Artificial intelligence inferred microstructural properties from voltage-capacity curves.

Area of Science:

• Materials Science
• Electrochemistry
• Artificial Intelligence

Background:

• Characterizing lithium-ion battery (LIB) microstructures is crucial for performance optimization but is currently expensive and time-consuming.
• Existing methods for quantifying microstructural properties hinder rapid advancements in battery design, quality control, and degradation tracking.

Purpose of the Study:

• To develop a fast and cost-effective deep learning approach for inferring LIB electrode microstructural properties.
• To utilize readily available cell voltage versus capacity data for microstructure analysis, bypassing traditional slow characterization techniques.

Main Methods:

• A novel framework combining two convolution neural network (CNN) models was developed to predict microstructural properties.
• The CNN models were trained and validated using porous electrode theory-generated voltage versus capacity plots for graphite|LiMn2O4 chemistry.
• The approach infers parameters like Bruggeman's exponent and shape factor from voltage-capacity curves.

Main Results:

• The CNN-based deep learning approach accurately predicted microstructural properties (Bruggeman's exponent and shape factor) with a 0.97 R² score in under 2 seconds per prediction.
• The method successfully distinguished between different particle morphologies, anisotropies, and particle alignments in LIB electrodes.
• The framework demonstrated the ability to accelerate the understanding of processing-property-performance relationships in LIBs.

Conclusions:

• Deep learning, specifically CNNs, offers a powerful and efficient tool for rapid microstructural characterization of LIB electrodes.
• This method significantly reduces the time and cost associated with traditional characterization, enabling faster LIB development and optimization.
• The developed neural network model is adaptable for analyzing both existing and emerging LIB chemistries.