Research On Incipient Fault Prediction Of CHR Power Electronic Circuits

With the leap-forward development of the railway industry in China, the security, stahility and reliability of the CRH (China Railway High-speed) was overwhelmed by the attention. In high-speed running state, any small and subtle failure may cause major accidents. While that the power supply system works properly is key for CRH trains to run normally, the power electronic circuits are key to the power system. Therefore it has a great significance to realize the real-time state monitoring and prediction as well as the incipient fault diagnosis for the power electronic circuits. In this paper the following contents are focused.(1) The state-of-art of fault prediction and incipient fault diagnosis is surveyed, with varieties of the fault prediction and diagnosis methods being commented according to their advantages and disadvantages.(2) The major characteristic parameters and the aging trends are introduced for the major kinds of components of a power electronic circuit including the electrolytic capacitors, the inductors, the MOSFET and the power diodes. The failure probability tables of such components are listed with the working and failure mechanisms being analyzed with the focuses on the characteristics of the electrolytic capacitors.(3) Attention is paid to the Buck circuit. Based on investigation to its working principle and trends resulting from its component’s aging, the characteristic parameters describing its trend are obtained and verified by means of the figures, tables and data given out by simulation. The health condition of this circuit is revealed by analyzing such characteristic parameters.4) The Particle Swarm Optimization algorithm (PSO), the grey prediction theory and the RBF Neural network are investigated respectively. Efforts are made to improve their performances and to integrate them together. Based on such improvement and integration, the state monitoring/prediction and the incipient fault diagnosis are studied for the power electronic circuit. Combining components aging characteristics and Buck circuit operation, the average output voltage and the output ripple voltage are obtained by both theoretical analysis and simulation. It has been shown that by integrating varying methods, more accurate prediction could be obtained. This proposed method combines the advantages of the gray theory, PSO and RBF network. As compared to the pure RBF neural network, the experimental results have shown better performances in non-linear tracking and the accuracy. Finally the research in this dissertation is summarized and further discussed.