Ordinary and Functional Principal Component Analysis for Relaxation Voltage Profiles and State of Health Estimation in Lithium-Ion Batteries

Relaxation voltage profiles methodologies have been proven to successfully estimate the State of Health (SoH) of lithium-ion batteries. These methods are particularly interesting since relaxation curves do not depend on the charging process and can be obtained directly from battery management systems, without the need for additional devices. Usually, features are extracted from relaxation voltage profiles using standard statistics, like mean, variance, maxima, minima, skewness, and kurtosis. In this paper, we propose alternative feature extraction methods based on principal component analysis (PCA) and functional principal component analysis (fPCA). The extracted features are then used as input for the eXtreme Gradient Boosting (XGBoost) algorithm to estimate the current SoH of the cells. The methodology is tested on a recent large public dataset containing 65 commercial lithium-ion NCA cells cycled under various conditions. The PCA and fPCA results provided valuable insight into the variability structure of the relaxation profiles and confirmed the effectiveness of the summary statistics variance, skewness, and maxima, which, despite their simplicity, remain the input configuration yielding the lowest test Root Mean Squared Error (RMSE) in the XGBoost algorithm.