Rolling Bearing Fault Diagnosis Based On Morphological Median Wavelet And Hilbert-huang Transform

Rolling bearing is the most common part in rotating machinery, its operating status affectsthe performance of the entire machine directly. Rolling bearing’s damage fault may causebearing failure, it can cause significant harm on the bearing system, therefore, its state detectionand fault diagnosis has important practical significance.Bearing vibration signals usually contain a lot of noise and impact ingredients, sodenoising should be carried out before analyzing in order to make a more accurate diagnosis.Traditional methods such as Fourier analysis is only applicable to the analysis of stationarysignals. Taking into account the excellent features between Mathematical Morphology andWavelet, Morphological Median Wavelet can better maintain the details of the signal whiledenoising, it is applicable for the analysis of non-stationary signals.Combining with the actual needs of rolling bearing fault diagnosis and on the basis of thesummary of conventional bearing fault diagnosis method, this paper combines MorphologicalMedian Wavelet method and Hilbert-Huang transform (HHT) method to form a new kind ofRolling bearing fault diagnosis method.This paper regards morphological median wavelet and Hilbert-Huang transform (HHT) asthe theoretical basis, expands the following themes: median morphological wavelet theory,signal de-noising, Hilbert-Huang Transform (HHT) theory, fault feature extraction and patternrecognition method, it uses the research route that combine theory with simulation. First, inorder to highlight the advantages of wavelet denoising, this paper implements mathematicalmorphology and wavelet de-noising noising simulation while preprocessing the signal withmedian morphological wavelet method; then, uses the Hilbert-Huang transform (HHT) methodto do feature extraction on rolling bearing vibration signal, first of all, do EMD decompositionand extract feature vector on the four states(ie: normal bearings, inner ring fault, race faultsand rolling element faults), and then select the the first six IMF components after EMDdecomposition to construct eigenvectors, and normalized them all; Finally, classify the datasobtained from the previously experiment with support vector machine (SVM) method, and getthe identification and classification results.