A better way is to create an output scoped to elements that exclude a layer of elements at the few singularity locations that coincide with the location of peak stress.
Many locations where there is a singularity may never have stress values that rise to a level of concern, so those don't need special treatment.
It could happen that the peak stress in the model occurs at a location that has no singularity, so filtering stress at that location would discard valid data.