Application of neural networks to predict locked cycle flotation test results

Abstract

The performance of a continuous flotation circuit is influenced by the flotation variables and the number of stages of a flotation circuit is dependent on flotation conditions, such that the interrelation between the flotation conditions and the number of stages must be carefully determined to obtain acceptable metallurgical performance from the circuit. The locked cycle test is a useful tool for simulation of continuous flotation circuits. However, it is a time-consuming and tedious procedure. A simulation method used to predict locked cycle test results from data from individual batch tests is available in the literature. In order to develop an optimum circuit configuration for a specific ore, several batch flotation tests for the first cycle of the locked cycle test have to be conducted to predict the metallurgical performance of various circuit types. Therefore, an integrated simulation method, which uses experimental data and the results of this simulation method has been developed to structure a neural network model for prediction of locked cycle tests results without additional experiment and calculation. In training and testing of the neural network model, results of the simulation method were used as the output data set and the flotation conditions of the batch tests were used as the input data set. Apart from the training and testing data, results of the LCT for several circuit types were predicted in order to validate the neural network model and to determine its performance on both: interpolation and extrapolation. Because the neural network model was trained using results of the simulation method, the use of the neural network model did not lead to any improvement in predictions of actual LCT results. However, the results of this study indicate that the neural network model can be used to simulate various circuit types with an error of less than 4%, instead of the simulation method. Consequently, the neural network model, as an alternative to the simulation method, can be used to determine the effects of changes in certain flotation variables on the number of cleaner and scavenger stages in a flotation circuit.