Towards a more accurate characterization of granular media: extracting quantitative descriptors from tomographic images
Imaging, epitomized by computed tomography, continues to provide unprecedented 3D access to granular microstructures at ever-greater resolutions. The non-destructive technique has enabled deep insight into the morphology and behavior of granular materials, in situ and as a function of macroscopic states, e.g., loads. However, a significant bottleneck in this paradigm is that it ultimately yields qualitative 'pictures' of microstructure. Hence, a major challenge is to extract quantitative descriptors of grain-scale processes, e.g., morphological description of particles, kinematics, and spatial interactions. Existing methods, including watershed and burn algorithms, are plagued with limitations related to image resolution and with the inability to sharply identify grain-to-grain contact regions, which is crucial for studying force transmission and strength in granular materials. In this work, we propose a method to overcome these drawbacks. Specifically, a novel way to extract grain topology in particulate materials via level sets is introduced. It is shown that the proposed method can sharply resolve the topology of grain surfaces near to and far from grain-to-grain contact regions with sub-voxel resolution, and is capable of grain extraction directly in three dimensions. The proposed method still relies on traditional techniques for input, but ultimately leads to much improved grain characterization. We validate the approach using three dimensional CT images of highly rounded (Caicos ooid) and highly angular (Hostun sand) natural materials, with excellent results.