Machine Learning Screening of Metal-Ion Battery Electrode Materials.

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• Rechargeable batteries are vital for renewable energy, electronics, and electric vehicles.
• Key electrode material properties include voltage, capacity, and volume change during cycling.
• Predicting these properties is crucial for designing efficient and stable batteries.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting average voltage and volume change of metal-ion battery electrode materials.
• To screen for novel electrode materials, specifically for sodium-ion batteries, with desirable energy density and stability.
• To assess the predictive power of ML models for materials beyond the initial training dataset.

Main Methods:

• Deep neural network regression models were trained using data from the Materials Project database.
• 10-fold cross-validation and independent test sets were used to evaluate model performance.
• ML models were applied to predict properties of hypothetical sodium-ion electrode materials.

Main Results:

• The ML models demonstrated good predictive accuracy for average voltage and volume change.
• Screening identified 22 promising sodium-ion electrode materials with high predicted energy density and low volume variation.
• ML predictions were validated against quantum-mechanics calculations, showing good agreement.

Conclusions:

• Machine learning is a powerful tool for accelerating the discovery of advanced battery electrode materials.
• The developed ML models can effectively screen for materials with optimized electrochemical performance and stability.
• This approach facilitates the exploration of new materials for next-generation energy storage solutions, particularly sodium-ion batteries.