3. Key terminology and standards

Uncertainty budgets for virtual crash assessment do not exist in a standards vacuum. Four bodies of guidance shape how we define, quantify, and defend uncertainty claims: the GUM (JCGM 100:2008, Guide to the Expression of Uncertainty in Measurement), the ASME V&V series (notably ASME V&V 10 and V&V 20), ISO 26262 (functional safety for road vehicles), and the crashworthiness validation frameworks that govern physical test correlation (FMVSS, ECE regulations, Euro NCAP, and OEM-internal sign-off procedures). Each uses overlapping but non-identical vocabulary. A credible virtual assessment claim must reconcile them rather than pick one and ignore the rest.

GUM: the metrological backbone

The GUM supplies the formal machinery for combining uncertainty contributions. Its central construct is the measurement model Y = f(X₁, X₂, …, Xₙ), where each input quantity Xᵢ carries an estimate xᵢ and a standard uncertainty u(xᵢ). The combined standard uncertainty follows the law of propagation:

u_c²(y) = Σᵢ (∂f/∂xᵢ)² u²(xᵢ) + 2 Σᵢ Σⱼ<ᵢ (∂f/∂xᵢ)(∂f/∂xⱼ) u(xᵢ, xⱼ)

Two GUM concepts matter most for crash simulation. First, Type A evaluation derives uncertainty from repeated observations (e.g., scatter across a DOE of 30 runs), while Type B evaluation derives it from prior knowledge—manufacturer tolerances, calibration certificates, engineering judgment. Second, the coverage factor k converts combined standard uncertainty into an expanded uncertainty U = k·u_c; k = 2 corresponds roughly to 95% coverage for a normal distribution. In crash work, a common error is treating solver-to-test deviation as a single Type B number when it is actually a composite of Type A scatter and Type B model-form error. The GUM's sensitivity coefficients ∂f/∂xᵢ are also directly useful: they tell you which input (shell thickness, strain-rate factor, friction coefficient) dominates the output variance, and therefore where to spend characterization effort.

ASME V&V: separating verification from validation

ASME V&V 10 (computational solid mechanics) and V&V 20 (computational fluid dynamics and heat transfer) draw a distinction that crash engineers must respect. Verification asks whether the equations are solved correctly—code verification (method of manufactured solutions, order-of-accuracy tests) and calculation verification (mesh convergence, time-step sensitivity). Validation asks whether the right equations are solved—comparison of simulation output against experimental data with quantified uncertainty in both. The ASME validation metric is not a simple percentage error; it compares the simulation prediction interval against the experimental measurement interval, and a validation comparison is only meaningful when both intervals are stated.

This is where the uncertainty budget becomes a validation artifact. ASME V&V requires that you document the validation uncertainty u_val, which combines numerical uncertainty u_num, input parameter uncertainty u_input, and (where applicable) model-form uncertainty u_model. The comparison error E = S − D (simulation minus data) is judged against u_val: if |E| ≤ u_val, the model is validated at that level of uncertainty. A crash model that matches peak acceleration within 5% but has u_val of 12% has not been validated—it has been coincidentally matched.

ISO 26262: the safety argument context

ISO 26262 does not prescribe uncertainty quantification methods, but it constrains what a virtual crash claim must support. The standard's ASIL classification (A through D) sets the rigor of evidence required for safety goals. For occupant protection functions, the relevant clauses concern hardware/software integration testing, confirmation measures, and the treatment of systematic versus random failure. Crash simulation enters as a tool for design verification and, increasingly, as part of the safety case for restraint deployment logic. ISO 26262-5 and -6 require that tool confidence be established (tool qualification, or a documented argument that tool errors cannot lead to undetected safety violations). The practical implication: your uncertainty budget must distinguish aleatory variability (manufacturing scatter, occupant posture) from epistemic uncertainty (model form, material card fidelity), because ISO 26262's safety argument treats them differently—aleatory variability is managed by design margin, epistemic uncertainty by evidence and validation.

Crashworthiness validation frameworks: where the claim lands

Regulatory and consumer frameworks (FMVSS 208, ECE R94/R95, Euro NCAP frontal offset and MPDB) define pass/fail criteria on measured physical quantities: HIC, chest deflection, femur load, intrusion. A virtual assessment claim is only useful if it maps simulation outputs onto those same quantities with a stated uncertainty. The alignment problem is therefore a traceability problem: every input in the GUM model must trace to a physical source (material coupon test, tolerance drawing, calibration record), every ASME validation metric must trace to a test configuration, and every output must trace to a regulatory or consumer criterion.

Reconciling the vocabularies in practice

The four frameworks use different words for related ideas, and a working uncertainty budget must translate between them. GUM's combined standard uncertainty maps onto ASME's validation uncertainty only when numerical and model-form contributions are explicitly included. ISO 26262's random failure maps onto GUM's aleatory variability, while its systematic failure maps onto epistemic model-form uncertainty—but ISO 26262 does not require a coverage factor, whereas GUM does. Crashworthiness frameworks care about margin to criterion, which is a decision quantity, not a metrological one.

A practical reconciliation uses a single uncertainty budget table with columns for: quantity, nominal value, standard uncertainty, distribution/type (A or B), sensitivity coefficient, variance contribution, and source reference. The source reference column is what makes the budget auditable under all four regimes—it points to a coupon test report (GUM Type B), a mesh convergence study (ASME calculation verification), a tool qualification record (ISO 26262), or a sled test (crash validation). When a reviewer from any of these communities asks "where does this number come from?", the same table answers.

Worked example: HIC prediction with a decomposed budget

Consider a frontal sled model predicting driver HIC. Suppose the nominal prediction is HIC = 620 against a criterion of 700. The budget decomposes as follows (values are illustrative, not measured):

Combining in quadrature gives u_c ≈ √(7.2² + 2.7² + 4.0² + 6.0² + 9.5²) ≈ 13.4 HIC units. With k = 2, U ≈ 27, so the claim is HIC = 620 ± 27 (95% coverage). The margin to the 700 criterion is 80, which exceeds U—the claim is defensible. Note that posture dominates; if the margin were tighter, the correct response would be to characterize posture more thoroughly (reduce Type A scatter) rather than to tighten the belt tolerance, because the sensitivity analysis shows where the variance lives.

The alignment across standards is therefore not a matter of choosing one vocabulary. GUM gives the arithmetic, ASME V&V gives the verification/validation discipline and the comparison metric, ISO 26262 gives the safety-evidence context and the aleatory/epistemic split, and the crashworthiness frameworks give the decision criteria. A virtual crash assessment claim is credible precisely when all four are visible in the same traceable budget.