3. Geometry preparation and CAD cleanup

Crash simulation quality is decided long before the first LS-DYNA run. A shell mesh built on dirty CAD will produce distorted elements, spurious penetrations, and hourglassing that no amount of solver tuning can fix. This chapter covers the geometry pipeline that feeds an LS-DYNA crash model: midsurface extraction, defeaturing, meshing strategy, and the quality gates that must be passed before the model is considered ready for assembly and load-case setup.

Why Crash Models Need a Dedicated Geometry Pipeline

Crashworthiness models are dominated by thin-walled stamped structures — front rails, B-pillars, rockers, floor pans, bumper beams. These are modeled with shell elements (ELFORM 2, 16, or 24 in LS-DYNA), which require a midsurface representation rather than the solid B-rep delivered by CATIA, NX, or Creo. A typical full-vehicle model contains 1.5–3.0 million shell elements, and every one of them inherits its shape from the midsurface. If the midsurface has gaps, overlaps, or sliver faces, the mesh will too.

The practical consequence: geometry preparation typically consumes 40–60% of the total CAE lead time for a new vehicle program. Treating it as a quick preprocessing step is the most common cause of late-stage model rebuilds.

Midsurface Extraction

Midsurface extraction collapses a thin solid into a surface located at the mid-plane of the wall thickness. In ANSYS SpaceClaim, HyperMesh (MidMesh), or Simcenter, the operation is semi-automatic: the tool detects opposing face pairs and offsets them to the center. The results are rarely clean on the first pass.

Key parameters and their effect:

• Thickness tolerance — pairs whose separation exceeds this value are rejected. For automotive stamped steel (0.6–2.5 mm), a tolerance of 3–4 mm works well; setting it too tight leaves holes, too loose merges unrelated walls.
• Minimum face size — faces below this threshold are ignored, which suppresses fillet slivers but can also drop small beads and stiffening ribs.
• Merge distance — controls whether adjacent extracted surfaces are stitched automatically.

After extraction, verify that the total surface area multiplied by the nominal thickness reproduces the solid mass within 2–3%. A quick sanity check in HyperMesh is Tools > Mass Calc on the midsurface with a uniform thickness assignment. If the mass is off by more than 5%, the extraction missed features or duplicated walls.

Defeaturing: What to Remove and What to Keep

Defeaturing is not cosmetic simplification — it is a deliberate trade between mesh quality and physical fidelity. The rule of thumb for crash models: remove anything whose characteristic length is below 3–5 times the target element size, unless it carries load path or crush-initiation behavior.

Defeaturing should be done on the midsurface, not on the solid. Working on the solid forces a re-extraction after every change and multiplies the effort. Use Geometry > Defeature in SpaceClaim or the equivalent surface-editing tools in HyperMesh to delete, extend, and stitch faces directly.

Meshing Strategy for Crash Shells

Crash models use a target element size driven by the smallest feature that must be resolved and by the stable time step. For explicit analysis, the stable time step scales with the smallest element dimension:

dt_stable ≈ L_min / c
c = sqrt(E / ρ)   (wave speed in steel ≈ 5100 m/s)

Example: L_min = 5 mm, steel
dt_stable ≈ 0.005 m / 5100 m/s ≈ 9.8e-7 s ≈ 0.98 µs

A 5 mm target element size gives roughly a 1 µs step, which is a practical baseline for full-vehicle crash. Local refinement to 2 mm drops the step to ~0.4 µs and can double or triple run time. Refine only where strain gradients demand it — typically the front rail crush zone, hinge pillars, and door latch regions.

Element type selection:

• ELFORM 2 (Belytschko-Tsay) — default, fastest, adequate for most body panels.
• ELFORM 16 (fully integrated) — use where hourglassing is severe or bending-dominated behavior matters.
• ELFORM 24 — improved for warped geometry; useful on complex stampings.

Aim for a mesh that is predominantly quad-dominant (≥ 90% quads). Triangles are acceptable in transition zones but should not exceed 5–10% of the total, and never in high-strain regions.

Quality Checks Before Assembly

Run these checks on each component before merging into the full-vehicle model. Catching defects at the part level is far cheaper than debugging a 2-million-element assembly.

In HyperMesh, use Check Elements > 2D > Quality Index with a crash-specific profile. LS-DYNA will terminate with a negative volume error if any element inverts during the run, so a negative Jacobian at setup time is a guaranteed failure.

Workflow Overview

Practical Notes on Tool Selection

HyperMesh remains the dominant preprocessor for LS-DYNA crash models because of its mature midsurface and defeaturing toolset, plus direct LS-DYNA deck export. ANSYS SpaceClaim is faster for midsurface extraction on clean parametric CAD but weaker on large assembly defeaturing. Simcenter 3D integrates well with NX-native geometry. The choice matters less than the discipline of running the quality gate on every component before assembly.

One final check before releasing geometry to the assembly stage: confirm that the midsurface thickness assignments match the drawing callouts. A 0.1 mm error on a 1.2 mm rail is an 8% stiffness error, which propagates directly into the crash pulse. Export the thickness map as a CSV alongside the mesh so the assembly team can audit it.