Keywords

Deep Learning, Convolutional Neural Networks, Distributed Sensing, Physics Informed Neural Networks, Convolutional Recurrent Neural Networks

Abstract

The development of small-scale local flow sensors distributed along the lifting surfaces allows new sensing methodologies to be implemented in the control of small unmanned aerial vehicles. Established methods achieve predictions of the aerodynamic parameters such as the angle of attack, sideslip angle, velocity, aerodynamic moments, etc. using artificial neural networks. Additionally, many traditional methods of calculating the aerodynamic moments exist using the underlying physics. It is proposed that these predictions could be streamlined by treating the collection of local flow measurements as an RGB “flow-field” image and performing image processing techniques to analyze the data. This data can be further collected into flow-field videos to learn time-dependency in the data. Instantaneous predictions of the angle of attack, sideslip angle, velocity, and flight mode are achieved via a convolutional recurrent neural network. Additionally, prediction of the aerodynamic moments is achieved using a novel convolutional neural network with a physics informed kernel. The performance of this new network scheme is compared to traditional methods of calculating the aerodynamic moments.

Completion Date

2024

Semester

Spring

Committee Chair

Xu, Yunjun

Degree

Master of Science in Mechanical Engineering (M.S.M.E.)

College

College of Engineering and Computer Science

Department

Mechanical and Aerospace Engineering

Degree Program

Guidance Control and Dynamics

Format

application/pdf

Language

English

Rights

In copyright

Release Date

November 2029

Length of Campus-only Access

5 years

Access Status

Masters Thesis (Campus-only Access)

Campus Location

Orlando (Main) Campus

Accessibility Status

Meets minimum standards for ETDs/HUTs

Download

Restricted to the UCF community until November 2029; it will then be open access.

Share

COinS