A data-driven surrogate model for non-electrical offline state-of-charge estimation of LiFePO₄ batteries using...

The aim of this study is to develop a non-electrical framework for accurate offline state-of-charge (SOC) estimation of lithium-ion batteries (LIBs), with particular focus on lithium iron phosphate (LiFePO₄, LFP) cells. Conventional voltage-based SOC estimation methods are inherently limited for LFP batteries due to their flat open-circuit voltage characteristics and are not well suited for offline measurements. To overcome these limitations, a data-driven SOC estimation approach based exclusively on guided ultrasonic wave (GUW) signals is proposed. Rich parametric features are first extracted from dispersive GUW responses using a matching pursuit (MP) algorithm combined with a customized asymmetric Hanning-windowed chirp dictionary, enabling isolation of first time-arrival wave packets and characterization of dispersion-related waveform features beyond conventional time-of-flight and amplitude metrics. The extracted parameters are used to train an Extreme Gradient Boosting (XGBoost) regression model capable of capturing nonlinear relationships between GUW features and SOC. Experimental validation is conducted on an LFP pouch cell under multiple discharge rates. The proposed method achieves a root mean square error below 4%, outperforming both voltage-based models and GUW-based models relying on limited signal features. Mutual information-assisted parameter screening further improves prediction robustness by identifying an optimal feature subset. These results demonstrate that feature-rich GUW signals provide a viable, non-destructive, and offline alternative for SOC estimation, offering strong potential for battery screening, recycling, and second-life evaluation.