PUBLICATIONS

This dissertation proposes a method to estimate the State of Charge (SOC) and State of Health (SOH) of lithium iron phosphate (LFP: LiFePO₄) batteries using deep learning, and to apply it for condition monitoring and lifetime prediction. The chemical structure and electrical characteristics of the LFP battery are theoretically analyzed, and voltage data varying with the battery’s charge level is collected through charge-discharge cycle experiments. The experimental data of the LFP battery is obtained using the Galvanostatic Intermittent Titration Technique (GITT), and the SOC and SOH are analyzed based on the Open Circuit Voltage (OCV) and diffusion coefficients. Changes in internal resistance (DCIR: Direct Current Internal Resistance) are correlated with battery health, and a regression-based estimation method is applied to estimate SOC and SOH. In addition, an equivalent circuit model-based estimation algorithm is used to dynamically track the battery state. A simplified 3-RC equivalent circuit model is used to simulate the actual operation of the battery, and its parameters are extracted from the GITT experimental data. These extracted parameters are used to estimate SOC and SOH using both the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF), and the results are comparatively analyzed. The proposed algorithm estimates the battery state and predicts lifetime using charge-discharge data from a battery pack composed of 24 LFP cells in series, by combining a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). CNN extracts abnormal cells based on voltage and current input data, while LSTM learns temporal dependencies and long-term patterns to predict battery health. The predicted battery pack state is validated by comparison with cell-level experimental data. The effectiveness of real-time state estimation and fault prediction for LFP batteries is verified through simulation and experimentation.