(753g) Data-Driven Process Synthesis, Design and Intensification

The complex physicochemical phenomena underpinning a majority of chemical processes can be accurately represented by equivalent physics-based first-principles models. The resulting ordinary differential equation (ODE)- or partial differential equation (PDE)-based models are adept at computing spatial and/or temporal variations in chemical systems. However, performing optimization studies while satisfying the first-principles models is computationally challenging as discretization of differential equations in space and/or time often result in large-scale nonlinear problems. Through a variety of chemical engineering case studies, we demonstrate how effective data-driven strategies can be conceptualized and implemented to obtain computationally-tractable optimization problems which can then be solved to obtain optimal solutions.

Firstly, we consider methanol and ammonia production flowsheets where the reaction unit operations are modeled using pseudo-homogeneous, one-dimensional, non-isobaric and non-isothermal ODE-based models [1,2]. The major objective is to obtain optimal reactor design and operation for which accurate surrogate models are developed using ALAMO [3]. The inputs and outputs of the reactor units are judiciously selected such that the accuracy of approximated models could be preserved while incorporating the crucial design and operation decision variables. The developed surrogate models are cross-validated and subsequently incorporated in the overall optimization formulation, which is solved to obtain globally-optimal solutions.

Secondly, we study dynamic process intensification (DPI) case studies where reaction and adsorption phenomena are combined in a single intensified column to yield cost-effective, efficient and sustainable processes compared to their non-intensified counterparts. The hybrid adsorption-reaction systems are represented through PDE-based one-dimensional, pseudo-homogeneous, non-isothermal, non-adiabatic, and non-isobaric models [4]. Due to the complexity of underlying physics and the resulting PDE-based model, it is difficult to develop accurate approximated surrogate models. Furthermore, as performing the simulations are computationally expensive, the optimization of DPI systems must be performed using as few simulations as possible. To achieve it, we develop an in-house simulation-based data-driven optimization algorithm which comprises of feasibility and optimization phases [5]. Both phases develop surrogate models on the fly for approximating black-box objective and constraints. A novel sampling strategy is also incorporated which can handle hard constraints and reduce overall number of simulations performed. The developed algorithm is applied to optimize several DPI case studies including sorption-enhanced steam methane reforming, sorption-enhanced methanol synthesis and carbon dioxide capture and conversion [6,7].

References

[1] A. Arora, J. Li, M.S. Zantye, M.M.F. Hasan, Process Design Frameworks for Economic Utilization of Small-scale and Unconventional Feedstocks, Found. Comput. Process Des. Accepted (2019).

[2] A. Arora, J. Li, M.S. Zantye, M.M.F. Hasan, Functionality-based Design Framework for Reducing Capital Intensity of Small-scale, Modular Processes, Submitted. (2019).

[3] Z.T. Wilson, N. V Sahinidis, The {ALAMO} approach to machine learning, Comput. Chem. Eng. 106 (2017) 785–795.

[4] A. Arora, S.S. Iyer, M.M.F. Hasan, GRAMS: A General Framework Describing Adsorption, Reaction and Sorption-Enhanced Reaction Processes, Chem. Eng. Sci. 192 (2018) 335–358.

[5] I. Bajaj, S.S. Iyer, M.M.F. Hasan, A trust region-based two phase algorithm for constrained black-box and grey-box optimization with infeasible initial point, Comput. Chem. Eng. 116 (2018) 306–321.

[6] A. Arora, I. Bajaj, S.S. Iyer, M.M.F. Hasan, Optimal Synthesis of Periodic Sorption Enhanced Reaction Processes with Application to Hydrogen Production, Comput. Chem. Eng. 115 (2018) 89–111.

[7] A. Arora, S.S. Iyer, I. Bajaj, M.M.F. Hasan, Optimal Methanol Production via Sorption Enhanced Reaction Process, Ind. Eng. Chem. Res. 57 (2018) 14143–14161.