{"id":209,"identifier":"LOR-00209","url":"https://lifeofresearch.com/theses/209","title":"Physics-Informed Machine Learning for Aerodynamic Flow Prediction and Optimisation of Next-Generation Aircraft","abstract":"The design of next-generation aircraft—blended-wing-body configurations, truss-braced\r\nwings, distributed-propulsion concepts and hydrogen-powered airframes—requires the\r\nexploration of design spaces that are far larger and less well understood than those of\r\nconventional tube-and-wing transports. High-fidelity computational fluid dynamics (CFD)\r\nremains the workhorse of aerodynamic analysis, but its cost limits the number of designs that\r\ncan be evaluated, and purely data-driven surrogates trained on CFD databases generalise\r\npoorly beyond their training distribution and offer no guarantee that predictions respect\r\nconservation laws.\r\nThis thesis develops a physics-informed machine learning (PIML) framework that\r\nembeds the governing equations of compressible, turbulent flow directly into the training\r\nof neural network surrogates, and couples these surrogates to gradient-based aerodynamic\r\nshape optimisation. Three contributions are pursued. First, a hybrid physics-informed neural\r\noperator is formulated that maps aerofoil and wing geometry, together with flight conditions,\r\nto the full Reynolds-averaged Navier–Stokes (RANS) flow field, using a composite loss\r\nin which sparse CFD data are regularised by PDE residuals, boundary conditions and\r\nintegral force constraints. Second, training pathologies that are known to afflict physicsinformed networks in aerodynamic settings—stiff loss landscapes, spectral bias towards lowfrequency solutions, and conflicting gradients between loss terms—are addressed through\r\nadaptive loss balancing, Fourier feature embeddings, boundary-layer-aware sampling and\r\nhard enforcement of wall boundary conditions. Third, the trained surrogate is embedded\r\nin a multi-fidelity, trust-region shape optimisation loop in which design sensitivities are\r\nobtained by automatic differentiation and predictive uncertainty, estimated with deep\r\nensembles, governs when the surrogate is refined with new high-fidelity CFD samples.\r\nThe framework is developed on two-dimensional aerofoil benchmarks and extended\r\nto three-dimensional transonic wings representative of next-generation transport aircraft.\r\nA potential-flow panel method implemented as part of the data-generation pipeline is\r\nverified against classical thin-aerofoil theory. The thesis concludes with a critical discussion\r\nof the accuracy, cost, robustness and certification implications of using physics-informed\r\nsurrogates in the conceptual and preliminary design of aircraft.","authors":[{"name":"Dominic Namachila","affiliation":"University of Cape Town","orcid":""}],"type":"thesis","license":"CC-BY-4.0","keywords":[],"publication_date":"2024-12-15","version":1,"views":77,"downloads":41,"created_at":"2026-11-10 12:41:39","doi":"","source":"","source_url":"","files":[{"id":229,"name":"PIML_Aerodynamics.pdf","size":997045,"mime":"application/pdf","downloads":41,"download_url":"https://lifeofresearch.com/theses/209/files/229/PIML_Aerodynamics.pdf"}]}