This paper presents the development of a deep learning-based thermal runaway detection system on the battery production line of the Dutch automobile manufacturer VDL Nedcar. Thermal runaway is a major safety issue that can lead to fire, explosion, and toxic gas emissions, making the development of an automated detection system crucial. Using optical and thermal image data collected from the production line, the research team simulated a baseline (no thermal runaway) and a thermal runaway condition using an external heat source and a smoke generator. Three deep learning models were evaluated: shallow convolutional neural networks, residual neural networks, and vision transformers. The models' ability to detect feature information was analyzed using explainability techniques. The results demonstrate that deep learning is an effective approach for thermal runaway detection on battery production lines.
