MULTI-CLASS FAULT DIAGNOSIS BASED ON SUPPORT VECTOR MACHINES WITH SEQUENCED BINARY TREE ARCHTECTURE
YUAN Sheng-fa1 CHU Fu-lei2
1.School of Material Science and Engineering, Wuhan University ofTechnology, Wuhan 430070, China;2.Department of Precision Instruments and Mechanology, TsinghuaUniversity, Beijing 100084, China
If the classifiers of support vector machines with binary tree architecture are arrayed randomly in the binary tree, their performance is not the best. A sequenced method in consideration of the sample range is proposed to rationally array the classifiers of support vector machines with binary tree architecture. A sample distribution radius and a sample distribution distance are introduced to estimate the sample range of all classes in the high-dimension characteristic space. The classes with bigger sample range are classified earlier in the higher nodal point of the binary tree architecture, and are given wider classificatory areas in the characteristic space. The experiment of multi-class faults diagnosis of the rotor shows that the method distinctly improves the fault recognition accuracy, the diagnosis speed and the generalization, and it is more suitable for practical application of multi-class fault diagnosis.