When system parameters change abruptly due to, for example, actuator damage in aircraft, or an unknown slung load disturbance in rotorcraft hover control, the use of neural network reconfigurable control has proven to be effective in mitigating the resulting disturbances and returning the controller performance to the desired baseline. Despite its continued success in both the theoretical and practical domains, there still exists a crucial missing link that thwarts not only acceptance in both government and industrial applications, but also the robustness guarantees that indicate how far the system is from instability, and how much time-delay can the feedback loops tolerate. Currently, there is no equivalent notion of phase and gain margin for neural network adaptive control. We therefore see a critical need to address this division with both sound theoretical analysis and experimental validation. This proposal seeks to (1) develop the analytical techniques necessary to overcome this division, (2) develop a software toolbox built on top of MATLAB to assist the engineer in applying these techniques to a wide range of control problems and (3) perform rigorous simulation based testing on an aircraft with three failure modes, and also on a rotorcraft with unknown sling load attached.
Keywords: Stability Margins, Reconfigurable Control, Neural Network Adaptive Control
