Flexible link control using multiple forward paths, multiple RBF neural networks in a direct control application

Abstract

The article presents a control scheme that uses multiple radial basis function neural networks (RBFNNs) as a direct controller for a flexible link robot. Each RBFNN is trained to specialize in one type of movement and a logical switch determines which neural network (NN) will be active for each update time. Unlike most NN controllers, this controller will be trained offline and inserted after the output error drops to an acceptable level. By training the NNs offline, the update speed of the controller is increased. The goal of this design is to produce a highly accurate controller that can be easily and inexpensively implemented in industry. Simulation results are presented when the controller is tested with an aluminum alloy link driven by a dc motor.