Real-world fault diagnosis of electric vehicle Lithium-ion batteries based on risk accumulation representation

AbstractWith the large-scale deployment of electric vehicles, battery safety risks have become one of the most challenging technical issues in the context of the energy transition. The frequent occurrence of thermal runaway events highlights the inherent deficiencies of existing state-based diagnostic methods that depend on real-time battery management system parameters. Under complex operating conditions, such methods fail to adequately represent risk accumulation, provide sufficient observability of internal states, or achieve reliable predictive generalization. Furthermore, the lack of a unified risk representation that can describe battery state evolution across vehicle models and operating conditions is a critical reason hindering the industry from establishing a highly reliable early-warning system. Motivated by this understanding, this study leverages large-scale real-world vehicle operational data to develop a novel mechanism-informed fault-diagnosis framework, which adopts risk accumulation as the core representation. By elucidating the coupled roles of temperature, state of charge, and charge/discharge rate in degradation and thermal runaway triggering, risk accumulation indicators are extracted, and a hybrid model integrating a comprehensive risk score with a deep neural network is developed to quantitatively map risk accumulation to thermal runaway risk. The framework achieves validation accuracies of 100%, 95.45%, and 95.45% on three real vehicle models, respectively, demonstrating strong cross-model robustness. Interpretability analyses indicate that the model can capture dominant risk-prone operating conditions and key usage patterns for different vehicle models, providing a new theoretical basis and research pathway for full life-cycle safety assessment and systematic optimization.