(364h) Solvent Mixture Design Using COSMO-RS Descriptors and Molecular Simulations

Liquid-liquid extraction is a low-energy separation technology, which relies on the selection of a solvent system in which a target solute will partition from its original phase [1]. A key challenge in developing liquid-liquid extraction processes is thus designing solvent systems. Solvent design methods exist using group-contribution based approaches [2]. These approaches often integrate molecular design and mixture design in a single problem formulation to meet with property and process requirements (e.g., solubility, solute recovery). Group contribution-based methods estimate compound and mixture properties based on the contributions of pre-defined groups in the molecular structure [3]. Given the enormous combinatorial space of functional groups, solvent design based on group contributions often leads to a computationally expensive problem, which is often solved using decomposition method and can identify commonly used solvent molecules. Meanwhile, group contribution parameters are obtained through experimental data fitting with potential uncertainties or errors, which can lead to inaccurate designs [4]. In addition, these designed solvent systems require fast and accurate validation and interpretation in terms of feasibility and properties [5]. Hence, research questions exist regarding 1) how to effectively explore the solvent space; 2) how to validate and interpretate designed solvent systems.

In this work, we propose a solvent mixture design method using descriptors of COnductor-like Screening Model for Real Solvents (COSMO-RS) and molecular simulations. Given a feed with compounds and compositions, we first prescreen solvent space based on molecular toxicity and safety. We sample solvent composition within the reduced solvent space and these solvent compositions along with the feed are input for COSMO-RS predictions of solute partition coefficients and activity coefficients. Solvent mixtures are represented using five sigma-moment descriptors based on mixture sigma-profiles [6]. These mixture sigma-profiles define the design space of solvent mixtures. Using the five sigma-moments and feed information as input, we develop a quadratic regression model connecting the input with predicted properties. We formulate a quadratic programming problem to identify optimal solvent mixture with economic or property objectives. These designed solvents are simulated using molecular dynamics (MD) simulations. We apply the proposed approach to bioproduct separation using liquid-liquid separation and show the interplay of mixture properties and separation performance.

[1] Shen, Z., Van Lehn, R. C. Solvent selection for the separation of lignin-derived monomers using the conductor-like screening model for real solvents. Industrial & Engineering Chemistry Research, 59(16), 7755-7764, 2020.

[2] Chen, Y., Kontogeorgis, G.M. and Woodley, J.M. Group contribution based estimation method for properties of ionic liquids. Industrial & Engineering Chemistry Research, 58(10), 4277-4292, 2019.

[3] Constantinou, L. and Gani, R. New group contribution method for estimating properties of pure compounds. AIChE Journal, 40(10), 1697-1710, 1994.

[4] Su, W., Zhao, L. and Deng, S. Group contribution methods in thermodynamic cycles: Physical properties estimation of pure working fluids. Renewable and Sustainable Energy Reviews, 79, 984-1001, 2017.

[5] Valencia-Marquez, D., Flores-Tlacuahuac, A., García-Cuéllar, A.J. and Ricardez-Sandoval, L., Computer aided molecular design coupled with molecular dynamics as a novel approach to design new lubricants. Computers & Chemical Engineering, 156, 107523, 2022.

[6] Austin, N.D., Sahinidis, N.V. and Trahan, D.W. A COSMO-based approach to computer-aided mixture design. Chemical Engineering Science, 159, 93-105, 2017