Date of Award

Fall 12-15-2022

Access Type

Thesis - Open Access

Degree Name

Master of Science in Aerospace Engineering


Aerospace Engineering

Committee Chair

Hever Moncayo

Committee Advisor

Hever Moncayo

First Committee Member

Kadriye Merve Dogan

Second Committee Member

Richard Prazenica


As the development of challenging missions like on-orbit construction and collaborative inspection that involve multi-spacecraft systems increases, the requirements needed to improve post-failure safety to maintain the mission performance also increases, especially when operating under uncertain conditions. In particular, space missions that involve Distributed Spacecraft Systems (e.g, inspection, repairing, assembling, or deployment of space assets) are susceptible to failures and threats that are detrimental to the overall mission performance. This research applies a distributed Health Management System that uses a bio-inspired mechanism based on the Artificial Immune System coupled with a Support Vector Machine to obtain an optimized health monitoring system capable of detecting nominal and off-nominal system conditions. A simulation environment is developed for a fleet of spacecraft performing a low-Earth orbit inspection within close proximity of a target space asset, where the spacecraft observers follow stable relative orbits with respect to the target asset, allowing dynamics to be expressed using the Clohessy-Wiltshire-Hill equations. Additionally, based on desired points of inspection, the observers have specific attitude requirements that are achieved using Reaction Wheels as the control moment device. An adaptive control based on Deep Reinforcement Learning using an Actor-Critic-Adverse architecture is implemented to achieve high levels of mission protection, especially under disturbances that might lead to performance degradation. Numerical simulations to evaluate the capabilities of the health management architecture when the spacecraft network is subjected to failures are performed. A comparison of different attitude controllers such as Nonlinear Dynamic Inversion and Pole Placement against Deep Reinforcement Learning based controller is presented. The Dynamic Inversion controller showed better tracking performance but large control effort, while the Deep Reinforcement controller showed satisfactory tracking performance with minimal control effort. Numerical simulations successfully demonstrated the potential of both the bioinspired Health Monitoring System architecture and the controller, to detect and identify failures and overcome bounded disturbances, respectively.