Reduced-order model based on Volterra series with kernels extracted from neural networks for nonlinear unsteady aerodynamics

• Grant number: 18/23199-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: February 01, 2019
• End date: December 31, 2019
• Field of knowledge: Engineering - Aerospace Engineering - Aerodynamics
• Principal Investigator: Flávio Donizeti Marques
• Grantee: Guilherme Fernandes de Figueiredo
• Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

This research project proposes the generation of a reduced-order model (ROM) based on the multi-input multi-input Volterra series for the resolution of a non-linear unsteady aeroelastic problem with two degrees of freedom. The Volterra series kernels will be drawn from a time delay neural network (TDNN), trained from a database provided by a computational fluid dynamics code (CFD) and will be used to predict the aerodynamic loads acting on the aeroelastic model of the typical section airfoil. The project will focus on the identification and application of the so-called cross kernels, which allow the internal coupling of the system inputs: in the case of this Scientific Initiation, the degrees of freedom of pitch and plunge. The implemented neural network will allow the use of high order kernels, unlike conventional approaches. The reduced order model will be constructed from the substitution of the CFD codes to obtain the aerodynamic loads by the predictions coming from the Volterra series-based neural network, in the case of non-linear transonic regime.