Real-time electrochemical-strain distribution and state-of-charge mapping via distributed optical fiber for lithium-ion batteries.

Deep learning has opened new avenues for estimating the dynamic state of charge in batteries. However, hindered by macroscopic and limited electrochemical inputs, data-driven models still struggle to accurately capture local states inside batteries. This study aims to leverage distributed fiber optic sensing and long short-term memory to enhance the accuracy and reliability of the state of charge estimation and to visualize its distribution. First, the internal strain is closely linked to the lithium embedded state of graphite particles, with a coefficient of 0.992 between strain amplitude and capacity over long cycles. Incorporating strain as an additional physical parameter significantly improves accuracy. The proposed model achieves a prediction error of only 2.01 %, representing a 41.2 % improvement compared to models that exclude stain data. Factors such as current and ambient temperature are considered to prove the effectiveness of this method. Furthermore, by combining the trained model with internal strain distribution, the developed intuitive imaging for local state visualization reveals aging tendencies and uneven behavior in lithium-ion batteries. This predictive and imaging approach offers valuable insights for local state analysis and advances battery research and design. In this study, internal strain is correlated with electrochemical behavior via fiber-optic sensors, demonstrating that the strain parameter can more accurately reflect the state of charge (SOC) compared to external current and voltage. The strain distribution along the sensing network is obtained and subsequently fitted to an internal SOC mapping via a trained long short-term memory (LSTM) neural network, achieving 99.51 % fitting index and 2.01 % root-mean-square error. The imaging results show that inhomogeneity is highest in the mid-term and will increase as the pouch cell ages, with a tendency to concentrate in the central region of the electrode. [Display omitted] • Internal strain is monitored using distributed fiber optic sensing technology. • Strain evolution correlates with lithiation process, showing sensitivity to SOC. • A LSTM-based model predicts SOC by extra strain with a 2.01 % RMSE. • A dense sensing network visualizes SOC distribution, highlighting unevenness. • SOC imaging provides a novel method for detecting local anomalies and evolution. [ABSTRACT FROM AUTHOR]