A physics-informed neural network-enhanced material point method for regression and heat transfer modeling of solid propellant

In this study, we introduce a Material Point Method (MPM) for simulating the regression process of solid composite propellants. By employing a physics-informed neural network for solving gas-phase chemical reactions combined with a novel gas–solid coupling approach, our method accurately models the propellant burning rate while capturing complex solid-phase interface morphologies under various pressures. Traditional MPM, typically employed for large deformation simulations, is enhanced by our heuristic, data-driven approach, enabling predictive combustion modeling. Simulations of pure AP, AP/HTPB sandwich configurations, AP/HTPB-packed propellants, and AP-packed propellants with different AP size distributions revealed non-steady state burning rates with inherent oscillations. Our results showed <10% error below 4 MPa and <20% error between 4–7 MPa compared to experimental data. Fine thermocouple measurements of surface temperatures showed ≤15% deviation from experimental results, thereby validating the model's predictive capability. The method's multi-physics tracking capability enables accurate simulation of complex interface morphologies, including low-pressure adhesive layer depressions and high-pressure protrusions in sandwich propellants, as well as subsurface structures in AP spherical packed configurations. This research provides a new method for predicting the burning rate and interface morphology of composite propellant combustion, with future work aimed at refining energy balance algorithms and parameter settings based on experimental insights.