Fault Tolerant Deep Reinforcement Learning for Aerospace Applications

<p>With the growing use of Unmanned Aerial Systems, a new need has risen for intelligent algorithms that not only stabilize or control the system, but rather would also include various factors such as optimality, robustness, adaptability, tracking, decision making, and many more. In this thesis, a deep-learning-based control system is designed with fault-tolerant and disturbance rejection capabilities and applied to a high-order nonlinear dynamic system. The approach uses a Reinforcement Learning architecture that combines concepts from optimal control, robust control, and game theory to create an optimally adaptive control for disturbance rejection. Additionally, a cascaded Observer-based Kalman Filter is formulated for estimating adverse inputs to the system. Numerical simulations are presented using different nonlinear model dynamics and scenarios. The Deep Reinforcement Learning and Observer architecture is demonstrated to be a promising control system alternative for fault tolerant applications.</p>

Name thesis:degree_name

Master of Science in Aerospace Engineering

Level thesis:degree_level

Thesis - Open Access

Discipline thesis:degree_discipline

Aerospace Engineering

Year

2021

Author dc:creator

• Aoun, Christoph Elias

• deep reinforcement learning
• control systems
• aerospace applications
• Aerospace Engineering

Repository record dc:identifier

https://commons.erau.edu/edt/603

OAI identifier oai:identifier

oai:commons.erau.edu:edt-1610