I've got a model with contacts that is being run with several thousand cases in a Monte Carlo through the external solver. It runs statics and then dynamics. 99% of the cases run fine but a handful of them fail to obtain a static solution. The easiest way to resolve these failures is to change the contact faceting tolerance slightly (300 to 310 or something) and rerun. Is there a way to automate this behavior within an Adams solver script?
This is the behavior I want:
1) run sim/static
2) if failed, change faceting tolerance or some other setting and rerun sim/static
3) otherwise just continue to the dynamic solution
The only way to do this, as I understand, is to write a consub. Through the consub you can have full control over the simulation, down to step-by-step interaction if so is desired. But there is no acf command that allows this.
Fortunately the change is small enough that it doesn't appreciably impact the simulation results. I've tried changing the faceting tolerance globally and a different subset of cases will end up failing the static solution. It ends up being a bit of a whack-a-mole situation where a handful of cases (~20/2500) will end up in a slightly unstable configuration that the static solution can't work out, no matter what settings (ALIMIT, TLIMIT, MAXIT, TOLERANCE, etc) I use. For the cases that do fail changing the faceting tolerance slightly (could increase it or decrease it) changes the contact mesh enough that the instability goes away and they end up solving statics in 10-15 iterations.
Mike, try increase the STABILITY parameter from 1e-5 to 1e-4. That can solve a lot of stability problems in the static solver. It will converge slower, but more monotonously.
Unfortunately no luck with stability parameter changes. Ends up with a similar magnitude of equation error in the contact forces (~50N) but fails to converge even if I let it go for hundreds of iterations. A change of contact tolerance as small as going from 300.0028 (default) to 300.0128 is enough to let it solve. It makes me think that the contact mesh vertices are just lining up in some unfavorable manner (I am slightly dispersing the position/orientation of the contact geometry in the Monte Carlo).