Building-Block-Flow Model for Large-Eddy Simulation

计算流体动力学是气动/水动力飞行器设计和优化的重要工具。据估计，减少5%的运输阻力所产生的影响相当于美国风能产量翻一番。然而，由于复杂的流动物理现象无处不在，包括不利的压力梯度效应、流动分离和层流到湍流的转变，对现实车辆周围流体流动的计算预测提出了一个独特的挑战。虽然一些计算模型预测了一种或两种情况，但没有一个模型能准确地预测所有的流动现象。该项目将寻求为计算流体动力学设计一个统一的封闭模型，能够解释丰富的流动物理集合。该项目的目标是将基础物理和机器学习建模结合起来，建立一个新的计算流体模型。该项目还利用现有项目促进工程领域的多样性和包容性，包括参与年度夏季研究项目和本科生研究机会，以吸引女性和代表性不足的少数族裔。闭包模型提出的核心假设是，一组有限的简单规范流包含了预测更复杂场景的基本物理。该方法使用人工神经网络和大涡模拟来实现，并汇集了五个独特的进步：(1)该模型可直接适用于任意复杂几何形状；(2)该模型可用于预测不同的流动形式（零/有利/不利平均压力梯度壁面湍流、分离、统计上不稳定的平均三维湍流和层流）；(3)如果需要，该模型可按比例放大以捕获额外的流动物理（例如激波）；(4)该模型通过补偿数值误差，保证了数值离散化和网格化策略的一致性；(5)该模型的输出在预测中附带一个置信度分数，用于不确定性量化和网格细化。研究的案例范围从典型的平板湍流到复杂的流动，如现实的飞机结构。在这项工作中建立的基础将为计算流体动力学中的多种流动模式提供新的场所。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。

Computational fluid dynamics stands as an essential tool for the design and optimization of aerodynamic/hydrodynamic vehicles. It is estimated that the impact of reducing transportation drag by 5% would be equivalent to that of doubling the US wind energy production. However, computational predictions of fluid flows around realistic vehicles poses a unique challenge due to the ubiquity of complex flow physics, including adverse pressure-gradient effects, flow separation, and laminar-to-turbulent transition. While some computational models predict one or two scenarios, no model performs accurately across all flow phenomena. This project will seek to devise a unified closure model for computational fluid dynamics capable of accounting for a rich collection of flow physics. The goals of this project are to couple fundamental physics and machine-learning modeling for a new computational fluids model. The project also leverages existing programs to promote diversity and inclusion in engineering, including participation in annual summer research programs and undergraduate research opportunities to engage women and underrepresented minorities.The core assumption of the closure model proposed is that a finite set of simple canonical flows contains the essential physics to predict more complex scenarios. The approach is implemented using artificial neural networks with large-eddy simulation and brings together five unique advances: (1) the model is directly applicable to arbitrary complex geometries, (2) it is constructed to predict different flow regimes (zero/favorable/adverse mean-pressure-gradient wall turbulence, separation, statistically unsteady turbulence with mean-flow three-dimensionality, and laminar flow), (3) the model can be scaled-up to capture additional flow physics if needed (e.g., shock waves), (4) the model guarantees consistency with the numerical discretization and the gridding strategy by compensating for numerical errors, and (5) the output of the model is accompanied by a confidence score in the prediction used for uncertainty quantification and grid refinement. The cases of study range from canonical flat plate turbulence to complex flows such as realistic aircraft configurations. The foundations established in this work will enable new venues to model multiple flow regimes in computational fluid dynamics.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.