The Workflow StackThermal systems / Issue 01
Five checks before trusting a thermal model
A converged model can still support the wrong engineering decision. Before quoting a cooling load, component temperature, or COP, make a short record of what the model can defend.
1. Name the decision
Write the question in a form that can fail: “Will the hottest cell stay below the allowable temperature during the design drive cycle?” Record the operating envelope, output of interest, and acceptance criterion before looking at the result. A model good enough to rank concepts may not be good enough to sign off a limit.
2. Close the balances
Check mass and energy across the modeled boundary, using one sign convention. For a steady heat exchanger with negligible ambient loss, hot-side heat loss should agree with cold-side heat gain within measurement and model uncertainty. A small solver residual does not by itself close the physical energy balance.
For example, water cooling from 60 to 45 °C at 0.20 kg/s gives 12.54 kW using a constant heat capacity of 4,180 J/(kg·K). Cold water warming from 20 to 32 °C at 0.25 kg/s gives the same value. The agreement follows from these illustrative inputs; real measurements need an uncertainty check.
3. Test numerical influence
Refine spatial resolution and time step separately when the case is transient. Compare the output used in the decision, not only contours or residual plots. If maximum temperature moves by 3 °C across meshes and the design margin is 2 °C, the result is not ready for a pass/fail claim. Record conservation, near-wall resolution, local gradients, and any solver settings that changed.
4. Compare independent evidence
Use a hand calculation, held-out test point, equipment curve, or a second method. Match sensor location, averaging period, boundary conditions, and operating state before comparing values. Calibration data and validation data should be distinguishable. When predictions disagree with measurements, investigate inputs and model structure before adjusting parameters.
5. State uncertainty and limits
Vary important uncertain inputs over defensible ranges and report how they move the decision output. Mark the conditions outside the validated envelope, including extrapolated compressor speeds, ambient temperatures, or flow regimes. Close the review with the decision the evidence supports and the next measurement most likely to change it.
A record you can reuse
Keep the purpose, model version, input sources, numerical study, held-out comparison, uncertainty range, and unresolved questions on one review sheet. The validation guide and downloadable report template provide a starting point.
For deeper numerical guidance, see NASA’s CFD verification and validation tutorial and its grid convergence discussion.