Deep convolutional architectures for optimizing multi-element airfoil systems

Abstract

The investigation of aerodynamic properties in multi-element airfoils is historically challenging and impractical for real-time applications. Predicting specific angles to achieve predetermined lift and drag forces has been elusive. Advances in machine learning (ML) and deep learning (DL) offer new possibilities, but their application to multi-wing system (multi-element airfoils) is limited by scarce datasets. This study explores development, design, and comparative analysis of ML and novel DL algorithms to predict the aerodynamic coefficients generated by multi-element airfoils given airfoil shape, angle of orientation of each airfoil, and their velocity, a previously unexplored area. Using datasets generated via ANSYS Fluent, the proposed models predict aerodynamic coefficients based on variables such as velocity, airfoil angle, and images of the airfoil system with high accuracy, achieving a mean squared error of 0.0049. The proposed approach also significantly reduces the computational time for predicting the aerodynamic coefficients from 10 to 15 min in a traditional computational fluid dynamics simulation to 1–6 ms using our approach. This real-time prediction capability allows for assessing various angle combinations to find optimal orientations for specific drag and lift forces.