Joint recurrence based root cause analysis of nonlinear multivariate chemical processes. (July 2021)

• Record Type: Journal Article
• Authors: Ziaei-Halimejani, Hooman; Zarghami, Reza; Mostoufi, Navid
• Is Part Of: Journal of process control. Volume 103(2021)
• Page Start: 19
• Page End: 33
• Publication Date: 2021-07
• Subjects: Joint recurrence plot -- Root cause analysis -- JDET -- Fault diagnosis -- DBSCAN; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2021.05.008
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract: A novel method of diagnosis and causality analysis of faults in chemical processes is developed based on the recurrence theory. By applying and adapting the joint recurrence plot (JRP), the effective feature or variables of each operating condition are examined and then they are used to detect and diagnose faults using an unsupervised method (the absence of fault labels and prior knowledge about different operating conditions), DBSCAN. Also, a method based on the concept of delayed joint recurrence plot is developed, which well reveals the ability to identify the main root cause of each fault and predict the propagation pathway of the fault affecting different variables. In order to reveal the capability of the proposed method in case of nonstationary, unstable, nonlinear data, two different multivariate chemical processes, i.e., Tennessee Eastman and chemical looping combustion, are used. Also, compared with other methods, it was found that the proposed method shows the best performance in fault diagnosis and root cause analysis of complex nonlinear processes even for unobservable faults. Graphical abstract: Highlights: A novel fault diagnosis scheme was proposed based on joint recurrence plot. Nonstationary and nonlinear multivariate chemical processes were analyzed. DBSCAN method was utilized for unsupervised learning investigation. Joint recurrence plot was recognized unobservable faults of Tennessee Eastman process. The proposed method performed best for fault diagnosis and root cause analysis.