The Research Of Fault Diagnosis And Fault Prognostics Method For Power Electronic Circuits

With the rapid development of the power electronic technology, power electronic equipment structure is more complicated, the scale is also growing, the probability of failure is increasing. Power electronic circuits as an important part of the equipment, the entire electrical system failure, even paralysis and serious losses may be caused by its failure. So it is important to ensure the reliability and stability of the running system, find the fault in a timely manner and prevent the happening of the failure. It also makes the method of fault diagnosis and fault prognostics for power electronic circuits got more and more attention.Fault feature extraction and recognition of fault diagnosis, as well as the circuit fault prediction method were studied in this paper, including the following:For the circuit diagnosis, the fault features of the power electronic circuit were extracted by the principal component analysis, the principal components which contain the most of the information of original fault data were obtained, and combined them into new feature vectors, the dimensionality of the data was reduced and the charateristics were highlighted. And then fault identification method based on Fisher Discriminant Analysis(FDA) was studied, to take advantage of it, which can identify the fault features, the final result of fault diagnosis is obtained, the results were compared with the experimental results of RBF neural network identification method. The effectiveness of the proposed method is verified by the simulation results.Taking into account the impact of the nonlinear nature of power electronic circuits, and tolerance, the environment and other factors, the relationship between the characteristics of data are complex. In order to improve the fault feature identification, the fault features were extracted by High Order Cumulant(HOC), each sample's kurtosis and skewness was obtained, and they were combined into new fault vectors. And then the feature vectors were identified by the method of FDA, the accuracy of the diagnosis method was compared with the method in second chapter. Simulation examples show the effectiveness of the proposed method and it has a high diagnostic accuracy.In order to prevent the occurrence of the circuit failure and in a timely manner to take'predict'maintenance, fault prediction method for power electronic circuits was studied. First, the features of circuit output voltage signal under various states are extracted by principal component analysis and high order cumulant respectively and processed by the FDA, the fault indicator parameter which can reflect the health status of the circuit is structed by the data from the processing. Then an empirical model was obtained from the degradation trends of the fault indicator parameters, we studied how to use the particle filter to predict the circuit fault and estimate the remaining useful life. The method is simple in implementation and examples show the effectiveness of the proposed method.