PREDICTION OF PHOSPHORUS CONCENTRATION IN PRIMARY STEELMAKING
Luan Carlos de Oliveira1; Leandro Rodrigues Ramos2; Roberto Dalmaso2; Henrique Silva Furtado2; Cassius Zanetti Resende1; Daniel Cruz Cavalieri1
1 Federal Institute of Espírito Santo; 2 Arcelor Mittal
doi:10.20906/CPS/CILAMCE2017-0956
Resumo
As globalization of the economy continues, steelmakers are faced with increasing competition. Product quality is often a differentiating factor among producers, driving the demand for the development of advanced tools to improve quality during the steelmaking process. In certain applications, an elevated phosphorous concentration of the steel can be severely detrimental to the final product quality. In order to further improve quality, more advanced tools are required to better predict phosphorous concentration during the primary steelmaking process in the basic oxygen furnace (BOF) than currently exist. This work proposes the development of predictors for the concentration of phosphorous at the end of the BOF process, using supervised learning techniques Partial Least Squares (PLS), Principal Component Regression (PCR), Support Vector Regression (SVR), and Artificial Neural Network (ANN). Since steel refinement is an aggressive process conducted at high temperature (>1600°C), there are limitations in measuring the variables involved, thus only 12 static process variables were used to train the algorithms. The database used in this work contains approximately 2049 real batch processes of steel refining, and for each one there is a maximum allowable phosphorous concentration, and the resulting phosphorus concentration as determined by laboratory analysis. In the calibration and validation steps, k-fold cross-validation was used to guarantee good generalization capacity and to avoid overtraining. The root-mean-square error (RMSE) was used to compare the models. As a goal of the steel refining process is to achieve a concentration of phosphorous less than the maximum allowed concentration, the results can be classified based on the predicted concentration, its relation to the maximum allowable concentration, and the result of the laboratory analysis. If the predicted concentration is less than the allowed maximum and this is confirmed by laboratory analysis, then the result is considered a True Positive (TP). If th
Palavras-chave: Supervised Learning Techniques; Primary Steelmaking; Phosphorus Analysis