Optimal Settings for Internal Solver

Roli

Hi again,

I am calculating a model (linear static) with about 170.000 nodes. It is fast with the ccx solver (about 3 minutes). With the internal solver I do net get an solution even after one hour. I want to use the compression only support and thats why I depend on the internal solver. Are there possibilities for optimized performance with the internal solver (eg. computer settings)? I am using windows 8 with core i5 and 8 GB Ram.
Instead I tried with the CCX solver (non linear static) and as geometry modelled supports using contacts to have similar conditions. But the solver stops after one iteration with this error:

*ERROR in e_c3d: nonpositive jacobian
determinant in element

Knows anybody what to do on this? Refine the mesh on the elements does not work.

Thanks, Roland

Victor

What does the convergence graph look like? Is generally decreasing or oscillating between identical values?

If there's no graph yet, it might be stuck on the first iteration. It's possible that the model is big enough to cause the solver to use the disk instead of RAM. That makes it much slower. You can check Task Manager to see if there's heavy disk during the "Solving matrix..." stage. If there's almost no CPU use and no disk use, then it's probably crashed in some new way and will never finish. With enough RAM, I wouldn't expect that model size to take that long.

Here's some older data on solver performance:
http://mecway.com/forum/discussion/comment/2332/#Comment_2332
http://mecway.com/forum/discussion/comment/689/#Comment_689

Generally, I think CCX's contact performs better than the internal solver's compression-only support in terms of iterations till convergence as well as risk of not converging at all.

Nonpositive Jacobian means an element is severely deformed (partly inside-out). You can find the element using Edit -> Select elements by number. It might have become deformed after the first iteration because the material is too soft/strain too high.

Can you post the .liml file? If it's confidential, you can email it to me at the address at the top of the page.

Andrea

*ERROR in e_c3d: nonpositive jacobian
determinant in element

Free mesh is fast but can lead to neglect the quality. This error is often linked to a very small element volume.

Refine the mesh is generally a good solution but is better to made a partion of the body (or bodies) to have a gradual element dimension decrease. This strategy is valid for tetra elements too not only for hexa.

I think that many times we force the solver to make more and more calculus because the poor quality mesh

Roli

Very thanks!
Victor made it run with CCX and internal solver. The problem was probably the geometry I made for the supports (jacobian determinat) with CCX. Flat contacts are working now.

With the internal solver a little more RAM is needed and 8GB is not enough, thats why it takes such more time.

In general, the quality of my mesh is very poor. For the geometry of the welded assembly it is very hard to mesh it without doing it for all parts manual.