Xiao Cui1,Shijing Sun2,William C. Chueh1
Stanford University1,Toyota Research Institute2
Xiao Cui1,Shijing Sun2,William C. Chueh1
Stanford University1,Toyota Research Institute2
Formation plays a critical role in battery manufacturing as it significantly impacts the quality of the solid electrolyte interface (SEI) formed, which in turn affects battery performance. However, formation can be time-consuming, costly, and difficult to optimize. In this study, a dataset comprising 150 cells, 50 different formation protocols, and 6 formation parameters was generated. The results reveal a wide range of battery lifetime based on different formation conditions. Interpretable machine learning is used to systematically study the contribution of formation parameters to battery performance. Furthermore, since cycling aging is kept the same for all the cells, the value of using this dataset for feature testing is discussed, along with an investigation into the predictive origin of the dominant features. We highlight the multi-purpose of this dataset for both Bayesian optimization and feature testing.