Affiliations: [a] School of Mechanical and Transportation Engineering, China University of Petroleum, Beijing, China
| [b] School of Mechanical-Electronic and Vehicle Engineering, Beijing University of Civil Engineering and Architecture, Beijing, China
Correspondence:
[*]
Corresponding author. Jinjiang Wang, School of Mechanical and Transportation Engineering, China University of Petroleum, Beijing, China. E-mail [email protected].
Abstract: With movement toward complication and automation, modern machinery equipment encounters the problems of diversity and complex origination of faults, incipient weak faults, complicated monitoring systems, and massive monitoring data, which are all challenging current fault diagnosis technologies. Conventional machine learning techniques, such as support vector machine and back propagation, have disadvantages in handling the non-linear relationships and complicated structure of massive data. Deep learning (DL) methods have a greater capability to address complex and heterogeneous machinery signals, and identify faults more accurately. This paper presents a review of DL methods in emerging research in the machinery fault diagnosis field. First, common DL models are briefly described. Then, the application of DL to machinery fault diagnosis is described in detail, including the problems DL aims to solve and the achievements it has accomplished thus far. To demonstrate the capability of DL to handle the multiplicity and complexity of equipment faults and massive data, we examine experimental results for typical reciprocating compressor and bearing. Finally, the limitations and trends of further DL development are discussed.
