A new method for the segmentation of 3D grayscale images was developed during the structur.e project and was implemented in the GeoDict software.

With this method, Neural Networks (NN) are trained to segment nanoCT-scans of battery electrodes more precisely. The NN is trained by the user through marking (labeling) specific locations on the 3D grayscale images and assigning a material phase (active material, binder and carbon black [CBD], pore) to these locations. Based on this information, the NN then learns to perform this labeling for the entire scan set.

AI-based segmentation of a 3D gray-scale image with GeoDict.
Fig. 1 AI-based segmentation of a 3D grayscale image with GeoDict. The user marks (labels) certain areas and assigns these to a material phase. The AI learns the segmentation of the entire scan based on these training data.

A more accurate segmentation of nanoCT-scans of microstructured graphite anodes, produced as part of the project, is now possible with this method. These segmented digital 3D models constitute the basis for digital modeling, aiming to virtually test the best microstructuring processes and thus, provide a fingerprint for the manufacturing of the anodes.