Dissertation/Thesis Abstract

Error Analysis on Multi-Fidelity Design Spaces through Surrogate Modeling with an Emphasis on Scramjet Combustion
by Basore, Kevin Daniel, M.S., University of Colorado at Boulder, 2012, 157; 1511948
Abstract (Summary)

Hypersonic vehicle design is a challenging problem. The main obstacle that makes this work challenging is the non-linear interdependency between each of the vehicle's systems. Thus, when generating a vehicle design model for a hypersonic system, a large number of parameters must be considered. The addition of a single parameter into the design space leads to an exponential increase in the number of necessary single point solutions.

This thesis will examine hypersonic vehicle design in a unique way that decreases the computational power required to achieve a more accurate solution. This was done through the examination of error vectors between multi-fidelity design spaces using Kriging surrogate models and superimposing the computed difference onto a low fidelity solver. By adding this empirical correction onto the low fidelity data, the accuracy of the computed metrics was improved by an order of magnitude in some cases.

In proving this theory a simplified airfoil test case and a generic cavity scramjet, that included combustion chemistry, were investigated. Examination of both of these geometries required a sizable amount of computing resources. To meet this requirement a custom cloud controller that could dynamically load processor cores depending on individual computer workloads was developed. This computational cloud, as well as an automated case generation algorithm, were developed in order to process the hundreds of computational runs necessary.

Once processed, these design spaces were visually and numerically inspected to verify convergence within each of the solutions. Having a verified discrete sampling in each of the design space fidelity levels, Kriging surrogate models were constructed to produce a continuous spectrum on the error between them. This continuous approximation was then used as the empirical correction to the low fidelity solver in order to achieve the order of magnitude increase in accuracy.

Indexing (document details)
Advisor: Starkey, Ryan
Commitee: Argrow, Brian, Jansen, Kenneth
School: University of Colorado at Boulder
Department: Aerospace Engineering
School Location: United States -- Colorado
Source: MAI 50/06M, Masters Abstracts International
Subjects: Aerospace engineering
Keywords: Hypersonics, Surrogate modeling, Vehicle design
Publication Number: 1511948
ISBN: 9781267386939
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy