Research Of Rotating Machinery Vibration Fault Signal Weak Feature Extraction Methods

When the rotating machinery fault occurs in the operation, affected by changes in the working environment and working conditions and other factors, lead to mechanical equipment failure vibration signals are mostly complex and non-stationary. Complete and effective for non-stationary vibration signal feature extraction will directly affect the accuracy of fault diagnosis conclusion. To resolve the non-stationary signal problem of weak feature extraction, must resort to a modern, advanced information processing techniques and methods. For this purpose, we study two kinds of rotating machinery vibration signals weak fault feature extraction methods.First of all, in order to improve the soft and hard threshold method of certain defects, a new threshold denoising function was invented. The new threshold function not only had good continuity, and greatly reduced the constant deviation between the estimated wavelet coefficients and the true wavelet coefficients in the soft threshold function. The simulation results show that the new threshold function is more flexible, and more effectively eliminate the white noise. After the signal noise filtering, because the Morlet wavelet has a good time domain and frequency domain characteristics, and optimizing bandwidth parameters and scale parameters of Morlet wavelet, extracting weak fault characteristics of vibration signal. Simulation results showed that this method compared to other methods have better practicality and reliability.In addition, this paper adopted HHT method to extract weak feature based on singular value decomposition and wavelet transform. This method is equivalent to set up a pre-filter before EMD, using singular value decomposition patterns combined with wavelet transform method for signal preprocessing, and analyzed for EMD, using the HHT method for noise reduction signal feature extraction. The analysis revealed that doing so not only can effectively filter out random noise can also restrict EMD level, improve the efficiency and accuracy of EMD.Finally, we compare the two kinds of text used in weak feature extraction techniques,the analysis results showed that for different types and nature of the fault signal to take which kinds of approach is more appropriate for fault feature extraction.