Theoretical Control and Experimental Verification of Carded Web Density Part III: Neural Network Controller Design

Abstract

A nonminimum phase property of a carded web system introduces difficulties for both classical control and neural network inverse model control. In this part of the series of papers, a two-stage artificial neural network model, including a controlled system learning scheme and controller design, is illustrated by application to feedback control for uniformly carded web density. A learning scheme is introduced using the dynamic model for general learning of the neural network along with a modified error back propagation algorithm based on propagation of the output error through the plant. A performance comparison is made of conventional control versus the artificial neural network control scheme, and the advantages of the new control strategy are effectively revealed by computer simulations.