State-of-health estimation of lithium-ion batteries subjected to different ageing conditions via electrochemical impedance spectroscopy and machine learning

State-of-health estimation is essential for the safe and reliable operation of lithium-ion batteries, particularly under diverse real-world operating conditions. This study presents a hybrid framework that combines electrochemical impedance spectroscopy with machine learning for battery State-of-health estimation. Electrochemical impedance spectroscopy measurements were acquired from full lithium-ion cells aged under fourteen distinct operating conditions, providing a broad range of degradation trajectories for model development and evaluation. Two compact feature representations were compared: parameters obtained from equivalent circuit model fitting and band-compressed statistical descriptors calculated directly from the impedance spectra. These feature sets were used independently to train a stacking ensemble comprising XGBoost and Random Forest regressors with a linear regression meta-learner. Both feature representations achieved comparable state-of-health prediction performance, with root-mean-square errors below 3%. Feature-importance analysis showed that only a small subset of features contributed most of the predictive capability. The results show that compact electrochemical impedance spectroscopy feature representations can provide reliable and computationally efficient inputs for data-driven state-of-health estimation, offering a practical complementary approach for impedance-based battery diagnostics.