State of Health Estimation of LIB Under Partial Charging Conditions Using ICA Peak Features and Gaussian Process Regression

Lithium-ion batteries (LIBs) are indispensable energy-storage devices due to their high energy density and long cycle life. The rapid growth of the electric-vehicle sector and the pursuit of greater energy efficiency have intensified the need for reliable techniques that continuously assess battery performance. Because cycling progressively raises internal resistance and induces capacity fade, precise estimation of a cell’s State of Health (SOH), which captures its degradation level and remaining useful life, is essential for safe and durable operation. This study proposes a Gaussian Process Regression (GPR) framework for SOH estimation using partial-charge data. To improve estimation accuracy, feature vectors consist of (i) the mean and standard deviation of incremental capacity obtained from partial-charging segments during each CC-CV charging process and (ii) peak magnitudes with corresponding voltages extracted via Incremental Capacity Analysis (ICA), which elucidates degradation patterns. The approach was validated using five SAMSUNG INR 18650-25R cells, four for model training and one for independent testing. Predictive accuracy was quantified using mean absolute error (MAE) and root-mean-square error (RMSE), and the proposed model’s estimation uncertainty was quantified and presented as 95% confidence intervals.. The results demonstrate that the proposed GPR model yields accurate and reliable SOH estimates, underscoring its suitability for implementation in on-board battery-management systems.