Multivariable System Identification of a Continuous Binary Distillation Column

Distillation is a process that is commonly used in industries for separation purpose. A
distillation column is a multivariable system which shows nonlinear dynamic
behavior due to its nonlinear vapor-liquid equilibrium. In order to gain better product
quality and lower energy consumption of the distillation column, an effective model
based control system is needed to allow the process to be operated over a certain
operating range. In control engineering, System Identification is considered as a well
suited approach for developing an approximate model for the nonlinear system. In this
study, System Identification technique is applied to predict the top and bottom
product composition by focusing the temperature of the distillation column. The
process in the column is based on the distillation of a binary mixture of Isopropyl
Alcohol and Acetone. The experimental data obtained from the distillation column
was used for estimation and validation of simulated models. During analysis, different
types of linear and nonlinear models were developed and are compared to predict the
best model which can be effectively used for designing the control system of the
distillation column. Among the linear models such as; Autoregressive with
Exogenous Input (ARX), Autoregressive Moving Average with Exogenous inputs
(ARMAX), Linear State Space (LSS) model and Continuous Process Model were
developed and compared with each other. The results of this comparison reveals that
the perf01mance of LSS model is efficient and hence it was further used to improve
the modeling approach and compared with other nonlinear models. A Nonlinear State
Space (NSS) model was developed by the combination of LSS and Neural Network
(NN) and is compared solely with NN and ANFIS identification model. The
simulation results show that the developed NSS model is well capable of defining the
dynan1ics of the plant based on the best fit criteria and residual performance. In
addition to this, NSS model predicted the best statistical measurement of the nonlinear
system. This approach is helpful for designing the efficient control system for online
separation process of the plant.