Multimodal Temporal Feature-Driven Detection of Overcharge-Induced Thermal Runaway in Lithium-Ion Batteries

Lithium-ion batteries exhibit high energy and power densities; however, safety incidents caused by thermal runaway lead to immeasurable losses, with overcharging-induced thermal runaway representing one of the most critical safety hazards. Although traditional detection methods, such as single-parameter threshold alarms and thermochemical modeling, have been employed in industrial settings, these approaches struggle to detect thermal runaway accurately across batteries of varying capacities. To address this limitation, this study proposes a deep learning detection model based on multimodal time-series data, incorporating battery voltage, current, and surface temperatures from five measurement points as input features. Three critical stages (normal, temperature rise, and thermal runaway) are defined through comprehensive data analysis, and a deep learning model is developed to classify battery status based on a 10 -second time window of input data. Experimental results demonstrate that the proposed model achieves an identification accuracy of 95.94 % on a small-capacity battery dataset, and notably, the model trained on small-capacity batteries achieves a detection accuracy of 98.68 % when applied to larger-capacity batteries. Compared to traditional machine learning algorithms and existing deep learning architectures, the proposed approach exhibits superior performance. Further analysis confirms the model's ability to effectively capture key and critical features indicative of different thermal runaway stages.