Mesh Quality Metrics — Six Evaluation Criteria That Determine Numerical Accuracy

I understand how you feel. But in practice, only three are most important: Aspect Ratio (ideal is 1:1, acceptable below 3:1), Skewness (0 is ideal, 0.95 or above is NG), and Orthogonality (below 0.1 is NG). If you master these three, you generally won't run into major problems.

Of course, Jacobian, Warping, and Taper are also important, but they often "become bad in conjunction when the top three are bad." In other words, if you keep the three major metrics in good condition, the others will naturally improve in most cases.

Simply put, it's because "an element that is extremely slender in only one direction cannot properly capture changes in that direction." For example, if you model a car's front pillar and use an element that is very slender vertically (aspect ratio 20:1 or so) with only one element in the thickness direction, you won't be able to properly calculate the bending stress distribution.

Good question! Boundary layer meshes can have aspect ratios of 100 or more, but that is intentional. They need to be thin to resolve the steep velocity gradient in the wall-normal direction. Since the flow direction is almost uniform, high aspect ratio is not a problem. In other words, "aspect ratio in a direction aligned with the physics" is acceptable.

For a triangular element, it's a "flattened" element where the three vertices are almost in a straight line. For a tetrahedron, it's when the four vertices are almost on the same plane. Such elements have area/volume approaching zero, so the Jacobian also degenerates and the calculation fails.

That's a fail. If the Fluent mesh check shows cells with a minimum orthogonality below 0.01, you must correct them. Consider results from cells below 0.01 to have zero reliability. Ideally, the minimum should be above 0.1 and the average above 0.9. First, identify the location of the problematic cells and correct the mesh locally.

Jacobian ratio is the ratio of the minimum to maximum Jacobian value within an element. For an ideal shape, the value is the same at all integration points, so the ratio is 1.0. In practice, 0.3 or above is recommended. If 0.0 or below (negative Jacobian) appears, that mesh cannot be used for calculation. Abaqus or Nastran will stop with an error if they detect a negative Jacobian.

A common issue with quadratic elements. If you shift mid-side nodes significantly away from the edge midpoint, the mapping becomes nonlinear and the Jacobian can change sign inside the element. This happens especially with meshes following curved surfaces. The countermeasures are to restrict mid-side node positions to near the edge midpoint or to refine the mesh in areas of high curvature.

A typical case is modeling vehicle body panels or aircraft outer skins using shell elements. When QUAD elements are placed on curved surfaces, the four nodes may not lie on the same plane in areas of high curvature. If warpage exceeds 15°, the calculation accuracy for bending stiffness plummets. Countermeasures are to mix in TRIA (triangles, always planar) instead of QUAD, or to refine the mesh so the curvature can be approximated by planes.

Good question. The accuracy of the Finite Element Method is determined by how accurately shape functions can interpolate physical quantities. Let's look at the element stiffness matrix calculation.

When an element is distorted, accuracy degrades in three stages.

• Stage 1 (Distortion of shape functions): High aspect ratio makes gradients in the short direction dominant, failing to capture changes in the long direction.
• Stage 2 (Numerical integration error): High skewness makes Gaussian integration point placement non-optimal, reducing integration accuracy.
• Stage 3 (Deterioration of condition number): Poor-quality elements worsen the stiffness matrix condition number, slowing iterative solver convergence (or worst case, no convergence).

It's even more serious in CFD (Finite Volume Method). For cells with low orthogonality, non-orthogonal correction is needed to approximate diffusive fluxes. If the correction is large, it cannot be treated implicitly, making iterations unstable. When "Divergence detected" appears in Fluent, the cause is often cells with minimum orthogonality below 0.05.

Exactly. Polyhedral meshes have many faces, so the impact of non-orthogonality per face is distributed. However, they are not a panacea; boundary layer meshes should still be stacked with prisms (hexahedra) as a basic rule.

Precisely. y+ is the dimensionless height of the first cell off the wall, and can be thought of as a quality indicator for whether it meets the turbulence model's requirements.

Aspect ratio for boundary layer meshes can be intentionally high. However, pay attention to the Growth Ratio. The size ratio between adjacent cells should be 1.2 or less. If it exceeds 1.5, non-orthogonality deteriorates rapidly, turning into an orthogonality problem. In other words, y+ and aspect ratio are not a trade-off; Growth Rate and Orthogonality become the bottleneck.

Let me teach you the 5-step quality check flow used in practice.

1. Check Global Statistics — Check the min/max/average aspect ratio, skewness, orthogonality for the whole model. Look at the distribution with a histogram.
2. Identify Location of Worst Elements — Visualize where the worst-quality elements are. They are often concentrated at shape edges or areas of high curvature.
3. Check Alignment with Structurally Important Regions — Check if poor-quality elements overlap with stress concentration areas or wake regions. Poor quality in distant areas has little impact.
4. Perform Local Correction — Correct only the problematic elements. No need to re-mesh the entire model.
5. Re-check After Correction — Verify that the correction hasn't worsened quality elsewhere.

These six are your weapons for quality improvement. The theory is to try them in order from the top.