Multidisciplinary Shape Optimization of a Supersonic Jet by Reduced Order Models

The aim of this thesis, conducted in collaboration with Esteco S.p.A. and Optimad S.r.l., is to develop a preliminary study on the multidisciplinary and multi-objective shape optimization of a high-performance supersonic jet. As a reference baseline, a representative fighter aircraft featuring a conventional aerodynamic design will be considered, characterized by a single vertical tail plane, a double-surface horizontal tail plane and a truncated delta wing planform. In this study, the original configuration will be modified in order to optimize its aerodynamic and stealth observability characteristics. In detail, an investigation aimed at minimizing aerodynamic drag and radar cross section will be proposed, although constraints will be imposed in order to guarantee same lift and static longitudinal stability characteristics. This will be carried out by resorting to computational fluid dynamics and computational electromagnetism analyses, but, since multidisciplinary optimization techniques demands a high volume of design evaluations to identify non-dominated solutions, and given the considerable computational expense of these type of simulations, reduced order models will be developed to decrease computational time and cost while guaranteeing a satisfactory level of accuracy. The thesis will begin by introducing the theoretical foundations underlying the employed methodologies and tools. Subsequently, it will focus on selecting the most suitable models and simulation setups for the investigation. High-fidelity CFD and CEM analyses will be conducted to build a comprehensive database for training reduced order models. The design of experiments will involve geometrical modifications of the baseline model through surface morphing, achieved using free-form deformation algorithms. This process will alter key parameters of the main wing and horizontal tailplane, including span, dihedral and sweepback angles. Parameter sampling will be performed using a Uniform Latin Hypercube algorithm to ensure adequate design space exploration. The parameter variation range will be deliberately set across a broad spectrum to facilitate a comprehensive exploration of the solution space. This approach aims to conduct an in-depth analysis of the reduced order models properties, enabling the identification of their strengths and providing a framework for critically assessing potential limitations. ROMs will utilize a Proper Orthogonal Decomposition-based approach with interpolation, with Radial Basis Functions for CFD simulations and Gaussian Processes for CEM analyses used as regressors. The selection of these interpolation models, along with their respective kernels, will be justified by a comparative accuracy analysis relative to full order solutions. Finally, the Pareto-optimal solutions for the postulated multidisciplinary optimization problem will be identified using genetic algorithms, specifically the MOGA-II. The results achieved include designs with drag reductions of up to 43% with respect to the baseline and radar cross-section reductions of up to 21%. Calculation times have been significantly reduced - from 25 hours to 80 seconds for fluid dynamic analyses, and from 15 minutes to 10 seconds for electromagnetic analyses - while introducing errors of no more than 13% compared to full order model results.