The Application Of Wavelet Analysis In Motor Fault Signal Pre-Processing

Rolling bearings, an indispensable part of large-scale revolving equipment, are not only widely used in industrial manufacture equipment but also related to the entire production systems safety and stability directly. Therefore, the condition monitoring and fault diagnosis of rolling bearing is of great practical significance to avoid accidents and economic losses. Signal pre-processing(denoising) of rolling bearing, the first step of fault feature extraction algorithm, will make a difference to the accuracy of fault diagnosis result. In this paper, the threshold and threshold function selection algorithm of wavelet threshold denoising algorithm will not only be improved and optimized but also be applied to the motor fault vibration signal processing for the sake of efficiency improvement of the bearing denoising, classification and reason of motor fault.To get rid of noise in the motor vibration signal, two new methods based on Bayesian and adaptive wavelet threshold, about noise reduction of the signal during motor vibration and the self-adaption wavelet threshold noise denoising, are proposed. The new threshold, considered different signal denoising characteristics after wavelet transform in different scales after is more suitable for the situation of noise distribution. The new threshold function, improved to denoise the vibration signal, make up the disadvantage of the traditional threshold function. What’s more, the former one, guaranteeing the continuity of the threshold function, solving the problem of the deviation of the traditional threshold function brings, adjusting the threshold function flexibly to adapt to different noise conditions, will make the useful information among signals well protected. The high-frequency resonance components is highlighted after going through the following process steps: firstly, analog signals mixed with noise and real-time signal collection of motor vibration are denoised. Secondly, the signal components is decomposed and denoised by EMD. Third, signal is extracted with Mutual correlation coefficient and kurtosis criterion, which avoid the blindness of IMF component selection. This method was tested effectively to reduce interference from signal mixed noise and to detect the motor fault accurately according to the results of analysis.In this paper, the experimental data were simulated by MATLAB. The experimental results show that: Bearing fault vibration signals are decomposed through improved wavelet threshold denoising combined with EMD. IMF component with obvious fault features are selected by the principles of Mutual correlation coefficient and kurtosis criterion. The envelope spectrum of IMF component is analyzed to find out the obvious fault characteristic frequency spectrum and then determine the type of rolling bearing failure.