Physical knowledge guided state of health estimation of lithium-ion battery with limited segment data

• Published in: Reliability engineering & system safety Vol. 251; p. 110325
• Main Authors: Wang, Fujin, Wu, Ziqian, Zhao, Zhibin, Zhai, Zhi, Wang, Chenxi, Chen, Xuefeng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2024
• Subjects: Equivalent circuit model; Interpretability; Lithium-ion battery; Partial charging data; Physical features; Physics-guided neural network; State-of-health; Lithium-ion battery; Interpretability; Partial charging data; State-of-health; Physical features; Physics-guided neural network; Equivalent circuit model
• ISSN: 0951-8320
• DOI: 10.1016/j.ress.2024.110325

Accurate state of health (SOH) estimation is basis for safe and reliable operation of lithium-ion batteries. In practice, accurate and reliable SOH estimation remains a challenge due to complex and dynamic operating conditions. In this paper, we propose a physics-guided neural network (PGNN) for SOH estimation of lithium-ion batteries. The physical knowledge is embedded into neural network from both explicit and implicit perspectives. Specifically, we extract physically meaningful features from the relaxation voltage segment of a fully charged battery based on an equivalent circuit model (ECM). These features and the limited current segment during constant voltage charging mode form the joint inputs for PGNN, both of which are less affected by the charging/discharging strategies. During the model optimization, the properties of the battery degradation are considered so that the model can learn better feature embeddings. To validate the proposed method, battery degradation experiments are performed to generate data over the entire battery life cycle. Finally, the superiority and effectiveness of the proposed method are validated on two datasets. •We propose a physics-guided neural network for SOH estimation.•A limited charging fragment is used for SOH estimation.•The physical features are extracted from relaxation voltage curve.•We consider battery degradation properties during the training. The effectiveness is verified on two datasets.