Scientific Machine Learning for Prediction of Thermodynamic Behavior and Shock Propagation

Scientific machine learning (Sci-ML) combines the data-driven approach of traditional machine learning (ML) techniques with foundational mathematical principles governing processes such as chemical reactions, fluid dynamics, and structural mechanics. Within the realm of fluid dynamics, Sci-ML has shown promise in augmenting and accelerating well-established computational modeling-based paradigms. However, there exists relatively few applications of Sci-ML to detonation science and shock propagation problems. Here, we present two sets of results leveraging Sci-ML to both aid, and replace, aspects of traditional hydrocode-based modeling of detonation-induced shock propagation. First, we train a neural network to represent the thermodynamic characteristics of a detonating common explosive. This ML-based equation-of-state uses prior knowledge of existing empirical EOS models while offering a more flexible and robust functional form. Second, we demonstrate the application of physics- informed neural networks (PINN) to approximate the Euler equations governing shock propagation. These PINN models are shown to produce high-fidelity solutions to the fluid field while significantly reducing compute time with respect to traditional hydrocode counterparts. Advantages, challenges, and future progress of the preliminary results of the Sci-ML approach to detonation and shock modeling will be discussed.