Application of artificial neural network for basic oxygen steelmaking

In recent years, one of the research subjects which has been widely studied is Artificial Neural Network (ANN) Models. ANN researches are applied to various fields such as optimization, control, image processing, language acquisition and recognition, natural language processing and forecasting. In this study, firstly Structure of Artificial Neural Network and Basic Oxygen Steelmaking were briefly described. In experimental work, a model was created based on real plant data?s for ?7112 K? low carbon steel, which was obtained from BOF steel making unit working at domestic integrated iron and steel making plant. Real plant data?s used in this study for the prediction of outputs with the lowest error ratio as much as possible, which were included composition of molten pig iron, targeted steel composition and other inputs. MATLAB 7.0 R14 ? Neural Network Toolbox was used throughout this work as computer program for modeling and for the prediction of outputs. In the developed model the number of data used was 226 of which 176 of them were used for training and the rest of 50 data for allocated for testing. Normalization was carried out for inputs and outputs due to high number of data and their variability and inconsistency and tangent-sigmoid function was utilized as a function. The number of iteration used was 3000 for the Artificial Neural Network. Trail and error method was applied for Number of Hidden layers or Neuron number, which was in the range of 5, 10, 15, 20 and 25 and the lowest error ratio for Neuron number was found at 25. As a result of this prediction studies, mean value of all prediction error ratio was found to be 0.38%. Some of the predicted values and their error ratios were very close to real plant data but on the other hand, some error ratio results (such as C % and Mn %) were found to be very high. This may be due to the fluctuations of related input values. Consequently, although limited number of inputs and variables were used in the prediction and training stage, very promising prediction results that were comparatively very close to real data were obtained. Therefore it was determined that BOF steelmaking could be controlled and modeled by the application of artificial neural network.