Multiscale dynamic construction for abnormality detection and localization of Li-ion batteries

The safety of the electric vehicle is highly dependent on the safe operation of the battery system, where the thermal behavior becomes important since any abnormality could be an early sign of potential damage. Some fault diagnosis methods have been proposed for abnormality detection and health management of battery systems. However, the existing approaches heavily depend on the governing equations or many sensors for data collection, which limits the applications of these methods in practice. In this paper, a multiscale dynamic analysis method is proposed to detect and localize thermal abnormalities occurring in battery cells using fewer sensors. First, a two-dimensional spatial construction is designed to model the temperature field in the battery cell under sparse sensing. The abnormal detection is performed separately on spatial and temporal scales, and then integrated into a more comprehensive detection criterion. After probabilistic processing under kernel density estimation, the final decision can be made more reliably in practical situations with stochastic uncertainty. Furthermore, the location of the abnormality is figured out with the constructed spatial basis functions. Simulations and experiments are carried out on pouch-type Li-ion battery cells, demonstrating that the proposed method can effectively detect and locate internal short circuit (ISC) abnormalities before they develop into thermal runaway. This study makes the first effort to consider both spatial and temporal information for designing the detection index in battery systems. The optimality of the proposed spatial construction method is proved by theoretical analysis.