Spectral denoising random feature decomposition and its application in gear fault diagnosis

• Published in: Applied acoustics Vol. 231; p. 110562
• Main Authors: Liu, Feng, Cheng, Zhengyang, Kang, Xin, Cheng, Junsheng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2025
• Subjects: Adaptive frequency domain denoising; Gear fault diagnosis; Maximum peak envelope segmentation; Random feature energy spectrum; Spectral denoising random feature decomposition; Random feature energy spectrum; Maximum peak envelope segmentation; Gear fault diagnosis; Spectral denoising random feature decomposition; Adaptive frequency domain denoising
• ISSN: 0003-682X
• DOI: 10.1016/j.apacoust.2025.110562

•A new signal decomposition method called SDRFD is proposed for gear fault diagnosis.•In SDRFD, the SSR-based maximum peak envelope segmentation technique is developed.•SDRFD designs an adaptive denoising strategy to remove the redundant noise in the frequency bands.•SDRFD can not only avoid the fragmentation of fault frequency bands, but also significantly enhance fault information.•Simulation and experiments validate the effectiveness and superiority of SDRFD. In engineering practice, due to the existence of strong background noise in measured gear fault vibration signals, the conventional signal decomposition methods are often difficult to accurately extract weak fault features of gear for diagnosis. To this end, this paper proposes a new method called spectral denoising random feature decomposition (SDRFD). Initially, SDRFD builds the random feature energy spectrum of signal using sparse random feature model, thereby revealing the intensity of different components hidden in the signal. On this basis, a scale-space representation (SSR)-based maximum peak envelope segmentation technique is developed to adaptively determine the appropriate segmentation boundaries of spectrum, so that the frequency bands containing fault modulation information can be accurately located. Besides, an adaptive frequency domain denoising strategy is designed to remove redundant noise in the segmented frequency bands, thus enhancing the fault information of frequency band components. Eventually, by recombining the random features of each denoising frequency band, a given signal can be decomposed into several intrinsic random mode components (IRMCs). The analysis results of simulated and experimental signals show that SDRFD can more accurately separate and extract weak fault features of gear, and is an effective gear fault diagnosis approach.