Bridging Particle-Resolved and Point-Particle Based Simulation for Turbulent Particle-Laden Flow Using New Heterogeneous High-Performance Computer

1235974 PI：在工程、生物和环境等领域，充满固体颗粒、小液滴和气体微泡的湍流流动是普遍存在的。在这些应用中，颗粒通常悬浮在湍流载体流体中。分散相和载体流体相之间的相互作用影响悬浮颗粒的动力学（例如，分散、沉积速率、碰撞速率、沉降速度）和多相流的整体性质（例如，壁或表面阻力、湍流强度）。了解湍流颗粒流可以帮助改进工程设备，如煤燃烧器，并更好地预测自然现象，如暖雨和飓风。 在过去的20年中，计算方法已经开发出来，以解决这些复杂的多尺度流，主要使用基于点粒子的模拟，其中有限的颗粒尺寸的影响被忽略。然而，在许多应用中，颗粒尺寸与湍流尺度重叠，需要一种更好的方法，称为颗粒分辨模拟，以充分解决分散相和流体相之间的耦合。本研究的总体目标是开发一种有效的粒子分辨模拟方法，以研究一系列重要的物理问题，从颗粒尺寸和流动耗散尺度到粗粒度系统尺度。该研究将利用介观晶格玻尔兹曼方法，将其数据局部性和算法可扩展性有效地映射到配备多核CPU和高性能GPU的异构PetaScale计算机。将利用格子玻尔兹曼算法在处理相之间的界面相互作用的灵活性。几个基准情况下，将模拟验证的方法和代码。将系统地研究颗粒尺寸、颗粒与流体密度比、体积分数和重力对两相相互作用动力学的影响。然后，这些代码将被移植到不同的高性能计算机上，以实现一致的可持续扩展性。通过扩展合作，该方法将用于解决工业设备中的颗粒粗糙壁冲击动力学和海洋大气边界层中的海雾对表面阻力的调制。 开发的代码将被转换为公共领域的软件，以允许其他人将该方法用于许多其他工程，生物和环境应用。该计算工具可用于解决许多涉及湍流载体流中移动物体的应用，如流化床、沉积物输送、海雾和暖雨发展。该研究将有助于将复杂湍流多相流的计算转移到主流的高性能、GPU加速的多核异构计算机上。这些能力将影响未来的研究方向在湍流多相流和并行计算。该项目为一名研究生，一名博士后和两名早期职业合作者提供了跨学科的培训和指导。该项目将为特拉华州大学的计算科学与工程新的多学科研究生证书课程做出贡献，该课程每年可能会影响几十名学生。

1235974PI: WangTurbulent flows laden with solid particles, small droplets, and gas microbubbles are ubiquitous in engineering, biological and environmental applications. In these applications, particles are usually suspended in a turbulent carrier fluid. The interactions between the dispersed phase and the carrier fluid phase impact the dynamics of suspended particles (e.g., dispersion, deposition rate, collision rate, settling velocity) and the bulk properties of the multiphase flow (e.g., wall or surface drag, turbulence intensity). Understanding turbulent particle-laden flows can help improve engineering devices such as coal combustors and better predict natural phenomena such as warm rain and hurricane. In the last 20 years, computational methods have been developed to address these complex multiscale flows, primarily using the point-particle based simulations where the effect of finite particle sizes has been ignored. However, in many applications where the particle size overlaps with turbulent flow scales, a better approach known as particle-resolved simulations is necessary to fully address the coupling between the dispersed phase and the fluid phase. The overall goal of this research is to develop an efficient particle-resolved simulation approach to study a range of important physical issues from the particle size and flow dissipation scales to coarse-grained system scales. The study will make use of the mesoscopic lattice Boltzmann approach to efficiently map its data locality and algorithmic scalability to heterogeneous PetaScale computers equipped with both multicore CPUs and high-performance GPUs. The flexibility of the lattice Boltzmann algorithm in treating interfacial interaction between the phases will be exploited. Several benchmark cases will be simulated to validate the approach and the codes. The effects of particle size, particle-to-fluid density ratio, volume fraction, and gravity on the interaction dynamics of both phases will be systematically studied. These codes will then be ported to different high-performance computers to achieve a consistent sustained scalability. Through extended collaborations, the approach will be used to address particle-rough wall impact dynamics in industrial devices and surface drag modulation by sea sprays in marine atmospheric boundary layer. The developed codes will be converted to public-domain software to allow others to use the approach for many other engineering, biological and environmental applications. The computational tool can potentially be used to address many applications involving moving objects in a turbulent carrier flow, such as fluidized bed, sediment transport, sea sprays, and warm rain development. The research will help move the computation of complex turbulent multiphase flow to the mainstream high-performance, GPU-accelerated, multicore heterogeneous computers. These capabilities will impact future research directions in both turbulent multiphase flows and parallel computation. The project provides an interdisciplinary training and mentoring ground for one graduate student, one postdoc, and two early-career collaborators. The project will contribute to a new multidisciplinary graduate certificate program in Computational Science and Engineering at the University of Delaware, which could impact a few dozens of students each year.