ABSTRACT
Design processes nowadays rely more and more on automated optimisation methods to shorten the development cycle. Within those optimisation methods, gradient-free ones converge slower but rather to a global optimum, while gradient-based methods converge faster to a local optimum. Quite recently gradient-free methods have been assisted by metamodels to improve their convergence and gradient-based methods are making use of adjoints to speed up the gradient evaluation. In this article, we compare an adjoint-assisted gradient-based and a metamodel-assisted gradient-free method with respect to convergence, local/global optima and especially the computational time. On a constrained multipoint aerodynamic optimisation of a turbine inlet vane, gradient-based and gradient-free methods reached 22% and 24%, respectively, of total pressure loss reduction. The metamodel-assisted method reached a 2% higher objective value at double the cost of flow evaluations, an additional cost related mainly to the evaluation of an initial database.
Disclosure statement
No potential conflict of interest was reported by the authors.
Correction Statement
This article has been corrected with minor changes. These changes do not impact the academic content of the article.
Notes
1 Steven G. Johnson, The NLopt nonlinear-optimisation package, http://ab-initio.mit.edu/nlopt.
Additional information
Notes on contributors
Mohamed Hassanine Aissa
Dr Mohamed H. Aissa is a research engineer at the von Karman Institute for fluid dynamics. He obtained his PhD in Applied mathematics in October 2017 from the technical university of Delft for working on the acceleration of CFD solvers using high-performance computing. He focuses currently on metamodel-assisted optimisation and is involved in industrial projects mostly related to the automotive and aerospace industry.
Roberto Maffulli
Dr Roberto Maffulli obtained his DPhil at Oxford University studying the coupling between aerodynamics and heat transfer in gas turbines using numerical modelling. He has worked at von Karman Institute for Fluid Dynamics on machine learning tools for automated optimisation systems and is now a post-doctoral researcher at Oxford University working on multi-scale modelling of unsteady conjugate heat transfer. His research interests include optimisation algorithms, machine learning, computational modelling of turbomachinery flows, conjugate heat transfer.
Lasse Mueller
Dr Lasse Muller is a research engineer at the von Karman Institute for fluid dynamics. He obtained his PhD in mechanical engineering in December 2018 from the Universit7eacute; libre de Bruxelles for his work on adjoint-Based Optimisation applied to axial and radial turbines. He focuses currently on adjoint-based optimisation and is involved in industrial projects mostly related to the automotive and aerospace industry.
Tom Verstraete
Prof Tom Verstraete obtained his PhD in 2008 from the University of Ghent in collaboration with the von Karman Institute (VKI). Since 2008 he joined the faculty of the VKI. In the period 2015–2017, he spent a 2-year sabbatical leave as a visiting professor at Queen Mary University of London, where he performed further research on multidisciplinary adjoint optimisation methods. Since 2017 he is as well a visiting professor at the University of Ghent. His main research interest is multidisciplinary shape optimisation applied to turbomachinery.