Physics-conforming material models with artificial neural networks


Our group's latest article on "Polyconvex anisotropic hyperelasticity with neural networks" was just published in the reputable Journal of the Mechanics and Physics of Solids (JMPS).

This work introduces the first data-driven constitutive models for finite deformation hyperelasticity that fully consider all physical requirements from the continuum mechanics theory such as energy conservation, objectivity, material symmetry, ellipticity, and growth conditions directly in the model formulation, here using input-convex with neural networks. Congratulations to our doctoral student Dominik Klein, who lead the research in cooperation with our former postdoc Dr. Mauricio Fernández, as well as Dr. Robert Martin and Prof. Patrizio Neff from Duisburg-Essen University!

The article is freely available within the next 50 days at Otherwise, the accepted version can also be found on arXiv: