Physics-based machine learning in materials modeling and multiscale simulation
Abstract
The rise of Machine Learning (ML) has been continuously advancing the frontier of materials modeling and multiscale simulations. However, due to expense of physical or numerical experiments, material modelers & designers oftentimes cannot foresee all corner-cases that will arise in real environments, causing failure of most ML models that rely on data interpolation. In this talk, I will introduce our recent efforts on Deep Material Network (DMN), which presents a new way of embedding material physics into a network architecture. Its key features are the physics-based building block with interpretable fitting parameters, extrapolation capability for material and geometric nonlinearities with only linear elastic training data, and efficient online inference. An integration data-driven framework based on transfer learning and concurrent simulation in LS-DYNA will also be discussed.
