Development of an early-stage thermal runaway detection model for lithium-ion batteries

This paper presents the development of a fast-responding and accurate detection model for early-stage thermal runaway of a lithium-ion battery utilizing acoustics and deep learning paradigms. A series of single-cell battery tests with different state-of-charge and battery orientations is conducted to collect acoustic data. Using data augmentation, 1330 acoustic samples of early-stage thermal runaway are obtained. To facilitate the development of a detection model that can be used in real-life settings, 1128 acoustic samples, including various human activities, are also used. Utilizing 10-s acoustic data as the input and a convolutional neural network model structure as the backbone, excellent model performance is achieved. The overall accuracy is about 93 % with a precision and recall score of about 92 % and 97 %, respectively. Parametric studies are also carried out to evaluate the robustness of the proposed model structure and the effectiveness of the data augmentation methods. In addition, the model performance against two entire tests is assessed using leave-one test-out cross-validation. It is hoped that the proposed work can help to develop a robust detection device that can provide early warning of thermal runaways and allow users to have extra time to mitigate the potential extreme fire hazards and/or to safely evacuate.