A Framework for Normalizing Physical Features of Li-Ion Batteries to Form a Generic Health Estimation Model

This article presents a framework for health estimation of a Li-ion battery that is conceptually formed based on dimensionless and common physical characteristics of a Li-ion battery. The proposed normalized features and framework enable the formation of a data-driven model to be trained using an available dataset and applied to estimate the state of health (SoH) of another Li-ion battery. It is established based on introducing the new concepts of participation factors and normalized dimensionless features which are independent of a battery capacity and its dimensions. Effective training of a model using a conventional data-driven approach needs an enriched dataset that is not readily available, particularly in high-power and custom-made battery energy storage applications. The reason is that data generation needs time-consuming and expensive lifespan experiments using several samples of Li-ion batteries. To evaluate the performance of the proposed framework, four widely used Li-ion battery datasets with different capacities and dimensions are selected and compared with conventional
$z$
-score and min-max normalization methods. Test results show that the average error using the proposed method in the early stage of battery lifespan is 3.5% that is 10% less than the error in existing methods.