A novel adaptive resonant band detection method based on cyclostationarity for wheelset-bearing compound fault diagnosis

• Published in: Measurement : journal of the International Measurement Confederation Vol. 213; p. 112770
• Main Authors: Pan, Yanlong, Yi, Cai, Song, Xinwu, Xu, Du, Zhou, Qiuyang, Li, Yanping, Lin, Jianhui
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 31.05.2023
• Subjects: Empirical wavelet transform; Fault detection; Fourier spectrum segmentation; L2/L1 norm of envelope spectrum; Empirical wavelet transform; Fault detection; L2/L1 norm of envelope spectrum; Fourier spectrum segmentation
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2023.112770

•An adaptive resonant band detection method based on cyclostationarity is proposed.•The proposed method is guided by the L2/L1 norm of envelope spectrum to perform multi-level frequency band segmentation adaptively instead of based on fixed frequency band segmentation, which effectively improves the over-decomposition and under-decomposition problems.•The proposed method is guided by envelope spectral indices for both frequency band segmentation and fault resonant band localization, which provides better robustness to impulsive noise and interference.•The proposed method can perform composite fault diagnosis. Based on the results of frequency band segmentations, ICS2, a health index that can accurately characterize the cyclostationarity of repetitive transients, is introduced to guide the detection of fault resonant band. By using different multiples of rotation frequency as window length, the proposed method adaptively realizes multi-level band segmentation to extract the subband with the largest ICS2. Bearing condition is crucial to train operation safety. In order to accurately locate the resonant zone caused by bearing failure, a compound fault resonant band detection method is proposed. Firstly, a multi-level frequency band segmentation method based on L2/L1 norm of envelope spectrum is designed. Secondly, based on the results of frequency band segmentations, ICS2, a health index that can accurately characterize the cyclostationarity of repetitive transients, is introduced to guide the detection of fault resonant band. By using different multiples of rotation frequency as window length, the proposed method adaptively realizes multi-level band segmentation to extract the subband with the largest ICS2. Moreover, the proposed method has the capability of multiple fault diagnosis. Simulation signals and bench signals verify the effectiveness of the proposed method.