Application of deep learning image recognition for lithium battery State of Health assessment

Abstract Accurately estimating the State of Health (SOH) of lithium‐ion batteries is essential for ensuring their reliable operation. The constant‐current charging voltage curves of batteries at different aging levels show significant deviations. Traditional methods based on one‐dimensional time‐series data face limitations in capturing and characterizing these complex patterns. To address this issue, this paper leverages the one‐dimensional (1D) time series data of the lithium battery constant‐current charging voltage segment, selected using incremental capacity analysis. This data is then transformed into a two‐dimensional representation using the Gramian angular summation field algorithm. Utilizing the exceptional image‐recognition capabilities of ResNet, this approach achieves high‐accuracy SOH estimation. Validation using publicly available datasets from the University of Oxford and the University of Maryland demonstrates a significant improvement in battery SOH estimation accuracy compared to traditional techniques, which directly input voltage segments into the network.