Experimental and triboinformatics approach for wear and friction prediction of AA7075/B ₄ C/TiB ₂ composites at elevated temperatures using machine learning techniques

Abstract This study presents a hybrid approach that combines traditional methods with data-driven Machine Learning (ML) approaches to predict the wear behavior and friction coefficient of AA7075/B ₄ C/TiB ₂ composites. The composites were fabricated using stir casting technique and a dataset comprising 1036 observations was gathered under dry sliding conditions at elevated temperatures utilizing a pin-on-disc tribometer. Experimental results reveal that the addition of B ₄ C/TiB ₂ reinforcements enhances wear resistance and reduces friction coefficient by promoting uniform microparticle distribution, the formation of mechanically mixed layers and strong interfacial bonding. In view of the prior studies, Non-linear parametric model Artificial Neural Network (ANN), Non- parametric model Decision Tree (DT), Instance based model KNeighbors Regressor (KNR) and Ensemble models Gradient Boosting Regressor (GBR), Random Forest (RF) were used to predict the wear and friction characteristics. Among the models, GBR yielded superior predictive accuracy for wear rate and friction coefficient. A detailed wear maps further aids in extensively understanding wear behavior.