| Abstract Scope |
I will present progress in predictive modeling of material under dynamical loading that combines physics-based simulations with machine learning. We use molecular dynamics (MD) to simulate shock loading of composite materials with complex microstructures. These simulations capture complex phenomena like the plasticity, the collapse of porosity, and interfacial friction and predict temperature and stress fields, including hotspots and stress concentration that govern the subsequent chemical and mechanical response. Unfortunately, these simulations are computationally too intensive to be used on-the-fly with continuum simulations capable of reaching macroscopic scales. To bridge this gap, we develop a machine learning (ML) model based on convolutional neural networks capable of predicting the shock-induced temperature and pressure fields with microstructure and shock strength as inputs. I will demonstrate the use of this model to provide sub-grid microstructural information to a continuum simulation that models shock loading, mechanics, chemistry, and thermal transport. |