16 Replies Latest reply on Jan 5, 2018 5:57 PM by Qinghai Jin

    Extremely long solution time on nonlinear static shell model

    Jesse Blake

      Hello,

       

      I have a model of a large structure consisting of ~800 surface bodies and a few springs. The surfaces all mate up nicely and I have created split lines where bodies meet up. Contact is set to bonded global. It meshes with shells perfectly, matching nodes across all the joints, and has about 130k nodes in total.

       

      When I run a linear static analysis it meshes in 49sec and solves in 45sec. All the contacts/loads/restraints are applied correctly and the stresses/defections are what I would expect both.

       

      I would like to run a nonlinear static analysis on the structure to determine when it starts to buckle. To do this I ramp up a small load from t=0-0.1s to set the initial condition that will allow the onset of buckling. Then I apply an increasing load from t=0.1-1s. When the analysis runs it typically jumps to very small time steps (0.01s and less) initially, and takes 5-6 hours to solve each, reaching roughly t=0.03s over a period of 15+ hours. Looking at the linear static solution for the t=0.1s loads, the deflections are small and the stresses only ~10% of yield so its certainly not in a highly nonlinear area.

       

      I have gotten previous (and slightly simpler geometry) versions of this model to solve, taking 8-12hrs to reach t=1 in ~17 time steps. This still seems like a very long time considering each static mesh+solve cycle is <2min on its own.

       

      What could be causing this behavior?

       

      I am running Solidworks Simulation Premium 2013 x64 SP4.0 w/32GB ram.

       

      Thanks!

        • Re: Extremely long solution time on nonlinear static shell model
          Bill McEachern

          It sounds like you are using  force control. If there is are not any no penetration contacts in the model you might have more success with  displacement control. However, it might be bad elements in the mesh. Lower mesh densities on some problems tend to let the solution to progress more readily.

          • Re: Extremely long solution time on nonlinear static shell model
            Shaun Densberger

            As Bill pointed out, you might want to try an enforced displacement (if possible) instead of an applied load. Non-linear analyses usually use some form of Newton's Method, which can have stability/convergence issues with applied loads (especially if the slope of the force-deflection curve becomes negative).

             

            Regarding the discrepancy between your linear and non-linear trials, there is most likely some improvements you can make to the mesh and load step to obtain an accurate solution in less time. However, depending on what type of non-linear effects you're trying to capture, there can be a very, very large difference between a linear and non-linear analysis.

            • Re: Extremely long solution time on nonlinear static shell model
              Jared Conway

              it meshes and solves in 2 mins in linear

              and then you only switched to nonlinear and now it takes 15+ hours to solve just to 0.03s?

               

              have you tried just running the same problem as linear in nonlinear just to get an idea of what effect that has only

              • Re: Extremely long solution time on nonlinear static shell model
                Jerry Steiger

                Jesse,

                 

                I'm not very familiar with SolidWorks Simulation. When you say you run a nonlinear analysis, are you just talking about allowing for large deflections, or are you using a non-linear material model, or both?

                 

                Jerry S.

                  • Re: Extremely long solution time on nonlinear static shell model
                    Jesse Blake

                    Jerry,

                     

                    I am using time dependent loads to try to buckle a structure from the induced changes in geometry. At the moment I am using a linear material model. Thanks,

                     

                    Jesse

                      • Re: Extremely long solution time on nonlinear static shell model
                        Bill McEachern

                        Post buckling can be tricky and the fact that is solves in linear is not something that should provide a lot of faith in being menaingful. Consider that you have a strucutre that has, as it is loaded, a small insifginifcant peice of strucutre that buckles at almost no load but does not really compromise the integrity of the structure as a whole - some little bracket some where trips at a small fraction of the over all structures capacity. This can cause your solution to hit the ditch lickety split. Displacement and arc length controls allow you to avoid the solution stoping at some inflection such as this, the reasons for which you can do some research on. Now notwithstanding what Gerry (post edit: sorry I meant Shaun and not Jerry - sorry about the sp as well)  said, in SWX sim you can not apply a fixed displacement as a load in a displacement controlled solution - you need to apply a force. What happens is the the solution process sort of gets reversed. In the displacement control dialog you need to pick a node - a vertex in SWX Sim - and a single degree of freedom - displacement in the direction your are concerned about or mroe loosely the direction the load is applied in. Pick a spot whose deflection and direction would be meaninful in assessinghte buckling response of the structure. Then you need to specify how far this DOF needs to go to resolve (buckle in your case). then what happens is the solver advanced the dof and calculate the load required to get it to whatever increment it is attempting and converge the force to achieve the displacment. This avoids ttying to iterate to convergence onto a peak of hte force dispalcement trajectory where the force tangent has a slope of zero. Force control will always have some difficulty overcoming buckling if the force slope goes to zero.

                          • Re: Extremely long solution time on nonlinear static shell model
                            Jesse Blake

                            Bill,

                             

                            Thanks for the explanation! It is very possible that's what was going on. I switched to arc length control and am able to run the model in a reasonable length of time.

                             

                            Do you know of a good online resource describing how to properly set up timed loads and time steps for arc length controlled studies? I'm still a little confused on the correlation between the time stepping options, the applied load vs time curves, and what "load factor" is when viewing the results.

                             

                            Thanks!

                             

                            Jesse

                              • Re: Extremely long solution time on nonlinear static shell model
                                Bill McEachern

                                Time in a static simulation is not really time. It is more a way to control the sequence of things or process steps if you will. So a loading going from 0 to 1 can equally be modeled as 0 to 0.1 if auto stepping is enabled in say a force control. I just did a pre stress and then reload force controlled simulation that went  like this: at time 0 all loads are off, at time 1.5 I went to load factor 1.5 on a centrifugal load, at time 2 the centrifugal load went from LF=1.5 to 0. And then from time 2 to time 3 the centrifugal went back up to LF=1 as well as a pressure load went from 0 to LF=1 and the simulation stopped at time 3. This revealed the drop in stress due to compressively induced stress from yeliding that occured from the high centrifugal run on the normal operating conditions. Ther stess dropped from 40 ksi to about 23 ksi vastly reducing fatigue concerns.

                                In a displacemnt controlled solution I would typically put auto steping on and go from 0 to 1. I arc length you loose control of hte step size and SWX Sim doesn't auto increase the step size as it used to, at least in my recent expereicne it appears that way, so it can take a while to run. The perecent complete in a dispalcement solution is the fraction fo the displacement achieved by the control node. In an arc length solution it is the fraction of the max number of steps allowed. To find out more about the arc length method you could look  up more on Crisfield or Riks methods.

                                 

                                To clarify things on load factor set up some small little fast running experiments and explore the subtelties.

                                 

                                Hope that helps.