A novel multi-classifier for fault diagnosis of analog circuit based on the unsupervised binary tree and support vector machines

Aiming at to the characteristics of fault diagnosis of analog circuit with tolerances, noise, circuit nonlineaiities and small sample set, this paper proposed a novel multi-class classification algorithm which combined unsupervised binary tree (UBT) multi-classifler based on self-organizing map nerve network (SOMNN) clustering rough and support vector machines (SVM) classification accurately. The robustness characteristic of SOMNN based on the embedded separability between pattern classes and SVM based on the theory of statistic learning for the small sample set were integrated in the algorithm. The SOMNN was firstly applied to cluster as layers by which binary-tree multi-classifler structure for fault diagnosis was established, namely, the fault classes at each node of the tree were nailed down. Then according to the preprocess results of SOMNN, SVM were utilized to segment each decision node accurately. Multiple fault diagnosis of analog circuit with tolerances was experimented with the proposed method. The simulation results show us that compared with the several existent multi-class classification methods: the current algorithm has high accuracy and speed.