A model for the simulation of a cold rolling mill, using neural networks and sensitivity factors

Abstract

Rolling process mathematical modeling involves nonlinear parameters and relationships that usually lead to nonlinear equations of difficult numerical solution. Such is the case of Alexandre's model (1972), considered one of the most complete regarding rolling theory. For simulation purposes, Alexandre's model requires too much computational time, which prevents its use in online control and supervision systems. In order to obtain a model for the simulation of a cold rolling mill, it is necessary to obtain an expression to calculate the outgoing thickness and the rolling load. This function can be written in terms of the sensitivity factors and these can be obtained by differentiating an artificial neural network (ANN) previously trained, reducing the computational time necessary. In this paper, a model for the simulation of a cold rolling process based in ANN is presented. Simulation results and conclusions to show the application of the model are also presented.