Physics-informed data-driven cavitation model for a specific Mie–Grüneisen equation of state

Abstract

Unsteady cavitation, as observed in phenomena like underwater explosions, entails dynamically evolving boundaries and developing dimensions of cavitation before collapse. The classic one-fluid models often fail to accurately simulate this, as they either rely solely on physics or experimental data. In this study, we introduce a novel one-fluid cavitation model tailored for a specific Mie-Grüneisen equation of state (EOS) known as polynomial EOS, employing an artificial neural network. Our approach integrates physics-informed equations with experimental data through an optimization problem. This physics-informed data-driven model accurately predicts pressure within the cavitation region, where pressure tends towards zero along with density. We apply this model to complex compressible multi-phase flow simulations, including nuclear and underwater explosions. Validation against experimental data and comparison with existing models precedes its application in one- and two-dimensional cases. The simulation of three-dimensional cavitation phenomena near submarines during underwater explosions yields a high-quality numerical solution. The outcome underscores the significant engineering value of the novel cavitation model.