Fault diagnosis and abnormality detection of lithium-ion battery packs based on statistical distribution

JOURNAL OF POWER SOURCES

卷:482

Article

Xue, Qiao¹  Li, Guang²  Zhang, Yuanjian³  Shen, Shiquan¹  Chen, Zheng^1,2  Liu, Yonggang⁴ 

[1] Kunming Univ Sci & Technol, Fac Transportat Engn, Kunming 650500, Yunnan, Peoples R China

[2] Queen Mary Univ London, Sch Engn & Mat Sci, London E1 4NS, England

[3] Queens Univ Belfast, Sir William Wright Technol Ctr, Belfast BT9 5BS, Antrim, North Ireland

[4] Chongqing Univ, Sch Automot Engn, State Key Lab Mech Transmiss, Chongqing 400044, Peoples R China

关键词: Electric scooters;    Battery pack;    Fault diagnosis;    Abnormality detection;    Gaussian distribution;   

DOI  :  10.1016/j.jpowsour.2020.228964

来源: Elsevier

【 摘 要 】

Lithium-ion battery packs are widely deployed as power sources in transportation electrification solutions. To ensure safe and reliable operation of battery packs, it is of critical importance to monitor operation status and diagnose the running faults in a timely manner. This study investigates a novel fault diagnosis and abnormality detection method for battery packs of electric scooters based on statistical distribution of operation data that are stored in the cloud monitoring platform. According to the battery current and scooter speed, the operation states of electric scooters are clarified, and the diagnosis coefficient is determined based on the Gaussian distribution to highlight the parameter variation in each state. On this basis, the K-means clustering algorithm, the Z-score method and 3 sigma screening approach are exploited to detect and locate the abnormal cells. By analyzing the abnormalities hidden beneath the external measurement and calculating the fault frequency of each cell in pack, the proposed algorithm can identify the faulty type and locate the faulty cell in a timely manner. Experimental results validate that the proposed method can accurately diagnose faults and monitor the status of battery packs. This theoretical study with practical implications shows the promising research direction of combining data mining technologies with machine learning methods for fault diagnosis and safety management of complex dynamical systems.

【 授权许可】

Free