[Gmsh] Multilevel mesh partition element interface quality

Wed Feb 6 04:01:51 CET 2013

Hi Andre - Thanks for the detailed report. One solution will definitely be to post-process (optimize) the mesh, as currently we just leave it as is. Optimization will be available pretty soon.

On 04 Feb 2013, at 21:30, Andre Nicolle <andre at cfdandip.co.uk> wrote:

> Hi Christophe
> 
> Thank you for revisiting this issue. It is helpful to know that it is possibly not just an implementation problem. I have retried the mesh of the human STL. Summarising, the problem does not seem to have gone away with the latest version of gmsh. I tried both the self built svn version and the nightly build from the website (64bit) on the 2/2/2013 and found no difference between them. As a way of determining if there was any particular problem I also did a quick parameter study using different meshing algorithms "Mesh.Algorithm" and partitioning methods "Mesh.RemeshParametrization". in all cases the resolution was set to "Mesh.CharacteristicLengthFactor=0.05" which produces a 2d surface mesh of around 80,000 triangles.
> 
> As before the partitioning is around the waist line of the human as shown in the file attached error_partition.jpg. The meshing algorithm we had the most success with was the "Frontal" where the elements were evenly spread as can be seen in the attached files face_q.jpg and face_mesh.jpg.
> 
> I have found that Delaunay or MeshAdapt options have difficulty meshing around the human limb extremities such as the feet. The example file attached frontal_foot.jpg shows a good quality mesh which can be produced at any resolution. When using Delaunay or MeshAdapt I found that the elements get substantially coarser around relatively sharp edges as can be seen in file meshadapt_foot.jpg. This problem gets worse as the resolution is increased.
> 
> I have also tried each of the seven RemeshParametrization as I was interested to see the effect of this parameter.  It seems that options 7,1 both do not have bad elements around the partition regardless of the mesh resolution. However the problem with these options the arms and feet get cut off as can be seen in file mis_feet.jpg. This occurs regardless of the meshing algorithm.
> 
> The only RemeshParametrization option where this does not happen is 4,5 however both algorithms suffer from poor elements on the partition region.
> 
> I hope this information helps in some degree in solving the problem we are encountering. If I can help any further please let me know.
> 
> Thanks
> 
> Andre
> 
> On 01/02/2013 01:14, Christophe Geuzaine wrote:
>> Hi Andre - Could you try again with a recent build and tell us if things improved?
>> 
>> Thanks,
>> 
>> Christophe
>> 
>> On 31 Oct 2012, at 06:18, Andre Nicolle <andre at cfdandip.co.uk> wrote:
>> 
>>> Dear gmsh user group
>>> 
>>> I have been struggling with meshing a stl file of a human mesh. The stl surface is in good shape with a closed surface which when loaded into gmsh and meshed using the procedure outlined in https://geuz.org/trac/gmsh/wiki/STLRemeshing forms a three-dimensional mesh. Unfortunately the tetrahedron element quality is very poor. On further investigation we found that this was caused by poor quality elements on the 2D surface after the surface was remeshed. We tried using  Mesh.Algorithm: 1 = MeshAdapt, 2=Automatic, 5=Delaunay, 6=Frontal and found that in general Frontal gave us the best elements which were all very good quality except on the interface between the multilevel mesh partitions.  Information on the partition stage of 2-D meshing is given below.
>>> 
>>> Attached is a picture showing the line-up of poor quality elements around the waist of the human. Nearly all the other 2-D triangular elements have a mesh quality indicator gamma > 0.5, however elements on theinterface have gamma approximately 0.03 which we believe is where the 3-D mesh problem is starting from.
>>> 
>>> Could somebody give us some advice on how best to solve this problem as I imagine it is a mistake on our implementation. If further information is required please let me know what would be helpful to diagnose the problem.
>>> 
>>> Any help would be gratefully received
>>> 
>>> Andre
>>> 
>>> gmsh version 2.6.1
>>> 
>>> Info    : Meshing surface 200 (Compound surface, MeshAdapt)
>>> Warning : Wrong topology: Genus=0, Nb boundaries=0, AR=1
>>> Info    : -----------------------------------------------------------
>>> Info    : --- Split surface 200 in 2 parts with Multilevel Mesh partitioner
>>> Info    : Building graph...
>>> Info    : Partitioning graph...
>>> Info    : Launching Chaco graph partitioner
>>> Info    : Done partitioning graph
>>> Info    : *** Mesh partition: level (1-0) is ZERO-GENUS (AR=5, NB=1)
>>> Info    : *** Mesh partition: level (1-1) is ZERO-GENUS (AR=4, NB=1)
>>> Info    : Multiscale Partition SUCCESSFULLY PERFORMED : 2 parts (3.26 s)
>>> Info    : *** Starting parametrize compounds:
>>> Info    : Parametrize Compound Line (2) = 1 discrete edge
>>> Info    : Parametrize Compound Surface (203) = 201 discrete face
>>> Info    : Parametrizing surface 203 with 'convex map'
>>> 
>>> 
>>> Warning : Mesh generation error summary
>>> Warning :     1 warning
>>> Warning :     0 errors
>>> Warning : Check the full log for details
>>> 
>>> 
>>> <test.jpg>_______________________________________________
>>> gmsh mailing list
>>> gmsh at geuz.org
>>> http://www.geuz.org/mailman/listinfo/gmsh
> 
> <error_element.jpg><error_partition.jpg><face_mesh.jpg><face_q.jpg><frontal_foot.jpg><meshadapt_foot.jpg><mis_feet.jpg>

-- 
Prof. Christophe Geuzaine
University of Liege, Electrical Engineering and Computer Science 
http://www.montefiore.ulg.ac.be/~geuzaine