Design and implementation of real-time neural control systems

Abstract

This research concentrates on the implementation of different control strategies used to make a two degrees of freedom non-linear robotic manipulator track a path in polar co-ordinates on a horizontal table. It employs a method for controlling the torque delivered by the manipulator's D.C. motors in an inner control loop. Sliding mode and neural controllers are used to handle plant nonlinearity and uncertainties in a digital outer loop. Ultimately, all three control schemes are combined into one controller, where each part addresses a different problem. Special emphasis is directed towards identification of the non-linear unknown functions governing the manipulator's dynamics by the use of Gaussian radial basis function neural networks and the implementation of the latter on hardware. Both simulation and experimental results are presented, compared and evaluated where applicable. This research concentrates on the implementation of different control strategies used to make a two degrees of freedom non-linear robotic manipulator track a path in polar co-ordinates on a horizontal table. It employs a method for controlling the torque delivered by the manipulator's D.C. motors in an inner control loop. Sliding mode and neural controllers are used to handle plant nonlinearity and uncertainties in a digital outer loop. Ultimately, all three control schemes are combined into one controller, where each part addresses a different problem. Special emphasis is directed towards identification of the non-linear unknown functions governing the manipulator's dynamics by the use of Gaussian radial basis function neural networks and the implementation of the latter on hardware. Both simulation and experimental results are presented, compared and evaluated where applicable.