Triboinformatic modeling of dry friction and wear of aluminum base alloys using machine learning algorithms

Data-driven methods including machine learning (ML) algorithms can yield a better understanding of how tribological and material properties correlate. Correlations of friction and wear of aluminum (Al) base alloys with their material properties (hardness, yield strength, tensile strength, ductility, silicon carbide content), processing procedure, heat treatment, and tribological test variables (sliding speed, sliding distance, and normal load) studied using traditional and data-driven approaches. Five different ML algorithms, K Nearest Neighbor (KNN), Support Vector Machine (SVM), Artificial Neural Network (ANN), Random Forest (RF), and Gradient Boosting Machine (GBM) applied to experimental tribological data to predict the coefficients of friction (COF) and wear rates. Through performance analysis, we demonstrated that the ML models can satisfactorily predict friction and wear of Al alloys from material and tribological test variables data. Comparative analysis of model performance illustrated that RF outperformed other ML models in wear rate prediction, while KNN exhibited the best performance in COF prediction. Feature importance analysis further revealed that normal load, hardness, and sliding speed have the maximum influence in predicting the wear rate of the alloys. The variation in hardness of the alloys and sliding distance influenced the COF prediction the most.

Automated summary

Data-driven methods, including machine learning algorithms, were used to study the friction and wear of aluminum alloys, demonstrating satisfactory predictive performance. Among the ML models tested, Random Forest performed best in predicting wear rates while K Nearest Neighbor exhibited the best performance in predicting coefficients of friction.