This work conducts a comprehensive modeling study on the viscosity, density, heat capacity, and surface tension of ionic liquid (IL)-IL binary mixtures by combining the group contribution (GC) method with three machine learning algorithms: artificial neural network, XGBoost, and LightGBM. A large number of experimental data from reliable open sources is exhaustively collected to train, validate, and test the proposed ML-based GC models. Furthermore, the Shapley Additive Explanations technique is employed to quantify the influential factors behind all the studied properties. Finally, these ML-based GC models are sequentially integrated into computer-aided mixed solvent design, process design, and optimization through an industrial case study of recovering hydrogen from raw coke oven gas. Optimization results demonstrate their high computational efficiency and integrability in solvent and process design, while also highlighting the significant potential of IL-IL binary mixtures in practical applications.

Ionic liquids (ILs) have received much attention in both academia and industries due to their superior performance in many applications. Efficient recovery/recycling of ILs from their dilute aqueous solutions is essential for the acceptance and implementation of many IL-based technologies by industry. In this work, a practical and cost-effective hybrid process design method that combines aqueous two-phase extraction, membrane separation, and distillation operating at their highest efficiencies is proposed for the recovery of hydrophilic ILs from dilute aqueous solutions. The application of this hybrid process design method is illustrated through case studies of recovering two hydrophilic ILs, n-butylpyridinium trifluoromethanesulfonate ([C4Py][TfO]) (CAS number: 390423-43-5) and 1-butyl-3-methylimidazolium chloride ([C4mIm][Cl]) (CAS number: 79917-90-1), from their dilute aqueous solutions. For the recovery of 10 wt.% [C4Py][TfO] from aqueous solution, the hybrid process using (NH4)2SO4 as the salting-out agent could reduce the total annual cost (TAC) and energy consumption by 57% and 91%, respectively, compared with the pure distillation processes. In the case of recovering 10 wt.% [C4mIm][Cl] from aqueous solution, the reduction in TAC and energy savings of the hybrid process with salting-out agent (NH4)2SO3 could reach 49% and 87%, respectively, compared with the pure distillation process. Furthermore, uncertainty analysis through Monte Carlo simulations show that the proposed hybrid process design is more robust to uncertainties in energy prices and other material (e.g., equipment and solvent) costs.

Current azeotropic refrigerant R410A has the disadvantage of a high GWP. Thus, a reliable predictive model is necessary to be developed for screening of the alternative refrigerant mixtures with low boiling point. In this work, a database of binary refrigerant azeotropic mixtures is identified by experiments and used to verify the feasibility of using COSMO-based models to predict the VLE of binary refrigerant mixtures. It contains 1766 compounds, out of which 306 are known as refrigerants. Sixty-nine binary systems are then selected after excluding the systems containing water and those with incomplete and azeotropic temperatures exceeding 325 K. The COSMO-based models were tested and found to predict the azeotropic points qualitatively for the binary refrigerant systems. Thus, all the VLE data of binary refrigerant systems in the database can be predicted through the combination of COSMO-SAC and COSMO-RS models. Next, three binary azeotropic refrigerant mixtures that could be a potential replacement for R410A were identified and their performance verified through simulation of the refrigeration process.