probably another dumb question. I have been playing around with different meshes with this set up. I have been experimenting with bias type meshing in cubit and wished to have a look at how that behaved. I have quite a course mesh with first order tets so probably not a particularly good mesh.
what intrigued me was that that this model converged so fast with out any other steps or tricks like soft springs. My question centres around why is this? have I missed something somewhere? I wish my other contact analysis ran so easily. Is there any chance the structure of the mesh has had a play in it? If so how do i make it happen like this for my next model!
Yes, the mesh pattern is a key factor here. Hence the OPs issue (this case runs with the mesh from old ppm version with proper Netgen refinement, but may fail with even better mesh generated by Gmsh). Such force-controlled, underconstrained contact problems can sometimes run and sometimes not, depending on whether the solver manages to establish contact preventing RBMs (and that can be highly mesh-dependent). To help the solver and make it more consistent, we can use various stabilization techniques.
Another thing is that contact may often converge better with first-order meshes.
fascinating - thank you for such a quick reply - i wondered myself about if it was a function of the elements being first order but have since looked at 2nd order elements and that converges fast and trouble free also.
Normally, it’s better to avoid such underconstrained models (even though the other directions are constrained by symmetry here) and use displacement control (usually the best option) or other aforementioned approaches. But since the solver sometimes manages to establish contact anyway (even though it initially happens at a single point), we can avoid these workarounds if we are lucky. In the Hertz contact problem discussed on the CalculiX forum, I ultimately resorted to displacement control.
