Improved two-level monitoring system for plant-wide processes

Abstract

In modern industries, the plant-wide process has become more and more popular, which always consists of various operation units, equipments, workshops and even factories. Therefore, it is more difficult to monitor those processes, and the monitoring complexity is also much higher. While the traditional multivariate statistical analysis provides satisfactory monitoring results within a single part, e.g. operation unit, it may fail to catch the detailed cross-information among different parts of the process. In this paper, an improved two-level monitoring system is formulated for plant-wide processes. In the first level, the latent variable information is extracted by the principal component analysis (PCA) model, based on which a global latent variable matrix is generated by combining latent variables from different parts of the process. In order to characterize the cross-data information of the plant-wide process, an efficient support vector data description (SVDD) method is employed for modeling the relationships among the global latent variable matrices. Based on the results of the Tennessee Eastman (TE) process, an enhanced performance is obtained by the improved two-level monitoring system. Compared to the traditional PCA based monitoring strategy, the new method is useful to describe the cross-data information of the plant-wide process, based on which more accurate monitoring results can be obtained.