Study On Fault Diagnosis Of Six Pulse Converter Based On Compact-type Wavelet Neural Network

The cycloconverter is widely applied in production with its small volume, lightweight and without DC link and so on. Once it breaks down, the entire productionprocess will be interrupted, it can even cause equipments destroyed and casualties. So, itis of great theoretical and realistic significance.This paper mainly studies the application of the Compact-type Wavelet NeuralNetwork (CWNN) in fault diagnosis of six pulse cycloconverter. First, it expoundstheories of wavelet analysis, wavelet packet analysis and neural network, studies thecompact-type combining method of wavelet analysis and neural network, anddetermines to adopt two CWNN with Morlet and Mexican hat wavelet as excitationfunction of hidden layer. Research on initialization method of network parametersincluding the selection method of best wavelet function, determination of networkinitial values, and choice of hidden nodes number is done. Then, simulation model ofsix pulse AC-AC VVVF system is built and simulation of thyristor open-circuit fault isdone. Through analyzing output signal of simulation of the cycloconverter, this paperintroduces the output current of converter as sources of fault features, applies waveletpacket band energy analysis technology to extract fault features, and structures featurevectors as input vectors of the CWNN. The simulation data shows the effective-ness oftwo CWNN proposed in this paper, and comparing diagnosis effect of the two CWNNwith BP neural network, the result shows that the CWNN is superior to BP neuralnetwork on both convergence speed and diagnosis accuracy. Finally, converter faultexperiment is performed with six pulse AC-AC VVVF system in the laboratory, andsatisfactory results are obtained by using the two CWNN to diagnose fault respectively.The simulation and experiment results demonstrate that the two CWNN both can beeffectively used in fault diagnosis of six pulse cycloconverter.