Mesh quality matters most where gradients, fluxes, separation, wall shear or interfaces matter. A poor cell hidden in a stagnant corner is different from the same defect at a leading edge, jet, thin gap or thermal interface.
Problem: the mesh has bad cells, or the solver points to mesh quality
Typical symptoms include divergence after several iterations, pressure or velocity spikes, excessive diffusion, non-physical recirculation, poor conservation, noisy force histories or solver messages associated with invalid or highly distorted cells.
The correct question is not simply “Is the minimum quality acceptable?” It is “Do poor cells coincide with the physics that controls my quantity of interest?”
Quality metrics are warning signals, not universal pass/fail numbers
Non-orthogonality
Large departure between face-normal and cell-centre connection directions can degrade gradient and diffusion calculations, especially in strong-gradient zones.
Skewness
Strongly displaced face or interpolation locations can increase correction requirements and reduce accuracy.
Very small cell volume
Tiny accidental cells can impose severe transient restrictions and create local stiffness without adding useful resolution.
Abrupt size transition
Fast jumps between fine and coarse cells can dissipate wakes, smear interfaces and destabilize transport across refinement boundaries.
Extreme aspect ratio
High aspect ratio is not automatically bad—boundary-layer cells are intentionally stretched—but orientation must align with the resolved gradients.
Poor face validity
Collapsed faces, negative volumes, topology defects or near-degenerate cells are more serious than merely imperfect shape metrics.
Where bad cells usually come from
- Dirty or overlapping CAD, tiny sliver surfaces and unresolved geometric gaps.
- Local surface sizes that are incompatible with curvature or narrow passages.
- Prism layers forced into corners, concave regions or gaps that cannot accommodate the requested stack.
- Too-aggressive growth between local refinement and the core mesh.
- Boolean or imprint topology that creates very short edges and tiny faces.
- Moving, overset or deforming meshes whose quality degrades during motion rather than at the initial state.
Diagnose mesh quality in the same coordinate system as the physics
- Display the worst cells, not only the minimum statistic.
Create threshold parts or cell sets for the poorest few percent and inspect their locations.
- Overlay gradients and engineering monitors.
Check whether poor cells coincide with high velocity, pressure gradient, heat flux, phase interface, shocks, separation or rotating interfaces.
- Check geometric cause before changing solver controls.
If bad cells repeat along one surface, corner or gap, the geometry/meshing recipe is usually the root cause.
- Compare local and global refinement.
Refine the defect-producing region only. Uniform global refinement can multiply cell count without correcting topology or transition quality.
- Run a controlled mesh variant.
Change the suspect feature and compare the quantity of interest, conservation and convergence—not only the quality number.
Use mesh decisions as engineering inputs
Estimate near-wall and prism requirements before forcing the volume mesher to satisfy incompatible scales.
Fix order
- Remove pathological geometry features.
Repair slivers, duplicate faces, tiny gaps and accidental topology first.
- Resolve surfaces at the scale required by curvature and gaps.
A volume mesh cannot recover physics that the surface representation already destroyed.
- Reduce abrupt mesh-size transitions.
Use staged refinement and smoother growth where transport crosses mesh levels.
- Repair prism-layer conflicts separately.
Do not sacrifice the entire near-wall mesh because a few corners cannot support the requested layer stack.
- Only then adjust numerical robustness.
Under-relaxation or more dissipative schemes can help startup, but they should not hide a persistent meshing defect.
How to verify that the repair mattered
- The repaired mesh removes or relocates the worst cells from critical-gradient regions.
- Mass, momentum or energy imbalance improves rather than merely residual smoothness.
- The engineering quantity—pressure drop, force, temperature, flow split, etc.—moves toward mesh independence.
- Local contour spikes disappear without erasing physically expected gradients.
- A further targeted refinement causes a smaller change than the previous mesh step.
Related Abecator resources
Still unsure whether the mesh is controlling the answer?
Submit the non-confidential mesh statistics, screenshots and engineering quantity that is behaving suspiciously.