With the bulb configuration, the Singular Suface boundary condition will be involved. This kind of condition is not so stable, such as the computation could be well converged in most of cases but not sometimes. So, if possible and you do have difficulty in convergence, you should try to avoid using the bulb. Another option is to creat the mesh in bulb range separately so that you can make the first cell near the machine axis (R=0) being larger. When this cell is too small in the radial direction, its dummy cell might be have negative volume which may cause convergence problem.
As I known, NUMECA is working at making this boundary condition more robust.