Quantitative assessment of machine-learning segmentation of battery electrode materials for active material quantification

X-ray computed tomography (CT) is an important tool for studying battery electrode microstructures but relies on robust segmentation validity. Here, several approaches to applying accessible machine-learning software segment open-source lithium-ion (LIB) tomograms are followed identify the optimised methodology that minimises variation in active material volume fraction quantification across three users. Iterative, manual training seven cross-sectional slices (<5%) of a tomogram identified as optimal balance between variance and user interaction, where 10–25% each slice was trained. This approach applied lab-based CT data compared with obtained by focused-ion beam/scanning electron microscopy slice-and-view tomography. Variation users lower at least one these two (10% or 25%) when raw LIB cathode tomograms, versus unsupervised techniques such simple watershed segmentations. On average, absolute values closer acquired correlated technique, most closely matching high-resolution data. The present analysis provides using apply quantifying fractions cutting-edge electrodes, providing more route quantification.