Multi-Fidelity Surrogate-Based Methodology for Ultra High Aspect Ratio Strut-Braced Wing Conceptual Design
Please login to view abstract download link
This paper presents a multi-fidelity, surrogate-based design methodology aimed at integrating high-fidelity disciplinary information into the conceptual design of ultra high aspect ratio strut-braced dry wing configurations for hydrogen-powered transport aircraft. The emphasis is placed on the design methodology rather than on the specific aircraft configuration or model architecture. The proposed approach builds upon previous studies and is implemented within the open-source multidisciplinary aircraft design framework FAST-OAD. In overall aircraft design, high-fidelity aerodynamic and structural analyses usually represent one of the most computationally prohibitive elements for iterative conceptual studies. In this work, these analyses are replaced by Kriging surrogate models trained offline on dedicated datasets generated from CFD and structural optimization tools. The key feature of the methodology is the introduction of an iterative multi-fidelity strategy at the disciplinary level, which enables the systematic injection of high-fidelity effects into aircraft-level sizing while limiting the number of calls to high-fidelity solvers. Low-fidelity surrogate models are first used to perform a large initial design space exploration and to provide consistent first estimates of the coupled disciplinary variables. These estimates are then used to initialize higher-fidelity analyses, reducing the overall computational cost. In addition to the MDA resolution methodology description, particular attention is given also to buffet margin prediction and its integration within the conceptual design process. A dedicated buffet margin estimation tool is developed using wing planform parameters available at the aircraft level, allowing buffet constraints to be enforced at early design stages. Using this tool, wing planform optimization is performed to maximize the aircraft-level lift coefficient while satisfying buffet limitations through adaptive taper ratio adjustments. On the structural side, the higher-fidelity models used for surrogate training are enhanced with global buckling constraints, enabling a more realistic assessment of structural feasibility. Thanks to the extensive use of surrogate models, the framework enables large-scale design space exploration, with more than 2000 configurations evaluated efficiently at the final design iteration.
