94 / 2025-04-14 21:22:27

Advanced Thermal Runaway Prediction and Fault Diagnosis in Lithium-Ion Batteries via Integrated Model and Data-Driven Frameworks

Lithium-ion battery, thermal runaway, data-driven model, high-temperature shock

Thermal runaway (TR) in lithium-ion batteries under extreme high-temperature shock poses significant safety risks in energy storage systems. This study presents a hybrid framework integrating experimental validation with multiphysics modeling and deep learning to predict TR temperatures and enable fault diagnosis. A three-dimensional conjugate heat transfer-TR coupling model is developed, validated through flame shock experiments on NCM523 batteries, achieving a mean absolute percentage error (MAPE) below 7.3%. Key temperature characteristics-onset temperature of self-heating (T₁), TR triggering temperature (T₂), peak surface temperature (T₃), and ignition time (t₁)-are extracted to establish fundamental indicators for fault diagnosis. To address data scarcity under extreme conditions, virtual datasets generated by the model are combined with experimental measurements to train a CNN-BiLSTM-ATTENTION network, achieving temperature predictions with relative errors within 5%. The results highlight the critical role of state of charge (SOC), where higher SOC reduces TR onset temperatures and significantly increases heat release rates (HRR). This framework bridges high-fidelity simulations with real-time monitoring, providing a cost-effective solution for early TR warning and safety management in practical applications.