Enhancing the Reliability of Fault Diagnosis in Lithium-Ion Batteries Using a Hybrid Model-Based and Machine-Learning Approach

Lithium-ion batteries (LIB) are essential for electric vehicles (EV) and renewable energy storage, and their safe, reliable, and efficient operation is critical. As their use grows, robust fault diagnosis becomes key to maintaining availability and performance. A major challenge is that many battery faults exhibit similar patterns, making it difficult to distinguish and isolate specific fault types. This paper presents a hybrid approach that combines model-based methods with data-driven techniques to improve the safety and reliability of LIB systems. The proposed diagnostic architecture uses structural analysis (SA) to identify analytical redundancy for fault detection and isolation (FDI), while Extended Kalman Filter (EKF) algorithms generate residuals that capture inconsistencies between model predictions and measurements. These residuals are processed into statistical features and passed to a machine learning (ML) classifier for accurate fault detection and classification. In this series-hybrid structure, using SA–EKF residuals as classifier inputs significantly improves diagnostic performance compared to ML-only features, and among the tested classifiers, Residual-based RF achieves the highest accuracy. The method achieves 98.73% accuracy in distinguishing fault types, demonstrating high sensitivity across all faults and robustness to varying noise levels. Overall, the proposed hybrid methodology enhances fault diagnosis effectiveness and supports the safe operation of LIB systems, contributing to the transition toward clean and sustainable energy solutions.