2. Stress-strain measures and finite strain kinematics

Calibrating a constitutive model against experimental data is only meaningful when the strain and stress measures used in the simulation match those used to reduce the test data. In small-strain analyses this distinction is often ignored because all measures collapse to nearly the same value. Once a rubber coupon, a metal forming operation, or a crash component exceeds a few percent strain, the choice of kinematic measure and stress rate directly changes the calibrated parameters. This section establishes the finite-strain framework that every subsequent calibration workflow depends on.

2.1 The Deformation Gradient as the Primary Kinematic Variable

All finite-strain kinematics derive from the deformation gradient F, which maps an infinitesimal material line dX in the reference configuration to its deformed counterpart dx:

dx = F · dX,   F = ∂x/∂X = I + ∂u/∂X

Here u is the displacement field, X are reference (material) coordinates, and x are current (spatial) coordinates. The determinant J = det(F) measures volume change: J = 1 for incompressible or isochoric behavior, J > 1 for dilation, and J < 1 for compaction. For rubber-like materials, enforcing near-incompressibility (J ≈ 1) is a central calibration constraint, so J must be tracked explicitly rather than assumed.

The polar decomposition splits F into a rotation and a stretch:

F = R · U = V · R

where R is a proper orthogonal rotation, U is the right stretch tensor (material description), and V is the left stretch tensor (spatial description). This decomposition matters for calibration because R carries no strain information; only U or V should drive the constitutive response. A model formulated directly on F without separating rotation will produce spurious stress under rigid-body rotation.

2.2 Strain Measures and Their Differences

Several strain tensors are built from F, and each behaves differently at large deformation:

The Green–Lagrange strain is the natural conjugate of the second Piola–Kirchhoff stress S, and the pair (S, E) is work-conjugate. This conjugacy is why hyperelastic strain-energy densities are almost always written as W = W(C) or W = W(E). For a uniaxial test stretched to a stretch ratio λ = 1.5, the engineering strain is 0.5, the Green–Lagrange strain is ½(λ² − 1) = 0.625, and the logarithmic strain is ln(1.5) ≈ 0.405. These are not interchangeable: feeding engineering strain into a model calibrated with Green–Lagrange strain introduces an error of roughly 25% at this stretch level.

2.3 Rate of Deformation and Spin

The spatial velocity gradient L = Ḟ · F⁻¹ decomposes additively into a symmetric rate-of-deformation tensor D and a skew-symmetric spin tensor W:

L = D + W,   D = ½(L + Lᵀ),   W = ½(L − Lᵀ)

D is the spatial strain-rate measure used in rate-dependent (viscoplastic) models, while W describes instantaneous rigid rotation. In a tensile test at constant crosshead speed, D₁₁ is the true strain rate along the loading axis, and it is not constant even when the crosshead speed is constant, because the gauge length changes. This is why viscoplastic calibration should use true strain rate, not engineering strain rate, when the strain exceeds a few percent.

2.4 Objective Stress Rates

Cauchy stress σ is a spatial quantity, and its material time derivative σ̇ is not objective: it changes under superposed rigid rotation. Constitutive rate equations must therefore use an objective stress rate. The two most common are:

Jaumann rate:   σ̊ = σ̇ − W·σ + σ·W
Truesdell rate:  σ̊ = σ̇ − L·σ − σ·Lᵀ + (tr L)σ

The Jaumann rate is simple and widely implemented, but it is known to produce spurious oscillations in simple shear at large strain. The Truesdell rate is work-conjugate with the logarithmic rate and is preferred for metal plasticity at large strain. For calibration, the practical consequence is that a parameter set fitted using a Jaumann-rate formulation may not transfer to a Truesdell-rate solver, and vice versa. The choice must be documented alongside the calibrated parameters.

2.5 Practical Calibration Workflow

The following procedure ensures consistency between test data reduction and the FE model. It is written for a uniaxial tensile test on a ductile metal, but the logic transfers to compression, shear, and multiaxial tests.

1. Record raw data. Capture force F, original gauge length L₀, original area A₀, and current gauge length L (from extensometer or optical measurement). Do not rely on crosshead displacement for strain beyond yield.
2. Compute true stress and true strain. Using incompressibility as a first approximation: σ_true = F/A = F·L/(A₀·L₀) and ε_true = ln(L/L₀). For large strain, correct A from measured width and thickness rather than assuming constant volume.
3. Select the conjugate pair. For metal plasticity with logarithmic strain, use (ε_true, σ_true) and a Truesdell or logarithmic rate. For hyperelasticity, convert to (E, S) using S = J·F⁻¹·σ·F⁻ᵀ.
4. Fit parameters. For a Swift hardening law σ = K(ε₀ + ε)ⁿ, fit K, ε₀, and n to the true stress–true strain curve. For a neo-Hookean model, fit the shear modulus from the initial slope of S versus E.
5. Verify in the solver. Run a single-element simulation with the same strain measure and objective rate. Compare the simulated force–displacement curve to the test. A mismatch at large strain usually indicates a measure or rate inconsistency, not a bad parameter fit.

2.6 Common Failure Symptoms and Diagnosis

The central discipline is consistency: the strain measure used to reduce the test, the stress measure used in the constitutive equation, and the objective rate used in the solver must form a work-conjugate set. When they do, calibration becomes a well-posed parameter identification problem. When they do not, the optimizer will still return numbers, but those numbers will not predict the behavior of the component.