Data-driven Aerodynamic Shape Optimisation for Morphing Configurations

Nowadays, morphing aerofoil stands as a valuable configuration to accomplish the required cost reduction and greening objectives. Nevertheless, due to the numerous evaluations needed for the design of these morphing architectures, gradient and CFD-based ASO is prohibitive for these optimisations. Consequently, the state-of-the-art implements surrogate-based optimisation frameworks to strike a balance between accuracy and efficiency. However data-driven optimisation still encounters challenges in generalisation. Indeed, given the complexity of morphing architectures, their development is recommended for a mission profile comprising several flight conditions. This forces the researchers to constrain both the flight envelope and the design space of interest. It is noteworthy that while this approach enhances performance, the generalisation and applicability of the model are reduced. This thesis work presents an efficient data-driven framework for the ASO of two-dimensional morphing aerofoils, comprising both subsonic and transonic regimes. With the aim to address the limits of the existing deep-learning models, an extensive database comprising more than 140,000 samples, 1,200 aerofoil geometries and 120 flight conditions, is collected and used to train a network employed to accomplish the aerodynamic computations. Coupling the model with free-form deformation and genetic algorithm, significant drag reduction for morphing configurations is achieved, preserving the structural integrity of a wing box embedded in the profile. However, there are important discrepancies in the model predictions, particularly in subsonic flight conditions, which reveal challenges in learning the underlying physics of the field. Nevertheless, the overall framework shows promise, displaying convergence and efficiency, and establishing a solid foundation for future outcomes.