A New Data-Driven Method for Nonlinear Process Monitoring. Issue 14 (2019)

• Record Type: Journal Article
• Authors: Chen, Zhiwen; Liu, Chang; Peng, Tao; Yang, Chunhua; Yuan, Xiaofeng; Xu, Degang; Huang, Keke
• Is Part Of: IFAC-PapersOnLine. Volume 52:Issue 14(2019)
• Page Start: 171
• Page End: 176
• Publication Date: 2019
• Subjects: Process Monitoring -- Fault detection -- Just-in-time learning -- Canonical correlation analysis; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2019.09.183
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: In this paper, a new data-driven method called just-in-time learning canonical correlation analysis (JITL-CCA) for tackling nonlinearity in process monitoring is proposed. Canonical correlation analysis (CCA)-based fault detection method has been applied for linear static and dynamic processes. However, CCA has deficiency in coping with nonlinearity existing in real applications, as with other well-established multivariate analysis techniques. This deficiency is illustrated by a numerical example. In recent years, nonlinear analysis tools using kernel principles have been proposed. But the main problem lies in the parameter of kernel function is sensitive and difficult to select. This paper constructs JITL-CCA method to realize on-line learning and monitoring, to build local model and to detect faults with simple parameter setting. Based on T ² statistic in the feature space, JITL-CCA is validated by the simulation benchmark of CSTR.