Quantitative detection of alkali metal Na based on laser-induced breakdown spectroscopy technology

The alkali metal compounds released from high alkali fuels such as low-quality coal during thermal utilization can affect the normal operation of thermal equipment. Therefore, rapid online detection of alkali metal content in low-quality coal is of great significance for controlling alkali metal release during combustion. The alkali metal Na element is used as the detection object, and the mixed samples of graphite and sodium chloride powder with different ratios are used as experimental samples. The influencing factors of measuring Na element in samples using laser induced breakdown spectroscopy (LIBS) technology are studied. The impact of two signal strength calculation methods on signal stability is compared. The influence of experimental parameters on signal strength and signal-to-noise ratio is analyzed. The quantitative calculation models for Na element have been established. Research has shown that the characteristic spectral lines of Na element, Na I 588.995 nm and Na I 589.592 nm, are suitable as the main analytical spectral lines. Using the area intensity of the dual line characteristic spectral lines of Na element as the signal intensity can effectively improve signal stability. When the laser energy is 60 mJ and the delay time is 1000 ns, the relative standard deviation of spectral signal intensity is low and the signal-to-noise ratio is high. The quantitative calculation models are established using traditional calibration method, partial least squares (PLS) method, and support vector machine (SVR) with spectral signal intensity as the input and Na element addition in the sample as the output. The accuracy of each model is compared and analyzed. The results indicate that the PLS model may exhibit overfitting when the sample size is small and the input quantity is large. The fitting accuracy of the SVR model is 0.9783, the root mean square percentage error of the training set is 13.42%, and the root mean square percentage error of the test set is 13.51%. Compared with traditional calibration model, when the sample size is small, the SVR model can better correct the influence of matrix effects and improve the accuracy of alkali metal quantitative detection in low-quality coal.