Estamos realizando la búsqueda. Por favor, espere...


Machine learning for predicting the solubility of high-GWP fluorinated refrigerants in ionic liquids

Abstract: The development of technology to reduce the environmental impact of fluorinated refrigerant gases (F-gases) is currently of outmost importance. The capture of F-gases in ionic liquids (ILs) is envisaged as solution to avoid emissions of F-gases to the atmosphere, and many studies have been devoted to the experimental determination of the vapor-liquid equilibrium of F-gas/IL mixtures. However, this is an expensive and time-consuming task, so finding prescreening options that can reduce the experimental load would pose a significant advantage in the development of new industrial-scale processes. Here, we develop a prescreening tool based on the use of artificial neural networks (ANNs) to predict the solubility of F-gases in ILs from easily accessible properties of the pure compounds, such as the critical properties of the gases or the molar mass and volume of the IL. We have used the UC-RAIL database with more than 4300 solubility data of 24 F-gases in 52 ILs. The ANN resulting from this study is capable to predict the fed dataset with an average absolute relative deviation (AARD) and mean absolute error (MAE) of 10.93% and 0.014, respectively, and we further demonstrate its predictive capabilities showing the very accurate prediction of a system including R-1243zf, an F-gas that was not present in the training set because it had not been previously studied. Finally, the developed ANN is implemented in an easy-to-use spreadsheet that will allow to extend its use in the prescreening of ILs towards the abatement and recovery of high environmental impact refrigerant gases.

 Autoría: Asensio-Delgado S., Pardo F., Zarca G., Urtiaga A.,

 Fuente: Journal of Molecular Liquids, 2022, 367, Part B, 120472

Editorial: Elsevier Science

 Fecha de publicación: 01/12/2022

Nº de páginas: 9

Tipo de publicación: Artículo de Revista

 DOI: 10.1016/j.molliq.2022.120472

ISSN: 1873-3166,0167-7322

Proyecto español: PID2019-105827RB-I00

Url de la publicación: https://doi.org/10.1016/j.molliq.2022.120472