A fast screening framework for second-life batteries based on an improved bisecting K-means algorithm combined with fast pulse test. (October 2020)

• Record Type: Journal Article
• Authors: Zhou, Zihao; Ran, Aihua; Chen, Shuxiao; Zhang, Xuan; Wei, Guodan; Li, Baohua; Kang, Feiyu; Zhou, Xiang; Sun, Hongbin
• Is Part Of: Journal of energy storage. Volume 31(2020)
• Publication Date: 2020-10
• Subjects: Retired lithium-ion batteries -- Secondary usage -- Pulse test -- Clustering method -- Unsupervised learning; Energy storage -- Periodicals; Energy storage -- Research -- Periodicals; 621.3126
• DOI: 10.1016/j.est.2020.101739
• Languages: English
• ISSNs: 2352-152X

Abstract

Highlights: This paper develops a fast screening algorithm of a canopy based bisecting K-means integrated with a fast pulse test for second life battery. A systematic study is carried out to validate the effectiveness of the fast clustering, including self-tested 136 pulse data set. The fast pulse signals using data-driven algorithm are correlated with the battery electrochemical states from a set of 54 batteries. Two open battery data sets from NASA and Oxford are used to demonstrate the effectiveness and accuracy of algorithm. Abstract: Lithium-ion batteries with high energy density have been widely used in energy storages and electrical vehicles. After retiring, they usually contain 70%-80% of their primary capacity and can still be reused for secondary applications. However, the most essential problem before such secondary usage is how to classify large amounts of retired batteries into subgroups effectively. In this paper, the retired battery screening is treated as an unsupervised clustering problem, and a fast pulse test integrated with an improved bisecting K-means algorithm has been applied to reduce the feature generation time from hours to minutes. The improved bisecting K-means algorithm generates almost the same clustering results for two groups of features: benchmark features including voltage (U), resistance (R) and capacity (Q) from conventional charge-discharge tests (~5 h), and new features from fast pulse tests (~2 mins). Thus, the proposed fast pulse test integrated with the improved bisecting K-means algorithm can realize fast clustering of retired lithium-ion batteries. Finally, two open lithium-ion battery data sets from NASA and Oxford are used to demonstrate the effectiveness and accuracy of the proposed learning-based framework.