Fault diagnosis method for lithium-ion battery cells based on information granularity and dynamic segmentation slope

With the development of new energy technologies, electric vehicles are becoming increasingly popular. Early battery fault detection is crucial for ensuring personal safety and minimizing property damage. However, traditional fault diagnosis methods often have difficulty detecting early-stage faults. To address this challenge, in this study, we propose a novel electric-vehicle battery fault diagnosis method that integrates information granularity and the segmented-slope feature. First, the original voltage data are denoised using the locally weighted scatterplot regression algorithm. Second, segmented-slope features are extracted from the voltage signal to enhance the distinction between healthy and faulty cells. Next, information granularity theory is applied to select the reference cell. The quality of each individual cell’s feature curve was evaluated based on its granularity value, and the mean granularity of the selected cells was then used to construct the reference feature cell. Finally, a dynamic threshold model, constructed based on the Manhattan distance and three-sigma criterion, provides an adaptive threshold adjustment mechanism for early warning and fault location. In the current tests involving four labeled vehicles (one normal and three with internal short circuit faults), the accuracy of this method was very high.