Paper Title
Inverse Airfoil Design Using Deep Learning
Abstract
The task of determining the appropriate airfoil section for designing wings of airplanes or turbine blades has
historically been a difficult and time-consuming process, requiring numerous design optimizations and simulations.
However, the use of deep learning models such as Autoencoders and Multi-layer Perceptron (MLPs) presents a promising
alternative by significantly reducing computation time and power required. In this study, these models are used to
parameterize airfoils. By training them with data generated using JAVAFOIL, the models can generate new airfoils that
correspond to previously unseen design parameters. The use of Convolutional Autoencoders (CAE) further reduces
computation time and power during offline stages, allowing for the mapping of airfoil sections at various angles of attack to
their corresponding design parameters, such as Reynolds number and coefficients of lift, drag, moment, and pressure.
Keywords - Airfoil, Deep Learning, Multi-Layer Perceptron, Autoencoders