Spectral-ICA-Based Thermal Fault Detection for Large Format Lithium-ion Battery

Abstract

Thermal fault detection is critical to the safety of electric vehicles. Due to the uneven surface temperature, traditional lump-based fault detection methods are unsuitable for large format lithium-ion batteries. This paper proposes a spectral independent component analysis (spectral-ICA) based distributed thermal fault detection framework to solve this problem. It contains two stages: 1) In the offline training stage, the 2-D battery thermal process is first decomposed into basis functions and time coefficients using the spectral method. The time coefficients are further decomposed by ICA. Then, the dominant temporal components and the residual errors are formed as monitoring statistics, which are used to derive the confidence bounds through the kernel density estimation. 2) In the online stage, the thermal fault can be detected in real-time by comparing the updated monitoring statistics with the confidence bounds. Simulations on a pouch-type lithium-ion battery are used to verify the effectiveness of the proposed method.