Machine Learning Framework for Evaluating Microscale Interface Cohesive Zone Models of Polymer Composites.

Microscale interface cohesive zone models (CZM) calculated by molecular dynamics (MD) are a critical component of multiscale simulations in polymer composites. However, conventional MD methods remain constrained by prohibitive computational costs and inherent scalability limitations. This study proposes a machine learning (ML)- MD analysis framework to predict CZM parameters of a microscale interface in polymer composites. This framework proposed an innovative decoupling simulation method that decoupled the complex interfacial interactions (multiatom interactions) into fundamental interfacial interactions (single-atom interactions); key factors influencing the CZM parameters are identified: van der Waals interactions and hydrogen bonds. The CZM parameters predicted by this framework were validated against experimental results and literature data and successfully applied to finite element simulations of fiber pull-out behavior. This framework establishes a scalable methodology for preliminary interface assessment that can be used as input for microscale interface properties in multiscale analysis, directly reducing the reliance on MD workflows.