CDS&E: Collaborative Research: CDS&E: Advances in closure modeling for turbulent flows with finite-sized particles informed by massive simulations on heterogeneous architec

在许多工业系统和许多自然过程中，浸泡在流体（无论是气体还是液体）中的固体颗粒悬浮物都很重要。多相计算流体力学预测这些系统的行为依赖于对适当的方程进行近似。因此，一些重要的现象，如固体颗粒在流动中的聚集，不能准确地预测。气固流动中颗粒团簇的形成是颗粒与气体以及颗粒之间复杂相互作用的结果。这个协作的计算和数据驱动科学与工程（CDS&amp；E）项目将开发一个模型框架，通过使用高度可扩展的代码进行大规模模拟，实现气体流动问题中的粒子聚类。三个参与机构的研究人员开发的数值技术将被整合，以形成一个一致和可扩展的框架，用于解决粒子聚类的重要问题。研究结果将使研究人员能够解决涉及颗粒团簇的问题，如化学环燃烧、生物质快速热解和二氧化碳捕获应用。该项目产生的计算数据将通过一个基于网络的界面提供给研究人员。研究人员将利用该项目开发教育工具，以吸引广大公众，特别是中学生和高中生，展示气体中颗粒悬浮液流动的基本原理。粒子解析直接数值模拟（PR-DNS）可以解决微观尺度的粒子-流体相互作用，而点-粒子直接数值模拟（PP-DNS）可以模拟粒子间的碰撞，将与伪谱技术相结合，这有可能加速气固流动的计算。将开发高度可扩展的PR-DNS和PP-DNS代码，以利用以下方面的新算法进展：(a)在异构架构上使用伪谱方法进行湍流模拟；(b)在欧拉-拉格朗日多相流求解器中使用粒子数进行有效缩放，以在演示粒子聚类所需的尺度上对典型聚类流执行大规模PR-DNS和PP-DNS。将为体积过滤欧拉-拉格朗日公式以及PP-DNS和PR-DNS开发一致的可扩展闭包。将对PR-DNS数据应用一致的平均操作，该操作明确考虑了已分解的尺度，从而产生新的未封闭项，这些项尚未在两阶段上下文中量化。使用机器学习的新颖闭包技术将用于评估新术语的函数形式。为了最大限度地提高它们对模型开发的影响，本项目中生成的源代码和规范多相流数据将被存档在一个开放访问的数据库中，供更广泛的科学界使用。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Suspensions of solid particles immersed in a fluid – either gas or liquid – are important in many industrial systems and in many natural processes. Multiphase computational fluid dynamics to predict the behavior of these systems relies on making approximations to the appropriate equations. As a result, some important phenomena, such as the aggregation of solid particles in the flow, are not accurately predicted. The formation of particle clusters in gas-solid flows results from complex interactions of the particles with the gas and the particles with each other. This collaborative Computational and Data-Enabled Science and Engineering (CDS&E) project will develop a modeling framework for particle clustering in gas-flow problems enabled by massive simulations using highly scalable codes. Numerical techniques developed by the researchers at the three participating institutions will be integrated to formulate a consistent and scalable framework for solving problems in which particle clustering is significant. The results will enable researchers to tackle problems involving particle clusters such as chemical looping combustion, biomass fast pyrolysis and carbon dioxide capture applications. Computational data generated in the project will be available to researchers through a web-based interface. The researchers will use the project to develop educational tools to engage a broad public audience, especially middle and high school students, in displays of the fundamental principles underlying flows of particle suspensions in a gas.Particle-resolved Direct Numerical Simulation (PR-DNS), which can resolve microscale particle-fluid interactions and point-particle direct numeric simulation (PP-DNS), which can model inter-particle collisions, will be integrated with pseudo-spectral techniques, which have the potential to accelerate computations for gas-solid flows. Highly scalable PR–DNS and PP-DNS codes will be developed to leverage new algorithmic advances in (a) turbulence simulation using the pseudo-spectral approach on heterogeneous architectures and (b) efficient scaling with number of particles in Eulerian-Lagrangian multiphase flow solvers to perform massive PR–DNS and PP-DNS of canonical clustering flows on scales needed to demonstrate particle clustering. Consistent scalable closures will be developed for volume-filtered Eulerian-Lagrangian formulations and PP-DNS and PR-DNS. A consistent averaging operation will be applied on the PR–DNS data that explicitly takes the resolved scales into account, resulting in new unclosed terms that have yet to be quantified in a two-phase context. Novel closure techniques using machine learning will be employed to evaluate the functional form of the new terms. To maximize their impact on model development, the source codes and canonical multiphase flow data generated in this project will be archived in an open-access database available to the broader scientific community.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.