Research On Fault Diagnosis Methods Of Rolling Bearing Basesd On Multi-Wavelet Analysis

Multi-wavelet, which is also called vector wavelet, is a new kind of wavelettheory based on traditional wavelet analysis. As multi-wavelet space is expanded bymultiple base functions, it can satisfy symmetry, orthogonality, compact support andhigh order vanishing moments features simultaneously. As a result, multi-wavelet ismuch more advantageous in pattern recognition and signal denoising than singlewavelet function. Multi-wavelet will be applied in the fault diagnosis of rollingelement bearings in this dissertation, whose content is as follows.(1) Because vibration signal is one dimensional, it can not be used directly inmulti-wavelet transform. In order to make it be suitable with the r-dimensional spaceof multi-wavelet, preprocessing is discussed firstly. There are two kinds ofmultiwavelet function used in the dissertation, i.e. GHM and CL multiwavelet. At thesame time, two indices, i.e. reconstruction error and ratio of average low-frequencyenergy, are selected to evaluate the performance of the used preprocessing methods.Thereafter, the optimal multi-wavelet function and most suitable pre-processingmethod for simulation signal and real vibration signal are determined.(2) The denoising algorithm is studied based on GHM multi-wavelet integratedwith pre-processing method of repeated data with a coefficient. Due to the influenceof threshold selection on denoising result, the adaptive thresholding and singularvalue decomposition are combined with multi-wavelet decomposition respectively.Based on comparative analysis of simulation function, experimental signal andengineering data of rolling bearings, it can be conclued that multi-wavelet is moreeffective in denoising than single wavelet. Moreover, it is much easier to identify theincipient fault through multi-wavelet.(3) Multi-wavelet packet will be then studied so that more sub-bandinformaiton can be utilized in the fault diagnosis. On the basis of selecting Shannonentropy as cost function, the optimal basis for multi-wavelet packet can be identified.Similarly, the adaptive threholding and singular value decomposition are combinedwith multi-wavelet packet individually to complete the signal denoising. Thedenoising effect of above methods is validated through vibration data of faulty rollingbearings with different pitting size.(4) According to nine modes of bearing status, which include normal mode,pitting failure modes on inner race or outer race with different sizes between 0.5 and5mm, the fault severity index is constructed. In which, multi-wavelet packt decomposition and spectral kurtosis are used to extract the fault feature. Based on thecomplexity calculation of selected decomposition coefficients, the above fault modescan be distinguished. And the different changing trend for inner race and outer racefault can be found.(5) At last, a fault monitoring and diagnosis system for mechatronicalequipment is developed, whose main advantage over other system is the integration ofmulti-wavelet analysis. The practicability and reliability of such a system arevalidated by simulated fault experiments of 6307 rolling bearing.