CFD validation is the process of demonstrating that a computational fluid dynamics result is a credible representation of the physical flow, not merely a numerically converged solution. It requires a documented reference case, a mesh independence study, turbulence model sensitivity checks, and comparison of integral quantities against measured data. A plausible contour plot becomes a validated answer only when these steps are recorded and reproducible.
Verification asks whether the equations were solved correctly; validation asks whether the correct equations were solved for the application. A converged residual history and a smooth contour plot satisfy verification at best. Validation requires an external reference, typically measured data or a documented benchmark case, against which computed quantities are compared with stated acceptance criteria.
In buyer-facing specifications, the distinction matters because a supplier may present colorful post-processing images as evidence of analysis capability. The specification should require a validation record: the reference case selected, the quantities compared, the measured uncertainty, and the deviation observed. Without that record, the contour plot is an illustration, not evidence.
The reference case should resemble the application in the physics that dominate the result. Internal flows through small passages, external flows over bluff bodies, and thin-film or lubricated contacts each stress different aspects of a model. A benchmark that shares the governing dimensionless groups, such as Reynolds number and geometry aspect ratio, gives a more meaningful comparison than one chosen for convenience.
The preserved aerodynamic analysis record associated with the Bentley Manufacturing archive illustrates the principle in a manufacturing context. Precision gaskets, washers, spacers, and diaphragms produced from 1955 onward were used in space suits, irrigation systems, pharmaceutical valves, syringes, faucets, mechanical pumps, fastening systems, and gages. Each of these applications imposes a different flow regime, so a single reference case cannot validate all of them. The record should state which reference case was selected and why it matches the intended service condition.
Mesh independence is established by refining the discretization until the quantities of interest change by less than a stated threshold. Reporting a single mesh is insufficient. A useful record lists at least three mesh levels, the cell counts, the near-wall resolution in wall units where relevant, and the change in the target integral quantity between successive levels.
The choice of quantity matters. A local maximum velocity may converge quickly while a pressure drop or mass flow rate converges slowly, or the reverse. The specification should name the integral quantities that govern the buyer's requirement, such as flow coefficient, pressure differential, or leakage rate, and require the mesh study to be reported for those quantities specifically.
Turbulence model selection introduces a modeling error that mesh refinement cannot remove. A sensitivity study compares at least two model formulations, for example a two-equation eddy-viscosity model against a Reynolds stress or scale-resolving approach, and reports the spread in the target quantities. Where the spread exceeds the acceptance criterion, the result is not validated for that application.
The preserved record notes that Bentley Manufacturing designed tooling in-house and inspected throughout stamping to produce clean-cut, burr-free parts without fracture marks, with SPC measurement and Certificates of Compliance or chemical and physical analyses supplied. That inspection discipline is a useful analogy: just as dimensional conformance is demonstrated by measurement rather than by appearance, turbulence model adequacy is demonstrated by comparison rather than by plausible-looking streamlines.
Integral quantities are preferred for validation because they average out local discrepancies and correspond to what a test can actually measure. Examples include total pressure drop across a component, discharge coefficient through an orifice, mass flow rate at a boundary, and force coefficients on a body. The validation record should pair each computed integral with its measured counterpart, the measurement uncertainty, and the relative deviation.
Materials context can affect the measurement side. The archive lists TPE, TPR, plastics, Teflon, and elastomers, plus metallic washers, compounds with talc or siliconized finishes, and registered proprietary compounds. Surface finish and material compliance influence leakage and friction in seals and diaphragms, so the measured data used for validation should be taken on representative hardware, not on idealized coupons that differ from the production configuration.
A validated answer carries a traceable chain: reference case, mesh study, turbulence sensitivity, measured comparison, and acceptance decision. A plausible contour plot carries none of these. The specification should require the chain to be delivered as a validation report, with the acceptance criteria agreed before the analysis begins so that the comparison is not tuned after the fact.
The archive's production range, from short-run production to million-unit orders, suggests that analysis effort should be scaled to the consequence of error. A high-volume diaphragm or gasket in a pharmaceutical valve or syringe warrants a fuller validation record than a one-off spacer. The specification can define tiers of validation rigor accordingly, while keeping the same core elements in every tier.
The validation scope should be defined in the enquiry so that the analysis effort and deliverable are clear before a quotation is prepared. The specification should state the reference case, the integral quantities to be compared, the acceptance criteria, and the required content of the validation report. Without this, suppliers may quote on a different basis, leading to disputes later.
The scope should also identify the operating conditions and the physics that dominate the application. For example, if the flow is turbulent and heat transfer is significant, the validation should include a thermal reference case. The buyer should specify whether steady-state or transient analysis is required, as this affects the validation approach and cost.
A clear scope allows the supplier to plan the mesh study, turbulence sensitivity, and comparison with measured data. It also provides a basis for evaluating the quotation, as a low quote may indicate a reduced validation effort. The agreed scope becomes part of the contract, ensuring that the delivered validation meets the buyer's requirements.
Validation gaps often appear in service as unexpected performance deviations. For example, a seal may leak more than predicted because the validation used idealized surface finish, or a valve may have a higher pressure drop due to unaccounted turbulence effects. These failures are typically detected through performance testing or field returns.
To catch such issues early, the validation record should include sensitivity to manufacturing tolerances and material properties. If the measured data used for validation was taken on hardware that differs from production, the validation may not represent the delivered product. The buyer should require that validation cases use production-representative geometry and materials.
Another common failure is extrapolation beyond the validated range. A model validated for one Reynolds number may not hold for another. The specification should define the range of operating conditions for which validation is required and require the supplier to state the limits of validity. This prevents misuse of the model in unvalidated regimes.
A complete enquiry should include the application description, operating conditions, and the quantities of interest. The buyer should provide any available measured data, even if preliminary, as this can serve as a reference for validation. If no data exists, the buyer should specify a benchmark case that is acceptable.
The enquiry should also state the required acceptance criteria and the level of validation rigor. For high-consequence applications, a full validation report with mesh independence and turbulence sensitivity is needed. For lower-consequence, a simpler verification may suffice. The buyer should indicate the consequence of error to allow the supplier to propose an appropriate approach.
Additionally, the buyer should provide the geometry, material properties, and boundary conditions. If the geometry is complex, a CAD model should be supplied. The buyer should also state any constraints on the analysis, such as software compatibility or delivery format. This information enables an accurate quotation and reduces the risk of misunderstanding.
The cost of validation increases with the rigor required. A full validation with multiple mesh levels, turbulence model sensitivity, and comparison with measured data is more expensive than a simple verification. The buyer should assess the consequence of error to determine the appropriate level of validation.
For critical components, the cost of validation is justified by the risk of failure. For non-critical or low-volume parts, a reduced validation may be acceptable. The specification can define tiers of validation, each with a defined set of requirements, allowing the supplier to quote accordingly.
The buyer should also consider the cost of measurement data. If no data exists, the cost of testing may be significant. In such cases, a benchmark case from literature may be used, but its relevance should be justified. The trade-off between validation cost and risk should be documented in the specification.
A converged solution shows that the numerical residuals have fallen and the mesh has stabilized. A validated solution additionally shows agreement with an external reference, such as measured integral quantities or a documented benchmark, within stated acceptance criteria. Convergence is necessary but not sufficient for validation.
General engineering practice uses at least three systematically refined mesh levels, reporting the change in the target integral quantity between successive levels. The study is complete when that change falls below a threshold agreed in advance, typically a small percentage of the quantity of interest.
Integral quantities such as pressure drop, mass flow rate, and discharge coefficient correspond directly to measurable test data and are less sensitive to local modeling error. Local profiles are useful for diagnosis, but they are harder to measure with low uncertainty and can mislead when used alone as validation evidence.
A validation report should include the reference case description, mesh independence study with at least three mesh levels, turbulence model sensitivity results, comparison of integral quantities with measured data including uncertainty, and the acceptance decision. It should also state the limits of validity and any assumptions made.
The level of validation is determined by the consequence of error. High-consequence applications, such as those affecting safety or high-volume production, require a full validation with mesh independence and turbulence sensitivity. Lower-consequence applications may use a reduced validation, but the core elements of reference case and comparison with data should still be present.
Validation requires an external reference. If measured data is not available, a documented benchmark case from literature or a standard problem can be used, provided its relevance to the application is justified. The benchmark should share the governing dimensionless groups and physics of the intended application.
This page is an archival specification note for informational purposes only. The former plant is not quoting from this domain. Nothing here is a qualified procedure, an inspection report, or a design calculation; confirm current requirements with the issuing standards body and your material supplier.