Machine Learning‐Guided Analysis of Rebound Mechanisms in Thermoplastic Vulcanizates via Automated Co‐Simulation

ABSTRACT Thermoplastic vulcanizates (TPVs) combine the resilience of elastomers with thermoplastic processability, yet their use in dynamic sealing is limited by inadequate rebound performance. This study integrates numerical simulation and data‐driven methods within an ISIGHT‐ABAQUS co‐simulation framework to investigate the governing mechanisms. A two‐dimensional stochastic multiparticle model was developed to represent the TPV microstructure, with parameters extracted via automated Abaqus macro modifications. This co‐simulation generated a database to address data scarcity for machine learning (ML) prediction. ML models were established and compared using XGBoost, GBM, and random forest, with the random forest model achieving minimal prediction error and an optimal efficiency‐accuracy balance at 800 samples. Interpretability techniques (feature importance, SHAP, and PDP) analyzed features governing residual deformation, tensile strength, and average rubber‐phase rebound stress. The results show that TPV rebound is predominantly determined by average plastic strain, rubber‐plastic mass ratio, and rubber‐phase rebound stress. This work provides a novel approach for studying key factors in TPV rebound mechanisms, aiming to develop refined strategies for performance regulation.