Page 1 of 1

Poor mesh convergence 3D linear elasticity

Posted: 25 Mar 2019, 13:59
by eriksson
Hi!
I started using Elmer a few weeks ago, and I'm trying to incorporate it in a parameter optimization program, to evaluate different designs of springs automatically. I use GMSH for meshing a STEP file from Autodesk Inventor, which then is converted to Elmer format with ElmerGrid, and finally ElmerSolver is used to solve a static linear elastic analysis with a force applied to this spring, to determine the spring's stiffness and stresses in it.

It seems, however, that my results conform poorly to FEM simulations in other software (namely the built in FEM functionality in Autodesk Inventor) with the same type of mesh, using similarly sized elements. I've also done experiments which show that the other FEM software are within 10 % from reality, stiffness-wise. Using a mesh size that I would find reasonable, my results from Elmer are around 5 times stiffer than reality and other FEM results. Refining the mesh shows rapid changes in results, but even with a ridiculously fine mesh I am around 30 % off from reality. A picture of different mesh sizes and the results is attached, and also a picture showing the deformation to illustrate what the simulation is doing.

I've tried using second order elements (adding Element = "p:2" to the equation section in the .sif), but the results don't get any better for the same computation time. Am I doing something stupid that is causing this? It feels like my simulation should be a fairly simple one. Any pointers or help would be greatly appreciated.

I wish you all a good day,
Felix Eriksson

Re: Poor mesh convergence 3D linear elasticity

Posted: 25 Mar 2019, 15:38
by eriksson
Hi again!
Alright, it seems as though I had not properly used 2nd order elements, as I previously stated. I had to make GMSH generate a mesh with 2nd order elements, which is reasonable in retrospect, but I got a bit confused by the fact that there is a keyword for setting 2nd order elements in the .sif-file. The results are much better now, as good as results from other FEM software when comparing to experimental data. I would therefore draw the conclusion that shear locking was causing my issues.

Then only one question remains in my mind - what does the 'Element = "p:2"'-keyword do?

Sunny greetings from Gothenburg,
Felix Eriksson

Re: Poor mesh convergence 3D linear elasticity

Posted: 25 Mar 2019, 17:07
by mzenker
Hi,

see Solver Manual, Appendix E: "Higher-order finite elements". ;)

HTH,
Matthias