Kiln Flame Image Information For Modeling And Predicting The End Of The Judgment

The Basic Oxygen Furnace (BOF) steelmaking end-point’s accurate control hasalways been a complex problem in steel industry. At present, the BOF steelmaking isstill the most main efficient steelmaking process in steel industry. The highest goal ofthe BOF steelmaking is to improve the stability control and gain maximum benefit.The BOF steelmaking, however, also has certain limitation in application, this is dueto the instability of the joining raw material and the severe complex high temperaturechemical reaction in the process of the steel, in addition, the thousands of strictstandards steel grade, the BOF steelmaking is still difficult to accurately controlcarbon and temperature.In order to accurately control the BOF steelmaking end-point’s judgment, thispaper carried out a series of studies, the specific content is: With the BOF processforward the flame images are different in each state. Based on the flame images topredict the BOF steelmaking end-point, the key problem is how to accurately extractthe main characteristic of the flame, according to the actual situation, this paper chosethe invariant moments features, the texture features, as well as the color space, tostudy their change process. the BOF steelmaking end-point’s judgment problem ofteninvolves a large number of independent variables, the number of independentvariables, have an impact on the entire model, the more independent variables willincrease the work-hard and the speed of modeling, the less independent variables willinfluence the accuracy of the model. The basic idea is to resolve this contradiction inselecting the impact on the dependent variable under certain guidelines, the moresignificant independent variables, set up a regression model is both reasonable andsimple and practical.Finally, using the improved PB neural network will associated image featureinformation with the BOF steelmaking end-point’s accurate control. The trainingsample data is used to establish neural network prediction model for judgment, and toestablish a good model for comparing two different data, and the requirements of end-point judgment are achieved.