Higher Training Size, Increased Model Complexity or Both: A Novel Decision Framework for Cycle Life Classification of Lithium-Ion Cells

Among various prognostic and diagnostic applications of lithium-ion (Li-ion) batteries, their classification into high and low-cycle life cells is an important application. This paper focuses on two aspects of data-driven classification models: useful features and model complexity. Acquiring data that is rich in valuable features requires time and hardware resources. Moreover, the effectiveness of features varies from one dataset to another. In this paper, we explore these trade-offs using two different datasets. The results show that the lack of training data can be compensated by using a higher complexity of the machine learning model. The decision framework offers exciting insights for emerging businesses in deciding on the vendors by classifying their cells into low and high cycle life with limited training data.