纳微流动与传递过程广泛存在于化学工程领域，诸如多相介质的界面、催化剂颗粒中的多级孔道和微化工系统等。随着对化工工艺和设备精准设计与调控要求的不断提高，认识这些过程的机理变得越发重要。在纳微尺度，对动态过程的全面实验研究还存在诸多困难，而传统的连续介质模拟方法随着系统特征努森数的增大逐渐失效。在该尺度下，流体更多地表现出分子的离散性质，因此各种离散粒子方法的应用近年来受到广泛重视，但其计算速度与精度之间的矛盾一直没有很好地解决。为此，本论文通过不同离散粒子方法的耦合建立了一套能够从微观水平高效准确地模拟纳微尺度或其他高努森数条件下气体流动、扩散和反应过程的模型和算法框架。论文的主要内容如下：绪论部分分析了软球、硬球、拟颗粒和直接模拟蒙特卡洛等典型的离散粒子模拟方法及其各自的优缺点以及前人在耦合不同模型方面的工作，据此提出了本论文的研究思路。即：以严格的软球模型（或其组合）模拟稠密或接近界面的复杂过程，而以简化的硬球模型（或其组合）模拟稀薄和远离壁面的条件下气体的流动与扩散过程，而以结合两者优势的拟颗粒模拟提供其过渡以及硬球并行模拟中局部的近似。第二章首先改进了事件驱动的硬球模拟与时间驱动的拟颗粒模拟的耦合与并行方法，严格确定了硬球与拟颗粒物性间的转换关系，并在管内流动和扩散模拟中验证了其正确性，表明了该耦合能有效克服硬球模拟扩展性差和拟颗粒模拟对稀薄气体效率低的问题，实现高效准确的大规模并行模拟。该方法还成功应用于气体在纳微孔道内的非平衡扩散以及在复杂多孔介质内的扩散研究，表明了其在微化工过程和催化剂开发等方面实际应用的可行性。另外，应用该方法还初步开展了高超声速流动的模拟，说明了其在航天航空等其他领域的潜在应用价值。第三章从算法改进和并行优化两方面深入研究了软球模拟方法，创新提出了关于粒子搜索的多壳层邻居列表算法。该算法通过对粒子跨越各壳层可能性的简单预估，有效提高了搜索效率。在此基础上，发展了针对软球模拟的多线程和向量化并行、众核与多核处理器协同、计算/通信/存取重叠等方法，建立了一套高效的大规模并行程序。基于上述工作，第四章最终实现了软球、硬球和拟颗粒模拟三者的耦合。论文导出了软球与拟颗粒模型间的参数转换关系，提出了采用不同模型的区域间通用的连接模式，在保持软球区域外部有拟颗粒过渡层的条件下，可构建任意复杂的界面。论文还为耦合模拟建立了简单几何体和固定粒子等多种边界条件，提高了方法的实用性。论文还通过经典的管流模拟等验证了该方法的正确性。第五章应用上述耦合模拟方法通过简单的概念模型研究了气固界面反应中界面结构对扩散和反应过程的影响。研究发现，在给定的反应条件下，由于扩散和反应过程在不同条件下相互影响的不同方式，界面结构存在最有利于整体反应速率的特定形状参数。通过建立更真实与细致的反应物及界面模型，该耦合方法有望提供诸如催化剂孔道结构设计等方面的机理分析与优化工具。第六章概括了论文的主要结论和创新点，并展望了后续工作。 

Flow and transportation at nano-/micro-scales is ubiquitous in chemical engineering such as at the interface of different phases in the multilevel pores of catalytic particles and in micro-chemical engineering systems. With the ever increasing demand of chemical engineering for more precise design and control of the processes and equipment, deeper understanding of its behaviors and mechanism is becoming more and more important. However, comprehensive investigation of the dynamic processes at nano-/micro-scales is still difficult in experiments, and traditional simulation methods based on the continuum hypothesis gradually become invalid with the increase of the system’s characteristic Knudsen number (Kn). At such scales, the fluids show more molecular discrete nature, so discrete particle methods have received more and more attention in recent years. However, the contradiction between the computing speed and accuracy has not been solved satisfactorily. In this thesis, a set of coupled models and algorithms combining the merits of different discrete particle methods are developed which can effectively and precisely simulate flow, diffusion and reaction processes at nano-/micro-scales or other systems with high Knudsen numbers. The main contents of the thesis are summarized as the following: In chapter 1, the commonly used discrete particle models, such as soft spheres (SS), hard spheres (HS), pseudo particles (PP) and direct simulation of Monte Carlo (DSMC), are briefly introduced including their advantages and disadvantages and previous work about these methods, based on which the main ideas of the thesis are proposed: The complex process near the interface or in high density regions could be simulated with SS (or their combinations) while the flow and diffusion process far from the boundary or in low density regions could be simulated with HS (or their combinations). PP with some merits of HS and SS could bridge the two models and provide local approximations in the parallelization of HS simulation.In chapter 2, the coupling between the event-driven HS simulation and the time-driven pseudo-particle modeling (PPM) is optimized and validated with pipe flow simulation. Rigorous mapping between HS and PP properties is also provided. It is proven that the optimized method can carry out parallel simulations at large scales, which can enhance the scalability of HS simulation and improve the efficiency of PPM for rarefied gas. Non-equilibrium diffusion in nano-scale channels and in complex porous media is also simulated demonstrating its potential applications to micro-chemical engineering and catalyst development. The method is also applied to supersonic flows to reveal its prospect in the aerospace field.In chapter 3, SS simulation is improved in algorithm and optimized for parallelization. A multilevel-skin neighbor-list algorithm is developed, which can enhance the searching efficiency substantially by sorting and tracing the particles of the skin domain with a simple prediction procedure. A complete set of optimization schemes including parallelization using multi-thread and vectorization techniques, multi-core and many-core hybrid parallel computing, overlapping of data caching, transferring and computing is developed to build a set of high performance parallel programs which can be implemented at large scales. In chapter 4, HS-PP-SS simulation method is finally built. A universal coupling scheme for different models is provided which supports arbitrary interfaces under the condition of a layer of PP outside SS regions where particle properties are adjusted based on theoretically derived rules. Several types of boundary conditions, such as simple geometry structure or fixed particles are supported and can be used in combination, facilitating the practical application of the method. The method is validated in classical pipe flow simulation.In chapter 5, HS-PP-SS coupled method is applied to study the influence of interface structure on the coupling of the diffusion and reaction processes in gas-solid interfacial reactions under idealized reaction scenarios. It is found that, under given reaction conditions, an optimal interface structure exists for the highest overall reaction rate, reflecting the best coordination of the diffusion and reaction processes. By establishing more realistic and precise reaction and interface models, the coupled method will be useful for analyzing the mechanism of catalytic reactions and optimization of the catalyst and catalytic particle structures.Chapter 6 summaries the main conclusions and innovations of the thesis and provides prospects on future studies.