Variational Auto Encoder-Based DNN for Lift to Drag Ratio

The convolutional neural network (CNN) method is capable of image processing and

is widely used today for aerodynamic meta-modelling tasks. CNN can predict the

aerodynamic property of an air foil based on a large enough dataset. Variational auto

encoder (VAE) has a capability of image processing too, as VAE belong to the family

of Generative Adversal Networks (GANs) they are also capable of extracting the best

feature on the basis of two additional layers (Mean and Variance). The primary goal of

this thesis is to predict the lift-to-drag ratio of air foils with various angles of attack using

VAE (a). This thesis also employs the Convolutional Neural Network (CNN) method to

calculate the lift-to-drag ratio of all air foils. The efficiency and accuracy of these two

methods are compared and discussed in this thesis, and it is demonstrated that the VAE

method can maintain a relatively competitive level of accuracy while being far more time

efficient than the analytical method and CNN method.