| 초록 |
Ethylene and Vinyl Acetate can make different grades by changing ratio of two reactants and process conditions. In this process, fault detection model for each grade is usually made by using PCA. However, the transition occurs too often that it is difficult to find normal condition in EVA reactor. If shut down can’t be predicted, then the plant have to be stop which causes economical loss. So a fault detection model for all grade is needed. To do the fault detection, this paper uses DPCA which considers variable correlation and character of time dependency. At first, faulty case and grade transition case can be detected by using Hotelling’s T^2. After that, clustering analysis is used to divide normal process and abnormal process regardless to grades. As a result, grade transition states can be neglected. By using DPCA and clustering, fault detection is more efficient than using PCA. |
