Data-driven modelling of nonlinear aerodynamics in high-speed aircraft using machine learning

Abstract

This thesis explores how advanced data-driven techniques can improve the modelling of nonlinear aerodynamics in high-speed aircraft, particularly in the transonic regime. As modern workflows increasingly rely on high-fidelity simulations, computational costs escalate when analysing large design spaces or unsteady phenomena. In response, this research develops and validates reduced-order modelling frameworks that integrate machine learning algorithms to provide accurate aerodynamic predictions at significantly reduced cost.

Initial studies on two-dimensional unsteady transonic loads demonstrate that flutter boundaries and other nonlinear aeroelastic features can be efficiently modelled. Building on these findings, more complex strategies are introduced, including parametric Volterra series combined with neural network interpolation, which allow rapid reconstruction of unsteady aerodynamic responses over a design space. Uncertainties in both high- and low-fidelity data are incorporated through a multi-fidelity Bayesian framework, providing confidence intervals for aerodynamic predictions. For large-scale three-dimensional test cases, advanced deep learning architectures, such as graph neural networks and spatio-temporal convolutional models, are proposed. These methods excel at handling unstructured meshes and flow features such as shock waves and separated boundary layers.

Applications include canonical transonic airfoils, wings, and wing-fuselage configurations, with emphasis on reducing simulation time while maintaining high-fidelity in predicting aerodynamic forces and flow characteristics. The proposed machine learning-based surrogates can significantly accelerate and enhance nonlinear aerodynamic analysis under both steady and unsteady conditions. By handling complex, unstructured geometries, integrating uncertainty quantification, and facilitating design space exploration, this work establishes the foundation for efficient optimisation and robust analysis in the design of high-speed aircraft and beyond.

More information

Identifiers

ORCID for Gabriele Immordino: orcid.org/0000-0003-2718-0120

ORCID for Andrea Da Ronch: orcid.org/0000-0001-7428-6935

Catalogue record

Export record