Evaluating the feasibility of batteries for second-life applications using machine learning.

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Electric vehicle battery retirement presents a challenge for sustainable resource management.
• Determining the optimal fate of retired batteries (second-life use vs. recycling) requires accurate state-of-health assessment.
• Current evaluation methods can be time-consuming and may not capture the full potential for reuse.

Purpose of the Study:

• To develop and validate a machine learning algorithm for the prompt evaluation of retired electric vehicle batteries.
• To differentiate between batteries suitable for second-life applications and those for recycling.
• To provide a data-driven approach for extending battery operational life.

Main Methods:

• Feature generation from battery current and voltage measurements using statistical methods.
• Feature selection and ranking via correlation analysis.
• Gaussian process regression enhanced with bagging for predictive modeling.
• Validation on diverse, publicly available aging datasets (>200 cells).

Main Results:

• The algorithm effectively utilizes simple statistical features from voltage and current data.
• Gaussian process regression with bagging demonstrated high accuracy in performance prediction.
• Achieved low error margins, with worst-case Root Mean Squared Percent Error < 1.48% and Mean Percent Error < 1.29%.
• Validation across varied cell chemistries, charging rates, and operating conditions confirmed robustness.

Conclusions:

• The proposed machine learning approach enables rapid and reliable assessment of retired electric vehicle batteries.
• This method supports informed decisions regarding battery second-life applications, maximizing value and promoting sustainability.
• The findings offer a practical solution for managing the end-of-life phase of electric vehicle batteries.