模拟电路故障最优特征搜索与识别讲义.docVIP

模拟电路故障最优特征搜索与识别 田成来1，彭敏放1，宋丽伟1，谭 虎1，沈美娥2 （1.湖南大学电气与信息工程学院, 长沙 410082; 2. 北京信息科技大学，北京 100101） 摘要：为解决模拟电路故障诊断中故障特征选取的难题并提高故障识别率，本文提出一种故障最优特征矢量候选集的搜索算法并构造了模式最佳邻居查询规则来选取故障模式最优特征进行故障识别。该方法以故障信号小波包分解的频段能量值构造故障样本的初始特征矢量，搜索识别多种模式的特征子矢量生成最优特征矢量候选集，查询模式最佳邻居确定其最大邻域，在综合判据监督下选取故障最优特征完成故障识别。模拟诊断实例表明，选取的最优特征在诊断模拟电路故障时具有满意的准确率。 关键词：故障诊断；模拟电路；最优特征搜索；模式最佳邻居；小波包分解 中图分类号：TM131 文献标志码：A 国家标准学科代码：470.40 Fault Optimal Feature Search and Recognition in Analog Circuit Tian Chenglai1, Peng Minfang1, Song Liwei1, Tan Hu1, Shen Meie2 (1. College of Electrical and Information Engineering, Hunan University, Changsha 410082, China) (2. Beijing University of Information Science and Technology, Beijing 100101, China) Abstract: In order to solve the difficulty of fault feature extraction and to improve fault recognition accuracy rate in analog circuit fault diagnosis, a search algorithm of candidate sets including fault optimal feature is proposed and query rules of patterns’ best neighbors are constructed to select fault optimal feature and to recognize faults in this paper. Firstly, take bands energy values of wavelet packet decomposition of fault signals as the preliminary feature vectors of fault samples. Secondly, search adequate feature sub-vectors identifying many fault patterns to construct candidate sets. Thirdly, confirm their largest neighborhood in feature space by querying patterns’ best neighbors. Finally, choose every pattern’s optimal feature according to the comprehensive criterions when execute pattern recognition of fault samples. Analog diagnosis examples show that the optimal features can recognize fault samples with satisfying classification accuracy rate. Key words: fault diagnosis; analog circuit; optimal feature search; patterns best neighbors; wavelet packet transform 1 引 言 模拟电路的测试和诊断一直是其生产和维护的瓶颈，因此自六十年代开始针对该领域的理论和方法研究就成为热点，模糊理论、神经网络、小波分析等智能信息处理技术对模拟电路故障诊断理论的发展有一定的推进作用[1-9]。 故障特征提取和识别是智能诊断的关键[5]。故障特征提取的常用方法有主元分析法（PCA）和Fisher线性鉴别法(FLDA)等，其中主元分