Meshing method for Opti-Struct / voxel vs. tetra

Altair Forum User

Good afternoon, 

Right now I am trying to figure out what is the proper element type to use for running a topology optimization in a volume.

Does someone know the benefits of using a voxel mesh instead of a tetra mesh? What is the best way to decide which element to use.

 

Let's discuss.

regrds

 

Jannes

Altair Forum User

From the point of view of FE's quality, hexa elements are better than tetra. But you can't make hexa mesh with any geometry. Sometime tetrahedra are only way to work.

Altair Forum User

Dear Nguyen-Dai,

thank your for your thoughts!

 

I found the following in an artcile by Watanabe in 2012, pointing in the same direction of yours.

(Abstract)

'...In optimization for development of electromagnetic devices, a large number of FE analyses should be performed to evaluate
each different shape of device. In such cases, the fast mesh generation as well as fast convergence of linear solvers are
required. The voxel mesh is an approach to solve this difficulty. The shape of magnetic objects can be changed easily
by remapping the permeability of each voxel without changing the structure of elements and nodes. Moreover, the voxel ele-
ments made of orthogonal grid yield the good convergence of the system equations. However, the voxel mesh approach has
two serious weak points. The first one is that a large number of voxel elements are required to fill the air region. The second
one is a poor ability of expression for curved surfaces: The terraced surfaces deteriorate accuracy of FE analyses. The former
problem can be settled by introducing some numerical tech niques such as non-conforming mesh or the progress of computer resources.

In the latter problem, for example, spheres with small different radius are represented by same shape ex- pressed by voxel mesh.

Moreover, discontinuous changes of shape are happened by the small variation of the radius. This causes a difficulty to evaluate the fitness value and sensitivity
of fitness with respect to design parameters in the parameter optimizations.'

 

Since curviture is not a big problem in my case- curvature radius is very big-  I tried different scenarios involving different sizes of voxel-mesh and tetra mesh.

Using OptiStruct I found a slightly better 0-1 performance using Voxel-Mesh (based on same options), while using less CPU-Time. Mainly because for a 10mm Vox-Based Mesh,

has got less elements, than a 10mm Tet. based Mesh, while occupying same volume.

 

 

rgrds,

 

Jannes

Merula_20758

I also tried Voxel vs. Tetra and came to the same conclusions. Did you ever encounter Voxel Mesh exeeding the limits of the design space geometry ? Can we prevent this from happening ?

 

Thanks and best regards,

Merula