1 Reply Latest reply on Dec 7, 2010 8:44 AM by Ryan Werner

    contact convergence in large model

    Peter Osgood

      I have a large assembly with multiple contact surfaces.  The model will solve with a very coarse mesh.  When I refine the mesh in high stress areas, the solution will not converge.  I am using FFEPlus and when I look at the convergence plot in the solver, it bounces around near the convergence limit.


      In the solver, there is a place to enter max. steps and convergence limits, but they do not do anything.  I will set them and come back later to look at them and they have reverted to defaults.  I have also tried to cancel the solution, get the popup to save results and say "yes", but the results are not saved.  The only way I can terminate the solution is to go into the Task Manager and end the star.exe process.


      I have tried DirectSparse and that does not converge either.


      Is there anyway to set the convergence limit or maximum number of steps to stop the solution so I can see the results after a certain point?  Ant other suggestions?



      Pete Osgood


      I ended up making the mesh more coarse and adding springs to stabilize the parts.

        • Re: contact convergence in large model
          Ryan Werner

          Hi Pete,


          I believe the only way to set the max steps and such is how you have already done it.  Unfortunately, when you have contacts set up it seems that Simulation must reach the default convergence limit in order to show the results including those contact sets.  How much refinement have you made to the mesh and how long have you allowed your analysis to run?  In a large assembly with contact sets, especially No Penetration, it may not take much of a mesh refinement to greatly increase the solve time.  I would prehaps try doing the mesh refinement in smaller steps until you can get results you are happy with in an amount of time that is appropriate.


          As another option I believe you could try running this same analysis as a Non-Linear type.  This would give you the ability to specify the number of steps Simulation takes before returning results.  I have not tried this method myself but have read threads describing it.  Good Luck.


          Ryan W.