PHYSICS-BASED AND DATA-DRIVEN DEEP LEARNING MODELS FOR FLOW FIELD PREDICTION AND AIRFOIL DESIGN

Abstract

Computational Fluid Dynamics (CFD) techniques have emerged as powerful tools for exploring and solving real engineering problems in mechanical and aerospace industries. However, it is still relatively time-consuming for tasks such as optimization and fluid-structure interaction, where there is a requirement of large iterations. Hence, the current work develops data-driven deep learning models, which are fast yet accurate. First, the current work develops a deep learning-based framework for obtaining flow field over variable geometries. Further, a surrogate model-based approach is developed to perform fast optimization of airfoils. In addition, an approach is presented to perform inverse design of airfoil using deep learning. Besides, the current work explores Physics Informed Neural Networks (PINN) for obtaining flow solutions near the wall accurately with measurements (or sampling points) away from the wall. From the obtained results, it is observed that the developed methods are efficient and accurate.