Sparse representation of transients based on improved matching pursuit algorithm for gear fault diagnosis

Abstract

Localized faults on gears tend to result in periodic transient components under a constant speed operation. Extraction of such transient components is crucially important for gear fault diagnosis. Sparse decomposition based on matching pursuit (MP) is one of the effective methods to extract the weak feature contaminated by strong background noise and has been extensively used for transient feature extraction. In this paper, a practical and effective method is proposed for MP based sparse representation, which is enhanced in terms of sparse dictionary construction and computational complexity of MP. A special wavelet basis matching with the transients is developed to construct the sparse dictionary for fault vibration signal. The inner product operation in MP is replaced by cross-correlation implemented by fast Fourier transform (FFT) to address the problem of enormous computational complexity of MP algorithm. Simulation study shows that the transient feature can be effectively extracted and the computational efficiency is essentially improved by the proposed method. Application in transient components extraction of gearbox vibration signal from an automotive gearbox shows that the proposed method can extract the fault feature effectively.