Comparison between physical and machine learning modeling to predict fretting wear volume

The objective of this study is to compare the performance of machine-learning strategy versus a physical friction-energy wear approach to predict the fretting wear volume of a low-alloyed steel contact by varying several loading parameters. Then, an artificial neural network (ANN) is used to predict the wear volume at each loading condition. These predictions were compared versus a physics-based friction energy wear modeling considering the third-body theory and the contact-oxygenation concept. A parametric study is performed to compare the prediction errors as a function of the proportion of the experiments involved in the modeling process. The results suggest that the physical modeling is more performant than ANN when a restricted number of experimental data is available for the calibration process. •Comparison between ANN and physics-based energy wear model performance.•Physics-based energy wear model is highly performant even with small data base.•ANN is accurate when sufficient quantity of training data is used for calibration.