On-line State of Health Estimation Method for Lithium-ion Batteries based on Data-Driven Support Vector Regression

State of Health (SoH) estimation is becoming increasingly crucial in the automotive industry due to the rapid spread of electric vehicles. Innovative approaches for accurately estimating the SoH in real time are currently under research at numerous public and private institutions and companies. This work aims to serve as a starting point for future research by studying various Machine Learning (ML) estimation methods used in the literature for this purpose and, among these, selecting the one that best suits the onboard application requirements. The Support Vector Regressor (SVR) was used as the estimator for this work. Training data have been taken from two different datasets coming respectively from laboratories of Bologna University and from NASA. After an initial data preprocessing and feature extraction phase, the SVR was fine-tuned, trained, and tested with different feature sets pertaining only to partial charge intervals (in a real-world application, it is unlikely that the vehicle would be fully charged from a fully discharged state). The results demonstrate the SVR's effective capability to be used as SoH estimator. However, the insufficient level of generalization achieved by the algorithm in cross-battery validation indicates the need for further study of the features used to train the algorithm.