses a predictor/corrector method, which means, that it is first assumed
that contact does not occur and that if there is an overclosure at the end of
an increment, it will be corrected by modifying the acceleration.
Contact Evaluation In the pure master-slave approach one of the surfaces is the master and the other one the slave. When they come into contact,
the penetrations are detected and the respective contact constraints are applied. This approach will resist only penetrations of slave nodes into master
facets, penetrations vice versa may remain undetected.
The balanced master-slave contact applies the pure master-slave approach
twice, reversing the surfaces on the second pass and can thus resist penetrations of either case.
For the contact pair algorithm ABAQUS/explicit chooses automatically a
master-slave approach based on the nature of the two surfaces involved and
the constraint enforcement method used (kinematic or penalty). The general
contact algorithm however uses balanced master-slave weighting whenever
possible, while for node-based surfaces pure master-slave weighting is used.
Appendix F provides an explanation of the algorithm ABAQUS uses for
tracking the motions of the contact surfaces in the case of a contact pair
algorithm. The general contact algorithm uses a more sophisticated tracking
approach that does not require user control and was thus not explained there.
4.3
Adaptive Meshing
40
Chapter 5
Underlying Computer
Simulation Model
Everybody who already did finite element computations knows that there
are a lot of decisions to be made during preprocessing. Just to name a few:
one has to choose the kind of elements and their shape, the size of the elements, the way of meshing, the integration method etc. Sometimes people
choose something without exactly knowing what effect it has on the entire
computation and they generally do not have the time to find out. For this
reason companies have some kind of standards for their internal crash computations. These are recommendations they give their engineers such that
they do not have to think about what to choose again and again. And even
in software packages like ABAQUS one can find recommendations which
are generally the default settings. It is for instance said in the ABAQUS
manual that generally it is enough to take 5 integration points for integration with the Simpsons rule along the thickness of a shell element. But
who made this statement and what is it based on? Can one conclude that
these recommendations are valid for all crash computations or for specific
computations only? Since it is not clear under what conditions one came
to these conclusions, the main objective of the following chapters is to find
out how sensitive crash computations are on changing single parameters. It
is desirable to finally be able to give recommendations about which settings
to choose to get accurate results under which specific conditions. These will
be compromises substantially influenced by the computation time, which
should be minimised as far as possible without loosing too much of accuracy.
Due to the complexity and time intensity, the analyses leading to those
recommendations cannot be carried out on the basis of a simulation of a
whole car. It will be taken a very simplified principle example of a single
41
5.1
Model Description
The model comprises three parts, the impactor, the base support and the
double chambered aluminium extrusion, as displayed in Fig. 5.1. The impactor is the moving part in the model, that is given an initial velocity
causing the crash of the extrusion. The task of the base support is, as the
5.2. GEOMETRY
43
name might let guess, to provide the clamped support of the double chambered extrusion on its one end. Impactor and base support do not play any
specific role in the whole system. Their tasks are easy and they are not
meant to deform. Thus, they can simply be modelled as rigid bodies. Everything concentrates on studying the deformation behaviour of the double
chambered extrusion.
5.2
Geometry
As one can see in Fig. 5.3, the end of the double chambered aluminium
extrusion of the impactor is inclined towards one side. It is assumed, that
this works as a crush initiator, that means as a deformation aid (cf. Section 3.3.2).
5.3
5.3.1
Element Model
Double Chambered Extrusion Beam
To reduce the computational cost, generally only linear elements are taken.
ABAQUS/explicit offers only linear elements with the exception of the
quadratic beam and modified tetrahedron and triangle elements, which are
all not reasonable in this case. Additionally it is only possible to take
reduced-integration elements, again with the exception of the modified tetrahedron and triangle elements.
That means that there is only one reasonable element left with three nodes,
and one with four. These are the S3R and S4R elements. Both of them are
used to mesh the double chambered extrusion uniformly. The S3R element
is a 3-node triangular general purpose shell element with finite membrane
strains and 6 degrees of freedom. There are only a few S3R elements used
in the front part to get the inclined shape. For all other elements, the S4R
element is used, which is a 4-node general purpose shell element, including
reduced integration, hourglass control, finite membrane strains and 6 degrees
of freedom per node. The planar dimensions of the elements are an order of
magnitude larger than the shell thickness.
The disadvantage of only having one possible kind of element is that later on
in the subsequent studies one cannot vary the element type and it was no way
found to switch off the hourglass control as well as the reduced integration
option. One is completely restricted to this specific element type in contrast
to other commercial software.
The shell thickness is chosen with 2.5 mm. Along the thickness of the shell
section one can choose two different numerical integration schemes: Either
Gauss point integration or integration with the Simpsons rule. And one can
vary the number of integration points. In this example, one chose integration
with the Simpsons rule and 5 integration points, which is the default setting
in ABAQUS.
5.3.2
For the impactor and the base support elements named SFM3D4R are used.
These are 4-node quadrilateral surface elements including reduced integration and 3 degrees of freedom.
Surface elements are defined just like membrane elements, i.e. as surfaces in
space. They may be used to define rigid bodies, do not have inherent stiffness, but may have mass per unit area. This fits perfectly for the impactor
and the base support as they do not have any special properties. The base
45
support just acts as a support and the impactor does not need anything
apart from a mass.
All in all, the model is defined via:
Number of elements
Number of nodes
Total number of variables in the model
(DOF plus any Lagrange multiplier variables)
5.4
11,120
11,187
52,077
Material Model
For the impactor and the base support, the material is just defined via its
density of = 7.82E09 t/mm3 .
The material used in this model for the double chambered extrusion is an
extruded aluminium alloy EN AW-7108 T6, which behaves in an elasticplastic manner and has the following basic material properties, which are
given in the input file (cf. Appendix A and [ABAb]):
Youngs modulus:
Density:
= 2.70E09 t/mm3
Poissons ratio:
= 0.33
To obtain reliable predictions from crashworthiness simulations, it is essential to model damage initiation and progressive failure due to various failure
mechanisms as well as modelling accurate plastic deformation behaviour.
The failure model as well as the stress-strain curves for different plastic
strain rates used in this input file are based on the article A comprehensive failure model for crashworthiness simulation of aluminium extrusions
of Hooputra et al. [HHW04]. It is shown there what experiments they carried out, the respective results and how they developed the fracture model.
All failure criteria have been developed in a way to include the influence of
non-linear strain paths.
5.4.1
Elastoplasticity
In the article of Hooputra et al. [HHW04], it is said that this aluminium alloy
EN AW-7108 T6 displays plastic orthotropy and they could use the Barlat
symmetric yield locus to fit the experimental data. But, as it seems to make
no significant difference when comparing the crashworthiness simulations to
experimental observations, they neglected the orthotropy and assumed the
(5.1)
where 1 , 2 and 3 are the principal stresses and y is the yield stress.
The failure criteria are based on macroscopic stresses and strains. One set of
parameters is valid for only one strain rate regime (quasi-static or dynamic).
Thus, in the input file at hand, stress-strain curves are provided for 10
different strain rates, as displayed in Fig. 5.4. Those curves are obtained
from experiments [HHW04].
5.4.2
Metal sheets and thin-walled extrusions made of aluminium alloys may undergo damage and failure due to either one or a combination of the following
failure mechanisms [HHW04]:
Ductile failure due to nucleation, growth, and coalescence of voids;
Shear failure due to fracture within shear bands;
and failure due to necking instabilities.
A description of these failure mechanisms can be found in the following
sub-chapters.
47
pl
,
,
(5.2)
D
where = 3m /eq is the stress triaxiality, m = (1 + 2 + 3 )/3 is the
hydrostatic stress, eq is the von Mises equivalent stress and pl is the equivalent plastic strain rate.
The stress triaxiality in the principal stess space is computed as:
=3
m
(1 + 2 + 3 )
=q
.
eq
1
2 + ( )2 + ( )2
(
)
1
2
2
3
1
3
2
(5.3)
pl
D =
0.
(5.5)
pl
D , pl )
In the input file a table is provided with the equivalent plastic strain at
fracture pl
D and the associated stress triaxiality at the respective equivalent
plastic strain rates pl . The curves one gets from this data are displayed in
Fig. 5.5.
5.4.2.2
.
(5.6)
pl
s
S
Here s = (eq + ks m )/max is the shear stress ratio, max is the maximum
shear stress, and ks is a material parameter. A typical value of ks for aluminum is ks = 0.3 [HHW04]. As for the ductile fracture, it will also here be
forgone to explain the exact equation for computing pl
D for the same reasons.
Similar to the criterion for ductile fracture, the criterion for damage initiation is met when the following condition is satisfied:
Z
S =
d
pl
= 1,
pl
pl
s
S
(5.7)
where S is a state variable that increases monotonically with plastic deformation proportional to the incremental change in equivalent plastic strain.
At each increment during the analysis the incremental increase in S is
computed as
Z
pl
(5.8)
S =
0.
pl
pl
s
S
Also here, the input file provides a table with the needed data. Here, the
equivalent plastic strain at fracture pl
S and the associated shear stress ratio,
s at the respective equivalent plastic strain rates pl . The curves one gets
from this data are displayed in Fig. 5.6.
49
5.4.2.3
r
pl
= major
=
4
1 + + 2
3
minor
major
(5.9)
(5.10)
To specify the MSFLD damage initiation criterion, one can thus either provide the limit equivalent plastic strain as a tabular function of , or the
limit major strain as a tabular function of minor strain, i.e. the FLD.
In this example, the major and minor plastic strains at the initiation of
localised necking are given in tabular form in the input file, cf. Appendix A
[ABAb]. These values are based on the experimental work of Hooputra
(2005) [ABAb] and assumed to be valid in the quasi-static as well as the
dynamic strain rates. The forming limited curve for this example is displayed
in Fig. 5.8.
Figure 5.7: Transformation of the forming limit curve from traditional FLD
representation (a) to MSFLD representation (b). Linear deformation paths
transform onto vertical paths. [ABAb]
pl
pl
pl , , fi
MSFLD ,
.
(5.11)
51
5.5
Contact Definition
5.6
5.7
Chapter 6
Sensitivity Examination of
the Model
6.1
Procedural Method
54
for this work is the group in which failure is not considered. In the second
analyses, Chapter 6.3, consideration moves to failure, with comparison made
between the results in an effort to examine whether the same predictions for
accurate modelling can be made.
Another significant difference is whether the computation is done in a quasistatic or a dynamic manner. It is reasonably obvious that differences in
model results will exist when modelling a process, that in reality unfolds at
high speed, in a dynamic or a quasistatic manner. However, since crash-
55
ing is actually a dynamic high speed process, and because the underlying
input file was made for a dynamic computation, it was decided to analyse
dynamic models first, and compare them to a quasistatic computation in
Chapter 6.2.5 and 6.3.5.
The next decision that had to be made was which kind of mesh to choose.
ABAQUS offers several options. Options include Quad-elements only, triangular elements only, or a mixture of both. In addition, there is the option
to choose a free, structured or sweep mesh. All the meshes taken here were
structured, but two different meshes were examined to compare any differences between mesh structures. In the file provided by ABAQUS, four
triangular elements were mixed among the Quad-elements in order to allow
right-angled elements in the main part of the beam, up until the point where
the inclination starts, cf. Fig. 6.3. As this requires some additional effort
and is not as straightforward as just clicking on a structured mesh with
Quad-elements only, it was decided that it might be interesting to find out
if this mesh variation makes a difference, and what the underlying reason
for meshing in this manner was. But before that was checked in Chapter 6.2.4, the first analyses were carried out with a structured mesh built of
only Quad-elements, cf. Fig. 6.4.
56
6.2
6.2.1
The aim of this section was to find a reference model in a greatly restricted
field of variables. All the computations that were analysed in this section
were dynamic computations where Simpsons rule was used to integrate
along the thickness, with 9 integration points, and all of the models were
built of only Quad-elements. The only difference between them was the
element size of the double-chambered extrusion.
How to Change the Element Size The easiest way to change the size
of the elements, and their arrangement, was to perform the process directly
through ABAQUS. For this, only the geometry of the model was needed,
which meant the shape of the model could be meshed later on automatically. The geometry was not defined in the input file, where the nodes of
the elements were given directly through their coordinates, and the individual elements together formed the surface. Changing that manually would
have been very expensive. Hence, it was cheaper to create the model from
scratch. To avoid creating the whole model from scratch and to keep all
the other definitions exactly as before, it was sufficient to create a model
for the double-chambered extrusion that could be meshed automatically in
ABAQUS. After meshing the extrusion in ABAQUS an input file could have
been written from which all the data of the nodes and elements could be
taken out. To get an input file for the new model with smaller elements,
one could now take the old input file and replace the old nodal and element
57
data with the new one. In this way, the impactor, the base support and
everything else (e.g. the material and the contact definitions) remained the
same.
Criteria for Finding the Reference Model In order to find the reference model, several criteria had to be chosen against which convergence
could be monitored. One could have considered the velocity of the car (which
here is simulated by the impactor) and its deceleration (which would be of
interesting in the design of airbags). Another possibility would have been
checking for any kind of energy, for example in the study of the plastic energy, or the fraction energy. For designing the engine mounts, however, two
main criteria are generally used: The length by which they are compressed
and the force needed for the deformation. As already known from Chapter 3, the integral over the curve for the relation of the deformation force to
the deformation length, indicates the energy which is absorbed during the
deformation process.
One way of measuring the length by which the beam is compressed was to
look at the displacement of the impactor reference node in the longitudinal
direction, and at the force required for the deformation (for this the reaction force at the end of the base support was a good indication). Both of
these criteria would give the same results for different meshes of adequate
construction.
A suitable criterion for the quality of the mesh is the total energy distribution
over time. Since energy cannot be lost, but only transformed, the total
energy should remain constant during the whole process. This criterion was
checked for all analyses, but will not be considered further. There were some
obscurities that could not initially be explained, cf. Appendix G.
In the subsequent sections, consideration is given to the differences between
graphs for the displacement distribution, and the reaction force over time
for the respective element sizes, as well as a discussion of the conclusions
that could be drawn from the graphs.
6.2.1.1
The first graph that was looked at showed the displacement over time curves
for the element sizes of 10 mm, 5 mm, 2.5 mm, 2.25 mm and 2 mm in Fig. 6.5.
As clearly evident in the graph, different results were obtained for different
element sizes not a great surprise. All the deformation curves demonstrated an early linear portion, at the beginning, i.e. at least for the first
fifth of the whole process, and then reduced in gradient before reaching a
58
Figure 6.5: Displacement over time curves for different mesh sizes.
maximum at a certain time, dependent on the element size. The fact that
the curves decreased afterwards was due to the measurement of the deformation on the impactor which, after a rebound, moved backwards.
It is clear to see that the result for a mesh of elements with an edge length
of 10 mm was completely different to other results. The total deformation
was, at about 150 mm, much lower than for finer meshes with up to 215 mm.
The finer the mesh, the larger the total deformation which confirms that
coarse meshes are a lot stiffer than fine meshes. For the 5 mm mesh the curve
got closer to the curves for finer meshes, but did not yet converge. If only
observing the deformation, convergence was reached for meshes with an edge
length similar to the 2.5 mm elements, and smaller. Since the curves were
very similar for 2.00, 2.25 and 2.50 mm meshes, one assumed that further
mesh refinement would have also given similar results. However, it could be
observed that the curves for the 2.50 mm and the 2.25 mm mesh were almost
exactly the same, whereas that of the finer 2.00 mm mesh differed slightly.
To see whether convergence was truly obtained, another finer mesh should
have been checked. Due to time restrictions however, it was not possible to
complete the process beyond 2 mm. For clarity, therefore, consider Table 6.1,
where the computation times are listed for the different mesh sizes.
59
Before being able to analyse the reaction force over time curves properly, it
was advisable to filter the curves since they oscillate heavily, especially in
the beginning, cf. Fig. 6.6, where the unfiltered curve for the mesh size of
2.50 mm is displayed. The oscillations make it difficult to see the underlying
shape of the curve.
Figure 6.6: Unfiltered reaction force over time curve (2.50 mm mesh).
The filter used was the ABAQUS saeGeneralFilter1 , a low pass filter which
filters data for a user-specified cutoff frequency. The cutoff frequency is the
frequency above which the filter attenuates at least half of the input signal.
It was not the aim to cut off any specific frequency, rather it was important
to get rid of the oscillations that disguised the main information of the curve.
Different frequencies were tested and the results compared to find out which
cutoff frequency gave a result closest to the original curve, without loosing
any important information. From Fig. 6.7 one can observe that the curves
for the frequencies of 100 Hz and 150 Hz have a different shape to the other
curves. They are smoother but some local maxima and minima are missing,
whereas the curves for 200 Hz, 250 Hz and 300 Hz have exactly the same
number of maxima and minima, locally and globally.
It was decided to filter the curves with a cutoff frequency of 300 Hz, since this
curve was very similar to the original curve without having the oscillations,
1
60
Figure 6.7: Filtered reaction force over time curves with different cutoff
frequencies (2.50 mm mesh).
as seen in Fig. 6.8. All subsequent reaction force over time curves were
filtered in the same manner.
Following this filtering, it was possible to analyse the reaction force over
time curves, for the different mesh sizes, for convergence. The curves for
the meshes with different element sizes are shown in Fig. 6.9. Immediately
obvious was the fact that the same dependence as for the deformation convergence was seen: The 10 mm curve seems to be completely different from
the other curves, the 5 mm curve gets closer to the curves for the finer meshes
and the 2.50 mm curve and the ones for even finer meshes are very similar.
That means that in a similar fashion to the displacement, convergence is
observed for the reaction force from 2.50 mm meshes onwards. As for the
displacement over time curves however, a finer mesh should be checked as
well, since the curves for 2.50 mm and 2.25 mm are almost exactly the same,
and the one for 2.00 mm is, again, only slightly different in the last quarter.
It is thus not a hundred percent certain if further mesh refinement gives the
same differences in results.
Another observation that can be made when comparing the reaction force
over time curves (Fig. 6.9) to the pictures of the deformed beam of the different meshes (Fig. 6.10 to 6.14) is that each minimum in the curve represents
one fold of the deformed beam. The reaction force over time curve has
four minima for the examples with an element size of 2.50 mm and smaller,
representing the four folds one can observe in Fig. 6.12 to 6.14. Also, the
61
Figure 6.8: Unfiltered and filtered reaction force over time curve with a
cutoff frequency of 300 Hz (2.50 mm mesh).
Figure 6.9: Filtered reaction force over time curves for different mesh sizes.
deformation of the beam with the 5 mm mesh and its respective curve is
not too different, whereas the deformed beam with the 10 mm mesh shows
only two less than perfectly formed folds, and is therefore too coarse. Since
the number of folds increases for finer meshes it also indicates, as in the
62
63
64
Figure 6.15: Building of the first fold with a mesh of 2.50 mm-elements.
65
Computation time
[hh:mm:ss]
00:15:36
02:06:09
1 day 15:06:40
5 days 00:32:03
14 days 10:34:16
Table 6.1: Computation time for different mesh sizes without failure.
The computation with a 10 mm mesh is very fast but, as discussed in the
previous section, not very useful in terms of accuracy. However, for the
model with an element size of 2.50 mm the computation time increased considerably. Running just this single small file required more than 39 hours.
Modelling and computing a whole car in practise, with that element size, is
thus not feasible with current CPUs.
The smallest element size currently used in the industry is 5 mm [HHW04].
The computation time for the model being considered, built with a 5 mm
mesh, is around 2 hours remarkably less than the time taken for the same
model built with a 2.50 mm mesh.
Following the observations made from Sections 6.2.1.1 and 6.2.1.2, that the
deformation and reaction force over time curves do not change remarkably
when having meshes finer than 2.50 mm, the model with an element size of
2.50 mm is considered accurate enough for this model, and is thus chosen as
the reference model. However, since a model with 2.50 mm elements cannot
be realised in industry for the simulation of whole cars, all the analyses
carried out later on will be built from with a mesh size of 5 mm, and then
compared to the reference model.
6.2.2
66
the thickness of the shell) with the Simpsons rule had. It was checked to
see if one finds similar results for a lower number of integration points, along
with the influence of the amount of integration points on the computation
time. The aim was to find out if it was possible to save computational cost
while observing the same results. Since the mesh with elements of size 5 mm
is the most practical one, the whole analysis was done for that mesh size.
When reducing the integration points, one always took an uneven number,
such that one point was always located on the midline of the shell elements.
Figure 6.16: Displacement over time curves for 3, 5 and 9 SIMP (5.00 mm
mesh).
The displacement and reaction force over time curves have already been
seen for 9 SIMP from the previous section, and are compared in Fig. 6.16
and Fig. 6.17 to the results with 5 SIMP, like in the underlying ABAQUSmodel and 3 SIMP. The meshes are exactly the same for the three examples.
Additionally, the graphs include the curves for the reference model to see
which number of integration points the curve is closest to the reference
solution.
It can be observed from Fig. 6.16 and Fig. 6.17 that the results for 5 and
9 SIMP are almost exactly the same, which leads to the conclusion that
it is not necessary to take more than five integration points for a dynamic
computation without considering failure. An integration with 7 SIMP was
hence not necessary to obtain results from. It was assumed that they lay
somewhere in the middle between 9 and 5 SIMP.
67
Figure 6.17: Reaction force over time curves for 3, 5 and 9 SIMP (5.00 mm
mesh).
Generally, one would assume that when taking more integration points,
the results would be more precise. The dependence here seems not to be
linear, and thus not easy to derive. Further reduction to 3 integration points
leads to different results and, astonishingly, the displacement curve for the
5 mm mesh with 3 SIMP is closest to the reference curve. The maximum
deformation with 3 integration points is larger than when having 5 or 9
integration points. Thus, one can conclude that a reduction to 3 SIMP
makes the mesh a little softer.
The reference force over time curve, however, is different for both of the
meshes (5 mm mesh with 3 SIMP and 2.50 mm mesh with 9 SIMP). There
is one similarity to observe: The reaction force reduces to zero at the same
time for both meshes. This is explained alongisde the similarity in the
deformation curves. It is due to the fact that the maximum displacement
after which the impactor rebounds and lets the reaction force reduce to zero,
is reached at the same time for both meshes. Since the reaction force over
time curve has a strong connection to the deformation of the structure, it is
advisable to have a look at the deformed structures. Therefore it is enough
to compare the figures for 3 and 9 SIMP (Fig. 6.18 and Fig. 6.19), since it
has already been discovered that there are almost no differences between 9
and 5 SIMP.
It can immediately be seen that the mesh with 3 SIMP is softer than expected, as already assumed from the deformation curves (where it was found
68
out that the total length of deformation is the same for 3 and 9 SIMP). But
it is even more surprising that during the deformation of the 3 SIMP model
more folds are built. This is in accordance with the reaction force over time
curve, already discussed in Section 6.2.1.2, where the maxima and minima
represent the number of folds. It can be observed here that the minima and
maxima for the 3 SIMP curve are denser than for 9 SIMP. That means more
folds are built within the same time.
One may conclude from this that reducing the integration points along the
thickness to 3 SIMP compensates for the effect of having a coarser mesh.
Coarser meshes stiffen the structure and fewer integration points weaken it
again. The results, however, for the 2.50 mm mesh with 9 SIMP and the
5.00 mm mesh with 3 SIMP are not the same. Thus, one cannot say that
a computation with a fine mesh of 2.50 mm and more integration points (5
or 9 SIMP) in thickness direction, can be replaced by a coarser mesh of
5.00 mm and only 3 SIMP integration points.
69
To prove that the same effects are given for another mesh size and thus to be
able to make a more representative statement, the same analysis was done
for a mesh with elements of size 2.50 mm.
Figure 6.20: Displacement over time curves for 3, 5 and 9 SIMP (2.50 mm
mesh).
By looking at the displacement and reaction force over time curves in Figures 6.20 and 6.21, it can be observed that for the 2.50 mm mesh the curves
for 5 and 9 integration points are almost the same, and the one for 3 integration points is different. However, there is a significant difference compared
to the 5 mm mesh. The displacement curve does not have a maximum after
which the curve decreases again. That means there is no rebound. The
same can be observed from the reaction force time relation, since this
curve does not reduce to zero at the end. This means that for a finer mesh,
the structure is softer when having only 3 SIMP, and here even softer.
Like for the 5 mm mesh, here the minima and maxima are denser than for
9 SIMP and more folds are built within the same time, as can easily be seen
in Fig. 6.22 (3 SIMP) and when comparing it to Fig. 6.23 (9 SIMP).
To find out which combination of mesh size and integration along the thickness is best suited to represent reality, one should have some results from
real experiments, where the deformation and the reaction forces during the
deformation could be measured and where the number of folds that should
be represented by the simulation could simply be counted.
70
Figure 6.21: Reaction force over time curves for 3, 5 and 9 SIMP (2.50 mm
mesh).
6.2.2.2
Computation Time
By simply observing the computation time for the same mesh (5.00 mm elements) with the three different numbers of integration points in Table 6.2,
one can see that the difference between a large and a small number of integration points is not very distinctive. All the computations were done on
the same computer in this case, and can thus be compared side by side.
number of
int. points
9 SIMP
5 SIMP
3 SIMP
computation
time [hh:mm:ss]
02:06:09
01:53:27
01:37:54
71
6.2.3
72
integration points get closer to the surface the more integration points are
used, but they will never be exactly on the surface. This is important to
know, for example, when having a composite material built of different layers
where the output is required between the layers. To get that, Simpsons rule
has to be used.
In this section, the two numerical methods were compared. Therefore, 5 integration points were taken for either method2 . The results are displayed in
Fig. 6.24 and Fig. 6.25 and compared to the reference solution.
Figure 6.24: Displacement over time curves GP vs. SIMP (5.00 mm-mesh).
From Figures 6.24 and 6.25 it can be observed, that the solutions for the
Simpsons and Gauss integration do not differ greatly. The deformation over
time curve for the model integrated with Gauss quadrature is only slightly
closer to the reference solution. The significant difference is observed when
comparing the computation times, cf. Table 6.3.
As already mentioned, the 5 GP integration is slightly more accurate than
the 5 SIMP integration, in spite of having a shorter computation time. Since
using Gauss quadrature makes it possible to save computational time and
one does not need the solutions on the surfaces for these kinds of problems,
it is advisable to use Gauss quadrature for integration along the thickness
of the shell.
2
To compare models accurately, one should have taken more integration points for the
Simpsons integrated model to get the same accuracy than with Gauss quadrature, but
since there is hardly any difference between 5 and 9 SIMP (cf. Fig. 6.16 and Fig. 6.17) it
does not matter.
73
Figure 6.25: Reaction force over time curves GP vs. SIMP (5.00 mmmesh).
This, however, is only true when not considering failure. There might be
a difference when considering failure, cf. Chapter 6.3.3 since the maximum
and minimum stresses are on the surface when bending results. Thus, the
material might fail earlier when integrating with the Simpsons rule, since
the failure criteria might be reached earlier.
number of
int. points
5 SIMP
5 GP
computation
time [hh:mm:ss]
01:53:27
01:44:15
6.2.4
74
Obviously, the bigger part of all elements is due to the mixture of Quadand Tri-elements. But does it make a difference?
Figure 6.26: Displacement over time curves Quad vs. mixed (5.00 mm
mesh).
Figure 6.27: Reaction force over time curves - Quad vs. mixed (5.00 mm
mesh).
As in the previous section, the difference in the two kinds of meshes was
75
checked with an element size of 5.00 mm, and compared to the reference
model in Fig. 6.26 and Fig. 6.27. From these figures, only negligible differences could be observed, and it was thus concluded that it does not make a
difference which mesh is chosen.
There was also hardly any difference in the computation time. It took
02:06:09 for the mesh with only Quad-elements and only slightly longer for
the mesh with mixed elements, at 02:13:38.
6.2.5
The comparison of a dynamic and a quasistatic computation was also performed for a model with only Quad-elements of size 5.00 mm and 9 SIMP
along the thickness. The difference between a dynamic and a quasistatic
computation here lays only in the material definition. In the case of a dynamic computation, stress-strain-curves (Fig. 5.4) were given for different
strain rates (cf. Appendix A and B), whereas to compute quasistatically,
this rate dependence was simply removed. All data for stress-strain-curves
of plastic strain rates other than 0 were deleted from the input file.
76
folds was exactly the same in either case. A difference was only seen when
examining the curves very closly, or when comparing the deformation over
time curves. These curves are displayed in Fig. 6.29 for both computations.
Figure 6.29: Deformation over time curves from a dynamic and a quasistatic
computation (no failure).
77
for higher strain rates cross the others and decrease. That means that for
higher strains the material behaviour is vice versa: Softer for high plastic
strain rates, and stiffer for small plastic strain rates. It was therefore necessary to show the rank of strains in which one is located in both computations.
Figure 6.30: Stress-strain-curves for different plastic strain rates in the range
of small strains.
Unfortunately, the border where everything is ordered nicely (0 0.1)
is exceeded and the logarithmic strains go up to 1.62 in the dynamic computation, and up to 1.37 in the quasistatic computation. It is not even the
case that the main strains are within the rank of small strains; they are
distributed all over.
Therefore, it was decided to check which plastic strain rates the curves
are still above the one for a plastic strain rate of 0. A second dynamic
computation of the same file including only the curves for the plastic strain
rates, which lay above the one for a plastic strain rate of 0, should then
give an even stiffer response than the original dynamic computation. This,
therefore, would explain the connection between the stress-strain-curves and
the difference in the deformation length seen between a dynamic and a
quasistatic computation.
As can be seen in Fig. 6.31, for strains up to 1.60, only the stress-straincurves up till a plastic strain rate of 1 are above the stress-strain-curve for
a plastic strain rate of 0.
The second dynamic computation, thus, excluded all stress-strain-curves for
78
79
plastic strain rates higher than 1. Fig. 6.32 shows the same deformation over
time curves as in Fig. 6.29 plus the one from the second dynamic computation. Against all expectations, the second dynamic computation with fewer
stress-strain-curves did not give a stiffer response of the structure. The curve
(in yellow) lies in the middle of the ones from the previous computations.
An explanation for the difference in the deformation length from a dynamic
and a quasistatic computation, and the connection to the stress-strain-curves
could not be found. Further investigation is needed.
6.3
Including Failure
6.3.1
Similarly to Section 6.2, this process was started by trying to find a reference
model, to allow comparison with all further models. Again, that meant trying to find a mesh as coarse as possible, but one which when further refined
produced no significant changes. For simplicity, a mesh built of only Quadelements was used, and integrated along the thickness with Simpsons rule
using 9 integration points. The computations were dynamic and compared
to quasistatic computations later in Section 6.3.5. The comparison criteria
for finding the reference model were the displacement-time and the reaction
force-time relations, similar to Section 6.2.1.
The coarsest mesh that was started with, had an element size of 10.00 mm.
This was then refined to meshes of element size 5.00 mm, 2.50 mm, 2.40 mm,
2.30 mm, 2.25 mm and 2.00 mm, cf. Section 6.3.1.2. The reason for the
smaller steps between 2.50 mm and 2.00 mm was that all the computations
with fine meshes were aborted, with no obvious explanation. There were no
failure messages that could have given helpful information. Thus, smaller
steps were attempted, but with little success. Finally, it was discovered that
when the computations were run on two CPUs, they could be completed
successfully. The first investigation here, therefore, was to find out what the
difference was between a computation that was run on 1 CPU and one on
2 CPUs. This is discussed in the following subsection.
80
6.3.1.1
When computing on more than 1 CPU (parallel computing), ABAQUS offers two possibilities for performing the task, either loop level parallelisation
or domain level parallelisation. The loop level method parallelises low-level
loops that are responsible for most of the computational cost. [...] However,
the speedup-factor may be significantly less than with domain level parallelisation, [...] since not all features utilise parallel loops. [...] The domain
level method breaks the model up into topological domains and assigns each
domain to a processor. [...] The analysis is then carried out independently
in each domain. However, information has to be passed between the domains
in each increment because they share common boundaries. [rhp] The results of both methods do not depend on the number of processors used, but
they are dependent on the number of parallel domains. To obtain consistent analysis results from run to run, the number of domains used in the
domain decomposition should be constant4 [rhp]. However, there are some
limitations on domain decomposition, since even if the algorithm succeeds
in creating the requested number of domains, the load may be balanced unevenly. [rhp]
Since the computations of files with meshes finer than 2.40 mm were all
aborted after a certain time when computed on 1 CPU, and after it was
found that it was possible to run the whole files successfully on 2 CPUs, it
was assumed that the number of CPUs must have a significant influence.
But did the number of CPUs have an influence on the results?
The computations of fine meshes, however, were only successful with the
domain level parallelisation. It had already been observed that a mesh of
2.50 mm had been aborted under the loop level parallelisation, which could
have been computed on 1 CPU. Thus, the loop level parallelisation was
not considered further, as it offered no advantage when considering mesh
refinement.
For the comparison between a computation which was run on 1 CPU and a
computation on 2 CPUs (with domain level parallelisation), the model with
elements of size 2.50 mm was considered. It should be mentioned here that
the input file that was taken for both computations was exactly the same.
As evident from Figures 6.33 and 6.34, the deformed structures look completely different for both computations. In either case one can see a rolling
up of the separated sheets, but in a much more regular fashion when computed on 2 CPUs. Also, it can be observed that the total compression of the
beam that has been computed on 2 CPUs is much higher than the one which
4
It was checked whether (for the model with 2.50 mm elements, with the domain level
parallelisation, and with two domains on two CPUs) the computation gave the same results
when computed twice under the same conditions. It was concluded that it did.
81
Figure 6.34: Result of a computation on 2 CPUs with domain level parallelisation (2.50 mm mesh).
was computed on 1 CPU. The difference is seen right away when looking
at the pictures of the deformed structures. For a more developed proof,
however, Fig. 6.35 shows the difference in the diplacement-time relations.
The initial length of the beam was about 396.5 mm, whereas the final length
is 280.9 mm for the computation on 1 CPU, and 217.3 mm for the one on
2 CPUs. This shows a difference of 63.6 mm, which equates to 16 % of the
total length a significant amout.
From the reaction force over time curve in Fig. 6.36, it can be seen that
82
Figure 6.35: Displacement over time curves for 1 and 2 CPU (2.50 mm
mesh).
Figure 6.36: Reaction force over time curves for 1 and 2 CPU (2.50 mm
mesh).
the force needed to deform the structure is much less for the file that was
computed on 2 CPUs than for the one that was computed on 1 CPU. Hence,
the energy the beam absorbs during deformation is also different, and worse
83
When failure was not included, convergence was found for models with
meshes finer than 2.50 mm, and hence the reference model discovered. The
assumption that the same is true when failure is considered could not be
84
n of
CPUs
1
parallelisation
method
none
domain
loop
integration
method
9 SIMP
mesh
size
[mm]
10.00
5.00
2.50
2.40
2.30
time
increments
auto
auto
auto
auto
auto
2.30
small
7 GP
2.30
2.00
auto
auto
9 SIMP
2.50
2.30
2.25
2.00
1.50
auto
auto
auto
auto
auto
1.50
small
2.50
auto
9 SIMP
comments and
computation time
[hh:mm:ss]
o.k. (00:13:34)
o.k. (01:34:05)
o.k. (12:01:58)
o.k. (18:43:16)
aborted after
1 day 01:53:30
(50 % completed)
aborted after
1 day 08:08:20
(51 % completed)
o.k. (1 day 09:04:18)
aborted after 13:01:02
(20 % completed)
o.k. (18:59:38)
o.k. (20:12:53)
o.k. (21:10:01)
o.k. (1 day 15:45:54)
aborted after 18:13:59
(13 % completed)
cancelled, took too long
after 2 days 19:14:52 just
0.15 % were completed
aborted (17:56:31)
(57 % completed)
Table 6.4: Different approaches for refining the mesh (failure included).
validated, since a computation on 1 CPU was not possible for meshes finer
than 2.40 mm, and the results of computations that were run on 1 and
2 CPUs was not comparable. Apart from that, computations on 2 CPUs
with the domain level parallelisation did not give reasonable results, and
the loop level parallelisation did not even begin to work for a mesh of size
2.50 mm. Another possibility that was checked, in order to discover possible
solutions, was a reduction in the time increments, but again these computations were aborted sooner or later. When comparing integration along the
shell thickness with Simpsons rule and Gauss quadrature in Chapter 6.3.3,
it was found that it was possible with Gauss quadrature to compute finer
meshes, but only in a limited fashion. This can be seen in Table 6.4, which
is an overview of all the computations that were run to find a way of solving
models with fine meshes.
85
Figure 6.38: Displacement over time curves for different mesh sizes.
Figure 6.39: Reaction force over time curves for different mesh sizes.
Since all approaches to get reasonable results for fine meshes failed, it was
impossible to find a reference model. It can be observed from Fig. 6.38 and
Fig. 6.39, that convergence could not be reached for models computed on
1 CPU, leading to a mesh size of 2.40 mm. The reaction force over time
curves did not show any similarities and the total compression of the beam
86
increases as the mesh gets finer, without seeming to converge at all. The
finer the mesh, the softer the structure as seen in the case where failure
was not considered. Similar to Chapter 6.2.1 rebound can be observed for
the coarsest meshes from the displacement and reaction force over time
curves. The models with finer meshes get compressed more and more as the
structure is disrupted, even more so than in the case where failure was not
considered.
Although all computations were not run on the same computer, and thus
are not 100 % comparable, one can clearly see (when comparing Tables 6.1
and 6.4) that the computation time is much less when considering failure
than without. This is very surprising and needs further investigation. An
initial thought is that, when considering failure, the failed elements are
excluded from the computation and are not further considered, thus making
the computation faster. This, however, would mean that the curve for the
real time against the CPU-time would be exponential rather than linear,
which is not the case.
6.3.2
Since the smallest element size currently used in the industry is 5.00 mm
(cf. Chapter 6.2), the influence of the number of integration points along
the thickness of the shell, as well as the influence of all further parameters,
was checked for the model with elements of size 5.00 mm. But as there is no
reference model, this comparison is missing.
Figure 6.40 shows that when reducing the number of integration points,
the effect is exactly the same as that seen earlier in Chapter 6.2.2 when
considering the case without failure: The maximum deformation with 3
integration points is larger than when having 5 or 9 integration points. Thus,
one can conclude that for a computation including failure criteria, reduction
to 3 SIMP makes the mesh softer.
6.3.3
As already done in the previous section and in Section 6.2.3 two models with
5 mm elements were compared, one with 5 SIMP along the thickness and
the other one with 5 GP. To make this short, it can simply be stated that
the results are the same as in Section 6.2.3: There is no significant difference
between the two computations when failure is included. The main difference
is the computation time which was 01:14:06 for the Gauss integration - less
than the 01:22:52 for the integration with Simpsons rule.
87
Figure 6.40: Displacement over time curves for 3, 5 and 9 SIMP (5.00 mm
mesh).
6.3.4
As already found in Section 6.2.4, the kind of mesh one takes does not make
a difference, even when failure is considered. Proof of this has been forgone.
6.3.5
Similar to what was seen in Section 6.2.5, this section offers a comparison
of a quasistatic and a dynamic computation for the same mesh, but now
with failure included. In addition to the removal of the rate dependence of
the stress-strain-curves, in order to get a quasistatic computation, here the
rate dependency in the failure models (for the quasistatic case) had to be
removed. Since the instability criterion (MSFLD) is not dependent on the
strain rate (cf. Appendix E), this only had to be done for the shear and
ductile damage criteria. As can be observed from Appendix C and D, the
failure curves are only available for plastic strain rates of 0.0001, 0.001, 250
and 1000, i.e. not for a plastic strain rate of 0, as needed for the quasistatic
computation. However, it can also be seen that the curves are the same for
plastic strain rates of 0.0001 and 0.001, and respectively for 250 and 1000.
Hence, it was assumed that the curve for a plastic strain rate of 0 is the
88
As for results for the computation, one gets deformed structures as displayed
in Fig. 6.41. It can be see that in the dynamic case almost no folding occurs
and that the structure is rather disrupted, whereas the result of a quasistatic
89
computation shows very regular foldings, similar to the computations considered without failure. When comparing the deformed structures from the
computer simulations to pictures of quasistatic and dynamic compression
tests in Fig. 6.42, it is immediately seen that the simulation seems to give
realistic results, although they are not quite identical.
Figure 6.43: Deformation over time curve from a dynamic and a quasistatic
computation (failure included).
Without considering failure (cf. Section 6.2.5), the main difference between
a quasistatic and a dynamic computation was the total deformation, while
the deformed structures looked almost the same. When including failure,
one already knows that the deformed structures look anything but similar.
A glance at the displacement over time curves for computations including
failure, in Fig. 6.43, however, shows that the difference in the compression
length is almost negligible and thus the difference must be sought elsewhere.
The fact that failure is considered seems to allow the structure to deform
more. This makes sense when considering how disrupted the structure is
after crushing (in the dynamic case). Hence, one can say that unlike Section 6.2.5, the effect that a beam from a quasistatic computation deforms
more is compensated by including failure criteria. How the deformation is
influenced by the different failure criteria in the quasistatic and the dynamic
case, and especially which criterion predominates, will be explained in the
following section.
90
6.3.5.1
91
Figure 6.44: Fracture with respect to the individual failure criteria for quasistatic and dynamic compression.
The sensitivity analyses for the model regarding changes to single parameters are terminated here. More could have been explored and clarified,
but time prevented this endeavour. In the following chapter a summary of
92
Figure 6.45: Dynamic (upper) and quasistatic (lower) failure diagram for
extrusion EN AW-7108 T6. [HHW04]
all results seen in Chapter 6 is made, along with resulting open questions
and conclusions that could be drawn. Finally, a discussion of further work
is given - relating to analysis that could yet be performed and topics of
subsequent work.
Chapter 7
94
Recommendations for Crash Simulations Including Failure Modelling Assumptions that were made for computations without considering
failure are only conditionally transferable to computations where failure is
included. The statements that can be made about the mesh type, the number of integration points along the shell thickness and the integration rule
are the same as the ones previously made for computations without failure.
Regarding all other points, there are considerable differences.
Mesh Size: Convergence for meshes with element sizes 2.40 mm
could not be found. That means that finer meshes have to be used.
However, it cannot be stipulated how fine, since all computations with
elements smaller than 2.40 mm were aborted.
Quasistatic or Dynamic Computation: The results from quasistatic and dynamic computations differ considerably. The deformed
structure looks completely different under each computation. The one
from the dynamic computation however, represents reality better and
is hence recommended. The rate dependence of the material and of
the failure models seems to balance the difference in the length of
compression, cf. Section 6.3.5.
Failure Model: The combination of the three failure models (ductile,
shear and instability) should always be used. The results when ignoring one do not make sense e.g. when not considering the shear failure
criterion, the walls do not separate at the edges and at the T-joints.
95
For the problem at hand, there are still a great many investigations necessary and some aspects of the work may well remain difficult to be fully
answered. During my work I often encountered contradictions or inexplicable details. The crash simulations performed in the frame of this thesis
were strongly dependent on the input settings, and in some cases even on
which computer the model was computed. Hence, it is essential to establish
a validation procedure, perhaps including a quality assessment of the FEmodel. In addition, it is important to know the software well and to study
the effects individual changes have on the results, for example as was done
in Chapter 6 (by computing one file several times with slightly different settings and comparing the results). As this describes the main problems in the
computations undertaken in this work, it is suggested that checking of the
mesh size (is it fine enough), checking if the parallelisation works correctly,
and checking if the failure model gives reasonable results (does the deformed
structure look reasonable), is always conducted.
Failure is generally not considered in the industry, perhaps due to the obvious reasons discovered during this study. However, computations that
include and exclude failure give quite different results. Hence, I find it critical not to consider failure. Rather, it should be the focus of studies to
discover exactly where the problems with simulations that include failure
lie, such that in the future they can be avoided. The most important open
questions to consider, that still require clarification through further studies,
are listed in the following:
First of all, it would be interesting to find out why it makes such a big
difference for the total computation time whether failure is considered
or not.
It should be clarified why the energy decreases for fine meshes (especially when failure is not considered), cf. Appendix G.
In order to find convergence when reducing the mesh size (failure included), computations for small elements that successfully run are required.
And finally, the reason for the difference in computations run on 1 and
2 CPUs should be explored, since most crash computations are run on
more than 1 CPU due to the large size of the models.
Ideas for Further Analyses Initially, it was the aim of this study to
carry out further analyses in addition to those discussed here, see Chapter 6.
However, due to time restrictions and the problems that arose, this was
96
not possible. However, to complete this work, a list of the original ideas
this author had for further analyses is given below, and could be use as
motivation for future projects.
The elements used in this work (S4R and S3R) automatically consider
hourglass control. It is unknown if there is a way of finding a process
in ABAQUS to switch off hourglass control. It could be interesting to
see what difference this makes.
It was discussed in Chapter 4.3 that adaptive meshing generally is
not used due to parallelisation problems. Nevertheless, it would have
been interesting to find out on the small model discussed in this study
(which could be computed on 1 CPU) what difference this would have
made to the results.
It would also have been interesting to analyse the sensitivity to changes in the material definition. The friction coefficient, for instance,
generally is an empirical value, which usually no one checks (in order
to assess if that value is suited to the present material). Does a slight
change in value have an influence on the result, or is the effect small
and makes no difference?
Further analysis could be made with changes to the geometry. It was,
for example, simply assumed that the inclination at the front of the
beam was meant to be a crush initiator. What would happen without
the inclination? And what effect would other crush initiators have,
like the ones in Fig. 3.4. It could also be checked what would happen
to the deformation when the beam had some holes etc.
Appendix A
Input File
The original input file from the ABAQUS Example Problems can be found
here including explanations of the keywords.
A.1
Preliminary Definitions
*HEADING just defines the title of the analysis. Doesnt have any meaning
and thus could be anything.
*HEADING
Dynamic Axial Crushing of a Double Chambered Aluminum Extrusion
*PREPRINT is used to select the printout that will be obtained from the
analysis input file processor.
*PREPRINT, ECHO=YES, HISTORY=YES, CONTACT=YES,
MODEL=YES
*RESTART, WRITE
The parameter *RESTART, WRITE is included to specify that restart data
is to be written during the analysis.
A.2
The nodes are defined here via a certain node number and the respective x-,
y- and z-coordinates in each input line. They are all collected in a node set
named All Nodes.
*Node, Nset=All Nodes
78565, -1.000000e+00, -2.500000e+02, 2.500000e+02
97
98
A.3
Element Data
The elements are defined like the nodes. Each input line contains a number
labelling the element followed by the 4 node numbers that build the element
or 3 node numbers for a triangular element. Three element sets are created,
one for the base support, the impactor and the double chambered extrusion
respectively.
Base support and impactor are built of SFM3D4R-elements, which are 4node quadrilateral elements with reduced integration and 3 degrees of freedom.
SFM3D4R stands for: SF: surface, M: membrane-like, 3D: three-dimensional,
4: number of nodes, R: reduced integration.
*Element, Type=SFM3D4R, Elset=P2 BaseSupport
199105, 153203, 150892, 151252, 153521
199106, 153203, 153540, 151839, 150892
...
*Surface section, Elset=P2 BaseSupport, DENSITY=7.8200e-09
*Element, Type=SFM3D4R, Elset=P3 Impactor
197949, 79696, 78700, 78565, 78566
197950, 79398, 78701, 78650, 78651
...
*Surface section, Elset=P3 Impactor, DENSITY=7.8200e-09
Additional mass can be brought into the model by using surface elements.
The mass per unit area carried by the surface element is therefore defined.
Input File Usage: *SURFACE SECTION, ELSET=name, DENSITY=number
The double chambered extrusion is built of S4R and S3R elements, which
are:
S4R: 4-node doubly-curved general-purpose shell elements with reduced integration, hourglass controll and finite membrane strains
S3R: 3-node triangular general-purpose shell elements with finite membrane
strains
*Element, Type=S3R, Elset=P1 DoubleChamberExtrusion
191791, 226378, 226437, 226444
99
A.4
Material Definition
100
The failure initiation criteria is defined separately for ductile, shear and
instability failure. The data for these criteria is prepared in tabular form in
Appendix C to E.
CRITERION=DUCTILE specifies a damage initiation criterion based on the
ductile failure strain.
*DAMAGE INITIATION, CRITERION=DUCTILE
3.3238E+01, -3.3333E+00, 1.00E-04
3.3238E+01, -3.3333E-01, 1.00E-04
2.3381E+01, -2.6667E-01, 1.00E-04
...
*DAMAGE EVOLUTION, TYPE=DISPLACEMENT
0.1
101
102
A.5
Impactor
*Elset, Elset=RBody Impactor
P3 Impactor,
All elements of the impactor are assigned from the former element set
P3 Impactor to a new element set called RBody Impactor.
*Nset, Nset=RBody Impactor RefNode
233109,
A new node set called RBody Impactor RefNode is created including only one
node, the reference node which represents the impactor.
*Rigid body, Ref node=RBody Impactor RefNode,
Position=CENTER OF MASS,
Elset=RBody Impactor
In these lines, the impactor is defined as a rigid body. All the elements are
represented in the reference node, where all the rigid element properties are
defined.
*ELEMENT, TYPE=MASS, ELSET=MASSE 500 kg
1, 233109
*MASS, ELSET=MASSE 500 kg
0.500,
The whole mass of 500 kg is assigned to element 1 including node 233109,
which was defined earlier as the reference node for the impactor.
*ELEMENT, TYPE=ROTARYI,
ELSET=MASSE 500 kg Rot Inertia
2, 233109
*ROTARY INERTIA, ELSET=MASSE 500 kg Rot Inertia
0.2088e+005, 0.1431e+005, 0.1430e+005
The rotary inertia of the impactor about the local 1-, 2- and 3-axis, I11 ,
I22 , I33 respectively is defined here. As the rotary inertia is defined per
element set, a new set named MASSE 500 kg Rot Inertia is created, including
element 2 with node 233109, the reference node of the impactor, where the
whole mass is located.
Base Support
The same as above for the impactor is done here for the base support,
without the definition of a mass and the rotary inertias. The number of the
reference node of the base support is 233110.
103
A.6
For contact of the impactor and the base support with the double chambered
extrusion, one has to assign a master and a slave surface, which is done in
the following few lines.
*SURFACE,TYPE=ELEMENT,NAME=Contact 1 Master
P3 Impactor,
*SURFACE,TYPE=NODE,NAME=Contact Slave
P1 DoubleChamberExtrusion,
*SURFACE,TYPE=ELEMENT,NAME=Contact 2 Master
P2 BaseSupport,
As one can see from the given names, the surfaces of the impactor and the
base support are the master surfaces and the double chambered extrusion
is the slave surface.
Contact of the double chambered extrusion has to represent self contact
due to the foldings it is going to form during deformation. As with the
general contact algorithm all surfaces are automatically double-sided, see
Section 4.2, it will do just to define a simple surface.
*SURFACE, Type=ELEMENT, Name=Contact 3 Self
P1 DoubleChamberExtrusion,
*Surface Interaction, Name=Contact 3 Prop
*Friction
1.500000e-01,
As a surface interaction property, friction is defined with a friction coefficient
of 0.15 and will be referred to with the label Contact 3 Prop. This will later
be used where the contact interaction is defined in A.9.
104
A.7
The boundary conditions used in the present work are defined in ABAQUS
as follows:
1, 2 and 3 are the displacements in x-, y- and z-direction and
4, 5 and 6 are the rotations in radians about the x-, y- and z-axis
respectively.
Thus, the displacements for the fixed base support are defined as
*Boundary
RBody BaseSupport RefNode, 1, 6
where first the node set label is mentioned, followed by the first and last
constrained degree of freedom.
The moving part in the whole model is the impactor, which has to move in
one direction, which is here the x- or 1-direction. The boundary conditions
for the impactor are thus:
*Boundary
RBody Impactor RefNode, 2, 6
A.8
History Definition
All the following, until the end of the output definition in A.11, is included
in one step. The command for beginning a step is *Step, NLGEOM=YES,
which is used in the same input line, indicates that it should be accounted
for geometric nonlinearity during the step.
*Step, NLGEOM=YES
*Dynamic, Explicit, Element by element
,0.0501
A dynamic stress/displacement analysis is indicated, where explicit time
integration is used. To estimate the time increments one uses the elementby-element time estimator described in Section 4.1.2.3 without switching on
to the adaptive global estimation algorithm. The time period for the step
was chosen as 0.0501 s, which is the total time for the analysis.
*DIAGNOSTICS, CONTACT INITIAL OVERCLOSURE=DETAIL
*NSET, NSET=Output Refnodes
RBody Impactor RefNode, RBody BaseSupport RefNode
105
A.9
106
*Contact defines the property of general contact with the surface pairings
being included in the general contact domain. When the first and second
surfaces, for which the contact is defined, are the same, like it is the case
here, self-contact is defined as
*Contact property assignment
Contact 3 Self, Contact 3 Self, Contact 3 Prop
Here, contact properties are assigned. In the second line, the first and second
surfaces for which the contact is defined, are included, as well as the name
of the surface interaction property defined in A.6, the friction coefficient
= 0.15.
*Surface property assignment, Property=THICKNESS
Contact 3 Self, 2.500000e+00, 1.
A thickness of 2.5 mm is assigned to the surface of the DoubleChamberExtrusion. The number 1 after the thickness indicates a constant scaling factor,
which is also by default 1 and could thus be omitted.
*Contact controls assignment, NODAL EROSION=YES
There are several contact controls assignable to the contact definition. By
setting NODAL EROSION=YES one makes sure that a node of an elementbased surface from the general contact domain will be deleted once all contact faces and edges to which it is attached have eroded.
A.10
The moving part of the model is the impactor. It is defined, that the impactor moves with a velocity of 10, 000.0 mm/s in global x-direction towards
the double chambered extrusion, cf. boundary conditions in A.7.
*INITIAL CONDITIONS, TYPE=VELOCITY
RBody Impactor RefNode, 1, 10000.0
A.11
Output Request
107
In the following, as a nodal output one defined to get the displacement components, which is indicated by U. As an element output all stress components
(S), all infinitessimal strain components for geometrically linear analysis (E),
the STATUS of the element and all active components of the damage initiation criteria (DMICRT) were chosen. The STATUS of the element is 1.0 if
the element is active and 0.0 if not.
*Node output
U,
*Element output
S, E, STATUS, DMICRT
Subsequently, the history-type output is defined. The parameters are chosen
for which an output for each time interval of 0.0001 s is required.
*OUTPUT, HISTORY, TIME INTERVAL=1.0e-04
As before, there are nodal outputs and additionally several energy outputs.
The nodal output here is the displacement in x- or 1-direction (U1) of the
impactor and the reaction force (RF1) at the base support in the same
direction.
*NODE OUTPUT,NSET=RBody Impactor RefNode
U1,
*NODE OUTPUT,NSET=RBody BaseSupport RefNode
RF1,
*ENERGY OUTPUT
ALLKE, ALLIE, ALLSE, ALLPD, ALLVD, ALLAE, ALLWK, ETOTAL
*End step
With *End step the whole definition of the step that started with the history
definition in A.8 ends. The meaning of the several shortcuts for the energy
output are listed below.
ALLAE:
ALLCD:
ALLDC:
ALLDMD:
ALLFC:
ALLFD:
ALLHF:
ALLIE:
ALLIHE:
ALLKE:
108
ALLPD:
ALLSE:
ALLVD:
ALLWK:
ETOTAL:
Appendix B
Stress-Strain-Curves
Yield stress
[N/mm2 ]
3.1100E+02
3.1664E+02
3.2452E+02
3.3520E+02
3.4896E+02
3.6552E+02
3.8370E+02
4.0165E+02
4.1821E+02
4.3474E+02
4.5535E+02
4.8516E+02
5.2975E+02
5.9664E+02
6.9698E+02
6.9698E+02
Yield stress
[N/mm2 ]
3.1100E+02
3.1664E+02
3.2452E+02
3.3520E+02
3.4896E+02
3.6552E+02
3.8370E+02
4.0165E+02
4.1821E+02
4.3474E+02
4.5535E+02
4.8516E+02
5.2975E+02
5.9664E+02
6.9698E+02
6.9698E+02
Plastic Strain
pl = 0.00E+00
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
109
Plastic Strain
pl = 2.00E03
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
110
APPENDIX B. STRESS-STRAIN-CURVES
Yield stress
[N/mm2 ]
3.1254E+02
3.1817E+02
3.2606E+02
3.3673E+02
3.5050E+02
3.6706E+02
3.8523E+02
4.0318E+02
4.1975E+02
4.3627E+02
4.5688E+02
4.8668E+02
5.3127E+02
5.9814E+02
6.9843E+02
6.9843E+02
Plastic Strain
pl = 1.00E02
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
Yield stress
[N/mm2 ]
3.1577E+02
3.2140E+02
3.2929E+02
3.3996E+02
3.5372E+02
3.7028E+02
3.8845E+02
4.0639E+02
4.2294E+02
4.3945E+02
4.6001E+02
4.8975E+02
5.3421E+02
6.0085E+02
7.0066E+02
7.0066E+02
Plastic Strain
pl = 1.00E01
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
Yield stress
[N/mm2 ]
3.2083E+02
3.2646E+02
3.3434E+02
3.4500E+02
3.5874E+02
3.7526E+02
3.9338E+02
4.1123E+02
4.2762E+02
4.4388E+02
4.6404E+02
4.9309E+02
5.3629E+02
6.0052E+02
6.9540E+02
6.9540E+02
Plastic Strain
pl = 1.00E+00
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
Yield stress
[N/mm2 ]
3.2877E+02
3.3433E+02
3.4210E+02
3.5258E+02
3.6603E+02
3.8208E+02
3.9940E+02
4.1597E+02
4.3032E+02
4.4332E+02
4.5819E+02
4.7839E+02
5.0628E+02
5.4302E+02
5.8721E+02
5.8721E+02
Plastic Strain
pl = 1.00E+01
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
111
Yield stress
[N/mm2 ]
3.3306E+02
3.3846E+02
3.4597E+02
3.5604E+02
3.6882E+02
3.8376E+02
3.9929E+02
4.1299E+02
4.2283E+02
4.2892E+02
4.3319E+02
4.3702E+02
4.3971E+02
4.3844E+02
4.2834E+02
4.2834E+02
Plastic Strain
pl = 2.50E+01
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
Yield stress
[N/mm2 ]
3.4119E+02
3.4639E+02
3.5359E+02
3.6316E+02
3.7513E+02
3.8877E+02
4.0220E+02
4.1262E+02
4.1753E+02
4.1644E+02
4.1053E+02
4.0034E+02
3.8470E+02
3.6154E+02
3.2908E+02
3.2908E+02
Plastic Strain
pl = 1.00E+02
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
Yield stress
[N/mm2 ]
3.4568E+02
3.5083E+02
3.5796E+02
3.6741E+02
3.7920E+02
3.9254E+02
4.0550E+02
4.1519E+02
4.1903E+02
4.1641E+02
4.0844E+02
3.9559E+02
3.7675E+02
3.5024E+02
3.1499E+02
3.1499E+02
Plastic Strain
pl = 1.00E+03
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
Yield stress
[N/mm2 ]
3.4568E+02
3.5083E+02
3.5796E+02
3.6741E+02
3.7920E+02
3.9254E+02
4.0550E+02
4.1519E+02
4.1903E+02
4.1641E+02
4.0844E+02
3.9559E+02
3.7675E+02
3.5024E+02
3.1499E+02
3.1499E+02
Plastic Strain
pl = 1.00E+04
0.0000E+00
3.4373E03
8.5932E03
1.6327E02
2.7928E02
4.5329E02
7.1431E02
1.1058E01
1.6931E01
2.5741E01
3.8955E01
5.8775E01
8.8507E01
1.3310E+00
2.0000E+00
2.1000E+00
Table B.1: Data for the stress-strain-curves for the respective plasic strain
rate pl .
112
APPENDIX B. STRESS-STRAIN-CURVES
Appendix C
Stress triaxiality
3.3333E+00
3.3333E01
2.6667E01
2.0000E01
1.3333E01
6.6667E02
0.0000E+00
6.6667E02
1.3333E01
2.0000E01
2.6667E01
3.3333E01
4.0000E01
4.6667E01
5.3333E01
6.0000E01
6.6667E01
7.3010E01
8.5097E01
1.0237E+00
1.2435E+00
1.5058E+00
3.3333E+00
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
113
114
Equivalent fracture
strain at damage
3.3238E+01
3.3238E+01
2.3381E+01
1.6447E+01
1.1570E+01
8.1394E+00
5.7268E+00
4.0303E+00
2.8377E+00
2.0000E+00
1.4124E+00
1.0013E+00
7.1550E01
5.1920E01
3.8780E01
3.0480E01
2.6000E01
2.4758E01
3.0283E01
6.1953E01
1.9018E+00
7.5608E+00
7.5608E+00
8.4132E+01
8.4132E+01
4.7327E+01
2.6621E+01
1.4974E+01
8.4238E+00
4.7393E+00
2.6675E+00
1.5034E+00
8.5084E01
4.8779E01
2.9076E01
1.9263E01
1.6006E01
Stress triaxiality
3.3333E+00
3.3333E01
2.6667E01
2.0000E01
1.3333E01
6.6667E02
0.0000E+00
6.6667E02
1.3333E01
2.0000E01
2.6667E01
3.3333E01
4.0000E01
4.6667E01
5.3333E01
6.0000E01
6.6667E01
7.3010E01
8.5097E01
1.0237E+00
1.2435E+00
1.5058E+00
3.3333E+00
3.3333E+00
3.3333E01
2.6667E01
2.0000E01
1.3333E01
6.6667E02
0.0000E+00
6.6667E02
1.3333E01
2.0000E01
2.6667E01
3.3333E01
4.0000E01
4.6667E01
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
115
Equivalent fracture
strain at damage
1.8195E01
2.6576E01
4.4001E01
7.4341E01
2.0883E+00
9.2599E+00
6.1718E+01
5.9367E+02
5.9367E+02
8.4132E+01
8.4132E+01
4.7327E+01
2.6621E+01
1.4974E+01
8.4238E+00
4.7393E+00
2.6675E+00
1.5034E+00
8.5084E01
4.8779E01
2.9076E01
1.9263E01
1.6006E01
1.8195E01
2.6576E01
4.4001E01
7.4341E01
2.0883E+00
9.2599E+00
6.1718E+01
5.9367E+02
5.9367E+02
Stress triaxiality
5.3333E01
6.0000E01
6.6667E01
7.3010E01
8.5097E01
1.0237E+00
1.2435E+00
1.5058E+00
3.3333E+00
3.3333E+00
3.3333E01
2.6667E01
2.0000E01
1.3333E01
6.6667E02
0.0000E+00
6.6667E02
1.3333E01
2.0000E01
2.6667E01
3.3333E01
4.0000E01
4.6667E01
5.3333E01
6.0000E01
6.6667E01
7.3010E01
8.5097E01
1.0237E+00
1.2435E+00
1.5058E+00
3.3333E+00
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
Table C.1: Data for the ductile damage initiation criteria curves.
116
Appendix D
1.0000E+01
1.4236E+00
1.4625E+00
1.5013E+00
1.5401E+00
1.5789E+00
1.6177E+00
1.6566E+00
1.6954E+00
1.7342E+00
1.7730E+00
1.8118E+00
1.8506E+00
1.8895E+00
1.9283E+00
1.9671E+00
2.0059E+00
2.0447E+00
2.0835E+00
2.1224E+00
2.1612E+00
2.2000E+00
1.0000E+01
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
1.00E04
117
118
Equivalent fracture
strain at damage
2.7610E01
2.7610E01
2.6130E01
2.5300E01
2.5100E01
2.5510E01
2.6560E01
2.8250E01
3.0650E01
3.3790E01
3.7780E01
4.2690E01
4.8650E01
5.5810E01
6.4350E01
7.4480E01
8.6440E01
1.0053E+00
1.1710E+00
1.3655E+00
1.5937E+00
1.8611E+00
1.8611E+00
3.3382E01
3.3382E01
3.3361E01
3.3552E01
3.3955E01
3.4572E01
3.5409E01
3.6473E01
3.7765E01
3.9297E01
4.1077E01
4.3117E01
4.5430E01
1.0000E+01
1.4236E+00
1.4625E+00
1.5013E+00
1.5401E+00
1.5789E+00
1.6177E+00
1.6566E+00
1.6954E+00
1.7342E+00
1.7730E+00
1.8118E+00
1.8506E+00
1.8895E+00
1.9283E+00
1.9671E+00
2.0059E+00
2.0447E+00
2.0835E+00
2.1224E+00
2.1612E+00
2.2000E+00
1.0000E+01
1.0000E+01
1.4236E+00
1.4625E+00
1.5013E+00
1.5401E+00
1.5789E+00
1.6177E+00
1.6566E+00
1.6954E+00
1.7342E+00
1.7730E+00
1.8118E+00
1.8506E+00
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
1.00E03
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
119
Equivalent fracture
strain at damage
4.8038E01
5.0943E01
5.4171E01
5.7742E01
6.1678E01
6.6005E01
7.0762E01
7.5956E01
8.1630E01
8.1630E01
3.3382E01
3.3382E01
3.3361E01
3.3552E01
3.3955E01
3.4572E01
3.5409E01
3.6473E01
3.7765E01
3.9297E01
4.1077E01
4.3117E01
4.5430E01
4.8038E01
5.0943E01
5.4171E01
5.7742E01
6.1678E01
6.6005E01
7.0762E01
7.5956E01
8.1630E01
8.1630E01
1.8895E+00
1.9283E+00
1.9671E+00
2.0059E+00
2.0447E+00
2.0835E+00
2.1224E+00
2.1612E+00
2.2000E+00
1.0000E+01
1.0000E+01
1.4236E+00
1.4625E+00
1.5013E+00
1.5401E+00
1.5789E+00
1.6177E+00
1.6566E+00
1.6954E+00
1.7342E+00
1.7730E+00
1.8118E+00
1.8506E+00
1.8895E+00
1.9283E+00
1.9671E+00
2.0059E+00
2.0447E+00
2.0835E+00
2.1224E+00
2.1612E+00
2.2000E+00
1.0000E+01
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
2.50E+02
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
1.00E+03
Table D.1: Data for the shear damage initiation criteria curves.
120
Appendix E
Table E.1: Data for the MSFLD damage initiation criteria curve.
121
122
Appendix F
Tracking Approach
For tracking the motions of the contact surfaces, very sophisticated search
algorithms are used. In the contact pair algorithm, a global and a local
search are carried out for the tracking approach. The global search by default
is only performed every 100 increments and is conducted to determine the
closest master surface facet for each slave node and a tracked master surface
node, which is the node on the master surface closest to the slave node. In
Fig. F.1 node 50 is the slave node, element 10 is the master surface facet
and node 100 the tracked master surface node.
Since the cost of the global search is relatively high, additionally a local
search algorithm is used (compare Fig. F.2), where only the facets that are
attached to the previous tracked master surface node (100) are checked to
find out which is the closest one, here facet 10 is closer than facet 9. In the
123
124
next step it is checked if the second node on facet 10 (here 101) is closer to
the slave node. If yes, the same procedure goes on until the closest node
(here 102) is found. That is then the case, when the tracked master surface
node remains the same from one iteration to the next.
Usually that works well because time increments are very short and thus the
movement during one increment very small. But there are certain situations
that may cause the local contact search to fail. Then it might help to
perform a global contact search more often. Another case where a global
search should be carried out every few increments, is due to its generality
and complexity, the performance of self-contact of a single surface. It is thus
significantly more expensive than two-surface contact.
Appendix G
126
Figure G.1: Total energy for a mesh of different sizes with only Quadelements (Dynamic computation with 9 SIMP, failure excluded).
Figure G.2: Total energy for a mesh of different sizes with only Quadelements (Dynamic computation with 9 SIMP, failure included).
Coming back to the curves in Fig. G.1, where failure was not considered,
attempts were made to try and find out what the reason for the decreasing
energy was for small meshes. Hence, the total energy was analysed in more
127
detail for meshes of element size 5.00 mm, 2.50 mm and 2.00 mm. The energy
was therefore split into its individual contributions. As itemised in detail in
Appendix A, the total energy in ABAQUS comprises (components for heat
dependent problems excluded):
ETOTAL = ALLKE + ALLIE + ALLVD + ALLFD ALLWK
For each of the three analysed mesh sizes, a graph was prepared showing
the curves for all components of the total energy over time, plus the total
energy itself. These graphs are displayed in Figures G.3 to G.5.
128
have to be below the others, as it is in the case for the internal energy
curves. However, all that could be reasonably discovered so far, is that the
differences in the total energy derive from differences in the internal energy.
Why this is the case is not yet understood, and presents an interesting area
of future work.
129
Figure G.6: Internal energy for different mesh sizes (no failure).
Figure G.7: Kinetic energy for different mesh sizes (no failure).
130
Bibliography
[ABAa]
[ABAb]
[ADA]
[Ada95]
H. Adam. Untersuchungen zur Optimierung der Vorhersagbarkeit der Energieabsorptionseigenschaften von Karosseriestrukturen. Institut f
ur Kraftfahrwesen, Aachen, 1995.
[Ans00]
[AZT]
[Bat02]
[BIR01]
[CRA]
[DOT07]
[Dud08]
[EEV07]
Eevc final report, May 2007. Homepage of the European Enhanced Vehiclesafety Committee. http://www.eevc.org/wgpages/wg15/wg15index.htm, last
access 16 April 2008.
[ENC]
208. http://www.
last
access
131
132
BIBLIOGRAPHY
[IIH]
[Kra98]
[LVA89]
J. D. Hudson L. V. Atkinson, P. J. Harley. Numerical Methods with FORTRAN 77: A Practical Introduction. Addison-Wesley Publishing Company,
Wokingham, March 1989.
[MK06]
R. Fr
oming und V. Schindler M. K
uhn. Fug
angerschutz: Unfallgeschehen,
Fahrzeuggestaltung, Testverfahren. Springer Verlag, Berlin, 1 edition, October
2006.
[NCAa]
Homepage of the European NCAP; Test procedures for frontal impact. http:
//www.euroncap.com/tests/frontimpact.aspx, last access 16 April 2008.
[NCAb]
[NHT]
[NHT97]
[Tan]
[TB00]
[UNEa]
[UNEb]
Homepage of the United Nations Economic Commission for Europe, regulations 81100. http://www.unece.org/trans/main/wp29/wp29regs81-100.
html, last access 14 November 2007.
[Wal05]
[Wika]
[Wikb]
[Wri06]