PLS regression-assisted LIBS for simultaneous quantitative analysis of five REEs and TREOs in rare earth ores

Abstract

Rare earth elements (REEs) are national critical strategic resources, and the rapid and accurate quantification of individual REEs and total rare earth oxides (TREOs) in rare earth ores is of great significance for safeguarding resource security and facilitating industrial regulation. This study adopted a laser-induced breakdown spectroscopy (LIBS) combined with partial least squares (PLS) synergistic quantitative analysis strategy to achieve simultaneous and accurate determination of five key REEs (La, Ce, Pr, Nd, and Sm) and TREOs in rare earth ore samples. Pellet samples with concentration gradients were first prepared, and their spectra were collected. After systematically comparing different spectral preprocessing methods, it was found that the hybrid preprocessing method combining standard normal variate and wavelet transform (SNV-WT) could significantly improve calibration model performance. Subsequently, the competitive adaptive reweighted sampling (CARS) algorithm was further introduced for variable screening to optimize the calibration model. The finally established SNV-WT-CARS-PLS calibration model demonstrated excellent predictive performance and robustness: the R_p² for the five REEs and TREOs was all above 0.9345, the MRE_p were all below 0.1309, and the RPD values ranged from 3.9 to 5.5. External validation and 95% confidence interval analysis confirmed the calibration model's reliable predictive capability for different rare earth elements across a wide concentration range. The results indicate that LIBS combined with PLS can provide an efficient, accurate, and rapid quantitative analysis method for rare earth ores with complex compositions and wide concentration spans, offering strong technical support for on-site detection and process optimization in rare earth smelting.