GPU-accelerated peridynamic simulation for quasi-brittle materials: model development and efficient parallel framework

The prediction of crack initiation and propagation is of vital importance to engineering safety in geotechnical engineering. However, the nonlinear damage characteristics of quasi-brittle materials, coupled with the high computational cost of nonlocal methods, present significant challenges to the simulation of crack evolution. Thus, this study introduces a quasi-brittle peridynamic (QB-PD) model based on the classical prototype micro-elastic brittle peridynamic (PMB-PD) formulation and further develops an efficient GPU-accelerated computational framework. Unlike the PMB-PD model, which accounts for only linear elastic tensile failure, the QB-PD model incorporates a bond stiffness degradation function and a repulsive force mechanism between material points, enabling the characterisation of nonlinear damage and compressive failure. To enable large-scale simulations, a GPU-parallel computing framework is constructed leveraging the inherent parallel characteristics of peridynamics, with a point-pair mapping strategy as the core logic. This framework systematically optimises the data structure and computational workflow of the peridynamic solver. The QB-PD model is validated through several benchmark numerical examples, and the performance of the parallel framework is thoroughly evaluated in the final case. Results demonstrate that the proposed model can more accurately capture the crack paths and nonlinear mechanical responses of quasi-brittle materials. Meanwhile, the GPU-based framework significantly enhances computational efficiency, reducing memory usage by nearly 50 % and achieving millisecond-level preprocessing for million-point models. During the solution phase, it achieves up to 10 3 -fold speedup compared to serial execution and provides approximately 3 × 10 2 -fold and 5 × 10 1 -fold speedups over serial and OpenMP implementations, respectively, under high-precision settings. This approach greatly facilitates efficient, large-scale simulation of geotechnical failure processes.