4. Building the virtual buck model

The virtual buck is the first CAE artifact in the program that is expected to answer a design question rather than merely describe a CAD shape. Its purpose is to predict stiffness, load paths, and joint-level behavior well enough to compare concept variants before tooling money is committed. That purpose dictates every modeling decision: what to delete, how coarse the mesh may be, which material law is defensible at this maturity level, and how connections are represented. A virtual buck that is geometrically faithful but numerically unusable is worse than a simplified model whose assumptions are documented and bounded.

The reference object for this chapter is a seat-back frame buck: a stamped steel side member pair, two cross members, a recliner bracket on each side, and a foam-and-trim surrogate mass. The design question is comparative — does variant A (1.6 mm side members, two cross members) outperform variant B (1.2 mm side members, three cross members) under a 1.5 kN rearward load applied at the upper cross member, with the recliner brackets grounded? All numbers below are illustrative teaching values, not measured results.

Geometry simplification: delete what cannot change the answer

Start from the CAD assembly and apply a deletion test to every feature: if removing it changes the predicted load path or local stiffness at a monitored location by less than the noise floor of the comparison (here, roughly 3%), it does not belong in the buck. In practice this removes fastener holes below M8, hem flanges that carry no load path, trim attachment bosses, and all cosmetic radii under 2 mm. Welding nut and bolt head geometry is replaced by equivalent connected patches. The foam-and-trim surrogate is not meshed at all; it becomes a distributed mass with an offset center of gravity, because its stiffness contribution is negligible relative to the steel frame at this load level.

Two simplifications deserve explicit sign-off rather than silent acceptance. First, mid-surface extraction: stamped members of 1.2–1.6 mm thickness are extracted to shell mid-surfaces, which shifts the geometric boundary by half the thickness. At bracket interfaces this offset matters, so the bracket footprint is projected onto the mid-surface and the connection is defined there. Second, the recliner bracket is modeled as a rigid body with a grounded reference point, not as a deformable part — justified only if bracket deformation is not the quantity under study. Both decisions go into the model assumption log with the date and the reviewer, because a later analyst will otherwise re-litigate them.

Mesh strategy: sized by the question, not by the geometry

Shell meshing dominates this buck. The governing rule is that element size must resolve the shortest structural wavelength you care about — typically the distance between spot welds or the width of a load-introduction flange. With welds spaced at 40 mm, a 10 mm target element size gives four elements between welds, which is adequate for load-path comparison. A global 10 mm quad-dominant mesh with 5 mm refinement on the recliner bracket flanges and around the load introduction point is a defensible starting point.

Quality gates should be set before meshing, not after: aspect ratio below 5, Jacobian above 0.6, warpage below 15°, and no more than 2% of elements failing any criterion. Run a mesh convergence check on the single most important output — here, the vertical deflection at the load point. Three meshes at 14 mm, 10 mm, and 7 mm should show the deflection changing by less than 5% between the last two; if it does not, the joint representation, not the mesh, is usually the culprit. Record the converged size in the model log so variant B is meshed identically. Comparing variants meshed at different densities is the most common way early-stage CAE produces a confident wrong answer.

Material models: elastic first, plasticity only where it earns its place

At concept maturity the dominant uncertainty is geometry and joints, not the constitutive law. Use linear isotropic elasticity for all steel parts: E = 210 GPa, ν = 0.3, ρ = 7.85 × 10⁻⁹ t/mm³, with the shell section defined by the actual gauge per variant. Do not introduce plasticity unless the predicted peak equivalent stress exceeds roughly 80% of yield somewhere that matters — and if it does, that is itself a finding worth reporting before you complicate the model. If plasticity is required, use a tabulated true stress–plastic strain curve for the specific grade (for example DP590) rather than a bilinear approximation, and state the strain-rate assumption, since seat-back loading is quasi-static.

Two material-model traps recur in bucks. First, using nominal rather than minimum specified thickness: the 1.2 mm variant may be rolled at 1.14 mm minimum, and a 5% thickness error propagates almost linearly into bending stiffness. Second, treating the foam surrogate as a soft solid with guessed modulus — this creates a stiffness path that does not exist and quietly changes the load split between cross members. The offset point mass avoids the problem entirely.

Joint representations: the dominant error source

Spot welds are modeled as rigid spiders connecting a small patch of shell elements on each member to a common node, with the patch diameter set to the actual weld nugget (about 5–6 mm). This is acceptable for load-path comparison but overestimates local stiffness. Where weld-line stress is a monitored output, switch to a deformable joint — a short beam or solid nugget with the weld metal's properties — at least in the bracket region. Bolt joints at the recliner are represented by a grounded rigid body plus a distributed coupling over the bracket footprint; do not use a single-point constraint at one node, which creates an artificial stress singularity and makes the reaction force meaningless.

Validate the joint representation before trusting any comparison. A useful check: apply a 100 N lateral load at the upper cross member and confirm that the reaction at each grounded bracket is within about 10% of the hand-calculated split based on relative stiffness. If one bracket takes 80% of the load in a nominally symmetric buck, a joint is mis-connected — usually a spider whose patch missed the mid-surface after extraction.

Model assumption log (deliverable)

The buck is complete when the assumption log is signed, the mesh convergence check is recorded, and the joint sanity check passes. Only then should variant A and variant B be run and compared — and the comparison is only valid if both variants share the same mesh density, material card, joint definitions, and load case. Any deviation between the two models is a confound that will be read as a design effect.