Validation and analysis of linear distillation models for controller design

Abstract

It is a nontrivial task to decide whether a model describes the main dynamic characteristics of a process and to determine which model is to be preferred for controller design. Techniques for examining lineal models for controller design are described in Section 2. One focus of this paper is on tools for analysing multivariable processes in the frequency domain, for example, condition number and dynamic relative gain array (RGA) analysis. The latter tool is extended by the phase information of the RGA. A novel evaluation tool involving the singular directions over frequency of the respective processes is introduced. In Section 3, a new linear low-order model is compared to five models from the literature using the analysis tools introduced in Section 2. The new model directly exploits wave propagation phenomena, and, therefore, will be called the wave model. Finally, Section 4 shows briefly result of the H∞-coutrolled column where the controller is designed on the basis of the wave model. Simulations with a detailed nonlinear distillation model show that very good control action can be achieved with this model. Moreover, the linear model allows to predict the nonlinear closed loop behavior quite accurately.