Development of a supervised machine learning model for predicting the friction coefficient of tribological pairs

Machine learning (ML) has become increasingly prevalent in various research and industrial fields. This study applies eight different ML models to predict the friction coefficient (COF) in rotationally tribological pairs and compares the results. Experimental friction data under different operating conditions between high-speed steel samples (W9Mo3Cr4V) and aluminium brass samples (HAl61-4-3-1) is used to train ML models developed using Python programming. ML models namely, K Neighbors Regression (KNN), Gradient Boosting Regression (GBR), AdaBoost Regression (ABR), Linear Regression (LR), Random Forest Regression (RFR), Extra Trees Regression (ETR), Bagging Regression (BR) and Decision Tree Regression (DTR) are established for predicting the COF and the results are evaluated and compared. The proposed model uses Load force, Rotate speed, Radius and Test duration as the input parameters. The prediction results are assessed by the Root Mean Square Error (RMSE) and Coefficient of Determination (R2) indicators. The results show that the ETR performs well in predicting COF. The number of features influences the accuracy of the model, and in the ETR model, the R2for COF with four features and three features show minor variations (both above 0.9). This shows that three features can also effectively predict COF.