Remaining Useful Life Prediction for Exciter Rolling Bearing Based on Self-Attentive CNN–GRU

Rolling bearing is one of the key components of the shaker, which is prone to failure under vibration shock loads, and its operational stability plays a crucial role in the operation of the shaker. Aiming at the problems of existing rolling bearing remaining useful life (RUL) prediction methods, such as the single feature extraction capability and the inability to fully utilize the spatiotemporal information embedded in the data, an RUL prediction method based on self-attentive convolutional neural network (CNN) and gated recurrent unit (GRU) is proposed. The method first inputs different time-domain metrics of vibration signals into the improved self-attentive CNN module to extract spatial feature information among the different metrics while performing self-attentive weighting to enhance the feature extraction effect. Next, the data extracted from the CNN layer are input to the GRU layer for life prediction. The experimental results show that the CNN–GRU model reduces the RMSE value by 35.75%–60.83% and elevates the Score by 0.9%–6.7% compared with CNN and GRU.