Adaptive Fault Tolerant Nonlinear Flight Control Using Neural Networks
by
Rolf Rysdyk
Georgia Institute of Technology
Coauthors: Dr. Anthony J. Calise

The work to be presented here is a cooperative effort between the flight controls lab. under supervision of Dr. Anthony Calise at Georgia Tech and Dr. Robert Chen at NASA Ames Research Center1. This work applies 'learning-while-controlling' neural networks to augment the dynamic model inversion control of the XV-15 tiltrotor aircraft. The network update law is based on a Lyapunov analysis, which guarantees boundedness of network parameters as well as tracking errors. Previous applications include, normal acceleration command augmentation for an F-18 fighter2, trajectory control for an AH-64 helicopter3, angle-of-attack augmentation for an agile missile4, and fault tolerant control of an X-33 type aircraft with multiple control surfaces.

The control architecture provides the aircraft with consistent handling qualities throughout its operational envelope, including conversion. Furthermore, it inherently provides fault tolerance to failures that might aversely affect the control of the aircraft. Conventional flight control systems, including applications of feed back linearization control, require extensive gain scheduling or require high-fidelity non-linear force and moment models that must be inverted in real time. The control architecture presented here can alleviate these requirements and thus can potentially reduce development cost and time. Most applications in the adaptive control literature assume a system that is affine in the control input. This assumption is not valid for a tiltrotor aircraft. The neural network augmented model inversion control applied here allows for non-linearities and uncertainties in the controls as well as in the states. The dynamic model inversion is linear, based on a single operating point of the aircraft. It assumes only rough knowledge of the stability and control derivatives, preferably somewhere within the operating envelope. The application in this work is based on the XV-15 in helicopter configuration, in 30 Kts. level flight at 1000 ft. in standard atmosphere. The demonstration uses NASA's Generic Tiltrotor Simulator from their XV-15 and V-22 development programs. This simulation model was extended to include actuator dynamics and non-linearities.

The control architecture is constructed as a model reference direct adaptive control scheme. It lends itself ideally to provide for Level 1 handling qualities. The command filter allows for straight-forward implementation of handling qualities. Its dynamics are frequency separated from the network update dynamics, by design. Yet, this design also prevents interference between the update dynamics and the control surface actuators.

In addition, we expect flight test results from an application to a remotely piloted Yamaha R-50 helicopter. This will include Attitude Command Attitude Hold augmentation for the pitch and roll channel, and Rate Command Heading Hold augmentation for the yaw channel. Most pertinent references:

1. Calise, A.J., and Rysdyk, R.T., "Research in Nonlinear Flight Control for Tiltrotor Aircraft Operating in the Terminal Area, " NASA NCC 2-922, November 1996.
2. Kim, B.S., and Calise, A.J., "Nonlinear Flight Control Using Neural Networks, " AIAA Journal of Guidance, Control, and Dynamics, Vol. 20, No. 1, 1997.
3. Leitner, J., Calise, A., and Prasad, J. V. R., Änalysis of Adaptive Neural Networks for Helicopter Flight Controls, " AIAA Journal of Guidance, Control, and Dynamics, Vol. 20, No. 5, 1997.
4. McFarland, M., and Calise, A.J., "Neural-Adaptive Nonlinear Autopilot Design for an Agile Anti-Air Missile", in Proceedings of the AIAA Guidance, Navigation, and Control Conference, San Diego, CA., 1996.

Date received: January 15, 1998

Copyright © 1998 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # caav-04.