A new approach to apply realistic fluid confining pressure has been developed for modeling triaxial tests on intact rock using the discrete element method (DEM). The new approach applies updated force boundary rather than a "wall" boundary to realistically simulate the confining pressure applied by a flexible membrane. The applied force only acts on the boundary particles, which are identified and updated periodically. Comparisons between rigid-wall boundary and membrane boundary were performed. The results have shown that rigid-wall boundary can significantly influence the material responses especially for the strength.
The discrete element method (DEM) can be generally viewed  as a method that allows finite displacements and rotations of discrete bodies, and update contacts automatically as the calculation progresses. Nowadays, DEM is widely used in geomechanics from soil (particulate type) to intact rock (relatively continnum type), to rock masses (assemblies of blocks) with applications in many areas, such as rock engineering, soil mechanics, mining and petroleum engineering [2-7]. When modeling the behavior of intact rock, the elements used in the model do not represent the actual material particle size and they are bonded to each other with a specific strength. DEM is attractive in modeling bonded geomaterials because it can naturally deal with the material failure by modeling the failure evolutionary process from micro crack development to macro failure without any complex constitutive models. One of the difficulties in modeling triaxial tests in DEM is applying confining pressure to realistically represent the test condition. Currently, the conventional periodic, rigid, and flexible boundaries are commonly used to simulate triaxial tests. Firstly, Periodic boundary is usually used in simulations with box specimens, and is implemented by copying the boundary particles to the opposite side of the box. The confining pressure is achieved by compacting the specimen
a specific pressure . The periodic boundary is difficult to implement in
cylindrical specimen. The way to achieve the desired confining pressures by compacting specimen can significantly increase the computational effort, and it is difficult to keep the confining pressure constant. Secondly, rigid boundary treats boundaries as rigid either in plane or cylindrical walls without any deflections. The confining pressures are applied by moving these boundary walls to reach the desired average stresses calculated at the boundaries [9, 10]. The drawback of rigid boundaries is that the boundary particles tend to be artificially aligned with the boundary wall, and the material failure process and deformation may be overly constrained by such a boundary, and hence, are not fully representative of the actual test conditions. Lastly, flexible boundaries emulate the boundary in the conventional triaxial setup by using adjoining triangular plate elements, whose corners connect the centers of neighboring particles. Prescribed confining pressure is applied to these plate elements, and the forces are then distributed among the three neighboring particles that form the plate element . Flexible boundary tried to mimic the rubber membrane function, but may fail to accurately determine the direction for applied confining pressures especially for cylindrical specimens.