Discharge Capacity Prediction from Unstructured Cathode Material Data in Li-Ion Batteries

In the process of material development such as battery materials, structured tabular data such as Excel is widely utilized to integrate data such as material information, process information, and analysis information. Recently, various machine learning techniques have been applied to analyze such structured data.1) In particular, tree-based machine learning models and deep learning models are mainly used to analyze structured data.2) Among the models, deep learning models can (1) directly process unstructured data such as cathode materials and (2) consider the relationships between interrelated features such as composition and crystal information - LiNi0.6Mn0.2Co0.2O2 and lattice constants from X-ray diffraction. In this study, we developed a deep learning model to predict the discharge capacity from data collected during the development of cathode materials, which is originally unstructured data. The features were grouped by name to consider the relationship between the features. Each groups, for example composition, were embedded into a representation, and used to predict the discharge capacity.<br/><br/>References<br/>1) Guanyu Wang, et al. ACS Cent. Sci. (2021) 7, 9, 1551-1560<br/>2) L. Grinsztajn, E. Oyallon, and G. Varoquaux. (2022) arXiv:2207.08815