Discovering new Refrigerants with Machine Learning

Refrigeration and cooling accounts for about 7.5% of global carbon emissions and 30% of that is attributed to the direct emission of leaked refrigerant fluids. Refrigerant molecules are potent green-house gases with Global Warming Potentials (GWPs) on the scale of 1,000-10,000 kg of CO2-Equivalent/kg of refrigerant. As such, there is a need for novel refrigerant discovery. However, there are challenges to ensuring the safety and environmental properties of candidate molecules. This research project focuses on the quantitative prediction of refrigerant flammability using cheap explainable ML models.