Professional Documents
Culture Documents
Hyperelasticity
1 Introduction
In a series of solver tests we are trying to compare performance and credibility of the new SAMCEF
based nonlinear FE solver SOL 402 relative to the established ADINA based solver SOL 601.
Here we focus on the modeling of hyperelastic materials. SOL 402 does not support Arruda-Boyce
and Sussman-Bathe models. Therefore we confine our considerations on models using Ogden
materials.
In the first place we compare performance using a medium sized model of torsional loading a
rubber part.
Finally we test the 8-node hexahedral element for a nearly compressible material using fitted
material constants. The underlying elastomer test data are from the well-known experiments of
Treloar (1944).
Up to now there is no comprehensive documentation of SOL 402. So we use default settings for
both solvers if not otherwise mentioned.
Femap 12.0.1 was used for modeling, NX Nastran 12.0.2 for solving.
2 Performance
We used the FE model of rubber part of moderate complexity. An Ogden model of fourth order was
used. The inner side of the part is loaded as a 3° enforced rotation relative to the fixed outer side.
This torsion leads locally to equivalent strains up to 83,3 %.
The part was meshed with 323,122 linear hexahedral elements leading to 508,066 nodes.
For both solvers we used 2 time steps, which seemed to be optimal for SOL 402 in this case.
Sparse solvers and automatic incrementation was used in both cases. No other settings were non-
default.
Stress results output is shown in Figure 1 and Figure 2. Noticeable is: SOL 402 exhibits higher
stress values in the strained regions of the part: up to a factor of 3.25. The question arises which
stress measures are used by the different solvers.
SOL 402 vs. SOL 601: Hyperelasticity
26.04.2019 Revision A Thomas Grauer
SOL 601 outputs Cauchy stresses and Green-Lagrange strains for hyperelastic materials.
According to Nastran 12.0.2 release guide SOL 402 should output Logarithmic strain and Cauchy
stress as well. This seems not to be the case.
Performance is shown in Table 1. A Workstation with a six-core Xeon E5 CPU and 128 GByte of
RAM was used. The ADINA solver is 4.5 times faster in this case.
where
Cauchy stresses:
1 𝜕𝑊𝑖𝑠𝑜 𝜕𝑊𝑖𝑠𝑜
𝜎1 − 𝜎2 = (𝜆1 − 𝜆2 )
𝐽 𝜕𝜆1 𝜕𝜆2
1 𝜕𝑊𝑖𝑠𝑜 𝜕𝑊𝑖𝑠𝑜
𝜎1 − 𝜎3 = (𝜆1 − 𝜆3 )
𝐽 𝜕𝜆1 𝜕𝜆3
2⁄ −1
𝑃𝑖 = 𝐽 3 𝜆𝑖 𝜎𝑖
1⁄ −2
𝑆𝑖 = 𝐽 3 𝜆𝑖 𝜎𝑖
These three benchmark cases are characterized by the following principal stretches and stresses:
Uniaxial: 𝜆1 = 𝜆, 𝜆2 = 𝜆3 = 1/𝜆 ; 𝜎2 = 𝜎3 = 0
Equibiaxial: 𝜆1 = 𝜆2 = 𝜆, 𝜆3 = 𝜆−2 ; 𝜎1 = 𝜎2 , 𝜎3 = 0
We fitted Treloar’s measured data against a third order Ogdon material model. Elastic energy
densities for this model is
𝜇 𝛼𝑘 𝛼𝑘 𝛼𝑘
𝑊𝑖𝑠𝑜 = ∑3𝑘=1 𝛼𝑘 (𝜆1 + 𝜆2 + 𝜆3 − 3)
𝑘
We did a least square fit of all measured data using First PK stress (see Figure 3). Results are:
6,00
Test Uniaxial
First Piola-Kirchhoff Stress [MPa]
Test Biaxial
4,00
Ogden Biaxial
1,00
0,00
1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00
Stretch
Figure 3: Least Square Fitting for Ogdon model
The resulting element stress values were compared with expected values according to chapter 3.
The first result was that SOL 402 outputs stress results as Second PK tensors, in contrast to the
documentation.
A comparison of relative errors between FE results versus expected values is shown in Table 2,
Table 3 and Table 4.
SOL 601 yields acceptable results for any simulation in the observed parameter range.
The strange outcome is: SOL 402 yields similar acceptable stresses in the loaded directions. But
the unconstraint directions deliver nonzero stress values. These parasitic stresses may exceed the
real stresses by several orders of magnitude. Even for strains below 100 percent these spurious
stresses may be unacceptable high, e.g. for biaxial load and smaller compressibility. This behavior
cannot be explained by element volume locking. Is there something wrong with element
formulation?
We did the same comparison (using a Mooney-Rivlin model) for SOL 106 and SOL 402. SOL 106
does not show up with these parasitic stresses.
Reliable results of SOL 402 can be expected only for high compressibility ( > 0,499) and strains
below 100%.
6 Conclusion
1. SOL 402 outputs Second Piola-Kirchhoff stress results – in contrast to documentation.
2. Performance of SOL 601 in hyperelasticity calculations, using default settings, may be five
times better.
3. SOL 601 yields acceptable results in the range = 0,48 to 0,49999 and strains up to 500%.
4. SOL 402 may yield parasitic stresses in non-constraint directions, if not confined to strains
below 100% and higher compressibility.