Visual Explainable Convolutional Neural Network for Aerodynamic Coefficient Prediction

Recently, aerodynamic performance analysis has been widely studied due to its importance in aircraft design. Most works adopted computational fluid dynamics (CFD) simulation compute the forces, which is time consuming. To reduce time, several proposed use deep learning model as surrogate of CFD simulation. However, explainability models poor and criticized, limits further development analysis. In this paper, a novel neural network predict forces airfoils. improve explainability, circular padding replace traditional zero convolutional layers. Moreover, saliency map predicted force on input airfoil shown more intuitive way. manner, influence different parts final can be easily analyzed. Extensive experiments data sets show that our work efficient effective. importantly, these results explain potential relationship between force.