Comparison of Numerical Models for Prediction of Iron's Weight Efficiency, Case Study Gol Gohar Dtp Flotation Circuit

"The aim in flotation systems is to control the grade and recovery of the product., numerical method or sampling and analysis is used to achieve this goal. Weakness of measuring instruments, lack of process knowledge, deficiencies in management and data processing in the control of flotation cells, has led the academic and industrial units to use numerical models. In this regard, using the results obtained from the experimental design carried out in the DTP line of Gol Gohar industrial and mining complex, a regression model was developed based on the analysis of variance. To present this model, after variance analysis of factors and interactions, all models were investigated and the final model was created based on 75% of the data. The model was tested with the remaining 25% of the data to confirm the goodness of the fit. The coefficient of determination for training and testing resulted 71% and 74%. Moreover, a feed forward back propagation neural network with different layers and neuron was used to model the results. The best network structure was 6-27-1 and the coefficient of determination for this network was 99% and 98% for training and test respectively. Given the complexity of neural network models and the relatively low accuracy of the regression model, neural network has a priority assuming we have access to a specialist. In addition, the results obtained from neural network can be used as expert systems to control and optimize the flotation cells.INTRODUCTIONThe primary objective of the flotation cell control is to maintain the product's (in this case tailing) recovery and grade, which represents the index for process efficiency and product quality. Continuous estimation of these indexes requires a significant amount of maintenance, and calibration of the continuous analyzers, in order to maintain an acceptable level of accuracy. Weakness of measurement equipment, lack of process knowledge and the general deficiency in management and processing of data in the flotation cell control, will lead industrial units to use smart methods such as mathematical models, regression and neural networks in near future (Panahi, Abdollahzadeh, & Sam, 2010). Artificial neural networks are inspired by the very complex structure of human brain in which millions of nerve cells solve problems or store information by communicating with each other. The duty of neural networks learning. It should be noted that learning in artificial neural networks is limited, and what is practically considered is the computing power of the network. A neural network consists of units called neurons, and has the capacity to handle a set of input data to produce a desired output data set. Each of these input and output data sets can be assumed as a vector (Soltanmohammadi & Noparast, 2006)."