Machine Learning-Based Lifetime Prediction of Lithium-Ion Cells.

Area of Science:

• Materials Science
• Electrochemistry
• Computer Science

Background:

• Precise lifetime predictions for lithium-ion cells are essential for advancing electric vehicle technology and sustainable mobility.
• Current limitations in predicting cell degradation stem from complex interactions influenced by design, operation, and storage conditions.

Purpose of the Study:

• To develop a machine learning framework for accurate and interpretable lithium-ion cell lifetime prediction.
• To overcome limitations of existing methods by utilizing physically interpretable models derived from aging data.

Main Methods:

• A machine learning framework based on symbolic regression via genetic programming was developed.
• The evolutionary algorithm inferred physically interpretable models from cell aging data without requiring prior domain knowledge.
• The approach was validated through case studies using cycle and calendar aging data from 104 automotive lithium-ion pouch cells.

Main Results:

• Predictive accuracy for extrapolations over storage time and energy throughput improved by 38% and 13%, respectively.
• Error reductions of up to 77% were achieved for predictions involving other stress factors.
• The generated aging models demonstrated applicability, generalizability, and interpretability.

Conclusions:

• Evolutionary algorithms offer a powerful approach to enhance the accuracy and interpretability of lithium-ion cell aging predictions.
• This novel method holds significant potential for improving battery development and accelerating the transition to electric mobility.