文献类型：期刊文献

英文题名：Fault Detection and Diagnosis for Nonlinear and Non-Gaussian Processes Based on Copula Subspace Division

作者：Ren, Xiang[1,2];Zhu, Kunping[3];Cai, Ting[4];Li, Shaojun[1]

机构：[1]East China Univ Sci & Technol, Minist Educ, Key Lab Adv Control & Optimizat Chem Proc, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Dept Math, Shanghai 200237, Peoples R China;[3]Rutgers State Univ, Dept Chem & Biochem Engn, Piscataway, NJ 08854 USA;[4]Rutgers State Univ, Dept Environm Sci, Piscataway, NJ USA

年份：2017

卷号：56

期号：40

起止页码：11545

外文期刊名：INDUSTRIAL & ENGINEERING CHEMISTRY RESEARCH

收录：;EI(收录号：20174204281077);WOS:【SCI-EXPANDED(收录号:WOS:000413057000018)】；

基金：X.R. gratefully acknowledges the financial support from the TA-GA Professional Development fund for research at Rutgers University. S.L. appreciates the National Natural Science Foundation of China (No. 21676086 and No. 21406064), National Natural Science Foundation of Shanghai (14ZR1410500), and Fundamental Research Funds for the Central Universities under Grant 222201717006. The authors would like to thank the anonymous reviewers for their valuable comments and suggestions to improve the quality of the paper.

语种：英文

外文关键词：Distribution functions - Chemical analysis - Fault detection - Gaussian noise (electronic) - Independent component analysis

摘要：A novel copula subspace division strategy is proposed for fault detection and diagnosis. High-dimensional industrial data are analyzed in two elemental subspaces: margin distribution subspace (MDS) modeled by joint margin distribution, and dependence structure subspace (DSS) modeled by copula. The highest density regions of two submodels are introduced and quantified using probability indices. To improve the robustness of the monitoring index, a hyperrectangular control boundary in MDS is designed, and the equivalent univariate control limits are estimated. Two associated contribution indices are also constructed for fault diagnosis. The interactive relationships among the root-cause variables are investigated via a proposed state chart. The effectiveness and superiority of the proposed approaches (double-subspace and multisubspace) are validated using a numerical example and the Tennessee Eastman chemical process. Better monitoring performance is achieved compared with some conventional approaches such as principal component analysis, independent component analysis, kernel principal component analysis and vine copula-based dependence description. The proposed multisubspace approach fully utilizes univariate-based alarm data with a dependences restriction modulus, which is promising for industrial application.