Mesh Refinement #7
Description
Hi Karthik,
Following up on our discussion today, I thought it might be helpful to start a discussion here on mesh refinement.
To summarize for others, we found that, for this particular device geometry and loading conditions, a relatively fine mesh resolution is required to achieve asymptotic convergence of our "quantity of interest" (i.e., strain amplitude). Here's a link to a figure summarizing our approach and mesh refinement results for the high-strain fatigue coupon at the most-extreme planned experimental conditions (1.5% strain amplitude):
https://github.com/kenaycock/Generic-IVC-Filter/blob/master/Fatigue_coupon/Rev1.2/02_high-strain_fatigue_coupon/02_high-strain_fatigue_coupon_FEA_mesh_refinement_submodel_approach.pdf
As requested, I extended the plot to include coarser mesh resolutions. Here are the results:
For C3D8R elements, the solution initially diverges, then begins to converge asymptotically after an effective mesh resolution of 64x64 elements across the strut cross section. The C3D8I elements begin to converge asymptotically at a more coarse mesh resolution (16x16 elements).
Using Richardson extrapolation theory (e.g., see ASME V&V-10, section 7.2), the predicted numerical error in strain amplitude for the finest C3D8I case (128x128) and values extrapolated to surface nodes is approximately 0.5% of the predicted value (% error, not % strain).
Let us know if you have any thoughts or insights. I think this problem illustrates the importance of carefully examining mesh convergence. But, the actual physical impact of the predicted peak strain amplitude on fatigue life is not clear (for reference, with the 128x128 mesh, the peak strain amplitude occurs in a 1 x 2.5 x 2.5 micron element...).
Regards,
Kenny