Application of modern control techniques to a distillation column

Abstract

Modern control techniques have been applied to a distillation column. Three control techniques were selected for evaluation. These are; a decoupling and disturbance rejection control scheme; an estimator aided control techniques using a Kalman filter; and an implicit generalised minimum variance self tuning control. A 10 tray pilot scale binary distillation column, interfaced with a microcomputer, was used for investigation of the process control techniques. A non-linear model of the column was developed. The reliability of this model was demonstrated. The model was therefore used for the design, analysis and screening of control systems for the pilot plant distillation column. The results of extensive simulations on linearised state variable models of the column simulator demonstrate that the decoupling and disturbance rejection controller works in the presence of load disturbances and setpoint changes. The proper choice of the values of a diagonal matrix in the precompensator of the controller required for accurate setpoint tracking has also been shown. By analogy with PI control, integral and derivative modes have been introduced into the controller to equip it with the ability to remove offsets. Simulation results demonstrate that the sensitivity of the controller to non-linear effects makes the controller inoperable on the column simulator, as well as on the pilot plant. Therefore, the use of an adaptive form of the controller is necessary to compensate for the non-linear effects and other model errors for on-line application to be practical on the pilot plant. On-line implementation of the Kalman filter algorithm using a linear state variable model of the column simulator as the filter model, was not possible because of the large memory requirement of the software, long execution time and the inability to produce satisfactory estimates of all the tray compositions. Simulated and experimental studies for both single temperature control and dual temperature control of the distillation column, demonstrated that self tuning control can provide tighter control of the products of distillation columns than PI control. An algorithm, called the Simplified Correction (SPC) method, has been implemented to prevent the parameters of a self tuning controller from reaching unsatisfactory values when the closed loop system is not sufficiently excited. Simulations show that the SPC can provide significant improvements even when only a subset of the controller parameters are prevented from attaining bad values. The findings in this work verify the degrading effects that model errors have on controller performance. Areas for future work have been suggested in the case of the on-line implementation of the control schemes selected m this work.