Real-Time Thermal Runaway Detection of a Lithium Battery Using X-Ray Video Frames

This letter addresses the problem of lithium battery thermal runaway to make accurate detection performance via contactless monitoring device, such as X-ray video stream. The problem can be formulated as state estimation by utilizing Kalman filter (KF) algorithms to estimate the evolution of individual pixels over time. In principle, estimating the motion of all pixels from moment to moment requires a high computational time; therefore, the compressed-sensing technique, such as dynamic mode decomposition (DMD), was implemented to create reduced pixel space because such technique were useful in terms of reducing the state vector to a few dominant modes to be estimated at each time step. The estimated dominant modes were employed to detect thermal runaway. In addition, the proposed approach was compared to machine learning methods.