Ice shape image prediction based on transfer learning and residual network
Among the ice shape prediction methods of neural networks,the ice shape data of wind tunnel experiment has high-precision ice shape features,but the experiment cost is expensive and the data obtained is limited so it has not been fully utilized,and most of the research is focused on numerical calculation data.To this end,an image prediction method combining transfer learning with residual network is proposed.A deep neural network prediction model is established by taking airfoil cross section image and icing condition parameters as input and 2D ice shape image as output,which realizes the high-precision 2D ice shape prediction.This method obtains the pre-training mod-el through a large amount of numerical calculation data,and then fine-tunes it with a small amount of wind tunnel experiment data to achieve ice shape prediction.The results show that the proposed method can predict high-precision 2D airfoil icing images,and most of the relative error between prediction and wind tunnel experiment data in ice shape feature parameters is kept within 15％.
ice shape predictiontransfer learningresidual networkwind tunnel test
NETL
NSTL
万方数据
