Aero-structural multi-fidelity coupling methodology for preliminary aircraft design

Overview & Synopsis

The increasing environmental and efficiency demands in aviation necessitate the exploration of novel aircraft configurations and advanced design methodologies. High Aspect Ratio Strut-Braced Wing (HARSBW) configurations present a promising solution by reducing fuel consumption and aerodynamic drag. However, their complex aero-structural interactions require sophisticated Multi-Disciplinary Analysis (MDA) and Multi-Disciplinary Analysis and Optimization (MDAO) methodologies. This thesis develops an aero-structural multi-fidelity coupling methodology that integrates surrogate modeling and dimension reduction techniques to enhance computational efficiency without sacrificing accuracy. The research begins with a bibliographical review outlining the challenges in aircraft design and the potential of MDAO to overcome them. A key issue is the high computational cost of high-fidelity simulations required for accurate MDA, motivating the need for surrogate-based modeling approaches. One of the challenges regarding a surrogate-based approach is to deal with high dimensional coupling variables corresponding to the vector of aerodynamic forces and structural displacements. To address this, the thesis first develops the parametric models for aerodynamic and structural simulations tailored to HARSBW. These models include low-fidelity potential fluid and high-fidelity compressible Euler aerodynamic solvers, as well as a linear elastic structural model. Next, a direct solver-based MDA framework is constructed and analyzed at different fidelity levels. The trade-offs between computational cost and accuracy are assessed. Based on the low and high-fidelity direct solver-coupled MDA schemes, this research constructs single and multi-fidelity dimension-reduced surrogate model-based MDA schemes, combining Proper Orthogonal Decomposition (POD) and Gaussian Process (GP) regression. This framework captures high-dimensional coupling variables efficiently, significantly reducing reliance on expensive high-fidelity solvers. A first attempt of further refinement of the multi-fidelity dimension-reduced surrogate model is carried out through an iterative enrichment algorithm that selectively incorporates additional data in high-uncertainty regions for the training of the POD basis and the GP surrogate models. Application of the developed algorithms on the HARSBW configuration illustrate the interest of such a proposed methodology in reducing computational costs while maintaining high-fidelity accuracy.