+1 on the mesh. How far along does the analysis get before it diverges? Since (I assume) it is a transient analysis, try to view the results just prior to divergence. Look for signs of oscillation, unnaturally high gradients near BC’s/walls, mesh dependence. Also, perhaps try a K-e model first, as it is generally more stable than the K-W.
According to SST k-omega model -- CFD-Wiki, the free CFD reference, K-e is used in the free stream, and K-w is used in the boundary layer. So if you can get it to converge with the K-e turb model, then you know your issue is in the boundary layer. At that point, investigate the mesh near the walls…