Tribological properties of non-equiatomic FeCoCrNiMn high entropy alloys: Molecular dynamics simulations and machine learning predictions

FeCoCrNiMn high-entropy alloys (HEAs), owing to their exceptional structural properties, exhibit significant potential for tribological applications. In this study, a combined molecular dynamics (MD) and machine learning (ML) approach was used to investigate the tribological performance of non-equiatomic FeCoCrNiMn HEAs. First, 20 non-equiatomic FeCoCrNiMn HEA models with varying elemental concentrations (5 %, 15 %, 25 %, and 35 %) were built using MD simulations. Subsequently, nanoscratch simulations were conducted to determine the friction coefficients and wear atoms, while dislocation evolution analysis was performed to elucidate the underlying deformation mechanisms. Four ML algorithms—K-nearest neighbors (KNN), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost)—were used to predict tribological performance, and the optimal model was selected through comparative analysis. Finally, the Shapley additive explanations (SHAP) study was conducted to determine the significance of elemental features. The results indicate that increasing Co and Ni content significantly improves the anti-friction and wear-resistant properties of FeCoCrNiMn HEAs. In contrast, higher concentrations of Mn, Fe, and Cr exacerbate friction and wear. Moreover, XGBoost exhibited the best predictive performance for friction coefficients, with an R² of 0.86. The KNN algorithm outperformed other algorithms in predicting wear atoms, with an R² of 0.96. SHAP analysis revealed that Co, Ni, and Mn have a significant impact on the friction coefficient and wear atoms of HEAs, followed by Fe, with Cr having the least impact.