This paper presents batternYNet, an Iterative Residual U-Net-based machine learning approach for detecting lithium structures in operando X-ray micro-computed tomography (μCT) datasets of solid-state batteries. The method addresses challenges posed by the low X-ray attenuation of lithium and large dataset size, enabling semantic segmentation of diverse lithium morphologies (dendrites, pits, deposited Li, redeposited Li). Quantitative analysis provides insights into volume and thickness changes of electrodes and deposited lithium, revealing spatial relationships between components and contributing to improved Li-metal battery design. The approach demonstrates transferability to other datasets and offers significant benefits for quality control.

Ying Huang, David Perlmutter, Andrea Fei-Huei Su, Jerome Quenum, Pavel Shevchenko, Dilworth Y. Parkinson, Iryna V. Zenyuk, Daniela Ushizima