CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   ANSYS Meshing & Geometry (https://www.cfd-online.com/Forums/ansys-meshing/)
-   -   [ICEM] comparison hexa and tetra (https://www.cfd-online.com/Forums/ansys-meshing/113593-comparison-hexa-tetra.html)

diamondx February 22, 2013 13:16

comparison hexa and tetra
 
hello my friends,

the reason why i'm starting this thread, i would like have your opinion on whether i'm right or not:
below is a chamber i meshed in hexa , and later in tetra:

the tetra mesh gave me around 600k node, including prism.

https://dl.dropbox.com/u/35161486/compairson_1.PNG

while in in hexa, i ended up with too too much node. over 500k node for the half of the geometry

https://dl.dropbox.com/u/35161486/compairson_2.PNG

I know the mesh density is not the same, but in hexa, lowering the elements number will result in bad quality.... that was the best i could do.

meshing in hexa does not implies having less nodes after all, for geometries like this !!!

what do you think ??

Far February 22, 2013 13:43

you made tetra mesh with octree?

diamondx February 22, 2013 13:43

yes... far

Far February 22, 2013 13:59

you can observe one thing very clearly that Hexa is resolving the important flow area very efficiently. Compare simulation results and then tell us the difference... Generally CFD people conclude that one can get similar results when tetra mesh is too fine.

diamondx February 22, 2013 14:03

i can't compare the result. the tetra mesh you saw is combined with the back of the chamber where there is an injector, and the outer casing. for the hexa mesh , nothing else is included and yet the number of number is just too big...

BrolY February 25, 2013 04:49

This debate is almost as old as CFD !

As you described, reffinement is a pain with hexa, because it can't be local. It has to be propagated all the way around. At the same time, quality is often better, especially compared to the prism. I don't speak about aspect ratio, which I consider as a reffinement indicator, instead of a quality indicator.

Some solvers are designed for Hexa, others for Tetra. That's important to keep in mind. But as we are doing RANS calculation, results are already an approximation of the "reality" ! At the end, it's really up to you ;)

Far February 25, 2013 05:05

Quote:

Originally Posted by BrolY (Post 409877)
Some solvers are designed for Hexa, others for Tetra. That's important to keep in mind.

which softwares are designed for Hexa? You meant multi-block hexa or unstructured hexa?

I usually try the Hexa first, if un-successful then revert back to tetra+prism.

There are many indicators on which one can decide which mesh type to use:

1. Complexity of model and relative impact on accuracy. For example it is not recommended to go for tetra for LES or DNS

2. It is well known that memory requirements are higher for tetra mesh due to higher cell count for same no. of nodes.

3. You need far finer mesh in tetra to resolve the flow features sharply as compared to hexa.

4. It is also told if your hexa is not aligned with flow then you will loose the advantage of hexa.

5. As far as propagation is concerned, you can avoid this by employing advance blocking strategy. For example check ICEM Hexa mesh of AIAA drag perdition workshop.

BrolY February 25, 2013 05:46

I was thinking of Code_Saturne for example. and I think, maybe I'm wrong, that CFD++ gives good results with tetra+prism. Star CCM+ is designed for polyhedrall meshes. etc ..

5) Sometimes you can, sometimes you can't and that's the limit of Hexa (I'm talking of blocking here of course).

Maybe add another point (which is related to 5)):
6) It requires more time to create a blocking tahn creating a tetra+prism mesh. This is even more true when the geometri is getting more complicated.

Far March 1, 2013 03:59

I would like to quote the lines from a project on mesh generation at University of Cambridge

Mesh Generation
This project explored the solution of differential distance function equations on unstructured moving and overset meshes. Flow solutions using such grids are increasingly common. However, robust mesh generation still presents a significant challenge. This is especially so if use is made of hexahedral cells and the higher numerical fidelity that they provide. The differential equations explored will include the hyperbolic Eikonal, Hamilton-Jacobi and elliptic Poisson. Especial attention will be paid to the economical and accurate solution of these equations on moving, unstructured over-set grids with mixed element topologies using finite element, finite volume & boundary element methods.

http://www-diva.eng.cam.ac.uk/direct...esh-generation


All times are GMT -4. The time now is 04:15.