I am working on a full aircraft CFD simulation and transitioning from ICEM CFD to SALOME for meshing. I have multiple design iterations, so automation via scripting is required.
My workflow is:
Create 2D surface mesh → Smooth surface mesh → Generate 3D volume mesh → Smooth again → Generate viscous layers
This works fine manually, but when I script the same steps, the process stalls during smoothing. I also end up meshing multiple times, which seems inefficient.
Is it possible to smooth only the surface mesh or only the volume mesh, instead of the entire mesh object?
Is there a recommended workaround for automating smoothing and viscous layer generation (e.g., sub-meshes, groups, or separate mesh objects)?
Any guidance on Laplace vs Centroidal smoothing, iteration counts, or aspect ratio limits for aircraft meshes would be appreciated.
From my knowledge in Salome you can not smooth 3d. Only 2d, only the surface mesh that will be used for the creation of the 3d.
You can identify the external faces of your domain (wing and boundaries of your domain) and then with some logic keep only the ones of your wing or your external boundaries and when defining these boundary layer you can say apply them to thes faces or apply it to everything except these fsces. This will help you easy selection of them. At least in geom, it is feasible.
Laplace and centroidal for tri mesh will give you good results. Centroidal a little bit more I would say from experience but as the doc says centroidal can generate wrong faces on concave angles such as the tailing edge, I would recommend the laplacian maybe. But I had in the past successfully made wrong elements with it lol. Smoothing in Salome is quite poor and specially it does not check for any constraints of quality therefore you will not be able to be sure and specially plug it into a automatic script.
I would remove the smoothing step if I where you. Unstructured mesh will be good enough also you can use gmsh that has an internal smoothing inside of it, I would go with that.
I am currently using an unstructured mesh. Euler runs are fine, but when I attempt RANS with prism layers and NETGEN 3D, many distorted pyramids are formed, causing SU2 to diverge.
I tried Gmsh for surface meshing, but the quality was poor. Using NETGEN 1D–2D for the surface and Gmsh for 3D works reasonably well, but I am unable to generate viscous layers in that setup. Please let me know if you recommend a different approach.
Previously, for isolated wing studies, I used Dakota for parameters, generated geometry in FreeCAD, meshed in SALOME, and ran cases in SU2 successfully. Now, with a full aircraft configuration and custom control over leading/trailing edges, meshing often stalls.
I also saw an earlier post from @virengos mentioning effective mesh quality improvement and smoothing, but I am unsure whether similar issues were encountered.