Connectionist approach for modeling the dry roll magnetic separator

Abstract

Magnetic separation is an age-old technique for mineral beneficiation operations. The existing mathematical models for magnetic separators are quite complicated, and it is very difficult to measure the theoretical separation efficiency. The connectionist approach of modeling, to establish the relationships of the inputs and outputs, is preferred for this kind of problem. Experiments were carried out on a high-intensity roll magnetic separator to develop a database using four input variables (magnetic flux intensity, particle size, splitter position and roll speed) and two output variables (weight recovery and Fe recovery). The artificial neural network (ANN) and statistical methods (MLRG) were used to model the magnetic separation process. For the ANN, the regression coefficient (R2) values between the predicted and measured results were 0.89 and 0.94 for Fe recovery and weight recovery, respectively. Compared to ANN, the multivariable linear regression analysis showed inferior performance in predicting the weight recovery and the Fe recovery. The ANN models for the roll magnetic separator can predict the separation efficiency up to an acceptable limit. These models can be used to modify the operational parameters at mineral beneficiation plants to minimize the effects of variations in the raw material characteristics.