Remaining Useful Life Prediction for Circuit Breaker Based on Opening-Related Vibration Signal and SA-CNN-GRU

In order to realize the prediction of the remaining useful life (RUL) of the low-voltage conventional circuit breaker mechanical system in the whole life cycle, a prediction method based on opening-related vibration signal and stage attention hybrid neural network is proposed. The opening vibration signal contains richer mechanical state information. Therefore, the variational mode decomposition (VMD) based on energy loss is used to reconstruct the signal to increase the time-domain identification of the shock characteristics of the vibration signal, and the double threshold method (DTM) based on short-time energy entropy ratio (STEER) is used to realize the segmentation of the opening-related vibration signal. A stage attention mechanism (SA) is proposed, combining the convolutional neural network (CNN) and the gated recurrent unit (GRU) to construct the SA-CNN-GRU hybrid model. The first-stage distributed attention (DA) acts on the CNN to refine the mining of the information sensitive to degradation in the time and feature dimensions, and the second-stage time-step attention (TA) acts on the GRU to learn the time-series gradual information of the degradation process in a targeted manner and eliminate the influence of the degradation uncertainty of the complex system. Using the vibration data of three circuit breakers to test, the results show that the proposed method can solve the RUL prediction problem of circuit breaker mechanical systems and have high prediction stability, and the introduction of SA can effectively improve the prediction performance of the model.