Multiple FEM analyses using open‑source tools...

mjb

Dear Colleagues,
a few months ago, I faced the task of performing several thousand calculations using the finite element method. For this purpose, I chose well-known open-source applications: Salome Platform, Gmsh, Code_Aster, and ParaView. In addition, I took advantage of Python and the Spyder environment. If anyone is interested in how to perform, for example, 528 calculations as a one process of a simple notched specimen with variable geometry — diameter and notch radius — you’re welcome to visit my GitHub: https://github.com/ennamepro/fem_mcip. If you have any questions, feel free to get in touch.
Regards,
MJB

mjb

Dear Colleagues,
What can be done with approximately 9,111,300 (more than nine million) data points following a FEM calculation? How should one verify and evaluate them?
Machine Learning (ML) tools are exceptionally versatile for this purpose. Another project on GitHub - https://github.com/ennamepro/elasf_fc - demonstrates how to prepare data FEM (stress state triaxiality) for verification and evaluation.
If you have any questions, feel free to get in touch.
Regards,
MJB

mf

mjb Hello,

thanks, this is all super interesting and I will definitely look at this in a more detailed fashion. Having worked a long time ago in fracture mech I know this is quite a hassle. So, my question is, why did you do this when analytical formulae for such specimens can be found in literature like Murakami (I haven't checked but I am quite sure)? Or is this in regard to uncracked specimens (blunt notches)? Also, I hope you are aware of the mesh dependency of your results?

Mario.

mjb

Hi Mario,
Thank you for your interest and for raising these very pertinent questions.
You are absolutely right regarding the literature. I am well acquainted with the classic and fundamental works of Prof. Murakami, Peterson’s stress concentration factors, and Roark’s formulas.
The resources I published on GitHub are actually just a small fraction of a much larger body of work. I intentionally chose these standard notched specimens (and yes, these are uncracked/blunt notches) for this public demonstration precisely because they can be quickly verified and benchmarked against the established analytical data found in the literature you mentioned. It served as a perfect validation case to prove that the automated script pipeline and the Machine Learning models are working correctly.
The real value and ultimate goal of this methodology (combining parametric FEM with ML) lies in its scalability. Once the framework is rigorously validated against these classic analytical cases, the exact same workflow can be applied to any arbitrary structural components and complex geometries for which no analytical formulas or literature data exist.
Regarding mesh dependency: yes, I am fully aware of this crucial aspect, as the mesh effect is highly significant in these analyses. However, in this methodology, determining the scalar field of the stress triaxiality and subsequently calculating the statistical measures of its distribution allows us to homogenise the obtained results, making them much more robust.
Thanks again for your insightful comments.
Regards,
MJB