On the GPGPU-accelerated thermomechanical coupling finite-difference method / hybrid finite-discrete element method (FDM/FDEM) model for geomaterials

Abstract

Thermomechanical (TM) coupling analysis is one of the hot research topics in computational geomechanics. To address the large computational costs associated with high-accuracy simulations, General-Purpose computing on Graphics Processing Units (GPGPU) acceleration is a cutting-edge technique that enhances numerical efficiency. This study presents a transformative approach by presenting the CUDA-based TM coupling model that integrates the finite-difference method (FDM) with a hybrid finite-discrete element method (FDEM). FDM grids are used for thermal field calculations, while FDEM meshes are employed to compute mechanical responses. Bi-linear interpolation is applied to transfer the FDM results to the FDEM model. The versatile model supports both rectangular and circular FDM boundaries. To address the challenge of contact heat transfer problems, we introduce a heat pipe scheme for the FDM-FDEM model. Several CUDA kernel functions have been developed to calculate node forces, perform contact detection, compute contact forces, and evaluate the thermal field. A series of numerical cases involving rock plates, rock heaps, wellbore stability, ceramics, and radioactive waste repositories were conducted to validate the model's reliability. Comparisons of computational efficiency demonstrate that our model can achieve an overall speed-up ratio in the hundreds, significantly outpacing traditional models implemented in C language. These results highlight the potential of our approach to advance TM coupling algorithms and FDEM modeling.