Artificial neural networks based on committee machine to predict the amount of sulfur and phosphorus in the hot metal of a blast furnace

Steel is an alloy of iron and carbon containing less than 2% carbon and small amounts of elements such as silicon, manganese, phosphorus, and sulfur, which together do not exceed 1% of the total. Sulfur and phosphorus are undesirable elements in steel because they cause brittleness. The best way to control sulfur and phosphorus content is during the production of cast iron in blast furnace. In the field of simulation and modeling, several models have been proposed for the simulation of blast furnace, which allow progress and detailed information about the fluid flow and mass and heat balances of the blast furnace. However, there are few mathematical models for the prediction of sulfur and phosphorus content. In this context, the main objective of this work was to develop an artificial neural network for predicting the sulfur and phosphorus content in cast iron. A mathematical model was developed based on a committee machine using 8 different artificial neural networks simultaneously. The artificial neural networks with a single hidden layer had neurons varying in 10, 20, 25, 30, 40, 50, 75 and 100 neurons per layer. Pearson's correlation coefficients, RMSE and MAE confirmed that the hidden layer with 25 neurons gave the best results. The conclusion is that high values of mathematical correlation demonstrate the good statistical performance of ANN and show that the mathematical model is an effective predictor of sulfur and phosphorus.