A New State-of-Health Estimation Method for Retired Lithium-Ion Battery Based on Expansion

To address the challenges of long detection times and low classification efficiency in sorting retired lithium-ion batteries, this paper proposes a new state-of-health estimation method. The approach is based on the mechanical characteristics of retired cells. The cell expands more obviously than new cell. Taking pouch cell as an example, the study explores the relationship between battery aging and both reversible and irreversible expansion. The features such as irreversible expansion, increment of irreversible expansion, rise of reversible expansion, reversion of reversible expansion and temperature are extracted to construct state-of-health feature data. A hybrid network model is developed by integrating a 50-layer residual network with pyramid squeeze attention and connecting it to a gated recurrent unit, which is then used to estimate the state-of-health. The model has been validated on public datasets. In the public dataset, it achieves a root mean square error of approximately 1%, a mean absolute error of around 0.5%, and a mean square error of about 1%. Furthermore, comparative experiments demonstrate the method’s superior stability and robustness. The relevant error sources are discussed and further improvements are proposed.