469050 Model Predictive Control of Semicontinuous Distillation Process

Thursday, November 17, 2016: 4:09 PM
Monterey II (Hotel Nikko San Francisco)
Vida Meidanshahi, Brandon Corbett, Prashant Mhaskar and Thomas A. Adams II, Chemical Engineering, McMaster University, Hamilton, ON, Canada

Semicontinuous distillation is a process intensification technique introduced in 2000 [1]. It can purify ternary mixtures to desired specifications. Where conventional continuous distillation requires two distillation columns for ternary separation, semicontinuous system utilizes only one distillation column and a simple storage tank to perform the same task. Therefore, with lower total direct costs, semicontinuous systems have lower total annualized costs for low to intermediate production rates compared to continuous configuration. Semicontinuous distillation operates in stabilized cycles. Initially, the storage tank is charged with fresh feed. The tank feeds the column and the side stream of the column is recycled back to the tank. The light and the heavy boiling point components are removed in the distillate and bottom streams of the column, respectively, during the cycle. As these components are removed from the system, the concentration of the intermediate components reaches the desired purity in the tank. Finally, the tank is discharged and the third product is collected. Subsequently, the tank is recharged and a new cycle begins.

The system is dynamic and control driven and the cycle time is a function of initial tank hold up, column diameter, number of trays and controller parameters [2]. In the last 16 years, the main focus on semicontinuous research was on the design of the process. These studies have shown that the semicontinuous system can economically outperform the conventional continuous configuration for low production rates [3]. In all these studies, proportional integral (PI) controllers were used. PI controllers are easy to implement and show satisfactory performance for the system but tuning its parameters is tedious since the conventional tuning rules cannot be implemented for this dynamic system. On the other hand, due to high variable interactions and nonlinearities of the system, advanced control strategies such as model predictive control (MPC) can improve the performance and the economics of the system.

In this study, for the first time, the implementation of MPC on a semicontinuous system is studied. The separation of benzene, toluene and o-xylene was chosen as a case study. The first principal model of the semicontinuous system was simulated in gPROMS using its public model library (PML) [4]. The simulation results were sent to MATLAB (where the MPC algorithm was implemented) through the gO:MATLAB feature of gPROMS. The subspace identification method was adopted to identify a linear time invariant state-space model to be used in the MPC [5]. The model was identified using the input-output data collected from the process. Subsequently, a shrinking horizon MPC was implemented to obtain the desired purity of intermediate component in the tank and the average purities of the distillate and bottom streams by the end of the cycle while reducing the operating cost. In this implementation, the PI controllers drive the process and stabilize the cycle while the MPC determines the set-points of those PI controllers. Compared to the optimal PI control configuration (the best previously known), the MPC strategy resulted in an 11% improvement in the energy consumption of the system.

[1] Phimister JR, Seider WD. Semicontinuous, middle-vessel distillation of ternary mixtures. AIChE J. 2000;46(8):1508–1520.

[2] Meidanshahi V, Adams II TA. A new process for ternary separations: Semicontinuous distillation without a middle vessel. Chemical Engineering Research and Design 2015;93:100–112.

[3] Adams II TA, Pascall A. Semicontinuous thermal separation systems. Chem. Eng. Technol. 2012;35(7):1153–1170.

[4] Meidanshahi V, Adams II TA. Integrated design and control of semicontinuous distillation systems utilizing mixed integer dynamic optimization. Computers and Chemical Engineering, 2016;89:172–183.

[5] Corbett B, Mhaskar P. Subspace identification for data-driven modeling and quality control of batch processes, AIChE Journal, 2016:62(5):1581–1601.


Extended Abstract: File Not Uploaded
See more of this Session: Process Control Applications II
See more of this Group/Topical: Computing and Systems Technology Division