Coupled modeling of aluminum agglomeration and combustion in composite propellants

This study develops a coupled Material Point Method–Lattice Boltzmann Method–Neural Network (MPM–LBM–NN) framework to model aluminum agglomeration and combustion in multi-component composite propellants. The method incorporates binder decomposition, molten aluminum coalescence under surface tension, and detailed gas-phase chemistry accelerated by a neural-network-based chemical solver. Micro-CT imaging provides realistic initial particle packing, while digital holography validates agglomerate size distributions during combustion. Simulations reproduce the observed bimodal agglomerate size spectra and predict burning rates across a wide range of aluminum particle sizes, AP gradations, and operating pressures with mean errors below 15%. Parametric analyses reveal that increasing HMX particle size and reducing aluminum particle size suppress agglomeration, while increased coarse AP fractions amplify pressure sensitivity and bimodal aluminum distributions moderate burning-rate pressure exponents. These findings quantitatively link propellant microstructure to combustion response, offering guidance for optimizing formulations and improving predictive combustion models.Novelty and significance statementThis study presents a fully coupled Material Point Method–Lattice Boltzmann Method–Neural Network (MPM–LBM–NN) framework for modeling the combustion of multi-component composite propellants. The model unifies condensed-phase regression, gas-phase reactive flow, and detailed chemical kinetics within a single physics-based formulation, enabling direct prediction of burning rates, temperature fields, and pressure-coupled responses. Unlike traditional models relying on empirical correlations, this approach resolves binder pyrolysis, oxidizer decomposition, and phase-change heat transfer with dynamic gas–solid coupling. Micro-CT-derived microstructures and digital holography measurements are incorporated for multiscale validation, ensuring consistency between simulated and experimental combustion behavior. The neural-network-based chemical reactor accelerates large kinetic mechanisms while maintaining accuracy, bridging the gap between detailed reaction chemistry and multidimensional flow simulation. This framework provides a predictive tool for linking microstructural features to macroscopic performance, advancing the design, optimization, and stability assessment of next-generation high-performance solid propellants.