Real-time tracking control of a DC motor using a neural network

Abstract

Neural networks are well-suited for the modeling and control of complex physical systems because of their ability to handle complex input-output mapping (through supervised learning) without detailed analytical model of the systems. This paper investigates a Multilayer Neural Network (MNN) for the real-time identification and control of a DC motor. The MNN is first trained to learn the inverse dynamics of the system and after the training is complete, the neural network is used as a feedforward controller to generate the input voltage for the motor to follow a pre-selected trajectories in position and speed. The training data is generated from the hardware setup (DC motor and servo amplifier) by applying a control voltage to the servo amplifier and observing the system output (motor speed). The main advantage of this scheme is that it does not require any knowledge of the system dynamics (and its nonlinear characteristics) and therefore treat the system as a black box. Experimental results show that the MNN is capable of identifying the motor system accurately and is able to control its position and speed with high degree of accuracy, even in the presence of disturbances.