Multi-model collaborative early warning for battery systems with integrated fiber optic sensors

Abstract The safety assessment of battery systems is constrained by sparse temperature measurements and monolithic diagnostic frameworks in conventional battery management systems (BMS). To address this, we propose a multi-model collaborative thermal-runaway early-warning framework using distributed temperature sensing. An integrated sensing scheme increases temperature sensing points within the battery pack from 32 to 540 with minimal wiring complexity and is validated in real-world vehicle operation. We establish a functional relationship between temperature Shannon entropy (SE) and average temperature variation to derive a dynamic warning boundary incorporating error tolerance. Integrating this boundary with thresholds for temperature, temperature change rate, and SE change rate enables collaborative early warning. Compared with conventional BMS warning benchmarks, the framework provides maximum warning lead times of 706 s in laboratory modules and 554 s in real-world vehicle operation. These findings demonstrate the potential of distributed sensing and collaborative diagnostics for proactive thermal-runaway risk identification.