Control Methods for a Hot Steel Rolling Mill: An Application of Learning Theory and Pattern Recognition

Abstract

This paper presents roll force control methods to be used with the predictive force setup model of the finishing stands in a hot steel rolling mill. Current mill practices achieve a desired strip gauge by using a predictive force model to setup the roll gaps on the finishing stands. At any time before the steel enters the first finishing stand a human operator may modify the roll gap settings if it is felt that under the present conditions the force predicted by the setup model is going to be unacceptable. In this paper, the decision process of the operator is modelled by pattern recognition methods to obtain this extra degree of feedforward control. In addition, feedback control is provided from one steel run to the next by an adaptive controller which uses a linear reinforcement learning scheme to adjust its parameters. Results are presented from actual mill data.