Methodology for Constructing Three-Dimensional DEM Model of Rock Specimen from CT Scan

Developing numerical simulation models that accurately represent actual features of rocks, including the distribution of minerals, voids, and fractures, is crucial. The advancement in non-destructive evaluation methods, especially X-ray computed tomography (CT), allows for detailed analysis of the internal structures of rocks. This study uses the Res-VGG-UNet deep learning model to segment CT images, distinguishing between components that appear similar but are distinct. Integrating the findings from deep learning into PFC^3D software enables the creation of rock models that truly mimic the features of actual samples. The simulation results obtained from these PFC^3D models show a high degree of agreement with experimental outcomes, highlighting the effectiveness of these methods in developing accurate rock numerical models.