Loading...

Utilizing Gaussian Processes to Learn Dynamics of Unknown Torques Acting on a Spacecraft

Baradaran, Behdad | 2021

541 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53800 (45)
  4. University: Sharif University of Technology
  5. Department: Aerospace Engineering
  6. Advisor(s): Kiani, Maryam
  7. Abstract:
  8. Accurate and fast attitude estimation of a rigid body plays an essential role in the performance of a vehicle’s control system, especially aerospace vehicles. Ample works have been done to increase the accuracy and speed of the attitude estimation process, but all have been developed according to a model-based approach. This approach assumes that the torques acting on the body have a known dynamical model that is used for the attitude estimation. The purpose of the present research is to estimate the attitude via a model-free approach, i. e. dynamics of the torques acting on the body are no longer assumed to be known, and its learning is the next step. Thus, the problem formulation of this research for a rigid spacecraft is as follows. It is assumed that after performing a specific maneuver, time histories of the system state-space variables are available through a set of spacecraft sensors that have specific dynamic models. It is also assumed that in the spacecraft dynamics modeling, only the external torques are unknown. The aim is to use this history to estimate the attitude of the spacecraft with appropriate accuracy and speed, and then to learn the dynamics of the deterministic part of unknown external torques. This problem is solved by replacing the unknown dynamical model with a specific Gaussian process; where this specific Gaussian process has a special linear state-space representation that allows the use of tools such as the Kalman filter and the Rauch-Tung-Striebel smoother. This method is utilized due to its fast and simple-to-implement calculations as well as its high flexibility for modeling a wide class of functions. The main innovation of this research is the improvement and employment of this method in a wide variety of simulations and getting desired results
  9. Keywords:
  10. Gaussian process ; Kalman Filters ; Attitude Estimation ; System Identification ; Spacecraft Guidance ; Rauch-Tung-Striebel Smoother ; Random Systems with Linear State Space ; Physical Systems Learning Dynamics ; Spacecraft Attitude Estimation

 Digital Object List

 Bookmark

No TOC