Design Space Exploration and Machine Learning Prediction of Hydrofluorocarbon Solubility in Ionic Liquids for Refrigerant Separation.

Abstract

Ionic liquids (ILs) are promising solvents for the separation of hydrofluorocarbon (HFC) mixtures due to their tunable solvation properties and negligible vapor pressure. We present a computational study of R-32 and R-125 solubility in over 341,000 ILs. These HFCs are widely used in refrigerant mixtures such as R-410A (50/50 wt % R-32 and R-125). Using COSMO-RS based molecular simulation, we compute infinite-dilution activity coefficients that reveal a broad spectrum of solubility and selectivity across IL families. Dimensionality reduction techniques, such as PCA and t-SNE, uncover distinct regions in IL design space with varying potential for HFC absorption. While traditional IL selection for R-410A separation primarily depends on R-32 selective ILs, our analysis reveals many R-125 selective ILs. Building on thermodynamic insights, we also propose a new geometric measure for rapid screening of ILs as solvents for R-410A separation. Furthermore, we develop machine learning (ML) models that accurately predict infinite dilution activity coefficients of R-32 and R-125 in ILs. We develop a binary classifier to further distinguish R-32- vs R-125-selective ILs with over 95% precision and recall. These models provide rapid prediction of infinite dilution activity coefficients, thereby facilitating the identification and design of promising ILs for refrigerant separation involving R-32 and R-125, and are available at https://github.com/aiftakher/HFC-IL-ActivityCoefficient.