卷积神经网络在流场结构检测中的应用研究  

Application of Convolutional Neural Networks In Flow Field Structure Detection

作　　者：苏伟[1] 秦绪国[1] 王薇 李强[1] 刘文伶[1] SU Wei;QIN Xuguo;WANG Wei;LI Qiang;LIU Wenling(Beijing Institute of Space Long March Vehicle,Beijing,100076;The PLA Rocket Force Equipment Department the 1th Millitary Representative of Beijing,Beijing,100076)

机构地区：[1]北京航天长征飞行器研究所,北京100076 [2]火箭军装备部驻北京地区第一军代表室,北京100076

出　　处：《导弹与航天运载技术（中英文）》2023年第4期137-141,共5页Missiles and Space Vehicles

摘　　要：针对飞行器气动设计流场结构分析需求,提出了一种基于卷积神经网络的流场结构自动检测的方法。该方法充分利用神经网络对非线性系统隐藏特征的提取能力,通过学习高维度流场结构特征建立的网络可从CFD流场仿真结果中自主检测气动设计关心的流场结构。该方法采用监督学习和交叉验证方法对网络进行训练和验证,同时采用金字塔结构方法对待检测流场进行预处理,解决了结构检测尺度不匹配的问题并提高了检测正确率。以二维涡检测为例对方法进行了验证,最终识别了大部分区域的涡结构,结果表明该方法是有效的。该方法可进一步应用于三维分离涡、激波边界层干扰等复杂流场结构检测。To meet the needs of flow field structure analysis in flight vehicle aerodynamic design,an automatic detection method of flow field structure based on convolution neural networks is proposed.This method makes full use of the ability of neural networks to extract hidden features of nonlinear systems.By learning the high dimensional flow field structure features,the networks established can independently detect the flow field structure concerned by aerodynamic design from the CFD flow field simulation results.The method uses supervised learning and cross validation methods to train and validate the networks.Additionally,it employs a pyramid structure approach to preprocess the target flow field,effectively addressing the issue of scale mismatch and improving detection accuracy.The method is validated by a two-dimensional eddy detection example,and the successful identification of a majority of vortex structures shows that the method is effective.This method can be further applied to the detection of complex flow structures such as three-dimensional separated flow with vortices and shock wave boundary layer interference.

关 键 词：气动设计 卷积神经网络 结构特征提取 流场结构检测 金字塔结构方法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] V211[自动化与计算机技术—控制科学与工程]