Abstract:
The use of models to predict coke quality is commonly practiced in steel industries, which aims to adjust coal percentages in the mix to ensure coke quality for blast furnaces. In this context, a statistical model was developed to predict coke quality at Usiminas. The data set, consisting of operational parameters and the characterization of the individual coals, used to obtain the coal mix, was evaluated. The data analysis was performed by univariate and multivariate statistics, obtaining the main variables that influences metallurgical parameters related to coke quality. Furthermore, multiple regression analysis and the ability to predict results according to real value were used. The validation of the model was performed using 90-day industrial data, comparing the predicted values with real values (daily average) of the three main coke quality parameters (CRI, CSR and DI15-150). The comparison between predicted and real results revealed that the average error for the CRI is ± 5,70%; for the CSR is ± 5,60% and for the DI15-150 is ± 1,86%. The model implementation allows to increase the coke quality rate to be produced, and it should be used with the coal mix formulation model.