A high-accuracy intelligent fault diagnosis method for aero-engine bearings with limited samples
Zhenya Wang,Qiusheng Luo,Hui Chen,Jingshan Zhao,Ligang Yao,Jun Zhang,Fulei Chu
DOI: https://doi.org/10.1016/j.compind.2024.104099
IF: 10
2024-05-02
Computers in Industry
Abstract:As a crucial component supporting aero-engine functionality, effective fault diagnosis of bearings is essential to ensure the engine's reliability and sustained airworthiness. However, practical limitations prevail due to the scarcity of aero-engine bearing fault data, hampering the implementation of intelligent diagnosis techniques. This paper presents a specialized method for aero-engine bearing fault diagnosis under conditions of limited sample availability. Initially, the proposed method employs the refined composite multiscale phase entropy (RCMPhE) to extract entropy features capable of characterizing the transient signal dynamics of aero-engine bearings. Based on the signal amplitude information, the composite multiscale decomposition sequence is formulated, followed by the creation of scatter diagrams for each sub-sequence. These diagrams are partitioned into segments, enabling individualized probability distribution computation within each sector, culminating in refined entropy value operations. Thus, the RCMPhE addresses issues prevalent in existing entropy theories such as deviation and instability. Subsequently, the bonobo optimization support vector machine is introduced to establish a mapping correlation between entropy domain features and fault types, enhancing its fault identification capabilities in aero-engine bearings. Experimental validation conducted on drivetrain system bearing data, actual aero-engine bearing data, and actual aerospace bearing data demonstrate remarkable fault diagnosis accuracy rates of 99.83 %, 100 %, and 100 %, respectively, with merely 5 training samples per state. Additionally, when compared to the existing eight fault diagnosis methods, the proposed method demonstrates an enhanced recognition accuracy by up to 28.97 %. This substantiates its effectiveness and potential in addressing small sample limitations in aero-engine bearing fault diagnosis.
computer science, interdisciplinary applications