Fault Diagnosis of Rolling Bearing Based on SDAE and PSO-DBN

Abstract

A new fault diagnosis method for rolling bearing based on two-step cascaded system with two deep network is presented in this paper. In view of the low accuracy of traditional diagnosis algorithm, Stacked Denoising Auto-Encoder (SDAE) model as the first network is used to extract the basic and shallow feature of the fault signal; in order to acquire more robust and deep feature representation, Deep Belief Network (DBN) is configured as the second network. However, as for specific fault diagnosis problems, the number of hidden layer nodes, learning rate and momentum factor will directly affect the diagnosis result of DBN model. Therefore, this paper adopts particle swarm optimization (PSO) algorithm to adaptively select the hyper-parameters of DBN to determine the optimal structure of network, finally realizes the classification of multiple faults. Rolling bearing fault simulation and experiments have been conducted under single load condition to verify the effectiveness of the proposed algorithm. Experimental results obviously demonstrate that, from the aspects of generalization capability and classification performance, this algorithm is superior to support vector machine (SVM), back propagation neural network (BPNN) and grey relational analysis (GRA).