Quantitative analysis of copper-molybdenum slurries based on low energy pulse laser, combined with artificial neural network and principal component analysis

Abstract

Listen

Translate article

In the mining industry, flotation serves as a critical step in the on-stream processing of slurry. A prompt and accurate acquisition of target elemental concentrations is fundamental for evaluating the efficiency of the flotation process. In this study, an algorithm of principal component analysis (PCA) combined with an artificial neural network (ANN) was proposed. This algorithm was based on a low energy pulse laser and a micro-spectrometer, incorporating the principle of laser-induced breakdown spectroscopy (LIBS). The method optimized the number of features for training the models, assisted by cross-validation. The copper concentrations of 36 slurries containing mixed copper-molybdenum powders were quantified. Compared to the full-spectrum model, the full-spectrum PCA model reduced the number of features from 271 to 5 while maintaining an appropriate fit. Meanwhile, the effect of the continuum on prediction accuracy was also analyzed, revealing that the continuum improved the prediction accuracy. This improvement can be explained by the fact that the continuum contains essential information, which serves as a correction for the intensity of the spectral lines. The results indicated that the prediction accuracy of the full-spectrum PCA model was significantly enhanced in comparison to the peak intensity and full-spectrum models, with the correlation coefficient (R2) of the test set improving from 0.894 to 0.985 and the root mean square error (RMSE) decreasing from 0.18% to 0.08%.