Rotor-Stator Rub-Impact Fault and Position Identification of Aero-Engine Based on VMD-MF-Cepstrum-KNN

• Published in: Tribology transactions Vol. 66; no. 1; pp. 23 - 34
• Main Authors: Chen, Wangying, Yu, Mingyue, Wu, Peng
• Format: Journal Article
• Language: English
• Published: Philadelphia Taylor & Francis 02.01.2023; Taylor & Francis Inc
• Subjects: Aerospace engines; cepstrum; Decomposition; Frequencies; K-nearest neighbors algorithm; KNN; margin factor; Rotors; rub-impact positions identification; Rubbing; Stators; transfer path; VMD
• ISSN: 1040-2004; 1547-397X
• DOI: 10.1080/10402004.2022.2131665

To precisely estimate rub-impact fault between a rotor and stator in an aero-engine and identify a rub-impact location, a method combining variational modal decomposition (VMD), margin factor (MF), and cepstrum has been proposed (VMD-MF-cepstrum). Firstly, to effectively separate rub-impact fault information, VMD algorithm was recruited to decompose original vibration acceleration signals due to its excellent time frequency locality, and the result provides an intrinsic mode function (IMF) from different frequency bands. Furthermore, in view of the sensitivity of the margin factor to the wear extent of the equipment, it was used to select the sensitive IMF that can best embody the characteristic information of rub-impact faults, which is labeled the optimal IMF. Moreover, considering the consistency and difference of transfer path of rub-impact fault characteristic information shown by the signals obtained by the sensors installed in different positions or the same position when the rubbing position is the same (or different). Cepstrum was employed to manifest the feature of transfer path and the transfer path from the optimal IMF was extracted. Finally, according to the characteristics that correspond to the transfer path and the k-nearest neighbor (KNN) algorithm, the fault and locations of rub-impact were judged and identified. Analyses from two independent experiments conducted in 2 days and comparison with classical cepstral method indicated that the proposed method is ideal for unknown samples. Compared with the classical cepstral method, the correctly identified percentage that corresponds to two independent experiments was increased by 52.00% and 43.25%.