A new artificial intelligence framework developed at Cornell can accurately predict the performance of battery electrolytes while revealing the chemical principles that govern them, providing engineers with a new tool for designing better batteries.
The framework, published Feb. 19 in Nature Computational Science, focuses on high-performing lithium-ion batteries that use nonaqueous electrolytes - liquid or gel-like materials that enable higher energy storage. It uses the power of AI to predict how salts, solvents and operating conditions work together to enable ion transport.
"Battery chemistry involves many coupled variables, and understanding how they interact is essential for rational design," said co-author Fengqi You, the Roxanne E. and Michael J. Zak Professor in Energy Systems Engineering at the Cornell Duffield College of Engineering. "We are developing AI tools that improve prediction while also providing insight into the underlying chemistry."
Conventional AI frameworks treat electrolyte formulations as uniform collections of input variables and learn statistical correlations between those inputs and properties such as conductivity. Those AI systems are optimized to predict outcomes, but not to explain the chemistry behind the electrolytes.
To address this challenge, You and Zhilong Wang, a postdoctoral researcher who is first author of the study, developed a dynamic modeling framework that treats salts, solvents and operating conditions as distinct but interacting contributors to conductivity. Rather than funneling all information through a single model that produces outputs without revealing its internal logic, the framework processes chemically meaningful descriptors for those three components separately, and then adaptively integrates them.
When applied to a large experimental dataset of lithium-ion electrolytes, the framework reduced prediction error by more than 65% compared with leading machine-learning methods. And unlike conventional models, it remained accurate across the full conductivity range, including rare high-conductivity formulations that are most relevant for next-generation batteries.
"For energy materials, it is not enough to rely on black-box predictions," said You, who is also a senior faculty fellow at the Cornell Atkinson Center for Sustainability. "Interpretability and integration with physics are critical for building reliable and scalable design tools."
The work reflects ongoing efforts within the Cornell AI4S Initiative to bring together faculty and students from across campus to apply AI to energy, materials and environmental challenges. It also complements a recent Science Advances article by You and Wang that outlines a broader framework for integrating AI into solid-state battery research through a coordinated framework combining machine learning, simulations and experimental feedback.
"These two studies address both an overarching framework and a specific modeling approach, reflecting our effort to connect strategy with practical implementation in AI-driven battery research," You said.
The studies were supported in part by the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship, a program of Schmidt Sciences.
Syl Kacapyr is associate director of marketing and communications for Duffield Engineering.
