论文部分内容阅读
研究目的是根据机翼各项气动参数,快速准确地预测出符合气动条件的机翼外形.用PARSEC方法对剖面翼型进行参数化处理,得到表征其物理特性的外形参数;用守恒型全速势方程进行流场计算;建立包含214组机翼几何及气动特性的专家数据库.人工神经网络方法对数据库进行分类,训练和测试.先用SOM(Self-Organizing Map)神经网络按气动参数对数据进行分类,再分别用BP(Back Propagation)神经网络,RBF(Radial Basis Function)神经网络和GRNN(General regression Neural Net)进行训练和测试.机翼由6个翼剖面组成,每个翼剖面包含11个PARSEC特征参量,扭转角以及相对厚度,总共78个独立的外形参数.预测值和预期值的相关性分析以及误差分析表明,GRNN的预测结果相比于BP和RBF更为准确;在预测模型的升阻比的平均相对误差的绝对值时,BP的相对误差为2.37%,RBF是0.97%,GRNN是0.40%.
The purpose of this study is to predict the aerofoil profiles that meet the aerodynamic conditions quickly and accurately based on the aerodynamic parameters of the aerofoils.The PARSEC method is used to parameterize the profile airfoils to obtain the shape parameters that characterize their physical properties. Equation is used to calculate the flow field.An expert database including 214 groups of wing geometric and aerodynamic characteristics is established.The artificial neural network method is used to classify, train and test the database.At first, the data is processed according to the aerodynamic parameters using the SOM (Self-Organizing Map) neural network And then trained and tested respectively with Back Propagation (BP) neural network, Radial Basis Function (RBF) neural network and GRNN (General Regression Neural Net) .The wing consists of 6 wing sections, each wing section contains 11 PARSEC characteristic parameters, torsion angle and relative thickness, a total of 78 independent shape parameters.Prediction and expected value of the correlation analysis and error analysis showed that, GRNN prediction results compared to BP and RBF more accurate; in the prediction model When the absolute value of the average relative lift-drag ratio is compared, the relative error of BP is 2.37%, the RBF is 0.97% and the GRNN is 0.40%.