Early Voltage-Based SOH and SOC Estimation for EOL Detection and Enhanced Battery Cell Balancing

This paper presents a data-driven methodology for estimating the state of health (SOH) and state of charge (SOC) of lithium-ion cells using early voltage data. A pivotal discovery in the Oxford dataset revealed that cells with SOC depletion times below a threshold of 2900 seconds consistently exhibited SOH values below 80%, signaling their transition to the second-life phase. This insight not only facilitates early end-of-life (EOL) detection but also aids in dynamic cell balancing and the initiation of second-life battery applications. Random Forest (RF) emerged as the best-performing model for SOH estimation across various cross-validation strategies. For SOC depletion time estimation, incorporating SOH and the estimation of linear regression of this time as input parameters allowed RF to outperform linear regression, Gaussian Process Regression (GPR) and Feed-Forward Neural Networks (FNN). To validate the robustness of the approach, the methodology developed for the discharge phase was extended to the charge phase, producing consistent results over multiple cycles. Additionally, the method was validated on a separate dataset exhibiting distinct physiochemical behaviors. This research contributes to advancing battery management systems by improving the reliability of energy storage and supporting second-life applications.