Critical Factors Governing the Frictional Coefficient in Mg Alloys—Learn From Machine Learning

Abstract

Data-driven methods are emerging as a promising approach in discovering the correlation between tribological properties, composition, and mechanical properties of engineering materials. In the present study, the capability of several ML models in predicting the coefficient of friction (COF) of magnesium alloys is studied. To this end, first 1400 data points are extracted from prior studies through an extensive literature review. The collected data is then used to train models for the following two scenarios: (i) COF prediction using composition, processing parameters, and tribological variables; (ii) COF prediction using mechanical properties (hardness, yield strength, ultimate tensile strength, ductility, and elastic modulus), and tribological variables. After preprocessing, the data is partitioned into train and test datasets where the train dataset is used for model training and hyperparameter tuning, K-fold cross-validation, and the test dataset is used for evaluating the best trained models. The results indicate that light gradient boosting (LGBM) accurately predicts COF of magnesium alloys using the processing procedure, heat treatment, alloy composition, and tribology variables with an R-squared value of 0.89. Further, the gradient boosting method (GBM) achieves an R-squared score of 0.87 for predicting the COF using mechanical properties and tribological variables, showing a promising performance. In addition, a comparative analysis between alloying elements, manufacturing process, heat treatment, mechanical properties, and tribological test variables is performed using feature importance in the trained random forest (RF) models. Our findings highlight the importance of normal load, elastic modulus, and content of Zn in determining the COF in magnesium alloys, which helps improve materials and mechanical system design for effective COF control.