Abstract:Lithium-ion battery failure is the main cause of electric vehicle fire accidents. In this article, we propose a fault analysis framework for Big Data-driven fault trace extraction based on the whole-life-cycle charging data of onboard lithium-ion batteries. First, battery voltage features strongly correlated with faults are mined and automatically selected by a random forest algorithm from the last-one-cycle operation data before sample accidents. Second, by usage of the vector sample points composed of selected features, density clustering is applied to identify faulty cells, and their fault traces in the whole life cycle are tracked through utilizing the Gaussian mixture model. This work uses more than ten real vehicle data for verification. The results show that the proposed method can detect the abnormality of one fault cell at least dozens of cycles in advance, or even in the earliest stage. By classifying traces, this paper also preliminarily proposes a method to distinguish faults caused by battery intrinsic and operative abnormalities, conducive to discriminate accident liability.

A Data-Driven Fault Tracing of Lithium-Ion Batteries in Electric Vehicles