Hi,

I wonder if anybody have experiences of getting different solutions by using different mesh in Fluent?

I have a quite simple pipe geometry with about 90 degree bend and rapid expansion. The diameter ratio on both ends of the bend is slightly larger 3. The tetra mesh predicted a flow field with strong swirl component but the prism/hex mesh did not predict such swirl. Pressure loss is certainly very diferent. All three cases start from zero and converge well. I am almost certain that this difference is not caused by mesh density. I checked with the tetra mesh case. Questions I have in mind are: 1) if the two different solutions are merely caused by mesh type, should I trust one by hex mesh or by tetra mesh? Why? 2) if there exist multiple solutions of the governing equation set for this case, is there a way to determine their stability in Fluent?

Thank you!

This is quite a difficult question to answer. Need a bit more detail on the problem first.

Is this a tubulent simulation? What discretisation schemes are you using, 1st, 2nd order etc? Is the tet mesh a hybrid mesh (do you use hex in boundary layer)?

In general tet meshs are more prone to numerical diffusion and it is difficult to resolve boundary layers with tet meshes. However, if the model is setup correctly, Fluent should yield solutions with very small discrepancy. I would recommend using the hex mesh if possible because its less likely to converge on unrealistic solutions.

Thank you for your reply. Yes. Turbulence is modeled by k-epsilon model. All discretizations are 1st order upwind scheme. There is no boundary layer mesh. I am pretty sure the model has been set up correctly because we have done many of them but didn't notice this issue until recently. Maybe this is just for this geometry, I don't know. I tend to trust solutions by hex/prism mesh because they preserve the symmetry quite well (the ProE geometry has a plane of symmetry). Tetra mesh solution, howerver, showed a strong swirl. The sense of rotation is probably, I guess, caused by the slight imperfection of the ProE model.

you have to use at least 2nd order discretisation schemes with an unstructured grid. This is probably the cause of the rotation.

You should also probably put about 10 layers of hex cells on the boundary layer to capture the sharp gradients and fill the rest with tet cells.