CCX version updated to 2.23 vs 2.22

FatmacFatmac
I have seen some differences in the solution generated from the same model in Mecway when I use 2.23 and 2.22 ( specifically a non linear contact involving bolts and pretension). In some cases 2.22 converges easily and 2.23 requires a different analysis set up to converge whatso ever. Has this been reported else where and are there any potential reasons for this? which solution is likely to be more reliable? I have attached the file 2.22 solves with out the need for automatic time stepping but 2.23 does not - the maximum stress is also different. I can well believe my approach to this particular problem may be questionable but I'm surprised at the difference in the solutions.
Incidentailly is there a way to be able to use the PARDISO solver with 2.23 ?

Comments

  • VictorVictor
    File not attached.

    That is a worry if they both converge with different solutions. I'm looking forward to seeing your example.

    You can use PARDISO by building from the source included with Mecway, or possibly download a compiled binary from dhondt.de
  • FatmacFatmac
    I have attached the file and some pictures. In the end i got myself a little confused - in one case i appear to have got different solutions with the same solver but one using the keywords and one using the default which suggested it was using the same solver! Quite possibly I have got hold of the wrong end of the stick (again!)
  • FatmacFatmac
    and the file
  • JohnMJohnM
    I ran it in 2.20, it wanted the auto time stepping like 2.23 or it would not converge.
    I would believe the results the required the better time stepping.

    If you run the executables availailable on dhondt.de, switching to PARDISO in CCX Modify keyword

  • FatmacFatmac
    Yea auto time step certainly helps convergence. The bit that surprised me was the very differing solution with the same solver but different version- see pics in last posting
  • VictorVictor
    edited March 23
    From your pictures, it looks like Spooles doesn't converge in either version so there's no inconsistency there, and Pastix is the one that gives different results. Is that right?

    Pastix can give random results with mixed precision, which is the default, so make sure you turn that off using the environment variable. Even with double precision, I'm not sure it tests for convergence properly, so maybe it's giving results when it should have failed.

    I only test Mecway against the version of CCX that's included with it, as well as Pardiso built from the source included. Other builds might not work as well.
Sign In or Register to comment.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In Register