Lithium-ion battery remaining useful life prediction using a two-phase degradation model with a dynamic change point

An accurate remaining useful life (RUL) prediction plays a crucial role in the prognostics and health management of lithium-ion (Li-ion) batteries. Current studies on the RUL prediction of Li-ion batteries commonly use single-phase degradation models, which result in inaccurate RUL predictions due to their insufficient capabilities in capturing various degradation patterns. The existing two-phase degradation models can divide battery degra-dation into two phases using a change point, a slowly decreasing phase, and a rapidly decreasing phase. The change point in the current two-phase degradation models is usually modeled in two ways. First, the change point is treated as a random variable and that however greatly increases the computational complexity. Second, a fixed change point is assigned for all battery cells for model simplification, which may not be realistic in practice. For example, battery cells' degradation data collected from our laboratory tests show a two-phase degradation pattern with different change points. By considering such differences in change points, this study first utilizes binary segmentation to identify the change point of a battery cell and then proposes a two-phase capacity degradation model with a dynamic change point. Further, variations have been observed in the degradation behaviors of tested battery cells. Therefore, by using the proposed two-phase degradation model, we develop a particle filtering-based framework considering uncertainties to predict the RULs of battery cells. Finally, the proposed framework shows superior prediction performance compared with the existing degradation models by providing the RUL prediction with an average absolute estimation error percentage of 27 % for laboratory data and an average absolute estimation error percentage of 24 % for NASA battery data.