An Improved Sparse Representation Based on Local Orthogonal Matching Pursuit for Bearing Compound Fault Diagnosis
Cai Yi, Le Ran, Jiayin Tang, Hang Jin, Zhe Zhuang, Qiuyang Zhou, Jianhui Lin
Abstract
Convolutional sparse representation based on local orthogonal matching pursuit (SR-LocOMP) can recover wheelset-bearing fault impulses without being affected by random slippage and plays an important role in wheelset-bearing fault diagnosis. However, the performance of SR-LocOMP at a low signal-to-noise ratio (SNR) is not satisfied. In addition, when the analyzed signals contain compound faults, the fault feature extraction ability of SR-LocOMP will be greatly affected. To further improve the performance of SR-LocOMP, this article proposes an improved SR-LocOMP algorithm, named improved SR-LocOMP. The proposed method introduces scale-space representation (SSR) for frequency band segmentation and then performs SR-LocOMP for signals in each segmented frequency band. Reconstructed signals corresponding to each segmented frequency band are obtained separately. Envelope spectral <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${L}2/{L}1$ </tex-math></inline-formula> norm is used to measure the fault information content of each reconstructed signal to extract more accurate and richer impulse response features caused by bearing faults. The proposed method not only has the ability to detect compound faults but also can effectively reduce noise interference. Through simulation and experimental analysis, the superiority of the proposed method is verified.