Neural Networks Applied to Adjustment and Combination of the Control Actions for the Cold Rolling Process

Abstract

The cold rolling process involves several parameters as back and front tensions, friction coefficient, among others. Any alteration in any of them will affect the output thickness of the strip being rolled. Each operation region demands a different control action. The action can be through gap, back or front tensions or, more effectively, through the combination of them. The metallurgical industry is still dependent on the operator skill, whose actions can act on several control parameters, but not simultaneously. In this work, a technique to choose the combination of the most adequate control action is presented. The technique uses a neural representation, the operator background and also the sensitivity equations of the process, obtained through the differentiation of the previously trained neural network. The expert knowledge about the choice of the control actions combined is represented through a matrix, using the concepts of fuzzy sets.