Validation and Uncertainty of Material Data
Theory: the fitted test vs the deformation you need
Overview
We set material constants by fitting curves to test data. Isn't that enough?
It can be, if the deformation matches the fitted test. The problem is when the part deforms differently. Rubber-like materials respond differently in tension, compression, shear and in-plane biaxial stretching, and some constant combinations match one mode but miss another badly. A clean curve fit is reassuring, but it only reproduces the fitting data and proves nothing about prediction. So you check predictions against tests not used in fitting. That's material data validation.
Reference material (Treloar's rubber, three-term Ogden)
P is nominal stress and λ stretch. Constants μ = 0.618, 0.00118, −0.00981 MPa and α = 1.3, 5.0, −2.0 (Ogden, 1972); initial shear modulus 0.414 MPa. Values from these formulas are treated as "test data".
Mooney–Rivlin model
With C01 = 0 it becomes neo-Hookean. Least-squares fits used uniaxial stretches 1.05–3.0 and equibiaxial 1.05–2.5.
Treloar's data
In 1944 Treloar tested vulcanised natural rubber in uniaxial tension, equibiaxial tension and pure shear, measuring stress to large stretches. As a classic dataset of different deformation modes on one material, it is still the benchmark for comparing new hyperelastic models. Ogden's 1972 model demonstrated on these data that one set of constants could represent all three modes well.
Worked examples
Example: how fitting choices change predictions
| Model and fit | C10 (MPa) | C01 (MPa) | Max uniaxial error | Max equibiaxial error | Equibiaxial stress at λ = 2.0 |
|---|---|---|---|---|---|
| Neo-Hookean (uniaxial only) | 0.160 | 0 | 22.2% | 26.6% | 0.629 MPa |
| Mooney–Rivlin (uniaxial only) | 0.112 | 0.112 | 6.6% | 309.5% | 2.209 MPa |
| Mooney–Rivlin (uniaxial + equibiaxial) | 0.160 | 0.009 | 17.9% | 20.1% | 0.774 MPa |
The reference equibiaxial stress at λ = 2.0 is 0.806 MPa. At uniaxial λ = 3.0 the uniaxial-only Mooney–Rivlin fit gives 0.863 MPa, exactly matching the reference.
The Mooney–Rivlin fit that matches uniaxial best is the worst in biaxial.
C10 and C01 act in similar ways in uniaxial tension, so uniaxial data barely determines how they split. Least squares picks the split that best fits the uniaxial curve, which here inflated C01. In equibiaxial tension C01 is multiplied by λ² and behaves very differently, so an oversized C01 multiplies predicted stress. That's why a 6.6% uniaxial fit was 2.7 times off in biaxial. Including biaxial data keeps C01 small and both errors near 20%. Using uniaxial-only constants is risky for parts dominated by biaxial deformation, such as rubber seals, tyres and inflated membranes.
Steps to validate material data
- From part analyses, identify main deformation modes and ranges of strain, temperature and strain rate.
- Gather tests covering those ranges (uniaxial, biaxial, shear, volumetric).
- Fit tests simultaneously and check constants are physically reasonable (stability etc.).
- Check predictions against tests or component tests not used in fitting.
- Estimate constant uncertainty from test scatter and assess its effect on results.
“Seal contact pressure fell short”
A rubber O-ring leaked in prototype despite analysis showing ample contact pressure. Its constants had been fitted to the supplier's uniaxial tension data alone. In a compressed O-ring cross-section, material near the surface deforms almost biaxially. Refitting with added biaxial and planar tests lowered predicted contact pressure by about 30%, explaining the leak. Increasing squeeze in the design fixed it.
Common mistakes
Mistakes and fixes
| Mistake | Effect | Fix |
|---|---|---|
| Fitting uniaxial only | Large errors elsewhere | Fit multiple modes |
| Calling a good fit validation | Predictive power unknown | Test on unused data |
| Using beyond test range | Instability on extrapolation | Check ranges |
| Ignoring temperature and rate | Fails under other conditions | Test per condition |
| Ignoring scatter | Overconfident margins | Quantify uncertainty |
I'd like to learn related topics.
Related pages include ASME V&V guidelines, boundary condition checks, model review checklists, the NAFEMS quality system and dimensional analysis.
Related Topics
Experience the theory firsthand with the interactive simulator for this field
All Simulators