摘要: |
针对传统的核主成分分析方法(KPCA)无法解决在故障样本交叠严重时多分类性能较
差的问题,提出一种基于改进KPCA的特征提取和类峰值特征辅助识别分类相结合的模拟电路
故障诊断方法。在预处理阶段,提出了一种图像混合欧氏距离用于建立核函数,进行核主成
分分析特征提取,克服了传统KPCA的局限性;并且设计了一种用类峰值特征识别的方法进行辅
助识别预分类,提高分类速度。标准电路的故障诊断仿真和结果分析表明,该方法较好地
克服了交叠样本给分类带来的困难,具有很好的故障识别速度和正确率。 |
关键词: 模拟电路 故障诊断 主成分分析 欧氏距离 类峰值特征 |
DOI: |
|
基金项目:国家自然科学基金资助项目(61004002) |
|
An Analog Circuit Fault Diagnosis Method Based on KPCA and Class Peak Characteristics |
TANG Jing,HU Yun-an,XIAO Zhi-cai |
(Department of Control Engineering,Naval Aeronautical and Astronautical University,Yantai 264001,China) |
Abstract: |
Traditional KPCA methods can not solve the problem
of poor multi-classification performance when fault samples overlap s
eriously. So, this paper presents a metho
d of analog circuit fault diagnosis based on improved KPCA and class peak characteristics
. In the data pre-processing stage,an Image Mixed Euclidean Distan
c
e(IMED) kernel principal component analysis method for data dimensionality reduction is proposed,which
overcomes the limitations of traditional KPCA methods.Then,feature recognition of the class peak method is designed to perform
pre-classification so as to improve classification speed.The circuit fault diagnosis shows that the method can
overcome the difficulty caused by overlap samples and is featured by good fault failure recognition speed and
accuracy. |
Key words: analog circuit fault diagnosis PCA euclidean distance class peak characteris |