Online Fe grade monitoring of iron ore slurry by Morse wavelet transform and lightweight convolutional neural network based on LIBS

Abstract

The Fe content or grade in ore concentrate slurry was a critical basis in the pricing and controlling of the iron ore flotation process. Therefore, our laboratory developed a LIBS-based slurry analyzer named LIBSlurry to realize the online monitoring of the Fe content. The LIBSlurry online analyzer has been running stably for over half a year after its successful installation in an iron ore flotation plant. However, the prediction error should be less than 1% according to the actual requirement of process control, and it is challenging to achieve such a long-term, stable, high-accuracy result based on the traditional linear/nonlinear analysis methods. This work proposed a new quantitative analysis method combining Morse wavelet transform (MsWT) and lightweight convolutional neural network (LCNN), named MsWT-LCNN for short. To sufficiently demonstrate the accuracy improvement of the slurry data analysis by our new research, spectra input PLS (spec-PLS), MsWT transformed image input PLS (MsWT-PLS), and spectra input LCNN (spec-LCNN) were also developed for comparison. We successively obtained 605 labeled samples from June to August 2022, in which the calibration set contained 484 samples from June to July for modeling, and the evaluation set contained 121 samples from August for verifying. The proposed MsWT-LCNN method obtained a mean absolute error on the evaluation set (MAEp) of 0.93%, which is the best and the only method with a MAEp of less than 1%. The results show that our proposed MsWT-LCNN method successfully processed and analyzed complex spectral data of iron ore slurry, realizing the successful application of the LIBS analyzer in the iron ore flotation process.