Skip to content

Meshing tools

seed pillar 7 · Domain codes and toolchains

Cheatsheet page

Gotchas, not tutorials. Write the entry the day you fight the tool.

The landscape

Tool Type Notes
Gmsh open source, scriptable .geo scripts or Python API; good for research
Pointwise / Fidelity commercial strong structured and hybrid meshing, glyph scripting
snappyHexMesh OpenFOAM castellate/snap/layer from an STL; awkward but free
cfMesh OpenFOAM often easier than snappy for the same job
AMReX / block AMR in-code no external mesh; refinement is a runtime criterion

Gmsh

gmsh model.geo -3 -format msh2 -o model.msh       # 3D mesh, MSH2 for older readers
gmsh -check model.geo                              # geometry sanity

import gmsh
gmsh.initialize()
gmsh.model.add("duct")
gmsh.model.occ.addBox(0, 0, 0, 1, 0.2, 0.2)
gmsh.model.occ.synchronize()
gmsh.model.mesh.field.add("Distance", 1)           # refine near a surface
gmsh.model.mesh.field.setNumbers(1, "FacesList", [1])
gmsh.model.mesh.field.add("Threshold", 2)
gmsh.model.mesh.field.setNumber(2, "InField", 1)
gmsh.model.mesh.field.setNumber(2, "LcMin", 0.002)
gmsh.model.mesh.field.setNumber(2, "LcMax", 0.02)
gmsh.model.mesh.field.setAsBackgroundMesh(2)
gmsh.model.mesh.generate(3)
gmsh.write("duct.msh")
gmsh.finalize()

Mesh size fields are the right mechanism for graded meshes — setting characteristic lengths point by point does not generalise and does not reproduce.

snappyHexMesh

Three stages, each toggled in snappyHexMeshDict: castellatedMesh (cut the background hex mesh to the STL), snap (move points onto the surface), addLayers (insert prism layers). Layer insertion is the stage that fails, and it fails by silently producing fewer layers than requested — always check the log for the achieved layer count per patch, not just for success.

Quality targets

Repeated from meshing and grid quality, because these are the numbers to type into the tool:

Metric Target
Non-orthogonality max \(< 60\)–70°
Skewness max \(< 4\) (OpenFOAM) / \(< 0.85\) (normalised)
Aspect ratio \(< 100\) in the bulk; \(10^3\)–\(10^4\) is normal in boundary layers
Expansion ratio \(< 1.2\) between neighbours
Prism layers enough to cover \(\delta\) — often 25–50, not the default 10
Failure modes

Prism layer count achieved, not requested. snappyHexMesh reports how many layers it actually inserted; in tight corners it silently inserts fewer or none. The mesh looks fine in a slice and the boundary layer is unresolved exactly where the flow separates.

Prism stack ending inside the boundary layer. Covered in meshing and grid quality — a size jump where gradients are steepest.

STL not watertight. Gaps or flipped normals cause the mesher to leak into the interior, producing a mesh of the wrong region. Check with surfaceCheck (OpenFOAM) or Gmsh's geometry check before meshing.

Characteristic lengths set per-point rather than by field. Works for one geometry, does not survive a parameter change, and cannot be reproduced. Use size fields.

MSH format version mismatch. Gmsh 4 writes MSH4 by default; many readers expect MSH2. The error is usually a parse failure far from the real cause. -format msh2.

Mesh independence never demonstrated. A single mesh is a single data point. See V&V — and note that refining a wall-function mesh changes the turbulence modelling, not just the discretization.

Units. CAD in millimetres, solver in metres. A factor of 1000 in length changes Reynolds number by the same factor, and the run proceeds happily.

Connections

Sources

  • Gmsh reference manual (the size-field section especially); OpenFOAM snappyHexMeshDict documentation.
  • Books — Roache for grid convergence.