Retired Battery Screening Based on Markov Transition Field and Swin Transformer

With the widespread popularity of electric vehicles, the secondary utilization of retired batteries is of particular importance. A prerequisite for the secondary utilization of retired batteries is the screening of retired batteries. The ideal health feature should be informative and easy to collect. However, this requires access to the complete charging or discharging process. Features require a priori knowledge and validity is often limited to specific data. To avoid feature engineering, we propose a screening method by converting partial voltage into the image with a Markov transition field and Swin transformer. Firstly, the data from the time-series voltage constant current charging are converted into a Markov transition field image. The image is resized by aggregating and compressing to be compatible with computational efficiency and image information. Then, a Swin transformer is selected to classify the transformed images. Compared to other networks, the Swin transformer offers flexibility in modeling hierarchical structures at all scales due to its unique structure of hierarchical and shift windows. Finally, we conducted cyclic experiments on 143 retired batteries. Comparing different methods and partial voltage, the experimental results demonstrate that the proposed retired battery screening method has high robustness and accuracy.