On the Use of Convolutional Neural Network for Aerodynamic Data Prediction

We employ Convolutional Neural Network to predict the lift coefficient for a new airfoil shape. Our proposed con-volutional neural network efficiently uses an artificially created image to map it to its lift coefficient. The RMSE obtained by our proposed model is around 0.13 which is quite comparable to those of Multi-layer Perceptron (MLP) and k-means-based MLP. The RMSE goes down to 0.068 if we marginally increase the size of the dataset. Furthermore, the RMSE drops to 0.1098 if we alter the model architecture by adding convolution/pooling and dense layers.