(555e) A Combined Machine Learning and Thermodynamic Modeling Approach for Designing Low Melting Molten Salt Eutectics

Molten salts based eutectic mixtures are ideal candidates for battery electrolytes and solar energy storage applications due their stability, thermophysical properties, and low toxicity. However, the high melting temperatures of these mixtures limit their practical applicability due to the costs involved in operating these systems. Hence, there is a necessity to design molten salts eutectics that melt at lower temperatures compared to the existing mixtures. Here we present a high throughput computational method combing thermodynamic modeling and machine learning to design novel eutectics of molten salts with target melting temperature. We develop and validate this combined model using a database of experimental eutectic melting temperatures. We show that the approach has higher accuracy than the thermodynamic or the machine learning approach applied separately. Using this hybrid model, we perform high throughput screening of million of molten salts from binary to hexanary mixtures and devise the design rules that lead to low melting eutectics. The design rules are identified using dimensionality reduction and clustering techniques to derive interpretations of the modeling predictions. Finally, we identify several novel low melting eutectic mixtures based on the computational approach.