Model Verification: Proving the Model Is Solved Correctly
A structured verification framework — reaction checks, hand calculation comparison, mesh convergence, limiting cases and sensitivity studies. Verification asks: did we solve the equations right?
Principle
Verification asks whether the model has been solved correctly. It is the analyst's responsibility and does not require test data. Verification is distinct from validation, which asks whether the model represents reality.
Verification: are we solving the equations right? Validation: are we solving the right equations? Both are required. They are not interchangeable.
Why it matters
Without verification, there is no basis for trusting any result. A model that has not been verified may contain errors in boundary conditions, loads, material properties or mesh that produce plausible but wrong results. Verification is the first line of defence against these errors.
Minimum verification checks
- Unit consistency confirmed throughout model — Check length, force, mass, time, temperature
- Applied loads verified against loads report — Direction, magnitude, distribution
- Reaction forces balance applied loads to < 1% — Sum in each direction
- Mass/weight matches design document — Including non-structural mass
- Deformed shape physically sensible — No through-mesh, no spurious rotation
- Nominal stress matches hand calculation — At a simple section away from features
- Mesh convergence demonstrated for the output of interest — Not just displacement
- No singularities reported as design stress — Check mesh-refinement behaviour at peak stress
- Contact status verified (if applicable) — Contact where expected, opening where expected
- Element quality metrics acceptable — Aspect ratio, skewness, Jacobian
Verification methods
- Reaction force check: sum of reactions must equal sum of applied loads in each direction — this is the most fundamental equilibrium check
- Free-body check: cut the model at a section and verify internal forces balance the external loads
- Hand calculation comparison: compare FE stress or displacement with an analytical solution at a simple location
- Mesh convergence: refine the mesh and verify the output of interest approaches an asymptote
- Limiting case check: simplify the model to a case with a known solution and verify the FE result matches
- Sensitivity study: vary a key parameter and check the result changes in the expected direction and magnitude
Warning signs
- Reaction forces do not balance applied loads — indicates a modelling error
- Deformed shape is non-physical — over-constraint, under-constraint or wrong load direction
- No hand calculation is available for comparison — the FE result is unverifiable
- No mesh convergence study — the reported stress may be mesh-dependent
- The analyst reports "the model ran without errors" as the only verification — solver convergence is not verification