A GPU parallel scheme for accelerating 2D and 3D peridynamics models

•A GPU parallel scheme is proposed to accelerate PD models while ensuring computational accuracy.•The bond-mapping unrolls the inner loop of the particle-mapping for stronger performance.•Three different types of PD numerical experiments are performed to analyze the performance of parallel codes.•Exploring the initiation and propagation behavior of fatigue cracks in rails with hole defects.•The code is hosted on Github https://github.com/XiaoMmingW/PD_Gpu1.git.