Objective In the mineral industry, the quality of ore depends on the content of ore components (mass fraction); the accurate analysis of the types and contents of elements in the ore lays the foundation for mining and beneficiation. Traditional detection methods rely on chemical methods with high accuracies, such as potassium dichromate volumetric method and flame atomic absorption spectrometry (AAS). However, it cannot simultaneously determine multiple elements, which is time-consuming and laborious and difficult to analyze a large number of samples in a short time. The emission of analytical reagents can easily cause environmental pollution. AAS is widely used for detecting ores in the laboratory. It has high accuracy and low limit of detection (LOD) in the detection of trace elements. However, self-absorption effect of high content elements leads to the failure of Lambert-Beer law, which is not suitable for detecting samples with high content elements. The above methods need to dissolve minerals in a strong acid or alkali, which are destructive to the samples and cannot be used in the industrial field. Laser-induced breakdown spectroscopy (LIBS) has been widely used in the field of multi-element analysis due to its advantages of without complex sample pretreatment, fast, and real-time detection of all elements. LIBS is a good choice for in situ and online quantitative analyses of ore elements.
Methods The quantitative analysis of Fe in iron, manganese, and chromium ores is conducted using LIBS. Due to the complexity of mineral composition, a specific algorithm is used to preprocess the spectrum to reduce the spectrum fluctuation caused by laser energy fluctuation and unstable ablation of the sample also to improve the signal-to-noise of the spectrum signal. Classification and quantitative analysis methods are combined to improve the quantification accuracy. The spectrum is preclassified by a support vector machine. Then, the linear relationship between spectral data and typical element content is established using the principal component analysis combined with multiple partial least squares regression. In this process, the numbers of input variables and principal components are optimized by the selection criteria of correlation variables and the number of principal components.
Results and Discussions The abnormal spectra are eliminated using the Pauta-Criterion. The relative standard deviation of most spectral lines decreased after removing the abnormal spectrum [Fig. 3(a)]. The principal component analysis combined with a support vector machine is used to classify the three kinds of ores [Fig. 4(a)]. The sum of the contributions of the top ten principal components of ores reaches 92.42%, indicating that these can cover most of the ore spectrum information. The score chart of the first three principal components shows that characteristic points of the same type of ore appear as obvious aggregation, and the three types of ores are very easy to distinguish [Fig. 4(b)]. The classification accuracy of the three types of ores is 100%. By full spectrum partial least squares regression analysis of unclassified ores, root mean square error of prediction (RMSEP) and average relative error (ARE) are 3.227% and 31.75%, respectively. The RMSEP of iron, manganese, and chromium ores decrease to 0.975%, 0.418%, and 0.123%, respectively, and ARE decreases to 1.46%, 6.72%, and 1.09%, respectively, after classification combined with regression-partial least squares (R-PLS), indicating that the method significantly improves the accuracy of quantitative analysis (Table 3).
Conclusions In this study, an innovative spectral pretreatment and quantitative analysis method based on LIBS analysis of ore composition is proposed. The semiquantitative classifications and quantitative regression algorithms are combined, and the selection range of regression variables is determined accurately using the category judgment, improving the prediction accuracy of the regression model. The effective spectrum is selected using the Pauta-Criterion to reduce the spectral line fluctuation caused by the experimental process. To improve the spectral signal stability, the spectral background integral intensity is used to normalize the spectral; the peak shift is corrected and the missing peak is completed. It is obtained that ores are first preclassified using the LIBS spectral analysis method. Then, the partial least square method based on the correlation variable selection mechanism is used to quantitatively analyze the content of each element in the ore, significantly improving the accuracy and robustness of the quantitative analysis. The classification accuracy for 35 kinds of ore reaches 100%, and RMSEP of prediction for Fe content in the ore is reduced to less than 1%. This study provides a theoretical and experimental basis for online rapid detection of ore.
