Improving the understanding of unsteady aerodynamic flows via high-fidelity simulations, analytical modeling and deep learning techniques

• Grant number: 22/09196-4
• Support type: Scholarships in Brazil - Post-Doctorate
• Effective date (Start): September 01, 2022
• Effective date (End): August 31, 2023
• Field of knowledge: Engineering - Aerospace Engineering - Aerodynamics
• Principal researcher: William Roberto Wolf
• Grantee: Renato Fuzaro Miotto
• Home Institution: Faculdade de Engenharia Mecânica (FEM). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated research grant: 13/08293-7 - CCES - Center for Computational Engineering and Sciences, AP.CEPID

Abstract

Unsteady aerodynamic loads generated during dynamic stall play a critical role in determining both the mechanical life span and performance of unsteady lifting devices such as helicopter rotors and wind turbine blades. To control these loads, it is required both an understanding of the unsteady flow conditions as well as providing mechanisms for prescribing the ensuing flow-wing interactions. However, the temporal and spatial complexity of unsteady separated flows renders them difficult to fully characterize and understand. It comes as no surprise that, thus far, unsteady separated flows have defied obtaining general analytical solutions. Although potentially accurate, numerical simulations of unsteady separated flows require large amounts of computational time and the application of intricate numerical algorithms. Experiments are also complex and time consuming, besides requiring expensive apparatus. As such, the broad parameter space encompassed by unsteady separated flows hamper exhaustive characterization by computational or experimental means. Fortunately, the emerging of machine (deep) learning techniques could provide more ideas for leveraging models. The era of big data and the significantly improved computing power have laid a good foundation for employing machine learning techniques to unsteady fluid mechanics applications. For this reason, here, these techniques are used with the hope of finding the mapping relation between the flow structures and the underlying airfoil responses. Dynamic stall is taken here as the main application since the flow has a common structure owning certain complexity as well. Despite that, the concepts applied herein can be easily extended to other branches of unsteady fluid dynamics. That said, based on any fluid property from the unsteady flowfield, the network between the existing flow structures and some concerned flow feature will be constructed. Here, we also aim at reusing or transferring information from previously learned tasks to extract quantitative information from available flow visualizations. Despite substantial advances in experimental fluid mechanics, the use of measurements to reliably infer fluid properties, like density, velocity, pressure or stress components, is not a straightforward task. This information, though, comes natural to CFD practitioners. So, in our study, we address the question of using the information learned from numerical simulation datasets to extract fluid properties from experiments that until then would be very complicated or even impossible to obtain. (AU)