4 Replies Latest reply on Jun 28, 2012 8:27 PM by Kevin Corr

    Nonlinear Analysis running slow - Did I miss any settings?

    Chris Nelson

      Howdy all,

       

      Let me start by saying that I usually set up a static analysis to check my analyses first.  That lets me see if I'm in the ballpark with my loads & restraints.  Once everything looks good, I'll create a non-linear analysis & copy everything over (except the material properties, which I want to be non-linear).  Press the run button & let it solve overnight.

       

      Now this system has worked fine for me most of the time.  But I've seen a few cases where one night of solving isn't even close to enough time.  Instead of solving overnight, I'll come back the next day & it's only 1% done.  Like it needs multiple weeks to solve.  The thing is, these analyses aren't really that much bigger in element count or complexity.  Slightly bigger sure, but not nearly to this scale.  It's like I've crossed some sort fo phantom line in the sand. 

       

      I've checked all the usual suspects I can think of.  Plenty of RAM, 64 bit machine, SSD drives.  Watching the ressource monitor shows that the core is pegged at full usage.  I'd believe this was a swap space / page file issue, but the memory usage isn't even close to needing to swap (analysis is requesting half a gig of RAM, machine has 32 gigs).  Is there some sort of limitation in SW that would cause this behavior?  Any setting I need to be changing?

       

      -Thanks

      -Chris

        • Re: Nonlinear Analysis running slow - Did I miss any settings?
          Roland Schwarz

          You might revisit your linear results and see if any of your materials are going past yield or UTS.  If stresses are getting into high-curvature parts of the stress-strain curve, it could take a while.

            • Re: Nonlinear Analysis running slow - Did I miss any settings?
              Kevin Corr

              I agree. Other usual suspects in the nonlinear study are curves and no penetration contact. Auto-stepping the time increment vs. user defined time step is also a frequent peril. I would animate the last time step's displacement plot and see if it behaves as intended. If buckling or other sudden behavior is observed, the arc-length control option may help reconcile big jumps in displacement over small time steps.

            • Re: Nonlinear Analysis running slow - Did I miss any settings?
              Chris Nelson

              Y'all seem to have nailed the problem exactly.  The stresses should be right near the top of the curve, and I've defined some no penetration contact sets as well.  Sounds like that combination is going to result in a slow analysis. 

               

              Anything I can do to help it along?  I'm vaugely familar with time steps, but I'm unaware of any guidelines as to how or when to change it from autostepping.  I'm even less familiar with the arc-length control option.  What exactly does it do?  Is there a time when I should NOT use it?

               

              -Thanks

              -Chris

                • Re: Nonlinear Analysis running slow - Did I miss any settings?
                  Kevin Corr

                  What did the displacement animation inform? I would halve or quarter the model, mesh and re-run. You may have an element problem where two nodes are "tangled". Try remeshing with draft elements to get faster return during troubleshooting. Scale back up to full model when the problem is solved and apply the fix everywhere.

                   

                  Others can delve into arc-length control's mathematics. ("...Equilibrium path..." ) Strategically, its used in situations where very large stiffness changes occur during the study, as in buckling. By default it is not used because it has more number crunching to do per step.  I brought it up mainly because stalled solving after t=0 reminded of its use. That simulation was predicting buckling after arc-length was switched on was an eye opener in many cases.

                   

                  By curve, I meant the relationship between loads and time ie. force vs. time. By default the relation is 1 to 1. By user definition, it can be exotic and can be a source of problems. By the same token, it can be very gentle.

                   

                  From the linear study you said you did prior, the clue to find is which no penetration contact is producing the biggest nonlinearity. Process of elimination can be used if all of them look equally nonlinear. If just one is found to cause the problem, one possibility is to increase the friction coefficient.

                   

                  If symmetry is employed as a boundary condition, use loads proportional to the reduced area caused by symmetric geometry.