Serial correlated–uncorrelated concurrent space method for process monitoring. (September 2021)

• Record Type: Journal Article
• Authors: Song, Bing; Shi, Hongbo; Tan, Shuai; Tao, Yang
• Is Part Of: Journal of process control. Volume 105(2021)
• Page Start: 292
• Page End: 301
• Publication Date: 2021-09
• Subjects: Process monitoring -- Serial correlations -- Principal component analysis -- Slow feature analysis -- Fault detection; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2021.07.016
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract: Due to the closed-loop feedback control system, the data collected in the industry process is always temporal serial correlated. Moreover, different types of sensor data (such as temperature, pressure, liquid level, flow and so on) would show different temporal serial correlations. Meanwhile, the current process monitoring method considers all types of sensor data equally, ignoring the difference in temporal serial correlations caused by various sensor types. In response to this, serial correlated–uncorrelated concurrent space method is proposed in this work, where process variable original space is divided into serial correlated space and serial uncorrelated space via the degree of temporal serial correlations. In addition, the proposed method not only considers the local monitoring within space via principal component analysis and slow feature analysis respectively, but also monitors the information between-space based on the moving window and the mutual information. Finally, on the basis of the monitoring statistic within space and between-space, the comprehensive monitoring indicator is constructed via the local outlier factor method. The advantage and superiority of the proposed method is illustrated via testing on Tennessee Eastman process and Continuous Stirred Tank Reactor process. Highlights: A novel SCUCS method is developed for refined process monitoring. A novel space partition method is developed on the basis of the data temporal serial correlations. To monitor the relationship changes between-space, a novel method via the moving window strategy and the mutual information is proposed. A novel LOF based comprehensive monitoring indicator is proposed on the basis of within space and between-space monitoring statistics.