Application of an enhanced fast kurtogram based on empirical wavelet transform for bearing fault diagnosis

Abstract

The fast kurtogram (FK), as a fast and effective method for fault diagnosis, is well accepted by many experts and scholars. However, the FK can only estimate the bandwidth and central frequency which come from resonance modulation of the signal. Sometimes useful information (containing faults) may be lost due to the inaccuracy of the estimated center frequency or bandwidth. In this paper, a novel method named empirical scanning spectrum kurtosis (ESSK), based on empirical wavelet transform (EWT), is proposed. Constructed by the principle of EWT, a set of filters with varying bandwidth scan and filter the whole frequency domain from low to high and a series of empirical modal components are obtained. Then, the spectral kurtosis (SK) of these components is calculated. The center frequency and bandwidth corresponding to the component which has the maximum SK are selected as the optimal center frequency and bandwidth. This method can adaptively and accurately find the frequency band containing rich fault feature information, and extract the corresponding component. Multiple simulation signals and experimental signals are used to verify the effectiveness of the proposed method. The results show that the method can maximally extract the components which contain the periodic pulse information and accurately diagnose the faults of the rolling bearing. In addition, comparisons with three popular signal processing methods, including the sparsogram, fir-based FK and short-time Fourier transform (STFT)-based FK are conducted to highlight the superiority of the proposed method.