Untitled

Abstract

In designing and developing airfoils, confirmation of proper design performance under various flow conditions is vital. Experimental studies using wind tunnels or numerical simulations can often utilize. In some cases, numerical studies have a weakness in computational time. This study focuses on predicting the drag coefficient of the airfoil using the CNN machine learning architecture. Starting with a numerical simulation of 500 types of NACA airfoils with a Reynolds number of 4000 using XLRF5 software to obtain image data, lift and drag coefficients. The training, test, and validation dataset uses numerical simulation results as labels. ReLU is the activation function used in this study, with Adam optimizer and MSE loss function. It achieved a relative error of 8% in predicting the drag coefficient. With the results obtained, aircraft designers can use the method to predict the drag coefficient value from various geometries.