X-ray computed tomography (X-ray CT) is a non-destructive characterization technique that in recent years has been adopted to study the microstructure of battery electrodes. However, the often manual and laborious data analysis process hinders the extraction of useful metrics that can ultimately inform the mechanisms behind cycle life degradation. This work presents a novel approach that combines two convolutional neural networks to first locate and segment each particle in a nano-CT LiNiMnCoO2 (NMC) electrode dataset, and successively classifies each particle according to the presence of flaws or cracks within its internal structure. Metrics extracted from the computer vision segmentation are validated with respect to traditional threshold-based segmentation, confirming that flawed particles are correctly identified as single entities. Successively, slices from each particle are analyzed by a pre-trained classifier to detect the presence of flaws or cracks. The models are used to quantify microstructural evolution in uncycled and cycled NMC811 electrodes, as well as the number of flawed particles in a NMC622 electrode. As a proof-of-concept, a 3-phase segmentation is also presented, whereby each individual flaw is segmented as a separate pixel label. It is anticipated that this analysis pipeline will be widely used in the field of battery research and beyond.
Keywords: computer vision; convolutional networks; lithium-ion batteries; mask R-CNN; nano X-ray tomography.
© 2022 The Authors. Small Methods published by Wiley-VCH GmbH.