Synergistic Fusion of Multimodal Information for Retired Lithium-Ion Batteries Sorting Based on Bidirectional Cross-Attention Mechanism
点击次数：
影响因子：6.3
DOI码：10.1109/TMECH.2026.3708599
教研室：W. Ma, G. Luo, H. Xiao, M. Lin, J. Wu
发表刊物：IEEE/ASME Transactions on Mechatronics
关键字：Battery sorting; echelon utilization; electrochemical impedance spectroscopy (EIS); lithium-ion battery (LIBs); multimodal learning
摘要：The rapid increase of retired lithium-ion batteries has posed critical challenges for sustainable energy management and intelligent recycling systems. Efficient echelon utilization requires accurate sorting strategies to ensure the consistency and reliability of reassembled packs. However, existing methods primarily rely on single-modal charge–discharge data, neglecting the internal electrochemical states that fundamentally influence battery behavior. In this study, we propose a multimodal sorting method that enables synergistic fusion of external dynamic responses and internal electrochemical characteristics for enhanced decision accuracy. Electrochemical impedance spectroscopy (EIS) is introduced as a complementary modality to conventional discharge curves, offering insights into complex internal reactions and electrode interface changes within batteries. EIS signals are encoded into 2-D representations via Gramian angular field to achieve feature augmentation. A dual-branch residual network is designed to accommodate different modal input forms, while a bidirectional cross-attention mechanism is employed to facilitate deep feature interaction across modalities. Comprehensive experiments on retired commercial lithium-ion batteries demonstrate that the proposed model performs excellently across multiple evaluation metrics, attaining an average classification accuracy of 94.00%, which highlights its effectiveness and potential for echelon utilization.
论文类型：期刊论文
学科门类：工学
文献类型：J
页面范围：Early Access
是否译文：否
发表时间：2026-07-15