Interpretable Learning of Accelerated Aging in Lithium Metal Batteries

Area of Science:

• Materials Science
• Electrochemistry
• Machine Learning

Background:

• Lithium metal batteries (LMBs) offer high energy density crucial for electric transportation.
• Rapid capacity decay and safety concerns hinder practical LMB application due to complex degradation patterns.

Purpose of the Study:

• To develop an interpretable machine learning framework for predicting accelerated aging in LMBs.
• To identify key factors influencing LMB degradation and propose strategies for life extension.

Main Methods:

• Utilized a comprehensive dataset of 79 LMB cells with diverse chemistries and parameters.
• Employed machine learning on early-cycle (first 10 cycles) data to predict aging knee points.
• Analyzed the impact of discharge depth on LMB aging rates.

Main Results:

• The framework accurately predicted aging acceleration points using only early-cycle data.
• Identified the critical role of the last 10% of discharge depth in LMB aging.
• Proposed a universal descriptor for rapid electrolyte evaluation based on early electrochemical data.
• Developed a dual-cutoff discharge protocol extending LMB cycle life up to 2.8 times.

Conclusions:

• Interpretable machine learning can effectively predict and understand LMB degradation.
• Early-cycle data provides valuable insights into battery aging mechanisms.
• Optimized discharge protocols significantly enhance the cycle life of lithium metal batteries.