基于深度学习的超分辨率重构方法在CAARC标模绕流流场重构中的应用  

Applications of deep learning-based super-resolution for reconstruction of flow around the CAARC benchmark model

作　　者：梁仍康 张伟[1,3] 杨思帆 黄刚 李朗 LIANG Rengkang;ZHANG Wei;YANG Sifan;HUANG Gang;LI Lang(School of Civil Engineering and Architecture,Southwest University of Science and Technology,Mianyang 621010,China;Hypervelocity Aerodynamics Institute of China Aerodynamics Research and Development Center,Mianyang 621000,China;State Key Laboratory of Aerodynamics,Mianyang 621000,China)

机构地区：[1]西南科技大学土木工程与建筑学院,绵阳621010 [2]中国空气动力研究与发展中心超高速空气动力研究所,绵阳621000 [3]空天飞行空气动力科学与技术全国重点实验室,绵阳621000

出　　处：《空气动力学学报》2023年第11期116-126,I0002,共12页Acta Aerodynamica Sinica

基　　金：国家自然科学基金(11902278);光合基金A类项目(202302015072);空天飞行空气动力科学与技术全国重点实验室项目。

摘　　要：基于深度学习的超分辨率重构方法是近年来发展的一种有效的流场精细化方法。本文超分辨率重构模型以卷积神经网络为基础,结合了混合下采样跳跃连接多尺度模型,并应用于CAARC标准建筑模型表面风压场和建筑绕流速度场的重构。通过对比分析对不同欠分辨率流场的高分辨重构能力,结果表明该深度学习模型重构高分辨率流场具有良好的精度,重构效果优于原始的卷积神经网络模型和传统的双三次插值方法。该方法具有一定的普适性,可推广应用到具有复杂湍流流动的任意建筑结构风场的超分辨率重构。The deep-learning-based super-resolution reconstruction methods developed in recent years are effective methods to obtain detailed flow fields.A deep learning-based super-resolution reconstruction method was applied to reconstructing high-resolution wind field of flow around building structures in this paper.The super-resolution reconstruction model was based on the convolutional neural network(CNN)and combined with the mixed downsampled skip-connection multi-scale(Multi-scale CNN)model.The super-resolution reconstruction model was applied to the reconstruction of the surface pressure on and the velocity field around the CAARC benchmark model.The reconstruction ability of the deep learning-based model for different under-resolution flow fields was investigated.The results show that the proposed deep learning model can greatly enhance the super-resolution reconstruction performance and the reconstruction accuracy is better than the original convolutional neural network model and the traditional bicubic interpolation method.Due to its universal applicability,this method can be extended to super-resolution reconstruction of wind field of any building structure with complex turbulent flow.

关 键 词：深度学习 超分辨率重构 卷积神经网络 CAARC 

分 类 号：O355[理学—流体力学] TP391.7[理学—力学] TU972.8[自动化与计算机技术—计算机应用技术]