Mesh generation using Parallel Sculpt


I am looking to generate all hex mesh for a geometry defined as follows:

a cube of side 1000 units with randomly distributed spherical bodies (155 in number) of varying radii varying between 20 to 35 units. Additionally there is an interface of thickness 5 units around each of these spherical entities. I am using Sculpt tool for this purpose.

Aim I want to achieve is having coarse mesh throughout but refined mesh near the interface region (particularly for this problem I am aiming to have total number of elements not to be more than 1 lakh). Now the problem is I have tried many iterations and combinations of cell size with refinement type/ levels/ threshold in order to limit the number of cells but everytime I am getting some segmentation fault error ! I am not quite sure why such an issue is coming ! Is there something I am doing wrong ?

For above case, I was using 250 as cell size, refinement type “volfrac_difference”, level 8, threshold 0.3 . According to log file, error generally comes during insertion of transition levels after refinement has been done.

Second issue I am concerned about is, for further investigation I need to increase the size of my geometry (by factor of 5 , 10, 50, 100) whereas keeping the interface thickness to be constant i.e 5 units. Here as well I want to keep total number of mesh elements to as minimum as possible and only want some refinement near the interface. But, I found and as I quote from a research paper ( :
“Mesh-first methods are limited to capturing geometric features with the available resolution of the selected base mesh”.
Will it be a problem to generate meshes for my geometry of such large dimension? I cannot found much resource where they have talked more about this resolution , so I would be grateful to have more information in this regards as well.

Looking forward to get some helpful response.



I don’t know what other applications are installed on your system, but sometimes other applications install their own versions of MPI libraries in the system paths. We work very hard to keep Cubit self-contained and not modify your path or the system libraries. You could try locally modifying your PATH environment variable by removing other MPI applications. You could also try temporarily adding the Cubit path at the start of your PATH environment and see if Cubit will pick up the correct MPI libraries from there.

Generally, you will want to keep the mesh size smaller than the features you want to capture. As the paper you cited mentions, the Sculpt algorithm will not detect features that are smaller than the global mesh size.

It looks like you can get a reasonable mesh using a mesh size that is double the same size as your smallest feature and a two-level refinement. In general, the adaptive refinement option will give a coarser mesh than you would get using a smaller global mesh size.

If your smallest feature is a constant 5 units you will need to set that size regardless of the scale of the other objects. This will necessarily create more mesh elements as you scale up. The example in the image below, with a single sphere radius 20 and of radius 25 in a 1000 unit cube, created 4 929 192 elements. This example used a global size 10 and a 2 level adaptive refinement.