Polyhedral particles are ubiquitous in natural and industrial processes. Recent advances in GPU computing have greatly enhanced the feasibility of discrete element method (DEM) simulations for polyhedral particles, yet accurately simulating their collective behavior remains computationally intensive for large-scale simulations. In this study, a distributed parallel DEM simulation framework with multiple-GPU computing for polyhedral particles is developed to achieve high-performance large-scale simulations. The framework integrates the Message Passing Interface (MPI) with NVIDIA's Compute Unified Device Architecture (CUDA), in which the main compute pipeline, including domain decomposition, neighbor-list construction, contact search, and evaluation of contact, is executed on GPUs. The proposed method is validated through both numerical and experimental studies. Numerical stability is verified through simulations of particle–wall impact and wall force evaluation at different mesh resolutions. The simulated static packing structures and velocity fields in a quasi-two-dimensional rotating drum show good agreement with experimental measurements. A scalability test involving 2 × 107 polyhedral particles on 16 GPUs demonstrates excellent parallel performance, achieving a 14.8 times speedup compared with the single-GPU case. Moreover, the large-scale applications, e.g., the silo deposition and fixed bed containing cylindrical catalyst particles, further demonstrate the capability of the proposed framework for industrial-scale applications.
扫码关注我们
求助内容:
应助结果提醒方式:
