A multigroup fault detection and diagnosis framework for large-scale industrial systems using nonlinear multivariate analysis. (15th November 2022)

• Record Type: Journal Article
• Authors: Yu, Enliang; Luo, Lijia; Peng, Xin; Tong, Chudong
• Is Part Of: Expert systems with applications. Volume 206(2022)
• Publication Date: 2022-11-15
• Subjects: Multigroup framework -- Fault detection and diagnosis -- Large-scale industrial system -- Nonlinear multivariate analysis -- Variable grouping; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.117859
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: A multigroup fault detection and diagnosis (FDD) framework for industrial systems. The gradKPCA and gradKCCA methods for the analysis of large-scale data sets. A method for dividing system variables into groups using the mutual information. Intra-group and inter-group FDD methods based on gradKPCA and gradKCCA. Abstract: In a large-scale industrial system with numerous variables, the relations among variables are often nonlinear and very complicated, due to material, energy and information flows throughout the entire system. In such systems, fault detection and diagnosis (FDD) suffer from the strong interference of fault-free variables and hence become more difficult, especially for minor faults. To achieve high-performance FDD in large-scale nonlinear industrial systems, a multigroup FDD framework is proposed in this paper. In this framework, system variables are divided into groups firstly, and then FDD are implemented in the form of variable groups. The whole framework consists of three components: a variable grouping method based on mutual information (MI), intra-group FDD methods based on gradKPCA, and inter-group FDD methods based on gradKCCA. The MI-based variable grouping method obtains optimal variable groups by maximizing the MI of variables in every group. The gradKPCA and gradKCCA are used for extracting nonlinear variable relations within each group and between groups, respectively. Except for the ability to cope with nonlinear variable relations, this multigroup FDD framework also has other advantages, such as the improved detection ability to minor faults and the ability to reveal fault transfers between variables and groups. These advantages are demonstrated with a numerical example and an industrial case study.