| Powered aircraft engine as the aircraft components,due to their high sages ratio,to constantly toward the direction of high precision,lightweight,this leads to the aircraft engine internal structure more compact,the interference effect between blade and more intense,the design requirements of aircraft engine has become more and more high,so,It is very important to analyze aeroengine blade aerodynamic force.At present,the blade aerodynamic data to obtain the need to pass a special test and simulation,but the traditional experiments and simulation modeling cycle is long,the consumption of resources,a reduced order model of aerodynamic force limited by way of data acquisition,so difficult to blade aerodynamic force for rapid analysis of the complex working conditions,obviously,it is not enough for aircraft engine design research.According to current situation of the research and the demand of the engine design,this paper puts forward the research based on image recognition CFD data and experimental data collecting,a reduced order model building method of the aerodynamic force diagram by collecting literature data,use the method of image recognition fetching CFD data and experimental data in literature,and use these data to establish a convolution neural network,a reduced order model aerodynamic The rapid analysis method of blade surface aerodynamic force under different working conditions is studied.The main contents of this paper are as follows:1)Roberts operator,Prewitt operator,LOG operator,Sobel operator and Canny operator were used to detect the edge of blade aerodynamic literature data images respectively.After that,Hough linear detection algorithm was used to detect and mark the coordinate axes of the blade aerodynamic literature data images.Finally,the color of the images was extracted by the METHOD of HSV space conversion.The analysis results show that the document data extraction model based on image recognition method can extract different color curve data more accurately,but it has certain limitations for the data extraction of the same color curve.2)Based on convolutional neural network,a reduced order model of blade aerodynamic force of different flow Angle,speed,to predict the total pressure under the condition of the blade aerodynamic force,and the convolution of the neural network output and a reduced order model blade aerodynamic model of literature aerodynamic data extraction based on image recognition method to obtain the CFD simulation results were compared with experimental results.The analysis results show that the blade aerodynamic force is used as training signal to train the convolutional neural network model under multiple working conditions,and the reduced order convolutional neural network aerodynamic force model obtained can predict blade aerodynamic force quickly,and the error of prediction results can be controlled within 20%.3)In this paper,feature extraction method was used to reduce the dimension of sample data,and the convolution neural network model was combined to establish the blade aerodynamic order reduction model under the data dimension reduction method.In this paper,PCA method and parameter pooling method were used to extract features from aerodynamic data samples of aeroengine compressor blades,and the prediction results of the convolution neural network aerodynamic order reduction model based on the two sample feature extraction methods were compared.The results show that: The convolution neural network model of compressor blade aerodynamic force reduction based on PCA sample feature extraction method has a large jitter and relatively low accuracy.The convolution neural network model based on feature extraction of parameter pooling sample has strong prediction ability,and the maximum error of prediction results is within 15%.Compared with the model without sample feature extraction method,the calculation speed of compressor aerodynamic reduction model is 5% higher,and the model has stronger generalization ability. |
PDF Full Text Request