MODEL DECOMPOSITION BASED ABNORMAL PARAMETER ESTIMATION FOR DISTILLATION COLUMN

Abstract

Parameter estimation method can produce useful physical parameters in finding abnormal causes, but nonlinear model makes this method computationally intensive and non-robust for distillation scenario. In this paper, we propose a model decomposition based parameter estimation method for distillation column diagnosis purposes. Nonlinear first principles dynamic model is divided into some disjoint submodels through occurrence matrix analysis. The whole model is used to monitor distillation process and the submodel that gives the highest contribution to the generated residual is selected to perform abnormal parameter estimation. Application results from stripping tower in the popular Tennessee Eastman challenge problem show that the model decomposition based diagnosis scheme is more time-saving and robust than pure nonlinear model based scheme.