Hits:

Indexed by:期刊论文

Date of Publication:2018-08-25

Journal:Lecture Notes in Control and Information Sciences

Included Journals:EI

Volume:480

Page Number:203-213

ISSN No.:9783030043261

Key Words:Aircraft engines; Clustering algorithms; Data mining; Engines; Feature extraction; Fighter aircraft; Learning systems; Object oriented programming; Takeoff, Auto encoders; Binary classification; Detection and diagnosis; Fault detection and diagnosis; High dimensional data; Operation conditions; Relevance Vector Machine; Rotating stalls, Fault detection

Abstract:Under various operation conditions, the take-off process of aero-engine is regarded as a typical positive system. Meaning while, the aero-engine surge caused by exerting force in the take-off process brings catastrophic risk to the flight safety and affects overall aero-engine performance. Therefore the precise forecasting of aero-engine rotating stall development process under complex conditions is an effective method for the detection and diagnosis of aero-engine surge fault. In order to avoid the roughness result of the binary classification and the difficulty of feature extraction within high dimensional data for traditional machine learning (ML) approaches, SDA-RVM is developed to provide an accurate rotating stall detection and a surge warning window. Firstly, the SDA is implemented to extract the implicit feature beneath the high dimensional data. Then, the RVM is carried out to calculate the stall trigger probability under the reconstructed vector input. Finally, the surge alert window is identified according to the stall probability. The result of various ML algorithm is compared with the data of on service aero-engine, demonstrating the efficacy of the proposed SDA-RVM approach. © 2019, Springer Nature Switzerland AG.