基于GPU的离散模拟在颗粒流动与混合机理研究中的应用
文献类型：学位论文
| 作者 | 戚华彪 |
| 学位类别 | 博士 |
| 答辩日期 | 2014-06 |
| 授予单位 | 中国科学院研究生院 |
| 导师 | 李静海 |
| 关键词 | 离散单元法 滚筒 螺旋输送器 混合 相似性 |
| 其他题名 | Application of GPU-based Discrete Simulation to Flow and Mixing Mechanisms of Granular Materials |
| 学位专业 | 化学工程 |
| 中文摘要 | 离散单元法（Discrete Element Method）是颗粒流模拟研究的主流方法，且发展迅速、应用广泛，能有效的揭示出实验难以获取的流动信息。然而，其计算量大、计算时间长、模拟规模小等缺陷严重制约了其应用和发展。针对此问题，本论文在前期及合作工作基础上设计并实现了面向多尺度异构超级计算系统的高效大规模离散单元模拟程序。应用该方法及程序定量研究了三维水平滚筒中颗粒流动速度场的相似性问题以及工业尺度的螺旋输送器中物料的混合问题，从机理和应用研究两方面阐明了该方法及模拟程序的有效性。 所开发的耦合使用中央处理器(Central Processing Unit, CPU)和图形处理器(Graphics Processing Unit, GPU)的高效大规模DEM模拟程序，在单节点测试中较单纯使用CPU快达25倍，并在多节点CPU/GPU并行时具备良好的可扩展性，为随后的实例研究奠定了基础。 论文以此研究了处于Rolling流型的三维水平滚筒，重点考察了颗粒的几何参数及各种物理属性对其稳态速度场的影响。研究表明：这些参数的影响程度不同，而粒径比具有一定的支配作用。同时，颗粒的滑动摩擦系数和杨氏模量对速度场的影响较大。对不同的滚筒直径，只要保证模拟系统的粒径比一致，稳态速度场将基本保持不变。由于颗粒物质的离散属性以及其相互作用的复杂性，即使在相对简单的Rolling流型中，滚筒的稳态速度场也呈现丰富的内部结构。 本文还对工业尺度螺旋输送器中物料的流动状态进行了大规模模拟。通过统计物料的混合指标和停留时间分布考察了其几何结构和操作条件对物料混合效果的影响。研究表明，螺旋输送器的混合性能强烈地依赖于操作条件和结构尺寸，转速和物料的添加速率对物料的整体混合和轴向混合效果影响最大，其次是混合段螺距和混合区域的长度。该结论对在螺旋输送器的工业设计中根据生产的具体要求合理地选择技术参数、降低能耗具有指导意义。 论文最后总结了所获得的主要成果，展望了离散单元模拟在研究颗粒流动的内在机理和工业应用方面的前景。 |
| 英文摘要 | In contrast to physical experiments, the discrete element method (DEM) could readily reveal the details of granular flows at the particle level, and thus has been extensively employed in both fundamental research and engineering practice. Nevertheless, most applications are restricted to small scale problems due to its large computational cost. Based on previous work, a parallel simulation software is developed, coupling GPU(Graphics Processing Unit)-CPU(Central Processing Unit) computing for non-linear DEM models, and is implemented on a multi-scale heterogeneous supercomputer system. With this software, the similarity of velocity fields in three dimensional horizontal drum and the mixing mechanisms of an industrial scale screw conveyor are investigated. In Chapter 2, the implementations of this DEM software are reported in details, which combine the advantages of both CPU and GPU computing with extensive optimization. In the performance test based on a single computer node, this code runs almost twenty-five times faster than pure CPU code. Additionally, it gives quite good scalability in the parallel computations with multiple nodes of CPU/GPUs. In Chapter 3, the similarity of velocity fields in the rolling regime of rotating drums is assessed for various size ratios and material properties, using a rigorous statistical method. It is found that both the size ratio and material properties affect the velocity fields, but the size ratio has a dominant effect. Besides, the sliding friction coefficient and Young’s modulus have relatively larger impact than other factors of material properties. In regard to the effect of drum diameter, the velocity fields would almost remain unchanged when a constant size ratio is guaranteed. The mixing mechanisms of an industrial-scale screw conveyor are examined in Chapter 4. The influences of geometry and operating conditions on mixing performance are evaluated with respect to the mixing index and residence time distribution. It is found that the screw speed and feed rate have dominant effects on the overall mixing and axial mixing, while the lengths of screw pitch and mixing region have a secondary effect. Therefore, to guarantee both reasonable energy loss and good mixing performance, the relevant parameters in the design as well as operations of screw conveyor should be selected according to specific requirements. In the final chapter, the important findings of this thesis are summarized, and the future work on DEM simulation for fundamental research and industrial applications is prospected. |
| 语种 | 中文 |
| 公开日期 | 2015-07-08 |
| 源URL | [http://ir.ipe.ac.cn/handle/122111/15519]  |
| 专题 | 过程工程研究所_研究所（批量导入） |
| 推荐引用方式 GB/T 7714 | 戚华彪. 基于GPU的离散模拟在颗粒流动与混合机理研究中的应用[D]. 中国科学院研究生院. 2014. |
入库方式： OAI收割
来源：过程工程研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。