GENERATIVE AI APPROACHES TO AIRFOIL FLOW FIELD PREDICTION
TL;DR
It is shown that despite relatively higher computational costs when compared with alternative deep learning methods, generative models offer a promising approach for obtaining higher-fidelity flow fields and more accurate uncertainty modelling.
Abstract
Airfoil flow field prediction is central to turbomachinery design, yet Computational Fluid Dynamics (CFD) imposes a computational bottleneck on iterative design and optimisation. Recent studies have explored deep-learning-based surrogate models to reduce the computational cost associated with aerodynamic analyses. Among these are generative models, a class of models that include diffusion models, Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). With a growing number of generative models in airfoil flow field prediction, there is limited systematic assessment of their performance and suitability. The purpose of this paper is to review the state of the art of generative models for airfoil flow field prediction, providing an overview of recent contributions, methodological advances, and outstanding challenges. We show that despite relatively higher computational costs when compared with alternative deep learning methods, generative models offer a promising approach for obtaining higher-fidelity flow fields and more accurate uncertainty modelling. Our analysis reveals that research in this area remains in its early stages, but applications point towards advancement beyond simple two-dimensional airfoil flow fields towards three-dimensional and multi-row configurations reflective of real-world turbomachinery.