Distributed process monitoring for large‐scale processes based on MJMI‐weighted DKPCA

Abstract

AbstractAlthough the distributed monitoring model has been widely employed in monitoring large‐scale processes, the dynamic nonlinear property in process data is rarely investigated. Given the complex dynamic nature of industrial processes, different process variables interact with each other over time. In order to describe the correlation of dynamic variables more accurately, this work proposes a novel dynamic nonlinear fault detection framework based on maximum joint mutual information (MJMI)‐weighted dynamic kernel principal component analysis (WDKPCA). After dividing the process variables using the MJMI scheme, the proposed dynamic weighting method defines the weight of time‐delayed variables, which allows the dynamic characteristics of these variables to be characterized more comprehensively. By these means, the processes can be decomposed into multiple subblocks, and a distributed monitoring scheme based on DKPCA is thus established. Then, the Bayesian fusion strategy is used to fuse the monitoring results of different subblocks. Through a series of experiments on the Tennessee Eastman (TE) process, the results indicate that MJMI‐WDKPCA has superior process monitoring performance.