No Cover Image

E-Thesis 239 views 213 downloads

Non-intrusive reduced order modelling for aerodynamic applications / KENSLEY BALLA

Swansea University Author: KENSLEY BALLA

DOI (Published version): 10.23889/SUthesis.58705

Abstract

During the design and optimisation of aerodynamic components, the simulations to be performed involve a large number of parameters related to the geometry and flow conditions. In this scenario, the simulation of all possible configurations is not af-fordable. To overcome this problem, the present wo...

Full description

Published: Swansea 2021
Institution: Swansea University
Degree level: Doctoral
Degree name: Ph.D
Supervisor: Hassan, Oubay ; Sevilla, Ruben ; Morgan, Kenneth
URI: https://cronfa.swan.ac.uk/Record/cronfa58705
Tags: Add Tag
No Tags, Be the first to tag this record!
Abstract: During the design and optimisation of aerodynamic components, the simulations to be performed involve a large number of parameters related to the geometry and flow conditions. In this scenario, the simulation of all possible configurations is not af-fordable. To overcome this problem, the present work proposes a novel multi-output neural network (NN) for the prediction of aerodynamic coefficients of aerofoils and wings using compressible flow data. Contrary to existing NNs that are designed to predict aerodynamic quantities of interest, the proposed network considers as output the pressure or stresses at a number of selected points on the aerodynamic surface. The proposed approach is compared against the more traditional networks where the aero-dynamic coefficients are directly the outputs of the network. Furthermore, a detailed comparison of the proposed NN against the popular proper orthogonal decomposi-tion (POD) method is presented. The numerical results, involving high dimensional problems with flow and geometric parameters, show the benefits of the proposed ap-proach.The proposed NN is used to accelerate the evaluation of the objective function in an inverse aerodynamic shape design problem. The optimisation algorithm uses the gradient-free modified cuckoo search method. Applications in two and three dimen-sions are shown, demonstrating the potential of the proposed framework in the con-text of both optimisation and inverse design problems. The performance of the pro-posed optimisation framework is also compared against existing frameworks where the more traditional NNs are employed.
Keywords: neural network, proper orthogonal decomposition, reduced order model,geometric parameters,NURBS, shape optimisation, inverse design
College: Faculty of Science and Engineering