DETECTION OF THERMAL RUNAWAY IN EV BATTERY PACK USING MULTI-MODAL SENSOR FUSION (THERMAL IMAGING, GAS & CURRENT) AND NEURAL NETWORK

Abstract: In electric vehicles (EVs), battery safety is very important, and thermal runaway (TR) is due to multi-physics failure mechanism that begins with electrical failure followed by chemical reactions. Traditional battery management systems only detect TR after there is a significant rise in temperature. This article proposes a novel multi-modal monitoring system of the safety of the battery, which detects thermal runaway precursors in EVs using thermal, gas and current sensing. The thermal image is segmented into hot spots using a U-Net Convolutional Neural Network (CNN) which is a type of semantic segmentation network. Gases are sensed by MQ-135 and current is sensed using ACS712 Hall Effect current sensors. A novel Fusion Index is introduced as the weighted sum of the thermal, gas and current anomaly score. The Fusion Index also determines when to cool the battery and isolate the battery cell using a MOSFET. The efficacy of the proposed multi-modal battery safety monitoring system was validated using a MATLAB simulation, which was an electro-thermal digital twin. The simulation demonstrated that thermal runaway was detected 30 to 90 seconds earlier using the proposed system compared to the existing thermal/gas systems. The multi-modal battery safety monitoring system has successfully delivered a significant improvement over traditional thermal/gas battery management systems.

Keywords: Thermal runaway, Electrolyte Decomposition, Convolutional Neural Networks, U-Net, Semantic segmentation, Active cooling, Anomaly, Battery pack Isolation, Fusion Index.