Sections of the image information of soil-rock mixture are obtained by CT scanning

the point cloud data of gravel surface are extracted by technology of image binarization and boundary recognition. By using Geomagic Studio

a reverse engineering software

three-dimensional geometric model of the gravel is reconstructed and the geometric model of gravels of the soil-rock mixture imported to ANSYS is quickly obtained. With the features of powerful and flexible function of mesh dividing

ANSYS ICEM CFD is adopted to mesh and generate orderly tetrahedral elements to obtain 3D numerical model of the soil-rock mixture. The gridding along the boundary of the soil and the gravels is densified. By using the interface code

3D numerical model is imported into FLAC3D for numerical simulation. Comparison with the results of nongravel model

it is shown that the established fine 3D model can reflect the structure characteristic of the soil-rock mixture. Especially

it can precisely describe the stress and the deformation of the soil

as well as the boundary behavior of soil and gravel. The method presented can reduce the error caused by model simplification and improve the accuracy and reliability of numerical simulation.