Prediction Of Airfoil Ice Accretion And Aerodynamic Characteristics Based On Deep Learning

Aircraft ice accretion is easy to occur when the aircraft flies in icing weather conditions.The ice accretion will effect the aerodynamic characteristics of the aircraft,which will lead to flight accidents in serious.Therefore,it is of great significance to carry out the research on the icing principle of the airfoil and predict the icing and aerodynamic characteristics of the airfoil.At present,the existing research methods mainly include flight test,wind tunnel test,numerical simulation calculation and so on,but these methods have the problems of long prediction period and large consumption of resources,which can not achieve fast and efficient prediction.This paper uses the method of deep learning to carry out relevant research,for the follow-up research to provide a fast and efficient airfoil ice accretion and aerodynamic characteristics prediction method,and taking NACA0012 airfoil as an example verifies the validity and reliability.The main work of this paper includes:(1)Analyzes the principle of airfoil ice accretion and the main factors affecting airfoil ice accretion,investigates the research on airfoil ice accretion prediction at home and abroad,summarizes the existing research methods and their advantages and disadvantages,and focuses on the numerical simulation method of computational fluid dynamics,which provides theoretical and data support for the follow-up prediction research of this paper.(2)The research of airfoil ice accretion prediction based on deep learning is carried out.According to the prediction object is divided into ice-shaped prediction and ice-shaped geometric characteristic parameters,in which ice curve parameters and ice-shaped geometric characteristic parameters are obtained through the processing of wing coordinate conversion and Fourier series expand.Then the prediction network is built on the basis of deep belief network and stack self-auto encoder,and some data experiments are selected to select the optimal prediction network model.At the same time,the BP network is selected for network comparison.The prediction results use the deep learning network model to predict the ice-shaped similarity factor of more than90%,and the average error of the ice-shaped geometric characteristic parameter prediction is less than 4%,which is better than BP network.In addition,migration learning and wind tunnel test data are used to further validate network reliability and generalization.(3)The aerodynamic characteristics prediction of airfoil ice accretion based on deep learning is studied.According to the different prediction methods,the prediction of aerodynamic parameters based on ice-shaped graphs and the prediction of aerodynamic parameters based on ice-shaped geometric characteristics parameters are divided into predictions,in which the prediction based on ice-shaped graphs obtains ice-shaped shapes through data conversion,uses convolution neural networks to extract ice-shaped image features,and establishes a nonlinear relationship between data.The prediction results show that the average relative error of aerodynamic coefficients in the fixed angle of attack is within 5%.Based on the prediction of ice-shaped geometric feature parameters,the four parameters with the greatest influence on aerodynamic characteristics are selected as inputs,and the prediction model is established by using the deep full-connection network,and the prediction results show that the average relative error of the prediction is about 10%.