New article on inelastic homogenization through machine learning

I would like to share another article from my group just published in CMAME. I think this is an interesting approach towards the automatic generation of constitutive laws for arbitrary microstructures:

https://authors.elsevier.com/a/1bxlSAQEIt1ls

The article focuses on how to systematically create constitutive laws using only:

1) Available microstructural models

2) Machine learning techniques

The approach is very generic, as it does not make any assumptions about the material response. It only needs input-output histories of strains and stresses. I think the most interesting interesting contribution of this work comes from using Gaussian Processes in the invariants of the strain tensor to dictate the strain histories for the training set. We were then able to incorporate the resulting constitutive law into a traditional FEM solver just by changing one line of code, the call to the constitutive model.