3 Replies Latest reply on Nov 25, 2010 12:00 AM by Anthony Botting

    Meshing Dilemna

    Sean Leo

      I have been working on a linear static simulation for a few weeks now. This whole time I have been using draft quality elements so that I could get an idea of areas that needed refinement without wasting too much time. I recently ran a simulation with draft quality curvature elements, try not to laugh but here are the specs of the mesh (439695 DOF's, 146726 Nodes, 853279 Elements). FYI I am already taking advantage of symmetry, my model is acutally a quarter size of the actual. The simulation took 2 days to run and returned pretty good results.


      I know the answer to this question is of course yes but here goes anyway. Should I now re-run the simulation with high quality elements? You can already see the massive size of my mesh and that it took 2 days with draft quality elements, I cannot fathom how large the mesh would be and how long it would take if high quality elements were used or if my computer could even handle it. My computer specs are: Intel Pentium 4 HT 3.00GHz, 2 GB of RAM, ATI Radeon HD 4300/4500 Series.


      Here are my questions:


      1) What is better a draft quality mesh with very good refinement in areas of interest or a high quality mesh with much less refinement?

      2) What are your opinions of using h-adaptive and p-adaptive?

      3) Curvature elements take longer to solve right? Should I maybe just go back to standard elements?

      4) What are your computer specs and how fast do you think you could solve a problem like mine?

        • Re: Meshing Dilemna
          Stuart Moore

          I can't really answer your questions but your computer specs are really poor.  The processor is several generations of architecture out-of-date, you don't have enough RAM and your graphics card doesn't look like a supported model.  This is not a good SolidWorks machine.


          A general rule of thumb with meshing is that you should refine the mesh until it makes no difference to the result - not very practical in your present situation.  SW 2011 has a nice feature whereby you can set a minimum and maximum mesh size and the software will vary the size it uses (between those limits) depending on the geometry.  A rather nice enhancement which will be useful in future.  

          • Re: Meshing Dilemna
            Ryan Werner

            Hi Sean,


            Here are my two cents:


            1) Since it took 2 days to solve this problem I am not sure I would rerun it if you are satisfied with the results you have already gotten.  Do these results seem realistic?  Can you compare them to some previous results or hand calcs to make sure they are in the ball park?  In the end it is all about the results, if you are happy with them and think they are accurate than maybe you don't need to redo it.


            2) I actually never really use the h and p adaptive features.  That is not to say they are not useful it is just that all the analysis I run are on larger, multibody models and assemblies so they just wouldn't really help me out all that much.


            3) Curvature elements are great but can increase solve time so I would stick to the standard elements unless I specifically needed to make a change due to mesh failure, geometry, etc.   


            4) My specs are a 64-bit machine with an Intel(R) Core(TM)2 Duo CPU T9550 @ 2.66 GHz with 8 GB of RAM.  It is by no means a monster but it runs SolidWorks and Simualtion pretty well.  Not having seen you actual model or how your analysis is set up it is a little hard to estimate how long it might take to solve on my machine but since your DOF's are under 500,000 I would think I could solve it in under an hour, maybe even as quickly as 10 mins.  If nothing like No Penetration contacts or Connectros and the like are present and it is just a Curvature Based Mesh with that number of DOF's than it shouldn't take too long.


            Ryan W.

            • Re: Meshing Dilemna
              Anthony Botting

              1) The draft quality element mesh at a higher density than a high quality element mesh, at a lower mesh density should theoretically give similar answers at some limit of mesh density, which is problem-dependent.

              2) H-Adaptive is nice to have (but I believe it only works on static, linear analyses) because there is a "coarsening" switch in the properties panel that allows for a reduction in the mesh density in areas of low gradients. The effect is to "optimize" the mesh so that more elements are present in higher gradient areas, and fewer elements exist in lower gradient areas. P-Adaptive is nice, too, because it increases the order of the displacement functional in areas of high gradient -effectively similar to H-Adaptive. However, the stiffness matrix in P-Adaptive is more complex than that for H-Adaptive (mathematically more difficult to solve for the computer), so typically you might run both approaches and see which uses less resources (again, problem-dependent).

              3) The "curvature elements" you are referring to: I believe you mean employment of the curvature-based meshing algorithm. The curvature-based meshing algorithm is just an approach to meshing that looks at the curvature of local geometry and, if the radius is small, for example, you get more elements there, and less elements where radius values are large. This has its drawbacks. I recently ran into a problem where two parts meshed but failed to contact correctly. One part had high curvature, so a lot of elements where placed there, and the contacting part was a rectangular shape, so very few elements were placed there - so, the upshot is that the curvature-based mesher created an incompatible mesh (the analysis failed because the parts separated). Large elements existed on the rectangular part, and small elements were placed on the highly curved part, so the node positions did not match and the mesh was incompatible (we used a switch in the software to write compatibility equations to get around the limitation).  The bottom line is the curvature-based mesher does not use different elements than the "standard" mesher - they both use the same element types (depending on your choice of High or Draft Quality elements). We alternately resorted to the standard mesher algorithm for this case because it resulted in a compatible mesh and contact between parts was successful. The question concerning longer or shorter solve times is really related to the number of elements and the type of element -draft or high quality.

              4) The computer you have seems a bit low on resources. If you could get a 64-bit machine and load it up with 8 GB RAM and perhaps a quad-core ( I read about AMD's new 6-core which is now available!), depending on the problem type, multiple cores can benefit a great deal. I have run a 1/2 million DOF's (near your number) in a few minutes on a dual-core, 64-bit with 4GB RAM but it was a natural frequency extraction problem, and again with a linear static problem, so - it is really problem-dependent. Nonlinear problems in particular can take a while, and if you're requesting results at each time step, for stresses, strains, displacements, at more than one location on the model - well, I hope you get the picture (for that case you need a very high speed hard drive to write the results at each time step). It is substantially problem-dependent.