A recent comparison test between two alternate shapes used in an assembly simulation has me wondering if averaging listed stress values within some percentage of peak values is a viable practice in determining the more favorable outcome. I've heard of this practice before but I'm not clear on what elements in the overall test should be removed from the stress output used to calculate the average, i.e. how far do you go in removing the outlyers.

Regards,

Dave Trent

In the real world you don't get singularities - you get non-linearities. The other option I use is to put in a continuous approximation (a fillet typically) and find out what sort of radius produces the limit I am looking for with a mesh independent solution. That should provide some insight. Then its all about the gradient and how fast the crack will propagate.