how can I obtain convergence with the RSM model in a cyclone separator simulation? Is there a relation between mesh quality and convergence?

If your mesh quality is poor then you will never get a converged simulation, especially not for the RSM which is very sensitive to highly skewed cells. In my experience you should make sure the maximum equiangle skew is no greater than 0.9 and ideally below 0.85.

You may have 20million cells and all it takes is a few poor ones and convergence is impossible. It will almost certainly be that giving you problems. Try finding out where the poor elements are and re-mesh with a finer spacing in those regions. Usually poor mesh occurs because of tiny faces in the geometry, again you should try to avoid them when you create a model.

Once your mesh quality is better you can also smooth it in fluent (Grid->Smooth) and gradually reduce the skewness even more, just be careful you dont smooth it too much or you can get some problems. If your simulations dont converge still then run a k-epsilon model to get a converged solution and then use this result to start off a simulation with RSM. Also, you can reduce the underrelaxation factors to improve convergence (Try making the momentum factor 0.5 instead of 0.7 and reduce pressure from 0.3 to 0.2).

Good luck!

Carlos.

In addition to the comment of carlos, use hex mesh only. Cyclone separators is having a complex physics and gives results with hex mesh. Use Presto for pressure, and Quick for others as discretization scheme.