(676e) Combining Machine Learning and Optimization for the Inverse Design of Ionic Liquids for Refrigerant Separation

The design of sustainable, energy-efficient, and cost-effective chemical processes is influenced by the selection of enabling materials, which have an impact on overall process performance [1]. This is particularly important for solvent-based separation processes where the selection of solvents dictates the feasibility of the separation process. One such example is the separation of mixed hydrofluorocarbons (HFCs), such as R-410A, R-404A, and R-407C, into their pure components. This is an important separation problem because mixed refrigerants that are commonly used in indoor cooling and industrial refrigeration, also possess high global warming potential (GWP). Moreover, the presence of azeotropes makes their separation especially challenging, rendering conventional distillation processes energy intensive. Ionic liquids (ILs) have emerged as a promising solution to overcome these challenges by facilitating an intensified separation process known as extractive distillation [2]. This process, which combines solvent extraction with thermal fractionation, allows for the breaking of azeotropes and the efficient production of high-purity HFCs. ILs are constructed by a cationic core, an anion, and organic side chains, which makes the selection of optimal ILs combinatorially complex. Given the vast array of potential ILs, systematic screening and selection are critical for identifying candidates that meet specific process requirements, including high purity, recovery, low energy consumption, and cost-efficiency.

Despite advances in mathematical programming-based approaches, such as optimal computer-aided molecule design [3] and continuous molecular targeting (CoMT-CAMD) [4], solving a computer-aided molecular and process design problem is challenging. This is primarily due to the multiscale nature of the problem stemming from nonlinear and nonconvex property and process models. To address these issues, we propose a framework for the discovery of new ILs for separating mixed-HFCs. We first develop machine learning models to accurately represent the nonlinear properties of ILs, such as melting point, viscosity, and density. We also employ COSMO-RS to compute the activity coefficients of R-32 and R-125 (components of R-410A) across over 360,000 ILs. These data are then used to train an artificial neural network (ANNs) that serves as a surrogate model for R-410A solubility in ILs. We then formulate an inverse design problem where the trained ANN models for both pure component and mixture properties are reformulated into mixed integer linear program (MILP) and directly embedded into the optimization model. We incorporate process performance indicators via the objective function, informed by process-level insights. By optimizing extractive distillation configurations for hypothetical solvents to maximize energy efficiency, we find a direct correlation between IL selectivity and energy efficiency. Thus, we determine the objective to maximize IL selectivity. The solution from the optimization model is then validated by the generation of sigma profiles for the predicted IL-structures, as well as computation of R-410A selectivity in those ILs, thereby enabling iterative refinement of the model prediction. Through this framework, we successfully identify a set of IL candidates exhibiting superior separation performance for R-410A. Our approach, while demonstrated on HFC separation, is versatile and can be applied to IL-based carbon capture and other solvent-based separations, which may offer a holistic strategy for identifying ILs that achieve high performance and purity in separation processes.

Keywords: Computer-aided molecular and process design, Group contribution, Machine learning, Optimization, Inverse design.

References:

[1] Iftakher, A., Monjur, M. S., and Hasan, M. M. F, 2023, An Overview of Computer-aided Molecular and Process Design, Chemie Ingenieur Technik 95(3), 315-333.

[2] Monjur, M. S., Iftakher, A., and Hasan, M. M. F, 2022, Separation process synthesis for high-gwp refrigerant mixtures: Extractive distillation using ionic liquids. Industrial & Engineering Chemistry Research 61(12), 4390-4406.

[3] Liu, Q., Zhang, L., Liu, L., Du, J., Tula, A.K., Eden, M., and Gani R., 2018, OptCAMD: An optimization-based framework and tool for molecular and mixture product design, Computers & Chemical Engineering 124, 285-301.

[4] Bardow, A., Steur, K., and Gross, J., 2010. Continuous-Molecular Targeting for Integrated Solvent and Process Design. Ind. Eng. Chem. Res., 49(6), 2834—2840.