摘要
采用离散元素法(discrete element method, DEM)进行颗粒系统运动仿真时,其模拟计算量大、计算效率低下,所采用的传统中央处理器(central processing unit, CPU)并行计算模型难以实现较大规模模拟。文章提出了一种基于图形处理单元(graphics processing unit, GPU)和统一计算设备架构(compute unified device architecture, CUDA)的并行计算方法;以球磨机的介质运动仿真为例,利用DEM方法结合CUDA并行计算模型,充分利用GPU众核多线程的计算优势,同时将颗粒属性信息存入GPU的常量存储器,减少信息读取的时滞,将筒体和衬板视为圆柱面和平面,简化了筒体与颗粒的接触判断,实现每个线程处理1个颗粒的相关计算,大幅提高计算速度;对颗粒堆积、筒体内2种尺寸颗粒运动进行仿真,并与基于CPU并行计算的结果进行对比。研究结果表明:在同等价格的硬件条件下,该文的方法可以实现10倍以上的加速比;对于含有复杂几何模型的仿真,如多尺寸颗粒和带衬板筒体的仿真,加速比会减少,但仍然可以实现数倍的加速。
When the discrete element method(DEM) is used for particle system motion simulation, it is difficult to achieve large-scale simulation using traditional central processing unit(CPU) parallel computing model for the large computational complexity and low computational efficiency. A parallel computing method based on graphics processing unit(GPU) and compute unified device architecture(CUDA) is adopted. To simulate the media motion in the ball mill, the DEM method is combined with the CUDA parallel computing model to make full use of the computational advantages of vast computing cores of the GPU, and the particle attribute information is stored in the GPU constant memory to reduce the time lag of information reading. The drum and linings are regarded as cylindrical surfaces and planes, which simplifies the contact judgment between the drum and the particles. Based on the above statements, an algorithm which processes the calculation of one particle per thread is realized. By simulating the particle packing and the motion of multi-size of particles in a rotating drum respectively, it is found that using hardwares of the same value, the method of this paper can achieve an acceleration ratio of more than 10 times. However, for simulations with complex geometric models such as multi-size particles and drums with linings, the acceleration ratio will be reduced, though the acceleration effect is still evident.
作者
付帅旗
黄鹏
丁逸飞
FU Shuaiqi;HUANG Peng;DING Yifei(School of Mechanical Engineering,Southeast University,Nanjing 211189,China)
出处
《合肥工业大学学报(自然科学版)》
CAS
北大核心
2019年第12期1602-1607,共6页
Journal of Hefei University of Technology:Natural Science
基金
国家自然科学基金资助项目(51775109
51005047)