Development of a 3D Hybrid Finite-Discrete Element Simulator Based on GPGPU-Parallelized Computation for Modelling Rock Fracturing Under Quasi-Static and Dynamic Loading Conditions

Abstract

As a state-of-the-art computational method for simulating rock fracturing and fragmentation, the combined finite-discrete element method (FDEM) has become widely accepted since Munzijia (2004) published his comprehensive book of FDEM. This study developed a generalpurpose graphic-processing-unit (GPGPU)-parallelized FDEM using the compute unified device architecture (CUDA) C/C++ based on the authors' former sequential two-dimensional (2D) and three-dimensional (3D) Y-HFDEM IDE (integrated development environment) code. The theory and algorithm of the GPGPU-parallelized 3D Y-HFDEM IDE code are first introduced by focusing on the implementation of the contact detection algorithm, which is different from that in the sequential code, contact damping and contact friction. 3D modelling of the failure process of limestone under quasi-static loading conditions in uniaxial compressive strength (UCS) tests and Brazilian tensile strength (BTS) tests are then conducted using the GPGPU-parallelized 3D Y-HFDEM IDE code. The 3D FDEM modelling results show that mixed-mode I-II failures are the dominant failure mechanisms along the shear and splitting failure planes in the UCS and BTS models, respectively, with unstructured meshes. Pure mode I splitting failure planes and pure mode II shear failure planes are only possible in the UCS and BTS models, respectively, with structured meshes. Subsequently, 3D modelling of the dynamic fracturing of marble in dynamic Brazilian tests with a split Hopkinson pressure

Three-Dimensional Combined Finite-Discrete Element Modeling of Shear Fracture Process in Direct Shearing of Rough Concrete–Rock Joints

“…However, due to the non-parallelized nature of the code, its previous applications were limited to small-scale 2D problems using relatively rough meshes. Recently, to overcome these limitations, a parallel programming scheme using GPGPU controlled by compute unified device architecture (CUDA) C/C++ was incorporated to develop the GPGPU-parallelized Y-HFDEM IDE code by the authors [47][48][49][50][51] for 2D and 3D FDEM modeling. Since the full details of the Y-HFDEM IDE code can be found in the literature [49,50], we focus in this section on the essential parts of the 3D FDEM in order to model the fracture process in the direct shearing of the rough concrete-rock joint.…”

“…Recently, the combined finite-discrete element method (FDEM), originally proposed by Munjiza [35,36], has attracted significant attention in the field of computational solid mechanics. Since then, various FDEM codes have been developed in several research groups across the world [37][38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53][54]. The FDEM can model the three important aspects of solid fracture, i.e., the continuous deformation in the intact regime, the transition from continuum to discontinuum, and the very complex contact mechanics between solid surfaces, including newly created macroscopic fracture surfaces.…”

Benchmarking Numerical Methods for Impact and Cratering Applications

“…This work presents a code-to-code comparison between a hydrodynamics code (hydrocode) and a finite-discrete element method (FDEM) code. The hydrocode approach is a more traditional approach for addressing impact cratering problems [4][5][6][7][8][9], while the FDEM approach has been traditionally used for brittle fracture in geomaterials [10][11][12]. These approaches are operative on similar spatial scales, but the underlying formulation, particularly in how each approach accounts for damage, is very different.…”

Combined Viscoplasticity-Embedded Discontinuity Model for 3D Description of Rock Failure Under Dynamic Loading

“…There is also the class of hybrid methods combining the advantages of discontinuum and continuum approaches in a single code (see e.g. Klerck et al 2004;Mahabadi et al 2010;Mahabadi 2012;Mardalizad et al 2020;Fukuda et al 2020). Although the hybrid methods capture the fracture processes very well and are easier to calibrate than pure particle methods, they still inherit the computational labour of the particle methods, which render them often unattractive.…”