AirfRANS

代理模型对于优化物理动力学中的有意义量至关重要，特别是在递归数值解法的情况下，其计算成本往往过高。这在流体动力学和纳维尔-斯托克斯方程的解析中尤为明显。然而，尽管数据驱动模型在物理系统中的应用领域不断扩大，但缺乏代表现实世界现象的参考数据集。在这项研究中，我们介绍了AIRFRANS数据集，该数据集用于分析二维不可压缩稳态雷诺平均纳维尔-斯托克斯方程在亚音速条件下的翼型，以及不同攻角下的情况。我们还通过检查几何表面上的应力力和可视化边界层来评估模型的能力。最后，我们针对四个机器学习任务提出了深度学习基线，以研究AIRFRANS在不同约束条件下的性能，包括大数据和稀缺数据、雷诺数和攻角外推。

Surrogate models are critical for optimizing meaningful quantities in physical dynamics, especially for recursive numerical solvers where computational costs are often prohibitively high. This is particularly evident in the analysis of fluid dynamics and the Navier-Stokes equations. However, despite the growing application of data-driven models in physical systems, there remains a lack of reference datasets that represent real-world physical phenomena. In this study, we introduce the AIRFRANS dataset, which enables analysis of airfoils under subsonic conditions using two-dimensional incompressible steady Reynolds-averaged Navier-Stokes equations across a range of angles of attack. We also evaluate model performance by examining stress forces on geometric surfaces and visualizing boundary layers. Finally, we propose deep learning baselines for four machine learning tasks to investigate the performance of AIRFRANS under various constraints, including large-scale and scarce data scenarios, as well as Reynolds number and angle of attack extrapolation.

搜集汇总

背景与挑战

背景概述

AirfRANS是一个空气动力学数据集，专注于二维不可压缩稳态雷诺平均纳维尔-斯托克斯方程在亚音速条件下的翼型分析，涵盖不同攻角情况。该数据集旨在为物理动力学中的代理模型优化提供现实世界参考数据，以解决计算成本过高的问题，并包含深度学习基线用于多个机器学习任务评估。