VISCOSITY PREDICTION OF C-A-S AND C-A-S-M-SLAG SYSTEMS BY MEANS OF AN ARTIFICIAL NEURAL NETWORK

摘要

Slags are of fundamental importance in the steel production, influencing deeply its quality; among the several properties of steelmaking slags, viscosity stands out as one of the most important. Viscosity is a function of slag composition andtemperature and is determined experimentally or using mathematical models.ln this work a study is made todetermineviscosity by means of an Artificial Neural Network using Wekasoftware developed at the University of Waikato.Weka offers several algorithms for machine learning, including those for neural network training. Here, the Multilayer Perceptron model (MLP) was used andtraining was carried out by means of the error back-propagation algorithm. For this introductory analysis, the composition of the chosen slags falls within the most frequently used quaternary oxide system CaO-MgO-SiO_2-AI_2O_3 (also known as C-S-A-M) and the ternary C-S-A, while the temperature is kept constant at 1600 °C.Because of the facilities offered, the primary viscosity input data for selected slags was provided by means of FactSage - a software specialized in the field of metallurgical thermodynamics (nevertheless, literature or experimental data can be used as well). A refined predictive system for viscosity could be established by ANN with correlation coefficients between predicted and observed values (R2) greater than 0.99 for both systems. The model can be of benefit to the steel industry and may contribute to the production of quality steels.