9 Replies Latest reply on Apr 26, 2018 9:14 AM by Chase Evans

    Solver failed. Restart analysis?

    Eric Christison

      I'm running a linear static analysis for a contact problem. The solver is using large displacements.

       

      It keeps failing, but when it does it asks me if I want to restart the analysis.

       

      What happens when the analysis is restarted? Can I rely on the results it gives me?

        • Re: Solver failed. Restart analysis?
          Chase Evans

          What exactly is the error message it provides? And what kind of magnitude of displacements are we talking about relative to the size of your model?

            • Re: Solver failed. Restart analysis?
              Eric Christison

              What I put in the subject is pretty much what it told me. I'm running SW just now so can't go back, but from memory "Solver failed: do you want to restart the analysis?".

               

              Deflection is about 2% of the thickness of the plate I am analysing.

                • Re: Solver failed. Restart analysis?
                  Chase Evans

                  The reason I asked was trying to figure out whether it was failing under the standard linear solver and asking if you wanted to re-start with large displacement. It sounds like you started the analysis from the beginning with large displacement - what is the reason for that? 2% of the thickness seems relatively small and well within the capability of the standard linear solver, I would un-check "large displacements". If you truly feel that you need large displacement calculations I would move to a non-linear analysis if you have solidworks premium as checking "large displacements" inside a linear analysis is a psuedo-nonlinear approach and a poor substitute for a true nonlinear analysis. Either way I would run the linear static simulation you have current WITHOUT large displacements and see what you get.

                   

                  Also if you are not already I would utilize the direct sparse solver instead of the iterative solver for contact problems. Direct sparse is much, much faster for contact problems - in fact i've had multiple situations where using the iterative solver for contact either totally failed to converge or took an inordinate time to converge and the direct sparse would solve in a tiny fraction of the time, its just not the way to go for contact analysis.

                  • Re: Solver failed. Restart analysis?
                    Simon Yang

                    put some screen shots here maybe better to see.

                    • Re: Solver failed. Restart analysis?
                      Chase Evans

                      Any luck with the simulation? Did my suggestions help?

                        • Re: Solver failed. Restart analysis?
                          Eric Christison

                          Hi Chase,

                           

                          In some respects your suggestions did help in that I could get the problem to solve in linear analysis. I'm not convinced about the results though. The problem is reasonably simple, a steel needle with a spherically radiused tip against a piece of flat glass, and SW predicts stresses and deflections about 50% greater than hand calcs. Trying to refine the mesh (wouldn't more control over meshing be great) just got me to a point where I got some very peculiar stress distributions under the contact surface.

                           

                          Non linear isn't working. The solver failed at 4.8% of the whole load. I reduced the force by a factor of 10 and it failed around 50% of full load which I guess makes sense. Given the non linear nature of the geometry I see no point in reducing the load as I won't be able to simply factor the stresses back up.

                            • Re: Solver failed. Restart analysis?
                              Chase Evans

                              Eric,

                               

                              Glad my input was able to help somewhat. Now that you have described the problem more fully, it seems what you are looking at analyzing is surface/subsurface stress due to point loading (essentially round sphere on flat plate) which is in the realm of hertzian contact. From what I gather solidworks simulation is not very good at modeling hertzian contact stresses - I am not sure about the exact reasons, but I just don't think their solvers handle those particular contact situations very well. You are correct in that the nature of hertzian contact is nonlinear however I don't think that the nonlinear solver will produce better results as I believe you will run into similar issues with inaccuracies in the solver. I believe your hand calculations are probably going to be more accurate than any solution solidworks will generate - there are probably more advanced (and more costly) FEA packages out there which may be able to handle these contacts.

                               

                              Also, although it is probably outside the realm of what you are looking for, I'm not sure how most FEA packages would handle or accurately model a material like glass seeing as it is extremely brittle and susceptible to crack/notch formation under loading - especially in a hertzian contact as the local stresses are extremely high. I guess as long as the loading is relatively low or below a certain threshold it might be fine but I would definitely do some research in this area.

                               

                              FYI - just passing on some knowledge that I recently learned. I know I suggested direct sparse for the linear solver for contact analysis (this suggestion still stands for LINEAR analysis), however since you are looking into nonlinear analysis I just wanted to let you know that apparently for nonlinear analysis the FFEPlus (iterative) solver will be more efficient in many cases. See the help file for more information:

                              2018 SOLIDWORKS Help - Analysis Solvers http://help.solidworks.com/2018/english/SolidWorks/cworks/c_analysis_solvers.htm?verRedirect=1