基于聚类和SVM多分类的容差模拟电路故障诊断

Fault Diagnosis of Analog Circuits with Tolerance Based on Clustering and SVM Multi-Classification

针对模拟电路故障的容差特性,提出了基于聚类和二叉树SVM多分类相结合的诊断方法。在每个故障特征样本子空间中,建立所有样本与样本重心的空间方向相似度并对其进行升序排列,把排列后的空间方向相似度作为聚类对象,根据聚类分析结果选择故障特征样本,利用所选样本训练二叉树SVM多分类器,实现故障特征样本的分类决策。用故障测试样本进行检验,实验表明采用该诊断方法可以解决容差模拟电路故障模式的识别问题。

Based on clustering method and binary tree SVM multi-classification, a new approach was proposed for the uncertainty characteristic of analog circuits with tolerance. In faulty sample subspace, the spatial direction similar degree of each sample with samples gravity was calculated and which of ascending sort was clustered to select faulty characteristic samples. The binary tree SVM multi-classifier was trained by faulty characteristic samples of preselection to implement classification criterion. By faulty testing samples verification, experiment shows the diagnosis classifiers based on clustering method and binary tree SVM multi-classification can solve the essential recognition problem for analog circuits with tolerance fault types.