Prediction performance of tribological behaviour of steel-glass reinforced polymer composites using machine learning models

This study examines the comparative effects of key operating variables (applied load, sliding velocity and sliding distance) on the tribo-performance of glass fiber-reinforced epoxy composites with and without steel wire mesh. A computational and data-driven machine learning (ML) approach has been used in this work to analyse and predict the performance output in a dry sliding wear mode. Predictive ML models have been proposed to estimate specific wear rates (SWR) under different test conditions. Four ML techniques such as decision tree (DT), random forest (RF), gradient boosting machine (GBM) and extreme gradient boosting (XGB) are considered. These trained models effectively predict tribo-performance of the fabricated composites and the feature score analysis indicates that variations in sliding velocity and normal load significantly impact SWR. Among all the ML models, XGB demonstrated the highest accuracy, with coefficient of determination (R 2 ) values of 0.91522 for composite G11 (11 layers of glass fiber), 0.97802 for composite G7S4 (seven layers of glass fiber and four layers of steel wire mesh), and 0.97803 for composite G4S7 (four layers of glass fiber and seven layers of steel wire mesh). The mean absolute error percentage is about 3%, confirming the prediction accuracy and reliability of the model. Major wear mechanisms are also identified using scanning electron microscopy.