Yangang Liang1,Juran Noh1,Hieu Doan2,Heather Job1,Lily Robertson2,Lu Zhang2,Rajeev Surendran Assary2,Karl Mueller1,Vijayakumar Murugesan1
Pacific Northwest National Laboratory1,Argonne National Laboratory2
Yangang Liang1,Juran Noh1,Hieu Doan2,Heather Job1,Lily Robertson2,Lu Zhang2,Rajeev Surendran Assary2,Karl Mueller1,Vijayakumar Murugesan1
Pacific Northwest National Laboratory1,Argonne National Laboratory2
Solubility is an essential physiochemical property of redox-active molecules and closely associated with the energy density of redox flow batteries. However, the lack of large-scale and high-fidelity experimental solubility databases limits the potential application of data-driven approaches to accelerate electrolyte materials discovery. Here, we describe a highly efficient and robust automated workflow that combines a high-throughput experimentation platform with active learning to optimize the solubility of redox active molecules in organic solvent systems. We successfully demonstrate this platform by identifying several solvents that yield high solubility (> 6 M) of a model redox active molecule, 2,1,3-benzothiadiazole, among 2,101 solvent candidates. More significantly, our combined approach only requires solubility measurement of less than 10% (~200) of the solvent candidates. Our findings indicate that binary solvent mixtures based on 1,4-dioxane significantly enhance the solubility of 2,1,3-benzothiadiazole, providing promising prospects for high-energy density redox flow battery applications. While this work focuses on flow battery solvent formulation, our AI-guided high-throughput robotic platform can be used as a versatile tool for discovering energy materials with desired properties in a cost-effective and efficient manner.