A Bayesian approach to robust early-stage aircraft wing design under uncertainty with reduced order surrogate models

Walton, David B., Featherston, Carol ORCID: https://orcid.org/0000-0001-7548-2882, Coggon, Simon and Kundu, Abhishek ORCID: https://orcid.org/0000-0002-8714-4087 2025. A Bayesian approach to robust early-stage aircraft wing design under uncertainty with reduced order surrogate models. Presented at: AIAA SCITECH 2025 Forum, Orlando, FL, USA, 6-10 January 2025. 10.2514/6.2025-2816 Item availability restricted.

PDF - Accepted Post-Print Version Restricted to Repository staff only until 3 January 2026 due to copyright restrictions. Download (816kB)

Abstract

Industrial design of aerostructures is a complex process stemming from challenges associated with quantifying and managing the uncertainty in loading and operating conditions, variable levels of design immaturity and the potential variability in manufacturing and assembly. Setting the margins on design parameters, particularly at the early design stage of aerostructures poses significant challenges. For aircraft wing design, the loads envelopes are generated with a set of industry standard simulations of aircraft maneuver and aerodynamic loads, but these simulations are expensive and time consuming and unsuitable for rapid exploration of the design space. The study presented here develops an industrial design workflow with the aims of providing a holistic Bayesian approach for design optimization and uncertainty management in early stages of design. In order to explore the limits of the design space, it is essential to create a Bayesian data-driven surrogate model that maps the design parameters to vector-valued outputs of load-bearing and aerodynamic performance. The surrogate model is then utilized in an inverse problem setting to optimize a set of important design parameters based on user-defined operational constraints. Data compression methods have been employed to reduce the dimensionality of high-dimensional vector-valued outputs in order to effectively manage the computational overhead and ensure optimal fitting of multi-output Gaussian process regression model. Optimization of the parameters of interest is performed with Bayesian inference with novel objective functions to accommodate the variability introduced due to the presence of stochastic variables. The proposed approach is applied to an aircraft wing, parametrized by jig twists and bending and torsional stiffnesses across its span. The training data is produced as a design of experiment from coupled structural mechanics and computational fluid dynamics models that generate coefficients of lift over drag, as well as shear force, bending moment, and torque responses at a selection of flight envelopes and discrete points along the span. The results demonstrate multiple design scenarios with performance objectives and constraints for various flight envelope. It has been shown that it is possible to produce probabilistic estimates of design parameters with a Bayesian approach, while leveraging the computational efficacy of a pre-trained a stochastic surrogate model. The proposed framework therefore provides industrial design teams with information regarding potential limits of particular design choices, and help with setting design margins for later detailed design stages.

Item Type:	Conference or Workshop Item (Paper)
Date Type:	Published Online
Status:	Unpublished
Schools:	Schools > Engineering
Funders:	EPSRC
Date of First Compliant Deposit:	3 March 2025
Date of Acceptance:	2 December 2024
Last Modified:	03 Mar 2025 11:15
URI:	https://orca.cardiff.ac.uk/id/eprint/176252

Actions (repository staff only)

Edit Item