Battery internal short circuit detection based on curvilinear Euclidean distance assessment and adaptive clustering method analysis in multi-factor coupling scenarios

Based on the difference of reference objects, the isolation way of internal short circuit (ISC) can be divided into two types, i.e., vertical inter-comparison of different cells and horizontal self-comparison of one cell. These two methods have low reliability due to cell inconsistency and external conditions, respectively. Therefore, we innovatively propose a battery ISC detection method via curvilinear Euclidean distance assessment and adaptive clustering method analysis, which is still able to accurately identify the fault cell in multi-factor coupling scenarios. First, by means of the variable voltage window mechanism, the partial cell charging voltage curves (CVCs) smoothed by the wavelet transform method are selected as reliable research object, enhancing the adaptability of the strategy to user charging habits. Then, based on the proposed hypothesis of uniform cell CVCs for adjacent cycles, the curvilinear Euclidean distance method, which can achieve the accumulation of anomaly risks, is used to evaluate the similarity between voltage segments. Further, by comparing the similarity indicators of different cells, ISC anomalies can be amplified to a significant level. The two-level comparison strategy ensures the reliability of the extracted features. Finally, an adaptive DBSCAN algorithm that can adapt to different datasets has been implemented for automatic diagnosis of ISC cells. Validation and comparison results based on experimental data and real fault sample demonstrate that the above strategy exhibits extremely high robustness to cell consistency, ambient temperature, and charging mode, without requiring complex feature engineering or model training. Ultimately, a 0 % false alarm and missed alarm rate are achieved, with a single computation cost of only 0.21 s.