State of Health Estimation of Lithium-Ion Batteries Combining Electrical and Ultrasonic Signal Features

State of Health (SOH) serves as a key metric in assessing the performance of lithium-ion batteries. It is challenging for a single sensing signal to fully characterize the multi-physics evolution characteristics during battery degradation, which limits the accuracy and robustness of SOH estimates. Therefore, a lithium-ion battery SOH estimate method combining electrical and ultrasonic features with a frequency-enhanced decomposed Transformer (FEDformer) is proposed. To begin with, a multi-condition battery aging dataset is constructed through experiments, comprising electrical and ultrasonic signal data from 7828 cycles. Subsequently, 20 electrical and ultrasonic features are extracted from multiple perspectives, and 12 strongly correlated features are selected via the Spearman correlation coefficient. Finally, the FEDformer is employed to establish the SOH estimate model, where the accuracy, robustness, and generalization of the SOH estimates are comparatively analyzed across different input features, models, and cross-aging conditions. The results demonstrate that, compared to using electrical features alone, the combined ultrasonic features improve the estimation performance by more than 40% on average. Furthermore, in the cross-aging datasets, the mean absolute error and root mean square error of the SOH estimates are 0.52% and 0.63%, respectively, validating the robustness and generalization capability of the proposed method.