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

Area of Science:

• Physical Chemistry
• Materials Science
• Chemical Engineering

Background:

• Hydrofluorocarbons (HFCs) are crucial refrigerants, with mixtures like R-410A (R-32 and R-125) facing separation challenges.
• Ionic liquids (ILs) present a promising, low-volatility alternative solvent class for HFC separation due to their tunable properties.

Purpose of the Study:

• To computationally screen a vast library of over 341,000 ionic liquids (ILs) for their potential to separate R-32 and R-125, key components of R-410A.
• To develop predictive models for rapid identification and design of ILs with high selectivity for R-32 or R-125.

Main Methods:

• Utilized COSMO-RS based molecular simulations to compute infinite-dilution activity coefficients for R-32 and R-125 in numerous ILs.
• Employed dimensionality reduction techniques (PCA, t-SNE) to analyze IL design space and identify promising solvent families.
• Developed machine learning models, including a binary classifier, for accurate prediction of HFC solubility and selectivity in ILs.

Main Results:

• Identified a wide range of ILs exhibiting varying solubility and selectivity for R-32 and R-125, challenging the traditional focus on R-32 selective ILs.
• Discovered numerous ILs demonstrating high selectivity for R-125, expanding the scope for R-410A separation strategies.
• Achieved over 95% precision and recall in classifying ILs as R-32 or R-125 selective using machine learning models.

Conclusions:

• Computational screening and machine learning models significantly accelerate the discovery of effective ionic liquids for refrigerant separation.
• The study provides a new geometric measure for rapid IL screening and identifies novel IL candidates for R-410A separation.
• The developed predictive models and curated IL data are valuable resources for designing next-generation refrigerant separation processes.