3. CAE model setup for crash simulation

In a crash correlation study, the CAE model is not merely a numerical replica of the physical test — it is a hypothesis about how the structure deforms, where energy flows, and which failure modes dominate. Every modeling decision, from element size to contact penalty stiffness, leaves a fingerprint on the simulated response. When the simulated acceleration pulse deviates from the sled or barrier measurement, the root cause is often traceable to a specific setup choice rather than to the solver itself. This chapter examines the modeling decisions that most strongly influence correlation quality and shows how to make them defensible.

Mesh: Element Size, Type, and Hourglass Control

Mesh density governs the resolution of bending waves, the fidelity of local buckling, and the stability of explicit time integration. For a full-frontal vehicle model, a common baseline is a 5–8 mm target element size on energy-absorbing structures (crash boxes, rails, bumper beam) and 10–15 mm on non-critical body panels. The critical time step in an explicit solver scales with the smallest element dimension and the material wave speed:

Δt_crit ≈ L_min / c,   c = sqrt(E / ρ)

Example: steel rail, E = 210 GPa, ρ = 7.85e-3 g/mm³
c ≈ sqrt(210000 / 0.00785) ≈ 5170 mm/ms
L_min = 5 mm  →  Δt_crit ≈ 0.97 µs (before mass scaling)

If the smallest element forces an impractically small time step, selective mass scaling on non-critical regions is acceptable, but the added mass must be reported and kept below roughly 1–2% of total model mass in the energy-absorbing path. Shell elements are standard for thin-walled structures; use fully integrated shells where hourglassing is a risk, or reduced integration with hourglass control (e.g., Flanagan–Belytschko stiffness-based) and verify that hourglass energy stays below 5% of internal energy. Solid elements are appropriate for castings and thick sections, while spot welds are typically modeled as deformable beam clusters or constrained hexahedra — the choice directly affects joint stiffness and therefore the force path into the occupant compartment.

Materials: Strain-Rate Sensitivity and Failure

Crash-relevant steels and aluminum are strain-rate sensitive. A piecewise-linear plasticity card (MAT_PIECEWISE_LINEAR_PLASTICITY in LS-DYNA, or the equivalent Johnson–Cook formulation) should include rate-dependent yield scaling. As an illustrative example, a mild steel might show a 15–25% yield increase at 100 s⁻¹ and 30–40% at 1000 s⁻¹ relative to quasi-static values; these multipliers must come from coupon data, not assumption. Failure modeling deserves equal care: a simple plastic strain limit will over-predict tearing in bending-dominated regions, whereas a stress-state-dependent criterion (e.g., GISSMO or Johnson–Cook damage) captures the difference between shear, tension, and plane-strain loading. For correlation work, calibrate failure parameters against component-level tests (three-point bending, axial crush) before applying them to the full vehicle.

Contacts: Penalty Stiffness, Friction, and Self-Contact

Contact definitions determine how impact forces transfer between components. The most consequential settings are:

• Penalty stiffness scale factor — default values (often 0.1 for soft, 1.0 for stiff) can allow visible penetration in stiff contacts; check maximum penetration in the d3hsp or contact energy output and keep it below one element thickness.
• Friction coefficient — 0.1–0.15 for lubricated steel-on-steel, 0.3–0.5 for painted or coated surfaces; friction affects sliding of the bumper beam against the barrier and the post-impact rebound.
• Self-contact — folded sheet metal will contact itself; define single-surface or automatic general contact with a sensible search depth (typically 1.5–2× element size) to avoid missed contacts during severe folding.
• Tied interfaces — spot welds and adhesive bonds modeled as tied contacts must use consistent stiffness; overly stiff ties artificially raise local frequencies and can shift the acceleration pulse.

Boundary Conditions and Initial Velocity

For a frontal barrier impact, the vehicle is given an initial velocity (e.g., 50 km/h = 13.89 m/s) and the rigid barrier is fully constrained. For a sled test, the pulse is applied to the sled table via a prescribed acceleration curve or an initial velocity with a deceleration boundary. Two common correlation errors arise here: applying the pulse as a velocity rather than acceleration (or vice versa), and omitting gravity or suspension preload, which changes ride height and therefore the impact point. Define the coordinate system explicitly — SAE vehicle coordinates (X forward, Y left, Z up) are conventional — and confirm that the initial velocity vector matches the test direction, including any offset or angle specified in the test protocol.

Solver Settings: Time Step, Output, and Energy Balance

Explicit solver settings influence both stability and the comparability of results. Use a consistent time step control (e.g., TSSFAC = 0.9), enable automatic time step recalculation, and monitor the energy ratio: total energy should remain approximately constant, with hourglass and sliding energy each below 5% of internal energy. Output frequency matters for correlation — acceleration signals should be sampled at 10 kHz or higher to capture the pulse peak, and the same filter (SAE J211 CFC 60 or CFC 180) must be applied to both test and simulation before comparison. A mismatch in filtering alone can produce apparent correlation errors of 10–20% in peak acceleration.

Bringing Setup Decisions into the Correlation Workflow

The practical discipline is to treat each setup category as a controlled variable. Before comparing curves, run a model quality gate: confirm hourglass and sliding energy ratios, check maximum penetration, verify added mass, and confirm that the output filter matches the test. If the simulated pulse peaks earlier and higher than the test, the likely culprits are an over-stiff contact penalty, a missing strain-rate effect, or an overly coarse mesh in the crush zone. If the pulse is too soft and wide, suspect under-stiff ties, excessive mass scaling, or a failure criterion that triggers too early. Documenting these setup parameters alongside the correlation metrics (CORA score, peak deviation, phase shift) turns a one-off comparison into a traceable engineering record — and makes the next model update a targeted correction rather than a guess.