Study Of An Improved Wavelet Packet Threshold Denoising Algorithm Based On Information Entropy

The engineering practice signal contains abundant characteristic information but also often contains interference noise at the same time,the directly time-frequency analysis of noise signal may not be able to get the exact analysis conclusion,due to the good denoising effect and its understandable,wavelet threshold denoising algorithm has been widely used in engineering practice.However,although the denoising algorithm is widely used,but because of the discontinuous problems of the traditional wavelet packet threshold function of hard threshold denoising algorithm,after denoising,signal appear false gibbs phenomenon,denoised signal lost a lot original information;And due to the existence of constant error of soft threshold function,the denoised signal occurred over-compression phenomenon? And the existing threshold estimation algorithm are not be able to adaptively adjust based on the changing situation of noise in wavelet packet coefficient sequences,the layer number of wavelet packet decomposition has no selection criteria,it's usually based on the artificial experience,seriously affected the denoising effect.Therefore,the study of improving the traditional wavelet packet threshold denoising algorithm has very important engineering practical significance.From the above analysis,the wavelet packet threshold denoising effect is mainly related to the three parameters of denoising,respectively is: threshold functions and threshold estimation and determination of decomposition layers.To this,this article carried on thorough research,using information entropy algorithm and its relationship with the signal and noise to improve the above parameters,put forward an improved wavelet packet threshold denoising algorithm based on information entropy.(1)Analyzed and compared three kinds of commonly used information entropy algorithm: Shannon entropy,permutation entropy,sample entropy and their correlation between signal and noise(mixed noise of Gaussian white noise and color noise and periodic random noise).Mainly analyzed the correlation between information entropy and noise size,the correlation between information entropy and data length,the correlation between information entropy and signal inherent characteristics(periodic).And the above three kinds of information entropy is used to represent the noise characterization of simulation noisy signal wavelet packet coefficients.It is concluded that using information entropy analysis amplitude modulation simulation noisy signal wavelet packet coefficients,the analysis effect from the best to the worst are: sample entropy,permutation entropy,Shannon entropy.(2)analyzed the traditional wavelet packet threshold denoising algorithm and its defects,using sample entropy algorithm to improve the traditional algorithm of threshold functions and threshold estimation and determination of the layer number respectively of wavelet packet decomposition method,proposed that using the sample entropy to characterize the noise of the wavelet packet coefficient distribution,obtained sample entropy sequences corresponding to wavelet packet coefficient as a threshold function adjust parameters of the new and improved wavelet packet threshold function,the simulation analysis and ultimately the final experimental results show that the signal analysis methods make thresholding function has the adaptability of noise distribution;Using the the difference between denoised signal and the original signal as noise estimation,and determine the optimal threshold is the threshold that making noise estimation sample entropy took the maximum entropy,determine this threshold removes the noise most complete and keeps most original signal,analysis results show that the method is barely affected by the noise standard deviation,has nothing to do with the length of the signal,suitable for different size and scale of noise signal denoising;Determine that between adjacent decomposition scale wavelet coefficients after the difference between the sample entropy is small to some degree,in the larger decomposition adjacent layer for the optimal decomposition scale decomposition layers,different noise content simulation signal were analyzed,the contrast result proves the method is effective.(3)Using the method to analyze denoising simulation vibration signals and compared with the traditional wavelet packet threshold denoising algorithm and improved algorithm of other literature,introduce three kinds of commonly used denoising effect parameters as an evaluation standard of denoising effect: signal to noise ratio,root mean square error,and smoothness.The results show that the method can remove the noise better,and retains the original signal components,the fundamental frequency of the simulation signal and the simulation fault frequency were restored better,and fewer spectrum interference in frequency,the denoising parameters has made a larger signal-to-noise ratio with smaller root mean square error and smoothness.(4)Experimental analysis of the outer ring fault rolling bearing vibration signal denoising results show that the method in this paper,effectively restored the bearing base frequency and the doubling frequency 1000 Hz and 2000 Hz and 3000 Hz,and the bearing outer ring fault frequency and doubling frequency 2552.9Hz,5150.8Hz,and the bearing outer ring fault frequency modulation components 3552.9Hz.And some of the noise frequency components in the high frequency is well suppressed,denoising effect is perfectly.To sum up,this paper mainly studied the information entropy characterize of the noise and the wavelet packet threshold denoising algorithm improved problems,the correlation of information entropy and hybrid noise in different aspects is analyzed,and use the typical sample entropy as noise characteristic parameter of wavelet packet coefficient,the traditional wavelet packet threshold denoising algorithm is improved,and by verification of the simulation mechanical vibration signal,proved the feasibility of this method,applied to the outer ring fault rolling bearing vibration signals with noise signal denoising and fault diagnosis,the denoising results show that the improved algorithm of this paper is feasible and effective,the denoising results compared with other algorithms,this method has many advantages,provides a new idea for signal wavelet packet denoising pretreatment,and can effectively improve precision of the machine condition monitoring and fault diagnosis accuracy.