Modeling the Vapor-Liquid Equilibrium of CO2-H2O-[CNC1Im][NTf2] Blends Using Machine Learning Techniques to Determine Feasible Solvent Conditions for CO2 Absorption in a Gas-Sweetening Unit

In the pursuit of sustainable development, ionic liquids (ILs) are of high research interest because of their properties, tuneability, and environmental-friendly features. One of the most notable applications of ILs is in CO2 capture. Solvents with methylimidazolium-based cations partnered with the [NTf2] anion are being investigated due to their high affinity towards CO2. However, these ILs are costly to synthesize and difficult to apply in industrial settings due to their high viscosity. As such, there is a need for pre-synthesis experiments and analysis. This study used machine learning techniques to predict the vapor-liquid equilibrium (VLE) of systems containing CO2, H2O, and [CNC1Im][NTf2]. Data for N = 1 to N = 4 were generated using the PC-SAFT Equation of State and were then used to train independent models for CO2 vapor composition and equilibrium pressure. The trained models could reliably predict the vapor's equilibrium composition and the equilibrium pressure for up to N = 5. Further, the machine learning models were used to solve a simple design problem, where an absorption tower used to sweeten the biogas produced from distillery press mud was considered. Solvent composition and the number of carbons yielding lower minimum solvent flow rate requirements were determined. It was found that high IL concentrations in the liquid and an increased number of carbons in the cation produced lower values for the minimum solvent flow rate required for the 99% separation of CO2 from the inlet vapor.