Quality evaluation extremely slow for triangular meshes of NURBS surfaces

Coreform Cubit Meshing
#1

Coreform Cubit Version: insert version you are using (2024.3/2024.8)
Platform: insert platform, incl. version (Fedora 39/MacOS 14.7)

Extremely slow quality evaluation for triangular meshes of NURBS surfaces
When trying to evaluate the quality or perform smoothing for triangular meshes on NURBS surfaces, quality evaluation for a moderate sized mesh (e.g., ~20k elements) takes an extremely long time, typically an hour or more. This also applies when attempting to smooth a mesh. Note that this pertains to more complex quality metrics (I have tried Condition Number for both Quality and Smoothing, and Smart Laplacian for Smoothing). If I export the mesh and evaluate the quality in Paraview (using Condition Number, for example), the computation is instantaneous. Do you have any ideas on what I can do to improve performance? I can provide a journal file that demonstrates the issue, if desired.

#2

Hi @charlesw,
could you please share the geometry so that we can take a look at this?

#3

Hi @Norbert_Hofbauer,
I am attaching a journal file that creates a net surface and then creates a triangular mesh on it.

Thanks,
Charles
test1.jou (66.4 KB)

#4

Hello @charlesw,
thanks for the journal. We can reproduce this but i don’t have a solution for you yet. The dev’s are informed and will track this issue.

#5

Thanks very much, Norbert. I’ll be interested to see what the problem is.

#6

Meshing speed can be improved by using the default trimesh algorithm instead of the tridelaunay algorithm. The tridelaunay algorithm is an older algorithm that is used for cases where the newer trimesh algorithm fails. You want to make sure that you turn the automatic geometry sizing off option off to give results that are similar to the tridelaunay algorithm.

set trimesher geometry sizing off
surf all size 5000
surf all scheme trimesh
timer start
mesh surf all
timer stop

We will release a new version of Coreform Cubit this quarter that will have improved speed. You can download and test the latest development release from the Downloads page.

Karl

#7

Hi,
I’m just curious whether there has been any more progress on this. I appreciate the info from Karl about using the trimesh algorithm; however, my problem is with quality evaluation and smoothing for surface meshes.
Thanks,
Charles

#8

I haven’t had time to evaluate this further. I am curious to why quality calculations would be slow unless you are generating millions of elements on the surface.

I can understand why smoothing would be slow. We have to calculate a move to surface for every node that moves. That can be computationally expensive for large meshes on complex surfaces.

Karl

#9

None of the meshes are that large – typically a few thousand triangles on a surface (or less). I do believe it is slower when the surface is a NURBS surface, though. I can’t remember now the details of the journal file I provided last year, but I believe that demonstrates the issue.

Thanks,
Charles