Prediction of concentrated phosphorus grade of iron ore using mathematical analysis

The Choghart iron ore is one of Iron deposits located in central Iran in the volcanic-sedimentary basin of Bafgh. Like other natural occurrences of the earth, it has grade fluctuations throughout the deposit. Understating and awareness of these fluctuations is vital in order to increase the recovery and concentration of iron. Choghart mineral processing factory has been designed and optimized based on factory input feed tolerant to some degree of variations in grade. The aim of this paper is predicting phosphorous concentration after mineral processing based on the analysis of raw samples applying mathematical models. As a result, non-linear regression model including quadratic polynomial model showed high accuracy and by employing stepwise regression the number of predictor variables were kept as low as possible. Also, discriminant analysis has separated high and low phosphorus samples having 88% accuracy. Also, the artificial neural network could predict phosphorus grade effectively and accurately. The natural occurring earth materials including mineral resources usually include fluctuations in parameters like. Knowing and awareness of these fluctuations is vital in order to apply the necessary measures to increase the recovery. Mineral processing factories are optimized and designed based on the feed which its fluctuations should be inside an allowed limit. Most studies have been done on grade changes identifications, which mostly focused on the errors caused by different, staged of sampling and laboratory, block variance and grade dispersion. The aim of the present study is considering plant feed changes to predict the output phosphorous grade of the Choghart mineral processing factory at the Iran Central Iron Ore Company (ICIOC), Bafgh. Different statistical and artificial intelligent techniques have been employed for prediction of concentrated or based on analysis of raw materials. In the present study 110 samples (including 7 duplicates) were collected from boreholes. Samples were sent to Zarazma laboratory at Tehran for analysis and the remaining samples were returned for test with Davis tube at ICIOC. The concentrated ore after magnet separator were send to chemical laboratory in ICIOC to determine Fe and P content of samples. Different techniques including mathematical models, discriminant analysis, regression and artificial neural network have been employed in the present work to predict the P content in concentrate after Davis tube from elemental contents in the original samples. A mathematical model is applied in order to better and more effective control the grade fluctuations. In this paper methods including regression, discriminant analysis and artificial neural network were used in order to predict the target variable. Although the non-linear regression containing quadratic polynomial has a high accuracy. The step wise regression has a better performance because of the limitation number of predictor variables. Discriminant analysis and neural network have divided high and low-p content samples with proper accuracy.