Extreme learning machine based transfer learning for aero engine fault diagnosis

Abstract

Aero engine fault diagnosis is very important to ensure flight safety. However, the lack of fault data limits aero engine fault diagnosis. Fortunately, transfer learning can transfer data information from other fields to the target field, so as to solve the problem of lack of data in the target field. Therefore, according to the idea of transfer learning, this paper proposes two cross-domain aero engine fault diagnosis methods, one-stage-transfer-learning ELM (OSTL-ELM) and two-stage-transfer-learning ELM (TSTL-ELM). These methods are both based on extreme learning machine (ELM), which has fast training speed and good real-time diagnosis. OSTL-ELM uses one stage to extract information from two domains, and TSTL-ELM uses two stages, which realizes the individually adaptation of the target domain. The network weights of the two methods are generated by calculation rather than iteration. Only a small amount of target domain data is needed, and high diagnostic accuracy can be obtained. Finally, the simulation experiment of aero engine degradation state fault diagnosis is designed, and the simulation experiment is carried out under four working conditions. The results show that the two methods are effective in fault diagnosis, and TSTL-ELM has better performance.