I am writing a code based on SIMPLE method. I am following the approach mentioned by peric. I have been observing one thing, that when I use structured meshes I get good results, but as soon as I shift to unstructured grids, the solution becomes unstable after few iterations, say 10 or 15. if I stop the solution before 10 iterations and observe the velocity-pressure field, I can see that in some parts the velocities are just too high. The pressure field also do not resemble the correct values.

Perics books mentions that the problems could come on tet meshes and fluent�fs manual also mentions this thing that on tet meshes we can see this kind of behaviour. What both the sources do not mention is what is the cure for this problem.

Can anybody give me some suggestion for this problem. I am using segregated solver. Thanks in advance.

Since you're using fluent to generate the grids, the problem is not with grid generation. If your code works on a structured grid and not on an unstructured one, there could be some problem with your spatial discretization.

actually on unstructured grids, the code works if the meshes are not too distorted, i mean if the mesh has some control volumes very small compared to neighbouring volumes, and if the angle is not good , the solutions goes off, otherwise it works on unstructured grids if the grids are simple, i checked with the descetization and i could not find anything that i am doign wrong, anyway i will keep looking, there might be small logical mistake in code, can eat the whole thing

Well, I'm developing a FEM code for incompressible flows, but with quadrilateral mesh. If the elements - or cells, to be more generic - are too deformed, then you'll probably have covergence problems. I have never used fluent, but commercial codes often have a tool for checking the mesh quality. I have experience with ANSYS only, and it performs a quality check just after the grid generation. If you must have control-volumes with different sizes, try to make this transition as smooth as possible, and it may works. Anyway, grid generation is an important step in CFD analysis, just because of problems like this. If the cells are too distorted, it's very likely that some problem will occur.

Hope it helps

Márcio Ricardo

thanks recardo, you might have given me the clue that i needed, i think smoothing will work, i am going to apply a correction to variables ensuring that they are smooth.

and about fluent, the same mesh on fluent works fine, thouhg i used to think it might be because of multigrids they use, but then i put the coarsening level to zero and still in fluent the things are fine.

anyway, i can not give more than this week to this finite volume, my main aim is to apply meshless type solver, (first i shall treat fluent't mesh as meshless for testing), this finite volume code is just to learn what it takes to write a cfd solver. and the meshless types solver i am thinking all the control volumes are overlapping and are just simple cubes, so all this skewness etc is not goign to bother.

anyway thanks again