Weak fault detection method in complex strong noise condition based on empirical wavelet transform

Abstract: Empirical wavelet transform (EWT) is able to extract the intrinsic modes of the signal by completely adaptive wavelet basis, which has a complete theoretical base as well as the classical wavelet transform. When the large-scale mechanical equipment in the industrial field is diagnosed and analyzed, and the analyzed vibration signal collected from the equipment often contains complex strong noise, especially a lot of pulse noise. Some recent methods, like the empirical mode decomposition (EMD), propose to decompose a signal accordingly to its contained information. Even though its adaptability seems to be useful for many applications, the main issue with this approach is its lack of theory. Using the adaptive methods to analyze a signal is of great significance to find sparse representations in the context of fault diagnosis. Aiming at the complicated problem of detecting nonstationary vibration signal of weak fault and compound fault with a large amount of background noise, EWT is introduced to improve the accurate diagnosis rate. A detection method for weak faults in complex strong noise condition based on EWT is proposed. By using the peak characteristic of autocorrelation function to judge the periodicity of the decomposed signals, the most obvious decomposition signal is being as the characteristic signal to be detected. The steps of this method are as follows: 1) The original signal is decomposed by EWT; 2) The first sub signal is decomposed continuously by EWT, and then the trend signal in the original signal is obtained until the variance change is less than 0.01; 3) Using the peak characteristic of autocorrelation function to judge the periodicity of each signal, the most obvious signal is the characteristic signal. The EWT analysis of the simulated signal with complex strong noise and the actual signal is carried out, and by the comparison with the EMD, the feasibility and effectiveness of weak fault detection by EWT are verified. Finally, the conclusions are obtained through this research: 1) To the unsteady fluctuating phenomenon of the actual vibration signal in industrial field, EWT can remove the trend item almost perfectly to get more clear spectrum; 2) EWT can suppress impulse noise, and reduce unwanted noise interference as far as possible; 3) Compared with the EMD method, the theory on EWT is more rigorous; 4) EWT is suitable for analyzing nonstationary multicomponent signal, and can extract the mono- component signal. If it is combined with the Hilbert transform, the instantaneous frequency and instantaneous amplitude of the equipment can be obtained, so it can be used to monitor the time frequency condition of the equipment which has the nonstationary vibration of frequency fluctuation and amplitude fluctuation in the industrial field.