Evanthia Kallou
(Advisor: Prof. Dimitri N. Mavris)
will defend a doctoral thesis entitled,
A Set-Based Methodology for Aircraft Design Space Exploration Enriched with Higher Fidelity Data
On
Wednesday, April 26th at 9:00 a.m. EDT
CoVE, Weber SST II
and
Abstract
The aircraft design process is constantly evolving and subject to changes in political, industrial, and economic environments, which can affect performance and cost requirements. Decisions made early in the design process have downstream effects on the total project cost and design decisions. Additionally, novel configurations lack historical data, which makes early decisions uncertain. As the design progresses from conceptual to preliminary, more variables are needed to describe the design, making it difficult to perform optimization. Additionally, during preliminary design, there are challenges such as organizational barriers, uncertainty propagation, and the need to choose appropriate model fidelity. To address these challenges, a novel multi-level Set-Based Design Space Exploration (SBDSE) method is proposed, which uses classification and supervised dimensionality reduction methods to define and communicate design sets between disciplines and subsystems.
The objective of this methodology is to establish an effective framework that facilitates both horizontal and vertical design communication and integration, while addressing the challenges inherent in design space exploration for preliminary design through the use of the SBDSE. The benefits of using the SBDSE methodology will be demonstrated through a multi-level aircraft design problem. Furthermore, the situations in which each communication scenario is preferable are identified. The consideration and documentation of physics assumptions, the computational budget, and the analysis code’s accuracy are crucial, while setting up the MDAO for each system and subsystem. The first research area regards the comparison between the multi-level communication of a point design versus a design set with hypercubic bounds. The second research area aims to tackle the issue of increased dimensionality in the input design space when analyzing lower-level subsystems and components. Dimensionality reduction methods can assist in identifying critical design variables and directions to sample from. If the dimensionality of the input design space for a subsystem is too computationally complex, the design of experiments (DoE) should be for the latent space. Dimensionality reduction should be based on the MDAO outputs to establish correlations and should occur before generation of approximation models. The number of design samples required for training approximation models is determined by the dimensionality of the design space. Finally, the third research area highlights the importance of selecting suitable analysis tools for each design group within a decomposition level, based on factors such as the complexity of the design, the type of analysis required, and the level of fidelity needed.
The expected contributions of this thesis are to provide a framework for more efficient design communication and integration, tackle challenges in design space exploration, demonstrate the benefits of SBDSE on a multi-level aircraft design problem, and emphasize the importance of documenting and considering physics assumptions, computational budget, and analysis code accuracy in an MDAO setup.
Committee
- Prof. Dimitri N. Mavris – School of Aerospace Engineering (advisor)
- Prof. Graeme Kennedy – School of Aerospace Engineering
- Prof. Daniel P. Schrage – School of Aerospace Engineering
- Dr. Burak Bagdatli – School of Aerospace Engineering
- Mr. Simon Coggon – Airbus UK