Multi-scale Computer-aided molecular design of Ionic liquid for absorption heat transformer based on Machine learning

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

9 Scopus Citations
View graph of relations

Related Research Unit(s)


Original languageEnglish
Article number115617
Journal / PublicationEnergy Conversion and Management
Online published28 Apr 2022
Publication statusPublished - 1 Jun 2022


Absorption cycles have attracted considerable attention for utilizing renewable energy and waste heat to achieve carbon neutrality. Absorption Heat Transformers (AHTs) using Ionic Liquids (ILs) as absorbents show great potential in avoiding crystallization and corrosion problems. To screen the optimum candidate for AHT systems, a multi-scale computational screening method by Computer-aid Molecular Design (CAMD) is applied, and its feasibility is verified with high accuracies. For 18 typical operating conditions, the optimum IL molecules are identified from 26,360 candidates generated by the CAMD method. The relationship between the AHT cycle performance and IL molecular structures is systematically analyzed to clarify the significant IL structural factors that affect the Coefficient of Performance (COP) primarily. The correlation analysis shows that IL molecules containing main-cations ([C1N], [C1P], or [C1S]), anions ([Br], [Cl], [Lac], or [OAC]), and sub-cations with short alkyls have the highest potential to improve the cycle performance. The optimum ILs screened by the multi-scale CAMD method perform better than the currently investigated ILs, even outperforming LiBr in some specific unfavorable conditions. At an absorber temperature of 120 °C, heat source temperature of 75 °C, and condenser temperature of 25 °C, [C,1,1,1,1,OHN][OAC] yields a COP of 0.458, much higher than LiBr (0.419). By distinguishing the applicable and inapplicable operating conditions, the comparison shows that the applicable operating ranges of the selected ILs are wider than that of LiBr due to their crystallization-free nature. To accelerate the molecular design process in various operating conditions, eight Machine Learning (ML) algorithms are integrated to predict the cycle performance based on molecular descriptors with less computation cost. To verify the accuracy of the proposed multi-scale CAMD approach, experimental results of [C1,1,2,2,OHN][Br], one of the best-performing ILs, are compared with the simulation results of the multi-scale CAMD. The agreement between experiment and simulation illustrates the high accuracy of this screening method. The multi-scale CAMD approach achieves the high-throughput computational screening of optimum ILs for high-efficiency absorption systems.

Research Area(s)

  • Absorption heat transformer, Computer-aided molecular design, High-throughput computational screening, Ionic liquid, Machine learning