Internal Strain and Temperature Sensing-Assisted Neural-Network-Based Battery SOC Estimation

Lithium-ion batteries are pivotal for electric mobility and energy storage, where safe and efficient operation relies on accurate state-of-charge (SOC) estimation. To enhance estimation accuracy, embedded fiber Bragg grating (FBG) sensors are employed in this study to obtain in situ, real-time measurements of internal strain and temperature, providing direct indicators of electrode expansion and thermal responses. Distinct C-rate-dependent strain and temperature trends are observed, highlighting the value of embedded sensing in capturing internal battery dynamics. Building on this foundation, this study systematically investigates the impact of these internal sensing features on SOC estimation accuracy in Li-ion batteries using two representative neural network-based architectures, a fully connected neural network (FCNN) and a temporal convolutional network (TCN). Incorporating internal strain and temperature features significantly improves accuracy over electrical-only inputs, reducing mean absolute error by up to 71.9%. Robustness and generalizability are further examined under five practical scenarios, including unseen C-rates, dynamic load profiles, temperature generalization, cross-cell transfer and noisy measurements. While the TCN achieves superior performance in cross-cell transfer and noisy conditions, the FCNN performs better for unseen C-rates, temperature generalization and comparably under dynamic loads. These findings demonstrate that high-fidelity embedded sensing can offset the need for highly complex architectures, providing a practical pathway to accurate and efficient SOC estimation for next-generation battery management systems.