Fault Diagnosis Of Motor Bearing Based On Wavelet Theory And Improved PSO-LSSVM

Due to advances in modern science and technology and the development of automation technology,asynchronous motors are spread across the industrial base.Rolling bearing as an important mechanical parts of the motor,the state can directly affect the whole industry system to work properly or not.Therefore,the rolling bearing fault diagnosis has important practical significance and economic value.Based on the study of wavelet theory,this paper employs wavelet packet analysis inciuding energy entropy to analyze the fault characteristics of extraction signal.The fault feature extraction is especially important in the fault diagnosis of rolling bearings,which is the basis of bearing state recognition.For the fault of bearing vibration signals with non-stationary characteristics,this paper uses wavelet packet for rolling bearing signal decomposition.The energy entropy of the various frequencies obtained by decomposition is processed,getting the fault characteristic signal.Based on the theory of the least squares support vector machine(LSSVM),the fault diagnosis of the extracted fault characteristics was chosen by the gaussian radial kernel function.By solving a linear equation using LSSVM,reduce the amount of calculation,the simulation results of different kernel function parameter,punishment factor obtain high diagnostic accuracy,and prove the importance of the two parameters for diagnosis.Because the kernel function parameters of LSSVM and the penalty factor needs to be optimized,a diagnostic algorithm based on Simulated Annealing Improved Particle Swarm Optimization(SAPSO)is proposed by optimizing LSSVM.Using the simulated annealing(SA)algorithm,based on probabilistic inferior to move,to a certain probability to accept a bad solution,it can make the mixed PSO algorithm jump out of local optimum and achieve the global optimal ability.Thus,it can improve the efficiency of algorithm optimizing LSSVM parameters and the accuracy of fault diagnosis.The experimental simulation shows that the LSSVM classifier,which is optimized by this algorithm,can train and test faster,and can obtain quite high recognition accuracy.