Identification of Distillation Process Dynamics Comparing Process Knowledge and Black Box Based Approaches

Abstract

A distillation plant equipped with a heat pump separates a mixture of isopropanol and methanol. The mixture contains some water as impurity. The model development aims at dual composition control design, where top and bottom compositions should follow the setpoints, and disturbances should be rejected. Disturbances may occur in feed low rate and feed composition. Identification is performed using multivariable linear discrete time model structure development tools: a process knowledge based and a black box approach. In the process knowledge based approach, the model structure is developed from qualitative process knowledge which presently may require modification to guarantee identifiability. The black box approach is based on pseudocanonical MFD model representation, where the model stracture is determined by a set of structure indices. The identifications are performed on experimental data obtained in closed loop operation of the distillation plant. In the present work, the two approaches are compared in terms of how well the model fits, and predicts the data, the conditioning of the model parameter estimation, and convenience of usage.