|M.Sc Student||Daniel Melman|
|Subject||Effect of Noise on the Performance of an Adaptive|
Neural Network Based Output Feedback Controller
|Department||Department of Aerospace Engineering||Supervisor||Professor Idan Moshe|
This work addresses the effect of measurement noise on the performance of a recently derived direct output feedback adaptive control scheme for uncertain nonlinear systems. The controller design analyzed is comprised of feedback linearization, linear compensation of the ideally linearized system, and an adaptive neural network based element constructed to compensate for system linearization errors. Each of those elements is susceptible to measurement noise that may affect the controlled system performance, tracking accuracy and even stability. This work provides analytical bounds for these effects. The stability analysis of the controlled system is carried out using the direct Lyapunov method. It is shown that ultimate boundedness of all signals in the closed loop system can be achieved in the presence of the measurement noise. The stability analysis also provides guidelines and constraints for controller parameters’ tuning to guarantee stable and satisfactory performance. In addition to the qualitative analytical results, the quantitative effect of bounded measurement noise on the closed loop performance is computed using two numerical examples. The first example is a flight-path-angle control application for an F-16 jet fighter. The simulation results show the effectiveness of the examined controller, in the presence of measurement noises. Furthermore, it is shown that the actual tracking error dynamics is indeed bounded by the analytical upper bounds derived. The second example examines pitch angle control application of an AT-300 agricultural aircraft. The dynamic model used for simulations and evaluation is a full nonlinear longitudinal model. The simulation results show good tracking performance in the presence of bounded measurement noise. Again, it is verified that the actual pitch angle tracking error is bounded by the analytical bound derived.