Oak Ridge National Laboratory scientists have developed a method leveraging artificial intelligence to accelerate the identification of environmentally friendly solvents for industrial carbon capture, biomass processing, rechargeable batteries and other applications.
The research targets a class of solvents known for being nontoxic, biodegradable, highly stable, cost-effective, and reusable.
The scientists developed a method to predict solvent viscosity — a key property impacting performance for industrial applications. They compiled nearly 5,000 data points on 672 solvents, evaluated quantum chemical features that guide solvent molecular interactions, and deployed an algorithm called categorical boosting to quickly parse the data and determine the best candidates.
“We reduced computational time and complexity with our approach, while still incorporating all possible molecular interactions,” said ORNL’s Mohan Mood.
ORNL’s Michelle Kidder said, “Interpretable machine learning helps us to design solvents with desired properties for carbon capture by reducing experimental time and cost in the laboratory.”
