Research On Feature Extraction And Diagnosis Of Rolling Bearing Fault Based On Wavelet Packet

Rolling bearings are the most common components in rotating machinery.Their operating environment is relatively harsh,resulting in higher frequency of bearing failures and equipment cannot be operated safely.Therefore,early detection of rolling bearings is extremely important.The rolling bearing vibration signal has the characteristics of non-stationary,non-linear,impact and modulation,which makes the rolling bearing vibration signal extremely complicated,and it is difficult to process and analyze it by the traditional method.In the early stage of bearing failure,the vibration signal usually has weak fault signal and low signal-to-noise ratio.Therefore,how to accurately extract the characteristic information of the fault from the sampling signal,find the appropriate fault classifier and realize the effective diagnosis of the fault has always been the difficulty and hot spot of the rolling bearing fault diagnosis research.This paper focuses on the problem of non-stationary feature fault signal extraction for typical rolling bearing equipment.Starting from the wavelet packet analysis method,the fault feature extraction and diagnosis of rolling bearing are mainly completed.The main research contents are as follows:(1)For the wavelet frequency aliasing problem,the amplitude-frequency characteristics inherent in the wavelet filter,as well as the interval sampling and the interpolation of the points in the reconstruction and decomposition process are analyzed,and the improved wavelet packet algorithm is understood.The effectiveness of the single sub-band reconstruction wavelet packet algorithm is verified by simulation experiments and fault signal analysis experiments,and the improved method is applied to fault diagnosis.(2)Starting from the wavelet packet threshold denoising,based on " 3? " wavelet denoising and FCM layered threshold denoising,a wavelet packet correlation threshold denoising method is proposed.The threshold calculation is based on wavelet packet coefficients and noise coefficients,the high-frequency coefficient(sub-band with less energy)after the wavelet packet decomposition is taken as the noise introduction threshold calculation.(3)A denoising method(LMD-AWPT-SVD)based on local mean decompositionand single sub-band reconstruction to improve wavelet packet correlation threshold and singular value decomposition is proposed.Make full use of the advantages of local mean decomposition,select the PF component with reasonable indicators,and discard the part with less useful information to ensure that the useful information(fault feature information)is not lost.The improved wavelet packet threshold denoising in the paper is used as the pre-filter for singular value decomposition,determining the effective rank by determining the number of useful signals by singular value difference spectrum,the random noise and pulse interference contained in the signal is small under the premise of ensuring the preservation of large singular values,so that the modulation characteristics of the fault signal in the rolling bearing can be fully presented,and thus effectively extract fault feature information.(4)Aiming at the problem that the feature vector is difficult to accurately extract in mechanical fault diagnosis,a new fault diagnosis method based on improved wavelet packet energy entropy and genetic algorithm optimization support vector machine(GA-SVM)classification algorithm is proposed.The improved wavelet packet analysis method is used to decompose the collected signal,extract the energy entropy of wavelet packet,and form the feature vector of the fault diagnosis.As the input,the fault diagnosis model of the GA optimized SVM is established to realize the state recognition of the rolling bearing.Status recognition of different parts.