Rolf T. Rysdyk

Nonlinear adaptive flight control using neural networks

The effectiveness of a controller architecture, which combines adaptive feedforward neural networks with feedback linearization, has been demonstrated on a variety of flight vehicles. The boundedness of tracking error and control signals is guaranteed. The architecture can accommodate both linear-in-the-parameters networks, as well as single-hidden-layer perceptron neural networks. Both theoretical and experimental research is planned to expand and improve the applicability of the approach, and to demonstrate practical utility in the areas of cost reduction and improved flight safety.

FAULT TOLERANT FLIGHT CONTROL VIA ADAPTIVE NEURAL NETWORK AUGMENTATION

Anthony J. Calise Fellow AIAA Professor Georgia Institute of Technology Atlanta, GA This work is an extension of previous research at Georgia Tech. It demonstrates the fault-tolerance capabilities of a non-linear adaptive controller architecture. The XV-15 tiltrotor in landing configuration is used for demonstration. The failures considered affect longitudinal and lateral channels. Augmentation is provided successfully in all channels.

Adaptive Model Inversion Flight Control for Tilt-Rotor Aircraft

Neural network augmented model inversion control is used to provide a civilian tiltrotor aircraft with consistent response characteristics throughout its operating envelope, including conversion flight. The implemented response type is Attitude Command Attitude Hold in the longitudinal channel. Similar strategies can be applied to provide for Rate Command Attitude Hold in the roll channel, and Heading Hold and Turn Coordination for the yaw motion. Conventional methods require extensive gain scheduling with tiltrotor nacelle angle and speed. A control architecture is developed that can alleviate this requirement and thus has the potential to reduce development time, facilitate the implementation of handling qualities, and compensate for partial failures. One of the key aspects of the controller architecture is the accommodation of modeling error. It includes an online, i.e. learningwhile-controlling, neural network with guaranteed stability. The performance of the controller is demonstrated using the nonlinear Generic Tiltrotor Simulation code developed for the Vertical Motion Simulator at the NASA Ames Research Center.

Dynamic neural networks for output feedback control

A dynamic neural network is designed to estimate velocities from displacement measurements for a nonlinear system. A linear-in-parameters NN is used for state reconstruction. Conditions are provided under which the estimation error is guaranteed to be ultimately bounded. Subsequently, this observer is integrated into an adaptive controller architecture. The controller is based on model inversion and is augmented with a second learning-while-controlling neural network. Conditions are derived which guarantee ultimate boundedness of all the errors in the combined observer-controller feedback system. Open loop and closed loop simulations for a Van Der Pol oscillator are used to illustrate the results.

Nonlinear Adaptive Control of Tiltrotor Aircraft Using Neural Networks

Neural network augmented model inversion control is used to provide a civilian tilt-rotor aircraft with consistent response characteristics throughout its operating envelope, including conversion flight. The implemented response types are Attitude Command Attitude Hold in the longitudinal channel, and Rate Command Attitude Hold about the roll and yaw axes. This article describes the augmentation in the roll channel and the augmentation for the yaw motion including Heading Hold at low airspeeds and automatic Turn Coordination at cruise flight. Conventional methods require extensive gain scheduling with tilt-rotor nacelle angle and airspeed. A control architecture is developed that can alleviate this requirement and thus has the potential to reduce development time. It also facilitates the implementation of desired handling qualities, and permits compensation for partial failures. One of the most powerful aspects of the controller architecture is the accommodation of uncertainty in control as well as in the states. It includes an online, i.e. learning-while-controlling, neural network. Lyapunov analysis guarantees the boundedness of tracking errors and network parameters. The performance of the controller is demonstrated using the nonlinear Generic Tilt-Rotor Simulation code developed for the Vertical Motion Simulator at the NASA Ames Research Center.

Robust adaptive nonlinear flight control applications using neural networks

Online neural networks are used to enhance feedback linearizing controllers of uncertain nonlinear systems. These systems may be non-affine with respect to the control. The controller architecture is based on inversion of a linearized plant model. A multi-layer perceptron neural network is added to compensate for the inversion error. Online weight update laws are derived from Lyapunov analysis, which guarantees boundedness of all signals. Robustness with respect to unmodeled input dynamics is provided by the addition of a nonlinear damper. The controller is demonstrated using the full, nonlinear models of the XV-15 tilt-rotor and the Caltech ducted fan. Numerical results demonstrate the effectiveness of the control in response to model inversion error, and degraded actuator performance.

Development of an intelligent flight propulsion and control system - Nonlinear adaptive control

This paper describes the combination of Approximate Feedback Linearization with Neural Network augmentation to provide transport aircraft with a backup flight control system as recommended by the NTSB. The direct adaptive control does not rely on on-line parameter identification or on-line controller redesign. The on-line neural network compensates for unmodeled nonlinearities and trim errors. The architecture is applied to a midsize transport aircraft using propulsion only. Changes in vehicle configuration and operating conditions are handled without extensive tuning or controller redesign.