Effectiveness of ML Algorithms for Prognostics of Bearings in Industry 4.0

The purpose of this article is to investigate the effectiveness of different machine learning regression algorithms for predicting remaining useful life (RUL) of rolling element bearings. A dataset obtained from experimental activities conducted by the engineering department of the University of Ferrara was used. The dataset consists of six run-to-failure vibration signals from self - aligning double row ball bearings. The procedure used in this study consists in four steps: health indicator construction from RMS of signal, health status determination, feature engineering and RUL prediction. RMSE, MAE and MAPE are used as performance metrics. Ensemble modeling is implemented to improve robustness of the prediction. Regardless of the machine learning model employed, the importance of correct data acquisition and pre-processing to obtain good predictions is demonstrated.