Return to search

Modelling and Advanced Multivariable Control Strategies For a Pilot Scale Extractive Distillation Column

<p>A pilot scale distillation column was built jointly with other students to study the separation of an azeotropic mixture of acetone and methanol. An extractive distillation process is used to break the azeotrope using water as a solvent. This is a nonideal, nonlinear, multicomponent system. A VAX 11/750 computer with DPM-23 industrial front-end system was used in a distributed control configuration for control and data archiving of the column. An on-line Varian-6000 gas chromatograph (G.C.) is used for the top and bottoms product analysis.</p> <p>A complete tray to tray steady state and dynamic mass and energy balance models were developed. The validity of these models has been explored using experimental data over a wide operating region. The simulated behaviour from the dynamic model was compared with the experimental column behaviour for various open loop steps in the reflux flow rate, feed flow rate, feed composition and reboiler duty. The nonlinear behaviour of the system was simulated by the models. A simple model for vapour liquid equilibrium (VLE) calculations was developed and its validity was proven.</p> <p>The grouping of several trays together as a compartment has a thermodynamic basis and reduces the total number of model equations to be solved. Models with 3, 4 and 5 compartments showed good agreement with the experimental data and with a tray to tray model. A saving of 70% in the computation time was achieved over the tray to tray model. The 5 compartment model was used to develop a reduced order bilinear model of the system using the reflux flow rate and the reboiler duty as manipulated variables. The dynamics of the column were investigated for steps in steam pressure, reflux flow rate and feed flow rate. The system behaviour using the bilinear model and a complete tray to tray model were similar to the actual plant data.</p> <p>PI control strategies were simulated for single and dual composition control for four different kinds of disturbances. From the simulation studies it was found that for the control strategy with two single loop the conventional pairing of the reflux flow rate and reboiler duty is suitable for only certain kinds of disturbances. Selection of the water flow rate and reboiler duty (steam pressure) as a manipulated variable pairing showed better performance. This was also shown experimentally.</p> <p>A bilinear optimal controller design was developed which would result in an asymptotically stable system. The optimal controller for a bilinear system is nonlinear in nature, but the control law depends upon the objective function selected. The proof of optimality is given. It was also shown that an optimal controller based on a bilinear equation has a larger region of application than a controller designed from a linearized model. The optimal controller, along with a reduced order bilinear model, was applied to a simulation and on the pilot plant for dual composition control. Reflux flow rate and reboiler duty were used as the manipulated variable pair. The weighting matrices from the objective function and the constant system matrices from the model were used to solve a Liapunov equation to obtained the matrix used in the control law. The tray temperatures were used to predict compositions for the experimental control studies. A recursive least squares (RLS) method was used to update the predictor parameters when the G.C. analyses were available. An optimal controller with the composition predictor was implemented successfully on the pilot scale extractive distillation column for the servo control problem. A bilinear controller was also applied to the regulatory problem of a feed flow disturbance. It is found that even a multivariable controller based on the reflux flow rate and reboiler duty did not perform well. Subsequently, a bilinear model for the extractive distillation was changed to use water flow rate and reboiler duty as manipulated variables. The controller resulting from this model was implemented successfully on the simulation for 20% feed flow disturbance. It should be noted that the system performance depends upon the performance index used in the design of the bilinear controller.</p> / Doctor of Philosophy (PhD)

Identiferoai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/6709
Date January 1988
CreatorsRajput, Kumar Sateesh
ContributorsTaylor, P.A., Wright, J.D., Chemical Engineering
Source SetsMcMaster University
Detected LanguageEnglish
Typethesis

Page generated in 0.0012 seconds