Nonlinear plant-wide process monitoring using MI-spectral clustering and Bayesian inference-based multiblock KPCA. (August 2015)

• Record Type: Journal Article
• Authors: Jiang, Qingchao; Yan, Xuefeng
• Is Part Of: Journal of process control. Volume 32(2015:Aug.)
• Page Start: 38
• Page End: 50
• Publication Date: 2015-08
• Subjects: Nonlinear plant-wide process monitoring -- Multiblock kernel principal component analysis -- Mutual information-spectral clustering -- Bayesian inference; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2015.04.014
• Languages: English
• ISSNs: 0959-1524

Abstract

Highlights: A novel totally data-driven multiblock process monitoring method is proposed. Mutual information-spectral clustering is proposed for block division. Both linear and nonlinear relations are considered in block division. KPCA is employed to model the nonlinear relations in each block. Results in all blocks are combined together by Bayesian inference. Abstract: Multiblock or distributed strategies are generally used for plant-wide process monitoring, and the blocks are usually obtained based on prior process knowledge. However, process knowledge is not always available in practical application. This work aims to develop a totally data-driven distributed method for nonlinear plant-wide process monitoring. By performing mutual information-spectral clustering, the measured variables are automatically divided into sub-blocks that account for both linear and nonlinear relations among variables. Considering that the variables in the same sub-block can be nonlinearly related, kernel principal component analysis (KPCA) monitoring model is established in each sub-block. The sub-KPCA models reflect more local behaviors of a process, and the monitoring results of all blocks are combined together by Bayesian inference to provide an intuitionistic indication. The efficiency of the proposed method is demonstrated using a numerical example and the Tennessee Eastman benchmark process.