弹片是解决翼型流动分离的重要技术手段，合理的弹片参数对翼型表面压力分布尤为重要。基于数据驱动的深度学习方法与计算流体力学（Computational Fluid Dynamics，CFD）相结合，可快速有效地完成对复杂流场特征的识别与提取。本文提出一种基于卷积神经网络（Convolutional Neural Network，CNN）的翼型表面压力分布预测方法，通过提取流场的尾流速度、压力等流动特征构建翼型表面压力分布的预测模型。首先，通过数值模拟计算了8种不同抬起角度的NACA 0012弹片翼型的流场；其次，采用提取的流场数据建立CNN预测模型；最后，将预测值和CFD计算值进行对比。结果表明：基于CNN的预测模型对翼型表面压力系数分布有较高的预测精度，其中尾流速度模型在弹片抬起角度为15°时的预测均方根误差仅为0.1，说明尾流速度中包含丰富的流场信息。

Flap is an important technical tool to solve the flow separation of airfoil, and reasonable flap parameters are especially important for the pressure distribution on the airfoil surface. The combination of datadriven deep learning method and computational fluid dynamics (CFD) can quickly and effectively complete the feature identification and extraction of complex flow fields. In this paper, we proposed a convolutional neural network (CNN)based method for predicting the pressure distribution on the airfoil surface, by extracting the flow features such as wake velocity and pressure of the flow field to build a prediction model for the pressure distribution on the airfoil surface. Firstly, the flow fields of flap of NACA 0012 airfoil with eight different lift angles were calculated by numerical simulation; secondly, the CNN prediction model was built using the extracted flow field data; finally, the predicted values were compared with the CFD calculated values. Results show that the convolutional neural networkbased model has a high prediction accuracy to pressure coefficient distribution on the airfoil surface, and the predicted root mean square error (RMSE) of wake velocity model is only 0.1 when flat lift angle is 15°, indicating the wake velocity contains abundant flow field information.

TK83

国家自然科学基金资助项目（51976131，52006148，52106262）