Finite strain FE2 analysis with data-driven homogenization using deep neural networks

A data-driven deep neural network (DNN) based approach is presented to accelerate FE2 analysis. It computationally expensive perform multiscale analysis since at each macroscopic integration point an independent microscopic finite element needed. To alleviate this computational burden, DNN surrogates are proposed for nonlinear homogenization that can serve as effective macroscale material models. probabilistic considered surrogates’ development, and efficient data sampling strategy from the deformation space used generating training validation datasets. Frame indifference of behavior consistently handled, two methods – regular where only input/output pairs included in dataset via L2 loss function, Sobolev derivative also with function compared. Numerical results demonstrate leads a higher testing accuracy compared training, DNNs accurate