气固两相流拟颗粒并行模拟—二维区域分割法
文献类型：学位论文
| 作者 | 卢健新 |
| 学位类别 | 硕士 |
| 答辩日期 | 2008-06-12 |
| 授予单位 | 中国科学院过程工程研究所 |
| 授予地点 | 过程工程研究所 |
| 导师 | 葛蔚 |
| 关键词 | 并行计算 拟颗粒模拟 颗粒流体系统 气固流态化 |
| 其他题名 | Parallel Simulation of Gas-solid Systems in Pseudo-Particle Modeling—Two Dimensional Spatial Decomposition |
| 学位专业 | 化学工程 |
| 中文摘要 | 拟颗粒模拟自1996年提出以来在颗粒流体系统的离散模拟上得到了很多重要的应用，但串行计算限制了模拟的规模，为了将拟颗粒模型应用于更大规模体系的模拟，并行化成为必然选择。 本文在一维区域分割气固拟颗粒模拟的并行程序基础上实现了二维区域分割的并行算法。在本算法中，粒子在每一个时间步内独立运动，运动后移出进程区域边界的粒子经Shift通信模式传递到相应的目标进程。对于粒子碰撞的计算，首先进行进程区域内部粒子碰撞的计算；然后经过边界信息交换，相邻进程共享边界粒子信息后，再进行边界部分粒子碰撞的计算。针对基于硬球作用模型的拟颗粒模拟，本算法保证了在时驱算法下边界粒子在相邻的各进程内碰撞顺序的一致性。算法的实现使并行模拟的拓扑分区不再受流场区域分割数目的限制，适应于更大规模体系的模拟。 同时，本文针对上述并行算法进行了效率分析和优化，通过对程序中主要函数的耗时分析，发现进程分配的粒子数不均匀和粒子链表结构复杂是影响程序运行效率的重要原因。在程序的优化上，选择Intel®icc编译器且参数选择“-fast”时，可以使所测算例的程序耗时减少约20%。同时，通过在程序中使用动态负载均衡和数据链表重排，可以使程序运行时间在相同计算规模上减少了30%-50%。最后提出了进一步优化的方向。 作为应用，本文分别模拟了含11550个固体粒子、2.43×106个气体拟颗粒和含102400个固体粒子、1.8×107个气体拟颗粒的流态化体系，得到了与文献相一致的结果。最后利用所得模拟结果对能量最小多尺度模型(EMMS)稳定性条件成立的空间尺度进行了初步探索。 |
| 英文摘要 | Pseudo-Particle Modeling (PPM) has been applied in many particle fluid systems as a discrete simulation method since it was developed in 1996. But high computational cost is a main obstacle for large scale application of PPM. Parallel computing becomes a natural choice for this problem. In this dissertation a 2-D spatial decomposition parallel algorithm is developed and implemented based on a former program implementing 1-D spatial decomposition. In this algorithm, the particles move simultaneously in each step. The collisions of particles in the inner regions are processed first and boundary particle collisions are then processed after a parallel exchange of boundary particle data. As a hard-sphere model, the execution order of the collisions in PPM is significant to the results. This algorithm can assure the same time order for a boundary particle in its local processor and the neighboring processors. With this progress, the parallel simulation of PPM is no longer restricted to the number of decomposition of the flow field and can be applied in larger systems. To optimize the parallel algorithm, the time costs are recorded and analyzed. The inequity of the workload in different processors and the increasing complexity of data structure during the simulation process are part of the factors that affect the parallel efficiency. To deal with these problems, Intel®icc compiler is used and by using the parameter “-fast” the time cost of the test example can be reduced by about 20%. Also, the dynamic load balancing (DLB) and data array reset are introduced to the algorithms. The time costs are reduced by 30%-50% in the simulation with these treatments. For application, two fluidization systems are simulated, one contains 11550 solid particles and 2.43×106 pseudo particles and the other larger one contains 102400 solid particles and 1.8×107 pseudo particles. The results agree well with the results reported in literature. Also, the simulation results are used for preliminary exploration on the valid scale of stability condition of the Energy Minimization Multi-Scale (EMMS) model. |
| 语种 | 中文 |
| 公开日期 | 2013-09-13 |
| 页码 | 87 |
| 源URL | [http://ir.ipe.ac.cn/handle/122111/1209]  |
| 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 | 卢健新. 气固两相流拟颗粒并行模拟—二维区域分割法[D]. 过程工程研究所. 中国科学院过程工程研究所. 2008. |
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。