Hypermesh
Hypermesh
MARCUS ANDERSSON
PETTER LIEDBERG
Department of Applied Mechanics
Division of Material and Computational Mechanics
CHALMERS UNIVERSITY OF TECHNOLOGY
G
oteborg, Sweden 2014
Masters thesis 2014:40
Cover:
Mesh of a corrugated composite specimen used in crush simulation
Chalmers Reproservice
G
oteborg, Sweden 2014
Abstract
In order to enable further implementation of composite structures in future crashworthy designs, a good way
of modeling composite materials needs to be defined. Many composites exhibit an even energy absorption
during crushing, which is advantageous during e.g. a car crash, since this leads to a constant deceleration of an
impacting object. This is what makes composite materials interesting in crashworthy structures, apart from
the fact that composites exhibits a high stiffness-to-weight ratio. Damage in fiber reinforced polymers can
occur with a combination of many failure mechanisms, such as delamination, fiber kinking, fiber pullouts and
matrix crack propagation. During crushing of composites, most of these modes can be observed. There are
many software using different techniques and material models for fiber reinforced polymers, consequently, the
various methods need to be evaluated.
As an initial part of the project, a literature study of relevant research and development concerning crash
modeling of composite structures was conducted in order to obtain an overview of the existing possibilities in
commercially available finite element software. Furthermore, simulations have been performed using the software
LS-DYNA and RADIOSS. The theory and behavior of three different material models were investigated. The
investigated material models are: MAT54 and MAT262 in LS-DYNA and LAW25 in RADIOSS. Simulations
that were performed were compared with results from a physical test. More specifically, the force-displacement
and energy absorption-displacement were compared.
All three material models use different material parameters, failure models and degradation schemes. MAT54
is a simple model, for which many input parameters lack direct physical meaning. Failure is based on the
Chang-Chang failure criterion. MAT262 is an advanced material model based on fracture toughness and
includes a continuum damage behavior. LAW25 in RADIOSS is also an advanced material model and involves
a plasticity formulation which gives the user the possibility to define a complex stress-strain curve.
All tested material models could be tuned to match the test data, but with different accuracy. The predictability
could not be commented for the material model MAT262 and LAW25, since vital material data could not be
obtained. All material data was known for MAT54 and the material model exhibits a poor predictability.
Keywords: composite failure, crush, crash, Finite element method, LS-DYNA, MAT54, MAT262, RADIOSS,
LAW25
ii
Preface
This Masters thesis project is a part of the research project Modelling crash behaviour in future lightweight
composite vehicles financially supported partly by Vinnova and partly by the industrial partners of the project
Volvo Car Corporation, AB Volvo, Autoliv, Altair Engineering, DYNAmore Nordic,
AF Industry AB, Semcon
AB and Escenda Engineering AB (50%). The work was carried out during the spring semester of 2014 (30
higher educational credits, at the CAE department at
AF office in Gothenburg, Sweden. The examination
was done by the department of Applied Mechanics at Chalmers University of technology and the work was
supervised by
AF.
First of all we would like to thank Jonas Zachrison for giving us the opportunity to carry out this Masters
thesis at
AF. Furthermore, our supervisors Henrik Nilsson and Teodor Bengtsson deserves a huge thanks,
for their support and guidance throughout the project. We would also like thank all employees at Technical
Analysis at
AF who made us feel welcome from the very first day. A special thanks goes to Martin Fagerstrom,
our examiner at Chalmers University of Technology, for the great patience, support and knowledge he has given
us. Lastly, we would like to thank our families and friends for their devotion and encouragement.
Marcus Andersson and Petter Liedberg
Gothenburg, June, 2014
iii
iv
Contents
Abstract
Preface
iii
Contents
1 Introduction
1.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Delimitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1
1
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
3
4
8
9
9
3 Theory
3.1 What is a composite? . . . . . .
3.1.1 Fiber reinforced polymers . . .
3.2 Main failure modes of composites
3.2.1 Delamination . . . . . . . . . .
3.2.2 Matrix failure . . . . . . . . . .
3.2.3 Fiber failure . . . . . . . . . .
3.3 Crushing of composites . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
12
12
12
13
13
14
14
15
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4 Crush test
18
5 LS-DYNA
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.1 Simulation model of the corrugated specimen in LS-DYNA . . . . .
5.1.2 Available material models in LS-DYNA . . . . . . . . . . . . . . . .
5.2 Simulations with MAT54/55 - Enhanced composite damage . . . . . .
5.2.1 Theory of failure model and degradation scheme in MAT54 . . . . .
5.2.2 Description of Material Card MAT54 . . . . . . . . . . . . . . . . .
5.2.3 Uniaxial simulations with one element . . . . . . . . . . . . . . . . .
5.2.4 Tensile dog-bone simulations with different element sizes . . . . . .
5.2.5 Implementation of MAT54 in crush simulations . . . . . . . . . . . .
5.2.6 Crush simulation using baseline model with MAT54 . . . . . . . . .
5.2.7 Investigation of mesh dependency in crush simulations with MAT54
5.3 Simulation with MAT262 - Laminated fracture Daimler Camanho . .
5.3.1 Theory of failure model and degradation scheme in MAT262 . . . .
5.3.2 Description of material card MAT262 . . . . . . . . . . . . . . . . .
5.3.3 Uniaxial simulations with one element . . . . . . . . . . . . . . . . .
5.3.4 Tensile dog-bone simulations with different element sizes . . . . . .
5.3.5 Implementation of MAT262 in crush simulations . . . . . . . . . . .
5.3.6 Crush simulation using baseline model with MAT262 . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
21
21
21
23
24
24
26
31
33
34
36
38
40
40
43
46
48
50
50
6 RADIOSS
6.1 Introduction to RADIOSS . . . . . . . . . . . . . . . . . . . .
6.1.1 Simulation model of the corrugated specimen in RADIOSS
6.1.2 Composites and anisotropic materials in RADIOSS . . . . .
6.2 Simulations with material LAW25 . . . . . . . . . . . . . . .
6.2.1 Theory and material parameters of LAW25 . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
55
55
55
56
57
57
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6.2.2
6.2.3
6.2.4
6.2.5
. . . . .
. . . . .
. . . . .
LAW25
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
65
67
68
70
and RADIOSS
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
74
74
75
8 Concluding remarks
77
9 Future work
78
References
79
Appendices
A Appendix - MAT54
A.1 Material card used in base line model using MAT54 . . . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Parameter study - Sensitivity of material model parameters in MAT54 . . . . . . . . . . . . . . . .
A.3 Parameter study - Sensitivity of simulation model for MAT54 . . . . . . . . . . . . . . . . . . . . .
I
I
II
X
B Appendix - MAT262
XII
B.1 Material card used in the simulation with uniform mesh model using MAT262 . . . . . . . . . . . XII
B.2 Material card used in the simulation with irregular mesh model using MAT262 . . . . . . . . . . . XIII
C LAW 25
XIV
C.1 Sensitivity of material model parameters in LAW 25 . . . . . . . . . . . . . . . . . . . . . . . . . . XIV
C.2 Influence of crushing velocity with RADIOSS model . . . . . . . . . . . . . . . . . . . . . . . . . . XVI
C.3 Material Parameters of the Baseline Model with LAW 25 . . . . . . . . . . . . . . . . . . . . . . . XVIII
vi
Introduction
Focus in this Masters thesis is on benchmarking existing commercial CAE software regarding crushing of
composites materials. Crash simulation tools with a truly predictive capability is one of the most prominent
enablers for the introduction of fiber reinforced polymer composites in structural automotive parts, thereby
allowing for significant weight reduction at maintained or even improved crashworthiness. The industrial
relevance is manifested by the substantial industrial interest in the current project.
1.1
Purpose
The main purpose of this thesis was to give relevant information regarding composite modeling to the research
project Modelling crash behaviour in future lightweight composite vehicles. This was done by investigating and
evaluating how commercial CAE software can be used to model crushing of composite materials. Results from
the CAE software were compared with results from a physical crushing test. In order to fulfill the specified
purpose following subgoals were stated:
Present how the theory regarding failure of composites is implemented in commercial software, especially
during crushing.
To model a quasi-static crushing test using both LS-DYNA and RADIOSS and compare with experimental
results obtained from a physical crush test, reported in the literature.
To assess the tested software regarding flexibility, level of detail and required material parameters.
To find and compare advantages and disadvantages for the tested software.
1.2
Delimitations
To stay within the time-frame of the master thesis project the following delimitations have been specified:
The project has focused mainly on two commercial software, LS-DYNA and RADIOSS, since the companies
who distributes these software are involved in the research project Modelling crash behavior in future
lightweight composite vehicles which this Masters thesis project is a part of.
Each software has several material models for composites. The focus will be on one simple and one
advanced material model in LS-DYNA and one advanced material model in RADIOSS.
Only single-shell models will be investigated, since solid models tend to be too expensive regarding CPU
cost in full vehicle crash simulations.
Other types of advanced element modeling, such as several shells stacked on top of each other with
cohesive elements in between each shell, have not been tested.
1.3
Method
In the beginning of the project, a literature study was performed, to investigate what damage mechanisms
occur during crushing of composites, as well as how the energy is absorbed in the material. A second literature
study was performed to investigate other projects, where FE-software have been used to solve the same type of
crushing problems.
Tutorials and courses were taken for the software that were used in the project. The software that have been
used are the preprocessor ANSA (version 15.0.0), as well as LS-DYNA (version R7.1.0) and RADIOSS (version
12.0). The software that was used for post-processing was HyperView (version 12.0).
1
In ANSA, the geometry and mesh of the specimen from the crush test were created. The composite stack-up
and material model were also defined in ANSA, with the keywords and functions of the respective software
considered. The loads and boundary conditions were determined in order to match the physical crush test and
the results from the simulations were compared with the results from the same test.
For LS-DYNA, two material models were tested, MAT54 and MAT262. The chapter covering LS-DYNA
simulations thus includes an introductory part where the FE-model specifics (used in simulations with both
material models) are presented. This is followed by a section covering the theory and description of material
model specifics of MAT54. Results from simulations with MAT54 are also presented here. Then, a section
regarding the theory and description of material model specifics of MAT262 follows. The results from simulations
with MAT262 are included here as well.
For RADIOSS, only one material model was tested, LAW25. The RADIOSS chapter is structured in the same
way as the LS-DYNA chapter. First, the descriptions of the FE-model specifics are presented, followed by
the theory and description of material model specifics. Then the results from simulations with LAW25 are
presented.
The strategy for simulations with each material model is presented in the list below:
Investigate the theory of the material model.
Describe the material card and material model parameters.
Perform uniaxial tests on a simple model, to verify the material behavior.
Perform simulations on dog-bone specimens, to investigate element size dependency.
Perform simulations on the corrugated specimen that was used in the actual crush test.
Investigate sensitivity of material model parameters.
Composite materials is a field in which a lot of research is going on, especially when it comes to failure models
and fracture mechanics. The interest in the subject comes from the possibilities that composites can provide.
For example, composites often exhibit high stiffness (and strength) to weight ratio and high energy absorption
in axial crushing. Much of the mechanical research addresses the problem of how to make it possible to capture
all phenomena that occur during failure of the material. Different approaches on how to do this can be found.
As pointed out in [1] there are two major branches, the micro-mechanical and the macro-mechanical approach.
There are many reports where the micro-mechanical behavior is observed and methods developed to take these
mechanisms into account. Micro-mechanical modeling, will put high demands on computer performance and is
thus unpractical for engineers in crash analysis [1]. However, some researches express that the micro-mechanical
behavior needs to be included in the material models in order to accurately capture the complex mechanics
in composites. Many software today use macroscopic strength criteria to estimate the strength in composite
materials. The limitations in todays computer performance makes the macroscopic perspective favorable, but
in many cases it leaves uncertainties whether all aspects of composite crash behavior are captured and if the
models are truly predictive.
Many of the studies performed in the literature have been made on components with simple geometry, such
that a real test can be performed in a controlled manner, where all necessary data can be measured. The data
obtained from simulations are then verified against the data from the real test. The following section will
describe how different software have been used by others in the field of crash analysis of composite components.
2.1
The Federal Aviation Administration, FAA, is a national aviation authority in the United States. One of FAAs
major function is to define the certification methods and standards regulating civil aviation. Composites are
non-standard technology with limited shared databases, methods and guidelines. In 2003, FAA announced the
initiation of a joint award between the Wichita State University and the University of Washington to create
the Joint Advanced Material and Structures Center of Excellence, JAMS.
JAMS is a major player in the development of composite materials and methodology. One of its projects is
to examine the crashworthiness of composite structures which is a relative new topic for the FAA certification[2]. The objective is to develop a guidance certification analysis and test protocol for composite fuselage
crashworthiness certification in order to increase the confidence and level of safety of composite components.
The crashworthiness project started in 2009 and is an ongoing project divided into four stages, see Figure
2.1. Stage 1 is to perform axial crushing tests on corrugated specimens and investigate the predictability and
correlation for different commercial software. Stage 2 is to repeat the axial crushing test but for a number
of different c-channel specimens. Stage 3 is to investigate the the characterization of joints and strain rate
effects on both quasi-static and dynamic tests. Stage 4 is to simulate dynamic crushing of sub-assembly models
e.g. fuselage support with bolted joints. The software investigated throughout these four stages is mainly
LS-DYNA, RADIOSS, PAM-CRASH and ABAQUS.
3
(a)
(b)
(c)
(d)
Figure 2.1: Stage 1 to 4 in JAMS crashworthiness project [2]. (a) Axial crushing on corrugated specimen. (b)
Axial crushing on c-channels. (c) Strain rate sensitivity and joints. (d) Sub-assembly simulation
The JAMS crashworthiness project involves many of the big mechanical companies in the United States. e.g.
General Motors, Ford, and Boeing Research & Technology.
The specimen, used in stage 1 of the JAMS crashworthiness project, is the same as is used in this master thesis,
see Figure 2.2. Simulations have been performed with different software and different material models. A lot of
the detailed results from the investigated software are not published or found. So far the result of the project
determines that none of the software are truly predictive but all of the software can reproduce the test with
a variation in accuracy [2]. This project with its well known partners has influenced other researchers and the
same geometries are recurrent in many other projects.
2.2
Projects in LS-DYNA
Methods trying to capture the crushing behavior have been tried with different material models and approaches
in LS-DYNA, e.g. with shells, tiebreak contact, cohesive elements and solids.
With the approach, made by Feraboli et al., a progressive row-by-row failure was achieved. The reaction force,
between the corrugated specimen and the rigid plate, will increase to a high peak value and drop down to
zero, as an element row is deleted. The spiky reaction force is filtered with a SAE600 filter, see Figure 2.4a.
In Figure 2.4b the filtered reaction force is compared to test data and show a good correlation. The study
shows that a simple shell definition using MAT54 (with all material data known) can be tuned to get a good
correlation, but is not a predictable model.
(a)
(b)
Figure 2.4: Reaction force from simulations by Feraboli et al. [3]. (a) Raw data compared to filtered data
(SAE600 filter). (b) Filtered data (SAE600 filter) compared to test data
Simulation of impact on a pressure vessel using thick-shells and solid elements with MAT22
In [4], a carbon fiber reinforced plastic vessel was modeled during a crash test. To simplify the testing and the
modeling the vessel was cut into a cylinder and fixed to a stand, see Figure 2.5. A sled device with an impactor
was crashed into the side of the vessel segment. Modeling this thick composite structure with only shell elements
will produce inaccurate results. A cost efficient way to increase the accuracy through the thickness is to use
thick-shells instead of letting solid elements represent each ply. Thick-shells (type 3) utilize a 3D stress state,
instead of only in-plane stresses. The test was performed with both thick-shells elements and solid elements.
Each thick-shell represented a few of the used plies in the test, whereas the solid element model had a layer
of solid elements representing each individual ply. In all simulations the simple material model MAT22 was
used. MAT22 is based on a reduced Chang-Chang failure criterion, where elements are deleted when the
Chang-Chang criterion is fulfilled.
5
Since each individual layer is modeled as a thick shell the computational time is increased radically compared
to if regular shell elements would have been used, thus modeling all plies at once. Between each layer a tiebreak
contact is defined. The tiebreak contact transmits both compressive and tensile forces with optional failure
criterion. Each ply is thus glued together and has a failure criterion to model the delamination behavior of the
laminate. In Figure 2.6 a cross section of the vessel can be seen with thick-shell and solid elements. At the top
of the figure, the impactor comes into contact with the laminate.
(a)
(b)
Figure 2.6: Cross-section of the pressure vessel [4]. (a) Thick shell model with tiebreak contact. (b) Solid
element model.
Neither of the performed simulations can reproduce the reaction force from the sled test accurately, as can be
seen in Figure 2.7. A more advanced material model should have been used in order to enhance the accuracy.
Both thick shells and solid elements are good tools to capture the behavior through the thickness since they
do not smear the material properties as the shell model do. Even though the size of the model is increased
the thick-shells are more cost efficient, than letting solid elements represent each individual ply. This material
model do not include delamination, therefore a tiebreak contact is a good way of introducing delamination to
the model.
6
Figure 2.7: Comparison of reaction force for different element formulations in the simulation with the pressure
vessel [4]
Simulation of impact attenuator on a racing car using shell elements with MAT55
In [5] and [6], computational crash simulations on an impact attenuator has been performed using LS-DYNA.
The same simulation has also been performed with RADIOSS [5]. During the development of a small formulastyle racing car an impact attenuator is made of a carbon fiber composite material, see Figure 2.8. The race
car is part of the yearly Formula SAE race. Formula SAE is a student design and manufacturing competition
organized by the Society of Automotive Engineers, SAE. To ensure drivers safety in high speed crashes, special
impact structures are designed to absorb the kinetic energy during a crash situation.
The front impact attenuator is a thin walled crash box made of carbon fiber. In the simulation a rigid wall with
the mass of 300kg is crashed into the impact attenuator with a speed of 7m/s. To trigger a correct crushing
behavior the number of plies is reduced in three steps, first 7 plies (red), then 9 plies (green) and last 10 plies
(blue), see Figure 2.8b. The impact attenuator is model with shell elements and material model MAT55. Some
of the strength reduction parameters implemented in MAT55 are used to further improve the progressive failure
behavior.
(a)
(b)
Figure 2.8: Illustration of the placement and shape of the impact Attenuator [5]. (a) Formula SAE racing car.
(b) Impact Attenuator.
The result from the simulations correlate quite well to the performed test, seen in Figure 2.9. This shows that
under right circumstances quite simple models can model a crush situation. Whether the result is predictive or
just adjusted to correlate with the test is not discussed.
7
Figure 2.9: Reaction force from test and simulation for the impact attenuator [5]
2.3
Project in RADIOSS
The crush simulations performed on the race car attenuator described earlier in Section 2.2, see Figure 2.8a and
2.8b [5], was also done with the explicit finite element solver RADIOSS. Modeling of the composite material
was done by using Type10-Composite Shell. With this property multiple layers of composite material can be
modeled with constant layer thickness and material definition, but variable material direction. Material model
LAW25 has been used, with the built in Tsai-Wu plasticity formulation, which enables the material to yield in
all in-plane directions and hardening of the material is isotropic.
With LAW25, a layer in the composite can get completely damaged and loose its mechanical properties either
due to tensile strain limits or maximum plastic work limits. The complete failure of a layer due to tensile
failure strains is preceded by material stiffness degradation after a certain strain value is reached in tensile
fiber, or transverse, direction. The maximum plastic work limit can be specified to be taken into account in all
directions, including compression. Plastic work is defined as the work that is done during plastic strain.
The final deletion of elements can be set to occur if either one layer, or all layers has reached maximum plastic
work limit, but also if tensile failure strains are exceeded in fiber or transverse direction. In [5] they have used
element deletion if the maximum plastic work limit is reached in one ply. The force-displacement results can
be seen in Figure 2.10. Apart from the initial peak force, the force from simulation follows the trend of the
experiment curve and the authors of the article are pleased with the results.
Figure 2.10: Force displacement curve from simulations, with RADIOSS, performed on the attenuator compared
with the experimental results [5].
2.4
Project in ABAQUS
A similar corrugated composite specimen that was used by the JAMS project [7] has been modeled by 3DS
Simulia [8]. Simulia have used their own commercial software, Abaqus/Explicit, to demonstrate its benefits.
They state that their model consider both in-plane fracture mechanisms as well as the interlaminar delamination
mechanism. Their approach was to model 8 layers of individual continuum shell elements (SC8R) with a
surface based cohesive contact between each layer of shell elements. The continuum shell elements act similar
to conventional shell elements but have the geometry of three-dimensional solid elements. The cohesive contact
between the shell elements enables the model to simulate delamination between the layers.
The in-plane behavior is considered by using a built-in user subroutine (VUMAT), which is a constitutive
model for fabric-reinforced composite material. This model uses stiffness degradation after damage initiation
in either fiber or matrix cracking. For the damage behavior, both initiation parameters, such as tensile and
compressive strengths of fiber, and damage evolution coefficients are needed. The damage evolution coefficients
are based on fracture energies per unit area. In shear loading, plastic deformation is available.
Delamination, due to failure of the interface between the shells, is initiated by defining the quadratic nominal
stress criterion (QUADS) in the cohesive contact. This criterion is based on maximum allowed normal and
shear stress. After damage is initiated, the degradation of the contact was done via an exponential softening
law.
The model was meshed with elements of 1x0.5 mm2 giving a total of one half million degrees of freedom. The
results from simulation compared to those from test are quite satisfying, see Figures 2.11a and 2.11b. However,
the time required for the simulation is not presented.
(a)
(b)
Figure 2.11: The model and results from [8]. (a) The model after some crushing. (b) Load-displacement results
2.5
Project in PAM-CRASH
In [1], the commercial explicit finite element software PAM-CRASH is used to model two different approaches
on crushing of a square composite tube. The first model was built with several shells stacked on each other and
the second was modeled with only one shell. The test that provided the reference data for the simulations can
be seen in Figure 2.12.
9
Figure 2.12: The test performed in [1]. Left figure shows test specimen in the test rig. The right figure is the
test specimen after crushing.
In the first model different layers of shell elements have been joined together using cohesive elements. The shells
are of PAM-CRASH material model 131, with property type 7, which is a fabric composite global ply model.
When loading the material its in-plane properties will be degraded due to micro-cracking before ultimate failure.
The cohesive elements are one-dimensional of material model 303. Two adjacent shells are tied together with a
penalty formulation, where a slave node on one shell is connected with a master segment on the other shell.
Failure of the inter laminar elements is based on the model proposed by Pickett et al. [9]. This delamination
approach is implemented in the PAM-CRASH code. The failure will occur when critical elastic fracture energies,
stored in a crack, are exceeded. Mode 1 and 2 (opening and in-plane shear mode respectively) fracture is
considered in this model, but also mixed mode delamination. In the material model, for the cohesive elements,
maximum allowed stress is defined, with which the elastic energy limits for both modes are computed [1].
Mesh dependence was tested by adopting two different mesh sizes. One of 7x7 mm2 and one of 4x4 mm2 .
The results differed in how the crushing load oscillated around its average force. For the larger mesh size the
oscillations were greater, but both mesh sizes resulted in same average crushing force.
Different amount of inter laminar element layers (3, 4 and 5) was tested. The results with the stacked shell
model indicates an underestimation of energy absorption. The reason for this is discussed and assumed to be
due to the models inability to accurately take friction effects into account. The friction along already crushed
material is not included. Sliding between the shell layers is also not included due to the fact that two adjacent
shells are actually not in contact (cohesive elements in between). However, deformation of the cohesive inter
laminar elements, who are placed in between the layers of shells, give rise to some energy absorption. Another
comment made is that the model was not robust in the sense that it could give completely different crushing
behavior if just a small modification of parameters was made.
The second model in [1] was built with a single layer of shell elements. Same material as in the stacked
shell model is used in this simulation (material model 131), but in addition a crushing zone model is applied.
Crushing zone is where the model first go into contact with the plate, which it is crushed against. The crushing
zone formulation is progressively activated for elements that are located in the region of crushing. The material
response is changed to give an elastic-perfectly plastic behavior in the crushing direction. The element is
deleted when a minimum allowed ratio between the current element area and the initial area is reached.
This model gave stable results which were very close to the reference data. The single layer shell model was also
less time consuming to simulate. At least 30% less time was required. However, the model does not capture
the fracture mechanisms that occur in a real test. Figures 2.13a - 2.13d illustrate a small excerpt of the results
obtained with the stacked shell and the single shell models.
10
(a)
(b)
(c)
(d)
Figure 2.13: The different models and corresponding results obtained in [1]. (a) Stacked shell model, with 5
inter laminar layers, after crushing. (b) The propagation of crushing in the single shell model. (c) Results from
the tests with 5 inter laminar layers and variations in values on critical energy release rates. (d) Results from
the single shell model.
11
3
3.1
Theory
What is a composite?
When two or more materials are combined on a macroscopic-scale they form a composite. A typical example
of composite material is ply-wood, where thin layers of wood are glued together with the wood grain of the
layers in different angles to gain strength and stiffness in more than one direction. Another example is concrete,
where aggregate particles are mixed with a binding material, which glues the aggregates together. This thesis
is however only to discuss the composites of fiber reinforced polymers, FRP. Common materials to use in FRPs
are fibers of carbon or glass in a matrix of epoxy or polymer thermosets. The FRPs can be divided into two
different groups, short fiber reinforced polymers and long fiber reinforced polymers. Another delimitation in
this thesis is that only long fiber composites are considered, and the FRP acronym will henceforth only refer to
the long fiber group.
3.1.1
FRPs can be of single-layer or multi-layer type. The single-layer type also includes composites where several
layers are stacked with the material orientation in same direction. The multi-layered composites are where
several layers are stacked with fibers orientation in different directions. Each layer can be unidirectional, where
all fibers in that specific layer are aligned in one direction. They can also be woven, where the fibers have two,
or more, directions. The idea of having layers with different material orientation is to give different strengths
and stiffness of the laminate to various directions.
In contrast to many other engineering materials, such as metals, composites are not isotropic. They are
anisotropic in the sense that the properties (stiffness, strength, toughness etc.) differ in different directions.
Layered FRPs are rather orthotropic, which means that properties are different i three perpendicular directions.
The orthotropic directions can have different notations. The longitudinal fiber direction is often denoted as
direction 1, A, or L. Transverse direction is denoted 2, B or T . The out-of-plane direction is denoted 3, C or
T 0 , see Figure 3.1. These directions are local to each ply in the stack-up. If the properties in all transverse
directions to the fiber are the same, the material is defined as transversely orthotropic.
For a unidirectional ply, such as that in Figure 3.1, the mechanical properties are highly dependent on the fiber
fraction. The properties of the matrix and the fibers are combined, where the fiber fraction is what determines
the longitudinal and transverse stiffness properties for that ply. When plies with different angles for the fiber
direction are stacked, the contributions from each layer are transformed to follow a global coordinate system,
see Figure 3.2.
When the transformation for each layer is done, the property contributions from each layer can be added up in
three different matrices, the extensional stiffness matrix, coupling stiffness matrix and bending stiffness matrix
(A, B and D respectively in Equations 3.1 and 3.2). Where the contribution from each ply in the two latter
matrices depends on the distance to the mid-plane of the laminate. These matrices are what couples forces and
moments to strains and curvatures in the laminate. Two relations can be written as in Equation 3.1 and 3.2
Nx
N
y
Nxy
Mx
M
y
Mxy
0x
0
= A y
0
xy
0x
0
= B y
0
xy
kx
k
+ B y
kxy
kx
k
+ D y
kxy
(3.1)
(3.2)
Where N are the forces, M are the moments, k are the curvatures and 0 are the mid-plane strains. Further
information on how the stiffness matrices are derived and constructed can be read in [10].
3.2
When a composite is subject to loading, several different failure modes are present. Most of the software used
today in crash analysis of composites are based on macro-mechanics and can not capture all of the failure
modes on a microscopic scale. However, quite accurate estimations of global energy absorption can be obtained,
but it is of importance to be aware of the possible failure modes in crushing of composites in order to assess
the quality of simulations.
According to [11] the main failure modes of fiber-reinforced composites are delamination, matrix compressive
failure, matrix tensile failure, fiber compressive failure and fiber tensile failure. Each of these modes require
some explanation. The schematic Figures 3.3 and 3.4 show how different loading give different angle on the
crack.
(a)
(b)
Figure 3.3: Matrix failure modes. (a) Shear matrix failure due to compression in transverse direction. (b)
Tensile matrix failure perpendicular to tensile load in transverse direction.
(a)
(b)
Figure 3.4: Fiber failure modes. (a) Shear fiber failure due to compression in fiber direction. (b) Tensile fiber
failure perpendicular to tensile load in fiber direction.
3.2.1
Delamination
When it comes to fracture mechanics and crash analysis one of the interesting characteristics is the amount
energy the material will be able to absorb. In [10] it is described that during loading, energy is absorbed in
13
two different main mechanisms. The first is deformation of the material and the second is formation of new
surfaces, such as cracks. If a composite material undergoes deformation to such extent that delamination cracks
are present, these cracks will absorb a significant amount of energy when they propagate. A delamination crack
is often preceded by a crack who has propagated through a ply until it has reached the fibers of an adjacent ply.
Then the crack may branch off to go along the interface in between the plies, causing the delamination crack
to form. Delamination is often found in specimens who have been subject to bending, due to predominant
out-of-plane shear stresses. Figure 3.5 illustrates a cutout of a specimen subjected to impact, causing it to
bend, where matrix cracking in the intermediate transverse ply has grown to the interfaces between the plies
causing delaminations.
Figure 3.5: Delamination crack in specimen due to impact which has caused bending [12]
3.2.2
Matrix failure
When a composite fail in compressive matrix mode, the crack is most often at an angle to the loading direction.
This failure mode is therefore primarily a shear matrix failure, as stated in [11], an illustration of which can be
seen in Figure 3.3a.
Failure due to transverse tensile loading can depend on either matrix tensile failure or constituent debonding.
In contrast to compressive matrix failure, the fracture plane in tensile matrix failure is perpendicular to the
loading as illustrated in Figure 3.3b [10].
3.2.3
Fiber failure
If a composite is loaded in compression in the fiber direction, different failure modes can occur. The Poissons
ratio effect, causing tensile stresses in transverse direction, can initialize cracks at the interface between fiber
and matrix. This mode is called transverse tensile failure. Another compressive fiber failure mode is fiber
micro-buckling. The micro-buckling of two adjacent fibers can buckle in-phase or out-of-phase. However, it is
most likely that the buckling of fibers occur in-phase, see Figure 3.6a, if the fibers are densely packed. This,
in-phase mode is also called shear mode, since the matrix exhibits mostly shear strains. For carbon fibers, the
angle of the crack is often at approximately 45 to the loading axis [10]. The shear mode is typically triggered
at imperfections like voids in the matrix, making the fiber loose its lateral support. It can also be triggered
where the fibers are misaligned. This mode cause kink bands to form [11]. Figure 3.6a illustrates a kink band,
shear mode failure, through a single bundle of fibers, while Figure 3.6b illustrates kink bands through several
plies and matrix cracking in the transverse intermediate plies.
14
(a)
(b)
Figure 3.6: Kink band formations due to compression in fiber direction [13]. (a) Kink band through a single
bundle of fiber. (b) Kink bands through several bundles of fibers and matrix cracking in the intermediate
transverse plies.
When fibers are loaded in longitudinal direction, they will break in a brittle behavior at their weakest cross
sections. Fibers, who are brittle, fail at low fracture strains. Even though the high strength contribution in
a composite comes from the fibers, the energy absorption from fiber breakage when a composite fails is very
low. But, as stated in [10] the presence of fiber influences the various failure modes and thereby also the total
absorbed energy. When fibers break the bonds between matrix and fiber can also break and cracks can go
along the fibers. This behavior is what cause fiber pullouts.
3.3
Crushing of composites
When it comes to crash of structures, especially in vehicle applications, it is of great interest to have a design
and material which can absorb the energy required to achieve an impact at an acceptable level of deceleration.
Tests have shown that composites generally are good at absorbing energy during a somewhat constant crush
load, since many different energy absorbing mechanisms occur simultaneously. Therefore composites are useful
in crashworthy applications. It is however important for the structure to be designed in such way that the
structure fails in a progressive energy absorbing manner. For example if the structure fails in global Euler
buckling the energy absorption will be much lower than if it fails in progressive crushing [14].
Many crush tests on composites have been performed, where the failure modes, energy absorption and crushing
force is observed. In order to avoid catastrophic failure, such as Euler buckling, long thin-walled structures
should not be used. In many tests a chamfered edge is used where the crushing is to be initiated.
Figure 3.7 shows an example of crushing where the progressive behavior is triggered by a chamfered edge. The
figure is based on an actual test performed by Hull [15], from which he describes an average behavior of the
crushing at different stages for a cross section of the tube. In this test a stack-up of [90/0]2S has been used. In
the first stage the inner 90 layer is compressed and fails in shear and is forced to move inwards, marked with
(a) in Figure 3.7. As this layer is forced inwards, compressive stresses along its fiber direction will eventually
make it buckle and kink.
When the middle axial layers come into contact with the rigid plate, which the tube is crushed against, buckling
and kink bands take form in those layers, (b). As crushing proceeds, crushed material forms a debris wedge, (c).
A debris wedge can be seen in most crush tests similar to this, and is what cause both inter- and intralaminar
transverse cracks parallel to the axial fibers in the middle. This will split the material and force it towards the
inside and outside of the tube. A central crack will cause mode I delamination. The splitting of the plate is
called splaying and results in fronds. The plies who exhibit splaying do not fracture, they are rather exposed to
extensive bending, which cause interlaminar and intralaminar shear cracks to form, (d) [16]. Delamination in
the splay, between the 90 and 0 ply, will most likely be mode II delamination, which is a shearing mode.
15
The outer 90 layer will, apart from transverse shear cracks in the matrix due to compression, experience tensile
stresses in its fiber direction, as the material is forced outwards. This will eventually make it fail in tensile fiber
mode.
Figure 3.7: A schematic illustration of how crushing of a tube with chamfered edge can occur, modified from
[15].
Figure 3.8a shows a crushed tube from another crush test where a braided composite tube has been crushed.
The fronds from splaying of the material is clearly visible. Figure 3.8b shows a cross section of the same tube.
(a)
(b)
Figure 3.8: Tube crush test, from [17]. (a) Tube after crush test. (b) A cross section of the crushed tube.
A typical force-displacement curve of axial crushing can be seen in Figure 3.9. Usually there is an initial force
peak, followed by stable crushing where the force are kept at an almost constant level. The initial peak is higher
since no failure modes have yet been triggered. The zone between the force peak and the stable crush zone
is called the transition zone. The transition zone is highly dependent on how the crushing is triggered. The
constant force in the stable crush zone makes the deceleration of an impacting object constant. This, together
with the high stiffness-to-weight ratio, is what make composites good in applications where a crashworthy
structure is important.
16
R
0
F ds
(3.3)
Where is the crush distance and F is the measured force. The specific energy absorption, Es can then also
be determined, according to Equation 3.4, as the energy absorbed per unit mass.
ES =
W
Al
(3.4)
Where is the density, A is the cross-sectional area, l is the crushed length. The specific energy absorption
depends on more than material properties. It is also highly dependent the stack-up of plies and geometry of the
specimen. In cases where the weight of the structure is an important parameter the specific energy absorption
is a good value to use in comparisons between materials and geometries.
17
Crush test
The test, which has provided the reference values for the simulations of this thesis, was performed by Feraboli
and co-workers [3]. A corrugated specimen was used, see Figure 4.1. The specimen was crushed axially between
two hardened polished steel surfaces and the force needed to crush the specimen at a constant velocity of 25
mm/min (1 in/min) was measured. By having the corrugated profile, the specimen is self-supporting. Thus, no
other support is required to keep it from buckling.
The material was of unidirectional tape T700 carbon fibers and 2510 epoxy prepreg. The lamina stack-up was
[0/90]3s (i.e. 12 plies) with a total thickness of 2.0 mm (0.079 in). On the end, where crushing is desired, the
edge is chamfered 45 . This helps to initiate crushing and avoid initial spikes in crush loads. The height of the
plate is 76.2 mm (3 in) and the width is 50.8 (2 in). More details regarding dimensions of the plate can be
found in Figure 4.2
Many tests on same type of specimen were performed. Figure 4.3 illustrates one specimen after crushing. The
force-displacement and the energy absorption-displacement curves from one of the tests can be seen in Figures
4.4a and 4.4b. The force curve has a high initial peak followed by a somewhat constant crush load. The near
constant force is reflected in the linear shape of the energy absorption curve.
18
(a)
(b)
Figure 4.4: Results presented in [3]. (a) Force - Displacement curve from crush test (Note that axes show
pounds and inches). (b) Energy Absorption - Displacement curve from crush test (Note that axes show Joule
and inches)
The material data that was presented by Feraboli for unidirectional tape of the material used in this test, can
be seen in Table 4.1.
Table 4.1: Parameters adopted from [3] to be used in simulations.
Property
Symbol
Value
Density
1.52 g/cm3
Modulus in direction 1
E11
127 GPa
Modulus in direction 2
E22
8.41 GPa
Shear modulus
G12
4.21 GPa
12
0.309
21
0.02049
1t
u
2200 MPa
2t
u
48.9 MPa
1c
u
1470 MPa
2c
u
199 MPa
Shear strength
12
u
154 MPa
19
These material parameters are those that have been used as input parameters for the simulations of this
project and have been considered to be correct, i.e. these values have been used consistently in all simulations.
Furthermore, the model in simulations was built, to resemble the test specimen, with the dimensions according
to Figure 4.2, with the same stack-up of plies (i.e. 12 plies, [0/90]3S ). The 45 chamfer on one side of the
specimen was modeled by reducing the thickness by 0.5 in one row of elements in the finite element mesh.
20
5
5.1
LS-DYNA
Introduction
5.1.1
To know how to set up your simulation model, the physical behavior of the test needs to be understood.
The setup of the model can be divided into three main parts, the geometry, the material and the boundary
conditions. Focus in this project is to analyze the material definition, but to achieve an accurate simulation, a
good combination of all simulation options are required. This section aims to describe the simulation model and
detailed software-specific issues for LS-DYNA. The implementation of material card MAT54 and MAT262, will
be described in Subsection 5.2 and Subsection 5.3, respectively. It should be noted that all values of parameters
have been transformed to fit the following units: Stress in [MPa], force in [N], distance in [mm], energy in
[Nmm] and time in [s]. For detailed information of LS-DYNA specific keywords described in this section, cf.
[18].
In order to use a composite material in LS-DYNA a property called PART COMPOSITE has to be defined.
Here is the lamina stack-up of the composite defined, together with lamina thicknesses and element formulation.
The material cards are then coupled to the PART COMPOSITE. The material direction of each element is
either defined in the material card or with an additional card called ELEMENT SHELL BETA, which will
override the material direction in the material card. Note that not all material cards can handle all types
of element definitions. Usually the element definitions are divided into shell, thick-shell and solid elements.
Within each definition there are many different element formulations, e.g. Belytschko-Tsay, fully integrated
shells etc. They differ in number of nodes, integration points, degrees of freedom, and kinematics.
Both MAT54 and MAT262 are compatible with regular shells and thick shells, as well as with solids. In this
simulation, regular shell elements are used, since this is a common approach to reduce costs in crash simulations
and this project aims to investigate the usage of composites under these circumstances. The usage of shell
elements lead to limitations in capturing delamination and gives a low accuracy of out-of-plane stresses, which
are important parameters in fracture mechanics.
The corrugated specimen is meshed with fully integrated shell elements of type 16, with four in-plane integration
points. From an accuracy standpoint the fully integrated elements are preferred over the under-integrated
element formulation, e.g. type 2, since the fully integrated do not suffer from hourglass modes. The underintegrated elements tend to be too soft and can have problems with hourglass forming, but are cost efficient
since only one in-plane integration point is used. The shell definition through the thickness for composite
materials is defined to one integration point for each layer, hence the corrugated specimen with its 12 plies has
a total of 4 times 12 integration points.
Just as in the crushing test the simulation model needs a trigger to initiate the failure where the specimen is
crushed to the ground. The trigger is modeled as a single element row with reduced thickness at the crush
front of the corrugated specimen, with the only purpose to initiate failure. Both the main laminate and the
21
trigger part has the same material with the same amount of plies, but the thickness of the trigger is reduced
from 2.00mm to 0.50mm. Different triggers, element sizes and mesh layouts have been tried before choosing a
baseline model.
The movement of the laminate is defined with a prescribed motion, BOUNDARY PRESCRIBED MOTION,
applied to the nodes at the top of the laminate, see Figure 5.1. The motion is purely vertical and the remaining
five degrees of freedom are fixed for the constrained nodes at the boundary condition. The prescribed velocity
is ramped up to a constant value of 2000mm/s in order to speed up the simulation. The effect of using
a higher velocity than the performed quasi-static crushing test will be discussed in Subsection 5.2.6. By
further reducing the cost of the simulation, mass scaling is also used in order to maintain the time step at an
affordable level, i.e. to keep the total simulation time low. To reduce vibrations and shockwave propagation
caused by the high velocity, 10% damping is added to the stiffness matrix of the system, with the function
DAMPING PART STIFFNESS. This stiffness damping is an effective way of suppressing high frequencies in
the system.
A predefined standard contact condition, CONTACT AUTOMATIC NODES TO SURFACE, is used between
the ground and the laminate, where the nodes of the laminate are slaves to the master surface of the ground.
The contact condition utilize a penalty method. Both dynamic and static friction is used to minimize oscillation
of the laminate. The ground is constructed as a shell, fixed in space with the rigid body material card, MAT20.
Even though MAT20 defines a rigid body it includes density, Youngs modulus and Poissons ratio which are
used in the computation of the reaction force in the contact condition. The ground is assumed to be made of
steel.
An important parameter to specify when using the shell definitions for composites in LS-DYNA is the LAMSHT
flag in the control card, CONTROL SHELL. This parameter activates laminate theory in the calculation of the
shear strain through the thickness. Otherwise the shear strain is assumed to be constant through the thickness
of the shell. An assumption that can be grossly incorrect if there are drastic differences in the elastic properties
from ply-to-ply, e.g. if sandwich type composites is used. Generally the result tend to be too stiff if this
correction flag is not used [19]. Also activate the NFAIL4 parameter in the CONTROL SHELL allowing highly
distorted element to be deleted before they lead to instabilities causing error termination of the simulation.
Note that the whole element is deleted not just the integration points of the plies.
Additional helpful parameters unique to each material model and which are strongly recommended to activate,
are the history variables. The history variables include damage parameters in different modes for each ply and
crashfront visualization. In order to get the history variable printed out, the IMATL flag has to be active
in the DATABASE BINARY D3PROP section. The DATABASE card controls what and how often certain
parameters are written to the result files. As default the result from only three integration points through
the thickness are written to the result files. In the DATABASE EXTENT BINARY section the NEIPS and
NEIPH parameters controls how many of the integration points that the result to the history variables are
written. NEIPS is used for shells and NEIHP is used for solids.
22
5.1.2
The simulated material behavior is governed by the material model. The input parameters to the material
model are defined in the material card, which usually include some elastic anisotropic material properties,
material coordinate definitions and other model parameters such as failure criterion. Each material model
utilizes a different model strategy for failure initiation prediction, material properties and degradation schemes.
Specific model parameters are also included in the material card. These parameters do not always have an
immediate physical meaning but are needed to run the simulation and tune the material behavior. The amount
of unphysical parameters varies between the material models, but are included in all models.
Within the failure surface, the composite materials behaves orthotropic linear elastic, although some models
can include plastic behavior. When failure is reached, the material properties are degraded according to the
degradation scheme. Some material models utilize a discontinuous failure where the material properties are
set to zero when a ply reaches a certain failure condition, i.e. the ply is either undamaged or fully damaged.
The degradation is generally on the elastic anisotropic material parameters, e.g. Youngs moduli, shear moduli
and Poissons ratios. Other material models utilize a continuum damage mechanics model where the material
properties are reduced after failure is initiated, in order to capture a more physical behavior. The damage of
the ply goes from a undamaged state to a fully damaged state according to the degradation scheme. A ply is
removed when the fully damage state is reached, i.e. the integration points of the ply will no longer contribute
to the element stiffness.
Available material cards for composite modeling can be seen in Table 5.1 [19].
Table 5.1: Available materials for composite modeling in LS-DYNA
Material Laws
Description
Elements
Material card
Failure condition
Elements
Comments
MAT22
Reduced Chang-Chang
Shell, Thick
shell, Solid
54: Chang-Chang
55: Modified Tsai-Wu
and reduced ChangChang
Shell, Thick
shell, Solid
Improvement of MAT22.
Crashfront algorithm.
Strength reductions parameters.
Modified Hashin
Shell, Thick
shell
Similar to MAT54.
Smooth stress strain relation.
Continuum damage model with
exponential softening.
Crashfront algorithm.
Non-Linear shear behavior.
Modified Hashin
Shell, Thick
shell, Solid,
SPH
Similar to MAT58.
Super imposed linear isotropic
viscoelasticity.
No failure
Shell
Modified Hashin
Shell, Thick
shell
Same as MAT58.
Rate sensitive.
Hashin
Shell, Thick
shell, Solid
Composite Damage
MAT54/55
Enhanced Composite Damage
MAT58
Laminated Composite Fabric
MAT59
Composite Failure Model
MAT116/117/118
Composite Layup
MAT158
Rate Sensitive Composite Fabric
MAT161/162
Composite MSC
23
Description
Elements
MAT261
Pinho
Shell, Thick
shell
Camanho
Shell, Thick
shell
MAT262
Laminated Fracture Daimler
Camanho
5.2
Material model MAT54 is one of the most tested composite material definitions in LS-DYNA. It is is often
used in dynamic failure simulations since it requires a reduced number of measured material input parameters.
Material card MAT54/55 is an enhanced version of the simplified material model MAT22. MAT54 makes use
of the full Chang-Chang failure criteria where failure can occur in tensile fiber mode, compressive fiber mode,
tensile matrix mode and compressive matrix mode. In MAT55, the tensile and compressive fiber mode are
treated as in the Chang-Chang failure criterion, but for the matrix modes the Tsai-Wu criteria is used instead.
In MAT54/55 an elastic-plastic stress-strain behavior is used, allowing a more ductile failure process. The
stress is increased linearly until the failure criteria is reached in one mode. Then the stress-strain behavior is
either ideal plastic or the maximum stress is reduced to a residual stress (since version R5), depending on how
certain parameters is defined in the material card. The integration point in a certain layer is removed when a
specified failure strain is reached.
In addition to the regular material behavior, a crashfront algorithm may be added, which offers a softening
factor to the strength in the elements of the crashfront. This can be used to suppress global buckling and trim
a progressive failure. Strain rate dependent strength can also be implemented in the model, via the function
DEFINE CURVE. The material models can be used for shells, thick-shell and solids elements.
5.2.1
MAT 54/55 uses an orthortopic elastic stress-strain relation, until failure is reached in any of the four failure
modes described in Section 3.2. All failure modes are defined in the plane of the shell, since the shell definition
and MAT54/55 works under the assumption of plane stress. MAT54 makes use of the full Chang-Chang
criterion, seen in Equations 5.1-5.4, which is formulated as a separate criterion for each of the four failure
modes. MAT55 makes use of a modification of the Tsai-Wu criterion in matrix failure, see Equation 5.5, and
uses the Chang-Chang criterion in the tensile and compressive fiber mode. The terms that are dependent of
the longitudinal stress, in the Tsai-Wu criterion, are removed. Only terms dependent on the transverse and
shear stress are used in the matrix mode for MAT55.
In the degradation scheme, used in MAT54/55, the affected parameters are either undamaged or fully damaged.
When failure is reached in one of the four failure modes, some of the orthotropic elastic material parameters are
set to zero, such as the Youngs moduli, Poissons ratios and shear moduli. Which of the material parameters
set to zero, depends on the reached failure criteria. The degraded parameters are shown together with each
failure criteria in Equations 5.1-5.5. The degradation of Youngs modulus causes the stress-strain relation to
become ideal-plastic after failure, see Figure 5.2, for an explanatory stress-strain relation. If unloading occur,
the failure criteria will not be fulfilled and the degraded parameters are undamaged again.
24
In the tensile fiber mode, Equation 5.1, the effect of the shear stress is added to the failure criteria. The shear
stress weighing factor is used to explicitly define the influence of the shear stress in the fiber tension mode. If
is set to zero the maximum-stress-failure criteria is used in the tensile fiber mode and a value of one gives
the Hashin failure criteria. Since tensile and shear stresses have a less affect on the result than the compressive
stresses the parameter will not have a such a big influence of the simulation.
a
XT
2
+
ab
SC
1
0
<0
failed
if a > 0
(5.1)
elastic
a
XC
2
1
0
<0
failed
if a < 0
(5.2)
elastic
b
YT
2
+
ab
SC
2
1
0
<0
failed
if b > 0
elastic
25
(5.3)
b
2SC
"
2
+
YC
2SC
2
1
b
+
YC
ab
SC
2
1
0
<0
failed
if b < 0
(5.4)
elastic
2 2 2
0
b
(YC YT )b
ab
e2md =
+
+
1
<0
YC YT
SC
YC YT
failed
(5.5)
elastic
5.2.2
The material card for MAT54/55 has eight data block columns and nine rows. The blue variables, which
constitute most of card 1 and card 2, are elastic orthotropic material parameters for the longitudinal, transverse
and normal direction. Since the assumption of plane stress is made for MAT54 and for the LS-DYNA shell
definition, the out of-plane-stress 33 is zero. Hence the out-of-plane material parameters are not used. This
implies that MAT54 can not predict delamination, a failure mechanism highly dependent on the out-of-plane
stress.
The red variables control the material coordinate system for each element. It is essential to define the material
direction to control the anisotropic behavior of the material.
The green variables are failure modeling parameters. Card 5 consists mainly of strain parameters and unphysical
tuning parameters. The strain parameters control that the element is kept after the failure criterion is reached
and is not deleted until the failure strain is exceeded. Card 6 consists of strength limits to the Chang-Chang or
Tsai-Wu failure criterion and card 7-9 are optional tuning parameters.
Table 5.2: Material card MAT54/MAT55
Card 1
MID
RO
EA
EB
(EC)
Card 2
GAB
GBA
GCA
(KF)
AOPT
A1
Card 3
PRBA
(PRCA)
A2
A3
MANGLE
(PRCB)
Card 4
V1
V2
V3
D1
D2
D3
DFAILM
DFAILS
Card 5
TFAIL
ALPH
SOFT
FBRT
YCFAC
DFAILT
DFAILC
EPS
Card 6
XC
XT
YC
YT
SC
CRIT
BETA
Card 7
PFL
EPSF
EPSR
TSMD
SOFT2
Card 8
SLIMT1
SLIMC1
SLIMT2
SLIMC2
SLIMS
NCYRED
SOFTG
Card 9
LCXC
LCXT
LCYX
LCYC
LCSC
DT
26
In Table 5.3 there is a short description of each parameter in MAT54. The more important parameter will be
explained further in this section.
Table 5.3: Description of input parameters in MAT54/MAT55
Variable
Definition
MID
RO
Mass density
EA
EB
(EC)
PRBA
PRBC
(PRCB)
GAB
GBC
GCA
(KF)
AOPT
A1, A2, A3
MANGLE
V1, V2, V3
D1, D2, D3
DFAILM
DFAILS
TFAIL
ALPH
SOFT
FBRT
YCFAC
DFAILT
DFAILC
EFS
XC
XT
YC
Definition
YT
SC
Shear strength
CRIT
BETA
PFL
EPSF
EPSR
TSMD
SOFT2
SLIMT1
SLIMC1
SLIMT2
SLIMC2
SLIMS
NCYRED
SOFTG
Softening reduction factor for transverse shear stiffness for crashfront elements
LCXC
LCXT
LCYC
LCYT
LCSC
DT
28
The LS-DYNA theory manual [20] and keyword manual [19] do not contain detailed information on how to use
the model parameters. Thus it is important to provide additional explanations and recommendations here.
Failure strains
The strain-to-failure criterion is active if the maximum strains are defined in the material card. It is always
recommended to specify the maximum strains, since the elements are not deleted otherwise. The maximum
strains are defined for all in-plane directions: DFAILT for maximum fiber tension, DFAILC for maximum fiber
compression, DFAILS for maximum tensorial shear strain and DFAILM for maximum matrix straining. Note
that the maximum matrix strain is defined for both tension and compression. The matrix strain limit is often
different in tension and compression for ordinary composite materials, thus the choice of DFAILM must be
estimated if two of these limits are known for the material. The maximum strains have to be higher than the
failure strain in order not to affect the Chang-Chang criterion. By definition DFAILC must be negative and
all other maximum stresses positive.
Residual stresses
In addition to the ideal plastic behavior a residual stress can be defined in all directions: SLIMT1 for the
fiber tension, SLIMT2 for matrix tension, SLIMC1 for fiber compression, SLIMC2 for matrix compression and
SLIMS for shear. After the failure criterion is reached the stress will drop down to the residual stress instead of
maintaining its maximum stress level.
CRIT
The CRIT parameter in the material card controls which of the two failure criterion is used. If the Chang-Chang
criterion is used the material card is called MAT54 and MAT55 if the Tsai-Wu criterion is used. It is, at the
time of publication of this report, recommended to use MAT54 instead of MAT55, since not all of the most
recent enhancements has been implemented in MAT55 yet (version R7.0) [21].
TFAIL
Element deletion is controlled by either the Chang Chang failure criterion or the maximum strains, if these
parameters are active in the material card. The model parameters, TFAIL, offers an additional way of element
deletion. In general, highly distorted elements that do not carry any load are not deleted by the failure criterion.
To handle highly distorted elements in the analysis, a very small time step is required. The simulation will
thus be costly and the time step can decrease rapidly. As a remedy, TFAIL can be activated to take the time
step in considered for element deletion. Further more, it should be remarked that the crashfront algorithm
only works if TFAIL is greater than zero. Without the crashfront algorithm the softening reduction factor
SOFT is inactive. It is recommended to always have TFAIL active, even if additional element deletion is not
required in the simulation. TFAIL can either be defined in units of time or as quotient of the initial time step
in the simulation. Costly elements which requires a time step beneath the TFAIL condition are deleted. The
initial time step set by LS-DYNA will give a rough indication of time steps used throughout the simulation and
TFAIL should be set to a much smaller value.
tf ail 0
No element deletion by time step size.
0 < tf ail < 0.1 TFAIL is defined in units of time. Elements are deleted when their
t
0.1
TFAIL is defined as a quotient of the initial time step. Elements are
f ail
deleted when their required time step causes the quotient between
the current time step and the initial time step to be lower than TFAIL.
29
ALPH
In plane-stress, which is used with a modification in MAT54, the strain is expressed as a function of the stress
terms. An additional shear term is added to the plane-stress condition, according to Equation 5.6. The ALPH
variable is a weight factor for the non-linear shear stress term and shown as in Equation 5.6 below. The
factor could be measured experimentally in the range of 0 < 1, but is calibrated by a trial and error
approach, since no such test have been performed. In the current work, where the test specimen is crushed
axially the shear stress does not have a big influence.
1 =
1
E1 (1
1 2 )
2 =
1
E2 (2
2 1 )
212 =
1
G12 12
(5.6)
3
+ 12
SOFT
A important parameter with a great influence on the result is the softening reduction factor, SOFT. It reduces
the material strength in the elements immediately behind of the crashfront. Elements sharing same nodes
as one of the deleted elements will get reduced strength. How the strength reduction is implemented is not
specified in the LS-DYNA manuals, [18, 19]. According to Feraboli et al. in [22] the strength reduction is
implemented according to Equation 5.7.
[XT0 , XC0 , YT0 , YC0 ] = SOF T [XT , XC , YT , YC ]
where
(5.7)
0 < SOFT 1
The default SOFT value is 1, i.e no strength reduction, and the strength is reduced as the SOFT value is
lowered. The value of SOFT needs to be found by trial and error.
Figure 5.3 illustrates the spread of the SOFT parameter. In Figure 5.3a, three elements in a uniform mesh
are damaged. When the elements are removed, the crashfront algorithm will be activated and the softening
factor will affect the neighboring elements, see Figure 5.3b. SOFT is an unphysical parameter, but the strength
softening could be seen as a way to imitate the effect of delamintion and other fracture mechanisms. In crush
simulation SOFT can prevent global buckling and promote progressive crushing behavior. The reaction force
and the energy in the system are then lowered, since the crashfront elements can not maintain the same amount
of stress as their original strength.
(a)
(b)
Figure 5.3: Illustration of the how the strength softening is initiated. (a) Elements are damage and will be
deleted according the their strain limit. (b) Spread of crashfront algorithm with initiated strength softening.
The impact of the SOFT parameter is highly dependent on the shape of the mesh. With a uniform mesh,
where many elements are in contact at the same time, the crashfront algorithm will spread rapidly along an
element row and the SOFT parameter will have significant influence on the result. Elements are removed row
by row. If the mesh has a more random pattern, the crashfront algorithm will not spread so easily and the
SOFT parameter will have a reduced effect.
An optional orthogonal softening reduction factor, SOFT2, can also be defined. If SOFT2 is active, the
SOFT parameter will only reduce the strength parallel to the crash direction while the SOFT2 parameter
reduces the strength orthogonal to the crash direction, see Figure 5.4. Elements in between the SOFT and
30
SOFT2 reduction direction will be reduced by the mean value of SOFT and SOFT2. For softening reduction of
the transverse shear stiffness the SOFTG can be defined.
PFL
As default all layers of an element must fail before the crashfront algorithm is initiated and the softening
reduction factor is applied. By using the PFL parameter the crashfront algorithm can be initiated before the
element is removed. PFL specifies how many percent of the layers that must fail until strength is reduced in
the neighboring elements. As an example, PFL = 80 corresponds to 80% of the layers in the element must fail.
Adding the PFL parameter to the simulation will increase the spread of the softening strength.
Additional strength reduction parameters
In MAT54 the FBRT and YCFAC parameters can reduce the fiber strength in tension and compression in
the remaining plies, once failure has occurred in the matrix compression mode for one ply. This is a simple
way of simulating the effect of strength reduction when the matrix start to crack. The strength in the tensile
fiber direction is reduced according to Equation 5.8 and the compressive fiber strength is reduced according to
Equation 5.9.
XT0 = XT F BRT
(5.8)
XC0 = YC YC,F AC
(5.9)
The FBRT factor is defined in the range of 0 < FBRT 1. For the YCFAC factor the range is dependent
of the strength in fiber compression and matrix compression. Note that YCFAC must be defined so that the
reduced strength does not exceed the original strength in fiber compression, YC YC,F AC XC .
5.2.3
A simple test is performed to validate and demonstrate the stress-strain behavior for a single layered composite
specimen, see Figures 5.5. The laminate is constructed of a single element with MAT54 and a prescribed
motion is applied in the fiber direction. Separate tests with both tensile and compressive load were conducted.
Also the effect of defining a residual stress and the unloading behavior, was investigated.
(a)
(b)
Figure 5.5: Uniaxial simulation with one single element in MAT54. (a) Tensile load. (b) Compressive load
The stress-strain behavior for fiber tensile load are shown in Figures 5.6a and 5.6b, where the latter figure has a
residual stress defined. For the case without any residual stress, the material behaves ideal-plastic without any
work-hardening during the plastic deformation. The stress level continues to be constant until the maximum
strain is exceeded and the element is deleted. When residual stress is defined, with the parameters SLIMT1,
31
the stress will drop to the defined residual stress after failure is initiated and keep a constant stress level until
the element is deleted by the maximum strain limit.
(a)
(b)
Figure 5.6: Stress-strain behavior in tensile fiber direction. (a) Stress-strain behavior in fiber direction - Tensile
load. (b) Tensile load and residual stress (SLIMT1) defined.
In Figure 5.7, the stress-strain relation during compressive fiber load is shown, without and with residual stress.
The same ideal-plastic behavior is shown in all directions. Since the strength and failure modes are different
depending on the direction, the stress levels will differ as well.
(a)
(b)
Figure 5.7: Stress-strain behavior in compressive fiber direction. (a) Compressive load. (b) Compressive load
with residual stress (SLIMC1) defined.
A cyclic load case has also been tested, in order to investigate the unloading behavior. If the element is
allowed to unload, the failure criteria will be deactivated since the stress is decreased. The previous damaged
parameters are then set to an undamaged state, and the unloading will follow the elastic material parameters.
In Figure 5.8a the element is stretched until failure occurs in tensile fiber mode. During the ideal-plastic
stress-strain behavior, the element is allowed to unload. When the element is stretched again, the stress will
increase elastically until the element fails according to the tensile failure mode.
If the same cyclic test is carried out with a residual stress defined, the unloading will differ from the ideal-plastic
case, see Figure 5.8b. The unloading will occur with a lowered Youngs modulus, thus allowing the strain the
reach the same level as in the ideal-plastic case.
(a)
(b)
Figure 5.8: Stress-strain behavior in tensile fiber direction during unloading. (a) Unloading in a ideal-plastic
state. (b) Unloading during a state with residual stress (SLIMT1).
32
This smooth and controlled stress strain behavior can only be seen with a few elements and with a stack-up of
only one ply. If many elements interact, only some of the elements will experience damage while others will
unload elastically and never reach the maximum stress. When the stack-up is increased to a few layers, some of
the layers will reach failure before others and some of the elastic material parameters are degraded, affecting
the global element strain and stress levels. The stress and strain behavior is then chaotic with both unloading
and material degradations. The behavior is then to hard to analyze.
5.2.4
To investigate the mesh sensitivity of MAT54, a small tensile dog-bone test was performed, see Figure 5.9a.
The specimens are fixed at the left boundary and pulled with a prescribed velocity in the x-direction on the
right boundary. In the waist of the three specimens a 2x2 mm big square is placed. Each square has same
material definition, but is meshed with three different elements sizes: 4 elements with the size of 1x1 mm2 , 9
elements with the size of 0.67x0.67 mm2 and 16 elements with the size of 0.5x0.5 mm2 .
In Figure 5.9b, the deformed shape of the specimens is shown during a certain time step in the simulation,
where the specimen with the big elements in its waist is still intact while the specimens with the smaller
elements in the waist has failed. This is a clear evidence that the simulation with MAT54 is mesh dependent,
when small elements will experience failure earlier than larger elements.
(a)
(b)
Figure 5.9: Tensile dog-bone test with different element sizes using MAT54. (a) Undeformed state, before
applied load. (b) Deformed state, during applied load.
In Figure 5.10a, the absorbed energy for each square in the waist of the three specimens is shown as a function
of the displacement. The energy added to the system is increased as the specimens are stretched. Since the
measured elements in the waist are cut off from the moving boundary condition, no more energy will be added
to the system. One can easily see that the specimen with the smallest element size in the waist has failed first
at a very low energy level, then the middle sized elements and last the largest elements. The mesh dependency
is an effect of the lack of length scaling, when computing the stress strain relation for each element, further
discussed in Subsection 5.3.4. This is seen in Figure 5.10b where all three element sizes will have the exact
same stress strain relation, i.e. increase elastic until the failure condition is reached and then ideal plastic
until the maximum strain limit. Due to localization only some of the elements will take damage according to
the Chang Chang criterion. The other elements will stay in the elastic zone and unload, when the damaged
elements fail according to their maximum strain limit.
In Figure 5.10c, the force needed to stretch each specimen is shown. The stress levels and the cross-section
areas are the same for all specimens, giving the same force-displacement relation independent of element size.
However, since a large element can withstand a larger displacement before the strain limit is reached, the force
is allowed to operate over a long period compared to a small element.
33
(a)
(b)
(c)
Figure 5.10: Result from tensile dog-bone test for elements of different sizes in MAT54. (a) Absorbed energy
in each Part ID. (b) Stress-Strain behavior for elements of different sizes. (c) Force needed to stretch each
specimen.
According to these findings the energy needed to simulate a crack propagation through the specimen will
decrease, when the resolution of the mesh is refined. This mesh dependency is often noted for material models
without regularization.
5.2.5
The baseline model with MAT54 implemented, which is used to simulate the performed crushing test, is
described in the following section. The setup of the simulation model is described in Subsection 5.1.1 and this
section aims to describe more detailed material card specific issues for MAT54.
The first approach of finding a baseline model was an uniform mesh with an element size of 1x1mm2 . The
progressive failure behavior could be trimmed such that the trigger row was first to be deleted, thus initiating
the strength reduction parameter SOFT in the neighboring elements of the crashfront, see Figure 5.11a, where
the elements experiencing strength reduction is shown in red. When the strength is reduced in the element
row closest to the contact, a progressive row-by-row failure process is achieved. Since a whole element row is
deleted at approximately the same time, the laminate will loose contact with the ground. The reaction force,
measured in the contact condition, will reach a peak value when the laminate is in contact with the ground and
drop to zero when a row is deleted, see Figure 5.11b. The reaction force will then stay at zero for a short while,
until the laminate comes into contact again. This progressive row-by-row failure is achieved by calibrating the
reduction parameters YCFAC, FBRT, SOFT and maximum strain limits, by trial and error. The simulation
with the uniform mesh is very sensitive and a small change can lead to a failed simulation.
34
(a)
(b)
Figure 5.11: Result from simulations on a uniform mesh with MAT54. (a) Crashfront in an uniform mesh. (b)
The reaction force goes to zero.
The oscillatory progressive behavior, where the load is periodically dropped to zero, do not correspond to the
experimental behavior and can introduce additional shockwaves and vibration to the simulation. In an attempt
to achieve a more even crushing, the element size was reduced to 0.5x0.5mm2 . This mesh refinement however,
did not lead to an improvement in the results. Since MAT54 is mesh dependent the behavior of the coarse
model could not be transferred directly to the new finer mesh and the parameters had to be adjusted again.
The simulations with the finer mesh exhibited a higher sensitivity level and the failure could not be confined to
the contact between the laminate and the ground. Instead the failure had a tendency to propagate fast through
an element row either close to the prescribed motion at the top of the laminate, hence cutting off the boundary
condition, or cutting of a random row at the center of the laminate.
A successful attempt to get a more even progressive row-by-row failure, where the load never drops down to
zero, has been made in other work by [3] by introducing an user-defined contact. The contact is defined so
that several element rows are in contact with the ground at the same time, thus never losing contact when an
element row is removed. This is done by manually defining a reaction force as a function of the penetration in
the contact. Since this project aims to mimic industrial needs, where such contacts can not be used in crash
analysis, the user defined contact was left out. A predefined standard contact, applicable for crash simulations,
was used throughout this project.
Naturally, a simple way of maintaining a non-zero reaction force is to always have elements in contact with the
ground. By adopting an irregular discretization of the specimen, thus removing the possibilities of row-by-row
failure, a continuous progressive behavior was achieved. Tests with both a complete random mesh, attained
by using triangular elements, and an irregular mesh using quadratic elements, attained by a manual meshed
layout, was successful.
The irregular mesh with quadratic elements was used in the baseline model and its irregular mesh can be seen
in Figure 5.12. The mesh is symmetric about a vertical centerline, with each element row inclined towards the
center of the laminate. By constructing the mesh this way, it was possible to avoid that a whole element row
comes into contact with the ground at the same time. Please note however that the trigger row is kept as an
uniform mesh, since its intention is to trigger the failure behavior over the whole crushing front. In total 5266
quadratic elements was used in the model with an addition of 32 triangular elements in order to fill intricate
spaces. The element size is approximately 1x1 mm2 .
35
The numerical values used in material card MAT54 for the baseline model can be seen in Table A.1 in Appendix
A. From experience of the current project, it appears as if as long as the model exhibit a stable behavior many
of the model-parameters can be combined in a different way and still reach a good correlation to the test data.
In this baseline model the strength reduction parameters has been kept to a minimum in order not to lower the
reaction force to much. The aim of the calibration of the material parameters was to obtain a stable crushing
with periodic oscillations and low amplitude. This was achieved with the SOFT parameter set to 5% strength
reduction. A low strength reduction in comparison with the uniform mesh, which needed a strength reduction
in about 40% to achieve a progressive behavior. The FBRT and YCFAC were set to 10% and 19% fiber
strength reduction respectively, when the compressive matrix mode is activated. Note that the maximum
strains had to be set to 10% to achieve a good behavior.
5.2.6
The chosen baseline model has a continuous progressive failure behavior and a good correlation with the test
data. Illustrations of the crushing behavior in the simulation can be seen in Figure 5.13, where the elements
affected by SOFT are shown in red. Since the mesh is irregular, approximately half of the elements in contact
with ground are deleted, while the other half stays in contact, produces a reaction force that will never go down
to zero. The baseline model is much more stable than simulations with the uniform mesh. Material parameters
can be changed in a greater interval without a failed simulation, e.g. buckling or an exploded laminate.
(a)
(b)
36
(c)
(d)
Figure 5.13: Simulation with the baseline model with MAT54. (a) Before crushing. (b) Trigger row removed.
(c) Continues progressive behavior. (d) Simulation ended.
The reaction force, measured as the sum of all nodal forces in the contact, can be seen in Figure 5.14a. Except
for the initial peak, where the trigger row is deleted, the correlation with the test data is very good, even
though the oscillations in the simulation have a larger amplitude. The same goes for the energy introduced to
the system, seen in Figure 5.14b. The initial force peak causes an initial increase of the energy in the system.
After the trigger row is removed, the rate of the energy absorption is consistent with the experiment.
(a)
(b)
Figure 5.14: Result from the baseline model using MAT54. (a) Reaction force. (b) Energy absorption in the
system.
5.2.7
To further verify the mesh dependency of MAT54, a comparison of four different mesh layouts is presented
here, see Figure 5.15:
Uniform mesh layout with quadratic shell elements, with an element size of 1x1 mm2
Uniform mesh layout with quadratic shell elements, with an element size of 0.5x0.5 mm2
Random mesh layout, achieved by using only triangular shell elements, with a side length of 1.5 mm
The baseline model with its irregular mesh was added for comparison, approximately 1x1 mm2
All simulations were performed with the same setup and material parameters (baseline values). The only
difference in this investigation was the layout of the mesh.
(a)
(b)
(c)
(d)
Figure 5.15: Comparison of 4 different mesh layouts using the same material data in MAT54. (a) Uniform
Quadratic mesh 1x1mm2 . (b) Uniform Quadratic mesh 0.5x0.5mm2 . (c) Baseline model - Irregular Quadratic
mesh 1x1mm2 . (d) Triangular mesh with a side length of 1.5 mm.
Both the simulations with the uniform mesh had the failure initiated in the crushing front. However, after a
short time interval the failure starts to propagate at the top element row where the motion is prescribed, thus
eventually removing the effect of the boundary condition, as can be seen in Figure 5.15a and 5.15b. In the
simulation with the triangular mesh a continuous progressive failure was achieved.
In Figure 5.16, the reaction forces and the absorbed energies are compared for the different cases. The reaction
force for the two cases with uniform mesh is dropped to zero, as the failure has propagated through the top
row and laminate is no longer in contact with the ground. No more energy is then introduced to the system.
The model with triangular mesh shows a stiffer response than the baseline model but has a stable crushing
behavior. However, the crash front algorithm never became active during the simulation, due to the random
38
mesh. Hence, the simulation with the triangular mesh do not experience any strength reduction by the SOFT
parameter.
(a)
(b)
Figure 5.16: Comparison of the results from the different mesh layouts with MAT54. (a) Reaction force. (b)
Absorbed energy.
(a)
(b)
Figure 5.17: Result from triangular mesh with adjusted parameters in MAT54. (a) Reaction force. (b) Absorbed
energy.
39
5.3
MAT262 is one of the most advanced composite material model in LS-DYNA and utilizes a physically based
failure criterion. The failure criterion is based on theory from Camanho et al. [23, 24] which is a continuum
damage model where the constitutive relation is degraded with an exponential evolution law. However, MAT262
exhibits a linear or bi-linear degradation and the implementation of the degradation scheme is not explained
in the theory manual. The degradation is activated by a set of damage activation functions, working in four
failure modes described in Section 3.2. When an activation function is active the degradation law start to act
in the corresponding direction.
A bi-linear, for fiber direction, and linear, for matrix direction, softening evolution is available for the stress
strain behavior. For shear, a non-linear behavior with hardening can be defined. The softening evolution
is based on fracture toughness, i.e. the energy needed for a material containing a crack to resist fracture.
In addition a crashfront algorithm with a softening strength parameter can be implemented, as in MAT54.
MAT262 is compatible with shells, thick shells and solids.
5.3.1
The failure model used by MAT262 is an orthotropic continuum damage model. The failure mechanisms are
represented with a set of scalar damage variables in each ply. When failure is activated, the scalar damage
parameters are reduced exponentially, in an interval of dM [0, 1], where (M = 1+, 1, 2+, 2, 6), dM = 0
is undamaged and dM = 1 is fully damaged. It should be remarked that the evolution of damage in each
failure mode is governed by a so-called smeared crack model, activated by the activation functions mentioned
above. The basic underlying assumption is that failure occurs in a cohesive manner forming a crack along
a critical failure plane, the orientation of which is determined by the angle that maximizes the respective
activation function. This cohesive behavior is then smeared of the entire element by scaling the associated
fracture toughness with the characteristic element length as discussed in Subsection 5.3.2 below. For complete
details of the model, please refer to [23, 24].
The damage variables, d1 + and d1 , are associated with tensile and compressive fiber failure, whereas the, d2 +
and d2 , damage parameters is associated with the transverse matrix cracking due to tensile and compressive
load. The shear damage parameter, d6 , damage variable is influenced by longitudinal and transverse cracks.
The contribution from tensile and compressive damage, for each failure mode, can never occur simultaneous
according to Equation 5.10. The damage is either due to tensile load or due to compressive load for each mode.
h11 i
h11 i
+ d1
|11 |
|11 |
h22 i
h22 i
+ d2
d2 = d2+
|22 |
|22 |
d1 = d1+
(5.10)
The damage parameters are included in the compliance tensor in the constitutive law, thus reducing the
material parameters and affecting the global material properties. In Equation 5.11, the constitutive law is
shown, which couples the effective stress tensor,
, to the strain, . The damage parameters are included in the
lamina compliance tensor represented in Voigt notation in Equation 5.12.
= H1
0 :
(5.11)
Where
1
(1 d1 )E1
12
H=
E1
21
E2
1
(1 d2 )E2
40
0
0
1
(1 d6 )G12
(5.12)
If the fully damaged state is reached for one of the failure modes the integration point is removed for that ply.
Damage occur when the activation functions M (M = 1+, 1, 2+, 2, 6) are activated, see Equation 5.15 to
5.19. The activation functions represent each of the four failure modes described in Section 3.2.
The activation function for the tensile fiber mode is derived from a maximum allowable strain criterion, with
the effective stress tensor
computed from the constitutive law.
The activation function for compressive fiber failure assumes that a kink band is triggered, under the circumstance
that the lateral support given from the surrounding matrix to the fibers, has failed due to the compressive
stress. The fiber misalignment is caused by the kink band formation and is included in the failure condition
by the stress tensor,
m , transformed into the angle, , of the fiber misalignment, see Equation 5.16. An
illustration of the fiber kinking and the stress represented in the fiber misalignment, can be seen in Figure 5.18.
11 12
22
XT
(5.13)
(5.14)
>1
1
1
1
m
m
|
12 | L
22
=
SL
failure activated
(5.15)
no failure
22
=
11 sin2 (C ) +
22 cos2 (C )cos(C ) 2|
12 |sin(C )cos(C )
m
12
= (
22
11 )sin(C )cos(C ) + |
12 |(cos2 (C ) sin2 (C ))
(5.16)
1
r1
exp[A1 (1 r1 )]
41
(5.17)
(1 g)
22
+g
YT
22
YT
2
+
2+
|
12 | L
22
=
SL
12
SL
2
>1
2+
1
2+
failure activated
for
22 0
for
22 < 0
(5.18)
(5.19)
no failure
1
exp[A2+ (1 f2+ (r2+ ))]
f2+ (r2+ )
(5.20)
v
u ef f !2
u
=t T
+
ST
Tef f
ST
>1
2
1
2
!2
failure activated
(5.21)
no failure
(5.22)
1
exp[A2 (1 r2 )]
r2
(5.23)
Where the additional variables, stated in [23, 24], used in the fracture model are:
XT , XC , YT , YC , SL are the material strength in all directions
E
dL
1+ , d1+ are auxiliary damage variables
42
5.3.2
The material card for MAT262 has eight data block colons and seven rows, presented in Table 5.4. The blue
parameters are elastic orthotropic material parameters for the longitudinal, transverse and normal direction.
The red parameters control the material coordinate system for each element.
The green parameters are model parameters. Flags to control the failure behavior are defined in card 2.
Strength parameters and fracture toughness for different directions can be found in card 4 and 5. Card 6
consists of the strength and the fracture toughnesses defining the bi-linear behavior in the fiber direction. The
non-linear shear behavior and strength softening parameters are defined in card 7.
Table 5.4: Material card MAT262
Card 1
MID
RO
EA
EB
EC
PRBA
PRCA
PRCB
Card 2
GAB
GBA
GCA
AOPT
DAF
DKF
DMF
EFS
Card 3
XP
YP
ZP
A1
A2
A3
Card 4
V1
V2
V3
D1
D2
D3
MANGLE
Card 5
GXC
GXT
GYC
GYT
GSL
GXCO
GXTO
Card 6
XC
XT
YC
YT
SL
XCO
XTO
Card 7
FIO
SIGY
ETAN
BETA
PFL
PUCK
SOFT
In Table 5.5 there is a short description of each parameter in MAT262, cf. [25]. The more important parameter
will be explained further in this section.
Table 5.5: Description of input parameters in MAT262
Variable
Definition
MID
RO
Mass density
EA
EB
EC
PRBA
PRBC
PRCB
GAB
GBC
GCA
AOPT
DAF
Flag to control if an integration point should fail based on longitudinal (fiber) tensile failure or not
DKF
Flag to control if an integration point should fail based on longitudinal (fiber) compressive failure or not
Continued on next page
43
Definition
DMF
Flag to control if an integration point should fail based on transverse (matrix) failure or not
EFS
XP, YP, ZP
A1, A2, A3
V1, V2, V3
D1, D2, D3
MANGLE
GXC
GXT
GYC
GYT
GSL
GXCO
GXTO
XC
XT
YC
YT
SL
Shear strength
XCO
XTO
FIO
SIGY
ETAN
BETA
PFL
PUCK
SOFT
44
GM
,
l
M = 1+, 1, 2+, 2, 6
(5.25)
For quadratic shell elements, the characteristic length is defined according to Equation 5.26. The characteristic
length is dependent on the area associated with the integration points of the element, AIP , and the angle of
the mesh line with the crack direction, .
l =
AIP
,
cos()
where
|| 45
(5.26)
The dissipated energy is used to define an area beneath the stress strain curve, thus introducing an element size
dependency to the stress strain relation. The fracture toughness is defined separately for each failure mode. A
bi-linear stress strain behavior can be defined for both tensile and compressive fiber failure. Two values for the
fracture toughness is then defined e.g. GXT and GXT O, giving two areas beneath the stress strain curve, see
Figure 5.19a. A inflection point stress, XT O is then required to control the boundary between the two areas.
For modeling matrix failure a linear stress strain softening behavior is used, thus only one fracture toughness
value is needed, see Figure 5.19b. The in-plane shear exhibits a ductile behavior and can be defined with a
yield stress, SIGY , a tangential stiffness, ET AN , and a hardening parameter, BET A, see Figure 5.19c.
(a)
(b)
(c)
Figure 5.19: Stress strain behavior in MAT262. (a) Fiber direction. (b) Matrix direction. (c) Shear direction.
45
2EM GM
,
2
XM
M = 1+, 1, 2+, 2, 6
(5.27)
5.3.3
The uniaxial test performed in Subsection 5.2.3, see Figure 5.5, to validate and demonstrate the stress-strainbehavior for a single layered composite specimen is repeated for MAT262. The laminate is constructed of an
single element with MAT262 and a prescribed motion is applied in the fiber direction. Separate tests with both
tensile, compressive and shear load was conducted for two different element sizes, 1x1mm2 and 2x2mm2 . The
same materials was used for both element sizes.
The stress strain behavior in the fiber direction for both tension and compression are shown i Figure 5.20. A
bi-linear stress strain relation has been defined with four values for the fracture toughness in the fiber direction.
Two values in fiber tension and two in fiber compression are needed. When the failure is activated, the softening
evolution begins. The stress is reduced linearly until the inflection point stress is reached and the softening
evolution changes its gradient. As expected a distinct difference can be seen between the two element sizes
caused by the length scaling of the fracture toughness.
(a)
(b)
Figure 5.20: Stress strain behavior in fiber direction for MAT262. (a) Tensile load. (b) Compressive load.
46
The stress strain behavior for the matrix direction only requires two fracture toughness values, one in tension
and one in compression. The linear softening evolution is much easier to control than the bi-linear behavior in
fiber direction, since only one parameters control the whole behavior. Also in this direction, a distinct difference
can be observed between the small and the big element model, see Figure 5.21.
(a)
(b)
Figure 5.21: Stress strain behavior in matrix direction for MAT262. (a) Tensile load. (b) Compressive load.
The shear strain exhibits a similar behavior as the matrix direction, except for the linear plastic behavior after
the yield stress is reached.
47
The critical elements size still gives a indication of the allowable maximum size of the elements and the risk for
snapback increase with large elements.
(a)
(b)
Figure 5.23: Illustration of snapback, even if the critical element size condition is fulfilled. (a) Snapback, due to
small GXT . (b) Snapback, due to small GXT O .
The length scaling, used in MAT262, prevents a mindset based on failure strain, since the failure strain changes
for different element sizes and fracture toughnesses.
5.3.4
To investigate the mesh sensitivity of MAT262, the tensile dog-bone test performed in Subsection 5.2.4 for
MAT54, is repeated. The left boundary of the specimens is fixed and a horizontal prescribed motion acts on the
right boundary. MAT262 uses length scaling which is dependent of the area associated with each integrations
point in each element. The length calculation is embedded in the FE-code and can not be controlled. Therefore,
two tests where the elements had different aspect ratios were performed, in order to examine if the length
scaling algorithm could handle distorted elements as well.
The undeformed tensile dog-bone specimens with perfectly square elements are shown in Figure 5.24a. When
the specimens are stretched one can easily see that the damage is localized to one element column, see Figure
5.24b. The remaining elements will unload elastically as the damaged element row is removed.
(a)
(b)
Figure 5.24: Tensile dog-bone test with different element sizes using MAT262. (a) Undeformed state. (b)
Deformed state, where damage is localized to one element column.
The energy needed to stretch the specimens can be seen in Figure 5.25a for the different element sizes. All
specimens break at the same time at approximately the same energy level. However, there is still a small
difference. Note that the specimen with the smallest element size have a slightly higher energy than the
specimens with the larger elements. In Figure 5.25b, the stress strain relation is shown for one of the failed
elements, which absorb all damage in each specimen. The effect of the length scaling is clearly visible. In
48
Figure 5.25c, the force needed to stretch each specimen is shown. As for the energy comparison, only a small
difference can be seen in the force-displacement relation.
(a)
(b)
(c)
Figure 5.25: Result from tensile dog-bone test for elements of different sizes in MAT262. (a) Energy Displacement in each dog-bone. (b) Stress-Strain behavior. (c) Force needed to stretch each specimen.
The simulation is repeated with elements with bad aspect ratio. Each element is elongated through the whole
waist of the specimens, see the undeformed specimens in Figure 5.26a. In Figure 5.26b, the specimens are
shown in a deformed state. Note that the two specimens with smallest elements have broken while the specimen
with the largest elements is still active.
(a)
(b)
Figure 5.26: Tensile dog-bone test with different element sizes with bad aspect ratio in MAT262. (a) Undeformed
state. (b) Deformed state, where the smaller elements have been deleted.
In Figure 5.27a, the energy needed to stretch each specimen is shown. The difference is larger between the
element with bad aspect ratio than in the simulation with perfectly squared elements. Once can also see that
the specimen with largest elements requires a higher energy to break than the specimens with smaller elements.
This corresponds to a behavior as if no length scaling would have been used, since larger element requires
higher energy to be stretched. A comparison between the two simulations shows that approximately 40%
49
lower energy is required to break the specimens with worse aspect ratio than those with squared elements.
In Figure 5.27b, the stress strain relation is shown for the deleted elements. Only a small difference between
the smaller and larger elements can be seen. The length scaling is not satisfactory for elements with bad aspect
ratio.
(a)
(b)
Figure 5.27: Result from tensile dog-bone test for element with different size and aspect ratio in MAT262. (a)
Energy - Displacement in each dog-bone. (b) Stress-Strain behavior.
One can conclude that the theory presented by Camanho et al. works well for different perfectly squared
element sizes, but the length scaling algorithm do not work satisfactory for badly sized elements. However, this
is only tested with tensile loading in the fiber direction. Additional testing is required, e.g. with an angled
fracture plane, but this is out-of-scope of the current work.
5.3.5
The simulation model used in the baseline model for MAT54, see Subsection 5.2.5, is used for simulations
with MAT262 as well. With this model, two aspects have been varied: the material card parameters and
different mesh layouts. Both simulations with a uniform mesh layout using 1x1 mm2 elements and the irregular
mesh layout used in the baseline model of MAT54, have been performed. Values of the fracture toughness are
the main parameters to vary in the material card of MAT262 in order to trim the behavior, since no other
additional parameters except for shear behavior and one softening strength parameters are available. It has
been observed that generally, a high fracture toughness gives a stiff response with risk for buckling. A too
low fracture toughness on the other hand gives crack propagation through element rows that are not at the
crushing front. Values for the parameters used in MAT262 can be found in, Table B.1 for the uniform mesh
and Table B.2 for the irregular mesh, in Appendix B.
Data on fracture toughness are hard to find, even though the behavior has been investigated by a vast number
of researches, cf. [10]. A large number of aspects may effect the fracture toughness in a laminate. Variables
such as the constituent properties, fiber volume fraction, fiber-matrix interface strength, laminate configuration,
stacking sequence, test temperature and fabrication procedures, cf. [10]. Because of the large number of
variables involved, the test result for one type of laminate is not necessary transferable to another laminate.
Data corresponding to the used composite material could not be found. Values of the fracture toughnesses
were arbitrarily chosen. These values fulfill the critical length criterion in Equation 5.27.
5.3.6
Simulations with MAT262 has shown a promising behavior, but since all fracture toughnesses are unknown, the
predictability for MAT262 can not be commented. The model for MAT262 was tested with different meshes,
one with uniform layout and one with irregular layout. Both cases could be trimmed such that the results
correlate well with the test data.
50
The simulations with the uniform mesh can achieve either a brittle progressive row-by-row failure behavior or a
ductile folding behavior, depending on how the fracture toughness parameters are defined. Generally, a low
fracture toughness gives a brittle response since the failure strain is lowered. On the same basis a high fracture
toughness gives a more ductile behavior since the allowable strain is increased. Both the amount of energy an
element can absorb and when a element is deleted, are controlled by the fracture toughness.
In the fiber direction, the stress behavior can not be controlled solely by increasing or decreasing the fracture
toughness, since three parameters interacts. The sum of, e.g GXT and GXTO, control the amount of energy an
element can absorb. If GXT has a large value and the inflection point is high, a high stress level is maintained
in the begin of the stress-strain relation. GXTO will get a lower proportion of the total fracture toughness
and the failure strain will be shortened. Therefore a rather brittle behavior is achieved, where elements are
removed early. If the material card is defined such that both GXT and the inflection point stress have low
values, the GXTO will get a bigger proportion of the total fracture toughness. The stress level will decrease
rapidly. However, higher failure strain is allowed. Observations have shown that a more ductile behavior is
achieved when the elements are kept longer in the simulations. The fracture toughness in fiber direction, can
thus achieve a brittle or a ductile behavior, depending on the setup in the material card.
In Figure 5.28a an example of brittle behavior is shown. Elements are deleted after a low failure strain, very
similar to a progressive row-by-row failure. In Figure 5.28b an example of a ductile behavior is shown. A
progressive folding behavior is achieved, where many of the distorted elements are not removed. The distorted
elements have been damaged and can not carry any significant load, but the elements are kept due to their
high failure strain.
(a)
(b)
Figure 5.28: Example of two different behavior in MAT262. (a) Brittle failure. (b) Ductile failure.
In Figure 5.29, the maximum compressive fiber damage is shown at the beginning and the end of the simulation.
The damage is localized to elements in contact with the ground and spreading upwards. A progressive failure
is achieved. Approximately half of the elements of a row are removed at the same time. At the end of the
simulation, the damage has spread and reached the top of the laminate. The damage at the top of the laminate
is not critical, but could increase if the simulation is run further. The simulation with MAT262 shows a stable
behavior, even if the parameters are varied in a large interval.
51
(a)
(b)
Figure 5.29: Simulation with uniform mesh, showing compressive fiber damage. (a) Damage short after trigger
row is removed. (b) Damage at the end of the simulation.
The mean value of the reaction force correlates well with the test data, see Figure 5.30a. Even though the coarse
mesh is used, the reaction force never drops down to zero. Although it is very close to zero, some elements are
still in contact. The reaction force exhibits high amplitudes, with some additional lower periods where many
elements are detached from the laminate at the same time. In the stable crush zone, the energy absorbed per
unit length has a good correlation with test data, see Figure 5.30b, but the initial energy absorption should be
lowered for perfect correlation.
(a)
(b)
Figure 5.30: Result from the simulation with uniform mesh. (a) Reaction force. (b) Absorbed energy.
The simulation with the irregular mesh worked well. At the beginning all damage is localized in the trigger
row, seen Figure 5.31a, where the compressive fiber damage is shown. When the trigger row is removed the
progressive failure is initiated and damage is spread through elements closest to the ground, see Figure 5.31b.
As the simulation proceed, the damage is only localized to the elements close to the ground. However, there
is some strange damage that occur in the middle of the laminate. Elements at the top of the laminate stays
undamaged throughout the simulation.
52
(a)
(b)
Figure 5.31: Simulation with irregular mesh, showing compressive fiber damage. (a) Irregular mesh - Trigger
row is damaged. (b) Irregular mesh - Continuous progressive behavior.
The reaction force for the irregular mesh, see Figure 5.32a, shows an even crush load. The amplitude of the
oscillations is close to the amplitudes from the test data. Except for the initial high peak, the reaction force
correlates very well with the test data. The same goes for the energy curve, see Figure 5.32b.
(a)
(b)
Figure 5.32: Result from the simulation with irregular quad mesh. (a) Reaction force. (b) Absorbed energy.
A major finding is that the crashfront algorithm does not work in MAT262, thus the softening reduction factor,
SOFT, is never activated. A parameters study has been performed, where the SOFT parameter was varied
from 100% down to 50%. In Figure 5.33 the reaction force is shown for different SOFT values. Note that all
simulation curves coincide in the figure. All SOFT values give the exact same result and the crashfront can not
be seen in the history variables. The implementation of the crashfront algorithm in MAT262 must be debugged
by DYNAmore support.
53
54
6
6.1
RADIOSS
Introduction to RADIOSS
RADIOSS, which is a product of Altair HyperWorks, is a finite element solver that over the years mainly has
focused on explicit large displacement and non-linear simulations, such as crash and impact analysis in the
automotive industry as well as manufacturing processes. However, in recent years implicit analysis is also a
possibility in RADIOSS.
In this project it is only the explicit feature of RADIOSS that has been tested. The model that has been used
for the simulations has, just as for the LS-DYNA model, been built in the preprocessor ANSA version 15.0. In
some cases keywords have been implemented manually in the text file, due to the fact that ANSA is not always
fully updated with all the keywords needed in the RADIOSS model.
6.1.1
The corrugated plate, on which simulations were performed with the LS-DYNA software, was also modeled and
simulated with RADIOSS. In this section, the important features of the model are described and how it was
built. Just as for the model in LS-DYNA, this RADIOSS model can be divided into four parts, which include
the boundary conditions, contact definition, the trigger row and the composite modeling. It should be noted
that all values of parameters have been transformed to fit the following units: Stress in [MPa], forces in [N],
distance in [mm] and time in [s].
The boundary at the top of the laminate was set to move with a constant velocity, pushing the laminate
downwards towards the rigid steel plate, with the RADIOSS keyword IMPVEL. The velocity was as before
ramped up from zero to a velocity of 2000 mm/s. As discussed briefly in the LS-DYNA Subsection 5.2.5, an
increase of velocity requires that extra damping is added to the model in order to dampen out shockwaves.
The dynamic relaxation keyword, DYREL, is added to attenuate the shockwaves. This damping keyword was
the first that was tested. Preliminary simulations showed that unphysical shockwaves were greatly reduced. A
study, to investigate the influence of increasing the velocity, was performed and is presented in Appendix C.2.
Between the composite laminate and the rigid steel plate a contact interface of TYPE7 was used. This is, as
described in the RADIOSS theory manual, a general purpose contact [26]. It is based on a penalty method,
where nodes on the laminate are slaves to the master surfaces on the ground.
For modeling of the composite stack-up, with layer thickness and material orientation, property TYPE11
(SH-SANDW) is used. By specifying a vector, the user can determine which direction should be the 0 direction
for the layers. Then for each layer the angle, thickness and material id for that specific ply is determined. This
property type allows to have different material cards on each layer. In the simulations of the corrugated plate,
the same thickness and material id have been used for each layer, only varying the orientation of the fiber
direction. The shell element formulation, Ishell , that was used was of fully-integrated Batoz elements.
For fully-integrated Batoz shell elements, 4 integration points are used in-plane and 1 integration point per
layer through the thickness, which means that there in the current model is a total of 4x12=48 integration
points per element. Using the DEF SHELL keyword, flags can be activated to incorporate thickness change
(Ithick = 1) and the calculation of strains for post-processing (Istrain = 1).
The trigger row, at the bottom of the corrugated plate, was modeled to have half the thickness of the main
laminate, but the same amount of layers and material properties. With the initial mesh size of 1x1 mm2 , one
element row was used for the trigger definition.
55
6.1.2
In RADIOSS, there are several different material laws available for composites. When choosing material law,
one has to make sure that the element type and element formulation that is going to be used in the simulation
are compatible with the chosen material law. Element types that are available for composite material laws
in RADIOSS are shells, thick shells and solids, where the different types can also vary in number of nodes
per element. In extension to that, different element formulations are available with different integration rules,
such as e.g. under-integrated Belytschko elements, or fully integrated Batoz elements. For more information
regarding element formulations see [27].
When modeling composites with solid elements in RADIOSS, one element per composite ply is needed [28].
A simple composite plate, which easily could consist of more than 10 plies, would require a huge amount of
elements as opposed to when modeling the composite with shell elements. When using shell elements, only one
element is required to model all layers, where each layer of the composite is represented with one integration
point through the thickness. This will reduce the simulation time compared to solid elements. The drawback of
using shell elements is that delamination can not be modeled. The different material laws which can be used to
model composites and anisotropic materials in RADIOSS are presented in Table 6.1.
Table 6.1: Available laws for composite and anisotropic materials in RADIOSS [29]
Material Laws
Description
Elements
LAW19
Shells
Shells
Fabric
LAW58
fabric
LAW28
Shells, Solids
Honeycomb
LAW68
Cosserat medium
LAW73
Shells
LAW74
LAW50
Crushable foam
56
Description
Elements
LAW32
Shells
Hill
LAW43
Shells
Hill (tabulated)
LAW57
Shells
LAW72
Anisotropic Hill
LAW12
LAW14
Composite Solid
LAW15
Composite Shell Chang-Chang
Shells
Elasto-plastic orthotropic composites
LAW25
Composite Shell
LAW53
Foam model
6.2
The choice of material law is important depending on the application and requirements on the simulation. There
are, as mentioned, several material laws for composites dedicated to different applications. For elastic-plastic
orthotropic composite shells there are two dedicated material laws, LAW15 and LAW25. They are similar,
since their plasticity models are based on the same criterion, Tsai-Wu plasticity model. LAW15 uses a fully
coupled plasticity model with a built in Chang-Chang failure criterion. In LAW25 it is possible to define the
yield limits in the orthotropic directions in tension, compression and shear. The failure criterion in LAW25 is
based on maximum plastic work or tensile strain in the orthotropic directions. If another failure criterion is to
be used with LAW25, it has to be added separately. There are different failure criteria available for composites,
such as Chang-Chang, Hashin, Puck and Ladeveze. The reference guide for RADIOSS recommends the user to
use LAW25 over LAW15 [29], thus LAW25 is the material law that has been tested for the simulations in this
project. In the section that follows, the theory regarding the parameters and properties of LAW25 is described.
6.2.1
There are two variants of LAW25. The two variants differ in plasticity model formulation. By setting the
formulation flag, Iform , to either 0 or 1, one chooses either a plasticity model based on Tsai-Wu formulation or
CRASURV formulation, respectively. The Tsai-Wu formulation leaves the user with less parameters to define
and could be seen as a simpler model.
With the CRASURV formulation, the user can specify different plasticity hardening parameters in all directions.
This is preferable when modeling a crushing problem. For example, compression in transverse direction of a
composite ply tend to yield a lot more before failure than compression in longitudinal direction. A more ductile
stress-strain curve can then be obtained by setting lower values for the hardening parameters for the material
in compressive transverse direction. By using the Tsai-Wu formulation, the user is restricted to have the same
57
plasticity hardening parameters for all directions. Another limitation with the Tsai-Wu formulation is that
softening of the material followed by residual stresses is not possible, whereas it is possible with CRASURV.
Due to the limitations with the Tsai-Wu formulation, CRASURV formulation was chosen to be investigated and
the hardening of this formulation will be described in this section. For further reading regarding the Tsai-Wu
formualtion, the reader is referred to [30]. Table 6.2 illustrates the structure and which parameters that are
included in LAW25 with the CRASURV formulation.
Table 6.2: Material card LAW25, CRASURV formulation
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
Card 1
Card 2
E11
E22
12
Card 3
G12
G23
G31
f 1
f 2
Card 4
t1
m1
t2
m2
dmax
Card 5
Wpmax
Wpref
Card 6
Card 7
t
1y
b1
Card 8
t1
1
t1
2
t1
rs
Wpmax t1
Card 9
t
2y
b2 t
n2 t
t
2max
Card 10
t2
1
t2
2
t2
rs
Wpmax t2
Card 11
c
1y
b1 c
n1 c
c
1max
Card 12
c1
1
c1
2
c1
rs
Wpmax c1
Card 13
c
2y
b2 c
n2 c
c
2max
Card 14
c2
1
c2
2
c2
rs
Wpmax c2
Card 15
t
12y
b12 t
n12 t
t
12max
Card 16
t12
1
t12
2
t12
rs
Wpmax t12
Card 17
ini
max
dmax
Card 18
Fsmooth
Ioff
Iflag
E33
Iflawp
n1
ICCglobal
t
t
1max
c1 t
c2 t
c1 c
c2 c
c12 t
Fcut
The blue parameters in Table 6.2 on card 1-3 are parameters that consider the elastic behavior of the material
in the orthotropic directions, such as Youngs and shear moduli. The cyan colored parameters on card 3-6 are
those related to the calculation of damage in the composite and failure of the elements. The black parameters in
card 6 and 18 are related to strain rate effects, which have not been used in the simulations of this project, since
the test that is modeled was performed under quasi-static conditions. In cards 7-16, parameters which control
yield, plasticity hardening and softening of the material can be found. Green parameters are those for tension
in both longitudinal and transverse direction, red for compression in both longitudinal and transverse direction
and the magenta colored ones for shear. On card 17 are parameters to be used for a delamination damage
calculation. This is however not applicable for shell elements, since the out-of-plane stresses are required.
In Table 6.3 there is a short description of each parameter in LAW25. The more important parameters will be
described below in this text.
58
Definition
Initial density
E11
E22
12
If lag
E33
G12
G23
G31
f 1
f 2
t1
m1
t2
m2
dmax
Composite tensile strength at maximum damage. When failure has occured, schematically,
= (1 dmax )E
Wpmax
Wpref
Iof f
Total element failure criteria for shells. Determines when shell is deleted.
= 0: Shell is deleted if Wp > Wpmax for one layer
= 1: Shell is deleted if Wp > Wpmax for all layers
= 2: Shell is deleted if for each layer Wp > Wpmax or tensile failure in direction 1
= 3: Shell is deleted if for each layer Wp > Wpmax or tensile failure in direction 2
= 4: Shell is deleted if for each layer Wp > Wpmax or tensile failure in direction 1 and 2
= 5: Shell is deleted if for all layers Wp > Wpmax or tensile failure in direction 1, or if for
all layers Wp > Wpmax or tensile failure in direction 2
= 6: Shell is deleted if for each layer Wp > Wpmax or tensile failure in direction 1 or 2
If lawp
= 1: activated. In direction 1, for both tension and compression, the plastic work rupture
ij
criterion becomes the specified value of plastic work, Wpmax
, for the corresponding
ij
direction. But in direction 2 it is the lower value of Wpmax and Wpmax .
c
59
Definition
ICCglobal
Global composite plasticity parameters flag for strain rate computation. By setting to 1,
2, 3 or 4 one can let the strain rate have different influence on maximum stress limits and
maximum plastic work.
t
1y
bt1
nt1
t
1max
ct1
t1
1
t1
2
t1
rs
t1
Wpmax
t
2y
bt2
nt2
t
2max
ct2
t2
1
t2
2
t2
rs
t2
Wpmax
c
1y
bc1
nc1
c
1max
cc1
c1
1
c1
2
c1
rs
c1
Wpmax
c
2y
60
Definition
bc2
nc2
c
2max
cc2
c2
1
c2
2
c2
rs
c2
Wpmax
12y
b12
n12
12max
c12
12
1
12
2
12
rs
12
Wpmax
ini
max
dmax
Fsmooth
Smooth strain rate option flag. =0: no strain rate smoothing. =1: strain rate smoothing
active
Fcut
61
By using the CRASURV formulation a nonlinear mixed hardening is enabled. The yield surface equation is
presented in Equation 6.1 and the associated Fij factors are presented in Equation 6.2.
2
F (Wp ) = 1 = F1 (Wp )1 + F2 (Wp )2 + F11 (Wp )12 + F22 (Wp )22 + F44 (Wp )12
+ 2F12 (Wp )1 2
1
1
c + t
1y
1y
1
1
F2 = c + t
2y
2y
1
F11 = c t
1y 1y
1
F22 = c t
2y 2y
1
F44 =
(12y )2
F12 =
F11 F22
2
(6.1)
F1 =
(6.2)
Note that the CRASURV formulation is on the same form as the Tsai-Wu yield criterion. The Fij factors are
functions of the yield stresses, but the yield stresses used in equation 6.2 are functions of the plastic work.
Thus, the yield surface is updated due to the hardening of the material and is a function of the plastic work. In
Equation 6.3, the hardening of the yield limits is presented.
))
0
c
c
= 20
(1 + bc2 (Wp )(n2 ) )(1 + cc2 ln( ))
0
c
t
= 10
(1 + bt1 (Wp )(n1 ) )(1 + ct1 ln( ))
0
t
t
= 20
(1 + bt2 (Wp )(n2 ) )(1 + ct2 ln( ))
0
c
c
c
1y
= 10
(1 + bc1 (Wp )(n1 ) )(1 + cc1 ln(
c
2y
t
1y
t
2y
(6.3)
))
0
c
c
t
t
It is the 10
, 20
, 10
, 20
, 120 parameters that are the initial yield limits and input parameters in LAW25.
If F (Wp ) < 1 in Equation 6.1 the material is in an elastic state. If F (Wp ) = 1 the material is in a plastic
admissible state and the state of stress is on the yield surface. When the current stresses in calculations exceeds
the yield surface, such that F (Wp ) > 1, they are plastic inadmissible stresses and must be projected onto the
yield surface in order to satisfy Equation 6.1, see Figure 6.2.
62
Figure 6.2: If the yield surface is exceeded, stresses are plastic inadmissible and must be projected onto the yield
surface [30]
In equation 6.3 one can see that, other than the dependency of plastic work, there is also a dependency on
the strain rate, .
If a material is deformed with high velocity it can be more resilient to deform. A higher
strain rate has been showed to give higher ultimate stress in composite materials [31]. The strain rate can thus
be important to consider when using LAW25 in dynamic applications. However, that part of the expression
in equation 6.3 has been ignored in the simulations of this work by setting the c parameters to 0, since a
quasi-static problem has been modeled.
Figure 6.3 illustrates an example of how the yield surface can change.
Figure 6.3: Example of how the yield surface can change [30]
When the material starts to yield, the strain in the material can be divided into two parts: the elastic strain
and the plastic strain. The plastic strain is the strain which causes the material to be unable to go back to
its initial state. The plastic work, Wp , is defined by the work that is done during the plastic deformation. A
schematic definition of plastic work is illustrated in Figure 6.4.
63
i ti
mi ti
for i = 1, 2 and ti
(6.4)
Where mi is the maximum strain. The stiffnesses, E11 , E22 and G12 is then reduced according to equation 6.5.
0
Eii
= (1 di )Eii
for i = 1, 2
(6.5)
The stiffness reduction proceeds until the damage factor reaches its specified maximum value, dmax (default
value in RADIOSS is 0.999 for this parameter).
Apart from the tensile rupture strain limits, which is the starting point for damage calculation, there are also
initial and maximum softening strains. Initial softening strains are denoted ti
1 and maximum softening strains
ti
.
The
tensile
rupture
strains
and
initial
softening
strains
have
a
similar
effect
on the material, but instead of
2
calculating damage to complete failure of the layer, the initial softening strains makes the stress fall down on a
ti
specified residual stress, rs
. The initial softening strain activates when the stress should be decreased and the
64
maximum softening strain is the strain, from which a constant residual stress level should be kept. Figure 6.5a
illustrates a stress-strain curve in tension with softening of the material followed by tensile rupture and damage
calculation up to maximum strain. Figure 6.5b illustrates a stress-strain curve in compression with softening of
the material followed by failure due to plastic work.
(a)
(b)
Figure 6.5: Stress-strain curves in tension and compression illustrating softening and failure behavior.(a) Failure
of layer in tension. (b) Failure of layer in compression.
Deletion of elements
An increase of plastic work will increase the yield surface until the maximum plastic work is reached. Then
what happens, depends on what value the Iof f parameter is set to. It is a flag, which can be set to a integer
from 0-6. All options for the Iof f flag are described in Table 6.3. Each value gives a deletion condition for the
elements, based on whether maximum plastic work limits or maximum tensile strains are reached for the layers
of the composite. If the user sets the value of Iof f to 6, which has been used in this work, the deletion of the
element occurs if maximum plastic work or tensile maximum strain in direction 1 or 2 is reached for each layer.
According to the RADIOSS theory manual [30] it is option 5 and 6 that give the most physical behavior.
To summarize the section above, the mechanical properties after yield in the composite are determined by the
plasticity model, strain limits and plastic work limits. Hardening is determined by the hardening parameters.
Softening strains control when residual stresses should be activated. Tensile rupture strain limits control when
the stiffness should be degraded with damage factors. Plastic work limits and the maximum strains control
when an element should be deleted.
6.2.2
Just as for LS-DYNA, a simple test on a single layered composite specimen was conducted with RADIOSS and
LAW25, see Figure 5.5a and 5.5b. The specimen was subject to both tension and compression in fiber direction.
Figure 6.6a presents the stress-strain results of the specimen. Initially the response is elastic, with a linear
stress increase. When the material reaches its yield stress, the slope of the curve is determined by the hardening
coefficients b and n, according to equation 6.3 (no strain rate effects has been included). Then the maximum
stress is reached, and the material behaves ideally plastic until it reaches the tensile rupture strain, t1 . However,
the plastic behavior can of course be suppressed by setting the tensile rupture strain to a value that occurs
before the yield stress is reached and thereby obtain a brittle failure.
When the maximum strain, mi , is reached the layer has a residual stiffness determined by dmax . If all layers
has reached mi and the Iof f parameter is set to 6, the element is deleted.
Figure 6.6b presents results from a simulation very similar to the one presented in Figure 6.6a. It is the same
elastic response, and the same plastic behavior until it reaches the specified initial softening strain, t1
1 . Then
the yield surface is decreased and the stresses are lowered accordingly in a linear manner to meet the point
65
t1
where the maximum softening strain, t1
2 , and residual stress,rs , coincide. The stress is then kept constant
at the residual stress until tensile rupture strain, t1 is reached and the damage procedure starts. The layer
reaches total failure at the maximum strain, m1
(a)
(b)
Figure 6.6: Two ways to define stress-strain curves in tension for a 0 lamina. (a) With ideal plastic behavior
at maximum stress: (a1) Elastic. (a2) Hardening. (a3) Maximum stress. (a4) Tensile rupture strain (a5)
Maximum strain. (b) Softening followed by residual stress: (b1) Initial softening strain. (b2) Maximum softening
strain. (b3) Tensile rupture strain. (b4) Maximum strain.
When the material is in compression the plastic behavior after yield is the same as in tension. What differs the
response in compression from the response in tension is that in compression there is no strain limit as t1 . The
failure criterion in compression depends only on plastic work. Figure 6.7a presents the results from a simple
compression test of an element with only one layer with fibers in the 0 direction. Just as the case for tension,
the material is elastic followed by plasticity after the yield stress is reached. Then the material reaches the
maximum stress in compression and have an ideal plastic behavior. Softening of the material then occurs at
the initial softening strain and the stress decrease to the residual stress, on which it is kept until the maximum
plastic work limit is reached and then the element has failed.
Figure 6.7b shows that the softening of the material can be set to occur at any strain. Here the softening is set
before the maximum stress is reached, making the stress go down to the residual stress, until the plastic work
limit is reached.
(a)
(b)
Figure 6.7: Two ways to define stress-strain curves in compression for a 0 lamina.(a) Ideal plastic behavior
before stress softening and residual stress. Element is deleted when plastic work limit is reached. (b) Stress
softening and residual stress. Element is deleted when plastic work limit is reached.
The general behavior of the material seen in Figures 6.6-6.7, can also be achieved in transverse direction.
LAW25 works in the same way for fiber and transverse direction, but differs between tension and compression.
66
6.2.3
In LS-DYNA, the dependency of element size was investigated trough conducting a tensile dog-bone test, where
three different element sizes were used. The same numerical test has been performed for LAW25 in RADIOSS.
Three mesh sizes were tested. The waist of the specimen is of quadratic shape and the number of elements in
the waists was set to 4, 9 and 16, giving the elements at the waists same aspect ratio. The element sizes are
1x1 mm2 , 0.67x0.67 mm2 and 0.5x0.5 mm2 respectively. In all three dog-bones the same material has been
used. Figure 6.8 illustrates the dog-bones when they all have failed.
(a)
(b)
Figure 6.9: Results from tensile dog-bone test for elements of different sizes. (a) Energy-displacement for each
dog bone. (b) Force-displacement for each dog-bone.
67
6.2.4
In order to find a good response for the corrugated specimen model, various setups of the parameters in the
material card was tried. Many of the simulations gave unwanted results, and the sensitivity of the model
was significant. A small change in a parameter setup, which gave good progressive crushing behavior, could
alter the response completely. Since the only available material data was elastic parameters and strengths,
parameters such as yield limits, hardening coefficients and strain limits were roughly estimated and tested
by trial and error. Preliminary simulations was carried out, but to find a more stable model a parameter
study was performed. The sensitivity in the simulations due to parameter variations in the material model was
investigated in the parameter study. This parameter study is presented in Appendix, Section C.1. From the
results of the study, a setup of parameters which gave a stable progressive crushing behavior was found. The
setup is presented in Table C.3 in Appendix C and is henceforth called the baseline model.
In the baseline model, there are some parameters that are not used. For example, there are no softening strains
or residual stresses utilized in tension. This is due to the fact that composites usually fail in a brittle manner
in this failure mode. That is also why the yield limits in tension have been set close to the maximum stresses,
since there is normally no notable plasticity in tension. In compression however, the yield limits are set to occur
at approximately half the maximum stress levels. In shear, the yield limit is set to occur at approximately a
third of the maximum stress, due to the fact that FRPs tend to yield much in shear [10]. Softening strains and
residual stresses are set in compression.
As mentioned in Subsection 6.2.1, where the plastic work limits were discussed, there is a flag, If lawp , which
enables the user to have different plastic work limits in the in-plane directions. This flag was not used in this
work since the values of the limits were hard to tune in and no good results were obtained, which is also why
that value is left blank in Table C.3.
Illustrations of the simulation with the baseline model can be seen in Figure 6.10. The contour in those figures
indicates different levels of accumulated plastic strains, i.e where at least one of the layers in the elements have
passed the yield limit. One can see that the plastic strains are located close to the crushing front.
(a)
(b)
68
(c)
(d)
Figure 6.10: Crushing process of the baseline model. (a) Before crushing. (b) Initial crushing. (c) Mid crushing.
(d) After crushing.
Figure 6.11a shows the force-displacement plot. The results, drawn with green color, are spiky and need to be
filtered in order to get a smoother curve and a better view of the results. The results are filtered in the time
domain with a SAE600 filter, and that curve is drawn in blue. The harmonic behavior of the curve reflects
that the simulation in a stable way removes one row of elements at a time. However, the average crushing
load is higher than the test data. The test data has an average crushing load of approximately 15 kN, whilst
the filtered result curve has an average of roughly 18 kN. The energy plot in Figure 6.11b shows a too steep
result, i.e. the energy absorption per unit time (which is proportional to unit length in this case) is too high.
Although many parameter variations to get closer to test data were made (maintaining stiffnesses and strengths
as before), these are the results that came closest to test data with a stable crushing behavior.
(a)
(b)
Figure 6.11: Results from baseline simulations. (a) Force-displacement curve of the baseline model. (b) Energy
absorption of the baseline model.
69
6.2.5
To investigate the mesh dependency, different meshes were tested. A mesh with finer uniform quad elements,
one with triangular elements (sidelengths of 1.5 mm) and one mesh with irregular quad elements (approximately
1x1 mm) were tested. In these simulations, the same material parameters as in the baseline model have been
used. Figures 6.12 and 6.13 shows the force-displacement and energy curves for each of the tested meshes.
Figure 6.12: Force-displacement curves for different meshes with same material.
Figure 6.13: Energy absorption curves for different meshes with same material.
The response for the meshes is obviously very different from each other. None of the curves are near to coincide
with another, except for the energy absorption of the two with uniform meshes. In Figure 6.14 close-up pictures
of the crushing for each simulation are presented. For the uniform mesh with element size of 1x1 mm2 , in
Figure 6.14a, the elements are deleted row-by-row at the crush front. For the other simulations, especially in
the case of the finer quad mesh in Figure 6.14b, elements which are not at the crush front fails. This cause
large pieces of the mesh to be detached from the plate. Thus, the plate will lose contact in that region. This is
reflected in the high peaks and low drops of the force-displacement curve for the uniform 0.5x0.5 mm2 element
size. The energy absorption is however, as mentioned, close to the curve for 1x1 mm2 element size, see Figure
6.13, but since the behavior of the crushing differs to much, no thorough conclusion can be drawn from this.
In the case for the mesh with triangular elements some elements are detached early in the simulation, see
Figure 6.14c, but as the simulation proceeds, the crushing force is stabilized and fairly good crushing behavior
is obtained, see Figure 6.12. However, the average crushing load is higher than for all other mesh variations.
The same behavior early in the simulation as for the case with triangular elements was also obtained with the
irregular quad mesh. But for the irregular quad mesh a larger piece of mesh is detached, leaving a smaller
70
region of elements in contact. This can be seen in Figure 6.14d. In the bottom right corner a large piece is
detached along the mesh lines. As the simulation proceeds a stable crushing of the elements still in contact is
achieved, but since less elements are in contact, the force is lowered significantly, see Figure 6.12.
(a)
(b)
(c)
(d)
Figure 6.14: Figures from crushing simulations with different meshes. (a) Uniform mesh with element size 1x1
mm2 . (b) Uniform mesh with element size 0.5x0.5 mm2 . (c) Triangular mesh. (d) Irregular quad mesh.
It may be noteworthy that none of the meshes had elements deleted at the top of the plate, where the velocity
boundary condition is applied. In the simulations with LS-DYNA, this was a problem with some of the meshes.
71
(a)
(b)
Figure 6.15: Crushing simulation with triangular elements. (a) Initial crushing. (b) After crushing.
The force-displacement and energy absorption curve for the simulation with triangular elements can be seen in
Figure 6.16. The force response indicates that the crushing is about to stabilize at a slightly too high average
crushing force. Which is also observed in the energy response, which has a slightly too steep curve, compared
to test data, at the end of the simulation.
(a)
(b)
Figure 6.16: Results from simulation with triangular elements. (a) Force-displacement. (b) Energy absorption.
(a)
(b)
Figure 6.17: Crushing simulation with irregular quad mesh. (a) Initial crushing. (b) After crushing. Large
piece of elements that were detached early in simulation has stayed in the same place.
During the remainder of the simulation, the detached piece of elements stays in the place where it first was
separated from the plate and since no self contact definition was used, the rest of the plate penetrates that
piece of elements. At the end of the simulation, the whole plate is in contact with the ground. At this point,
the force is stabilized very close to the test data curve, see Figure 6.18a. The energy curve in Figure 6.18b, also
shows that the energy absorption per unit length of the simulation curve is the same as that for the test data.
(a)
(b)
Figure 6.18: Results from simulation with irregular quad mesh. (a) Force-displacement. (b) Energy absorption.
73
LS-DYNA and RADIOSS have shown to be similar when it comes to modeling the composite structure, e.g.
the stack-up of plies and fiber direction. However, the investigated material models differ. This chapter will
highlight and compare the major differences between the three material models, MAT54 and MAT262 in
LS-DYNA and LAW25 in RADIOSS, which were investigated in this project.
7.1
Since the available material data was limited to Youngs moduli, shear modulus, Poissons ratio and material
strengths, many of the material parameters for the three material models had to be estimated. However, for
the case when using MAT54, all the required actual material data was available, but many unphysical model
parameters needed to be used in order to stabilize the model. These unphysical parameters need to be tuned,
no matter how much information is known about the material.
In MAT262, additional information about the material is required, i.e the fracture toughness in all relevant
failure modes and the non-linear behavior in shear. Experimental tests are required to determine these
parameters. Even though MAT262 is a more advanced material model than MAT54, it requires less input
parameters. Apart from the SOFT and PFL parameters, all of the utilized input parameters of MAT262 have
a physical meaning. The SOFT and PFL parameters were both inactive during the simulations.
LAW25, with the CRASURV formulation, is the material model that requires the most material data. It would
require extensive testing in order to determine all of the parameters. Assumptions, such as linear response in
tension, that were made in this work, could be done in order to reduce the amount of unknown parameters.
7.2
MAT54 is the simplest of the three tested material models, as it lacks a continuum degradation scheme.
However, the abrupt changes in the stress-strain curve can have a negative effect on the stability of the model.
The theory which MAT262 is based on is the most advanced of the three material models and it requires no
unphysical parameters. However, there have been simplifications of the theory implementation to LS-DYNA.
For example, the material degradation in fiber direction has been given a bi-linear response in LS-DYNA,
whereas the theory describes an exponential degradation.
LAW25 is the only of the three material models, in which yield limits and hardening can be specified in all
directions. Of the three tested material models, LAW25 gives most freedom to specify the stress-strain curve.
However, the vast number of parameters and the coupling between them makes it complex to achieve a desired
behavior.
Figure 7.1 illustrates a schematic example of how the stress-strain response can be modeled in tension with the
three models.
74
Figure 7.1: Schematic comparison of uniaxial stress-strain curves between the material models.
7.3
Through the modeling process and results from simulations in this project, with each of the material models,
advantages and disadvantages can be stated. These are compiled in Tables 7.1 - 7.3.
Table 7.1: Advantages and disadvantages when modeling with MAT54 in LS-DYNA
Simulation with MAT54
Advantages
Disadvantages
+ Simple model
Table 7.2: Advantages and disadvantages when modeling with MAT262 in LS-DYNA
Simulation with MAT262
Advantages
Disadvantages
- Length scaling of fracture toughness can be influenced incorrectly by bad aspect ratio of elements
75
Table 7.3: Advantages and disadvantages when modeling with LAW25 in RADIOSS
Simulation with LAW25
Advantages
Disadvantages
76
Concluding remarks
This project has shown the complexity to run explicit crush simulations of composite structures. Simulations
can give good results, close to test data, but there are uncertainties if the used material model parameters
represent the real physics. This makes it uncertain whether the material models are predictive or not. For
MAT54, all the required material data was known, but the model was very unstable and could not give
satisfactory results with a uniform mesh. A different shape on the mesh did however give a more stable model.
To change the mesh to be irregular, in order to obtain stable results is unwanted.
Using LAW25 did not cause the same problems with the uniform mesh. However, the absence of required
material data made it difficult to tune the material parameters to correlate the simulation results to the test
data. When using an irregular mesh it was easier to match the results with test data.
MAT262 did not show the same mesh dependency, due to the length regularization of fracture toughness. This
way of considering fracture, appears to be advantageous.
Another uncertainty regarding the simulations, is that the instabilities in the models can not be said to solely
come from parameter sensitive material models. Instabilities can also, to a large extent, come from other model
characteristics, such as badly tuned contact definition or incorrect damping in the model.
The delamination failure mechanism is not possible to capture using only a single-shell simulation model.
However, there are good estimations of how to capture the in-plane failure behavior. Delamination has great
influence on other failure mechanisms that will occur during crushing. The results from this project indicates
that additional failure mechanisms, such as delamiantion, are required to accurately capture the behavior of
composite materials during crushing.
77
Future work
If FE-software analysis is to be used when verifying composite designs in crashworthy structures, predictability
of the material models needs to be confirmed. For this purpose, additional experimental testing should be
performed to calibrate the model parameters and simulations of the corrugated specimen should thereafter be
repeated with these parameters to better be able to assess the predictability of each model.
In order to model delamination of composites, multiple shells tied together with either cohesive elements or
tie contacts should be used. This will increase the complexity and size of the model compared to single-shell
modeling, but by only using cohesive elements between some of the plies, the size increase of the model does
not have to be too great. This type of modeling should definitely be considered and investigated in future work.
Adding solid elements, or more advanced shell elements, could also be used in order to capture out-of-plane
stresses and may improve quality of the results. This should also be investigated further.
There are, apart from the investigated macro-mechanics based material models, other software plug-ins that
are partially based on micro-mechanics. Due to lack of time, these could not be investigated in this project,
but in future works they should definitely be evaluated and compared with the macro-mechanics based models.
Even though the CPU cost might increase, the results may be more accurate.
In order to fully understand the material models that are available, simulations of specimens with other
geometries and different load cases need to be performed. Dynamical effects, such as strain rate dependency,
should also be investigated.
78
References
[1] Modelling of {CFRP} crushing structures in explicit crash analysis. Composites Part B: Engineering
60.0 (2014), 725 735. issn: 1359-8368.
[2] The Joint Advanced Materials and Structures Center of Excellence (JAMS). Crashworthiness of composites
structures: Experiment and Simulation. AMTAS FALL MEETING 2012 (2012).
[3] P. Feraboli et al. LS-DYNA MAT54 modeling of the axial crushing of a composite tape sinusoidal
specimen. Composites Part A 42.11 (2011), 18091825.
[4] M. Chatiri, T. G
ull, and A. Matzenmiller. An assessment of the new LS-DYNA layered solid element:
basics, patch simulation and its potential for thick composite structures analysis. 7th European LS-DYNA
Conference (2009).
[5] G. Belingardi and J. Obradovic. Numerical crash analysis of composite racing car front impact attenuator
by use of explicit FEM codes. Applied Mechanics and Materials 82 (2011), 290295.
[6] J. Obradovic, S. Boria, and G. Belingardi. Lightweight design and crash analysis of composite frontal
impact energy absorbing structures. Composite Structures 94.2 (2012), 423.
[7] P. Feraboli. Development of a Corrugated Test Specimen for Composite Materials Energy Absorption.
Journal of Composite Materials 42.3 (2008), 229256.
[8] Simulation of the Quasi-static Crushing of a Fabric Composite Plate. Abaqus Technology Brief TB-11QSC-1 (2011).
[9] Computational methods for predicting impact damage in composite structures. Composites Science and
Technology 61.15 (2001), 2183 2192. issn: 0266-3538.
[10] B. D. Agarwal, L. J. Broutman, and K Chandrashekhara. Analysis and Performance of Fiber Composites.
Third edition. New Jersey: John Wiley & sons, Inc., 2006.
[11] S. T. Pinho et al. Failure models and criteria for FRP under in-plane or three-dimensional stress states
including shear non-linearity. NASA Langley research center (2005).
[12] Impact-induced delamination of composites: a 2D simulation. Composites Part B: Engineering 29.5
(1998), 589 602. issn: 1359-8368.
[13] L. Grauers, R. Olsson, and R. Gutkin. Energy absorption and damage mechanisms in progressive crushing
of corrugated NCF laminates: Fractographic analysis. Composite Structures 110 (2014), 110117.
[14] A. M. Robinson and J. J. Carruthers. Energy absorption capability and crashworthiness of composite
material structures: A review. Applied Mechanics Reviews (1998).
[15] D. Hull. A unified approach to progressive crushing of fibre-reinforced composite tubes. Composites
Science and Technology 40.4 (1991), 377421.
[16] G. C. Jacob et al. Crashworthiness of automotive composite material systems. Journal of Composite
Materials 36.7 (). issn: 0021-9983.
[17] M. Okano, A. Nakai, and H. Hamada. Axial crushing performance of braided composite tubes. International
Journal of Crashworthiness 10.3 (2005), 287294.
[18] LS-DYNA Keyword Users Manual - Volume I. Version R7.0. Livermore Software Technology Corporation,
February 2013.
[19] LS-DYNA Keyword Users Manual - Volume II. Material Models. Version R7.0. Livermore Software
Technology Corporation, February 2013.
[20] LS-DYNA Theory Manual. Livermore Software Technology Corporation, John O Hallquist, Mars 2006.
[21] Introduction to Composite Material Modeling with LS-DYNA. Stefan Hartmann, Thomas Kl
oppel, Cristian
Liebold, Dynamore, 2014.
[22] ().
[23] P. Maim et al. A continuum damage model for composite laminates: Part I Constitutive model.
Mechanics of Materials 39.10 (2007), 897908.
[24] P. Maim et al. A continuum damage model for composite laminates: Part II Computational implementation and validation. Mechanics of Materials 39.10 (2007), 909919.
[25] LS-DYNA Keyword Users Manual - Volume II. Material Models. Version R7.1 (revision: 5442). Livermore
Software Technology Corporation, May 19, 2014.
[26] Radioss Theory Manual. Large displacement Finite Element Analysis Chapter 8 - Interfaces. Version 12.0.
Altair Engineering, Inc., November 2012.
[27] Radioss Theory Manual. Large displacement Finite Element Analysis Chapter 5 - Element library.
Version 12.0. Altair Engineering, Inc., November 2012.
79
Appendices
Appendix - MAT54
A.1
Card 1
Card 2
MID
RO
EA
EB
(EC)
PRBA
(PRCA)
(PRCB)
1.52E-9
127000
8140
0.02049
GAB
GBA
GCA
(KF)
AOPT
4210
4210
4210
A1
A2
A3
MANGLE
Card 3
Card 4
Card 5
Card 6
Card 7
Card 8
Card 9
V1
V2
V3
D1
D2
D3
DFAILM
DFAILS
0.10
0.10
TFAIL
ALPH
SOFT
FBRT
YCFAC
DFAILT
DFAILC
EPS
8E-9
0.1
0.95
0.9
0.10
-0.10
XC
XT
YC
YT
SC
CRIT
BETA
1470
2200
199
48.9
154
MAT54
0.5
PFL
EPSF
EPSR
TSMD
SOFT2
SLIMT1
SLIMC1
SLIMT2
SLIMC2
SLIMS
NCYRED
SOFTG
LCXC
LCXT
LCYX
LCYC
LCSC
DT
A.2
The sensitivity of the model parameters in material model MAT54 are investigated in this section. A parameter
study has been performed where the baseline model is changed with one parameter at a time, see Table A.2.
Table A.2: Parameter study - Model parameters in MAT54
Parameter:
Baseline value:
Parametric variation:
SOFT
0.95
FBRT
0.9
YCFAC
0 (Off), 2, 4, 6, 7
SLIM(T1,T2,C1,C2,S)
PFL
ALPH
0.1
0, 0.10, 0.50, 1
BETA
0.5
0, 0.10, 0.50, 1
DFAILT
0.10
DFAILC
-0.10
DFAILM
0.10
DFAILS
0.10
SOFT
Even though the crashfront algorithm is partly suppressed for the base line model, compared to an uniform
mesh, the SOFT parameter is still important. Both the reaction force and the absorbed energy are considerably
lowered as the SOFT parameter is gradually reduced from 100% to 50%, see Figure A.1. A surprising conclusion
is that the progressive behavior can be achieved even when the SOFT parameter is turned off, not possible for
the case with uniform mesh.
(a)
(b)
Figure A.1: Parameter study with different SOFT values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
II
(a)
(b)
Figure A.2: Parameter study with different FBRT values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
(a)
(b)
Figure A.3: Parameter study with different YCFAC values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
III
Residual stresses
The residual stress for different directions; SLIMC1, SLIMC2, SLIMT1, SLIMT2, SLIMS was changed all at
once in the parameter study. A lowered residual stress will lower the energy in the system compared to the
default ideal-plastic stress behavior, but not as much as expected, A.4. A too low residual stress, i.e. beneath
50% of the strength limit, will either cut of big pieces, see Figure A.5a or make the laminate explode, see Figure
A.5b.
(a)
(b)
Figure A.4: Parameter study with different residual stresses. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
(a)
(b)
Figure A.5: Two illustrations of the behavior of the residual stresses. (a) SLIM=0.1 - Laminate will explode.
(b) SLIM=0.5 - Big pieces will be cut off.
IV
PFL
The PFL parameter exhibits an unexpected behavior. When the percentage of the layers needed to fail until
the crashfront algorithm is initiated, is lower below 100% the reaction force and absorbed energy is lowered
as well. But for all reductions, 90% to 49%, the response will be exactly the same. An explanation is that
half of the plies have a 0 fiber direction and the remaining half have a 90 fiber direction. Since the strain
is measured for the whole element, all plies with the same fiber direction will be deleted at the same time,
thus the crashfront algorithm can only distinguish between 100% or 50% of the deleted plies. The value of
PFL=49% should however effect the result. An explanation could be that the value is rounded up to 50%.
(a)
(b)
Figure A.6: Parameter study with different PFL values. (a) Comparison of reaction forces. (b) Comparison of
absorbed Energies.
Failure strains
Generally, a low strain limit gives a low energy response, whereas a high strain limit increases the energy in the
system as the elements are kept for a longer time. The maximum strain limits are vital parameters to specify
correctly. In these simulations gave strain limits for, DFAILT, DFAILM and DFAILS below 5%, a catastrophic
failure, whereas the DFAILC had 9% as the lowest limit.
By turning off DFAILC the whole simulation will failed completely. For strain limits beneath 9%, big pieces of
the laminate is cut off, see Figure A.8. Comparison of reaction forces and energies for DFAILC, can be seen
in Figure A.7. If DFAILM is turned off, the FBRT and YCFAC parameters are disabled, and the reaction
force and energy will be very high, see Figure A.9, but the simulation can still be run. For a too low strain
limit, beneath 5%, the laminate will be cut into pieces. Both DFAILC and DFAILM have a big impact on the
simulation according to the energy curves and should not be turned off.
If DFAILS is turned off the simulation can be run, but with big pieces cut off from the laminate and for a
strain limit beneath 5%, the whole laminate will explode, see Figure A.11. A similar behavior has also been
observed associated to DFAILT. If this option is turned off, global buckling occurs and for a too low strain
value, big pieces are cut off from the laminate, see Figure A.13. When a good progressive failure is achieved
the different values on the tensile strain limit will not have a big affect on the result. Comparison of reaction
forces and energies for DFAILS and DFAILT, can be seen in Figure A.10 and A.12, respectively.
V
(a)
(b)
Figure A.7: Parameter study with different DFAILC values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
(a)
(b)
Figure A.8: Two illustrations of the behavior of different DFAILC values. (a) DFAICS=0.01 - Big pieces will
be cut off. (b) DFAILC=0.05 - Big pieces will be cut off.
VI
(a)
(b)
Figure A.9: Parameter study with different DFAILM values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
(a)
(b)
Figure A.10: Parameter study with different DFAILS values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
VII
(a)
(b)
Figure A.11: Two illustrations of the behavior of different DFAILS values. (a) DFAILS=OFF - Big pieces will
be cut off. (b) DFAILS=0.01 - Exploded laminate.
(a)
(b)
Figure A.12: Parameter study with different DFAILT values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
VIII
(a)
(b)
Figure A.13: Two illustrations of the behavior of different DFAILT values. (a) DFAILT=OFF - Global buckling.
(b) DFAILT=0.01 - Big pieces will be cut off.
Variation of the ALPH and BETA parameters only have a small effect on the result, compared to the other
parameters. In Figures A.14 and A.15, the result from the parametric study can be seen for ALPH and BETA
parameter, respectively.
(a)
(b)
Figure A.14: Parameter study with different ALPH values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
IX
(a)
(b)
Figure A.15: Parameter study with different BETA values. (a) Comparison of reaction forces. (b) Comparison
of absorbed Energies.
A.3
The sensitivity of the setup of the simulation model is discussed in this section. A short parameter study has
been performed to analyze the effect of the used velocity. Different approaches for the modeling of the trigger
row have also tested in order to analyze the effect on the initial peak of the reaction force. In Table A.3, the
parametric variations can be seen.
Table A.3: Parameter study - Model parameter in MAT54
Parameter:
Baseline model:
Velocity
2000 mm/s
Trigger thickness
0.50 mm
t=0.50
t=0.50
t=0.50
t=1.00
t=1.00
mm,
mm,
mm,
mm,
mm,
Strength=25%
Strength=50%
Baseline values
Strength=50%
Baseline values
The quasi-static test was performed with a velocity of 0.42 mm/s (1 in./min). Such a low crushing speed would
be extremely costly to simulate. Since no strain rate dependent parameters are defined for the material model,
vibrations and wave propagation are the only dynamic aspect affecting the simulation by an increased crushing
speed. With added damping, the assumption was that the effect of the increased velocity would be canceled
out. However, a notable difference can be seen in the result, see Figure A.16. The highest velocity, 2000 mm/s,
gives the highest response and the lowest velocity, 20 mm/s gives the lowest response. The increased velocity is
necessary to reduce the simulation time. Increased damping could reduce the dynamic effects further.
X
(a)
(b)
Figure A.16: Parameter study with different velocities. (a) Comparison of reaction forces. (b) Comparison of
absorbed Energies.
(a)
(b)
Figure A.17: Parameter study with different triggers. (a) Comparison of reaction forces. (b) Comparison of
absorbed Energies.
XI
B
B.1
Appendix - MAT262
Material card used in the simulation with uniform mesh model
using MAT262
Table B.1: Material card MAT262 with values for simulation with uniform mesh
Card 1
Card 2
Card 3
Card 4
Card 5
Card 6
Card 7
MID
RO
EA
EB
EC
PRBA
PRCA
PRCB
1.52E-9
127000
8140
8140
0.02049
0.02049
0.3
GAB
GBA
GCA
AOPT
DAF
DKF
DMF
EFS
4210
4210
4210
XP
YP
ZP
A1
A2
A3
V1
V2
V3
D1
D2
D3
MANGLE
GXC
GXT
GYC
GYT
GSL
GXCO
GXTO
100
300
60
60
60
100
500
XC
XT
YC
YT
SL
XCO
XTO
1470
2200
199
48.9
154
400
800
FIO
SIGY
ETAN
BETA
PFL
PUCK
SOFT
53
XII
B.2
Card 2
Card 3
Card 4
Card 5
Card 6
Card 7
MID
RO
EA
EB
EC
PRBA
PRCA
PRCB
1.52E-9
127000
8140
8140
0.02049
0.02049
0.02049
GAB
GBA
GCA
AOPT
DAF
DKF
DMF
EFS
4210
4210
4210
XP
YP
ZP
A1
A2
A3
V1
V2
V3
D1
D2
D3
MANGLE
GXC
GXT
GYC
GYT
GSL
GXCO
GXTO
30
40
40
40
40
100
120
XC
XT
YC
YT
SL
XCO
XTO
1470
2200
199
48.9
154
300
300
FIO
SIGY
ETAN
BETA
PFL
PUCK
SOFT
53
XIII
C
C.1
LAW 25
Sensitivity of material model parameters in LAW 25
Preliminary simulations resulted in a parameter setup which gave a fairly stable model. With this setup, a
parameter study was executed. Important parameters were varied to investigate how the response would change.
The results from the parameter study gave a better understanding of the material model and an optimized
parameter setup could be defined. This setup of parameters is the baseline model, described in Section 6.2.4,
with which mesh dependency studies were performed, presented in Section 6.2.5.
The variation of parameters that was used in the study is listed in Table C.1. Since the model that was used
before the parameter study gave a too stiff response, the general goal was to investigate how the load could be
lowered, thus many of the parameter variations are lower than than the start values.
The mesh that was used in the parameter study was with uniform quad elements with a size of 1x1 mm2 .
Table C.1: Parameter study - LAW 25
Parameter:
Tensile rupture strains
Start values:
t1 = 0.010
t2 = 0.0085
m1 = 0.016
m2 = 0.015
c1
1 = 0.012
c2
1 = 0.017
Maximum
strains
c1
2 = 0.10
c2
2 = 0.10
softening
Wpmax = 4.75
Residual stresses
c1
rs
= 570
c2
rs
= 110
t12
rs
= 50
c
1y
= 710
c
2y
= 90
t
12y
= 45
XIV
Residual stiffness
dmax = 0.914
The results from the parameter study showed that variation of a single parameter is not sufficient to obtain a
reduced force in the force-displacement plot. The response could be altered completely. Many of the variations
was however done in pairs of parameters. For example, the tensile rupture strains in both fiber and transverse
direction were varied simultaneously. The results from varying the tensile rupture strains show that the response
changes quite much, even though small variations are made. This can be seen in Figure C.1 and C.2, where it
is clear that the model is very sensitive. The larger drops in the force-displacement curves indicates that there
has been deletion or detachment of more than one row at the crush front. This behavior is unwanted.
XV
Figure C.1: Force-displacement curves for variation of tensile rupture strains, t1 and t2 .
Figure C.2: Energy-displacement curves for variation of tensile rupture strains, t1 and t2 .
In order to find a better agreement with test data, more parameter variations have to be taken in consideration.
An idea was that decreasing the yield limits would give lower response in the force-displacement curve. This,
however, only led to instabilities as well, as for the case of other parameter variations.
C.2
An important variation to investigate was the velocity of the crushing, see Table C.2. An increase of velocity
was done in order to reduce simulation time. Therefore as mentioned in Subsection 6.1.1 damping was added
to reduce the dynamic effects, introduced by the increased velocity.
Table C.2: Veloctiy variations with the RADIOSS model
Parameter:
Velocity
Start value:
2000 mm/s
Figure C.3 and C.4 shows the results from the velocity variation. The difference in response at very low
displacement can be explained by the fact the ramping (i.e. the acceleration) of the velocity was not scaled. A
better way would have been to scale the ramping proportionally. The cases where the velocity is 20 mm/s
and 2000 mm/s give fairly similar results, while the case where 200 mm/s gives a little less stable response.
With these results at hand, it seems like the velocity increase actually can be troublesome, but the increase was
XVI
highly necessary. The simulation with 2000 mm/s took about 30 minutes on 8 cores, while the simulations for
200 and 20 mm/s took 75 minutes and 12 hours respectively on 8 cores.
XVII
C.3
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
1.52E-9
Card 2
Card 3
Card 4
Card 5
E11
E22
12
Iflag
127000
8410
0.309
G12
G23
G31
f 1
f 2
4210
4210
4210
0.06
0.06
t1
m1
t2
m2
dmax
0.011
0.016
0.009
0.016
0.91
Wpmax
Wpref
4.75
Ioff
E33
Iflawp
Card 6
c
0
Card 7
t
1y
b1 t
n1 t
t
1max
c1 t
2190
0.13
0.8
2200
Card 8
t1
1
t1
2
t1
rs
Wpmax
Card 9
t
2y
b2 t
n2 t
t
2max
c2 t
48
0.13
0.8
48.9
Card 10
t2
1
t2
2
t2
rs
Wpmax
Card 11
c
1y
b1 c
n1 c
c
1max
c1 c
730
0.13
0.81
1470
c1
1
c1
2
c1
rs
0.013
0.10
600
Card 12
Card 13
Card 14
c
2y
b2
100
0.12
0.8
c2
1
c2
2
c2
rs
0.018
0.10
120
n12 t
n2
Card 15
t
12y
45
2.2
Card 16
t12
1
t12
2
0.065
0.12
50
ini
max
dmax
Card 17
Card 18
Fsmooth
b12
ICCglobal
Wpmax
t1
t2
c1
0
c
2max
c2 c
199
Wpmax
c2
t
12max
c12 t
0.3
154
t12
rs
Wpmax t12
Fcut
XVIII