Fault Diagnosis Method Based On Support Vector Machine

Intelligent diagnosis technology stands for the development direction of diagnosis technology, and it is associated with artificial intelligent technology, and provides the possibilities of intelligentization for fault diagnosis. Support Vector Machine(SVM) based on Statistical Learning Theory(SLT), is a powerful method for machine learning. SVM realizes Structural Risk Minimization Principle, which can effectively solves the problems of nonlinearity, limited samples, and high dimension, usually provides good generalization ability, and has been successfully applied to various areas including fault diagnosis.In the thesis, some issues in the field of SVM for intelligent fault diagnosis are studied. Firstly, a brief introduction of fault diagnosis technology and SVM is given, then the research of SVM in theory and application is presented. Secondly, the foundational issues in machine learning and SLT are summarized, and the basic principles and performance of SVM are discussed, besides, the methods for multi-class based on SVM are analyzed. Thirdly, we study the multi-class classification method based on Decision Tree SVM(DTSVM). In respect that the performance's relation with the structure of DTSVM, Genetic Algorithm is used to solve combinatorial optimization problems during the formation of Decision Tree. Fourthly, we analyze the multi-class SVM based on Directed Acylic Graph(DAG). This method is equivalent to operating on a list and depends on node sequence in DAG, so separability measure is introduced to organize the list and sequence to design DAG. Finally, the classification method of SVM based on Rough Set(RS) is discussed. Using the concept of distribution weight ,we redefine dependency degree and attribute importance of RS, and improve the algorithm of attribute reduction.The study in this thesis is very important to intelligent diagnosis technology, and also useful for SVM.