0

I have many geometric primitives (100) which cut a cube. This could be modelled using CSG. Now I need a relatively coarse tetra mesh for finite element simulations. For generating the mesh I came across netgen (see the picture). But netgen will give me mesh with 2 million+ elements which can not be handled by the FE program I am using. Are there any reliable programs that could do this job? I already tried to only mesh the surface and coarsen it somehow but without any success.

user3528789
  • 103
  • 7
  • I have never tried csg meshing (only cad) with netgen, but have you tried changing the granularity option in Mesh->Meshing Options-> General? – Thomas Feb 03 '15 at 20:37
  • yes I tried to play around with the granularity (using always coarse) as well as with the max mesh size, mesh-size grading, elements per curvature (this seems to coarsen the most) ... I guess the major problem is the advancing front algorithm in combination with a relatively sharp geometry (the axis ration of the ellipsoids is 1/5). Therefore I was desperately looking for an alternative – user3528789 Feb 04 '15 at 20:17
  • You wrote "see the picture", but there is none. Thus i cannot understand what you say about the geometry. Just two more hints: there are mesh size options in the mesh-size tab as well which might do better than just setting the preset "coarse" (and btw, there is a very coarse option). Another question would be what your problem was with generating the surface and then coarsening? If the coarsening step is your problem i suggest you export the surface, import in MeshLab and try some of the coarsening filters. Then, you can also try TetGen (yes, with "T"), which can handle surface mesh inputs. – Thomas Feb 04 '15 at 23:27
  • Hi thomas, sorry for the miss leading picture comment. I am not allowed to upload a picture so you may follow this link (https://www.dropbox.com/s/sj5ahd115oqb5gs/Ellipsoids.png?dl=0) . I tried all the options in the netgen gui without too much success. My actual working procedure right now is to create a surface mesh with netgen and then use TetGen for volumes. But for coarsening TetGen does not work automatically. You have to specify nodes to be removed or a size function which is not possible. I will try meshlab. Surface generation is not working properly e.g. in salome because of the csg. – user3528789 Feb 05 '15 at 12:50
  • Based on your picture i would say 2 million is not that high for this geometry and i think you need to becareful not to loose too much computation accuracy - the tips of the ellipsoids can easily be "lost" in coarsening. Another question is why your FE program cant handle 2 million elements - is this a software or hardware restriction? – Thomas Feb 05 '15 at 13:00
  • I am doing finite deformation nonlinear constitutive law calculations with quadratic tetrahedral elements. 2million elements are simply too much for this kind of analysis. i am using the sophisticated fe-software abaqus. even on clusters I run in a serious memory problem and the calculation takes ages. I think the tips do not need to be meshed that fine. there should be another way. the major "problem" with netgen is the advancing front algorithm that creates too dense meshes in the neighbourhood of high curvature. I just had a look at meshlab... i think it is more suited for visual meshes. – user3528789 Feb 05 '15 at 13:11
  • Of course meshlab may not handle all aspects like element quality as restrictive as netgen of FE specific tools, nevertheless i think it would be worth a try and e.g. do a quadratic edge collapse decimation on the surface mesh and then recompute tets based on the result with tetgen which can add in a few steiner points in the last step again. Other than that, i am out of options, as i am not an FE expert. Maybe you could use a smaller, seamless (symmetric) rve? – Thomas Feb 05 '15 at 13:22
  • thanks for your considerations thomas. I will play a round with meshlab. the size of an rve is not sharply defined so I wanted to see a convergence to gain an idea on how big it should be for my simulations. – user3528789 Feb 05 '15 at 13:40

0 Answers0