Densely Connected Fully Convolutional Auto-Encoder Based Slewing Bearing Degradation Trend Prediction Method

Large slewing bearings are characterized by low rotational speed, high load bearing and long design service life, and their operating condition determines whether the rotating machinery can operate normally. Condition monitoring and prediction of degradation trends in slewing bearings have long been hot topics of research. Traditional health indicator construction and prediction methods require human extraction of features and huge amounts of state label data. To avoid these problems, a health indicator construction method is proposed that combines densely connected fully convolutional auto-encoder (DFCAE) networks with Hidden Markov Model (HMM) in this paper. The proposed method is verified by large-scale slewing bearing data from the highly accelerated life test. The proposed methodology is also compared with other common methods of constructing health indicators, and the results prove that the proposed methodology constructs better health indicators. Finally, machine learning and deep learning networks are used to predict the degradation trend of the test slewing bearing. The prediction results show that the proposed methodology can meet the prediction requirements in the actual operation of large slewing bearings.