(171a) Process Intensification of Ternary Distillation Using Dynamic Optimization Method and Data-Driven Approach

Authors: 
Yan, L., McKetta Department of Chemical Engineering, The University of Texas at Austin
Deneke, T. L., University of Helsinki
Heljanko, K., University of Helsinki
Harjunkoski, I., Aalto University
Edgar, T. F., McKetta Department of Chemical Engineering, The University of Texas at Austin
Baldea, M., The University of Texas at Austin
Witt, P., The Dow Chemical Company
Distillation is the dominant separation technology for liquid mixtures in chemical industry. Studies revealed that over 40,000 distillation columns in operation in the US are responsible for more than 90% of purification needs [1]. The thermodynamic efficiency of distillation can be as low as 10% [2], and the operation of distillation can account for half of the energy consumed by chemical plants [3].

Process intensification (PI) is by now an accepted means for improving the energy efficiency of distillation column (sequences). Through the implementation of innovative design changes, process intensification significantly lowers both capital and operating costs. However, tight design integration/integration comes with control challenges: a reduced number of control degrees of freedom and a more constrained operation window compared to conventional columns [4].

In this paper, we rely on a different approach to process intensification, implemented in the temporal domain, which we refer to as “dynamic process intensification (DPI).” DPI aims to improve process efficiency by exploiting and favorably manipulating in the time domain the nonlinear dynamics of existing columns [5-7]. The principle of DPI consists of imposing operational instead of design changes, aiming to meet product quality constraints while minimizing energy per production rate through implementing a transient (periodic) operating pattern. Thus far, such patterns were defined in terms of switching between a finite (generally, two) fixed operating points. In this paper, we introduce a novel approach based on dynamic optimization. We introduce a data-driven approach for learning the DPI optimization-relevant dynamics of a distillation process from its operational history and present a dynamic optimization-based DPI framework. We show that, among others, Hammerstein-Wiener (HW) models that have fewer parameters to regress and can be trained with a considerably smaller amount of data can accurately capture the dynamics of the distillation plant along with its control system [8-9].

We will apply this conceptual framework to an extensive case study separating cyclohexane, toluene and m-xylene mixture. Meeting all quality constraints, DPI can offer considerable energy savings compared to steady-state operation of the same column.

References

[1] Humphrey, J. L. Separation processes: playing a critical role. Chemical Engineering Progress, 1995, 91.na

[2] Humphrey, J. L.; Seibert, A.; Koort, R. Separation Technologies: Advances and Priorities,

tech. rep., Office of Industrial Technologies, Washington DC, 1991.

[3] Isopescu, R.; Woinaroschy, A.; Draghiciu, L. Energy reduction in a divided wall distillation column. Revista de Chimie 2008, 59, 812-815.

[4] Schembecker, G.; Tlatlik, S. Process Synthesis for Reactive Separations, Chem. Eng. Process, 2003, 42, 179-189.

[5] Baldea, M.; Edgar, T. F. Dynamic Process Intensification, Dynamic Process Intensification, Current Opinion in Chemical Engineering, 2018, 22, 48-53.

[6] Yan, L.; Edgar, T. F.; Baldea, M. Dynamic Process Intensification of Binary Distillation Based on Output Multiplicity, AIChE J, 2019, 65(4), 1162-1172.

[7] Yan, L.; Edgar, T.; Baldea, M. Dynamic Process Intensification of Binary Distillation via Periodic Operation, Industrial & Engineering Chemistry Research, 2019, 58(15), 5830-5837.

[8] Richard C Pattison, Cara R Touretzky, Ted Johansson, Iiro Harjunkoski, and Michael Baldea. Optimal process operations in fast-changing electricity markets: framework for scheduling with low-order dynamic models and an air separation application. Industrial & Engineering Chemistry Research, 55(16):4562–4584, 2016

[9] Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT Press, Cambridge, MA, 2016.