Regime-Aware Physics-Guided Early Warning of Lithium-Ion Battery Thermal Runaway Using Thermo-Mechanical Signals

Thermal runaway (TR) in lithium-ion batteries is a catastrophic failure mode often preceded by subtle mechanical and electrochemical changes. Traditional battery management systems rely heavily on temperature monitoring, which often triggers warnings too late to prevent disaster. This study investigates whether integrating mechanical signals (force and deformation) with electrical and thermal data can provide a more reliable, earlier warning system for TR under mechanical abuse.

The researchers proposed a regime-aware, physics-guided neural network architecture. The system operates in two stages: first, a lightweight convolutional classifier categorizes the battery's current state into 'safe,' 'warning,' or 'danger' regimes based on mechanical inputs. Second, these regime labels condition a causal temporal convolutional network (TCN) through feature-wise linear modulation (FiLM) and physics-biased attention mechanisms. This joint learning approach allows the model to simultaneously classify the safety regime, detect the onset of thermal runaway, and estimate the remaining time until failure.

The proposed framework was validated using 30 mechanical-abuse experiments across varying states-of-charge (10%, 50%, and 90%). The model achieved an F1 score of 0.89 and a detection success rate of 0.92. Notably, the system provided a mean lead time of 15.6 seconds, which is 69.6% longer than the strongest baseline model. Ablation studies revealed that force measurements are critical, as removing them reduced the warning lead time by over 60%, confirming that mechanical precursors are essential for early detection.

As energy density in lithium-ion batteries increases, the margin for safety errors shrinks. This study demonstrates that moving beyond simple temperature thresholds toward multi-modal, physics-informed sensing can significantly improve the reliability of battery management systems. By identifying mechanical precursors, this approach provides a viable pathway for preventing catastrophic failures in electric vehicles and large-scale energy storage systems.