Time Suboptimal Variable Structure Control with Neural Controller

In this paper a neural network controller is combined with the equivalent time suboptimal robust controllers with the goal of preserving the guarantees provided by the robust controllers and the further reduction of joint position tracking errors. Time suboptimal control properties are preserved by disenabling the neural controller outside the domain of error guaranteed by the robust controllers and the output gains of the neural network are selected such that the neural controller is unable to accelerate the system beyond the allowed maximum velocity. Neural network weights are adjusted using a genetic algorithm and the controller performance is tested on a tracking problem involving a three link revolute joint manipulator.