Fault detection for lithium-ion batteries of electric vehicles with spatio-temporal autoencoder

Fault detection of lithium-ion battery packs is crucial for the safe operation of electric vehicles. Autoencoder, as an advanced machine learning method, has significant potential for improving anomaly detection accuracy. However, the autoencoders that just use temporal features in the reconstructio...

Full description

Saved in:
Bibliographic Details
Published in:Applied energy Vol. 392; p. 125933
Main Authors: Li, Heng, Liu, Zhijun, Bin Kaleem, Muaaz, Duan, Lijun, Ruan, Siqi, Liu, Weirong
Format: Journal Article
Language:English
Published: Elsevier Ltd 15.08.2025
Subjects:
Battery pack
energy
Inconsistency
lithium batteries
spatial data
Spatio-temporal autoencoder
time series analysis
Two-stage fault detection method
Battery pack
Inconsistency
Two-stage fault detection method
Spatio-temporal autoencoder
ISSN:0306-2619
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Fault detection of lithium-ion battery packs is crucial for the safe operation of electric vehicles. Autoencoder, as an advanced machine learning method, has significant potential for improving anomaly detection accuracy. However, the autoencoders that just use temporal features in the reconstruction process are suffered from high false positive rate. In this paper, a spatio-temporal autoencoder is proposed to address this limitation by learning the complex time dependence of time series data while considering inconsistencies within the battery pack. Meanwhile, a multi-head attention mechanism is introduced to the autoencoder to fuse temporal and spatial information to better learn the dynamic features of the battery charging time-series. Finally, we propose a two-stage fault detection method that first detects the faulty battery pack and then locates the faulty cell based on a dynamic threshold. Experimental results demonstrate the effectiveness of the proposed spatio-temporal autoencoder. It achieves the value of 0.961 on F1-score and is able to warn of faults a month in advance, accurately locating the faulty cell. •A spatio-temporal autoencoder for battery fault detection is proposed.•A multi-attention mechanism is introduced to integrate spatio-temporal information.•The two-stage fault detection method can significantly reduce the false positive rate compared to traditional method.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0306-2619
DOI:10.1016/j.apenergy.2025.125933