Geometric learning for computational mechanics Part II: Graph embedding for interpretable multiscale plasticity

The history-dependent behaviors of classical plasticity models are often driven by internal variables evolved according to phenomenological laws. difficulty interpret how these represent a history deformation, the lack direct measurement for calibration and validation, weak physical underpinning those laws have long been criticized as barriers creating realistic models. In this work, geometric machine learning on graph data (e.g. finite element solutions) is used means establish connection between nonlinear dimensional reduction techniques Geometric learning-based encoding graphs allows embedding rich time-history onto low-dimensional Euclidean space such that evolution plastic deformation can be predicted in embedded feature space. A corresponding decoder then convert back into weighted dominating topological features observed analyzed.