Deep Learning-Based Screening of Retired Lithium-ion Batteries for Sustainable Lifecycle Management Hefei University of Technology

Format:

Book

Conference/Event

Author/Creator:

Xiao, Hualong, author.

Contributor:

Lin, Mingqiang

Luo, Gang

Wang, Li

Wu, Ji

Conference Name:

SAE 2025 International Conference on Battery Safety and Reliability (2025-10-23 : Shanghai, China)

Language:

English

Physical Description:

1 online resource cm

Place of Publication:

Warrendale, PA SAE International 2026

Summary:

Ensuring safety and consistent quality in lithium-ion battery manufacturing is essential for the reliable operation of electric vehicles and energy storage systems. Strict quality control measures during production not only enhance product safety but also reduce the number of defective units entering post-market recycling streams. However, variations in battery quality remain inevitable, making efficient downstream sorting an important complement to upstream manufacturing control. Efficient sorting of retired lithium-ion batteries is critical for battery second-life utilization and circular economy development. Based on 750 commercially recycled retired batteries, this study proposes a 1D CNN-Transformer hybrid deep learning framework for automatic screening of retired batteries. The framework first employs a 1D convolutional neural network to extract local features from timevoltage sequences and compress sequence length, followed by a Transformer encoder to capture global discriminative features during the charging process. Subsequently, a two-layer multilayer perceptron classifier produces the category predictions. Experimental results show that the proposed method achieves a classification accuracy of 95.33%, significantly outperforming conventional approaches. Further analysis reveals that the 1D CNN module improves accuracy by approximately 4% by providing efficient feature inputs for global modeling; charging data, compared to discharging data, offer richer information, boosting accuracy by 16.67%; incorporating temporal information under non-uniform sampling enhances time-series modeling effectiveness, yielding a 2.67% accuracy gain; and using only the first 412 minutes of charging data can still achieve 92.67% accuracy, indicating that the early charging phase carries high discriminative value. This study provides an effective technical solution for sorting retired batteries and offers valuable insights for advancing the battery recycling industry

Notes:

Vendor supplied data

Publisher Number:

2026-01-7007

Access Restriction:

Restricted for use by site license