Try to remesh with linear elements that have half the element size. The total number of nodes will be about the same, so the solve time will be similar, but linear elements might behave better than the quadratic elements you have.
You can also make the linear elements even smaller so that the hot spot that EKILL removes is at least four elements across instead of the one or three quadratic elements I see in the image above. That will take longer to solve, but might be more stable.
Anytime you have highly distorted elements, it is always important to make some edits under Analysis Settings, turn on Auto Time Stepping. You can configure that by Time Increment in seconds or Number of Substeps. The important point is to use much smaller Time Increments for the Initial and Maximum time steps so big time increments are avoided. You can allow the Minimum Time increment to be two orders of magnitude smaller to help the solver to converge each substep.
You are navigating away from the AIS Discovery experience