3. Simulation model development lifecycle

A crash simulation model is not a single artifact but the end product of a controlled pipeline. Each stage — geometry preparation, meshing, material modeling, and verification — introduces decisions that propagate forward and either preserve or destroy the model's fitness for its intended acceptance purpose. In the applicability-and-acceptance framing of this course, the lifecycle matters because a model can only be trusted within the boundaries its development evidence supports. A model that was never verified against a physical test cannot justify an acceptance decision, no matter how refined its mesh.

The lifecycle is best understood as a chain of transformations with explicit hand-off criteria between stages. The figure below shows the four stages, the principal inputs and outputs of each, and the verification gates that must be passed before the next stage consumes the output.

Geometry Preparation

Geometry preparation converts CAD into a form the solver can consume without introducing artifacts. The first decision is representation: thin-walled body structures are usually idealized as shell mid-surfaces, while thick castings, battery modules, and crash boxes with closed sections may require solid or mixed shell–solid modeling. Mid-surfacing is not a cosmetic operation — it changes bending stiffness if the offset is mishandled, so the offset direction and thickness assignment must be checked against the original CAD.

Defeaturing removes details that do not influence the crash response but do inflate mesh size and degrade quality: small fillets below the target element size, bolt holes that are represented by connectors, and logos or stamped text. The rule of thumb is to defeature only features smaller than roughly one-third of the local target element size, and to document every removal so the model's fidelity boundary is traceable. Part positioning follows: the vehicle must be assembled in the correct global coordinate system, with the impact direction, ground plane, and barrier orientation defined once and held consistent through meshing, material assignment, and post-processing.

Connections — spot welds, adhesive bonds, bolts, and seams — are prepared at this stage as connector definitions rather than as geometry. A common failure is leaving weld locations to be inferred later; instead, export the weld map with coordinates and part-pair identifiers so the meshing stage can place rigid or deformable spot-weld elements deterministically.

Meshing

Mesh design in crash applications is driven by the deformation physics, not by uniform refinement. Crush zones, hinge lines, and fold initiators need element sizes small enough to resolve the folding wavelength — typically 5–8 mm for automotive steel and aluminum structures — while non-deforming regions can use 15–20 mm. Shell elements dominate body-in-white models; solids are reserved for thick sections and for parts where through-thickness stress matters.

Quality metrics must be checked before the mesh is accepted. A practical acceptance set for crash meshes is: Jacobian above 0.6, warpage below 15°, aspect ratio below 5 in energy-absorbing regions, and no negative-volume elements. Hourglass control is a meshing-adjacent decision: reduced-integration shell elements are standard for speed, but they require hourglass stabilization, and the hourglass energy must later be confirmed to stay below roughly 5% of internal energy. If it does not, the mesh is too coarse locally or the stabilization is misconfigured.

Material Modeling

Material cards are where the model's predictive claim is most fragile. For ductile metals, a piecewise-linear plasticity curve with strain-rate dependence (for example, a Cowper–Symonds or tabulated rate factor) is the baseline. The curve must be calibrated from coupon tests at multiple rates, and the failure criterion — equivalent plastic strain at rupture, or a stress-triaxiality-dependent fracture locus — must be selected to match the failure mode the acceptance test will exercise. Using a tensile-coupon failure strain for a shear-dominated joint will over-predict ductility.

Polymers, foams, and composites require dedicated laws: crushable foam for energy absorbers, progressive-damage models for CFRP, and viscoelastic–viscoplastic formulations for glassy polymers. Each card should carry a calibration record: test type, rate, specimen geometry, and the residual error between simulation and coupon test. A material card without that record cannot support an acceptance argument.

Model Verification

Verification answers a narrower question than validation: does the model behave as intended numerically? Three checks are non-negotiable. First, mass and inertia: total model mass should match the CAD-derived target within about 2%, and the center of gravity should sit within a few millimeters of the reference. Second, energy balance: total energy must be conserved, added mass from contact must be small, and hourglass and sliding energy must remain bounded fractions of internal energy. Third, component-level correlation: a subsystem — a bumper beam, a B-pillar, a crash box — is simulated against its physical test before the full vehicle is trusted. If the subsystem correlation fails, the rework loop returns to geometry, mesh, or material as the evidence indicates, not to the full-vehicle model.

The lifecycle closes when the verified model is paired with a written statement of what it may and may not be used to decide. That statement — the validity envelope — is the direct input to the applicability and acceptance boundaries that the remainder of this course addresses.