Learning theory & Gaussian Process Regression for surrogate modeling, and a novel framework for Design Optimization under uncertainty. Application to an early-stage aircraft wing design

Dorado-Ladera, Jose-Luis 2021. Learning theory & Gaussian Process Regression for surrogate modeling, and a novel framework for Design Optimization under uncertainty. Application to an early-stage aircraft wing design. MPhil Thesis, Cardiff University. Item availability restricted.

Preview	PDF (M.Phil Thesis) - Accepted Post-Print Version Download (11MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (2MB)

Abstract

The aim of this thesis is to study the problem of regression for surrogate modeling, to develop a novel framework for design optimization that makes no assumptions on the model, and to give guidelines to apply the aforementioned theory to an early-stage aircraft wing design. The first part provides a summary of the mathematical foundations of learning theory for regression. The theory of Reproducing Kernel Hilbert Spaces is broadly covered. In addition, Gaussian Process regression is explained in detail. The second part introduces a novel framework for design optimization. Sampling from a probability distribution is at the core of this framework. Therefore,algorithms for simulating different distributions are described in detail. Furthermore, rejection sampling, the theory of Markov chains,and Metropolis-Hasting are explained as methods for simulating arbitrary distributions. The framework aims to optimize the parameters of a Probability Density Function in the input space of a surrogate model in order to satisfy prescribed performance in the output. Stochastic Optimization is suggested as the optimization process and a description of Simulated Annealing is included. This MPhil is part of a project in collaboration with Airbus. They provided a dataset with the goal of optimizing the jig twist of an aircraft wing. The last part analyzes this data and provides future researchers in the project with guidelines to train a Gaussian Process and apply the novel framework mentioned above to tackle the optimization problem

Item Type:	Thesis (MPhil)
Date Type:	Completion
Status:	Unpublished
Schools:	Engineering
Uncontrolled Keywords:	Surrogate modelling; Gaussian process; Stochastic optimization; Design optimization; Learning theory; Inverse design.
Date of First Compliant Deposit:	9 June 2021
Last Modified:	07 Jan 2022 02:12
URI:	https://orca.cardiff.ac.uk/id/eprint/141807

Actions (repository staff only)

Edit Item