Prediction of the tribological properties of a polymer surface in a wide temperature range using machine learning algorithm based on friction noise

Abstract

A signal processing method based on friction noise is introduced to predict the tribological properties of polymers in a wide temperature domain. Three different machine learning algorithms, XGBoost, LightGBM and CatBoost are used to establish a mapping relationship between the time–frequency domain features of friction noise and the friction coefficient. The friction coefficients of pairs of four polymers and seven metals are predicted at five temperatures and three load/speed conditions of point contacts in a wide temperature domain from − 120–25 ℃. Performance analysis reveals that the machine learning approach satisfactorily predicts the friction coefficients of different polymer–metal pairs in a wide temperature range from tribological test data. This method can be used for in-situ monitoring on tribological properties.