Accuracy enhancement of laser-induced breakdown spectroscopy by polarization spectrum fusion

Abstract

Laser-induced breakdown spectroscopy (LIBS) has been proven to be an effective way for online analysis of coal properties (such as ash content, volatile content, and calorific value), but the quantitative analysis accuracy still needs to be improved. Herein, a polarization spectral data fusion scheme was proposed to obtain more abundant spectral information and improve the performance of quantitative analysis models. First, the polarization effect of laser-induced coal plasma emission was analyzed using the Stokes parameters of the average spectra. Due to the partially polarized phenomenon of emission, fused data of PRLIBS (0°, 0) (horizontal plasma polarization spectra) and PRLIBS (90°, 0) (vertical plasma polarization spectra) was utilized to establish an efficient prediction model to obtain high prediction accuracy. For fusion models, the root mean square error prediction (RMSEP) of ash content, volatile content and calorific values were 1.497%, 0.595%, and 0.597%, respectively, and the coefficients of determination (R2) were 0.980, 0.968, and 0.978, respectively. Comparison with regression models for individual PRLIBS (0°, 0), PRLIBS (90°, 0), and LIBS data, the data fusion scheme significantly improved the performance of the prediction model. The improvement in model performance is due to the fact that there is no perfect correlation between PRLIBS (0°, 0) and PRLIBS (90°, 0), whilst polarization spectral fusion makes it possible to acquire more number of characteristic spectral lines with high importance score. The results demonstrated the feasibility of PRLIBS spectral fusion to provide better analytical results for ash content, volatile content, and calorific values of coal.