Has someone carried out compartive studies with the structured and unstructured version of Fluent for single phase cyclones, hydrocyclones, or in general swirling flows?. I have done some using the same models and getting grid-independent solution for every version and found big differences. Does anybody know the reason?

Be aware that standard k-eps models are not very good at swirling flows . Perhaps the wall functions (if you used them) are defined slightly different in the two versions.

Haven't done the kind of study you are mentioning, though..

We have done some of the same kind of tests and have the same problems. The unstructured code often give somewhat unphysical results, whereas the structured one seems much more reasonable.

We have tested this for k-epsilon, RNG and Reynold stress models, and the same problems occur.