Optimization of Elastomer-Metal Components with TOSCA and

ABAQUS
Dr.-Ing. Ralf Meske, Dr.-Ing. Jürgen Sauter
FE-Design GmbH, Karlsruhe
Dipl.-Ing. Matthias Friedrich
Freudenberg Forschungsdienste KG, Weinheim

ABSTRACT: To reduce time and costs in the development of elastomer-metal components while maintaining
a high product quality it is essential to use efficient optimization tools besides the already established nonlin-
ear FE analysis. FE-Design offers with the optimization system TOSCA an integrated solution for topology
and shape optimization with interfaces to major industry standard FE solvers. In combination with ABAQUS
it is now also possible to perform topology and shape optimization with moderate non-linearities. To apply
the optimization algorithms of TOSCA successfully for stronger non-linearities, an enhancement of the algo-
rithms is necessary. This task is approached in the German research project ELAnO (Entwicklung und Kon-
struktion von innovativen Leichtbauprodukten unter konsequenter Verwendung adaptierter Analyse- und Op-
timierungsmethoden) in cooperation with Freudenberg and several other companies. The life time and
durability of elastomeric bearing components will be increased by the systematic application of topology and
shape optimization with TOSCA and ABAQUS under full consideration of geometric, material and boundary
non-linearities. First results will be shown in the presentation and an outlook towards further developments
will be given.

New shapes and new materials often determine

the design of lightweight products. New materials
1 THE RESEARCH PROJECT ELANO1 require an accurate description of the material be-
haviour. A linear approach is often not sufficient and
Lightweight products are often critical with respect has to be replaced by a more realistic non-linear ap-
to their manufacturing costs. Due to the fact that a proach. The responsible dimensioning requires du-
major portion of the total costs is defined in the de- rability analysis. This includes an exact description
velopment phase, it is very important to use efficient of static and time-dependent loads and boundary
tools as soon as possible. This tools must be suitable conditions. The engineer can not rely exclusively on
for the special requirements of lightweight construc- his experience. The higher requirements for light-
tions and must provide an accurate prediction of the weight products lead to new approaches in the con-
physical behaviour of the component. Classic ma- struction, which have to be collected in design cata-
chine parts, which have not yet been investigated logues and information networks.
with respect to lightweight design, contain a high The production costs have to stay in focus despite
potential to increase their strength and stiffness and all these additional requirements, such that the suc-
to improve their dynamic behaviour by reducing the cess of these innovative products is not jeopardized.
mass of accelerated components. The success of the project can be measured in such a
Therefore the objective of the project is to estab- way that each end-user constructs products or com-
lish a continuous process chain for the development ponents which will be at least 20% lighter than the
of lighter, stiffer and more durable products, which existing design, which can still be manufactured
are nevertheless economic to produce. This will be economically and which satisfy additional require-
achieved by the consequent application of numerical ments like dynamic stiffness or durability.
methods and interdisciplinary structural optimization Comprehensive know-how in the field of product
and will be based on the development and construc- development, manufacturing, material description,
tion of several lightweight products. experiments, use of numerical methods, implementa-
tion in software and introduction of new techniques
is required for the successful realization of the pro-
1
www.elano.org
ject. The knowledge of the consortium of producers
of lightweight products, software developers and re- reads the results of the FE calculation and modifies
search institutions is extended by the establishment the FE input file according to the objective function
of a board of industrial consultants. The experiences of the optimization task until the convergence crite-
and results of the research project will be propagated rion is satisfied. The result of the optimization can
by publications and seminars, the industrial working be visualized with most common FE-postprocessors
group “structural optimization”, an associated and is available in various formats for the further
VDMA working group and the research institutions, processing in the virtual product development proc-
such that the results are made available to other in- ess [Sauter et al. 2000, Meske et al. 2001].
terested companies.
2.1 General capabilities
2 THE OPTIMIZATION SYSTEM TOSCA • Independent of finite element solvers and pre-
and postprocessing systems
TOSCA is a system for non-parametric structural • Support of continuum, shell and membrane ele-
optimization. TOSCA can perform topology and ments in the optimization
shape optimization of FE models with an arbitrary • Efficient handling of very large models
number of boundary conditions and load cases. A • Stable and fast optimization algorithms based on
parameterization of the model is not needed, which optimality criteria
reduces the modeling effort and allows greater flexi- • Optimization with an unlimited number of load
bility in the optimized structure. The optimization cases
algorithms are based upon mechanical optimality • Support of geometric and material non-linearity
criteria, which makes the optimization fast and ro- • Contact definition outside and at the boundary of
bust. the optimization domain
Structural optimization with TOSCA is an itera- • Definition of weight, stiffness and eigenvalue as
tive process. The structural response of the compo- objective function
nent is calculated in each iteration with an external • Definition of various constraints and design
FE solver. The high quality of the results is guaran- variable constraints
tied by using approved and accepted industry stan- • Support of multiprocessor solver calls
dard solvers. Further benefits for the user are that he • Supported platforms: UNIX and NT
can work with his favorite solver and that he does • Network licensing
not need additional training for a new solver. Al-
ready existing FE models can be used directly in the
optimization. Preprocessing (FE + Optimization)
TOSCA is based on the optimization kernel CA- $%$486&$( 06&3$75$1 RWKHU
OSS which is successfully used in industry since
*.inp *.par
1997 with an interface to MSC.Nastran under the
product name MSC.Construct [Bakhtiary et al. Optimization
CAD

1996]. Because several companies, which are using 726&$

ABAQUS as solver, were strongly interested in the $%$486

optimization with TOSCA, FE-Design has devel-

*.fil *.odb *.wrl *.stl
oped an interface to ABAQUS [Meske et al. 2000].
Postprocessing (FE + Optimization)
Besides common linear optimization it is now also
$%$486&$( 06&3$75$1 RWKHU
possible to perform topology and shape optimization
with moderate non-linearities.
The optimization procedure with TOSCA is Figure 1: The optimization cycle with TOSCA and ABAQUS
sketched in Figure 1. The optimization task is de-
fined in addition to the already known FE preproc-
essing in an optimization preprocessing step. This
can be just the selection of node or element groups
to specify the optimization domain, but also allows
the definition of manifold constraints for the optimi-
zation task. The optimization preprocessing can be
performed either in MSC.Patran together with the
FE preprocessing or in the GUI of TOSCA after the
FE preprocessing with any other available preproc-
essor. The result of this step is a compact parameter
file to control the optimization. The optimization it-
self is performed in an iterative process between
TOSCA and the FE solver. In each cycle TOSCA
 Figure 3 presents an example of a FEA calcula-
tion using the Freudenberg model for the rotationally
symmetric rubber spring of a hydromount under ax-
3 RUBBER MATERIAL MODEL ial compression. The deflection is large enough to
induce self-contact in the rubber contour. In Figure 4
The simpliest model to describe elastomer behavior both FEA results and measurements of the rubber
is a physically one, known as Neo-Hooke model. spring displacement are included. It clearly shows
When we assume incompressibility for elastomers – how well the deformation is approximated using the
which is a very good approximation – the material Freudenberg material model.
will be characterized by only one parameter, the Furthermore, a finite element analysis of the rub-
shear modulus G. The Neo-Hooke model, however, ber spring for typical displacements is performed to
can be used with sufficient accuracy only for small quantify the stress and strain levels in the rubber. As
medium strains. Mooney’s attempt to extend the a result of a FEA optimization, the magnification
Neo-Hooke model phenomenologically for large factor of the stress concentration versus the average
strains should be regarded criticially. Although Riv- stress level in the rubber is reduced. This signifi-
lin and Saunders had warned explicitly against a cantly increases the durability performance of the
“naive” use of the Mooney model in 1951, analyses mount.
are still frequently carried out using the Mooney
model. The reasons for this are to be found firstly in
an inadequate knowledge of the material behaviour
of elastomers, but also in the high testing effort for
later developed material models which are nowadays
integrated in various nonlinear FE codes. At Freu-
denberg, a physically based material model has been
developed. This model permits elastomers to be
characterised precisely over a very large deforma-
tion range. The model has been implemented as a
user subroutine in ABAQUS.
For the accuracy of the FEA results, the choice of
an adequate material law for the description of the
complex nonlinear behavior of the rubber is essen-
tial. As it can be seen in Figure 2, the load-
deflection curves calculated with the material laws
Neo-Hooke, Mooney-Rivlin and the Freudenberg
model for the case of a test specimen under tension
demonstrate significant differences, especially for
large deflections. However, large deflections com- Figure 3: Rubber spring at 20mm deflection
monly occur in front wheel drive vehicles, in which
case the mounting system has to carry a maximum
torque exceeding 2000 Nm.
3
Experiment
Neo-Hooke
Mooney - Rivlin
2 Freudenberg Model

Stress
[Mpa]
1

0
-0,5 0 0,5 1 1,5 2
Strain [100 %]

-1

Figure 2: Comparison of different material models Figure 4: Comparison between measured and computed load
deflection curve
 should not be changed too much, a constant volume
The use of a correct material model results in a cor- constraint was defined. The optimization yields a
rect stress distribution and stress level. Accurate re- significantly improved design after 10 optimization
sults for the stress calculation are necessary for cycles. The modified geometry is shown together
shape optimization because the in shape optimiza- with the maximum equivalent stress of both load
tion local responses of the structure are considered. cases in on the right side. The deformation of the ini-
tial and the optimized component is shown for both
load cases in Figure 6. The maximum equivalent
4 SHAPE OPTIMIZATION OF A 2D stress was reduced in both load cases for 26% each.
HYPERELASTIC SUPPORT The folding of the component in the compression
load case was prevented by the geometry change.
A substantial amount of the developed parts at the The capability to use own material laws which
Freudenberg Group are rubber-metal components. provide a high accuracy for even very high strains
The computation of rubber components inhibit in which often occur in rubber parts is seen as a high
general several non-linearities. Because of the large extension for the optimization of rubber-metal com-
deformation geometric non-linearity has to be taken ponents.
into account. Due to the incompressibility of the ma-
terial hybrid elements have to be used. The hypere-
lastic material law itself is non-linear. Sometimes
contact problems occur, which lead to non-linear
boundary conditions. These complexities imply that
accurate calculations for these components can only
be performed with a solver like ABAQUS which is
specialized for non-linear problems.
The Freudenberg Group has been using
MSC.Construct together with MSC.Nastran for
structural optimization for several years. Extensive
investigations of a semi-automatic linearized optimi-
zation with MSC.Nastran using a linear-elastic mate-
rial as substitute have been made, but did not lead to
an acceptable result in accuracy. The overhead for
the linearization of the problem and the nonlinear
control calculation of the optimization result was
very time expensive [2]. Therefore, there was a
strong interest to perform the optimization with
TOSCA in combination with ABAQUS.
Due to the integration of ABAQUS in the optimi-
zation with TOSCA the user now can take advantage
of the analysis capabilities of ABAQUS in the opti-
mization process with TOSCA. The shape optimiza-
tion of the following example of an hyperelastic
support was performed without difficulties in the
same way as with a linear-elastic material. The hy-
perelastic material behavior was modeled in this
case with the Freudenberg material law. The initial
configuration of the component is shown in Figure 5
on the left side. Due to the symmetry of the compo-
nent only the right half was modeled. The compo-
nent is fixed on the lower right side. In the radius
there is a steel ring which is not shown in the fig-
ures. A tension and compression load in y-direction
was applied on the steel ring. The compression load
led to a folding of the component in the initial de-
sign, which had a negative influence on the life span.
The nodes on the lower left contour of the com- Figure 5: Equivalent stress and geometry of initial and opti-
ponent were chosen as design nodes of the optimiza- mized design
tion. The optimization target was the minimization
of the von Mises equivalent stress along the design
nodes. Because the total stiffness of the component
 Figure 6: Deformation and equivalent stress of initial and optimized design.

First, relevant boundary conditions regarding the

component optimization and shape optimization
must be identified. In this case the axial load is criti-
cal (dark spot in, left side).
The optimization result calculated with TOSCA
5 SHAPE OPTIMIZATION OF A 3D RUBBER-
shows a more homogeneous stress distribution along
METAL COMPONENT
the component surface. The stress level could be re-
duced for 35%.
An experienced design engineer may be able to
Fig. 8 compares the different geometries of initial
modify a single ore multiple loaded 2D structure
and optimized design. The changes in both cross-
manually by modifying the geometry and calculate
sections of the component are very small, but with
the modified design. For 3D examples this is more
this change of the geometry the effect on the maxi-
complicated because first of all the preprocessing ef-
mum stress level is very high.
fort normally is higher. In second place the calcula-
tion time is higher, too. So, as it is not possible to
design only axisymmetric or extrusion parts, the in-
terest to have a high efficient optimization program
6 CONCLUSIONS
which is able to handle more complex 3D structures
is very high. To save manpower and calculation time
The use of TOSCA as non parametric optimization
it is necessary to have available an optimization pro-
program in combination with ABAQUS opens a new
gram which can handle 3D real world structures.
possibility for the solution of the optimization of
Figure 7 shows a rubber and metal component as an
nonlinear FE models. By now, shape optimization
example for 3D automatic shape optimization with
can be performed for “moderate” nonlinear prob-
TOSCA. The rubber and metal bearing doesn’t have
lems. The first optimization examples result in a
a rotationally symmetric geometry because of the
high stress reduction and will result in a higher life-
nonsymmetric holes.
span. The use of user defined material models espe-
The cross-section of the optimized model has to
cially for the description leads to accurate FE results
be the same over all the model. The border of the
even for large strains. For highly nonlinear
holes has to stay perpendicular.
 Initial Design Optimized Design

100% 65%

Figure 7: Deformation and equivalent stress of initial and optimized design.

Initial Design

Optimized Design

Figure 8: Geometry comparison of initial and optimized design.

Meske, R., Sauter, J., Güngör, Z. 2001. Recent improvements

problems with self contact and high deformations in topology and shape optimization and the integration into
the virtual product development process. NAFEMS World
the controller which TOSCA is based on has proba- Congress 2001, 24.-28. April 2001, Como, Italy.
bly to be adapted. This also will be necessary for the Sauter, J.; Fricke, H.; Güngör, Z.; Himmler, G.; Hougardy, P.;
optimization of contact regions. The adaption of the Lauber, B.; Müller, O.; Neithardt, W.; Schirrmacher, R.
controller is one of the objectives of the German re- 2000. Integrierte Topologie- und Gestaltoptimierung im
search project ELAnO. virtuellen Produktentstehungsprozess – Einbindung in die
Further optimization potential is seen from Freu- iViP Architektur und industrielle Anwendung. VDI Berichte
Nr. 1559.
denberg in the topology optimization of models
which contain nonlinear rubber material at least in
the non design regions.

REFERENCES

All publications are available as pdf from www.fe-design.de.

Bakhtiary, N; Allinger, P; Friedrich, M; Mulfinger, F; Sauter,

J; Müller, O; Puchinger, M. 1996. A new Approach for Siz-
ing, Shape and Topology Optimization, SAE International
Congress and Exposition 1996, 26.-29. February 1996, De-
troit/Michigan.
Meske, R., Sauter, J., Friedrich, M. 2000. Topologie- und Ge-
staltoptimierung mit CAOSS und ABAQUS, ABAQUS An-
wendertreffen, 28.-29. September 2000, Winterthur.