Focusing on the key problem in intelligent fault diagnosis, machine learning, principal component analysis (PCA) and distance separability rule are employed to extract main features from the fault information of turbogenerator units by reducing the dimension of data. It is proved by actual cases that PCA is effective on abstracting fault information with little information loss and no influence on the analysis results. Support vector machines (SVM) can achieve good performance when applied to small samples pattern recognition problems. The basis support vector machine is designed for two-class problem. In this paper, a new support vector machine with method of hierarchical clustering and decision tree, is proposed to solve the multi-class recognition problems. The result indicates that the algorithm is efficient in the fault diagnosis of turbogenerator unit with small samples. Mao JiPei (Control theory and Control engineering)Directed by Prof. Liu ChangLiang...
|