Active learning accelerates electrolyte solvent screening for anode-free lithium metal batteries.

Area of Science:

• Materials Science
• Electrochemistry
• Machine Learning

Background:

• Anode-free lithium metal batteries promise higher energy density but face challenges in electrolyte development for extended cycle life.
• Current electrolyte development relies heavily on inefficient trial-and-error methods due to a lack of universal design principles.

Purpose of the Study:

• To demonstrate active learning as a method to accelerate electrolyte discovery for anode-free lithium metal batteries.
• To overcome data-scarce and noisy label limitations in electrolyte development.

Main Methods:

• Implemented a sequential Bayesian experimental design with Bayesian model averaging active learning framework.
• Integrated experimental feedback to iteratively refine predictions using capacity retention in Cu||LiFePO4 cells as the target property.
• Explored a virtual search space of 1 million electrolytes starting with 58 data points.

Main Results:

• Identified four distinct electrolyte solvents that rival state-of-the-art electrolytes in performance after seven active learning campaigns.
• The active learning framework rapidly converged on optimal electrolyte candidates.
• Demonstrated efficient navigation of a large electrolyte chemical space.

Conclusions:

• Active learning offers a promising alternative to traditional methods for discovering electrolytes for next-generation batteries.
• This approach significantly accelerates the development cycle for advanced battery materials.
• Highlights the potential of AI-driven strategies in materials science research.