An Incipient Multifault Diagnosis Method for Lithium-Ion Battery Pack Based on Data-Driven With Incremental-Scale
Yue Wang, Yunlong Shang, Xin Gu, Jinglun Li, Chenghui Zhang
Abstract
The effective fault diagnosis method is a key measure to enhance the safety of lithium-ion batteries (LIBs). Nevertheless, it is challenging for conventional threshold diagnosis methods to detect minor faults in the early stages. Herein, an incipient multi-fault diagnosis method based on data-driven with incremental-scale is proposed. Firstly, a lightweight long short-term memory (LSTM) is used to obtain the healthy mapping relationship between the voltage-current interleaved sequence (VCIS) and cell voltage increment (CVI). Secondly, the faults are detected by determining the residual between the predicted CVI and the actual CVI. Eventually, the incremental diagnosis model (IDM) is constructed to distinguish the fault type and quantify the fault severity by evaluating fault resistance. The experiment results demonstrate that the proposed method performs excellent generalization to different battery types, fault degrees, and combinations of noise. Moreover, the minimum detectable fault CVI is as low as 15mV, and the mean error of fault resistance evaluation is 4.0%. Compared with conventional approaches, the proposed method has a higher fault detection accuracy and provides richer fault information. More importantly, the method is proven to work efficiently on battery management system (BMS).