Somebody give me some advice please, I'm really not sure what to do next.

I'm running large (1 million cell) cyclone simulations. No matter what I do, I can't get them to converge on second order equations (using RSM or KE, steady or unsteady). I run them until convergence on first order, but the second orders wont converge.

I tried playing with the relaxation factors to the extreme without any success.

By the way the mesh I'm using is a tet-hybrid hexcore.

Any suggestions?

Often it happnes that even after acheieving the first order convergence with second order schemes problems do not converge. However monitoring few parameters and watching their behaviour with iterations might be useful in such cases. If not satisfactory then you may like to see the higher residual locations, and you may do some grid refinement in that region.

Although this may be a different topic of discussion, solution using second order schemes may be much different than first order in case of Hybrid mesh. And in my opinion different for betterment. So it is better that you are opting for it.

All the best!

there are two thing and one main cause of problem.

When you switch from 1st order to 2nd order you introduce one more variable and this is gradient.

Fluent uses limiter on gradient but still it could give problems.

The gradients are used in diffusional term mainly and this is where the problems start. So the cause of all the problem is diffusional terms.

Since i said there are two issues and one cause (diffusional term) the other issue is hex core mesh. Though it is good type of mesh it is not really fit for diffusional term if one box is connected to mainly small boxes (in mesh). (it happens if there is refinement).

To tackle this diffusional term issue many people run with addional corrector steps. Though people say it helps, i never had to resort to it. Try switching to PISO or SIMPLEC and increase the corrector steps. Let us know what you find.

Okay I'll try them. As an aside, I remeshed one of the models without the hexcore using tet/hybrid Tgrid. The models solved (and solved sooner) using this model.

I like the results I got with the hexcore better though, but I'll give some of your suggestions a try and let you know.

i really did not mean that hex core is bad, but when a hexa's face is connected with many smaller faces of other hexa control volumes, you could see that the line connecting cell centers and the face between them create a very bad angle. This is problematic for diffusional term and actually area of research. Recently i saw a paper about how to improve this part. So researchers know that this is a problem and they are working on it. Hex core look like cartesian mesh and we feel that it has to be better than other elements. It is usually better but could be problematic in some cases.

For cyclonic flows you need :

-hex mesh only (dont use hexcore here) -RSM (due to anisotropy) -QUICK for pressure -double precision -cell based gradients

BTW 1million cells is a low count but check grid independancy. Use manual meshing/decomposition to ensure close boundary layer meshing.

I agree with Red Lemon - try to get a hex mesh. Although maybe you should start with the k-e model, and once you have a converged solutiuon for this try the RSM. Quick also is good.