| 초록 |
It is known that many of the monitored process variables are not independent. They could be combinations of some independent components that may not be directly measurable. Independent component analysis (ICA) can find these underlying factors from multivariate statistical data. ICA is a recently developed method in signal processing where the goal is to find a linear representation of non-Gaussian data so that the components are statistically independent. Whereas PCA can only impose independence up to the second order (mean and variance) while constraining the direction vectors to be orthogonal, PCA imposes statistical independence up to more than second order on the individual component and has no orthogonality constraint. Hence, ICA can reveal more useful information than PCA. Furthermore, the conventional SPM (statistical process monitoring) methods using PCA are based on the assumption that the measured values of product quality are normally distributed. However, such assumption is often invalid for the measurements gained from actual chemical processes because of their dynamic and nonlinear nature. In the present work, a new statistical monitoring method based on ICA and kernel density estimation is proposed. For investigating the feasibility of the proposed method, its fault detection performance is evaluated and compared with that of PCA monitoring method by applying those methods to the simulation benchmark of biological wastewater treatment process. |
