my previous example of plate with hole has eight segments of curve, maybe some feature of splitting a circle needed similar to periodic faces currently available.
Maybe, but it’s hard to say how many segments are needed. I guess it depends on the overall shape of the surface (since we’ve seen cases with holes that worked out of the box). It would be good to discuss this with the QuadWild team on their GitHub. I can ask there unless Matej figures it out first.
1 Like
I have found a problem in the mesher; it has problems with planar geometries. An additional parameter can be provided to solve it. I have added it. It still fails for the Synt’s model, with the default mesh bit, which works with a finer mesh but neglects split edges.
1 Like
I upload images of the job I am working on:
1- QUASI-STRUCTURED QUAD:
2- QUADWILD
Shell body is very similar for both algorithms, but the legs is much better QuadWild.
1 Like
can the part be isolated and saving and meshing individually, it seems the legs’ part shape is simple enough like box or L-sections but i did not know why Gmsh (quasi-structured quads) generate few distorted elements near corner.
Quasi-structured quad sometimes create such irregular transitions even for simple shapes, but it’s usually the case with parts containing some holes, circular boundaries or otherwise differing from rectangular shapes. Here it may just need some partitions to avoid that.
right sometime the flow is undesired and generate irregularity, edge or line partition needed for helping. However, this case of legs part looks simply but generate distorted element like shape near triangle and it can be rejected by the solver probably. I only guessed related to improper CAD preparations i.e edge being split with small value.
I am modifying the Qadwild to enable direct positioning of edge nodes, non-uniform mesh density, non-manifold geometry, better meshing around internal features (holes), and FEM optimized element shapes.
1 Like
it seems QuadWild did not have internal algorithm to control in reducing skew distorted element to be accepted by FEM solver, but next development version showed it does improvement.
Well, after all, it’s not necessarily meant for FEM. In fact, I would say that it seems to focus mostly on other applications, such as Blender-like models.
2 Likes
Also it depends on size. In case is very small, I have obtained negative jacobians.
Only with QuadWild and not with Gmsh Quasi-structured quad ? That must be another case where FEM-specific cheks are missing.
1 Like
Yes, the default Quadwild has these problems. It was improved as stated earlier, but will be introduced in the new PrePoMax.
Actually, I meant the new, improved QuadWild. Maybe I will call it QuadwildFEM.
Maybe you could publish a GitHub/GitLab repo with this modified fork of QuadWild or otherwise post it separately. Some users may want to try it as a standalone tool.
It seems to be a good idea to treat it as a FEM-oriented version of QuadWild since the default version is not really focused on this area.
1 Like
I will. But it is not finished yet. Still has some problems and needs more development to be useful.
The problem is that for sweep meshing, I need a quad algorithm that can adjust the mesh to the boundary/edge node positions. The Quasi-structured quad is problematic in this regard, so I am looking for a better quad mesher. Could not find anything really good. Quadwild is also not too good in this regard, but better than a quasi-structured quad.
3 Likes
That would be great, thanks. I think that adopting QuadWild was a really good idea. It can be useful also for shell meshing itself since Quasi-structures quad sometimes gives strange results.
Let me know if you need more testing for specific cases.
2 Likes
Hi one additional question. Is it possible to get the input Gmesh code used by Prepomax?
Thank you.