I'm using the different unstructured meshing techniques in ICEM CFD and finding that Hexa-Dominant gives me unusuable meshes. The tetra and hexa-core give me meshes which have nice stress distributions.

I'm using the mesh created by ICEM for a stress analysis problem using nonlinear materials with large strain applied.

The simulation is a soft tissue (non linear material) cylinder being compressed by two rigid platforms. The cylinder is experiencing uniaxial compression of 40% strain.

Has anyone else encountered problems with the HexaDominant mesh? I've tried smoothing the mesh, but to no avail.

Evan

Create the mesh as a block stuctured mesh and then convert it to a unstructured mesh. You'll have 100% hexas. Creating a cylindrical block structured mesh is trivial.

Thanks for the quick response.

I guess I should've explained a bit more about my project. I'm actually trying to get the Hexa-Dominant automesher to work since I'll eventually be meshing complex geometry (bones and soft tissue from MRI scans) and am trying to determine which auto meshing techinique is optimal.

It seems from the large strain simulations that I've run so far that Hexa-Dominant seems pretty useless for these types of problems (absolutely horrible stress distributions). Anyone else encounter this?

Evan

I have had little success with them too, for fluid flow meshes.

Another option is to do a tet only mesh *with no smoothing* and then use the tet->hex converter http://www-harwell.ansys.com/demoroo...s.php?id=39,10 . Again ... it hasnt worked well for me.

If you are going to be doing trabecular bone I would have thought tet meshing is the only viable option.

Why are you using ICEM to ceate FE meshes. Doesnt Abaqus have a decent in house mesher?

Are you using Abaqus? As I recall it has an adapttive mesher which could adapt the mesh to the lasgest stress gradients?

It can like in MSC Marc where mesh refinement works in relatively small area of cases. It can be true for Abaqus because both codes shares common roots.

Just did the following experiment:

Created 50 unit diameter sphere. Created two mesh densities that covered each hemisphere: One with 0.25 unit mesh size, one with 0.5 unit mesh size.

Global mesh size 1 unit.

Hexa-core generates a very high quality mesh with three symmetrical cell layers: -Outer tet mesh bordering on sphere surface. -Intermediate pyramid layer. -Inner symmetrical hex core.

Only a single hex size is present and is roughly the average of the two mesh density values.

Hexa-dominant with exactly the same setup generates complete rubbish. About 3-4 hexes (out of a million+ cell mesh), the rest are a horrid mismash of prisms, tets and pyramids. The core is made up of a few gigantic pyramids which completely violate the mesh density specs.

This would explain why I have only be able to use hexa-core succesfully in the past

Perhaps one of the Ansys employees could comment on what hexa-dominant meshing is usefull for. (Regular cuboidal geometries?)

The applications were purchased awhile ago: Ansys ICEM for the mesher and LS-DYNA for the solver. We don't have licenses for other FE software at the moment, so I have to make do with these for now.

I've also tried using the tet to hex converter, but have been mostly unsuccessful. Unless the mesh is very orderly, the conversion will fail to take place.

Thanks for the input.