TRAINING OF NEURO-CONTROLLER USING CLOSED LOOP ERROR

Abstract

Proportion-Intetegral-Derivative (PID), Proportion-Derivative (PD) and Proportion-Intetegral (PI) controllers have been used to control linear and non-linear plant for their robustness. However, the proper adjustment of the gains in these controllers is prime importance to obtain the desire response. However, a precise knowledge of the plant is essential for the tuning of the gains of these controllers. In practice, a precise model of the plant under control may not be available or may be difficult to obtain, especially in case of non-linear plants. As such, precise control may be difficult with PID, PD and PI controllers. Feedback error learning scheme has been used to solve these type of problems. This control scheme combines the learning capabilities of an Artificial Neural Network based Controllers and the conventional PD Controllers to control DC motor, single-link robot arm and two-link robot arm. This kind of a controller provides a better control of non-linear plants which otherwise would have been difficult to control without the proper tuning of gains of the conventional controllers. It consists of feedback controller, neural network invers plant model and plant. However, the ANN has to learn the plant behavior before it can actually control the plant. Artificial Neural Network can trained off-line or on-line; in case of the off-line control method, the training is done by supplying a set of test data obtained usually from an approximate model of the non-linear plant. However, the time spent in training the neural controllers and poor performance has relegated the neural controllers to the background. On the other hand in on-line training neural network learns as it controls. In on-line training neural network takes care of any plant parameters perturbation in real time. In addition to this there is no need of precise model of plant, no precise tuning and no retuning of PD controller to account for the plant perturbances. Simulation results are presented for the speed control of a DC motor and position control of single-link robot arm and two-link robot arm.