As part of the InSilicoHealth MSCA Doctoral Network, we recently held another Methods Club session, this time focusing on inverse finite element (FE) optimisation for identifying material parameters.
A big thank you to Rémy Benais for leading the session and sharing his expertise!
The presentation introduced the basics of inverse FE optimisation, an approach where model parameters are iteratively adjusted so that simulation results match experimental data. One interesting point that came up in the discussion was that, in practice, researchers usually fit their models to the full experimental time series, not just the final result. This helps capture material behaviour much more accurately.
We also explored a few alternative methods, such as the Virtual Fields Method, Euclid, and Constitutive Artificial Neural Networks. While they follow a similar idea of iteratively improving the model, they use different strategies and can sometimes be more efficient depending on the case.
Another engaging part of the discussion focused on the challenge of finding the right solution. In complex (non-convex) problems, it’s not always clear whether you’ve reached a global optimum or just a local one. Participants shared useful tips, like setting appropriate parameter bounds and choosing good initial guesses to improve results.
Finally, we touched on validation: how to make sure that the identified parameters actually reflect real material behaviour? This remains a crucial step in making these models reliable and useful.
Thanks again to Rémy and to everyone who joined and contributed to the discussion. The Methods Club continues to be a great space for sharing knowledge, exchanging ideas, and learning from each other!