An acoustic emission identification model for train axle fatigue cracks based on Deep Belief Network

Abstract Railway axles are safety-critical components of the railroad rolling stock and the consequences of possible in-service failures can have a huge impact. Axle fatigue cracks are relatively common defects during train operation, but how to intelligently identify axle fatigue cracks in running trains is still a great challenge. In order to identify axle fatigue cracks more intelligently, the problem that needs to be solved is how to overcome the manual extraction of features by manual experience as well as shallow networks. Therefore, in this paper, an acoustic emission signal identification method based on deep belief networks (DBNs) for axle fatigue cracks is proposed. In this method, a DBN model is constructed. The axle fatigue crack acoustic emission signal data were obtained by our designed acquisition experimental setup, and these data were used to verify the accuracy of the constructed DBN network model identification. The experimental results show that the method of identification of axle fatigue cracks based on DBN, compared with the traditional fault diagnosis method, eliminates the operations of data feature extraction, feature screening, feature fusion, etc and makes complete use of all the information contained in the fault data. The method can not only identify fatigue crack signals but also has a high identification rate of fatigue cracks at different stages. In the axle fatigue crack acoustic emission identification field, it can be seen that the proposed method in this paper will be a promising approach.