Identification of Internal Short Circuit on Cell-Level via Similarity Assessment of Partial Voltage Segments in Adjacent Charging Cycles

Differential comparison between faulty and normal cells is currently the main way to recognize internal short circuit (ISC) in the battery pack, which is prone to generate false alarms for inconsistent cells. In view of this, we innovatively propose a new method of ISC identification based on self-comparison of cell history data. The research data is the cell charging voltage with unified charging protocol after smoothing by wavelet transform method. The selection of research data is accomplished under the dynamic voltage window mechanism to ensure the robustness of the strategy to the different charging habits of the users. Based on the proposed voltage consistency assumption of adjacent charging cycles for one cell, the dynamic time warping algorithm is rationally chosen as the feature parameter extraction method to address the difficulty of similarity assessment for unequal length voltage series, which realizes the accumulation of voltage anomaly deviation. The improved density-based spatial clustering of applications with noise algorithm can realize the adaptive adjustment of hyperparameters according to the data discretization degree, which improves its adaptability to different datasets and avoids the misclassification of slight outlier samples. Based on experimental and real fault data, the final validation and comparison results indicate that the proposed strategy is robust to cell inconsistency and widely applicable to different types of batteries. An average accuracy of 99% and a one-time computation cost of 0.14 s are obtained, and potentially threatening hidden faults in real vehicles can be predicted up to 5 days in advance.