Research On Fault Diagnose Of Rolling Bearings Based On Least Squares Support Vector Machine

Rolling bearings is one of the important parts and used most commonly in rotating machinery, Its working state impacts directly on the whole machine whether it can properly work or not. The fault diagnosis of rolling bearing has a very important practical significance and economic value.For fault bearing generating vibration signals with characteristics of non-stationary, at the same time, failure information of bearing being extracted difficultly with the methods of traditional signal analysis, therefore, an algorithm which based on wavelet packet transform extracting the fault information of rolling bearing is proposed in this paper.For Artificial Neural Network classifier with disadvantage, such as local minimum and slow convergence rate, an algorithm which fault classification of the rolling bearing based on support vector machine(SVM) is proposed in this paper. Simulation and experiments show that training time and testing time of the algorithm are shorter than the Neural Network classifier's, recognition accuracy of the algorithm is higher than the Neural Network Classifier's.For training process of Support Vector Machine based on solving quadratic optimization, rate of the training will significantly drop when data sample is large size. So, an algorithm which fault classification of rolling bearing based on Least Squares Support Vector Machine is proposed in this paper to transform the SVM quadratic optimization into solving linear equations. Simulation and experiments show that training and testing rate of the algorithm are faster than the support vector machine's and the algorithm can get very high recognition accuracy.For the choices of the width coefficient of kernel function, regularization factor and parameters of the algorithm which were proposed being with uncertainties and randomness, an approach based on Bayesian inference methods to hierarchically determine these parameters is proposed in this paper. Simulation and experiments show that the algorithm improves the efficiency of the modeling and raise training and testing speed and high recognition accuracy.