Cloud-Based Fault Diagnosis of Lithium-Ion Battery Packs by Identifying SOC Abnormal Fluctuations Under Real-World Conditions

As electric vehicles advance, the safety of lithium-ion batteries has emerged as a pivotal concern. Battery safety faults are typically accompanied by abnormal discharge. Addressing this, we design an integrated monitoring algorithm for abnormal fluctuations in battery pack state of charge (SOC), enabling fault diagnosis. This approach integrates neural networks (NNs), primary–secondary filters, data integration, and fuzzy correction to ensure both precision and robustness under real-world operating conditions. The algorithm enables real-time SOC estimation and prediction, while also identifying abnormal discharge through monitoring of unpredictable residuals. The efficacy of our methodology in SOC estimation and prediction is validated under dynamic stress test (DST) conditions and is further corroborated by data collected from 500 vehicles. The results indicate that the maximum residual fluctuation in abnormal samples exceeds that of normal samples by over fourfold. The impact of different alarm thresholds on the true positive rate, false positive rate, and early warning lead time is also discussed. Our method achieves prewarning for certain abnormal samples more than 12 h before fault triggers, offering substantial benefits for safety and economic viability.