3. Establishing a CAE quality management system
A CAE quality management system (QMS) is not a single document or a software configuration. It is the controlled set of policies, procedures, records, and technical standards that makes simulation results defensible — to an internal reviewer, a notified body, or a regulator. Three external frameworks dominate the landscape, and they answer different questions. ISO 9001 asks whether the organization consistently controls its processes. ASME V&V 10 and V&V 20 ask whether a specific model is credible for a specific use. FDA and EMA guidance asks whether simulation evidence is acceptable inside a regulated submission. A working CAE QMS threads all three together without duplicating them.
What each framework actually governs
ISO 9001 is process-oriented and domain-agnostic. It does not prescribe simulation methods; it requires that you define your processes, control documented information, manage competence, and act on nonconformities. For CAE, the practical translation is that model development, verification, validation, and release must be described as controlled processes with defined inputs, outputs, acceptance criteria, and records. ISO 9001 certification alone proves nothing about model accuracy — it proves that you can demonstrate you followed your own defined process.
ASME V&V 10 (computational solid mechanics) and V&V 20 (computational fluid dynamics) are technical credibility frameworks. They introduce the vocabulary that makes simulation evidence auditable: the context of use, the validation metric, the validation comparison, and the distinction between verification (solving the equations right) and validation (solving the right equations). V&V 10-2019 structures credibility assessment around risk-informed activities, and V&V 20-2009 provides the CFD-specific companion with its own verification and validation procedures. Neither standard certifies a model as "valid" in general; both require you to state the context of use and demonstrate credibility within it.
FDA and EMA expectations are evidence-oriented. The FDA's 2023 guidance on computational modeling and simulation in medical device submissions, and the ASME V&V 40 standard it references, establish that a model's credibility should be commensurate with its influence on the regulatory decision. EMA's modeling and simulation qualification pathway serves a similar function for medicinal products. The common thread: regulators do not accept "the model is validated" as a claim. They accept a documented credibility argument that links the model's intended use, its verification and validation evidence, and its influence on the decision being made.
The credibility argument as the integrating artifact
The single most useful construct for integrating these frameworks is the credibility argument — a structured record that states the context of use, identifies the model's influence on the decision, lists the verification and validation activities performed, and justifies why the evidence is sufficient. This is where ISO 9001's document control meets ASME's technical rigor meets the regulator's risk-informed expectations.
The diagram below shows how the three frameworks feed into a single credibility argument, and how that argument drives the release decision.
Building the QMS: from procedure to record
A practical CAE QMS can be built in four layers, each traceable to one or more of the frameworks above.
Layer 1 — Process definition. Write a simulation procedure that names the phases (requirements capture, model build, verification, validation, uncertainty quantification, review, release) and the entry/exit criteria for each. This satisfies ISO 9001 clause 8.5 (production and service provision) and gives ASME V&V activities a home. Keep it short enough that engineers actually use it; a 40-page procedure that nobody reads is a nonconformity waiting to happen.
Layer 2 — Technical standards and templates. Define the templates that carry the technical content: a context-of-use form, a verification checklist (mesh convergence, time-step independence, solver tolerance, conservation checks), a validation plan that names the metric and acceptance threshold before results are seen, and an uncertainty budget. These templates operationalize ASME V&V 10/20 without requiring engineers to read the standards cover to cover.
Layer 3 — Records and traceability. Every simulation that supports a decision must be traceable to its inputs (geometry revision, material data source, boundary conditions), its solver version and settings, its verification and validation evidence, and the decision it informs. ISO 9001 clause 7.5 (documented information) requires control; FDA/EMA traceability expectations require that a reviewer can reconstruct the argument without interviewing the engineer.
Layer 4 — Review and release gate. A release gate is where the credibility argument is judged sufficient or insufficient. The gate should be risk-tiered: a low-influence model (e.g., a screening study that informs a design trade but not a safety claim) may need only verification and a sensitivity check; a high-influence model (e.g., a device performance claim in a 510(k)) needs a full validation package with quantified uncertainty and a documented sufficiency justification.
A worked example: risk-tiered release gate
Consider a finite element model of a polymeric implant component used to support a fatigue-life claim in a regulatory submission. The model's influence on the decision is high, so the credibility level required is high. The table below shows a release-gate checklist for this case; the acceptance criteria are illustrative and would be set by the organization's procedure.
| Gate item | Evidence required | Acceptance criterion (example) | Framework anchor |
|---|---|---|---|
| Context of use | Signed statement naming the decision, the quantity of interest, and the operating envelope | Reviewed and approved before model build | ASME V&V 10 §3 |
| Code verification | Method of manufactured solutions or benchmark comparison | Observed order of accuracy within 10% of theoretical | ASME V&V 10 §4 |
| Calculation verification | Mesh convergence study; time-step independence | Peak stress change < 2% between successive refinements | ASME V&V 10 §5 |
| Validation | Comparison against physical test data at relevant load levels | Validation metric within pre-defined tolerance; uncertainty band reported | ASME V&V 10 §6 |
| Uncertainty quantification | Input uncertainty budget; propagation to quantity of interest | Total uncertainty stated and shown not to overturn the decision | ASME V&V 20 / V&V 40 |
| Traceability | Record linking geometry revision, material data, solver version, and decision | Complete and retrievable by an independent reviewer | ISO 9001 §7.5; FDA guidance |
| Sufficiency justification | Written argument that the evidence is commensurate with model influence | Approved by independent reviewer | FDA 2023 guidance; ASME V&V 40 |
Two failure modes are worth calling out. First, validation performed after the decision: if the acceptance threshold is chosen after seeing the comparison, the evidence is not credible regardless of how good the numbers look. The validation plan must fix the metric and threshold in advance. Second, verification mistaken for validation: a converged mesh and a stable solver prove the equations were solved correctly, not that they represent the physics. A model can be perfectly verified and still be wrong for its context of use.
Practical implementation notes
Start with the highest-influence model in the portfolio and build its credibility argument first; this exposes gaps in templates, data availability, and reviewer competence faster than any gap analysis. Map each QMS procedure to the clause it satisfies so that internal audits and certification bodies can trace coverage without reinterpreting the system. Keep the context-of-use statement under version control and treat changes to it as changes to the model's validity claim — a model validated for one load regime is not automatically valid for another. Finally, train reviewers on the distinction between verification and validation, because the most expensive QMS failures are not missing documents but reviewers who accept the wrong kind of evidence.
The QMS is working when an independent reviewer can pick up a released model, read the credibility argument, and either agree with the release decision or identify precisely which evidence is insufficient — without needing to ask the original engineer anything.