Predicting ballistic impact behavior of silicon carbide/ultra-high molecular weight polyethylene composites through multi-scale modeling coupled with machine learning

This study presents a hierarchical multi-scale modeling framework to correlate the interfacial traction-separation responses in silicon carbide/ultra-high molecular weight polyethylene (SiC/UHMWPE) armored composites, advancing understanding of their ballistic performance. The novelty lies in coupling molecular dynamics (MD) simulations with machine learning in the form of a non-linear artificial neural network (ANN) to characterize the composite interfacial behavior and address strain-rate effects at the atomic scale. Additionally, the influence of polyurethane (PU) and epoxy (EP) resins on SiC/UHMWPE adhesion is evaluated for the first time through MD simulations. The ANN-predicted interfacial properties are implemented in finite element method (FEM) simulations of ballistic impacts using a cohesive zone model (CZM) framework. Validation against experimental data shows close alignment of interfacial strengths, residual velocity, and bulging depth, with a 7.1% difference in velocity-reduction relative to the corresponding experimental data. The results highlight the critical contribution of interfacial properties to penetration resistance and inter-layer delamination during impact. This integrated approach demonstrates an enhanced predictive capability for composite behavior under ballistic loading, thereby bridging the atomistic and continuum scales.