An Efficient Method for Monitoring Degradation and Predicting the Remaining Useful Life of Mechanical Rotating Components

Data-driven techniques have been widely adopted to monitor the degradation process and predict the remaining useful life (RUL) of mechanical rotating components. However, many existing methods exhibit an inadequate ability to determine the most effective features and their contributions for constructing a health indicator (HI), and they may select the failure threshold that determines the end of life irrationally. To address the abovementioned problems, this article presents the self-normalizing convolutional neural network and gated recurrent unit (SCNN-GRU)-based HI (SCG-HI). First, a complete feature set is constructed that fully reflects the health condition of mechanical rotating components based on raw vibration signals. Then, several high-quality features are that best describe the degradation process are selected to form an effective feature set based on the metrics of time correlation, monotonicity, and robustness. SCG-HI is constructed by implementing the SCNN and the gated recursive unit on the effective feature set. The method is fully evaluated on a benchmark data set, and the results show that the proposed method achieves better degradation monitoring accuracy and is more conducive to RUL prediction than its counterparts.