Hybrid Model for Metal Temperature Control during Hot Dip Galvanizing of Steel Strip

Abstract

The paper proposes a hybrid model for predictive control under step disturbances that lead to a sharp jump in the state of the process. Similar changes occur when controlling the temperature of the steel strip on continuous hot-dip galvanizing units. Periodic changes in strip gauge or strip speed result in abrupt changes in the temperature of the steel at the outlet of the annealing furnace. During such periods deviation control is difficult requiring introduction of tolerances that limit productivity and leading to excessive heating of the metal. The paper shows that the existing proposals for controlling the temperature of the steel strip are not effective enough with a sharp change in the state of the process. The reasons for this are unknown disturbances operating in a wide frequency range and having low-frequency and trend components, as well as many influencing factors. It is shown that the problems of representativeness of the initial accumulated data make it difficult to create complex empirical models, and the level of uncertainty of the processes in the furnace makes it difficult to create complex interpretable models. The proposed hybrid model involves combining two types of simplified interpretable process models, as well as an empirical model based on an artificial neural network. The errors of the interpreted models are shown to be effectively predicted by a neural network in the presence of an additional signal from an observer of unknown disturbances. Computational experiments carried out on the data of one of the units of MMK PJSC in Russia showed that the hybrid model provides high accuracy of steel strip temperature prediction during technological disturbances and does not require frequent reconfiguration.