摘要
针对工业过程的多模态和非高斯特性,提出一种基于改进局部熵主元分析(ILEPCA)的故障检测方法。引入k近邻的均值对局部概率密度函数进行改进,构造改进的局部熵数据剔除多模态和非高斯特性。对改进的局部熵数据建立主元分析(PCA)模型,根据核密度估计计算控制限。对于测试数据,运用改进的局部熵算法预处理后,向PCA模型上投影,计算统计量。通过比较统计量与控制限来进行故障检测。把该方法应用到数值例子和半导体过程故障检测,仿真结果表明,与PCA、核主元分析(KPCA)和局部熵PCA (LEPCA)相比,ILEPCA算法在具有多模态和非高斯特性的工业过程故障检测中具有明显的优越性。
A new fault detection algorithm based on improved local entropy principal component analysis (ILEPCA) was proposed for multimodal and non-Gaussian characteristics of industrial production processes. Mean values of k neighboring was introduced to improve local probability density function. The local entropy was constructed to eliminate the multimodal and non-Gaussian characteristics. The PCA model was established for the improved entropy data and the control limit was calculated based on kernel density estimation. The test data was preprocessed by the improved local entropy algorithm and projected onto the PCA model, and the statistics were calculated. Fault detection of multimodal processes was performed by comparing the statistics with the control limit. The new method was applied for fault detection of a numerical example and a semiconductor process. Simulation results show that compared with PCA, kernel principal component analysis (KPCA) and local entropy PCA (LEPCA) method, ILEPCA has higher accuracy in fault detection of industrial processes with multimodal and non-Gaussian characteristics.
作者
郭金玉
刘玉超
李元
GUO Jin-yu;LIU Yu-chao;LI Yuan(College of Information Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)
出处
《高校化学工程学报》
EI
CAS
CSCD
北大核心
2019年第4期922-932,共11页
Journal of Chemical Engineering of Chinese Universities
基金
国家自然科学基金重大项目(61490701)
国家自然科学基金(61673279)
辽宁省自然科学基金(201602584)
关键词
多模态过程
非高斯特性
故障检测
改进局部熵
主元分析
multimodal process
non-Gaussian characteristics
fault detection
improved local entropy
PCA
