پديدآورندگان :
Zandi-Atashbar Navid zandnavid@yahoo.com Laboratory of Iron Making, Mobarakeh Steel Company, Isfahan, Iran / Department of Chemistry, Isfahan University of Technology, Isfahan, Iran; , Hassanpour Javad Laboratory of Iron Making, Mobarakeh Steel Company, Isfahan, Iran / Department of Chemistry, Isfahan University of Technology, Isfahan, Iran
كليدواژه :
Pattern Recognition , PLS , DA , PCA , ECVA , Iron Ore Concentrate , Pelletizing , Reducibility
چكيده فارسي :
Iron ores are rocks and minerals from which metallic iron can be economically extracted. The ores are generally rich in iron oxides, where it can be formed from magnetite (Fe3O4), hematite (Fe2O3), goethite (FeO(OH)), limonite (FeO(OH).nH2O) and siderite (FeCO3). Different compositions of the iron ores as well as other oxides including TiO2, SiO2, Al2O3, CaO,, MgO, V2O5, P2O5, and MnOx represented different effects on the mechanical and chemical iron products, so, the identification of the origin of the iron ore can be influenced on the pelletizing and reduction processes [1]. The data obtained from x-ray fluorescence spectroscopy, classical determination of iron content, and carbon and sulphur determination by high frequency combustion method with infrared measurement were studied by the multivariate pattern recognition methods including the most usual unsupervised method principal component analysis (PCA) and the supervised methods partial least squares discriminant analysis (PLS-DA) and extended canonical variates analysis (ECVA) [2,3]. In this work, five various iron ore concentrates including Golgohar, Bafgh, Chadormaloo, Kimiamaaden, and Goharzamin were investigated. 350 samples were prepared from these iron ore origins. The dataset was randomly divided into three subsets of training (50% of the samples), evaluation (30%) and test (20%) for classification methods PLS-DA and ECVA. PCA and ECVA were showed their abilities in discrimination of iron ore concentrates. PLS-DA represented so close performances, giving CC% on the training, validation, and test sets in the ranges of 89.4–93.2, 89.1-91.9, and 89.0-91.2%, respectively. These iron ore concentrates were employed in pelletizing and sintering processes by addition of some binders such as limestone, dolostone, olivine and bentonite [4,5]. The iron pelletes represented different characteristics and reducibilities because of their compositions. Hence, the pattern recognition was applied as an efficient tool to classify these iron products into various categories based on the initial origins. Accordingly, the reducibility and other mechanical and chemical properties of iron product can be controlled by considering the iron ore composition. Moreover, these properties can be modelled rather than the expecting composition. In conclusion, the merge of raw composition-based data and chemometrics tools was applied to successfully anticipate the final product.
