Research On Fault Diagnosis Of Power Electronic Circuits Based On Wavelet Transform And Neural Network

As the fast development of power electronic technology, the problem of fault diagnosis becomes more and more outstanding. Consequently,it is very important for power electronic circuit to be examined and diagnosed fast.The main research work is as follows:(1) Optimal selection methods of test nodes for analog circuits are researched. Considering the existence of tolerance for analog circuits, a new method of creating ambiguity sets is proposed on the basis of present methods. Due to better time-frequency localizing characteristic of wavelet analysis, wavelet packet decomposition is made to simulation data in every faulty modes and available features information is extracted. In the following the ambiguity sets are created. It improves the accuracy. In addition, test nodes selection algorithm is improved and time consumption is reduced.(2) The methods of fault features extraction are researched. The wavelet transformation is used to preprocess the original data of the circuit under test, and extract available features information. Principal Component Analysis is utilized to extract the principal component information, reduce the dimension and standardize the features information. And therefore the neural network structure is simplified and training speed improved.(3) The applications of neural networks in faults diagnosis are researched. The radial basis function network has better generalization ability and self-adaptive learning ability. Besides, it could converge and realize global optimal solution quickly. As to the reasons above, a method of faults diagnosis for power electronic circuits based on wavelet radial basis function network is proposed. The experiment proves the method could improve training speed and diagnose hard faults better. And comparative analysis is made with other faults diagnosis methods.(4) Faults diagnosis methods based on information fusion technology are researched. The feature-level fusion could realize data condensation and is convenient for real-time disposal. Consequently, a method of faults diagnosis for power electronic circuits based on neural network feature-level fusion is proposed. The method uses features association to realize the correlation between voltage signal and current signal and neural network fusion classifier to recognize the faults. Finally it's validated by the experiment and influence on fusion results between different targets is analyzed.