QQ客服 官方微博 反馈留言

题名	气固两相流大规模直接数值模拟—GPU实现及统计性质分析
作者	熊勤钢
学位类别	博士
答辩日期	2011-05-31
授予单位	中国科学院研究生院
授予地点	北京
导师	葛蔚
关键词	直接数值模拟 气固两相流 介尺度结构 GPU计算 宏观拟颗粒方法
其他题名	Large-scale direct numerical simulation of gas-solid flow—its GPU implementation and statistical analysis
学位专业	化学工程
中文摘要	气固两相流是典型的非线性非平衡的系统，具有显著的动态多尺度结构，其中介尺度流动结构可表现为流体为主的稀相和颗粒聚集的密相(如以团聚物的形式)共存，并伴有错综复杂的动态变化。现有的基于连续介质的模拟方法，如双流体模型(two-fluid mode, TFM)，如不合理地考虑此介尺度结构的影响将使计算结果严重失真。虽然近年来出现了对传统双流体模型的改进，如细网格模拟等，但其理论基础和预测能力还受到诸多限制。同时能量最小多尺度(EMMS)模型虽然在气固系统的模拟中获得了较高的预测精度，其稳定性假设的验证和模型方程的改进还需要深入的系统微观信息。直接数值模拟(DNS)能在颗粒尺度以下获得流场的详细信息且不需要任何经验关联式，因此受到了越来越多的关注。鉴于这种状况，本文利用最新的GPU加速技术实现了气固悬浮系统的大规模DNS，并从多方面分析了系统的统计性质。在第一章文献调研与综述的基础上，论文第二章实现了改进的宏观拟颗粒方法在GPU集群上对包含上万个固体颗粒的气固悬浮系统的大规模DNS。发现在中等固/气密度比下，随着计算规模逐渐增大，模拟结果表现出明显的尺度效应，并在较大规模下初步展现出尺度无关性。在对高固/气密度比气固悬浮系统的DNS中，较完整地讨论了亚网格尺度的非均匀结构，定性地证明了EMMS模型的正确性。同时对团聚物大小和分布等进行了探讨，发现DNS的结果与实验吻合较好。论文第三章以格子Boltzmann方法(lattice Boltzmann method, LBM)，浸没移动边界法(immersed moving boundary, IMB)和离散单元法(discrete element method, DEM)为基础，在过程所最新的GPU超级计算系统—Mole-8.5上实现了更大规模(提升了两个量级)的气固系统DNS。这些模拟展示了团聚物的大尺度特征以及团聚物间的相互作用关系，为深入研究气固系统打下了基础。论文第四章以细网格TFM和DNS的对比为起点，阐明了即使在TFM中网格足够精细，介尺度结构的影响仍可能被低估。随后利用DNS在网格尺度上定量验证了EMMS模型中的稳定性假设，为EMMS模型在亚网格尺度的使用提供了依据。同时详细讨论了团聚物形成与演化的过程，并定性地表明了EMMS模型对气固系统的稀密相和相界面结构处理的合理性。论文第五章总结了本论文获得的主要成果，展望了DNS在气固系统介尺度研究方面的应用前景以及进一步研究的方向。
英文摘要	Gas-solid flow is a typical non-equilibrium and non-linear system, with complicated dynamical multi-scale structures. At the meso-scale, gas-rich dilute phase and particle-rich dense phase (e.g., clusters) co-exist, displaying complex dynamic behavior. The widely used continuum modeling approaches, such as two-fluid model (TFM), are unable to obtain accurate predictions unless meso-scale structures are reasonably accounted for. Though some improvements have emerged in recent years such as fine-grid TFM, their theoretical foundations and predictive capabilities are still under a lot of constraints. At the same time, though energy-minimization multi-scale (EMMS) model has achieved relatively better predictions, verifications of its stability hypotheses and further improvements of the model equations also need understanding micro-scale details. Direct numerical simulation (DNS), which can provide information below particle scale without any empirical correlations, becomes a hotspot recently. In view of this situation, the emerging GPU computing technology is employed to carry out large scale DNS and the simulation results are analyzed from different aspects. Based on the literature review in chapter 1, we implement in chapter 2 the revised macro-scale pseudo-particle model on a GPU cluster for the DNS of gas-solid suspensions containing more than ten thousand particles. It is found that in moderate solid/gas density ratios, distinct scale effect can be observed with the computational domain increases. As for relatively large domains, scale independence is obtained preliminarily. In high solid/gas density ratio suspensions, we discuss the sub-grid-scale structures in detail and verify the EMMS model qualitatively. Furthermore, the results about cluster diameter and its distributions are found to be in good agreement with experiments. In chapter 3, we carry out even larger scale gas-solid DNS on the newest GPU cluster established in Institute of Process Engineering, Chinese Academy of Sciences—Mole-8.5, based on lattice Boltzmann method (LBM), immersed moving boundary (IMB) and discrete element method (DEM), which elevates simulation scale by two orders. Large-scale clusters and their interactions are reproduced, which we believe is very useful for our further investigations of gas-solid systems. In chapter 4, taking the comparison between DNS and fine-grid TFM, we emphasize that even if the cell size in fine-grid TFM is chosen extremely small, meso-scale structure effects may still be underestimated. Then we quantitatively verify the stability hypothesis in EMMS model at element scale, which validates the application of the EMMS model at sub-grid scale. At the same time, we discuss the process of cluster formation and evolution in detail, showing the rationale of resolving a non-uniform gas-solid system into two phases to account for inter-phase interactions. Finally, chapter 5 concludes this study, and discusses future work on DNS of gas-solid flows.
语种	中文
公开日期	2013-09-23
页码	102
内容类型	学位论文
源URL	[http://ir.ipe.ac.cn/handle/122111/1691]
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	熊勤钢. 气固两相流大规模直接数值模拟—GPU实现及统计性质分析[D]. 北京. 中国科学院研究生院. 2011.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

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。