中文摘要：

针对模拟电路存在较多故障模式的诊断中易出现分类混叠的问题,提出一种小波分析和分层决策的故障识别方法。首先用小波变换方法提取电路的两种故障特征,模糊C均值算法分析故障特征数据的分布特性,以决策树的形式分割各故障子类。通过对决策树节点特征的优化选择,使各故障子类的区分得以最大化。最后按照决策树结构建立分级诊断的故障决策系统,分别以支持向量机和神经网络作为树节点分类器,有效地提高了故障的识别率。该方法应用于高通滤波器电路的故障识别,正确率高于99%,比经典支持向量机多分类方法具有更好的诊断性能。

英文摘要：

Aiming at overlapped recognition on analog circuit fault diagnosis with large number of fault categories, this paper presented a fault identification approach based on wavelet analysis and hierarchical decision. Firstly, extracted two types of fault features of circuit under test by using wavelet transform. Then processed clustering analysis for fault feature data sets by fuzzy C-mean algorithm, which separated fault sub-classes in form of decision tree. Partitioned the fault sub-classes maximally by optimizing the feature selection on each tree node. Finally, constructed a hierarchical fault decision system by combining multiple classifiers according to the structure of decision tree. Chose support vector machines and neural networks as classifiers for tree nodes to validate the proposed method and improved the fault identification accuracy effectively. The experimental results on a high-pass filter are higher than 99%, which is better than classical support vector machine methods.