Dense granular materials are difficult to reconstruct from X-ray computed tomography images because tight particle contacts, blurred interfaces, partial-volume effects, and spatially varying grayscale frequently cause particle merging or fragmentation. Reliable particle-resolved reconstruction is nevertheless essential for linking internal granular structure to macroscopic mechanical behaviour (Xing et al., 2021). This study presents GranulaX, a physics-guided workflow that converts grayscale image volumes into globally consistent particle labels for micromechanical analysis. The workflow integrates target-region extraction, image enhancement, initial particle segmentation, and uncertainty diagnosis using physically interpretable particle size and shape constraints. Uncertain regions are selectively refined, while overlapping subvolume processing and global label stitching enable scalable reconstruction of the full specimen.
Validation against manual annotations and independent laser granulometry demonstrated high accuracy in particle identification, three-dimensional reconstruction, and particle-size characterization. Beyond image segmentation, GranulaX establishes a particle-resolved structural representation from which particle-size redistribution, packing heterogeneity, contact-network organization, and their evolution during shearing can be quantitatively assessed. It therefore provides a direct pathway for linking X-ray tomography observations to particle-scale kinematics, including grain rearrangement, migration, and collective motion during deformation.
 

X-ray computed tomography; Dense granular materials; Particle-scale kinematics; Physics-guided image processing; Three-dimensional reconstruction