The quality of the computational mesh plays a crucial role in the accuracy of simulation results, especially with lower-order discretization approaches. Even a single distorted element can compromise the entire calculation. While structured grids offer advantages, they are limited in representing complex geometries. More flexible options include extruded unstructured meshes from hexahedrons or unstructured tetrahedral meshes, although the latter often pose issues in finite-volume methods.
Solution approaches include hybrid meshes, where structured elements are used near the surface and combined with unstructured meshes in the interior, as well as polyhedral meshes and hierarchical grids based on octrees. The meshing strategy can be boundary-fitted or embedded, depending heavily on the implemented method.
Quality criteria vary by method: for finite-volume methods, the angle between cell face and the line connecting cell centroids is critical, while for finite element methods, the determinant of the Jacobian matrix is essential. RANS simulations allow for coarser meshes with larger progression factors (1.10–1.20), whereas LES simulations require factors below 1.05.
To verify mesh independence, grid studies with at least one refined mesh are necessary. Identical model settings and boundary conditions should be used, and the mesh quality must meet the standards of the CFD software, solver, and turbulence models. Richardson extrapolation can aid in assessing grid independence but requires at least three different grids and assumes a monotonic progression of target variables.