Redes neurais artificiais LSTM aplicadas na previsão de padrões um forno de reaquecimento de tarugos.

Abstract

The billet reheating process within the rolling industry is a crucial step in obtaining products with the desired quality, as it ensures that the parts have the ideal mechanical properties, such as hardness, elasticity, ductility and tensile strength. The temperature inside the oven needs to be controlled within a range of 1000 to 1300°C, which varies according to the operation of the

oven, so that the parts do not melt or lose the desired properties, as well as waste of out-of- quality parts. and fuel used to promote heat transfer to the parts through combustion in the oven

zones. Thus, the present work presents the creation of Neural Networks of the LSTM type to predict patterns in the billet reheating process, so that it can be used in the construction of PI type controllers, to be used in the eight zones present in the furnace under study. Different methods were used to process the input data to create the model, such as normalization, sampling rate variation, elimination of outliers, in order to discover their influence on the accuracy of the RNA obtained. The results show that the developed RNA were satisfactory to predict data of complex processes that involve many variables, as the one studied, as well as presents the obtaining of a PI controller that allows to keep each zone of the oven within the desired temperature values for it, however. The RNA obtained for the sampling rate of 30 seconds showed a better learning of the process data. The controller developed was able to maintain the setpoint temperatures established for each zone, showing an effective tool to obtain parts with the desired quality for the mill, avoiding waste, increasing the mill's useful life, and consuming the necessary amount of fuels that feed the oven and promote the combustion necessary to transfer heat to the parts.