Professional Documents
Culture Documents
= =
, [2]
where F
ij
, here in indicial form, is commonly known as the deformation gradient tensor,
or simply the deformation gradient, F, in matrix form.
The right Cauchy-Green tensor, C, is obtained from the deformation gradient such that
ij mi mj
C F F = , [3]
and is used in this context to define the following strain invariants:
2 2 2
1 1 2 3
tr( ) I = + + = C ,
( )
2 2 2 2 2 2 2
1
2 1 2 2 3 3 1 1 2
tr( )
2
I I = + + = C ,
2 2 2 2
3 1 2 3
det( ) I J = = = C , [4]
where the
i
s are the principal stretch ratios. Note that I
3
is a volumetric constant and, in
the context of incompressibility, is always equal to unity, leading to a useful relationship
between the stretch ratios:
1 2 3
1 = . [5]
It is also necessary to clearly understand the relation between nominal (a.k.a.
engineering) strain and Hencky (a.k.a. natural or logarithmic) strain, since the
experimental data presented herein is measured in nominal strain, whereas ANSYS result
files for nonlinear analysis provide Hencky strain. The principal engineering (or
nominal) strain,
e
i
, can simply be defined as the change in reference length,
o
L L L = ,
over the reference length, L
o
, and can be obtained directly from the principal stretches as
follows:
1
e
i i
o
L
L
= = . [6]
The principal logarithmic strains,
n
i
, can also be obtained from the principal stretches,
such that
( ) ln
n
i i
= , [7]
A constitutive material law is said to be hyperelastic if it is defined by a strain energy
function. The strain energy, W, is usually expressed as a function of strain invariants,
W(I
1
, I
2
, I
3
), or principal stretch ratios, W(
1
,
2
,
3
), depending on the hyperelastic
model. The principal Cauchy stresses and the principal 1
st
Piola-Kirchhoff (a.k.a,
nominal or engineering) stresses for an incompressible material can be derived from the
strain energy function as follows:
i i i i
i
W
S p
= =
, [8]
or
1 2
1 2
i i i i
i i
I I W W
S p
I I
= = +
, [9]
where p is a necessary hydrostatic pressure term due to incompressibility.
3. CONSTITUTIVE LAWS OF HYPERELASTICITY
Rivlin (8) obtained a generic strain-energy-based model for incompressible
hyperelasticity, commonly referred to (namely in ANSYS) as the polynomial model,
which takes the following form:
( ) ( )
1 2
, 1
3 3
i j
ij
i j
W C I I
=
=
. [10]
The Neo-Hookean model, discussed by Treloar (9), can in fact be obtained from Rivlins
general formulation with
( )
10 1
3 W C I = , [11]
taking
10
1
2
C NkT = , where k is the Boltzmann constant, N is the number of chains of
molecules per unit volume and T is the absolute temperature. Mooneys (10) earlier work
can also be expressed in terms of Eq. [10], resulting in what is generally accepted today
as the Mooney-Rivlin model. Yeoh (11) suggested yet another model which can be
derived from Rivlins (8) generic formulation, under the assumption that the second strain
invariant, I
2
, is constant with stretch (12) and thus does not contribute in the strain energy
function.
( )
0 1
1
3
N
i
i
i
W C I
=
=
. [12]
Finally, some outstanding contributions should also be mentioned, namely by Ogden (13,
14), Arruda and Boyce (15) as well as Gent (16) who have each proposed hyperelastic
models known by their respective names.
4. CHARACTERISATION AND CURVE FITTING
Standard tests have been developed to measure the stress-strain response of rubber under
states of simple stress. Probably the most popular and widely used is the uniaxial tension
test, mainly for its simplicity. Another test, designed to overcome the difficulties involved
in performing a purely uniaxial compression experiment, is the equal-biaxial extension
test. Also designed to prevail over experimental complications, the planar tension test
may be used, when incompressibility is assumed, to simulate pure shear. A schematic
representation of these three states of simple stress is given in Fig. 4. Assuming
incompressibility, the stretch ratios are determined from Eq. [5] for each stress state.
The general procedure followed by Axel Products Inc. for all three aforementioned states
of simple stress is fundamentally the same. The loading is performed at a quasi-static
strain rate of 0.01 s
-1
, this rate is chosen to minimize thermal effects due to internal heat
generation. Each sample is first loaded to 5% nominal strain, then completely unloaded.
This cycle is repeated 9 more times in order to precondition the testpiece. By the 10
th
cycle, the sample has stabilised and is considered to be preconditioned for 5% maximum
strain. The same procedure is repeated on the same specimen for nominal strains of 10%,
15% and 25%. Three samples were tested in each strain state. A sample of the collected
nominal stress-strain data for a testpiece in uniaxial tension is given in Fig. 5(a). Note
how this figure reproduces the combined behaviour shown in Figs. 2 and 3 of Section 1.
Fig. 4: Schematic representation of the three states of simple stress: (a) uniaxial
tension (
1
=,
2
=
3
=
-1/2
,
1
= ,
2
=
3
= 0), (b) equi-biaxial extension (
1
=
2
= ,
3
=
-2
,
1
=
2
= ,
3
= 0) and (c) planar tension (
1
=,
2
= 1,
3
=
-1
,
1
= ,
2
0,
3
= 0), the triangles indicate fixed supports in the X
2
direction
while the system remains free in the other two directions.
The last cycle for each maximum nominal strain, i.e. the 10
th
, 20
th
, 30
th
and 40
th
loading
cycles, corresponds to a stabilized path for the respective 5%, 10%, 15% and 25%
maximum nominal strains. Four preconditioned loading curves are clearly shown in Fig.
5(a). Note that the virgin stretch curve is exact up to 5% nominal strain; it is then
approximated to the maximum stress-strain data point for each subsequent stretch cycles.
For the virgin stretch path to be exact, a separate testpiece should be stretched up to the
desired strain. Based on Fig. 2, one could expect the virgin stretch curve of Fig. 5(a) to be
slightly above the data for the 10%, 15% and 25% nominal strain cycles.
0
0.5
1
1.5
2
0 0.1 0.2 0.3
Eng. strain,
E
n
g
.
s
t
r
e
s
s
,
S
(
M
P
a
)
0
0.5
1
1.5
2
0 0.1 0.2 0.3
Eng. strain,
Fig. 5: (a) Original data (in grey) for the uniaxial tension test showing
preconditioning at 4 different strains: 5%, 10%, 15% and 25% nominal strain.
The preconditioned loading cycles (thicker lines) are extracted for manipulation.
(b) Manipulated preconditioned loading cycles. The dashed lines show the virgin
stretch curve.
(a) (b)
e
Since purely elastic behaviour will be assumed, the preconditioned stress-strain curves
are manipulated accordingly. The stabilised loading paths show considerable unrecovered
strain when stress-free, referred to here as the offset strain,
offset
. For example, from
Fig.5(a), the offset strain for the 25% preconditioned loading cycle is ~0.06 nominal
strain. The offset strain for each preconditioned curve,
offset
, is subtracted and all strain
values are divided by (
offset
+ 1) to account for the larger stabilized gauge length.
Similarly, all the stress values are multiplied by (
offset
+ 1) to account for the stabilized
gauge cross-sectional area. Following this operation, the (usually small) offset stress
value is subtracted from the rest of the stress values in order to have zero stress at zero
strain. Finally, the number of data points is evenly reduced in order to provide an easily
manageable file for the curve fitting tool provided in ANSYS.
ANSYS provides curve fitting tools to obtain material constants for hyperelastic
models from the characterisation data. The data can be fed to the FEA software in the
form of tab delimited stress-strain text files of the manipulated characterisation data for
uniaxial tension, equi-biaxial tension or planar tension. Care must be taken when fitting
characterisation data to hyperelastic models. The coefficients in the strain energy
functions may be viewed as material constants. They have a certain physical meaning,
which is not considered by the curve fitting tools which only concentrate on obtaining the
best mathematical fit.
The users can choose to generate material constants that fit a single state of simple strain,
two states of simple strain at a time, or even all three simultaneously. When fitting a
model to only one state of simple strain, it is important to verify that the stress-strain
curves are meaningful in the other two modes, particularly if expected to be applied in a
FEA simulation under multiaxial loads. Often one must choose whether it is best to have
a better fit to the uniaxial tension data while for another simulation, the fit in equi-biaxial
extension is critical. Clearly if one is performing a two-dimensional plane strain analysis,
then the fit to the planar tension characterisation data is the only one that matters. The
following section provides a solution to selecting the characterisation data best suited to
determine the appropriate material constants in a hyperelastic model.
5. BIAXIALITY TEST
It has been proven difficult to obtain one set of coefficients for a hyperelastic model that
would fit well in all three simple states of strain. Compromise may be necessary for a
globally acceptable fit. Otherwise, there is acceptable behaviour in one strain state but
not in the other two. For simple loading scenarios, it is simply a matter of carefully
selecting the hyperelastic model that best fits the dominant strain state. However, in the
context of FEA models under multi-axial loads, the aforementioned lack of accuracy may
lead to false results.
Let us define the biaxiality ratio, , as the ratio of the minimum principal stretch,
3
, over
the maximum principal stretch,
1
:
3
1
= , [13]
where
1
1 and
3
1 is always the case for incompressible materials. The value of
lies somewhere between 0 and 1, and will only equal 1 if there is no deformation. Note
that the biaxiality ratio evaluates the state of deformation in an element, irrespectively of
the state of stress.
Due to incompressibility, the biaxiality ratio of an element will have an upper boundary
when in uniaxial extension, and a lower limit, when in equi-biaxial extension for which
the respective biaxiality ratios are
3/ 2
1 U
= and
3
1 B
= , which is
somewhere in-between.
Two more biaxiality ratios can be arbitrarily identified,
2 1
3 3 U P P U
= + [14]
and
2 1
3 3 B P P B
= + . [15]
The five newly defined biaxiality ratios are plotted against the maximum principal
stretch,
1
, in Fig. 6.
0
0.2
0.4
0.6
0.8
1
1 1.5 2
alpha U
alpha U-P
alpha P
alpha P-B
alpha B
Fig. 6 : Defined biaxiality ratios plotted against the maximum principal stretch,
1
. It is
assumed that the biaxiality ratio of any incompressible element will have a value between
B
and
U
.
It may be desirable to work with the logarithmic biaxiality ratio, , which will be
defined as
( )
3
3 1
1
ln ln
n n
| |
= = =
|
\ .
. [16]
U
U P
P B
Biaxiality ratio,
Max. stretch,
1
The logarithmic biaxiality ratio, , is fundamentally the same as the biaxiality ratio, ,
but becomes particularly useful when working with software packages such as
ANSYS, since the principal logarithmic strains are readily available.
Table 1 : Classification of elements according to their biaxiality ratio (total of 2560
elements tested)
Dominant extension Biaxiality ratio range Number of elements Percentage (%)
Uniaxial
U P
> 607 24
Planar
U P B P
1543 60
Equi-biaxial
B P
< 410 16
Applying the biaxiality test in an FEA simulation may provide valuable information on
the dominant simple deformation mode for a model under multiaxial loads. Taking a
cantilevered plate under bending load as an example, a first analysis is conducted to the
desired deformation with an arbitrary hyperelastic model and with arbitrary material
constants. The biaxiality ratio for every element can then be determined when post-
processing the results. The elements are classified into three regions of dominant
deformation, as defined in Table 1: a region of dominant uniaxial extension between
U P
and
U
, a region of planar deformation between
B P
and
U P
, and a region of
equi-biaxial extension between
B
and
B P
.
Fig. 7 : Meshed FEA half model of a cantilevered plate subjected to a bending load
at its free end showing the regions of dominant deformation; 1) dominantly planar
deformations, 2) dominantly uniaxial extension type of deformation and 3)
dominantly equi-biaxial extension deformation.
If the plate is infinitely wide, all deformations are expected to be planar and thus the
biaxiality ratios of every element would equal to
P
. But when considering a plate of
finite width, regions of dominantly uniaxial (above neutral axis) and dominantly equi-
biaxial (below neutral axis) extension will appear near the ends, as shown in Fig. 7, the
dimension for which are given in Section 7. As seen in Table 1 and according to the
1)
2)
3)
1
2
mg
: Clamped boundary condition
: Symmetry boundary condition
defined criteria, the majority of the elements in the analysis are in the dominantly planar
deformation region. This suggests that the selected hyperelastic model should have a
particularly good fit to the planar tension characterisation data.
6. PRECONDITIONING
A preconditioning method is proposed in this section to approximate the effects of strain-
induced stress-softening in a carbon filled rubber at moderate strain. Returning to the
example of the cantilevered bending plate, it is well known that the material nearest the
neutral axis will deform very little, whereas the top and bottom surfaces will undergo the
largest strains. During cyclic loading, the plate will be preconditioned to different states
of maximum strain meaning that it will show different stress-strain behaviour across its
thickness.
A first analysis of the finite element model to the desired deformation must be performed
with a hyperelastic model that is globally acceptable, so that the general behaviour of the
rubber is captured in the hyperelastic model. For the sake of the argument, a 1
st
order
Yeoh hyperelastic model (refer to Eq. [12]), with a material constant, C
10
= 6, is
employed in the example of the cantilevered plate subjected to a bending load.
Following the first FEA analysis with a single hyperelastic material, ANSYS post-
processing commands are used to retrieve the average maximum principal Hencky
strains,
1
n
, for each element. Using Eqs. [6] and [7], the principal Hencky strains are
transformed into nominal strains,
1
e
, and classified depending on the range of strain they
are associated with. The elements are then given a new hyperelastic material according
to their maximum principal strain, as shown in Table 2. The offset strains for the 5%,
10%, 15% and 25% nominal strain preconditioning curves in uniaxial tension will be
different from the respective offset strains in the biaxial test and again for the planar
tension test. Nonetheless, they differ very little; so, for the purpose of the
preconditioning routine, the manipulated preconditioning strain values in all three states
of simple strain are assumed to be the same; thus, 0.05 becomes ~0.03, 0.10 becomes
~0.06, 0.15 becomes ~0.10 and 0.25 becomes ~0.18 (refer to discussion on offset strain
manipulations in Section 4).
Table 2 : Classification of elements according to their maximum principal strain
Classification 1
st
order Yeoh material
constant (C
10
)
Corresponding
preconditioned purve
First run 6 Virgin initial stiffness
1
0 0.03
e
<
5.8 5%
1
0.03 0.06
e
<
3.6 10%
1
0.06 0.10
e
<
2.9 15%
1
0.10
e
<
2.1 25%
Note that a 1
st
order Yeoh hyperelastic model, such as used here, is equivalent to using
the Neo-Hookean model (Eq.[11]). Note also that it is not necessary to precisely
reproduce the preconditioned curve, but simply to closely fit its initial stiffness; hence the
use of a single material coefficient model. The 1
st
order Yeoh hyperelastic model was
preferred to the Neo-Hookean model in this case since the curve-fitting tools in
ANSYS 9.0 do not allow fitting only planar tension data for the Neo-Hookean model.
The distribution of preconditioned hyperelastic material properties in the cantilevered
plate under bending load is shown in Fig. 8.
It could be argued that the material constants for preconditioned strain values could be
interpolated and that a different hyperelastic model could be attributed to each individual
element for a more accurate solution. This was not done for computational time-
efficiency and besides, the results are accurate enough that it probably would not be
efficient to try to improve them by attributing interpolated material properties to all the
elements.
Fig. 8 : Distribution of strain shown on the top of the half model of a cantilevered
plate subjected to bending load. Darker grey:
1
0 0.03
e
< ; dark grey:
1
0.03 0.06
e
< ; light grey:
1
0.06 0.10
e
< and white:
1
0.10
e
< .
7. EXPERIMENTAL BENCH TEST
An experimental bench-test was conducted, consisting of bending a cantilevered plate
made of the characterised rubber, in order to apply the biaxiality and the preconditioning
iterations. The portion of the plate above the neutral axis is in tension and that below is
in compression. Note that the neutral axis is not necessarily at mid-thickness. Larger
deformations occur at the top and bottom surfaces of the plate than near the neutral axis.
If subjected to cyclic loading, the properties of the rubber plate may change along the
thickness, making the bench test an ideal subject for the preconditioning iteration.
Furthermore, along the width of the plate, the extremities will show end effects, such as
shown in Fig. 7, and will be subjected to dominantly uniaxial loads (tension or
compression) while the mid-width of the plate will be dominantly under planar loading,
putting the biaxiality test to trial.
1
2
mg
: Clamped boundary condition
: Symmetry boundary condition
Fig. 9 : Schematic representation of the experimental bench test of a cantilevered
rubber plate under the bending load, mg. 1) Rubber plate, 2) clamped section, 3)
displacement gauge, 4) free section and 5) rigid bar.
Table 3 : Geometric parameters for the bending cantilevered rubber plate bench
test
Total length (X-axis) 201.5 mm
Free length (X-axis) 125.3 mm
Width (Z-axis) 151.0 mm
Thickness (Y-axis) 16.25 mm
Position of displacement gauge along length, from free edge 25.3 mm
Width of rigid bar (X-axis) 12.3 mm
The clamped plate was set horizontally and weights were hung at its free end; a schematic
representation is provided in Fig. 9. A rigid bar ensures that the load is evenly distributed
along the width of the plate. After each weight was added, a 2 minute interval elapsed
before measuring the displacement to let the viscous effects take place. The plate was
loaded and unloaded, and then reshaped manually until it was straight again. This
procedure was repeated several times in order to precondition the material. The geometry
of the plate with respect to the coordinates given in Fig. 9 is given in Table 3, along with
the position of the displacement gauge and the width of the rigid bar used to evenly
distribute the load at the free end.
8. DISCUSSION OF RESULTS
The objective of the FEA simulation was to obtain static load-displacement curves that
agreed with the experiments. Due to the results of the biaxiality test (refer to Table 1), the
hyperelastic models used were best fitted to the planar tension characterisation data.
1
2
mg
1
2
mg
mg
5)
3) 2) 1)
4)
X
Y
Z
X
Y
0
20
40
60
80
100
0 10 20 30
Displacement (mm)
L
o
a
d
(
N
)
0
20
40
60
80
100
0 10 20 30
Displacement (mm)
L
o
a
d
(
N
)
Fig. 10: Two sets of experimental bench-test results (open symbols) compared to
FEA results. (a) Single curve fit hyperelastic models: ( ), 3
rd
order Yeoh
model fitted to 5% strain preconditioned planar tension curve; ( ), 1
st
order
Yeoh model fitted to 10% strain preconditioned planar tension curve. (b) ( ),
FEA results after the preconditioning iteration, 1
st
order Yeoh planar tension
curves used.
Second- and third-order hyperelastic formulations that can be obtained from the Mooney-
Rivlin, Ogden and Yeoh models are not necessarily desirable in this context since the
available characterisation data does not go to high enough strain. Of course, there are
local stiffening effects for each of the preconditioned characterisation curves due to
Mullins effect, but the danger in attempting to fit these curves with higher order
hyperelastic models is that they immediately stiffen excessively if stretched beyond the
strain for which they were preconditioned.
As shown in Fig. 10(a) when attempting to solve the FE model with a 3
rd
order Yeoh
model fitted to the 5% preconditioned characterisation data, the load history is quite
accurate (within 10% error) at small displacements (less than 5 mm) but quickly sky-
rockets with loads of several orders of magnitude above the expected load at the final
value of displacement (30 mm in this case, but the curve has been stopped at 20 mm for
clarity in the figure). Fig. 10(a) also shows poor FEA results for a 1
st
order Yeoh model
fitted to the 10% preconditioned characterisation data, especially at small displacement
error of the order of 100%. However, when applying the preconditioning method to the
FE model, excellent results are obtained, within 10% of the experimental results, both at
small and large displacements, as shown in Fig. 10(b).
The issue with this particular material is not so much fitting the data in all three simple
states of strain, but rather being able to capture the strain-induced stress-softening of this
particular rubber. The preconditioning method allows doing this.
(a) (b)
9. CONCLUSION
The biaxiality test could potentially be a useful tool when applied to simulations with
much more complicated geometry and constraints, as long as compressibility remains
negligible, in order to determine the dominant state of simple strain present in the FEA
model. Applying the preconditioning method to a cantilevered plate subjected to bending
loads has proven to drastically improve the predicted loads for a static FEA simulation,
going from nearly unacceptable results with a single hyperelastic material model to
excellent agreement with the preconditioning method. This approach is nevertheless
limited to the errors incurred in the manipulations of the characterisation data. Clearly,
using an appropriate constitutive relation to model the unrecovered strain, the
considerable hysteresis as well as the strain induced stress-softening seen in the
characterisation data would be the next best solution. Implementing Dorfmann and
Ogdens (4) strain energy based pseudo-elastic theory, or a similar model, in ANSYS
TM
to predict Mullin effects and preconditioning could be a first step towards achieving a
cleaner result.
10. ACKNOWLEDGEMENTS
The authors gratefully acknowledge the support given to them by the Natural Sciences
and Engineering Research Council of Canada, ROI Engineering Inc. and Camoplast Inc.,
Research and Development Traction Group.
11. REFERENCES
1. J.S. Dick, How to Improve Rubber Compounds: 1500 Experimental Ideas for
Problem Solving, Hanser, Munich, 2004.
2. L. Mullins, Rubber Chem. Technol., 42, 339 (1969).
3. A.H. Muhr, J. Gough, I.H. Gregory in A.H. Muhr, A. Dorfmann, eds., Proceedings of
the 1
st
European Conference on Constitutive Models for Rubber, Vienna, 1999, p.181.
4. A. Dorfmann, R.W. Ogden, Int. J. Solids Structures, 40, 2699 (2003).
5. G.T. Mase, G.E. Mase, Continuum Mechanics for Engineers (2
nd
Edition). CRC
Press, New-York, 1999.
6. A.D. Freed, J. Eng. Mater. Technol., 117, 379 (1995).
7. R.W. Ogden, in G. Saccomandi, R.W. Ogden, eds., Mechanics and Thermomechanics
of Rubberlike Solids. Springer, Udine, 2004.
8. R.S. Rivlin, Philos. Trans. R. Soc. London, Ser. A, 240, 459 (1948).
9. L.R.G. Treloar, Trans. Faraday Soc., 39, 241 (1943).
10. M. Mooney, J. Applied. Phys., 11, 582 (1940).
11. O.H. Yeoh, Rubber Chem. Technol., 63, 792 (1990).
12. O.H. Yeoh in G. Allen, ed., Comprehensive Polymer Science, Elsevier, Oxford, 1996,
Chapter 12.
13. R.W. Ogden, Proc. R. Soc. London. Ser. A, 326(1567), 565 (1972a).
14. R.W. Ogden, Proc. R. Soc. London. Ser. A, 328(1575), 567 (1972b).
15. E.M., Arruda, M.C., Boyce, J. Mech. Phys. Solids. 41 (2), 389 (1993).
16. A.N. Gent, Rubber Chemistry and Technology, 69, 59 (1996).