PROJECT 002 · VN1 · REFERENCE STUDY · BIOMEDICAL FLOW
The FDA benchmark nozzle, in both directions, at five Reynolds numbers.
The CDRH benchmark nozzle compared against the inter-laboratory PIV measurement, framed as an ASME V&V 40 credibility assessment. This is the study we run to show a device buyer exactly what our evidence package looks like, on the one geometry their reviewer already knows.
five laboratory datasets per Reynolds number, both flow directions, published by the FDA
Conditions
- Geometry
- 12 mm tube, sudden contraction and 10 deg conical diffuser
- Throat
- 4 mm diameter, 40 mm long, both flow directions
- Fluid
- Blood analogue, density 1056 kg/m3, viscosity 0.0035 Pa s
- Range
- Throat Reynolds 500, 2000, 3500, 5000, 6500
- Solver
- OpenFOAM, with the transition treatment declared per regime
- Framing
- ASME V&V 40 context of use, question of interest, model risk
The comparator
Validation means agreement with a measurement this studio did not produce. Ours is named, cited, and downloadable by the reader.
FDA CDRH inter-laboratory PIV of the benchmark nozzle: five independent laboratory datasets per Reynolds number, both directions, with centreline and radial velocity, Reynolds stresses, wall pressure and wall shear stress
SOURCE: U.S. FOOD AND DRUG ADMINISTRATION, CRITICAL PATH INITIATIVE
https://github.com/OSEL-DAM/CFD-and-Blood-Damage-BenchmarksPublic domain, no registration. The nciphub.org address cited across most of the literature no longer resolves; this is the live source, and we link it so you can pull the same files we do.
How the study is run
- 01
Context of use, stated first
What the model is used to predict, the question of interest, and the model-risk placement with its reasoning, written in ASME V&V 40 form before any solver runs.
- 02
Thresholds from the measured spread
The published inter-laboratory RMS difference is about 20 percent at Re 3500 and 15 percent at Re 6500. Acceptance bands are set from that spread rather than from a flattering tolerance we chose ourselves.
- 03
Both flow directions, every Reynolds number
Ten cases, no omissions. The benchmark defines both directions, and running one of them is a different study.
- 04
The transitional regime treated honestly
Re 3500 is where the published round robin showed the field performing worst. We state the treatment used, and where a model cannot capture the regime we publish that at full size rather than selecting the Reynolds numbers that flatter it.
- 05
Claim discipline is automated
A lint gate blocks the vocabulary that would misrepresent a simulation study as a device conclusion. Nothing here supports a safety, efficacy or biocompatibility statement, and the copy is checked mechanically rather than by good intentions.
The bar we hold to
11 acceptance criteria in three bands. Every threshold is cited to the procedure or the published statistic it came from, written into the SOW, and fixed before the first production run. A threshold with no basis is not worth stating, and a threshold that moves after a result is read is not a threshold.
Verification
4 CRITERIAThat the numerical answer stands up on its own terms: convergence, grid, wall treatment, reproduction.
| ID | CRITERION AND BASIS | THE BAR |
|---|---|---|
| V1 | Thresholds committed before the measured data is openedMachine-checkable from the git history | Threshold commit timestamp precedes the dataset download timestamp |
| V2 | Iterative convergence at each Reynolds numberASME V&V 20 | Quantity of interest drift under 0.05 % over the final 2000 iterations |
| V3 | Solution stationarity at the higher Reynolds numbersThe sudden-expansion jet is naturally unsteady. Where a steady solution does not exist we say so and change the treatment, rather than reporting a residual floor as convergence. | A stationary solution, or a declared transient treatment with the reason published |
| V4 | Grid convergence with reported order of accuracyASME V&V 20 and Richardson extrapolation | Three grids, constant refinement ratio, monotonic convergence |
Requirements
3 CRITERIAThe commitments we hold ourselves to on every job, including the ones about how we behave with a threshold.
| ID | CRITERION AND BASIS | THE BAR |
|---|---|---|
| R1 | Claim-discipline lint passes on the published pageAutomated. The list covers safe, proves, certified, approved, validated used loosely, and any efficacy or biocompatibility object. | Zero occurrences of the banned claim vocabulary |
| R2 | Every run reproduces from a clean checkoutStudio release rule | One command, pinned container digest, SHA-256 per file |
| R3 | Both flow directions at every Reynolds numberThe benchmark defines both directions | Ten cases, no omissions |
Validation
4 CRITERIAAgreement with a measurement we did not produce. This is the band most published CFD leaves empty.
| ID | CRITERION AND BASIS | THE BAR |
|---|---|---|
| Va1 | Centreline axial velocity against the inter-laboratory bandPublished RMS difference between laboratories, about 20 % at Re 3500 and 15 % at Re 6500 | Inside the measured inter-laboratory spread at Re 500 and 2000 |
| Va2 | Jet breakup location at Re 3500The transitional regime, where the published round robin showed the widest scatter | Within one throat diameter of the measured location |
| Va3 | Wall static pressure distributionThe measurement publishes means with 95 % confidence intervals | Inside the 95 % confidence interval of the measurement at every tap |
| Va4 | Wall shear stress in the throatTime-resolved PIV with published uncertainty quantification | Inside the Monte-Carlo uncertainty band of the PIV-derived value |
What the study shows
- Agreement with an inter-laboratory measurement this studio did not produce, at five Reynolds numbers, against bands set from the measured spread.
- A grid convergence study with the observed order of accuracy and the Richardson extrapolation reported.
- The regime where a model fails, published at full size with the mechanism explained in plain language.
- An ASME V&V 40 credibility framing: context of use, question of interest, model risk, and what each credibility goal is supported by.
What it does not show
- Any device claim. This is an idealised benchmark geometry used to demonstrate a workflow, not an evaluation of a product.
- Any safety, efficacy, biocompatibility or clinical statement, and none can be derived from it.
- Blood damage or hemolysis prediction. The comparator is velocity, pressure and wall shear stress.
- Patient-specific anatomy or physiological pulsatility.
- Regulatory status of any kind.
Every study here states the limit of what it covers. A reviewer looks for that limit first.
The same method, on your geometry.
Biomedical devices engagements run the ladder from a two-week threshold study through to design closure. The check matrix you see here is the artefact you get first, written for your case, before anyone runs anything.