There are two questions and they do not substitute for one another:
- Verification: Am I solving the equations right?
- Validation: Am I solving the right equations?
The first is purely mathematical and needs no measurement. The second is physical and cannot be done without one. If a report says "the CFD result agrees with measurement to within 3%" but no grid independence study is shown, then nobody knows what that 3% means.
Why the order matters
In an unverified solution two errors can mask each other. If mesh error pushes the result up and model error pulls it down, the total lands very close to the measurement and the study looks successful. The moment you refine the mesh the agreement falls apart — because it was a coincidence to begin with.
The order is therefore fixed: verification first, then validation. Comparing against measurement without having demonstrated grid independence makes it impossible to separate the physics in the result from the numerical error.
What verification covers
- Grid independence. At least three meshes with a systematic refinement ratio. See the grid independence note.
- Time-step independence. As necessary as the mesh in transient solutions.
- Convergence. Residuals, monitored quantities and the conservation balance.
- Code verification. Solving a problem with a known analytical solution and checking that the expected order of convergence is recovered. Rarely done on the user side, but it should be when a custom model has been written.
What validation covers
Validation means comparing the result against measurement — but the comparison itself carries uncertainty:
- Measurement uncertainty. Sensor accuracy, calibration, repeatability. If the measurement is uncertain to ±8%, a 5% difference between computation and measurement is not a disagreement.
- Matching conditions. Is the boundary condition in the model the same as the real condition on the test rig? Inlet profile, surface roughness, ambient temperature, sealing.
- Choice of compared quantity. A temperature at a single point can agree by chance. Agreement along a profile or a curve is far stronger evidence.
When there is no validation data
Most industrial projects have no measurement to compare against. That does not license behaving as though validation had been done. What can be done:
- Compare a simplified sub-problem against an analytical solution or an established correlation
- Run the same method once against published experimental data for a similar rig in the literature
- Use the result comparatively rather than as an absolute — most of the model error acts in the same direction in both designs and cancels in the difference. See the note on significant digits.
Whichever was done is what goes in the report. "Validation was not possible; the result should be used comparatively" is far more valuable than implying a validation that never happened.
References
- ASME (2009, R2021). V&V 20 — Standard for Verification and Validation in Computational Fluid Dynamics and Heat Transfer. American Society of Mechanical Engineers, New York.
- Roache, P. J. (1994). Perspective: A method for uniform reporting of grid refinement studies. Journal of Fluids Engineering, 116(3), 405–413. DOI: 10.1115/1.2910291
- Celik, I. B., Ghia, U., Roache, P. J., Freitas, C. J., Coleman, H. & Raad, P. E. (2008). Procedure for estimation and reporting of uncertainty due to discretization in CFD applications. Journal of Fluids Engineering, 130(7), 078001. DOI: 10.1115/1.2960953