Multimodal Minor Short-Circuit Fault Diagnosis in Batteries via Low-Rank Subspace and Fiber-Optic Sensing

Minor short-circuit faults pose a critical threat to the safety of lithium-ion battery packs, yet their early-stage features are weak and difficult to detect under dynamic operating conditions. To address this challenge, a multimodal, unsupervised fault diagnosis method is proposed by integrating voltage-driven dynamic feature extraction with fiber-optic sensing information. First, the phase space of voltage sequences is reconstructed via time-delay embedding to capture intrinsic dynamic behaviors. A shared low-rank subspace is then established to quantify anomalous deviations among cells. Finally, a physics-consistent fusion mechanism incorporates fiber Bragg grating (FBG)-based strain and temperature signals to enhance anomaly sensitivity and suppress false alarms. Experimental results obtained under various operating conditions and fault severities demonstrate the effectiveness of the proposed method within the tested experimental scenarios, enabling the detection of minor short-circuit faults associated with voltage drops below 0.03 V. Compared with the Isolation Forest method, the proposed approach exhibits improved detection performance and robustness under the evaluated conditions. These findings support the potential of the proposed method for battery health monitoring and safety management.