Application of neural networks for real time control of a ball-beam system

This paper examines the application of neural network techniques to the control of an open loop unstable ball-beam system. Computer simulation and implementation using both conventional pole-placement and neural network methods of control have been undertaken. Two-layer networks have been used in both the simulation and the experimental implementation. The error backpropagation, the temporal-difference, and the reinforcement learning algorithms have been used in the neural network controllers. The results from simulation and also from real time control experiments show that the time required to balance the ball-beam system has been significantly reduced by using neural networks