Application of recurrent neural network to mechanical fault diagnosis: a review

• Published in: Journal of mechanical science and technology Vol. 36; no. 2; pp. 527 - 542
• Main Authors: Zhu, Junjun, Jiang, Quansheng, Shen, Yehu, Qian, Chenhui, Xu, Fengyu, Zhu, Qixin
• Format: Journal Article
• Language: English
• Published: Seoul Korean Society of Mechanical Engineers 01.02.2022; Springer Nature B.V; 대한기계학회
• Subjects: Algorithms; Complexity; Control; Data processing; Dynamical Systems; Engineering; Fault diagnosis; Industrial and Production Engineering; Intelligent manufacturing systems; Machine learning; Mechanical Engineering; Neural networks; Original Article; Recurrent neural networks; State-of-the-art reviews; Vibration; 기계공학; Fult diagnosis; Time series characteristics; Recurrent neural network; Structure optimization
• ISSN: 1738-494X; 1976-3824
• DOI: 10.1007/s12206-022-0102-1

With the development of intelligent manufacturing and automation, the precision and complexity of mechanical equipment are increasing, which leads to a higher requirement for fault diagnosis. Fault diagnosis has gradually transformed from traditional diagnosis algorithm to deep feature mining and expression of highly nonlinear, complex and multidimensional systems. At present, the mechanical fault signals of various equipment are mostly time series. In addition, recurrent neural network (RNN) has strong nonlinear feature learning and processing ability of time sequence information, which has achieved promising results in mechanical fault diagnosis and big data processing. Therefore, this study reviews state-of-the-art RNN method in mechanical fault diagnosis and introduces applications from two aspects: RNN and the combined neural networks which include RNN. Then, this paper discusses the challenges and future development of RNN based fault diagnosis.