It depends on your simulation's circumstances, for example, input boundary condition should be rechecked.

All turbulence models and separation calculations (e.g., k-e/weak or strong adverse pressure gradient) are based on a finite set (perhaps as few as one) experimental data sets. No doubt, one calculation may work very well in some cases and not in others. If it didn't ever work well, no one would use it. CFD is so complex and the circumstances of flow are so diverse that no model or method will always work.