Probabilistic Entropy EMD Thresholding for Periodic Fault Signal Enhancement in Rotating Machine

Abstract

Since the slight fault feature of incipient fault is usually polluted by heavy background noise, it is difficult to extract the weak feature signal in rotating machine. As an adaptive decomposing technique, empirical mode decomposition (EMD) based denoising methods have a good effect on the feature separation and noise elimination. However, for rotating machine with poor working environment, the components attributed to noise might have higher amplitudes, which restrict the efficiency of noise reduction in current EMD-based denoising methods. Therefore, a probabilistic entropy EMD thresholding algorithm for periodic fault signal enhancement in rotating machine is proposed in this paper. In this method, the entropy threshold of each IMF is constructed instead of the threshold applied to N sampling points of each IMF directly, which overcomes the shortcoming of the denoising effect limited by larger amplitude noise reservation and smaller amplitude feature signal reduction in the current denoising methods. Meanwhile, in order to make the amplitudes of all the IMF reduce in a smooth way, a multiscale thresholding algorithm based on quantile statistics to provide probability indexes is presented. Engineering application demonstrates that the proposed method is effective in the noise reduction and fault feature enhancement in the rotating machine.