Based On Morphological Wavelet And Least Squares Support Vector Machine (svm) Of The Rolling Bearing Fault Diagnosis Research

Roller bearing is one of the most widely used parts in mechanical, and it is a veryvulnerable part to go wrong. When the bearing fails, not only affects the enterprisesnormal production process, but also can cause accidents and have a life-threatening ifthe damage is serious. Therefore, the fault diagnosis research of the rolling is verynecessary.During the operation process of the damaged bearing, its vibrations tend to bestronger. In fact, it is unavoidable phenomenon in bearing running-time. The vibrationsignal of the fault bearing contains a rich set of fault information, so we can go on withthe research by analyzing the vibration signal.The vibration signal of fault bearing often contains noise and should be filteredbefore analyzing. Traditional analysis methods such as Fourier analysis only apply tostationary signal, but the failure signal of rolling bearing is time-varying andnon-stationary. Wavelet analysis is a new time-frequency analysis methods. It can beused for analysis of time-varying signals, and has a good frequency resolution at thelow-frequency, has a good domain resolution at the high-frequency. However, theinherent algorithm of wavelet analysis can cause folding frequency; it is detrimental forthe extraction of fault. Morphological Wavelet contain both the morphology andWavelet’s advantage, it can well maintain the detail of fault signal while de-noising. Inthis article Morphological Median Wavelet is used to extract the feature of the faultbearing, and the energy of various frequency bands of vibration signals is extracted aspattern recognition feature vectors. Comparing with the wavelet, the morphologicalWavelet is much more efficiency, and it can extract some characteristic frequency thatcannot be extracted by wavelet. Traditional neural network classification methods needa lot of training samples; it is not applicable as the data samples of bearing fault datausually are not enough. Classification using support vector machine training needs lesssamples, it can solve nonlinear problems by mapping to a high spatial and will not bringthe dimensions of disaster. Least squares support vector machine(LSSVM) is based on support vector machine, and has a higher efficiency and accuracy.