A physics-guided multi-source collaborative fault diagnosis framework for real-world lithium-ion batteries

Hits:

Impact Factor:10.7

DOI number:10.1016/j.est.2026.124389

Journal:Journal of Energy Storage

Key Words:Lithium-ion battery; Fault diagnosis; Physics-guided; Multi-source fusion; Domain adaptation

Abstract:Precise fault diagnosis for real-world lithium-ion batteries is critical to guaranteeing safe and reliable operation. To overcome the difficulties in early fault recognition, unreliable assessment, and limited generalization capability of existing methods, we propose a physics-guided multi-source collaborative diagnostic framework for lithium-ion batteries. First, we design a physics-guided cell-level healthy reference voltage reconstruction model to address the scarcity of fault samples and to provide a foundation for subsequent feature generation. Simultaneously, the model decouples time-varying implicit electrochemical parameters that reflect the health evolution of the battery pack. Then, we establish a physics residual driven dual-view representation enhancement strategy to improve the feature separability of different faults. Furthermore, we integrate the residual representations, implicit parameters, and statistical features through an adaptive gating mechanism to obtain a more comprehensive fault representation. At last, a cross-scenario multi-source diagnostic scheme is constructed to achieve highly accurate cross-scenario diagnosis with low-cost fine-tuning. Compared with existing algorithms, the proposed method demonstrates superior performance (with the Macro-F1 score consistently between 97.8% and 98.9%). Moreover, extensive experiments on four real-world electric vehicle datasets thoroughly validate its effectiveness and practical applicability, contributing to enhanced active safety capabilities in battery operation.

Note:中科院2区Top

Co-author:Li Wang,Lei Bao,Qilong Xie,Zhannan Wang

First Author:Muxing Li

Indexed by:Essay collection

Correspondence Author:Muyao Wu

Document Code:124389

Discipline:Engineering

Document Type:J

Volume:181

ISSN No.:2352-152X

Translation or Not:no

Date of Publication:2026-08-22

Included Journals:SCI、EI

Links to published journals:https://www.sciencedirect.com/science/article/pii/S2352152X26040533