Introduction. Sulphide Capacity Prediction of Molten Slags by Using a Neural Network ApproachSulphide Capacity Prediction of Molten Slags by Using a Neural Network Approach Bora DERIN,1)Masanori SUZUKI2)and Toshihiro TANAKA2) 1) Metallurgical and Materials Engineering Department, Istanbul Technical University, Maslak 34469 Turkey. 2) Division of Materials and Manufacturing Science, Osaka University, 2-1 Yamadaoka, Suita, Osaka 565-0871 Japan. (Received on December 28, 2009; accepted on April 28, 2010) In the present study, the neural network approach was applied for the estimation of sulfide capacities (Cs) in binary and multi-component melts at different temperatures. The calculated results obtained using neural network computation were plotted against the experimental values for comparison comparative purposes. Besides, iso-sulfide capacity contours on liquid regions of some ternary melt phase diagrams were generated and plotted by using neural network model results. It was found that calculated results obtained through neural network computation agree very well with the experimental results and more precise than those of some models. KEY WORDS: sulfide capacities; molten melts; neural network computation; estimation. Introduction Sulfide capacity (Cs) modeling is of great interest in metallurgical engineering, since it’s aim is to predict the desulfurization power of slags which directly affect the metal quality during the production stage. There are many empirical, semi-empirical and theoretical sulfide capacity models in the literature. For example, Reddy and Blander (RB) developed a model, a priori, based on a simple solution model and on knowledge of the chemical and solution properties of sulfides and oxides. Moretti and Ottonello proposed a model for calculating the sulfide capacity of simple and complex silicate melts using different polymeric solution theory. Both models are purely theoretical and do not require any experimental data. Sosinsky and Sommerville first proposed a Cs prediction model using optical basicity concept which is related with the ratio of electron donor power of the oxides in the glass to the electron donor power of “free” oxide ion defined by Duffy and Ingram. KTH model is based on an optimization of the experimentally determined Cs values of simple systems in order to obtain those of multicomponent slags. However, above models are in good agreement with experimental results only within a certain composition and/or temperature range due to either lack of thermodynamic data or complexity of the mathematical equations which are required in the model. There are also a number of derived equations mostly based on simple regressions of experimentally obtained Cs values of simple as well as complex melt systems.8–10) However, these equations are not flexible enough to produce reliable data far from the experimental zone, such as for predicting Cs iso-contours in complex systems. In the present study, an advanced empirical approach, neural network computation will be described briefly and applied for the estimation of sulfide capacities (Cs) in multicomponent slags and flux systems which consist of SiO2, CaO, Al2O3, FeO, MgO, MnO, Na2O, CaF2, CaCl2and TiO2at different temperatures.
|
