Rolling Bearing Fault Diagnosis Using Deep Learning Network

Abstract

Automatic and accurate fault diagnosis of rolling bearing is crucial in rotating machinery. Deep belief network (DBN) can automatically learn valid features from signals, which leaves out manual feature selection compared with traditional fault diagnosis methods. In this paper, a novel method called deep belief network with Nesterov momentum is developed for the diagnosis of rolling bearings. Nesterov momentum is used to accelerate training and improve precision. An experimental analysis is carried out using a dataset under different bearing health states from a test rig to substantiate the utility of the proposed DBN architecture. Results show that the method demonstrates impressive performance in bearing fault pattern recognition. Comparison analyses are further conducted to demonstrate that Nesterov momentum can improve the capability of DBN.