A Study Of Iron Ore Classification Based On Chemometrics Methods And Laser-induced Breakdown Spectroscopy

Iron ore is a significant raw material in steel-making industry, and there is a very large demand for iron ore in China. Hence, a rapid method for the classification and identification of iron ore should be established to improve the utilization efficiency of iron ore. In order to avoid the waste and loss of state funding and resources, iron ore analysis is an indispensable part for improving the quality of raw material. Iron ore analysis often include identification of iron ore and the analysis of iron. Conventional methods for iron ore analysis include atomic emission spectroscopy, inductively coupled plasma mass spectrometry(ICP-MS) and X-ray fluorescence(XRF), however, these methods often have complicated sample preparation and a longer analysis time, which impedes the rapid analysis of iron ore.Laser-induced breakdown spectroscopy(LIBS) is a new atomic spectroscopy technique, a low energy laser pulse is focused on the sample surface and plasma is generated, emission line with specific frequency is formed in cooling process of the plasma. These spectral line can be used to qualitative and quantitative analysis of samples. Due to LIBS technique have the advantages of rapid, on-site, real-time analysis and without complex sample preparation, it can be employed to the on-site and rapid analysis of iron ore.LIBS signal are often easily affected by environment and instrument, the spectral information often contains a lot of noise, and there is seriously affect for the classification accuracy without denoise. In this paper, several chemometrics methods combined with LIBS technique is employed to iron ore analysis. At first, wavelet transform(WT) and principal component analysis(PCA) are used to denoise and data reduction of LIBS spectra. It can keep effective information during the process of de-noising after the LIBS spectra preprossed by WT and PCA, which is helpful for the classification(or pattern recognition) of iron ore. And then, K-nearest neighbor(KNN) and artificial neural network(ANN) are employed to construct the classification model for five types of iron ore samples. The results shows that the two models both have good prediction ability, operation procedures of KNN is simple and rapid, however, ANN model shows a better classification performance for iron ore. Therefore, ANN method combined with LIBS can be used for the classification of iron ore, and contributes to the selection of raw material.