A New Method of Circuit Fault Diagnosis Based on Clustering Pruned Binary Tree SVMs
Aiming at to the characteristics of fault diagnosis of analog circuit with tolerances,noise and nonlinearities and fault small sample set,this paper presents a novel algorithm based on clustering pruned binary tree support vector machines (SVMs) for these problems. Advantages of clustering and support vector machine are combined by the algorithm. Firstly circuit respond signal is decomposed by wavelet packet transform (WPT) to extract the fault feature of circuit. State sorts similarity is determined according to the distance between different sorts and sorts distribution in feature space. According to the similarity a binary tree SVMs is constructed rationally to enhance fault diagnosis efficiency. The samples having bigger similarity are clustered as the same sort in turn until all samples are clustered as two-class ultimately. Then the two sorts are used as topmost root nods of SVMs binary tree. Meanwhile according to the current estimated state,sub-trees impossibly occurring are pruned and the pruned binary tree SVMs is reconstructed. The results of simulation experiments show that the proposed method ensures bigger classification margin and good generalization ability and has higher classification speed and accuracy compared with several existent fault diagnosis methods
circuit fault diagnosis WPT clustering pruned binary tree SVMs
Yuntao Hou Ran Tao
School of Information Science and Engineering,Northeastern University Shenyang,China Kyoto University,Kyoto,Japan
国际会议
西安
英文
106-110
2011-12-23（万方平台首次上网日期，不代表论文的发表时间）