Neuro Adaptive Control For Aerospace And Distributed Systems

Abstract

Nonlinear and adaptive control is generally considered one of the most effective techniques for stabilizing complex nonlinear systems, where linear control techniques may fail completely. Thousands of research papers are published on either theory or applications of nonlinear and adaptive control. But often one obvious question arises how to implement these techniques in real life model? The best answer that one can think of is to develop simple nonlinear control laws which are easy to implement. Moreover for controlling multi-agent systems, it is often required to distribute the control laws based on limited information available among the agents. This research provides some of these issues in the following way. Autopilot design for Aerospace systems: this research developes adaptive backstepping and dynamic inversion methods with internal dynamics stabilization for the quadrotor. Quadrotor helicopter models usually show two main characteristics. First, strong coupling among the system states and second, underactuation where many states are to be controlled with few control inputs. Due to these unique characteristics, the design of stabilizing control inputs is always challenging for quadrotor models. To confront these problems, first, a dynamic inversion technique with zero dynamics stabilization loop is introduced to a practical quadrotor model, second, an adaptive-backstepping technique is developed to a lagrangian quadrotor model. The stabilizing control laws for both of these techniques are developed using on Lyapunov based method; and Coordination of multi-agent systems: coordination among multiple agents is generally done based on balanced or bi-directed communication graph models. If the agents are nonlinear and passive then for a balanced graph model synchronization is possible. But, for other than balanced and bi-directed graph models, it is difficult to synchronize nonlinear systems. Moreover, the performance of synchronization is normally dependent on the second largest eigenvalue of the laplacian matrix of the network graph. This eigenvalue is also known as the Fiedler eigenvalue. This research shows how to implement distributed nonlinear and adaptive controllers using pinning techniques for generalized directed communication graph models. The dynamics of the agent at each node are non-identical and unknown. A Lyapunov based technique is adopted to show the tracking performance when the tracker dynamics are also considered unknown. It is also shown using pinning adaptive control that the synchronization speed no longer depends on Fiedler eigenvalue of the graph laplacian matrix. The research also develops duality properties of linear controllers and observers for general cooperative di-graph systems. The choice of gains for controller and observer using riccati equations ensures stable synchronization on general di-graphs. Finally the research is extended to decentralized control of HVAC systems using pinning control methodology.