| 초록 |
Fault diagnosis has been an important issue in chemical industry. If accurate and immediate detection is available, it can help reduce maintenance cost and optimal operation of process. However, because most of chemical plants are continuous process, the plant should be closed to adjust fault. Moreover, within the faulty state, operation performance and product quality of plant can be worse. Therefore, fault occurrence is a critical issue itself and fault prediction method is necessary to prevent occurrence of process fault. With multivariate analysis, such as PCA or PLS, data dimension can be reduced. In lower dimension space, a characteristic of fault data can be identified and the characteristic of fault data can be trained with machine learning method. In this work, multivariate data was collected from target process and classified the class and characteristic of data by multivariate analysis. The fault prediction system was also proposed by trained characteristic of fault data under lower dimension space with machine learning method. |
