Machine Learning Potential-Enabled Atomistic Elucidation of Interfacial Mechanics in Energetic Composites: A Case Study on NEPE Propellant

This is the author-accepted manuscript of an article accepted for publication in Energetic Materials Frontiers. Interfacial adhesion and failure between polymeric binders and rigid particulate fillers play a decisive role in the mechanical reliability of high solid loading polymer composites. Nitrate ester plasticized polyether (NEPE) propellants represent a chemically complex class of particle-filled energetic composites, in which a polyurethane-based binder interacts with multiple fillers, including Al/Al2O3, AP, and HMX. Despite their technological importance, the atomic-scale origins of binder-filler adhesion, debonding, and failure-mode transition remain insufficiently understood because conventional fixed-topology force fields cannot describe bond rearrangement, whereas conventional reactive simulations are computationally prohibitive for multicomponent interfaces of this complexity. Here, we develop a machine learning potential trained on ab initio calculation data and apply it to reactive molecular dynamics simulations of binder-filler-binder sandwich structures under different constraint configurations, strain rates, and temperatures. The simulations quantify a distinct hierarchy of interfacial adhesion, with work of adhesion values of 1127.6, 275.0, and 144.2 mJ·m-2 for the NEPE-Al/Al2O3, NEPE-AP, and NEPE-HMX interfaces, respectively, identifying the HMX-containing interface as the most susceptible site for interfacial debonding and crack initiation. A comparison of boundary constraints is further introduced to assess the contribution of bulk matrix deformation to the interfacial response, enabling direct comparison of filler-specific interfaces within a unified reactive simulation framework. Atomistic analyses reveal that the strong NEPE-Al/Al2O3 interface originates from strong attractive interactions at the alumina surface, whereas the weaker NEPE-HMX interface is dominated by physisorption. In addition, strain rate and temperature regulate polymer-chain relaxation, microvoid nucleation, and energy dissipation, thereby governing transitions between interfacial debonding and matrix-dominated failure. These results establish quantitative structure-interaction-mechanics relationships for complex binder-filler interfaces and provide atomistic reference parameters for subsequent multiscale calibration and interface optimization of particle-filled polymer composites.