I'm running a simulation of time dependent deformation and failure at a notch root in polyethylene. I'm using Jorgen's VUMAT, the TNM. When the model reaches the defined critical state (critical chain strain or ultimate true strain) it throws the failure flag that Abaqus sees, and Abaqus unloads the offending integration point. The problem is, the remaining integration points in the element survive and are apparently protected from failing by the unloaded "dead" material point. The result is that I get these organized stacks of elements that are partially dead, and this partial death spreads through the model instead of any elements being removed. Eventually the element stacks start deforming in impossible ways and the simulation breaks down. Has anyone seen this sort of frustrated deletion problem? Anyone have insight?
Also, I am scaling the mass a great deal because the elements are tiny (L=1e-6), but the process is essentially quasi-static so I think this is OK unless it brings in a problem that I don't recognize. I'm running it in double precision, so I think the calculations are OK, and therefore the problem is systemic.