基于聚类预处理和支持向量机的模拟电路故障诊断技术

Analog circuit faults diagnosis based on clustering preprocess and SVM

支持向量机(SVM)最初源于两种分类问题,用于存在较多故障模式的模拟电路诊断问题,易造成识别重叠区域。为此提出了利用动态聚类算法作为SVM预分类器的故障诊断方法,首先采用模糊C-均值(FCM)算法对训练样本进行聚类,然后分别对两大类进行内部的子聚类,每一次的聚类都产生两种模式并对各个模式内的故障模式样本训练产生对应的SVM网络,最后采用二叉树形式把所有的模式分开。实验结果表明,采用该方法对测试样本的诊断正确率可以达到99%以上。

Support Vector Machines （SVM） is invented from binary classification problem and it is difficult to be used directly in analog circuit faults diagnosis because of multi-fault modes and overlapped recognition areas. A diagnosis method was proposed by using dynamic clustering method as a preprocessor of SVM. Firstly, Fuzzy C-Means （FCM） algorithm was used to generate father and mother clustering gather, then their children were generated by binary tree structure, and every clustering would train a SVM network to be stored. The simulation experimental results show that accuracy of this diagnosis can be higher than 99%.