SEGV in my experience has always been segmentation fault,
and the problem may be in the ATLAS program or how it
deals with some content of your models.
Since you know when it came on and even a limited number
of possible actors, you should be able to debug at least down
to the specific offending file.
You could of course also ask the applications engineering group
for the software, who might have an actual chance of knowing
what's up.
I have a quantum theory of my own. To wit:
"If the description contains 'quantum' then there is a finite
probability that it's all bullsh!t".
Have anything to do with research grant funding on the
topic, and you cannot help but to agree.