CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   CD-adapco (http://www.cfd-online.com/Forums/cd-adapco/)
-   -   Whats going on? (http://www.cfd-online.com/Forums/cd-adapco/53374-whats-going.html)

Joel July 7, 2003 08:25

Whats going on?
 
i have run a few compressible non-newtonian flow simulations using a structured cylindrical mesh, the results seemed ok the solution converged and the world was happy, i did however notice a few problems with the simulations and i decided i was not entirely happy with the mesh. So i have built myself a nice looking hex/trimmed cell mesh using pro-am and everything has gone wrong. The simulation settings are exactly the same, the user subs are all copied straight across, the simulation is identical in every was except for the mesh. And yet it diverges after one iteration with pressure initiallization failures. So whats going on? I have used all these settings on a couple of cylindrical meshes before hand with no problems in the running, and yet i get onto this cartesean mesh and everything falls ot bits, why? Any ideas?

star-user July 7, 2003 08:35

Re: Whats going on?
 
try mapping earlier results on this mesh and restart. It may help.

cfdweirdo July 7, 2003 17:45

Re: Whats going on?
 
The trimmed cell (degenerate hex) type of mesh is essentially crap. You end up with elements that look like they have been run over by a mack truck.

Garbage mesh = Garbage answers

Stick to your nice cylindrical hex mesh!

CJ Tune July 7, 2003 21:21

Re: Whats going on?
 
Well, I avoid tet meshes because I often get cells that look as if they're still stuck to a Mack truck's tire, and usually they're at the interior, making repairs near-impossible. At least with trimmed cells you can 'pop the pimples' (hand-fix unresolved cells), and they're almost always at the surface.

When using trimmed or tet meshes, make sure you have prism layers generated at the wall boundaries. These kind of unstructured meshes cannot resolve the boundary layers properly without prism (extrusion) cells. Manual recommends 3 layers for hi-Re. You might also need to generate at least one layer of prisms for trimmed cells at boundaries for use with supersonic cases.

Also, I've had an experience once where the corner cells of a 'boxy' model were trimmed cells, ie. the corner tips weren't well-represented and my whole run was unstable. Hand-fixing those corner trimmed cells to hex cells saved the day.


All times are GMT -4. The time now is 06:04.