Modeling, Adaptive Control, and Flight Testing of a Lighter-than-Air Vehicle Validated Using System Identification
Open Access
- Author:
- Messinger, Steve
- Graduate Program:
- Aerospace Engineering (MS)
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- March 21, 2022
- Committee Members:
- Amy Pritchett, Program Head/Chair
Michael Andrew Yukish, Thesis Advisor/Co-Advisor
Simon W Miller, Committee Member - Keywords:
- Model Reference Adaptive Control
System Identification
Adaptive Control
Lighter-than-Air Vehicle
Vehicle Design
Sparse System Identification
Flight Testing
Modeling
Simulation
Control
Aerospace - Abstract:
- Lighter-than-air vehicles (LTAV), or airships, can be useful platforms to test the autonomy and swarming capabilities of collaborative multi-agent systems. Airships have many advantages compared with other unoccupied aerial vehicles such as simplistic construction, longer flight times, and slower dynamics. The complex dynamic effects due to buoyancy forces can create a challenging control problem with varied size, weight, and power considerations. For autonomous operation, a non-linear model of the system is needed to test the controller performance in simulation and explore model-based control techniques. In this work, a LTAV system is used to compete in the Defend the Republic competition --- an autonomous aerial match where two opposing teams of LTAVs are trying to capture and score neutrally buoyant game balls into opposing team's goals in an indoor arena. The first main contribution of this work is to build and design a prototype LTAV that is highly maneuverable and controllable in unknown environments with many disturbances. The system must have enough actuation for a human operator and an autonomous system to capture and score game balls. Designing an autonomous system requires state estimation and a control system. The corresponding hardware to accommodate the processes needed for autonomous operation must be selected and included on the platform. LTAVs present a challenge due to their low weight budget. Hardware selection was carefully selected to best balance weight, maneuverability, and compute power of the vehicle. The second main contribution of this work is to model the vehicle using system identification techniques. A simple non-linear model is formed using linear least-squares, then sub-sampling based threshold sparse Bayesian regression is used to accurately discover where the simple model is not accurate and update the corresponding coefficients. An analogous simulated model is developed and verified using additional flight testing data captured from the LTAV. The third contribution of this work is the use of the LTAV identified model to develop a neural network model reference adaptive control system. During gameplay, the control system will allow the vehicle to autonomously track a user-provided trajectory. In this work, the control architecture is explained, implemented on the simulated system, and implemented on the real system. Performance of the control system is tuned, assessed, and quantified on both the simulated and real system. The control system is shown to provide tracking of the commanded state in real life and simulation during multiple trajectories used in the Defend the Republic competition.