Machine Learning Prediction of Tribological Properties of Graphene-Reinforced Perfluoroelastomer Composites Based on Molecular Dynamics Simulations and Experimental Research

This study aims to develop high-performance polymer nanocomposites based on perfluoroelastomers (FFKM) using an integrated framework that combines molecular dynamics (MD) simulations, machine learning (ML), and experiments to investigate the factors governing tribological behavior. The friction coefficient shows negative correlations with temperature, normal load, and sliding velocity. Increasing the sliding velocity raises the average diffusion coefficient from 0.00422 to 0.04300. Increasing the normal load increases the transfer film thickness by 39.13% and reduces surface roughness by 36.19%. The SDM model achieves the highest prediction accuracy (R 2 = 0.9662), outperforming conventional algorithms. Among six friction test conditions, the minimum friction coefficient is 1.339. Two conditions with identical PV values of 0.1 exhibit similar friction coefficients (1.604 and 1.646). The experimental trends are consistent with the MD and ML results. Highlights A comprehensive research framework of MD simulation, ML, and experiments. The COF of FFKM is negatively correlated with multiple factors. A novel ensemble algorithm, SDM, is developed for friction prediction. Experimental results show clear and consistent trend indications.