(235d) A Generic Decomposition-Based Framework for Computer-Aided Molecular and Process Design | AIChE

(235d) A Generic Decomposition-Based Framework for Computer-Aided Molecular and Process Design


Iftakher, A. - Presenter, Texas A&M University, 3122 TAMU
Monjur, M. S., Texas A&M University
Gani, R., Technical University of Denmark
Hasan, F., Texas A&M University
Designing sustainable, energy-efficient, and cost-effective chemical processes depends on, among others, the choice of enabling materials that directly or indirectly influence the overall process performance [1]. This is especially true for various solvent-based separation processes. Even though many interesting design approaches have been reported that systematically identify solvents through the computer-aided molecular design (CAMD) approach, e.g., optimal CAMD [2] and continuous molecular targeting (CoMT-CAMD) [3], solution of the CAMD problem with complex but accurate and computationally expensive property models is still challenging due to the multiscale nature of data, models as well as decisions that need to be made at various levels. A framework allowing the solution of a wide range of CAMD and related problems is an option worthy of investigation [4].

In this paper, we propose a generic multiscale-multilevel framework for computer-aided molecular and mixture design that utilizes a decomposition-based or direct solution strategies, where the target product-process properties are decomposed, if necessary, into subsets based on computational complexity of the associated property models. For example, qualitatively correct but quantitatively less accurate and computationally fast group-contribution (GC) based pure component predictive models [5] are included in one subset, while COSMO-based phase equilibrium computations [6-7] are included in another subset. Also, machine learning based quantitatively very accurate models [5] are included in another subset. For any CAMD (molecular or mixture design) problem, an intelligent AI-based interface is used to identify the target properties, their desired values in terms of bounds, the needed property models, and the creation of subsets of target properties. A solution strategy and the corresponding design workflow is then created. For each step of the workflow, the needed data, models, and computational tools are retrieved from in-house libraries of databases, property models and numerical techniques.

Application of the framework is illustrated through the solution of three industrially relevant problems, namely, design of ionic liquids for separation of R-125 from R-32, which are the components of refrigerant R410a; separation of isomers by organic solvent-based extraction; and substitution of R410a with a more sustainable mixture. For the first problem, GC-based pure component properties of ionic liquids (IL) (melting point, density, viscosity) are included in the subset that is used to find promising IL-candidates by solving a typical CAMD problem through an MINLP solver. A set of promising candidates are obtained, for which solubility and selectivity of the IL-solvent candidates are determined by COSMO-based computation of vapor-liquid phase equilibrium. Note that for each IL-candidate, their sigma profiles need to be predicted before the VLE can be computed and used in process analysis. Ordering the candidates according to solubility or a function of solubility, identifies the best IL-solvent. For the second problem, organic solvents are considered for the separation of isomers of xylene. In the pure component properties subset, boiling point, melting point, solubility parameter are included as target properties. However, the CAMD problem is now solved through a database search method to identify promising solvent candidates. Accurate ML-based computationally complex property models [5] are used to fill the gaps in the solvent database and thereby providing a wider search range. Similar to the first problem, COSMO-based phase equilibria are computed to identify the best solvent. In the third example, using pure component properties as constraints, candidate binary pairs are first identified through a pattern search method with a database filled through data generated with the ML-based pure component property models. A computer-aided mixture design problem is formulated and solved to identify the binary pair and their composition in the mixture that gives the best performance, considering the refrigeration cycle operation.

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


[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] 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

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

[4] Mann, V., Gani, R., and Venkatasubramanian, V., 2023, Group contribution-based property modeling for chemical product design: A perspective in the AI era, Fluid Phase Equilibria, 113734

[5] Alshehri, A.S., Tula, A.K., You, F., and Gani, R., 2021, Next Generation Pure Component Property Estimation Models: With and Without Machine Learning Techniques, AIChE J., 68 (6), e17469

[6] Hsieh, C-M, Sandler, S.I., Lin, S-T., 2010, Improvements of COSMO-SAC for vapor–liquid and liquid–liquid equilibrium predictions. Fluid Phase Equilibria. 297(1), 90-97.

[7] Eckert, F., Klamt, A., 2002, Fast solvent screening via quantum chemistry: COSMO-RS approach. AIChE J., 48(2), 369-385.