A multi-input multi-output artificial neural network model to predict the separation characteristics of iron ore by a magnetic separator

Abstract

Due to the paramagnetic properties of hematite ore, its magnetic susceptibility naturally increases in an increasingly powerful magnetic field. One of the most popular and effective technique utilized in fine iron ore recovery is the Wet High Intensity Magnetic Separation (WHIMS). A Multi-Input-Multi-Output (MIMO), multilayer perceptron (MLP) based neural network model has been developed to predict the output parameters grade and recovery to characterize the separation behavior of a WHIMS system for processing iron ore in the particle size range of 75~300 μm. The input parameters in the Artificial Neural Network (ANN) model comprises of feed composition, %Fe, %SiO2, %Al2O3 and process parameters such as particle size, pulp density and magnetic field intensity. The neural network architecture has been optimized using highly effective Broyden-Fletcher-Goldfarb-Shanno (BFGS) network optimization algorithm to minimize the training error within few training cycles. The model is based on the data generated from WHIMS experimental investigations. There has been a very good agreement between the optimized model predictions with the measured values pertaining to recovery and grade during magnetic separation.