V&V Plan and Report Requirements
Concept: decide first, record afterwards
Overview
A customer asked for documentation showing our analysis is trustworthy. What should V&V documents contain, and how much?
Split it in two: a plan written before the analysis, and a report written afterwards. The plan fixes in advance what the analysis is for, which quantities matter and how close is good enough. The report records the verification and validation carried out per the plan, the pass/fail judgement and its basis. The yardstick for how much to write is whether a reader could reproduce the same judgement.
Plan contents
| Item | What to write |
|---|---|
| Purpose and intended use | Which decisions the results support (design changes, test reduction, etc.) |
| Quantities of interest | Quantities and locations to compare (peak stress, strain at a point, displacement) |
| Acceptance criteria | Numerical, e.g. |E| within 2u_val and difference within 5% |
| Verification methods | Code verification against known solutions; solution verification on three meshes |
| Experiments | Test conditions, measurement points, uncertainty estimation |
| Roles and records | Who performs and checks; archiving of input files |
Report contents
- Deviations from the plan: what changed and why.
- Code and solution verification results: differences from known solutions, observed order, numerical uncertainty.
- Validation results: simulation and experiment, difference, uncertainty breakdown, judgement.
- Domain of applicability: the range of validated conditions and cautions outside it.
Why “plan first” became a standard
Nuclear and aerospace have long required quality-assurance procedures because analyses support safety decisions. Even so, models kept being tuned after seeing results, and only the matching cases reported. ASME V&V 10 (2006), covering structural V&V, recommends planning first, fixing acceptance criteria, and keeping experiment and analysis teams somewhat separate. V&V 40 (2018) for medical devices added the idea of scaling the required evidence to how consequential the analysis's use is.
Example: steel cantilever bracket
Plan entries
An illustrative example. A steel cantilever 200 mm long, 20 mm wide and 10 mm thick (Young's modulus 205 GPa) carries 300 N at its tip. The quantities of interest are the strain 50 mm from the root and the peak stress at the root fillet. The acceptance criterion is “strain difference |E| within 2u_val”.
Code verification
Beam theory gives a second moment of area of 1,666.7 mm⁴, tip deflection 2.341 mm, root bending stress 180.0 MPa and 659 μ strain at the gauge. Confirm the solver reproduces these on a simple model without the fillet, and record the model and the differences in the report.
Solution verification
Suppose the fillet peak stress on three meshes, each halving the element size, is 212.4, 214.9 and 223.5 MPa from fine to coarse. The observed order is 1.78, the fine-mesh GCI 0.60% and the extrapolated value 211.4 MPa. Stress-concentration values can rise or fall with refinement, so showing the trend on three levels is the baseline.
Validation and judgement
| Item | Value (μ strain) | Notes |
|---|---|---|
| Simulation S | 659 | At the gauge |
| Experiment D | 640 | Strain gauge |
| Difference E | 19 (2.9%) | |
| u_num | 1.3 | From GCI |
| u_input | 16.1 | Load 1%, modulus 2%, thickness |
| u_D | 12.8 | Gauge 2% |
| u_val | 20.6 | 2u_val = 41 |
The difference is 19 and 2u_val is 41, so it meets the criterion and passes.
Yes. But the report should give the uncertainty breakdown, not just “pass”. Here input uncertainty (modulus and thickness) dominates and numerical error hardly matters, so if tighter criteria are needed the fix is material testing and thickness measurement, not the mesh. Also, only one strain point was validated; the fillet peak stress was not checked experimentally. Stating that in the domain of applicability is the honest way to write it.
Writing tips
Cautions
- Set acceptance criteria before seeing results, and record the date and approver.
- Archive input files, solver version and meshes so results can be regenerated.
- Report experimental scatter and instrument accuracy, not just the mean.
- Separate data used for calibration from data used for validation.
- Keep failures and unexpected results rather than deleting them.
“Could you read it three years later?”
The real reader of a V&V document is less the reviewing customer than whoever redoes the analysis for a design change years later — often yourself. What hurts then is not knowing where the mesh was refined, why a boundary condition had that value, or which test data were used. A one-line reason beside each decision saves a great deal of rework. Think of the report as a record of decisions, not an advertisement of results.
Common mistakes
Mistakes and fixes
| Mistake | Effect | Fix |
|---|---|---|
| Setting criteria afterwards | Judgement loses credibility | Fix them in the plan |
| Reporting only the difference | No basis for pass/fail | Add uncertainty and breakdown |
| Single-mesh report | Numerical error unknown | Show convergence on three levels |
| No domain of applicability | Misuse outside validated conditions | State the validated range |
| Not archiving inputs | Not reproducible | Version-control and archive |
I'd like to check the example calculations myself.
Try the beam deflection calculator for deflection and stress, the strain gauge simulator for strain measurement, the stress concentration tool for stress concentration and the Richardson extrapolation tool for GCI. Procedures are detailed in ASME V&V 10, code verification, solution verification and pedigree matrix.
Related Topics
Experience the theory firsthand with the interactive simulator for this field
All Simulators