文献类型：期刊文献

英文题名：Distributed process monitoring based on canonical correlation analysis with partly-connected topology

作者：Peng, Xin[1,2];Ding, Steven X.[2];Du, Wenli[1];Zhong, Weimin[1,3];Qian, Feng[1]

机构：[1]East China Univ Sci & Technol, Key Lab Adv Control & Optimizat Chem Proc, Minist Educ, Shanghai, Peoples R China;[2]Univ Duisburg Essen, Inst Automat Control & Complex Syst, Duisburg, Germany;[3]Tongji Univ, Shanghai Inst Intelligent Sci & Technol, Shanghai, Peoples R China

年份：2020

卷号：101

外文期刊名：CONTROL ENGINEERING PRACTICE

收录：;EI(收录号：20202408813571);WOS:【SCI-EXPANDED(收录号:WOS:000555039300022)】；

基金：This research was supported by the National Natural Science Foundation of China (61890930-3, 61925305 and 61803157), Shanghai Sailing Program, China (18YF1405200), the Fundamental Research Funds for the Central Universities, China (222201814041, 222201917006).

语种：英文

外文关键词：Fault detection; Distributed process monitoring; Canonical correlation analysis; Residual generation

摘要：In this work, a novel data-driven residual generation based process monitoring method is proposed for plant-wide process systems which can be partitioned into several sub-processes and partly communicated with each other. The focus of this method is to reduce the communication costs and risks caused by a centralized process monitoring method, while to avoid the significant decrease in terms of monitoring performance. In this method, canonical correlation analysis is used as the basic method for residual generation, for the reason of its optimum detectability under an affordable false alarm rate (FAR). Then, local models are constructed based on their own information and the quality information from relevant neighbors. In local models, each of key quality index is estimated if the corresponding real values are missing. The main contributions of this work are as follow, (1) Residuals generation based on regression models in a distributed fashion, (2) Process monitoring based on residuals instead of origin process data with improved monitoring performance, (3) Decomposition of the overall process and construct local models according to local measurements and neighborhood communication. Results in a simulated plant-wide process show that the monitoring performances of the proposed method are still satisfactory, but high communication cost is avoided.