汽车工程 ›› 2023, Vol. 45 ›› Issue (9): 1677-1687.doi: 10.19562/j.chinasae.qcgc.2023.09.016

所属专题: 新能源汽车技术-动力电池&燃料电池2023年

• • 上一篇    下一篇

云端数据驱动的锂电池故障分级预警研究

郭文超1,杨林1(),邓忠伟2,李济霖1,范志先3   

  1. 1.上海交通大学机械与工程学院,上海 200240
    2.电子科技大学机械与电气工程学院,成都 611731
    3.中通客车股份有限公司,聊城 252000
  • 收稿日期:2023-04-07 修回日期:2023-06-03 出版日期:2023-09-25 发布日期:2023-09-23
  • 通讯作者: 杨林 E-mail:yanglin@sjtu.edu.cn
  • 基金资助:
    国家自然科学基金(51875339);国家重点研发计划(2022YFE0102700);国家新能源汽车监测管理中心西南分中心和新能源汽车事故调查协作网资助

Research on Multi-level Fault Warning Method for Lithium-ion Batteries Driven by Cloud Data

Wenchao Guo1,Lin Yang1(),Zhongwei Deng2,Jilin Li1,Zhixian Fan3   

  1. 1.School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240
    2.School of Mechanical and Electrical Engineering,University of Electronic Science and Technology of China,Chengdu 611731
    3.Zhongtong Bus Co. ,Ltd. ,Liaocheng 252000
  • Received:2023-04-07 Revised:2023-06-03 Online:2023-09-25 Published:2023-09-23
  • Contact: Lin Yang E-mail:yanglin@sjtu.edu.cn

摘要:

目前还未有一种有效手段针对故障类型未知的车辆云端数据进行无监督式的故障预警,为此本文提出了一种云端数据驱动的锂电池故障分级预警方法。首先通过机理分析选取适用于云端数据特性的特征,构建6类差熵特征集进行多次混合聚类实现对电池健康度的打分评价。通过引入温度信息区分热相关故障并构建预警等级划分准则判断电池故障状态。利用5种现场故障案例进行验证,结果表明,该方法能准确识别故障并区分故障类型,且具有较高的超前性和适应性。

关键词: 锂电池, 无监督式故障预警, 差熵特征集, 预警等级划分

Abstract:

At present, there is no effective method for unsupervised fault warning for vehicle cloud data with unspecified fault types. Therefore, this paper proposes a multi-level fault warning method for lithium-ion batteries driven by cloud data. Firstly, the features suitable for the characteristics of cloud data are selected through mechanism analysis, and six types of differential entropy feature sets are constructed for multiple mixed clustering to achieve the score evaluation of battery health. Then, temperature information is introduced in to distinguish heat-related faults and the warning level division criteria are constructed to determine the battery fault status. Finally, five field failure cases are used for validation. The results show that the method can accurately identify faults and distinguish fault types, and is ahead of its time and highly adaptable.

Key words: lithium-ion battery, unsupervised fault warning, differential entropy feature sets, warning level division