Towards Intelligent Tribology: Predicting Wear and Friction in WC–Co Coatings with Machine Learning

Abstract

Abstract The present study aims to build Machine Learning (ML) models to predict the wear and coefficient of friction (CoF) for WC–Co-coated mild steel substrates. Tribological runs revealed that wear increased with sliding distance and load, ranging from 35.04 to 61.38 µm, while CoF varied from 0.0520 to 0.1795 under 20 N and remained more stable (0.1070 to 0.1186) under 30 N, indicating better frictional consistency at higher loads. Gaussian Process Regression (GPR) and Support Vector Regression (SVR) models were implemented on the experimental results and the evaluation metrics of wear prediction for both GPR (Training R 2 = 0.9999, Testing R 2 = 0.9998, RMSE = 0.0884, MAPE = 0.11) and SVR (Training R 2 = 0.9995, Testing R 2 = 0.9999, RMSE = 0.0616, MAPE = 0.10) were impressive with SVR displaying marginally more accuracy. In case of CoF, once again both GPR (Training R 2 = 1.0, Testing R 2 = 0.9999, RMSE = 0.0003, MAPE = 0.23) and SVR (Training R 2 = 0.9999, Testing R 2 = 0.9998, RMSE = 0.0004, MAPE = 0.27) performed well with GPR exhibiting marginally more accuracy. Confirmation experiments also validated that SVR best predicts wear, while GPR excels in CoF prediction. The findings prove that ML can reduce tedious experimental trials, enable optimum material selection, and optimize performance for industrial applications such as aerospace, automotive, and manufacturing. These results establish ML as a reliable tool for wear and CoF predictions and pave way towards data-driven intelligent tribology.