Organizational Unit:
Daniel Guggenheim School of Aerospace Engineering

Permanent Link
https://hdl.handle.net/1853/70742
Research Organization Registry ID
Description
Previous Names
Parent Organization
Parent Organization
Organizational Unit
Includes Organization(s)
ArchiveSpace Name Record
https://finding-aids.library.gatech.edu/agents/corporate_entities/106

Filters
1
2024 1
1
Reset filters

Settings
Author: Achour, Gabriel × Item Type: Publication × search.filters.applied.f.isSeriesOfPublicationTitle: Doctor of Philosophy with a Major in Aerospace Engineering × Type: Dissertation ×

Publication Search Results

Now showing 1 - 1 of 1
Thumbnail Image
Item

A Multi-Objective Deep Learning Methodology for Morphing Wings

2024-04-27 , Achour, Gabriel

Due to design constraints, conventional aircraft cannot achieve maximum aerodynamic performance when operating under varying missions and weather conditions. One of these constraints is the traditional approach of optimizing aircraft wings to achieve the best average aerodynamic performance for a specific mission while maintaining structural integrity. Previous studies have shown that changing the shape of wings at different points of a mission profile improves the aerodynamic performance of aircraft. As such, stakeholders have explored the viability and feasibility of changing or morphing the shape of aircraft wings to enable aircraft to adapt to varying missions and weather conditions. However, as with any other aspect of aircraft design, some challenges currently exist that hinder the development of conventional aircraft with morphing wings. First, the computational cost of flow solvers makes aerodynamic shape optimization time-consuming and computationally expensive due to its iterative nature. When designing a morphing wing, different configurations are computed for different points in the flight envelope, multiplying the computational cost necessary for morphing wing aircraft design. Consequently, a framework capable of performing shape optimization at a reduced computational cost is needed. Second, morphing can lead to a high variation of wing shapes, generating high aerodynamic loads and minimizing the aerodynamic benefits of morphing wings. Moreover, structural analyses are also computationally expensive, replicating the same challenges as aerodynamic optimization. As such, a multi-objective framework capable of optimizing morphing wings to increase aerodynamic efficiency while addressing aeroelastic constraints at a lower computational cost is needed. Finally, even though changing the shape of an aircraft’s wing at each segment of a mission profile is the most efficient approach to maximize the benefits of morphing wings, this is not ideal as flight and weather conditions are not constant throughout the flight segment. A framework that can adapt the wing shapes to varying flow conditions during the flight is needed. Consequently, this thesis aims to address these gaps by 1) developing a Conditional Generative Adversarial Network-based algorithm capable of generating optimal wing shapes of a morphing wing vehicle for each segment of a given mission profile, 2) training a Reinforcement Learning agent to modify the optimized shape and design the wing structure to ensure the structural integrity of morphing wings throughout the flight while maintaining a high aerodynamic performance 3) implementing a Meta Reinforcement Learning agent to make aircraft wings adapt their shapes to variations in flow conditions during each mission segment. The experiments outlined in this thesis involve designing each network architecture, collecting the training datasets, and training each model. These models are then applied to various aerodynamic and aero-structural optimization tasks across various demonstrated morphing wing mechanisms. Each model demonstrated accurate optimization results when compared to classical optimization methods. Additionally, the results indicate a significant reduction in computational power required by the deep learning models. As such, this thesis demonstrates the immense benefits of training and implementing deep learning models to perform various optimization tasks related to morphing wing aircraft design at a lower computational cost than traditional optimization algorithms. Furthermore, this thesis demonstrates the benefits of morphing wings throughout flight to maximize aerodynamic efficiency while minimizing structural constraints, which can lead to a non-negligible fuel consumption economy. Finally, this thesis demonstrates how meta-learning can be applied to continuously adapt the shape of a wing to unexpected changes in flow conditions throughout flight.

View