Research On Fault Diagnosis Of Asynchronous Motor Based On Neural Network

Asynchronous motor is widely applied and plays an important role in industrial and agricultural production as one of drive facilities. With the rapid development of modern industrial system,unit capacity of motor continuously increases and the load it drives grows more and more complex. Motor faults not only damage motor itself,in some serious cases but also stop motor in a sudden and make the production line collapse which causes great economic losses. Moreover,it may even pose seriously threaten to personal safety. Therefore,it is of great significance to precisely find and diagnose faults of running motors in the industrial production.The thesis analyses the status quo and the existential problems of fault diagnosis of motors home and abroad at present,intensively studies fault characteristics and the mechanistic of several common faults, including induction motors stator fault,broken rotor bars fault,bearing fault,insulation fault,etc. Based on detailed analysis of Fourier transform theory,the paper concludes the weak points when it is used to examine non-stationary signals,proposes an extraction method of fault signals characteristics based upon wavelet transform using energy value extracted in terms of wavelet as fault feature vector.With the development of neural network technology,neural network has already been extensively used in the field of fault diagnosis of motors. Firstly,the paper utilizes three-layer-neural network training from the standard BP algorithm in the simulation platform of MATLAB to diagnose the motor faults. Then it proposes two improvement methods when the test shows less impressive results. One is increasing momentum term to standard BP algorithm in order to be out of local minima when iteration occurs stagnation,which effectively solves the problem of BP algorithm being easy to fall into local minima,and makes it convergent to absolute minimum. Another is to examine faults of asynchronous Motors in terms of BP neural network based on Particle Swarm Optimization(PSO).To be more specific,it is to improve the value of weight and threshold of input layer-hidden layer and hidden layer-output layer of neural network by PSO. It is confirmed that PSO could overcome intrinsic shortcomings of BP neural network,including low learning efficiency, slow convergence rate, being easy to fall into local minima,etc.Finally,the paper extracts the fault characteristics of motor vibration and electric current signals,concludes corresponding fault samples, trains which by BP neural network based on PSO,and applies the trained network to the fault diagnosis of motors. After testing the actual fault data,the precision of this diagnosis method could be substantially improved,so as to solve the fault diagnosis problems of motors,and more precisely and intelligently. This method illustrates a good feasibility and a broad prospect of application.