Methods of Machine Learning and Spatial Stochastics for Characterizing the 3D Morphology of Battery Materials at Various Length Scales

Microscopy techniques like X-ray computed tomography (CT) or scanning electron microscopy (SEM) can provide detailed image data of the nano- and microstructure of functional materials at various scales. This allows for the investigation of structure-property relationships, i.e., how the nano-/microstructure influences material properties like electrochemical behavior. However, the structural characterization by means of measured image data often entails nontrivial processing tasks. In this paper, a workflow is shown for the holistic structural characterization of active material (AM) particles in Li-ion battery electrodes by means of stochastic modeling. For this purpose, image data are acquired at different length scales and with various measurement techniques, namely, CT, SEM, and focused ion beam (FIB) combined with electron backscatter diffraction (EBSD). To enable quantitative structural characterization through spatial stochastic modeling, machine learning methods for segmentation purposes are initially deployed. Then, a stochastic-geometry model for AM particles is calibrated to image data, overcoming the limitations of the different measurements (e.g., resolution vs. field of view). The model is fitted using CT data for the outer shell of AM particles and FIB-EBSD data for the polycrystalline grain architecture within. Then, the model is used to perform structural scenario analyses, i.e., to generate numerous digital twins with statistically similar shape and grain architecture as observed in measured image data. These digital twins are input for numerical (dis)charging simulations to investigate their degradation behavior, e.g., AM particle cracking due to repeated cycling.