The Research On Fault Diagnosis Of Power Electronic Circuits Based On The Waveform Analysis

Power electronic technology is a new technology which born in the later twentieth century, and it has been developing fast in twenty-first century. With its development, a variety of power electronic products with high performance appearances, more and more power electronic equipments (PEE) applied in practical. Furthermore, they often used in the key flows of production. So, it is of great practical significance to keep the research on the fault diagnosis of PEE. But because of the complexity and impracticability of the theory, although a series of diagnosis methods were formed after decades of development, a wide gap between reality and fiction remains unaddressed.In view of the current situation above, a method of fault diagnosis, which is based on the waveform analysis using basis function, was proposed in this paper. The output waveforms could be considered as a group of fundamental curves. The analytical forms of the curves were called basis functions, and could be expressed by the corresponding vectors. Thereby, all kinds of fault waveforms listed by their vectors. According to the different sentences, which correspond to the sequences, the fault attributes expressed by the sentences could be diagnosed. The fault dictionary can be obtained easier by this method, and the test junctions can be greatly reduced, thus this diagnosis method can be applied in many power electronic circuits.Another method of fault diagnosis for power electronic circuit, which is based on neural network, was also presented in this dissertation. A lot of research work of the errors in back propagation neural network (BP) had been carried out, and an improvement on the arithmetic of BP was proposed. Take advantage of the capability of neural network in association, memory and reasoning, and the characteristic of neural network in fault tolerant, robustness and nonlinear mapping, fault diagnosis can be realized better by this method. Takes the three-phase rectifying circuit for example, the simulation model of the system being diagnosed was constructed in the simulation software MATLAB. Kinds of faults in actual systems were simulated; the wrong waveforms when the faults happen were analyzed; the specimen which sampled from the fault waveforms was used to training the neural network which was then built to process power electronic rectifying circuit diagnosis, and at last, this trained neural network was applied in fault diagnosis. The simulation results indicate the feasibility of this method.