• CN: 11-2187/TH
  • ISSN: 0577-6686

Journal of Mechanical Engineering ›› 2023, Vol. 59 ›› Issue (16): 157-166.doi: 10.3901/JME.2023.16.157

Previous Articles     Next Articles

Early Stage Fault Diagnosis Method of Bearings Based on Nonlinear Sparse Blind Deconvolution

ZHANG Zongzhen1,2, WANG Jinrui1, HAN Baokun1, BAO Huaiqian1, LI Shunming2   

  1. 1. College of Mechanical and Electronic Engineering, Shandong University of Science and Technology, Qingdao 266590;
    2. College of Energy and Power Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 210016
  • Received:2022-07-05 Revised:2022-10-01 Online:2023-08-20 Published:2023-11-15

Abstract: Aiming at the early fault diagnosis of rotating machinery under big data environment, a fast and efficient method under strong interference based on nonlinear sparse blind deconvolution is proposed. Firstly, the fault expression ability of the objective function of unsupervised learning is studied theoretically. The results show that appropriate nonlinear scaling can improve the relative gradient and the stability of fault expression under abnormal impulsive interference. Then, the normalization and generalized nonlinear convolution activation are used to solve the distribution variation caused by the inconsistent scaling of the nonlinear function. The objective function of the nonlinear sparse blind deconvolution is constructed to build an unsupervised neural network.Adaptive fitting Gaussian window function is used to improve the impulse characteristics of the filters. The final filter is selected using frequency domain kurtosis. Finally, the early fault characteristics of the bearing are obtained by filtering and envelope analysis.The proposed method is verified and compared through the simulation of outer ring fault, distal inner ring fault and IMS test-to-failure bearing data. The results show that the proposed method can learn and enhance weak early faults automatically, and has superior noise adaptability, computational time and robustness. This study provides a good theoretical support for the realization of real-time online monitoring of mechanical equipment and shows a good application prospect.

Key words: unsupervised learning, sparse blind deconvolution, early fault diagnosis, rotating machinery

CLC Number: