基于偏置经验特征映射的电路故障诊断方法  

Electronic circuit fault diagnosis method based on biased empirical feature mapping

作　　者：杨智明[1] 俞洋[1] 乔立岩[1] 王钢[2] 

机构地区：[1]哈尔滨工业大学自动化测试与控制系,哈尔滨150080 [2]哈尔滨工业大学通信工程系,哈尔滨150080

出　　处：《仪器仪表学报》2013年第7期1595-1602,共8页Chinese Journal of Scientific Instrument

基　　金：装备预研基金(9140A17050111HT0128);航天支撑基金(2010-HT-HGD-03)资助项目

摘　　要：基于支持向量机的模拟电路故障诊断方法已经成为故障诊断领域的研究热点。然而,在该方法实用化过程中,故障样本集中存在的不平衡分布问题严重影响了该方法的整体诊断性能。针对该问题,提出一种基于偏置经验特征映射的故障诊断方法,该方法将故障样本集映射至经验特征空间,并在该特征空间中使用偏置判别分析准则作为核函数优化的目标函数,最大化所有正常样本同故障样本中心的距离,从而提高故障诊断方法的整体诊断能力。标准数据集以及真实电路上的实验效果表明,提出的方法可以大大缓解由于样本不平衡造成的支持向量机诊断效果下降的问题,从而提高了基于支持向量机的电路故障诊断法方法的适用范围。Analog circuit fault diagnosis method based on support vector machines has become a hot topic in research field of fault diagnosis. However, in the practical application process of this method, the imbalanced distribution problem occurred in fault sample dataset has greatly influenced the effectiveness of the method. Aiming at this problem,this paper proposes a new fault diagnosis method based on biased empirical feature mapping. In the new method, the fault sam- ple dataset is mapped into empirical feature space firstly;then, in this empirical feature space the biased diseriminant analysis criterion is applied as the object function for kernel function optimization, and the distance between the centers of the normal samples and fault samples is maximized, so that the overall fault diagnosis ability of the fault diagnosis method can be improved. The theoretical analysis and experiment result on UCI datasets and real electronic circuit fault diagnosis show that the proposed method can greatly relieve the problem of diagnosis effect degrading of SVM caused by imbalanced samples, so that the applicable area of the fault diagnosis method based on SVM is widened.

关 键 词：故障诊断 模拟电路 不平衡数据集 偏置经验特征映射 

分 类 号：TP2[自动化与计算机技术—检测技术与自动化装置]