Cerfacs Enter the world of high performance ...

Aerodynamic Optimization with Uncertain Shape Parameters

MEGEP (Mécanique, Energétique, Génie civil & Procédés) [Subject to defense authorization]

YouTube link: https://youtube.com/live/uiqGay2Xkgg?feature=share

High Aspect Ratio Wings (HARW) are a promising technology for next-generation aircraft. They offer improved aerodynamic efficiency, reducing fuel consumption and carbon emissions. However, during the early design stages, structural models are often not sufficiently mature. This creates an epistemic gap between idealized numerical simulations and real-world behavior. Such discrepancies can compromise in-flight performance. As a result, engineers must perform additional fine-tuning of the wing geometry to achieve the desired performance. Hence, the motivation of this thesis is to design aerodynamic shapes that are less sensitive to geometric variations. To achieve this goal, I investigate robust design frameworks, with a particular focus on robust optimization. By incorporating uncertainty quantification into the optimization process, robust optimization enables the development of more reliable and resilient designs. Gradient-based aerodynamic optimization using the adjoint method is a mature and efficient approach that has proven effective even during the early stages of shape design. However, it operates in a high-dimensional design space because accurate representation of aerodynamic shapes requires a large number of design variables to capture the relevant physical phenomena. Consequently, the optimization process can be computationally expensive. Moreover, when uncertainty propagation is performed at every optimization iteration, the computational cost of robust optimization increases significantly. This thesis therefore proposes novel strategies to reduce the computational burden of robust optimization. A multilevel approach has been developed to improve convergence rates. Furthermore, incorporating dimension reduction technique to surrogate model techniques to reduce the number of samples required for uncertainty propagation has been assessed. This thesis demonstrates the feasibility of robust aerodynamic shape optimization under geometric uncertainties. The developed approaches are first assessed on an analytical test case using a 100-dimensional Rosenbrock function with 10 uncertain design variables, and then applied to the DLR-F25 wing-body configuration with 94 design variables, across two scenarios. In the first scenario, four design variables are considered uncertain. To further challenge the proposed methodology, a second scenario is introduced in which the uncertain space is expanded to include additional non-optimized parameters.

Jury

Mme Paola CINNELLA	Sorbonne Université	Reviewer
M. Olivier LE MAÎTRE	Polytechnique	Reviewer
M. Mathieu BALESDENT	ONERA	Examiner
M. Gilbert ROGE	Dassault aviation	Examiner
Mme Aude RONDEPIERRE	INSA Toulouse	Examiner
M. Jean-Christophe JOUHAUD	Cerfacs	Thesis supervisor
M. Joel BREZILLON	Airbus	Invited