Mesh convergence problem

Project Support
1

Hello, everyone!

For the past few weeks, I have been trying to conduct a mesh convergence study for a static structural analysis, but I have been facing some problems. The simulation link is here: https://www.simscale.com/workbench/?pid=2633001207437297647

What happens is that my mesh does not seem to be converging on single value. Initially, I was trying to refine the mesh for the whole model, but it did not work. I thought that maybe this result was because the greatest tension gradient is located at a stress concentration spot. Then, I proceeded to refine my mesh just in this location, using the Region Refinement tool. However, the results that I have so far, do not seem to converge as well. I am using the strain value to acess the result, as this is the most important parameter in this case. The results I have so far are (element number / region refinement max. edge lenght / max. strain):

296k - 0.0002 - 2.387e-2
657.9k - 0.0005 - 2.866e-2
730.7k - 0.00045 - 2.865e-2
867.6k - 0.0004 - 3.05e-2
1.1M - 0.00035 - 3.165e-2

Could I have some guidance here? I am confused if I am doing some pre-processing step wrong or if my mesh simply just did not converge yet. I think the my case is not very complex, so I would expect the results to converge with less than or a little more than one million elements. The material I am using is ABS, which with that amount of strain would not behave as a linear material, and maybe that is the reason for the divergence, but I am not sure if this would have some impact in the values, once I stated for the software that it is a linear case.

I would really appreciate a help! Thank you so much!

2

Hi @srocketry,

Thanks for posting your question!

Yes, you are right. However, as you’re facing to with this singularity, the more the mesh is refined, the more the stress continues to increase and tend towards infinity. Here I see that the main thing is that you already have reached mesh convergency:

This is due to geometry and loading/boundary conditions creating artificial stress results, which are inaccurate. Got it, but how can you avoid that? From my side the response is quite simple, above you see that the solution does not change substantially with subsequent mesh refinements from 657.9k to 730.7k, then you can assume that the result is enough and use that for an inverse parameter identification study. In this case just get back the refinement and then take a proof that as low quality mesh is, it is far to the final value of 2.86e-2.

Please let me know if that helped, and also let me know if the project runned well with this inverse parametric identification method that I have mentioned.

I also would like to recommend this page as a theoretical guide in your study: What is Convergence in Finite Element Analysis? SimScale

best,