3. Experimental characterization for calibration
Calibration of automotive material models begins with experimental data, not with the solver. Before a single parameter can be tuned, the engineer must decide which tests to run, how to convert raw machine signals into true stress–true strain curves, and which portions of those curves will actually drive the calibration. This chapter covers the five test families most commonly used in automotive crash and durability work — uniaxial tension, compression, shear, bulge, and split Hopkinson pressure bar (SHPB) — and explains how to extract usable stress–strain data and identify calibration targets from each.
Test Selection by Stress State and Strain Rate
Each test produces a different stress state, and each material model responds differently to that state. A metal plasticity model calibrated only on tension data will usually over-predict yield in compression, because the tensile instability (necking) truncates the useful curve early. Conversely, a polymer model calibrated only on compression will misrepresent the dilatational response under tension. The practical rule is to select tests that cover the stress states and strain rates the component will actually experience.
| Test | Dominant stress state | Typical strain rate | Primary calibration target |
|---|---|---|---|
| Uniaxial tension | σ₁ > 0, others ≈ 0 | 10⁻⁴–10⁰ s⁻¹ | Yield surface, hardening, failure strain |
| Compression | σ₁ < 0, others ≈ 0 | 10⁻³–10⁰ s⁻¹ | Compressive yield, large-strain hardening |
| Shear | Pure shear τ | 10⁻³–10⁰ s⁻¹ | Shear modulus, shear yield, Lode dependence |
| Bulge | Biaxial tension | 10⁻³–10⁻¹ s⁻¹ | Biaxial yield, anisotropy coefficients |
| SHPB | Tension/compression/shear | 10²–10⁴ s⁻¹ | Rate sensitivity, dynamic yield |
For a typical automotive steel or aluminum alloy, a minimum calibration set is tension plus compression at quasi-static rate, plus SHPB compression at one or two dynamic rates. For anisotropic sheet metals, add bulge or a balanced biaxial test. For polymers and elastomers, add shear and, where hydrostatic sensitivity matters, a confined compression test.
From Raw Signals to True Stress–True Strain
The most common error in calibration is feeding engineering stress–strain data into a model that expects true stress–true strain. The conversion is straightforward but must be applied consistently, and the reference convention must be stated explicitly: here, tension is positive, engineering strain is ΔL/L₀, and true strain is defined as ε_true = ln(1 + ε_eng).
# Convert engineering to true stress-strain (tension, incompressible assumption)
# e_eng: engineering strain, s_eng: engineering stress [MPa]
e_true = ln(1 + e_eng)
s_true = s_eng * (1 + e_eng)
# For compression, signs flip:
# e_true = -ln(1 - |e_eng|)
# s_true = s_eng * (1 - |e_eng|)Beyond the conversion, three corrections matter. First, the elastic modulus must be measured from the initial linear region, not taken from a handbook, because grip compliance and machine stiffness inflate the apparent strain. Second, the yield point should be reported at a defined offset (typically 0.2% for metals) and the offset must be recorded. Third, the post-necking region in tension is not a material property — it is a structural response — so the calibration target should be truncated at the point of maximum load unless a reverse-engineering method (e.g., digital image correlation or an inverse FE fit) is used to extend it.
Extracting Calibration Targets
A calibration target is a scalar or a curve that the optimizer will try to reproduce. Not every point on the stress–strain curve is a good target. Good targets are sensitive to the parameters being calibrated and insensitive to noise. For a Johnson–Cook plasticity model, the targets are typically:
- A (initial yield): the 0.2% offset yield stress from the quasi-static tension test.
- B and n (hardening): fitted from the true stress–true strain curve between yield and the point of maximum load.
- C (rate sensitivity): the ratio of dynamic yield to quasi-static yield, extracted from SHPB tests at two or more rates.
- Failure strain: from the tension test, but only if the gauge length and strain measurement method are recorded, since failure strain is gauge-length dependent.
For anisotropic sheet metals, the Lankford coefficients r₀, r₄₅, and r₉₀ are extracted from tension tests cut at 0°, 45°, and 90° to the rolling direction. These become targets for the Hill or Barlat yield function. The bulge test provides the balanced biaxial yield stress, which constrains the equi-biaxial point of the yield surface and is essential for distinguishing between yield functions that fit the uniaxial data equally well.
Rate-Dependent Testing with SHPB
The split Hopkinson pressure bar extends the strain-rate range to 10²–10⁴ s⁻¹, which is the regime relevant to crash events. A standard SHPB setup consists of a striker bar, an incident bar, a transmission bar, and the specimen sandwiched between the incident and transmission bars. Strain gauges on the bars record the incident, reflected, and transmitted pulses. The specimen stress and strain are computed from these pulses using the one-dimensional wave equations:
# SHPB data reduction (compression, specimen between bars)
# e_r: reflected strain, e_t: transmitted strain
# A_b, E_b: bar area and modulus; A_s, L_s: specimen area and length
# c_b: bar wave speed
strain_rate = -2 * c_b * e_r / L_s
strain = integral(strain_rate, dt)
stress = E_b * (A_b / A_s) * e_tThree practical issues dominate SHPB data quality. First, the specimen must reach dynamic equilibrium — the stress at the front and back faces must be nearly equal — otherwise the computed stress is not a material property. Equilibrium is checked by comparing the incident-plus-reflected signal to the transmitted signal. Second, friction between the specimen and the bars creates a barreling effect that artificially raises the apparent flow stress; lubrication and a correct length-to-diameter ratio (typically 0.5–1.0) mitigate this. Third, pulse shaping with a copper or polymer shim is often needed to ramp the incident pulse and avoid dispersion, especially for low-impedance materials such as aluminum or polymers.
Linking Tests to Model Parameters
The final step before optimization is to map each test to the parameters it constrains. This mapping prevents over-fitting: if two tests constrain the same parameter, the optimizer may satisfy one at the expense of the other, and the resulting model will fail on validation. A clean mapping for a metal plasticity model is shown below.
Once the mapping is fixed, the calibration workflow becomes: process each test into true stress–true strain, extract the target values, run the optimizer against the targets, and then validate the calibrated model against a test that was not used in the fit. If the validation test is a different stress state or strain rate than any calibration test, the validation is meaningful; if it is a repeat of a calibration test, it only confirms the optimizer converged.
Practical Checklist for Data Extraction
- Record specimen geometry, gauge length, and strain measurement method for every test.
- Measure the elastic modulus from the test data; do not assume a handbook value.
- Apply the engineering-to-true conversion before fitting, and state the sign convention.
- Truncate tension data at maximum load unless an inverse method extends it.
- Verify SHPB dynamic equilibrium before accepting the curve.
- Keep at least one test per stress state out of the calibration set for validation.
The quality of the final calibrated model is bounded by the quality of these extracted targets. A solver can fit noisy or inconsistent data very precisely and still produce a model that fails in a full-vehicle simulation. Careful test selection, consistent data reduction, and a clean test-to-parameter mapping are what make the calibration both accurate and defensible.