Multivariable control of an industrial debutanizer distillation process

Abstract

Most processes encountered in the petrochemical industry are coupled and multivariable in nature. Control loops in multivariable control systems tend to interact with one another where a change in one input variable affects multiple other output variables. This is referred to as signal coupling due to process interactions. Control systems capable of providing satisfactory performance for such processes typically require the use of nontrivial multivariable controller design techniques. This thesis discusses the development of two control strategies suitable for multivariable processes; decentralized proportional-integral-derivative (PID) control and centralized model predictive control (MPC). Among the many control technologies available in the market today, the Proportional-Integral-Derivative (PID) controller is the most widely used controller in industry for its simplicity and ease of implementation with relatively low-cost hardware, providing satisfactory performance for most control applications encountered in industry. The decentralized PID control system is designed using mathematical tools such as the relative gain array (RGA) and the PID controller gain selection is facilitated using the internal model control (IMC) technique. The control loop interactions are compensated by making use of decoupling control techniques. This research presents an opportunity to better understand the important design features offered by the internal model control PID design technique that can be useful for industrial practitioners. Model predictive control (MPC) is an advanced control technique that makes use of a dynamic process model for prediction and process control. Model predictive control was first introduced in the late 1970s and has since found extensive use in the petrochemical industry, particularly in crude oil refining facilities. Centralized model predictive control is designed to handle process interactions inherently and to incorporate constraints on both the manipulated and controlled variables. This research provides the study of tuning parameter trade-offs that industrial practitioners often must make in designing model predictive controllers. The work performed in this thesis includes the development of a dynamic transfer function model of a debutanizer column from step response coefficients exported from an industrial real-life operating plant for study in the MATLAB/Simulink environment. Both control strategies developed in this thesis, decentralized PID control and centralized MPC control, are applied on the dynamic model of the industrial debutanizer distillation process that is part of a Gas Recovery Unit (GRU). A GRU forms a major part of a refinery’s Fluidized Catalytic Cracking Unit (FCCU). FCCUs convert a low value feedstock mixture into high value product streams. The main purpose of a gas recovery plant in the FCCU is to extract as much valuable liquid product from the overhead vapor stream of the FCCU main fractionator as possible to be treated into Liquefied Petroleum Gas (LPG) and gasoline product streams. The debutanizer distillation process studied in this research is used to separate butane and propane from pentane and heavier hydrocarbons used to produce gasoline. The work further develops a testbed for real-time implementation of a closed-loop system in a Hardware-in-the-Loop (HiL) configuration. Hardware-in-the-Loop configurations are essential in facilitating learning for process control students in the academic community to aid their understanding of theoretical concepts taught and the work developed in this research furthers such an objective.