OATAO - Open Archive Toulouse Archive Ouverte Open Access Week

Constrained bayesian optimization over mixed categorical variables, with application to aircraft design

Saves, Paul and Bartoli, Nathalie and Diouane, Youssef and Lefebvre, Thierry and Morlier, Joseph and David, Christophe and Van, Eric Nguyen and Defoort, Sébastien Constrained bayesian optimization over mixed categorical variables, with application to aircraft design. (2021) In: An ECCOMAS Thematic Conference on Multidisciplinary Design Optimization of Aerospace Systems (AeroBest 2021), 21 July 2021 - 23 July 2021 (Lisboa, Portugal).

[img] (Document in English)

PDF (Author's version) - Depositor and staff only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
3MB

Abstract

Multidisciplinary Design Optimization (MDO) methods aim at adapting numerical optimization techniques to the design of engineering systems involving multiple disciplines or components. Among MDO architectures, various ones are considering the resolution of the Multidisciplinary Design Analysis (MDA). In our study, the system of interest being an aircraft, the resolution of the MDA will be provided by the Future Aircraft Sizing Tool with Overall Aircraft Design (FAST-OAD), a point mass approach that estimates the required fuel and energy consumption for a given set of top-level aircraft requirements. In this context, a large number of mixed continuous, integer and categorical variables that arise from aircraft design has to be tackled by the optimization process. Recently, there has been a growing interest in mixed variables constrained Bayesian optimization based on Gaussian process surrogate models. In this setting, most existing approaches severely increase the dimension of the covariance matrix related to the surrogate. In fact, the construction of the Gaussian process model may not be scalable to practical applications involving a large number of mixed variables. In this paper, we address this issue by constructing a covariance kernel for the surrogate model that depends on only a few hyperparameters. The new kernel is constructed based on the information obtained from the partial least squares method. The obtained numerical results lead to interesting results for the optimization of a baseline aircraft and to reduce the fuel consumption of "DRAGON", a new hybrid electric propulsion aircraft, with a high number of mixed variables and for a small budget of time-consuming evaluations.

Item Type:Conference or Workshop Item (Paper)
Audience (conference):International conference proceedings
Uncontrolled Keywords:
Institution:French research institutions > Centre National de la Recherche Scientifique - CNRS (FRANCE)
Université de Toulouse > Ecole nationale supérieure des Mines d'Albi-Carmaux - IMT Mines Albi (FRANCE)
Université de Toulouse > Institut National des Sciences Appliquées de Toulouse - INSA (FRANCE)
Université de Toulouse > Institut Supérieur de l'Aéronautique et de l'Espace - ISAE-SUPAERO (FRANCE)
French research institutions > Office National d'Etudes et Recherches Aérospatiales - ONERA (FRANCE)
Université de Toulouse > Université Toulouse III - Paul Sabatier - UT3 (FRANCE)
Laboratory name:
Statistics:download
Deposited On:06 Oct 2021 13:57

Repository Staff Only: item control page