气固流动是典型的非线性非平衡系统，介尺度结构是此类系统研究的重点和难点。从实验和理论的角度研究介尺度结构还存在很大困难，数值模拟已经成为研究气固流动不可或缺的重要手段。但传统的模拟方法，如双流体模型和离散颗粒模型，由于理论基础和解析度的限制无法全面描述介尺度结构的形成和演化过程，模拟精度也受到很大限制。解析颗粒的直接数值模拟（particle-resolved direct numerical simulation，PR-DNS或简称DNS）能够跟踪每个颗粒的运动并在小于颗粒的尺度上描述颗粒周围的动态流场，对探索气固系统中介尺度结构的形成机理、准确描述系统在连续介质尺度上的行为以及为连续介质尺度上的模型提供可靠的本构关系具有重要意义。本论文开展了一系列气固流动的大规模DNS，以探究准确描述和表征介尺度结构的方法，特别是稀、密两相结构的划分方法，深入分析介尺度结构对气固系统底层统计性质和流体动力学特征的影响，提出了颗粒脉动速度概率密度函数和局部平均无量纲曳力的关联式。论文第一章概述了当前研究气固两相流的手段和识别其中介尺度结构的常用方法，回顾并总结了颗粒统计性质和流固相间作用的研究进展，包括颗粒脉动速度分布、均匀与非均匀曳力模型等。第二章论述了气固系统大规模DNS的常用方法并论证了格子Boltzmann方法、离散单元方法与浸入移动边界法的耦合对本研究的适用性。通过系综平均流体动力学量的网格无关性和流场的自相关性分析选取了合理的模拟参数。详细讨论了由此模拟的气固流动从初始均匀分布到失稳再到介尺度结构形成，直至达到统计稳态的动力学过程和宏观变量的演化。第三章为量化介尺度结构对统计性质的影响，分别利用由Voronoi多面体定义的局部颗粒浓度和基于Voronoi空间分割的聚类分析，将气固两相流中的非均匀结构划分为颗粒聚集的密相（团聚物）和流体聚集的稀相。通过分析颗粒的统计性质给出了稀、密两相划分的合理依据。第四章主要考察了介尺度结构对颗粒统计性质的影响。发现颗粒脉动速度分布以及颗粒温度具有显著的尺度依赖性和各向异性，气固流动呈现局部非平衡特征。提出了考虑稀、密两相中颗粒动力学行为差异的颗粒脉动速度分布函数，该分布函数在不同尺度上均与DNS结果吻合。第五章重点考察了介尺度结构对曳力的影响。从微观上揭示了气固流动中局部平均无量纲曳力的尺度依赖性，探究了在不同尺度上非均匀结构、颗粒温度对无量纲曳力的影响，利用DNS结果拟合了考虑气固流动尺度效应的曳力关联式。论文最后总结了本研究的主要成果，展望了利用大规模DNS研究气固流动系统底层机理的方向以及机器学习在其中的应用。;Gas-solid flows are typical non-linear and non-equilibrium systems and understanding the meso-scale structures in these systems is a challenging but attractive research topic for which experimental measurements and theoretical analysis are of great difficulty, while numerical simulation has been an indispensable approach. With little consideration to the effect of meso-scale structures, traditional simulation methods such as two-fluid model and discrete particle model can’t provide accurate prediction of the flow behavior. The particle resolved direct numerical simulation (PR-DNS or DNS for short) is able to track the motion of each individual particle and fully resolve the dynamic flow field around the particles. Therefore, it is very important for exploring the formation mechanism of meso-scale structures in gas-solid flows, predicting the behavior at continuum scale and establishing reliable constitutive relations for continuum models.Meso-scale structures in gas-solid flows are the focus of this dissertation, for which a series of DNS has been carried out, and the methods for describing and characterizing the meso-scale structures have been investigated together with the partition of between dilute and dense phases. The effects of meso-scale structures on both statistical properties at micro-scale and hydrodynamic characteristics have been analyzed, and the probability density function of particle fluctuating velocity and a correlation of local-averaged dimensionless drag have been proposed and validated.First, the traditional methods for simulating gas-solid flows and analyzing the state of the art of meso-scale structures therein are reviewed in Chapter 1. The statistical studies on particle phase properties and particle-fluid interactions are summarized, including probability density function of particle fluctuating velocity and drag models.In Chapter 2, the essential numerical procedure for large-scale DNS of gas-solid flows is presented, in which the lattice Boltzmann method, discrete element method and immersed moving boundary are coupled. The simulation settings and the parameters at the simulation cases are explained and validated by analyzing the scale-independence of the domain-averaged hydrodynamic variables on resolution and the auto-correlation function of the flow field. The evolution of the structures and macroscopic hydrodynamic characteristics in the process from homogeneous particle suspension, local instability, aggregates growth to statistically steady state is discussed in detail based on the DNS data.In Chapter 3, the gas-solid flows are partitioned into a gas-rich dilute phase and a solid-rich dense phase for quantifying the effect of meso-scale structures on statistical properties, by quantifying local solid volume fraction and clustering analysis based on Voronoi tessellation. An intrinsic threshold for the partitioning of dilute and dense phases is provided by analyzing the statistical properties of particle in dilute and dense phases.In Chapter 4, the effect of meso-scale structures on statistical properties of the particles are investigated. It is found that both the fluctuating velocity distribution and granular temperature are scale-dependent and anisotropic, and the flows are locally non-equilibrium. A probability density function of the particle fluctuating velocity is proposed by considering the difference of particle kinetic properties between dilute and dense phases, which is in good agreement with the DNS data at different scales.In Chapter 5, the effect of meso-scale structures on drag is investigated. The scale-dependence of local-averaged dimensionless drag is revealed by DNS data. The roles of heterogeneity and granular temperature on the dimensionless drag at different scale are clarified. By fitting the DNS data, a drag correlation with consideration of scale-dependence is obtained.Finally, the main results of this dissertation are summarized. The future of large-scale DNS for exploring the underlying mechanism of gas-solid flows and their constitutive laws through machine learning are prospected.