Construction of hierarchical diagnosis network based on deep learning and its application in the fault pattern recognition of rolling element bearings

Abstract

A novel hierarchical diagnosis network (HDN) is proposed by collecting deep belief networks (DBNs) by layer for the hierarchical identification of mechanical system. The deeper layer in HDN presents a more detailed classification of the result generated from the last layer to provide representative features for different tasks. A two-layer HDN is designed for a two-stage diagnosis with the wavelet packet energy feature. The first layer is intended to identify fault types, while the second layer is developed to further recognize fault severity ranking from the result of the first layer. To confirm the effectiveness of HDN, two similar networks constructed by support vector machine and back propagation neuron networks (BPNN) are employed to present a comprehensive comparison. The experimental results show that HDN is highly reliable for precise multi-stage diagnosis and can overcome the overlapping problem caused by noise and other disturbances.