Multivariable decoupling model predictive control of reactive distillation column for the production of ethyl acetate

Abstract

.

This work was carried out to develop a decoupling model predictive control algorithmfor a reactive distillation column for the production of ethyl acetate from theestcrification reaction between acetic acid and ethanol. The pilot scale of the columnthat was set up had the rectifying and stripping sections filled with rasching rings whilethe reaction section was filled with Amberlyst 15 solid catalyst. The system oftheoretical models and the prototype of the column were developed using the firstprinciples and IIYSYS 3.2 respectively. The I IYSYS prototype plant was simulated andoptimized taking the maximization of ethyl acetate mole fraction in the condenser as theobjective function. The optimum values of the input variables estimated using I IYSYSoptimizer were used to simulate the theoretical models numerically with the aid ofMATLAB R2010b and run the plant for steady state studies and. in all the cases, theoutput variables were recorded accordingly, fhe validity of the conceptually selectedcontrol configurations were verified by studying the dynamics of the plant. Alsoachieved from the dynamics studies of the reactive packed distillation column was thegeneration of plant input/output data for the development of nonlinear neural networksmodels. Three different nonlinear neural networks models were designed and simulatedfor the reactive packed distillation process with the aid of Neural Network Toolbox 7 ofMATLAB R2010b using Levenberg-Marquardt algorithm as the training algorithm. Thedecoupling of the 3 x 3 multi-input multi-output process was carried out by estimatingthe process transfer function matrix with the aid of System Identification Toolbox 7 ofMATLAB R2010b and formulating the interaction compensator matrix.The next line of action of this work is proceeding to the formulation of the multivariabledecoupling model predictive control algorithm. The algorithm will be used to controlthe process theoretically first before using it to control the real pilot plant for theproduction of ethyl acetate. The results of the algorithm will be compared to that of anundecoupled model predictive control algorithm and a classical Proportional-lntegralDerivative (PID) control algorithm