An Intelligent Method for Predicting the Pressure Coefficient Curve of Airfoil-Based Conditional Generative Adversarial Networks

Airfoil aerodynamic performance prediction using machine learning and surrogate modeling

“…While some intelligent methods aimed to reduce computational time, they frequently lacked enhancements and introduced graphical representations that could distort aerodynamic curves and data. Moreover, efforts to learn the relationship between airfoil shape and aerodynamic coefficients sometimes disregarded the influence of initial inflow conditions [ 10 ]. Portal-Porras et al [ 11 ] concentrated on enhancing wind turbine performance by implementing flow control devices on airfoils.…”

Towards Universal Parameterization: Using Variational Autoencoders to Parameterize Airfoils

“…Generative algorithms, notably Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs), have gained substantial attention in recent research [2][3][4] for aerodynamic surrogate modeling and flow prediction [5,6] and generation of new data points to augment sparse datasets [7][8][9]. These methods play a pivotal role in expanding the capabilities of computational design and data analysis.…”

Machine Learning in Aerodynamic Shape Optimization

“…Guo et al [386] used a convolutional autoencoder to provide a fast prediction of non-uniform steady laminar flow for twodimensional or three-dimensional shapes. Bhatnagar et al [255] applied a convolutional autoencoder Wang et al [395] to predict the velocity and pressure field in unseen flow conditions and airfoil shapes, and the model achieved a relative error of less than 10% over the entire flow field. Eivazi et al [390] showed that a nonlinear autoencoder network is more effective than POD in extracting low-dimensional flow features for the accurate prediction of unsteady velocity fields around a cylinder and an oscillating airfoil.…”

“…The conditional generative model was trained by 500 airfoils sampled around the RAE2822 airfoil and then was shown to roughly predict the flow fields of unseen airfoils. Wang et al [395] used GAN to predict the pressure coefficient distribution of different airfoil shapes at various M , Re, and α. Different from the two-step modeling approach (first perform dimensionality reduction and then model the reduced-order flow variables) in conventional reduced-order models, the nonlinear manifold for dimensionality reduction can be obtained together with the training of the prediction model in ANN.…”