CV Solving Speed...
I work with CFX-5 and I was wondering, which kind of element costs less time to solve? Tetrahedron, Pyramids, Prisms or Hexahedrons?
And another thing, how less accurate those elements are in comparison with the hexahedron one?
Re: CV Solving Speed...
Tetrahedrons are generally more costly. They require more memory (~400 real words per node, when solving u,v,w,p,k,epsilon, and ~100 integer words per node as compared to around ~200 real / ~ 60 integer for hexes) because there are a large number of neighbor connections. Assembly time per node is slightly more on a tet grid also. You have about 5-6 times as many elements per node, but 3 times less integration points. So, that works out to about a factor of 2 or so if you have the same number of nodes.
As far as accuracy goes when compared with hexahedrons the argument is still out somewhat. If you have a grid aligned flow the solution on hexahedrons is always really good, since errors cancel perfectly, even with a first order upwind advection scheme. With tetrahedrons the flow is never grid aligned, so you really have to use a decent second order scheme to get good answers. However, in most real situations the flow is never grid aligned, so you end up having to use a decent second order scheme on hexahedral grids as well.
It's also more difficult to judge grid independence with a tet mesh. i.e., what does it mean to double or triple your resolution.
Often hex grids are used to resolve boundary layers. To help this situation, CFX-5 can easily use prisms in boundary layer regions as well, which helps with accuracy on tet grids alot. I wouldn't do a calculation without them, especially if you are interested in things like drag or pressure drop.
I think hex grids are nice if you have the time to make one, but a hybrid mesh can do just as well. I suppose it depends on what sort of geometries your doing. Complex geometries with hex grids can be a real nightmare, and some of the new features in 5.5 (Edge and volume proximity in particular) make user intervention in the meshing stage minimal. In many cases there is little or no need to use mesh controls to resolve the geometry.
Throw mesh adaption into all this too and it even gets more complex. It makes my head hurt.
|All times are GMT -4. The time now is 15:32.|