Please use this identifier to cite or link to this item:
Type: Conference paper
Title: Derivative free global optimisation of CFD simulations
Author: Morgans, R.
Doolan, C.
Stephens, D.
Citation: Proceedings of the 16th Australasian Fluid Mechanics Conference, 2007 / Morgan, R., Lemckert, C. (ed./s), pp.1-4
Part of: Proceedings of the 16th Australasian Fluid Mechanics Conference, pp. 1432-1435
Publisher: University of Queensland
Publisher Place: CDROM
Issue Date: 2007
ISBN: 9781864998948
Conference Name: Australasian Fluid Mechanics Conference (03 Dec 2007 : Gold Coast, Australia)
Statement of
R.C. Morgans, C.J. Doolan and D. Stephens
Abstract: This work reports on the use of numerical optimisation techniques to optimise objective functions calculated by Computational Fluid Dynamics (CFD) simulations. Two example applications are described, the first being the shape optimisation of a low speed wind tunnel contraction. A potential flow and viscous flow solver have been coupled to produce a robust computational tool, with the contraction shape defined by a two parameter B´ezier curve. The second application is a simplified test case with a known minimum calculated using a commercial CFD code. For the optimisation of complex CFD simulations, it is sometimes advantageous to use an efficient derivative free global optimisation algorithm because of potentially long simulation times, the objective function may contain multiple local minima and it is often difficult to evaluate analytical or numerical gradients. The Efficient Global Optimisation (EGO) algorithm sequentially samples results from an expensive calculation, does not require derivative information, uses an inexpensive surrogate to search for a global optimum, and is used in this current work. For both applications, the EGO algorithm is able to efficiently and robustly find a global optimum that satisfies any constraints.
RMID: 0020075882
Published version:
Appears in Collections:Mechanical Engineering conference papers
Environment Institute publications

Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.