Journal of Northeastern University(Natural Science) ›› 2021, Vol. 42 ›› Issue (5): 665-672.DOI: 10.12068/j.issn.1005-3026.2021.05.009

• Mechanical Engineering • Previous Articles     Next Articles

Bearing Remaining Useful Life Prediction Method Based on Transfer Learning

WANG Xin-gang, HAN Kai-zhong, WANG Chao, LI Lin   

  1. School of Mechanical Engineering & Automation, Northeastern University, Shenyang 110819, China.
  • Revised:2020-10-26 Accepted:2020-10-26 Published:2021-05-20
  • Contact: WANG Xin-gang
  • About author:-
  • Supported by:
    -

Abstract: To address the problem that most bearing remaining useful life (RUL) prediction methods based on artificial intelligence cannot well predict bearing RUL under different working conditions, a transfer learning method was proposed to predict bearing RUL under different working conditions. Fourier transform was applied to the raw vibration signals of the bearing to obtain the frequency-domain signals, and convolutional neural network (CNN) and long short-term memory network(LSTM) were used to extract the features between data of the bearing′s frequency-domain signals and mine temporal information. The method of combining global and local domain adaption was adopted to reduce the distribution differences of the bearing data under different working conditions. The effectiveness of the method was verified by the existing bearing data. Compared with the traditional deep learning models, the proposed method improves the accuracy of bearing RUL prediction under different working conditions.

Key words: bearing; remaining useful life (RUL); deep learning; transfer learning; domain adaptation

CLC Number: