Multi-sensor multi-mode fault diagnosis for lithium-ion battery packs with time series and discriminative features

• Published in: Energy (Oxford) Vol. 290; p. 130151
• Main Authors: Shen, Dongxu, Yang, Dazhi, Lyu, Chao, Ma, Jingyan, Hinds, Gareth, Sun, Qingmin, Du, Limei, Wang, Lixin
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2024
• Subjects: Fault identification; Lithium-ion battery pack; Principal component analysis; Sensor fault; Lithium-ion battery pack; Sensor fault; Fault identification; Principal component analysis
• DOI: 10.1016/j.energy.2023.130151
• ISSN: 0360-5442

Sensor fault diagnosis is essential to guaranteeing the safety of lithium-ion batteries. To address the general drawbacks of the existing diagnosis methods, including the difficulty in determining the threshold, inability to handle multiple faulty sensors concurrently, and limited capacity in identifying fault modes, a multi-sensor multi-mode fault diagnosis method for lithium-ion battery packs is proposed. The proposed method utilizes time series and discriminative features to accomplish sensor-specific fault detection and fault mode identification. First, a total of 18 general time series features are extracted to characterize the measurements of each sensor during each charge–discharge cycle. Principal component analysis is then used to reduce the high-dimensional feature space to a two-dimensional space, such that fault detection can be carried out with the α-hull algorithm. For the detected faulty samples, a two-layer identification algorithm is designed based on three discriminative features, namely, correlation coefficient, impulse factor, and Hurst coefficient, to identify the specific fault modes. The diagnostics can decouple the information from different types of sensors so that the proposed method can effortlessly isolate current, voltage, and temperature sensors that are concurrently experiencing faults. Ultimately, experimental results from three scenarios, including simultaneous failure of multiple sensors, substantiate the effectiveness and feasibility of the proposed method. [Display omitted] •Sensor faults are detected without establishing models and setting thresholds.•Different types of sensors that malfunction simultaneously are effortlessly isolated.•The fault modes of the faulty samples can be accurately identified.•The proposed method can handle scenarios where multiple sensors fail simultaneously.•The diagnosis results in three scenarios prove the validity of the proposed method.