Nick N. Vlassis

Computational Mechanics & Artificial Intelligence Lab

Nick Napoleon Vlassis is an Assistant Professor in the Department of Mechanical & Aerospace Engineering at Rutgers University.

Google Scholar
News Feed

Sobolev Training of Thermodynamic-Informed Neural Networks for Interpretable Elasto-Plasticity Models with Level Set Hardening

Vlassis, Sun, CMAM, 2021.
We introduce a deep learning framework to train smoothed elastoplasticity models with interpretable components like stored elastic energy function, yield surface, and plastic flow that evolve based on deep neural network predictions. We use the Hamilton-Jacobi equation to deduce solutions for the hardening/softening mechanism and Sobolev training to ensure thermodynamic consistency and interpretability. Our numerical experiments using data from a 3D FFT solver show that this approach provides more accurate forward predictions of cyclic stress paths than other black-box models such as recurrent neural network, 1D convolutional neural network, and multi-step feed-forward model.

Read more →