Objective The steel industry is an important economic pillar of China. At present, sinter accounts for more than 70% of blast furnace iron- making charge. Maintaining the stability of each component of sinter is important to ensure smelting efficiency and quality. Presently, sinter composition analysis generally adopts off- line sampling, and then adopts chemical titration or X-ray fluorescence spectrometry to determine the complex samples after preparation. Some use neutron activation technology to perform online measurement, but because of radioactivity, they are not widely used. Laser-induced breakdown spectroscopy (LIBS) has been widely used in many fields owing to its advantages of real-time, online, in situ, no radiation contamination, and total element analysis. In the production process of sinter, the mixture contains a certain amount of moisture, and the level of moisture affects the intensity and stability of the LIBS spectrum. Studies on the influence of the sample moisture mainly focus on soil, coal, rock debris, etc.; however, studies on the influence of the sinter mixture moisture have not been reported. In this study, the influence of moisture on the quantitative analysis accuracy of Fe, Ca, Si and Mg mass fraction in sinter is discussed by configuring sintered mixture samples with different moisture contents, and a correction method to reduce the influence of the moisture content by spectral standardization is proposed. The spectral stability and quantitative analysis accuracy are improved by correcting the spectral differences under different moisture contents. Methods The moisture content in sinter has considerable influence on the strength of the LIBS spectrum as well as the accuracy and stability of the prediction of the quantitative analysis model. In the experiment, the spectral features were screened using the wrapping algorithm in combination with the ridge regression model, and the optimal subset selected was placed in the partial least squares (PLS) model to establish a quantitative analysis model. Further, a spectrum standardization method is proposed based on the spectral line intensity correlation matching to calculate the intensity correlation of each dimension spectral line between standard and unstandardized spectrum matrices. The absolute of the correlation was sorted in descending order. The former m-dimensional unstandardized spectral lines corresponding to the sorted absolute were extracted to fit the standard spectral lines. Thus, the spectral difference caused by the different moisture contents is corrected, and the influence of moisture content on the quantitative analysis is corrected. Results and Discussions By evaluating the relationship between the moisture content and mean value of the spectral intensity, the average spectrum with the mean intensity value between 50 and 70 under each sample was finally determined as standard spectrum under the sample, and the spectrum with the mean intensity value out of 50 to 70 was regarded as unstandardized spectrum under the sample. The standard spectrum in the training set was used to establish the quantitative analysis model of the four elements; the standard and unstandardized spectra in the test set were predicted separately. It was deduced that the measurement results of the two were far from each other, and the error bars of the results of the four elements predicted by the model were large (Fig. 4, Fig. 5). The SICMS spectrum standardization model was established under the optimal parameters, and the unstandardized spectrum under each sample in the test set was standardized. The difference in the spectral line intensity between the unstandardized and standard spectra was significantly reduced. After the quantitative analysis of the standardized spectra, it was deduced that relative standard deviation (RSD) of the quantitative analysis results of Fe, Ca, Si, and Mg generally decreased by more than half. For example, the RSD of Fe in sample No. 9 decreased from 6.01% to 2.32%, and the RSD of Si in sample No. 10 decreased from 8.42% to 0.81% ( Fig. 10). Additionally, the error bar of the mass fraction prediction value was significantly shortened, and the distance between the square and circular points in the Fig. 11 was significantly reduced. Conclusions To overcome the problem that moisture in the sintered mixture will affect the intensity of the LIBS spectrum and the accuracy and stability of the quantitative analysis, the relationship between water content and spectral intensity of the sintered mixture was first studied. It was deduced that the mean spectral intensity had a linear relationship with moisture content, and the mean spectral intensity could be used to indirectly characterize the moisture. A quantitative analysis model was then established, and an SICMS spectral standardization method was proposed to correct the effect of moisture. The results showed that compared with the measurement results before standardization, the RSD of the standardized spectrum was reduced by more than half, and the accuracy was also significantly improved. The SICMS spectral standardization method is not only suitable for the correction of the moisture effect of the sintered mixture, but also has reference significance for the correction of the moisture effect in other types of targets.
