Enhancement of fracture properties of amorphous polymers by nanoparticles: A machine-learning assisted coarse-grained model

Abstract

Polymer nanocomposites, formed by incorporating nanoparticles into epoxy matrices, exhibit exceptional thermo-mechanical and fracture properties, making them ideal for advanced engineering applications. This study explores the enhancement of fracture properties of epoxies by nanoparticles and develops a coarsegrained (CG) model to enable this investigation. We present a novel artificial neural network (ANN)-assisted optimization framework to calibrate CG molecular simulation models. The algorithm integrates particle swarm optimization with ANN predictions, where ANN accelerates parameter optimization by minimizing errors between CG simulation results and all-atom reference data. This process significantly reduces computational cost while ensuring accurate predictions of critical properties, such as yield stress and elastic modulus, over a wide temperature range, demonstrating excellent temperature transferability of the model. Large-scale CG simulations facilitated the analysis of nanoparticle agglomeration effects on fracture behavior, a challenge infeasible for all-atom simulations. Simulation outcomes were qualitatively compared with experimental findings, offering valuable insights into the influence of nanoparticle distribution on fracture properties. This integrated approach provides a robust pathway for designing and optimizing polymer nanocomposites for real-world applications.