Multifidelity Data Fusion Applied to Aircraft Wing Pressure Distribution

Anhichem, Mehdi, Timme, Sebastian ORCID: 0000-0002-2409-1686, Castagna, Jony, Peace, Andrew and Maina, Moira (2022) Multifidelity Data Fusion Applied to Aircraft Wing Pressure Distribution. In: AIAA AVIATION 2022 Forum, 2022-6-27 - 2022-7-1, Chicago, IL.

Text Paper-Aviation2022-MA.pdf - Author Accepted Manuscript Download (1MB) | Preview

Abstract

Designing an aircraft and analysing its performance requires uncertainty-aware and robust aerodynamic data. The three principal ways of acquiring such aerodynamic data are flight testing, wind tunnel testing and numerical analysis. They all can be expensive and are subject to multiple sources of uncertainty. A multifidelity data fusion framework applied to surface pressure data of a large aircraft wing model obtained from experiment and simulation is presented herein with the ambition to enhance its intended use in optimising wind tunnel campaigns and eventually in high-value design. Static pressure tapping and dynamic pressure-sensitive paint are exploited to obtain experimental data sets of different quality and quantity. These are complemented by numerical data ranging from a linear potential panel method to non-linear Reynolds-averaged Navier–Stokes simulations. The data fusion approach introduced by Lam, Allaire and Willcox (2015) and revisited here is non-hierarchical, meaning there is not an absolute hierarchy in terms of accuracy between the information sources. The confidence in an information source over the parameter space is defined through a fidelity function. The approach uses Gaussian processes to enable the fusion of experimental and numerical data into a single multifidelity surrogate model. The generated (and required) volume of data to study the surface flow and distributed aerodynamic loads on an aircraft wing leads to scalability issues. A suitable extension of Gaussian process regression based on stochastic variational inference has been adopted alongside the use of GPU architecture to enable the application of the data fusion framework on large data sets. The approach provides a surrogate model with a quantified uncertainty from uncertainty-aware disparate data sources. We explore (and adapt) the framework for a high-dimensional practical data set generated through industrial wind tunnel testing and numerical flow analysis.

Item Type:	Conference or Workshop Item (Unspecified)
Uncontrolled Keywords:	Generic health relevance
Divisions:	Faculty of Science and Engineering > School of Engineering
Depositing User:	Symplectic Admin
Date Deposited:	21 Jun 2022 08:45
Last Modified:	15 Mar 2024 08:59
DOI:	10.2514/6.2022-3526
Related URLs:	Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3156818