Comparative study of machine learning techniques for the state of health estimation of Li-Ion batteries

Lithium-Ion batteries play a crucial role in vehicle electrification to meet the goals of reducing fossil fuels. However, they deteriorate over time and it is thus needed to predict their rate of decay since, after a certain threshold, they are not suitable anymore for their designed application. One of the main indicators that displays the aging status of a cell is the State of Health (SOH), which expresses the remaining total capacity of the cell in comparison with its initial total capacity. For an accurate evaluation of the SOH, various methods have been proposed in the literature. In this article, a comparative study of nine data-driven techniques in terms of error and processing time is presented. These different algorithms have been tested on two public datasets, extracting selected features and performing an Incremental Capacity Analysis (ICA) to estimate the SOH of a cell using only partial discharges. The benefits of using Machine Learning (ML) methods for the models can be found in the processing times that are faster than experimental methods, making the entire industrial testing stage faster.