Roller bearing intelligent fault diagnosis method under speed fluctuation condition

Speed fluctuation condition is the key problem that affects the fault diagnosis of mechanical equipments, and the existing methods always have defects in computational efficiency and diagnosis error. Due to the strong ability of deep learning on automatic feature extraction and classification, it has attracted a lot of attention in the field of fault diagnosis. Therefore, a novel intelligent fault diagnosis method is proposed based on the advantages of deep learning. In the proposed method, frequency domain samples are extracted according to rotational speed information. And then the samples are employed to train the batch normalized deep neural network. The shift and scale parameters of batch normalization technique are able to solve the frequency shift and amplitude variation properties of frequency domain signals, and also can reduce the internal covariate shift problem of the deep neural network and accelerate convergence. Finally, two specially designed roller bearing experiments under speed fluctuation condition are adopted to verify the effectiveness of the proposed method. The results show that the proposed method can completely ignore the influence of speed fluctuation and achieve accurate identification of different fault types, and obtain a higher accuracy than other methods. © 2020, Nanjing Univ. of Aeronautics an Astronautics. All right reserved.