基于差分动态主元分析的多模态过程故障检测  

Multi-Modal Process Fault Detection Based on Differential Dynamic Principal Component Analysis

作　　者：李元 黄莹莹 LI Yuan;HUANG Ying-ying(Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学信息工程学院

出　　处：《沈阳化工大学学报》2019年第4期339-348,共10页Journal of Shenyang University of Chemical Technology

基　　金：国家自然科学基金重大项目(61490701);国家自然科学基金项目(61673279);辽宁省教育厅重点实验室项目(LZ2015059)

摘　　要：针对现代工业生产过程中的动态和多模态特性问题,提出基于差分动态主元分析(differential dynamic principal component analysis,Dif-DPCA)的故障检测方法.首先,在原始数据中找到每个样本的最近邻样本,将原始样本与最近邻样本进行差分处理,得到新样本.然后,根据时滞h对新样本数据进行增广,并对增广矩阵通过主元分析(principal component analysis,PCA)算法,建立PCA模型,得到平方预测误差(squared prediction error,SPE)和Hoteling T^2控制限,进而对工业过程故障进行在线检测.最后,通过多模态AR过程和实际半导体生产过程验证方法的有效性.Aiming at the dynamic and multi-modal characteristics in modern industrial production process,a fault detection method based on differential dynamic principal component analysis(Dif-DPCA)is proposed.First,the nearest neighbor samples of each sample are found in the original data,and the original samples and the nearest neighbor samples are differentially processed to obtain new samples.Then the new sample data is augmented according to the time lag h,and PCA model is established for augmented matrix through principal component analysis(PCA)algorithm,and the square prediction error(SPE)and hotelling T^2 control limit are obtained,and then the industrial process faults are detected online.Finally,the effectiveness of this method is verified by multi-modal AR process and actual semiconductor production process.

关 键 词：差分算法 多模态过程 动态主元分析算法 故障检测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]