最小生成树SVM的模拟电路故障诊断方法

提出最小生成树的支持向量机模拟电路故障诊断方法,通过小波分解提取电路故障特征,在特征空间中以故障类的可分性测度为权值构造最小生成树,得到具有聚类属性的故障子类划分,从而优化故障决策树节点的分布。按照最小生成树的结构建立具有较大分类间隔的多分类支持向量机,能够有效地提高模拟电路故障诊断的正确率。该方法简化支持向量机的结构,在实例电路的故障诊断中获得更高的诊断精度和效率,其性能优于常用的支持向量机方法。

Fault Diagnosis Approach for Analog Circuits Using Minimum Spanning Tree SVM

A fault diagnosis approach for analog circuits based on minimum spanning tree (MST) support vector machine (SVM) is proposed. Fault features of analog circuits are extracted by wavelet analysis method. By taking separability measure of fault classes as weights of edges in feature space, the MST is generated and the sub-class separation for fault groups with clustering property is achieved. The node distribution of fault decision tree is then optimized. Hierarchical multi-class SVMs with large margins are constituted according to the structure of MST, which can effectively improve the fault diagnosis accuracy of analog circuits. The presented approach simplifies the structure of multiclass SVMs. Case study shows that our approach achieves more precision and higher efficiency comparing with other conventional SVM methods in analog circuit fault diagnosis.