为进一步提升高雷诺数、大迎角(Angle of attack,AoA)和高马赫数下的翼型可压缩流场预测精度和效率,本文提出了一种基于坐标转换方法和UNet神经网络的机器学习推理方法。首先,提出了用于数据前处理的坐标转换方法,将计算流体力学中的物...为进一步提升高雷诺数、大迎角(Angle of attack,AoA)和高马赫数下的翼型可压缩流场预测精度和效率,本文提出了一种基于坐标转换方法和UNet神经网络的机器学习推理方法。首先,提出了用于数据前处理的坐标转换方法,将计算流体力学中的物理量和网格信息转换成神经网络空间信息,使流场信息的分布更符合神经网络的输入要求。其次,建立了新型深度UNet神经网络,使模型学习到翼型流场精细复杂的局部流动特征。本文将两种方法结合,建立了翼型可压缩流场机器学习推理方法,得到快速高精度的推理模型。最后,对不同种类翼型的流场与气动力进行预测分析,并与传统机器学习方法预测的结果进行比较。结果表明,本文提出的机器学习推理方法能够较好地预测翼型的可压缩流场,并且能够更好地捕捉高雷诺数下的复杂流动行为以及预测大迎角、高马赫数条件下的流动分离和激波现象。展开更多
The paper examines the dynamic stall characteristics of a finite wing with an aspect ratio of eight in order to explore the 3D effects on flow topology,aerodynamic characteristics,and pitching damping.Firstly,CFD meth...The paper examines the dynamic stall characteristics of a finite wing with an aspect ratio of eight in order to explore the 3D effects on flow topology,aerodynamic characteristics,and pitching damping.Firstly,CFD methods are developed to calculate the aerodynamic characteristics of wings.The URANS equations are solved using a finite volume method,and the two-equation k-ωshear stress transport(SST)turbulence model is employed to account for viscosity effects.Secondly,the CFD methods are used to simulate the aerodynamic characteristics of both a static,rectangular wing and a pitching,tapered wing to verify their effectiveness and accuracy.The numerical results show good agreement with experimental data.Subsequently,the static and dynamic characteristics of the finite wing are computed and discussed.The results reveal significant 3D flow structures during both static and dynamic stalls,including wing tip vortices,arch vortices,Ω-type vortices,and ring vortices.These phenomena lead to differences in the aerodynamic characteristics of the finite wing compared with a 2D airfoil.Specifically,the finite wing has a smaller lift slope during attached-flow stages,higher stall angles,and more gradual stall behavior.Flow separation initially occurs in the middle spanwise section and gradually spreads to both ends.Regarding aerodynamic damping,the inboard sections mainly generate unstable loading.Furthermore,sections experiencing light stall have a higher tendency to produce negative damping compared with sections experiencing deep dynamic stall.展开更多
文摘为进一步提升高雷诺数、大迎角(Angle of attack,AoA)和高马赫数下的翼型可压缩流场预测精度和效率,本文提出了一种基于坐标转换方法和UNet神经网络的机器学习推理方法。首先,提出了用于数据前处理的坐标转换方法,将计算流体力学中的物理量和网格信息转换成神经网络空间信息,使流场信息的分布更符合神经网络的输入要求。其次,建立了新型深度UNet神经网络,使模型学习到翼型流场精细复杂的局部流动特征。本文将两种方法结合,建立了翼型可压缩流场机器学习推理方法,得到快速高精度的推理模型。最后,对不同种类翼型的流场与气动力进行预测分析,并与传统机器学习方法预测的结果进行比较。结果表明,本文提出的机器学习推理方法能够较好地预测翼型的可压缩流场,并且能够更好地捕捉高雷诺数下的复杂流动行为以及预测大迎角、高马赫数条件下的流动分离和激波现象。
基金supported by the National Natural Science Foundation of China(No.12072156)the National Key Laboratory Foundation of China(No.61422202103)the Priority Academic Program Development of Jiangsu Higher Education Institutions(PAPD)。
文摘The paper examines the dynamic stall characteristics of a finite wing with an aspect ratio of eight in order to explore the 3D effects on flow topology,aerodynamic characteristics,and pitching damping.Firstly,CFD methods are developed to calculate the aerodynamic characteristics of wings.The URANS equations are solved using a finite volume method,and the two-equation k-ωshear stress transport(SST)turbulence model is employed to account for viscosity effects.Secondly,the CFD methods are used to simulate the aerodynamic characteristics of both a static,rectangular wing and a pitching,tapered wing to verify their effectiveness and accuracy.The numerical results show good agreement with experimental data.Subsequently,the static and dynamic characteristics of the finite wing are computed and discussed.The results reveal significant 3D flow structures during both static and dynamic stalls,including wing tip vortices,arch vortices,Ω-type vortices,and ring vortices.These phenomena lead to differences in the aerodynamic characteristics of the finite wing compared with a 2D airfoil.Specifically,the finite wing has a smaller lift slope during attached-flow stages,higher stall angles,and more gradual stall behavior.Flow separation initially occurs in the middle spanwise section and gradually spreads to both ends.Regarding aerodynamic damping,the inboard sections mainly generate unstable loading.Furthermore,sections experiencing light stall have a higher tendency to produce negative damping compared with sections experiencing deep dynamic stall.
