Date of Award

Summer 2023

Embargo Period

8-2024

Access Type

Thesis - Open Access

Degree Name

Master of Aerospace Engineering

Department

Aerospace Engineering

Committee Chair

Hever Moncayo

First Committee Member

Kadriye Dogan

Second Committee Member

Morad Nazari

College Dean

James Gregory

Abstract

The interest in utilizing multi-agent systems (MAS) has increased in the aerospace industry. Its scalability, efficiency, robustness, fault tolerance, and cost-effectiveness make it ideal for performing real-world missions that require more than one agent. However, in completing the tasks, the multi-agent systems are still vulnerable to environmental disturbances, cyber-attacks, and hardware failures. Therefore, an adaptive distributed fault-tolerant control architecture is needed to minimize the impacts of the previously stated circumstances and ensure the mission can continue successfully.

This thesis describes the development of an experimental setup for testing and validating an adaptive consensus algorithm and a bio-inspired health management architecture. The distributed adaptive consensus algorithm gathers information states from the agents, processes the shared information, and decides the subsequent action for each agent until they all agree on a particular state. This act is also known as consensus. The adaptation law in the algorithm mitigates the effects of disturbances on the agents when trying to reach a consensus by minimizing the error between the system states and the reference model state. The addition of the health management system in the architecture is for detecting abnormal conditions, such as hardware failures and disturbances, alerting the other agents, and starting a joint action that mitigates the adverse effects. Inspired by the biological immune system, it mimics the capability of the human immune system in classifying threats as non-self and normal conditions as self. A supervised machine learning method, called Support Vector Machine, is utilized to improve its classification capabilities.

These control architectures are implemented and flight tested on the developed testbed to analyze and verify that the intelligent architecture can drive the agents to reach a consensus and that the health management system can detect failures and mitigate their effects. The testbed consists of a multi-agent system of four Crazyflie 2.1 quadcopters, a VICON motion capture system for tracking the drones, and the Crazyswarm software architecture to send commands to the Crazyflie and receive data from the VICON system. In this thesis, a simulation environment incorporating the distributed control architecture and the nonlinear models of the quadcopters is also developed to perform preliminary tests and demonstrate that the desired results are obtainable.

Share

COinS