Cloud-Based Li-ion Battery Anomaly Detection, Localization and Classification
Aihua Tang, Zikang Wu, Yuchen Xu, Kailong Liu, Quanqing Yu
Abstract
Achieving comprehensive and accurate detection of battery anomalies is crucial for battery management systems. However, the complexity of electrical structures and limited computational resources often pose significant challenges for direct on-board diagnostics. A multifunctional battery anomaly diagnosis method deployed on a cloud platform is proposed, meeting the needs of anomaly detection, localization, and classification. First, the proposed method extracts four anomaly features from discharge voltage to indicate battery anomalies. A risk screening process is applied to classify vehicles into high, medium, and low-risk categories with these features. Next, these classifications and prior anomaly labels are utilized in the offline phase to train an anomaly classifier. Then, the types of faults are further segmented by a specially developed voltage cumulative difference mean model, the warning information is refined. Finally, the proposed method was validated on data from 25 real vehicles, achieving an anomaly detection accuracy rate that exceeded 98%, demonstrating its accurate detection capability. This article proffers an effective multifunctional vehicle anomaly detection method, providing a new approach to assist in-vehicle fault diagnosis with the support of a reliable cloud computing foundation.