Coupling molecular dynamics–derived energy fingerprints with random forest to reveal wear mechanisms in functionalized nanocomposites

This study constructs four functionalized graphene/nitrile-butadiene rubber (FGNS/NBR) molecular dynamics (MD) models under the COMPASS force field to evaluate mechanical properties and abrasion rates. For each composite, 50 independent shear simulations (200 samples) were run; averages of force-field energy components over the post–steady-state segment built a bonded/non-bonded energy-fingerprint library. Feature screening relied on the consensus across multiple importance measures, yielding 17 descriptors. A Random Forest regressor (RFR) was trained, and mechanistic interpretation combined SHapley Additive explanations with mutual information. The MD analyses indicate that functionalization generally enhances mechanical properties and improves wear resistance. Subsequently, the RFR explanatory model further revealed three principal determinants of wear performance—Fe layer van der Waals energy, total bond length after shearing, and the torsion–stretch energy of FGNS—and indicated that the interplay between chain extension and interfacial adsorption constitutes the dominant anti-wear pathway. Overall, the study introduces an interpretable comparative method focused on relative trends across functionalizations, offering a useful analytic perspective for small-difference systems and for considering functional group and grafting choices.