A graph-based approach for tracking grains in time-resolved 2D/3D microstructural datasets

Understanding how grains evolve in polycrystalline materials is critical for linking processing to structure and ultimately properties. However, grain tracking in time-resolved 2D/3D datasets remains a major bottleneck due to topological changes, grain translation, sample deformation, and registration errors. Here, we present two new registration-independent tracking algorithms designed to overcome these challenges: a coordinate-transformation-based method based on local misorientation neighborhoods, and a graph-based method that reformulates the microstructure as a network of grains (nodes) and grain boundaries (edges) and uses graph matching to resolve correspondences while preserving topological consistency. We benchmark both against two conventional approaches, volume overlap and centroid-proximity-based matching, using two large-scale 3D phase-field simulations (with and without a thermal gradient) and an experimental laboratory-based x-ray diffraction contrast tomography dataset containing over 2,200 grains. Across all datasets, the graph-based approach consistently achieves the highest tracking efficiency and accuracy, especially in cases with substantial microstructural evolution or sample deformation. We further demonstrate how graph-based representations facilitate the quantification of topological events and guide experimental design by estimating the maximum interval between scans for reliable tracking. These results establish a robust, scalable, and topology-aware framework for 4D materials characterization, with broad applicability to both synthetic and experimental studies.