The Application Of Induction-Based Learning Theory In Fault Diagnosis

Fault diagnosis is to discriminate the fault types, reasons and positions in power electronic circuits base on fault information detection and analysis of characteristic. At present, the methods of fault diagnosis in power electronic circuits can be divided into two categories, namely the method depending on mathematic model of the circuit to be diagnosed and independent of the mathematic model. In this paper, the study object is a large-scale rectifier, while the purpose is to develop and perfect automatic fault diagnosis methods. Several key technical problems in fault diagnosis system are studied thoroughly. Firstly, in the paper kinds of fault cases in large-scale rectification equipment are analyzed combining real industry background. The main research target is thyristor. According to characters of signals in power electronic circuits, the wavelet multi-resolution transform theory is applied to analyze the output DC voltage and output DC current. The squared wavelet coefficients are used to find a unique feature vector for each fault. Energy feature space for different fault patterns is thus constructed effectively. Based on multivariate statistical theory, principal component analysis method is used to reduce dimensions of energy feature vectors.Secondly, on the basis of abundant study of the prosperous induction-based learning theory-decision tree learning theory, decision tree classification method is applied in fault diagnosis of strong non-linear systems, such as power electronic circuits. Virtues of decision tree method, namely the powerful ability of dealing with non-linear problem, high classification speed , explainable performance and its simpleness, has been exerted fully. The existing fault diagnosis methods are also studied and used for reference.At last, the powerful simulation tool Matlab, especially Power System toolbox and Wavelet toolbox, together with PSCAD/EMTDC software are applied. The interface between PSCAD/EMTDC and Matlab is also studied and used. According to the proposed method, detailed simulation and test are carried out. The results validate the effectiveness of the proposed method.