计算机科学
可视化
特征(语言学)
短路
电池(电)
特征提取
数据可视化
人工智能
实时计算
目标检测
模式识别(心理学)
计算机视觉
电压
作者
Jia-Ni Shen,Xiaoyu Chen,Peng-Ju Li,Yijun He
标识
DOI:10.1109/tie.2026.3718133
摘要
Early and reliable online detection of internal short circuit (ISC) is crucial for risk control and safety warning in lithium-ion battery (LIB) systems. Existing data-driven methods usually require massive labeled data for training, and are susceptible to interference from inconsistency and aging. To address these issues, this study proposes a novel training-free visualization analysis method based on multiple dynamic feature collaboration, in which the change rates of internal parameters with the charge-discharge cycle are taken as ISC features. First, the voltage data is decoupled by a multidimensional scaling analysis algorithm, and a 3-D visualization distance matrix of internal resistance, capacity, and state of charge among cells is constructed. Then, the rates of change in the internal parameters between adjacent charge-discharge cycles are extracted as the ISC feature using a rolling strategy. Finally, to simultaneously identify the ISC cycle and locate the ISC cell, the multiple dynamic features are analyzed collaboratively using a median-based outlier algorithm and presented in a visual format. The effectiveness of proposed method is experimentally tested by series-connected ternary LIB pack with different ISC severity, sampling interval, operation temperature, typical operation profile and aging status. The results suggest that the proposed method can facilitate timely detection under various conditions when the ISC resistance up to 900$\boldsymbol{\Omega}$, thereby demonstrating a significant potential to enhance online detection capability of early ISC.
科研通智能强力驱动
Strongly Powered by AbleSci AI