Applicability of machine learning on predicting the mechanochemical wear of the borosilicate and phosphate glass

Abstract

Machine learning (ML) has become powerful tools for predicting wear properties of solid materials. Numerous studies have elucidated the prediction of mechanical wear of materials via various ML algorithms, but the prediction of mechanochemical wear of materials involving both mechanical and chemical effects via ML has not been well-addressed yet. In this study, both the mechanical effects (contact pressure, sliding speed, and cycles) and chemical effects (RH) on the experimentally wear volume of borosilicate and phosphate glass were considered as dataset for ML, and various ML algorithms are used to predict the mechanochemical wear properties of the two glasses involving the linear regression (LR), support vector machine regression (SVM), Gaussian process regression (GPR), K-nearest neighbor (KNN) and artificial neural network (ANN). The order of prediction accuracy of the five ML algorithms is as follows: ANN, KNN, GPR, SVM and LR. Further studies reveal that the prediction accuracy of ANN algorithm can be related to nature of repeated training of the input data and the existence of hidden layers. The influence of the inherent characteristics of the two glasses on the prediction accuracy under the synergistic effect of mechanochemistry is also discussed. The obtained results may be helpful for developing glass materials with controlled wear properties, thereby accelerating the development of novel functional glasses.