CFDGAN: A generative adversarial network for flow approximation

In aerodynamics and automobiles, Computational Fluid Dynamics (CFD) simulations are generally computationally exorbitant, high memory demanding and time consuming iterative process. Due to this, many design processes analyze their design after finishing the design stage, making the Innovation and new product development slower which in today’s concurrent development causes waste of time and energy. To solve this problem to some extent, we propose a Pix2Pix Generative Adversarial Networks (GAN) or as we name as CFDGAN, which takes 2D drawings as Images and gives the converged CFD results as output based on the Reynolds number and Velocity and Pressure. The models have been trained with the dataset of airfoils’ velocity and pressure flow field images with varying Reynolds numbers solved using OpenFoam. The trained Model has been deployed using flask as a web-app for running in cloud.