2. Simulation model verification and validation
Verification and validation (V&V) is the technical backbone of any credible CAE sign-off. Verification asks a narrow, internal question: are the governing equations being solved correctly on the mesh and model we built? Validation asks a broader, external question: does the model reproduce the physical reality it claims to represent? A model can pass verification and still fail validation — a perfectly converged solution on a poor idealization is still wrong. Reviewers must therefore treat mesh, boundary conditions, material data, and correlation evidence as one connected chain, not four independent checkboxes.
Mesh Quality: Convergence, Not Cosmetics
Mesh quality metrics are proxies for discretization error. The decisive evidence is not a single quality report but a mesh convergence study: refine the mesh systematically and show that a monitored quantity (peak stress, first natural frequency, drag coefficient, intrusion) asymptotes. A practical acceptance pattern is to run at least three refinement levels and require the change between the two finest to fall within a defined tolerance — commonly 3–5% for stiffness and stress, tighter for modal frequencies.
Key metrics to review, with typical automotive targets:
- Jacobian / element distortion: minimum scaled Jacobian ≥ 0.3 for structural hex/tet meshes; below this, stiffness is artificially stiffened or softened.
- Aspect ratio: keep below ~5 in critical stress regions; thin boundary-layer elements in CFD legitimately exceed this.
- Skewness / warpage: shell elements should stay below ~0.6 skew and ~5° warpage for accurate bending.
- Element size transition: avoid jumps greater than ~2:1 between adjacent regions to limit interpolation error.
A frequent failure symptom is a stress singularity at a re-entrant corner or a point load: refining the mesh makes peak stress grow without bound. This is not a mesh defect but a modeling defect — the reviewer should require a fillet, a distributed load, or a defined extraction path away from the singularity.
Boundary Conditions: The Most Common Source of Error
Boundary conditions encode engineering assumptions, and they are where models most often diverge from test. Every constraint and load must be traceable to a physical fixture, a test rig, or a documented load case. Reviewers should ask: what is constrained, in which DOF, and why?
Common pitfalls include over-constraining (fully fixing a face that in reality rotates), under-constraining (free rigid-body modes appearing as near-zero eigenvalues), and applying loads at nodes rather than through realistic pressure or coupling distributions. For a suspension lower control arm, for example, a bushing should be represented by a stiffness matrix or a connector element, not a rigid clamp — a rigid clamp can raise predicted stiffness by 20–40% and invalidate the correlation.
Material Data: Traceability and Rate Dependence
Material inputs must be traceable to a source: a coupon test report, a supplier datasheet, or a validated material card. For metals, confirm the elastic modulus, Poisson's ratio, yield definition (0.2% offset), and hardening curve. For plastics and elastomers, rate dependence and temperature dependence are not optional — a crash-relevant strain rate can shift yield stress by 50% or more. For composites, ply properties, fiber orientation, and failure criteria must match the manufacturing process.
Reviewers should flag any material card with no documented source, and any use of room-temperature properties in a high-temperature or high-rate load case. Where scatter exists, run a sensitivity study rather than a single nominal value.
Correlation with Physical Tests
Validation is demonstrated by comparing simulation output to measured test data at matched locations and conditions. The comparison must be quantitative. Define correlation metrics before running: peak value error, time-to-peak error, phase shift, and curve shape (e.g., a correlation coefficient or an objective function).
A reproducible correlation workflow for a frontal impact sled test might proceed as follows. Inputs: a validated material card set, a deceleration pulse measured on the sled, and accelerometer data at the B-pillar. Execution: apply the measured pulse as a prescribed acceleration to the sled base, run the explicit solver, and extract acceleration at the same B-pillar node. Expected output: a time-history overlay. Acceptance: peak acceleration within ±10%, time-to-peak within ±2 ms, and no unexplained secondary peaks. If the simulated peak is early and high, suspect excessive stiffness in the front rail or a missing trigger; if late and low, suspect under-represented mass or a soft material card.
An Integrated V&V Review Checklist
| Area | Evidence Required | Acceptance Criterion |
|---|---|---|
| Mesh | Convergence study, quality report | Monitored quantity change < 5% at finest level |
| Boundary conditions | Constraint map, load traceability | No spurious rigid-body modes; loads match rig |
| Material | Source card, rate/temperature data | Documented source; rate effects included |
| Correlation | Overlay plots, error metrics | Peak ±10%, timing ±2 ms, shape matched |
The figure below summarizes how these four elements feed a single validation decision, and where the loop closes when correlation fails.
The loop is the essential message: when correlation fails, the reviewer returns to the specific input — mesh, boundary condition, or material — that the error pattern implicates, rather than adjusting the result. Documenting each iteration and its justification is what turns a simulation into defensible technical evidence for sign-off.