Integrated approach of coarse-grained molecular dynamics calculations and machine learning for understanding mechanical properties of filler-filled polymer models

Rubber materials are functional polymers used in a wide range of applications, yet simultaneously controlling multiple properties presents significant challenges. In this study, we applied a combined approach of coarse- grained molecular dynamics (MD) simulations and machine learning to comprehensively analyzed the structural features that influence the mechanical properties of a filler-filled crosslinked polymer model. Through MD simulations, we observed that both polymer structure and filler morphology varied with filler loading conditions and polymer-filler interactions. A regression model was developed using explanatory variables related to these morphological features, with the stress values at different stages of uniaxial extension simulations as the objective variable. Our results indicated that the presence of voids significantly contributed to the stress. To explore this further, we performed a regression analysis using void volume as the objective variable, identifying polymer density, crosslink distribution, and polymer-filler interactions as controlling factors for void formation. Shapley additive explanations were employed to elucidate the contribution of these factors to mechanical properties, revealing their mutual dependence and non-linear effects. This integrated approach not only provides valuable insights into optimizing polymer composites for enhanced mechanical performance but also establishes a novel predictive framework that significantly advances the efficient design of high-performance polymer materials.