Accurate acquisition of coal quality information, including the chemical composition and proximate analysis results, is of great significance for coal mixing and combustion control in thermal power plants. Laser-induced breakdown spectroscopy (LIBS) is one of the potential and competitive technologies for rapid and on-line analysis of coal quality. However, due to the difficulty in controlling the stability of the generated plasmas, the measurement repeatability of LIBS needs to be further improved. In this paper, we propose a novel X-ray fluorescence (XRF) assisted LIBS method for high repeatability analysis of coal quality, which not only inherits the ability of LIBS to directly analyze organic elements such as C and H in coal, but also uses XRF to make up for the lack of stability of LIBS in determining other inorganic ash-forming elements. By combining the elemental lines of LIBS and XRF spectra, the principal component analysis (PCA) and the partial least squares (PLS) are used to establish the prediction model and perform multi-elemental and proximate analysis of coal. Quantitative analysis results show that the relative standard deviation (RSD) of C is 0.15%, the RSDs of other elements are less than 4%, and the standard deviations (SDs) of calorific value, ash content, sulfur content and volatile matter are 0.11 MJ kg-1, 0.17%, 0.79% and 0.41% respectively, indicating that the method has good repeatability in determination of coal quality. This work is helpful to accelerate the development of LIBS in the field of rapid measurement of coal entering the power plant and on-line monitoring of coal entering the furnace.

The combination of laser-induced breakdown spectroscopy and energy dispersive X-ray fluorescence spectroscopy in the coal quality analysis was reported formerly. But in the practical test of the prototype instrument in the real power plant, the X-ray fluorescence signals suffered from intensity fluctuations over long-time measurements. The long-term signal fluctuations cause lower efficiency on the establishment of the calibration model and relatively larger root-mean-squared error of prediction (RMSEP) for unknown samples. Therefore, the spectral intensity correction was performed in the measurements; a randomly selected sample was measured several times in the whole measurements, including the modeling samples and unknown samples, recording the signal fluctuations and searching for a set of factors suitable for the intensity correction of a full-spectrum–based partial least square calibration model. In addition, as the signals of the coal samples of the power plant showed the potential of classification, the piecewise models were also established in case of further enhancement of the model or prediction accuracy. The RMSEPs of the calorific value, ash, volatile, and sulfur were lowered from 0.68 MJ/kg, 1.62%, 0.32%, and 0.24% to 0.51 MJ/kg, 1.34%, 0.16%, and 0.14% after spectral intensity correction, respectively. The piecewise modeling with spectral intensity correction achieved similar RMSEP for volatile and sulfur prediction but with more accurate models. The spectral intensity correction showed the ability to reduce the long-term signal fluctuation, and piecewise modeling also showed more efficiency in the model establishments for volatile and ash determination.