We capture the geometries of sand grains via Micro-CT scanning images. Instead of using conformal meshes, we discretize the grains using cubic elements that compose surrogate domains. We concern two main problems, i.e. imposing boundary conditions and frictional contacts. The shift domain method is implemented to apply Dirichlet boundary conditions under the framework of an updated Lagrangian algorithm to simulate the damage of the grain. We also develop an unbiased Nitsche’s algorithm for frictional contacts using an implicit material point method enhanced by level sets. With these techniques, we can:
1. generate databases (constitutive laws considering real shapes) for machine learning;
2. introduce realistic mechanical behaviors of assemblies of grains to the multi-scale modeling.
Some applications are as follows.
(1) Soil fabric (clay structures in microscopic scales)
Flocculated structure Dispersed structure
(2) Isotropic compression of an assembly of four grains reconstructed from scanning images
(3) Macroscopic behaviors of granular assemblies considering stress-fields of individual grains (please click the figure, it is an animation)
The framework of force chains Macroscopic constitutive laws
(4) Brazilian test for an elastic disc (also click the figure)
(5) Shift domain method to apply boundary conditions
Scanning image Level set Surrogate domain Damage field