LAUSR

AbstractThis paper aims to predict the strength of materials similar to the ionic rare earth (IRE) ores [hereinafter referred as similar materials (SM)]. A 4 × Y × 2 back propagation neural network (BPNN) prediction model, based on 18 groups of samples of the SM with different mix proportions, was used to describe their strength. The BPNN modelling scheme includes four input layer neurons, representing the amounts of kaolinite, potassium feldspar, anorthose and mica, and two output layer neurons corresponding to the strength indices c and φ of the samples after 6 h leaching. Comparing the training and prediction errors, it is verified that the error in predicted strength is minimized when the number of hidden layer neurons Y equals 9. The correlation coefficient R of the prediction model is as high as 0.998, and the maximum relative errors of the strength indices (c and φ) are 4.11% and 4.26%, respectively. Orthogonal tests show that the BPNN is a reliable and accurate method to predict the strength of SM. Featuring uniform dispersion, comparability and nonlinear optimization, the proposed method sheds further light on the strength prediction of IRE ores.