Deep Learning Models for the Evaluation of the Aerodynamic and Thermal Performance of Three-Dimensional Symmetric Wavy Wings

Abstract

The present study initially evaluates the feasibility of deep learning models to predict the flow and thermal fields of a wing with a symmetric wavy disturbance as the passive flow control. The present study developed the encoder–decoder (ED) and convolutional neural network (CNN) models to predict the characteristics of flow and heat transfer on the surface of three-dimensional wavy wings in a wide range of parameters, such as the aspect ratio, wave amplitude, wave number, and the angle of attack. Computational fluid dynamics (CFD) is used to generate the dataset of the deep learning models. Various tests are carried out to examine the predictive performance of the architectures for two deep learning models. The CNN and ED models demonstrated a quantitatively predictive performance for aerodynamic coefficients and Nusselt numbers, as well as a qualitative prediction for pressure contours, limiting streamlines, and Nusselt contours. The predicted results well reconstructed the spiral vortical formation and the separation delay by the limiting streamlines. It is expected that the present established deep learning methods are useful to perform the parametric study to find the conditions to provide efficient aerodynamic and thermal performances.