BPS5234A Chap.8 Interpreting and Validating CFD Results
Interpreting and Validating CFD Results
Results are evidence only when they are read correctly
A simulation supplies values everywhere and at every instant, where an experiment supplies a few measured points. That is a strength and a trap: read carelessly, a handsome plot misleads the designer it was meant to help.
The course treats post-processing as an engineering task with a clear purpose, which is design improvement rather than presentation.
Choosing what to plot and what to compute
Contours and vectors are cut on planes that answer questions, such as pedestrian level outdoors, the occupied zone indoors, or vertical sections through openings. Design options are compared on identical colour scales.
Streamlines and path lines reveal short-circuiting and stagnant zones; iso-surfaces show volumes above a comfort or air-quality threshold.
Every picture is then backed by numbers such as flow through openings, heat transfer at walls, average temperatures or concentrations, and vertical profiles.
Verification and validation are different questions
Verification asks whether the equations are solved correctly: at least three progressively refined meshes, residuals cut a thousandfold, or a millionfold for energy, balance errors typically below one per cent, and second-order final results.
Validation asks whether the equations and models represent reality, by comparison with full-scale measurements, reduced-scale experiments or benchmarks. Calibration asks whether integral quantities are reproduced for a class of problems.
What counts as success
Perfect agreement is rare in buildings because wind and heat loads vary.
The course asks for correct dominant flow features, acceptable agreement in meaningful quantities, and a sensitivity analysis, documented through a seven-point quality checklist.
What this chapter covers
- 01
Choosing planes, plot types and colour scales
- 02
Surface integrals, profiles and comparative metrics
- 03
Turning results into design changes
- 04
Verification through grid independence and convergence
- 05
Validation, calibration and benchmark data
- 06
The quality-control checklist for a report
Worked example · free
Comparing simulated and measured pressure coefficients
- 2Compute the difference at each point as a share of the measured value.
- 2Identify where agreement fails and the physical reason.
- 2Propose the checks that would address it.
Key terms
- Verification
- Checking that the governing equations are solved correctly, through discretisation, convergence and implementation checks.
- Validation
- Comparing predictions with measurements or benchmarks to judge whether the models represent physical reality.
- Calibration
- Testing whether a model reproduces integral quantities such as airflow rate or mean temperature for a class of problems.
- Iso-Surface
- A surface of constant value, such as a chosen air speed, used to show the volume where a threshold is exceeded.
- Path Line
- The trajectory traced by a fluid particle released into the flow, useful for finding short-circuits and stagnant zones.
Interpreting and Validating CFD Results FAQ
Why compare design options on a fixed colour scale?
Automatic scales stretch each legend to its own minimum and maximum, so the same colour can mean different values in two plots. A fixed scale makes visual differences real differences and stops a worse option looking similar to a better one.
How many meshes does a grid-independence study need?
The course asks for at least three progressively refined meshes, with refinement concentrated where gradients are strong, and a comparison of quantities that matter, such as velocity profiles, temperatures, pressure drop or ventilation rate.
Is a converged residual plot enough to prove a result?
No. Residuals must fall at least a thousandfold, but whole-domain balances must also close within about one per cent and the monitored physical quantities must stop changing. Even then the result is only verified, not validated.
What if the simulation cannot match the measurements closely?
In buildings that is common because of uncertain wind directions and heat loads. The course asks for the dominant flow features to be reproduced, reasonable agreement in statistically robust quantities, and a sensitivity analysis that shows the conclusions survive reasonable changes.
Why are comparative studies more robust than absolute predictions?
Errors from the model, mesh and boundary conditions tend to affect two similar design options in similar ways, so their difference is more reliable than either absolute value. That is why many design questions are framed as comparisons.
Assessment move
For each run you report, write a three-line evidence block: the mesh comparison, the convergence record, and the benchmark comparison. If any line is empty, the run is not yet ready to support a conclusion, however good its plots look.
