11 Replies Latest reply on Feb 3, 2015 2:30 PM by Jared Conway

    Simulation / Meshing / Stress Error

    Mike Armstrong

      I have a relatively simple model with 3 solid bodies and have been running simulations to try and get consistent results. I started off with a very coarse mesh and to ensure the solver managed to converge.

       

      I have gradually reduced the mesh size in there areas of interest (pull connections and locations of applied restraint) and have ran into a number of problems. The maximum stress values continually rise the more I refine the mesh due to singularities and the stress error plots return extremely high values. I have used a number of techniques provided throughout the forum (not shown in this model), such as fillets, simplification of geometry with little success.

       

      So my questions are:

       

      • What mesh size should I be using at the areas of interest and the areas deemed not critical
      • What stress error is acceptable
      • What techniques are best to eliminate singularities

       

      Thanks in advance.

       

      Mike

        • Re: Simulation / Meshing / Stress Error
          Bill McEachern

          HI Mike,

          I did not download your model so these are just general remarks. The best way to deal with singularities in my humble opinion is to do a non-linear analysis and let the equilibrium iterations deal with them but you may not have the NL option. Try and turn on large displacement in the study properties - it will do the same thing but you won't have a plasticity model so they might still be high. Sim NL with plasticity and contact is not the most robust of combinations for getting a result to converge.

          Assuming your areas of interest are not where the singularities are then just make sure the stress in those areas has leveled off as you reduce the mesh. You can plot the error there but they should be ok once the stress converges.

          Also you don't really want your areas of interest near a restraint as they are idealized and typically overly stiff in some respect but this is not always the case for a well chosen restraint system.

            • Re: Simulation / Meshing / Stress Error
              Mike Armstrong

              Thanks for the response Bill,

               

              I do have NL option available but have to admit that I have not used it before so I will have to do a little research. Scouring the net I have just noticed that my analysis should meet certain criteria before using NL.

              • Has the maximum stress exceeded the yield stress for the material? Or are any of the material properties modified as a consequence of the stress level reached?
              • Are the deflections large? (rough guide: if deflections are close to the thickness of the material then the deflections are large).
              • Is stress stiffening/stress softening occurring?
              • Are there dynamic or inertial effects?
              • Is a no-penetration contact area expected to change shape or size during the simulation?
              • Is a material experiencing creep or viscoelasticity?
              • Does the load direction/magnitude vary due to the deflection of the model?

              I am not sure my model meets any of the above criteria, well it doesn't until I start refining the same.

                • Re: Simulation / Meshing / Stress Error
                  Bill McEachern

                  A linear analysis is a projection of the current state (unloaded) to the loaded state. The same analysis done with an  NL solution will do the projection and then iterate the projected solution till the solution satisfies the equilibrium constraint. The linear approximation is "less accurate" (but often close enough) than an NL solution all other things being equal. The equilibrium iterations allows the smoothing of the singularities with or without a plasticity model.

                    • Re: Simulation / Meshing / Stress Error
                      Mike Armstrong

                      Bill McEachern wrote:

                       

                      A linear analysis is a projection of the current state (unloaded) to the loaded state. The same analysis done with an  NL solution will do the projection and then iterate the projected solution till the solution satisfies the equilibrium constraint. The linear approximation is "less accurate" (but often close enough) than an NL solution all other things being equal. The equilibrium iterations allows the smoothing of the singularities with or without a plasticity model.

                      Thanks Bill I will try this when I get my home machine back working.

                       

                      Cheers for the explanation.

                • Re: Simulation / Meshing / Stress Error
                  Jared Conway

                  what is your expectation here?

                   

                  to remove the singularities all together? this is basically not possible, you can only minimize their effects

                   

                  use displacement to determine that your solution has converged

                   

                  add a fillet in the area of interest

                   

                  improve the mesh to the point where the solution is taking too long for you and look at a few elements away

                   

                  back up results with hand calcs