
Cet article présente une comparaison de techniques efficaces de quantification d'incertitude et d'optimisation robuste. Deux cas test basés sur un profil mince représentatif d'une pale de soufflante non carénée sont étudiés dans des conditions transsoniques. L'accent est mis sur l'estimation précise des statistiques des coefficients aérodynamiques. Grâce à deux modèles de substitution basés sur la CFD, les deux cas test ont été étudiés de manière approfondie sans engager le coût de calcul important normalement associé aux évaluations CFD directes. Pour traiter la haute dimensionnalité de l'espace d'incertitude, nous étudions deux approches différentes dans le cadre des polynômes du chaos généralisés et de leur construction par approximation des moindres carrés. La première approche utilise des techniques d'acquisition comprimée (compressed sensing), en particulier les algorithmes LARS et BPDN, et la seconde approche, appelée approximation des moindres carrés « améliorée par gradient », tire parti des capacités adjointes des solveurs CFD modernes pour un calcul de gradient efficace. Une combinaison des deux, associant les informations de gradient à l'acquisition comprimée, est également évaluée.
This paper presents a comparison among efficient techniques for uncertainty quantification and robust optimization. Two test cases based on a thin airfoil representative of an open-rotor airfoil are studied at transonic conditions. The focus is on the precise estimation of the statistics of the aerodynamic coefficients. Thanks to two CFD-based surrogate models, the two test cases have been studied extensively without engaging the large computational cost normally associated with direct CFD evaluations. To address the high-dimensionality of the uncertainty space, we investigate two different approaches within the framework of generalized polynomial chaos expansion and least-square approximation for stochastic surrogate modeling. The first approach utilizes compressed sensing techniques, specifically Least Angle Regression and Basis Pursuit Denoising methods and the second approach named 'gradient-enhanced' least-square approximation, takes advantage of the adjoint capabilities of modern CFD solvers for efficient gradient computation. A combination of the two, associating gradient information with compressed sensing is also benchmarked.
soufflante non carénée, Open Rotor, [SPI] Engineering Sciences [physics], quantification d'incertitude, optimisation robuste, Robust Optimization, Uncertainty Quantification, chaos polynomial, Polynomial Chaos, [PHYS] Physics [physics]
soufflante non carénée, Open Rotor, [SPI] Engineering Sciences [physics], quantification d'incertitude, optimisation robuste, Robust Optimization, Uncertainty Quantification, chaos polynomial, Polynomial Chaos, [PHYS] Physics [physics]
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 0 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Average | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Average | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Average |
