Subband averaging kurtogram with dual-tree complex wavelet packet transform for rotating machinery fault diagnosis. (August 2020)

• Record Type: Journal Article
• Authors: Wang, Lei; Liu, Zhiwen; Cao, Hongrui; Zhang, Xin
• Is Part Of: Mechanical systems and signal processing. Volume 142(2020)
• Publication Date: 2020-08
• Subjects: Subband averaging kurtogram -- Dual-tree complex wavelet packet transform -- Rotating machinery -- Fault diagnosis; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2020.106755
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: A new method named subband averaging kurtogram (SAK) is proposed. The dual-tree complex wavelet packet transform (DTCWPT) is incorporated with SAK. Subband rearranged DTCWPT is developed to arrange the subbands in order. Simulation and applications demonstrate the superiority of the proposed method to the original fast kurtogram (FK). Abstract: This paper presents a method called subband averaging kurtogram (SAK), incorporating with dual-tree complex wavelet packet transform (DTCWPT), to improve performance of the fast kurtogram (FK) for rotating machinery fault diagnosis. The proposed method first segments a signal into M sub-signals by a sliding window, then computes the kurtosis of subbands obtained by DTCWPT of each sub-signal. Finally, average kurtosis of corresponding subbands are calculated to obtain the SAK, which indicates the optimal frequency band for the envelope analysis. The FK is easily misled by non-Gaussian noise (e.g., sporadic impulse interferences) whereas the SAK can overcome this problem. Moreover, the DTCWPT simultaneously subdivides bands at high and low frequencies, offers the desirable property of approximate shift-invariance and meanwhile remains less computationally expensive. When the original DTCWPT iterates filter banks on the high-pass channel, the obtained subbands of a signal are not arranged in monotone order of the center frequency. This problem can be resolved by exchanging the inverted filter banks based on their band-pass properties. The proposed method provides improved performance compared to FK, in particular, for extracting periodic transients from noisy signals containing a variety of interferences. A simulation case and two applications to fault diagnosis of a planetary gearbox and a rolling bearing validate the effectiveness and improvements of the proposed method.