An Approach for HVCB Mechanical Fault Diagnosis Based on a Deep Belief Network and a Transfer Learning Strategy

Abstract

Traditional fault diagnosis for a high-voltage circuit breaker (HVCB) encounters the following problems: the fault features extracted by traditional shallow models is of weak expression ability, and the accuracy of fault identification can be affected by the lack of labeled training samples. To overcome these problems, we present a new approach for HVCB mechanical fault diagnosis based on a deep belief network (DBN) and a transfer learning strategy. This approach uses a DBN to achieve the deep mining and adaptive extraction of the inherent features of sample data, and combines the transfer learning method to improve the accuracy of the fault diagnosis model, which uses a large amount of selective auxiliary data to augment the tiny amount of target data learning by adjusting the weight of training samples. The target sample data are obtained by collecting the coil current signal of the HVCB from fault simulation experiments, and the auxiliary sample data are obtained through simulation based on the electromagnetic system mathematical model of the HVCB spring mechanism. The experimental results show that compared with the traditional feature extraction and fault diagnosis method, the DBN approach combined with transfer learning can achieve stronger feature learning and generalization ability.