Modelo para predição da redução da temperatura do aço, entre o forno panela e o lingotamento contínuo no processo siderúrgico

• Ano de defesa: 2009
• Autor(a) principal: Iara Campolina Dias Duarte
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Engenharia Química; Redes neurais (Computação); Fundição continua; Aço; Engenharia quimica
• Link de acesso: http://hdl.handle.net/1843/BUBD-AHSLSS

This work aims to develop a neural model to estimate the steel temperature reductio caused by heat loss between the process of ladle furnace 2 and continuous casting of billets at the Gerdau-Açominas steel mill. Currently, the model used by the plant operators to predict the reduction in temperature has low acting (accomplishment), resulting in losses and increased production costs. To start the development of theneural model, the variables that influences the reduction of steel temperature caused by heat loss were mapped. The selection of variables that had the greatest influence on the desired variable was based on the correlation analysis method. A neural network type"Multilayer Perceptron" trained by the algorithm Backpropagation,with a hidden layer was used. The number of neurons in this layer was set empirically, based on the correlation coefficient, obtained in the model validation step. The algorithm used to build the neural model was developed in the "software" Matlab version 5.2. The correlation coefficient of the neural model for the prediction of temperature reduction ofthe steel, caused by thermal losses from the values measured in the industrial unit is equal to 72.3%, being much higher then the model used by operators that has a correlation of 35.3%. In conclusion, we observed the validity of the use of an artificial neural network computer tool to build models for the steel industry.