Outlier Detection of Abnormal Cells in a Lithium-Ion Battery System Using a KNN-Based Multi-Feature Framework

The safety and reliability of lithium-ion battery systems are critical for the advancement of the new energy industry. Early detection of cell inconsistencies within a battery pack is paramount to preventing performance degradation and catastrophic failures. This paper proposes a data-driven fault detection framework based on the K-Nearest Neighbors (KNN) algorithm to identify abnormal cells. The methodology’s core contribution lies in the engineering of a comprehensive feature set comprising eight key parameters extracted from three distinct domains: statistical, time-series, and nonlinear. This multi-domain approach creates a rich, high-dimensional health signature for each cell. The framework was validated using a year’s worth of real-world charging data from a 20-cell ternary lithium-ion battery pack installed in an electric two-wheeler. The results demonstrate the successful identification of a distinct outlier cell (Cell #20) and an incipient abnormal cell (Cell #16), confirming the method’s efficacy. The proposed framework is robust, computationally efficient, and exhibits high transferability, making it suitable for a wide range of applications, including electric vehicles (EVs) and large-scale energy storage systems (ESS).