📚 Course ContentsChapter 3 Supplier qualification

3. Supplier qualification

Selecting a crash CAE supplier is not a procurement formality. A vendor that produces visually convincing animations but cannot reproduce a regulatory load case, or that reports peak acceleration without filtering per SAE J211, will inject errors that surface only after tooling is cut. Qualification therefore has to answer three distinct questions: Can the vendor do the work? (capability assessment), Can the vendor do our work correctly? (benchmark problems), and Is the vendor's organization trustworthy enough to depend on? (technical due diligence). These are sequential filters, not parallel checkboxes — a vendor that fails the benchmark should not proceed to due diligence regardless of how impressive its facility tour was.

Capability Assessment: What to Verify Before Any Model Is Built

Capability assessment establishes the vendor's technical envelope. It should be evidence-based: ask for solver versions, not solver names; ask for correlation plots, not correlation claims. The core dimensions are:

Solver and toolchain depth. LS-DYNA is the de facto standard for explicit crash, but confirm which versions are in production (e.g., R11.x vs. R13.x), whether the vendor maintains licensed seats for implicit follow-up (NVH, static stiffness), and whether pre/post-processing is standardized on ANSA/META, HyperMesh/HyperView, or a mix. Mixed toolchains are acceptable but require a documented translation and QA step.
Model build capability. Mid-surface extraction, shell meshing with target element quality (aspect ratio, warpage, Jacobian, min/max angle), spot weld and adhesive modeling methods (CWELD, ACM, solid hexa), and material card authorship. Ask which material cards the vendor writes from test data versus copies from the solver library — the answer reveals a great deal.
Correlation and validation experience. Request a correlation package for a component-level test (e.g., a bumper beam drop tower or a B-pillar three-point bend) showing test vs. simulation curves, not just a final percentage. The ability to explain why a curve diverges matters more than the divergence itself.
Regulatory and consumer-protocol familiarity. FMVSS 208/214/216/301, ECE R94/R95, and consumer programs such as Euro NCAP and C-NCAP each impose specific setup, instrumentation, and filtering conventions. A vendor that has never run a full frontal ODB at 64 km/h with a THOR or Hybrid III ATD will need supervision.
Capacity and staffing. Number of full-time CAE engineers, peak concurrent model capacity, and whether the vendor uses subcontractors. Undisclosed subcontracting is a common source of quality drift.

A practical screening artifact is a capability matrix scored against weighted requirements. The weighting should reflect program risk: for a greenfield EV platform with a tight timing plan, model build throughput and correlation depth outweigh cost.

Benchmark Problems: The Only Reliable Predictor

Benchmark problems are controlled, blind tests that reveal how a vendor actually works. They must be scoped so that the vendor cannot reverse-engineer the expected answer, and they must be graded against objective criteria. Three benchmark types are worth the effort:

Regulatory load case reproduction. Provide a complete, self-consistent model (mesh, materials, boundary conditions, initial velocity) for a simplified body-in-white or a sub-assembly, and ask the vendor to run a full frontal rigid barrier impact at 50 km/h and report occupant compartment intrusion, peak deceleration, and energy balance. The model should include at least one deliberate modeling ambiguity — for example, an unspecified spot weld failure criterion — to see whether the vendor flags it or silently guesses.
Correlation challenge. Supply raw test data (accelerometer channels, filtered and unfiltered) and a component model, and ask the vendor to correlate within a stated tolerance. Grade on the correlation methodology: time-step alignment, filtering per SAE J211 CFC 60 or CFC 180, and whether the vendor reports both phase and magnitude error.
Debugging exercise. Provide a model that terminates early with a negative volume error or an unrealistic hourglass energy fraction, and ask for a written diagnosis and fix. This tests the vendor's ability to read the d3hsp and messag files rather than simply re-running with a smaller time step.

Grading should use a rubric with explicit pass thresholds. A representative rubric for the correlation challenge might require: peak force within ±10%, pulse duration within ±5 ms, and a documented filtering rationale. Vendors that meet the numeric targets but cannot explain their filtering choices should be marked as conditional.

Benchmark scoping rule: The benchmark must be solvable within the vendor's stated capacity (typically 3–5 working days) and must not require proprietary OEM data that the vendor could not reasonably possess. If the benchmark requires your internal material cards, provide them — the test is about execution, not data ownership.

Technical Due Diligence: Process, Data, and Continuity

Due diligence moves from can they to will they, consistently. The focus is on process control, data integrity, and organizational resilience. Key areas:

Model QA process. Is there a documented checklist before a model is released? Does it cover element quality, mass balance, initial penetration, energy balance, and time-step stability? Ask to see a completed checklist from a recent project, redacted as needed.
Version control and traceability. How are model revisions tracked? A vendor using dated folders and email attachments is a risk; one using Git, SVN, or a PLM system with commit history is far easier to audit.
Data security. CAD and CAE data are among the most sensitive assets an OEM holds. Confirm NDA coverage, access controls, and whether the vendor's engineers work on shared or isolated hardware. Ask about incident history — a vendor that has never had a breach may simply never have looked.
Staff continuity. Key-person dependency is a real risk. Ask for the named lead engineer's tenure and the backup structure. A vendor with a single LS-DYNA expert and high turnover is a single point of failure.
Reporting cadence and escalation. Define what a weekly report contains (model status, open issues, correlation progress, risks) and how problems are escalated. Vague reporting is a leading indicator of vague engineering.

The due diligence output should be a risk register with mitigations, not a pass/fail verdict. For example: "Vendor uses dated folders for version control — mitigation: require weekly model snapshots delivered to OEM PLM." This keeps the qualification actionable.

Putting the Three Stages Together

The qualification sequence should be run as a gated process. Stage 1 eliminates vendors that lack the basic technical envelope — this is cheap and fast, often done through a questionnaire and a reference call. Stage 2 is the expensive part: it consumes vendor engineering time and requires your own engineers to prepare models and grade results. Budget 2–3 weeks for a proper benchmark cycle. Stage 3 is a documentation and interview exercise that can run in parallel with commercial negotiation, but its findings should be able to veto the award.

A common failure mode is to treat the benchmark as a formality and award on price. The cost of a missed negative volume error or an unfiltered acceleration peak is not the vendor's invoice — it is the program delay and the potential recall exposure. The qualification process exists to make that risk visible before it is expensive.

Practical tip: Keep a qualification dossier for each vendor containing the capability matrix, benchmark submission, rubric scores, and risk register. When a program is later audited or a dispute arises, this dossier is the evidence that the selection was technically justified.