Data-driven model enhancement of late-life lithium-ion batteries

Battery models require parameter adaptation to account for degradation during their lifetime. Current parameter estimation methods need an accurate pre-defined OCV curve, which can be expensive and time-consuming to obtain if not available. Furthermore, the shape of the OCV curve changes as the battery degrades, making measurements at the beginning-of-life insufficient at later stages of the battery lifetime. This work introduces a data-driven approach to build a lithium-ion cell model using only operational data. It enhances an equivalent circuit model with Gaussian process regression to fit the OCV curve and the non-linear SOC dependency in the cell’s internal resistance. To put it to the test, it is compared to a state-of-the-art method in a model fitting benchmark, using a dataset of cells with SOH ranging between 100% and 70%. While the conventional method loses accuracy with cell degradation, the proposed method accurately reconstructs the OCV curve, estimates the cell impedance and achieves a high accuracy over the whole lifetime.