Model order reduction for parameterized multidisciplinary analysis using disciplinary surrogates: application to non-linear solvers - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2024

Model order reduction for parameterized multidisciplinary analysis using disciplinary surrogates: application to non-linear solvers

Résumé

Multidisciplinary design analysis (MDA) is a useful tool when dealing with systems involvingseveral disciplines. Nonetheless, its use in applications requiring several MDA analysis can resultin intractable computational costs if high-fidelity solvers are used directly in the system. Oneway to reduce the computational burden is by replacing the disciplinary solvers by surrogatemodels. These surrogates can however be difficult to build when the coupling variables of thesystem are high-dimensional vectors whose ranges are unknown. Recently, a methodology calledDisciplinary Proper Orthogonal Decomposition and Interpolation (DPOD+I) has been proposed,which provides a solution to this problem by coupling a linear dimension reduction strategywith Gaussian Process interpolation. In this work we show that the DPOD+I strategy canencounter some challenges when one or more of the obtained disciplinary POD bases contain asignificant number of vectors. To address these challenges, we propose an alternative strategyfor obtaining a disciplinary reduced order model, based on the interpolation of pointwise, localPOD bases. Application to an aeroelastic engineering test case which couples two high-fidelitydisciplinary solvers is used to confirm the interest of the proposed approach. The obtainedresults show that the method efficiently reduces the dimension of the coupling variable spaceand is capable of finding the MDA solution at a reduced computational cost, with an averagerelative error of less than 1%.
Fichier sous embargo
Fichier sous embargo
0 5 25
Année Mois Jours
Avant la publication
vendredi 27 décembre 2024
Fichier sous embargo
vendredi 27 décembre 2024
Connectez-vous pour demander l'accès au fichier

Dates et versions

hal-04627277 , version 1 (27-06-2024)

Identifiants

Citer

Inês Cardoso, Sylvain Dubreuil, Nathalie Bartoli, Christian Gogu, Michel Salaün. Model order reduction for parameterized multidisciplinary analysis using disciplinary surrogates: application to non-linear solvers. AIAA SCITECH 2024 Forum, AIAA, Jan 2024, Orlando, United States. ⟨10.2514/6.2024-1407⟩. ⟨hal-04627277⟩
0 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More