(444a) Swelling Constrained on-Line Level Control of Industrial Batch Reactors

This contribution presents the on-line model predictive level control of a batch reactor. The on-line control algorithm implements a first-principle model based approach in order to maximize the reactor productivity at the end of the batch without causing reactor content overflow, despite the reaction rate disturbances which arise due to catalyst dosing uncertainty. In the beginning of the process operation, until the complete dissolution of the solid component, the reactor system consists of three phases: solid, liquid and gas. The reversible catalyzed reaction takes place in the liquid phase and a gas phase product is formed. The reaction rate is dependent on the concentrations, temperature and the fed catalyst mass. The catalyst decomposes in the reactor; therefore it is fed at several time instances during the reaction. The gas phase product is removed at the top of the reactor and it produces a certain void fraction in the liquid phase. In the case of increased reaction rates the liquid level rises and eventually will overflow the reactor, thus it enters the piping and heat exchanger. After such an event the reaction is stopped and the piping is cleaned, therefore production time is lost. The on-line strategy is required to accommodate the reaction rate disturbances which arise due to catalyst dosing uncertainties (catalyst mass, feed time). In order to cope with the reaction rate disturbances an on-line non-linear model predictive level control strategy is implemented. The controlled variable is the true reactor level, and the manipulated variable is the reactor temperature. The first-principles model consists of a reaction kinetics model linked to the Churn-turbulent hydrodynamic model which calculates the true level in the reactor. The implemented on-line control strategy is based on the shrinking horizon approach. The solution of the optimal control problem is based on the control vector parameterization using a piecewise-constant approximation over equally spaced intervals. The optimal control problem is solved in the sequential way, the numerical optimizer is the pattern search algorithm, and the path constraint (maximum liquid level) violation is included in the objective function. The implemented on-line non-linear model based strategy was able to accommodate the reaction rate disturbances, thus controlled the level without causing excessive reactor content swelling or sub-optimal operation.