10 Replies Latest reply on Nov 11, 2013 10:38 PM by Jared Conway

    Draft quality mesh

    Chris Fernald

      Hi,

       

      I was wondering when it was/wasn't appropriate to use draft quality meshing.  The help files do a good job at explaining the differences between the two, and the lack of accuracy with draft meshing.  However, I very rarely use draft meshes; only when I want to test a model for stability and convergence.  I am wondering if there are other cases where the draft mesh might be appropriate.  For example, I thought maybe the reaction forces would still be accurate - but I am finding this is not the case (at least in models with no penetration contact).  Any thoughts?

       

      Thanks!

        • Re: Draft quality mesh
          Jared Conway

          best use is for troubleshooting. make sure you're model is working in a shorter period of time.

          thermal analysis could theoretically work with draft quality elements without too much penalty.

          otherwise unlikely to give very good results because it changes the stiffness of the structure and the way the stress is mapped. so use with caution for "final" analysis studies unless you're using the data to understand trending.

          • Re: Draft quality mesh
            Mike Pogue

            I would add that the draft mesh is likely to reproduce displacement and temperature reasonably well, but stress and heat flux poorly. It's a little complicated, but it depends on the differentiability of both the function and the element shape functions. Draft quality elements are linear. Linear elements can only be differentiated once before you hit zero. The high quality elements are quadratic. They can be differentiated twice before you bottom. The bottom line is both will converge to the same answer as you decrease element size, but the draft elements will converge more slowly, especially with respect to flux or stress.

             

            Another thing to consider is geometry. If you put draft (linear) elements around a circle, you are actually modeling a polygon. The answer will be a poor approximation unless you have a very large number of elements around the circle. With high quality (quadratic) you get a pretty good approximation of a circle with about 8 elements (as a starting point, you still have to examine the convergence). The reason is obviously that a circle is a quadratic function and you are modeling the geometry correctly.

              • Re: Draft quality mesh
                Jared Conway

                i've seen draft quality elements under predict bending displacements as well. so overall, be real careful with them if you're going to use them.

                  • Re: Draft quality mesh
                    Mike Pogue

                    That's a good point.

                      • Re: Draft quality mesh
                        Chris Fernald

                        Thanks for the responses, guys.  This helps reinforces what I have heard in the past regarding draft quality meshes.  Mike, in general I find the draft elements converge more quickly as Jared suggested as a means for time-efficient troubleshooting.  Considering the effect on geometry is also interesting - curvature would be poorly modeled with draft elements.  I will continue to stick with higher-order elements unless I'm testing for trends or troubleshooting boundary conditions.

                         

                        Thanks!

                          • Re: Draft quality mesh
                            Jared Conway

                            they are solving faster because they have less DOF

                            • Re: Draft quality mesh
                              Bill McEachern

                              Doing tests with similar properteis can provide a lot of insight if you choose to use tet 4's - elements through the thickness and curvature refinement essentially. See how many element you need to get a decent response. I have seen large models that will solve with Tet 4's but will not solve with Tet 10's. For contact & NL I most always debug in tet 4's. Sometimes only a tet 4 mesh will solve - then it really pays to have a good feel for their performance. From what I understand is that any higher order element tends to develop a lot of "noise" in the convergence of mid side nodes so they take a lot longer to settle down. The FFE solver graph frequently looks a lot noiser with a tet 10 solve vrs. a tet 4.

                              • Re: Draft quality mesh
                                Mike Pogue

                                Chris,

                                 

                                Don't mix up convergence of the iterative solution with convergence of the mesh.

                                The convergence you are talking about is the trial and error of the computer as it works to solve a given mesh. This will always be faster for draft elements, all else being equal.

                                The convergence I'm referring to is how quickly the solution approaches the correct answer as you reduce element size and rerun the solution. This will always be slower for draft elements, all else being equal.

                        • Re: Draft quality mesh
                          Jerry Steiger

                          Chris,

                           

                          Sometimes hyperelastic models will have problems with elements having midside nodes and will solve properly with elements without midside nodes.

                           

                          Jerry S.