Quantitative Analysis And Classification Of Iron Ores Via Laser-Induced Breakdown Spectroscopy Combined With Chemometrics

Iron ores are raw and processed materials in industry. Different types of iron ores have different physical and chemical properties, which affects the other substances ratio in steel smelting. In order to ensure the production need, efficient and effective analysis of detection should be implemented in each link, such as beneficiations, smelting, processing and transportation. However, the current traditional analysis testing technology has popular disadvantage of complex pretreatment of sample, time-consuming and offline analysis, which hinder the operation and development of the steel industry.Laser-induced breakdown spectroscopy (LIBS) is an atomic emission spectroscopy, which is determined the elemental composition and content of the sample by laser plasma emission spectra. As an emerging technology, LIBS has many advantages that can be achieved in situ on-line analysis, trace analysis, non-destructive analysis. In the steel industry, LIBS has been studied extensively for industrial process control, quantitative analysis and classification.Based on the traditional concept of chemical analysis, the external calibration in conjunction with internal standardization (ECIS) coupled with LIBS technique was proposed to perform the quantitative analysis of Fe content in iron ore. The plasma temperature and the electron number density were calculated to prove that the plasma was under local thermodynamic equilibrium (LTE) conditions and to ensure that the integral intensities of Fe I lines were reasonable. It was determined by four calibration curves, drawn for four emission lines of Fe I normalized by Mn I line (405.52nm), based on the ECIS method which can eliminate the influence of matrix effect and improve the accuracy of quantitative analysis, compared with the standard addition method. It shows a good performance to analyze the Fe content of iron ore in combination with LIBS.LIBS integrated with random forest (RF) was developed and applied to the identification and discrimination of ten iron ore grades. In addition, two parameters of the RF were optimized using out-of-bag (OOB) estimation. Finally, support vector machine (SVM) and RF machine learning methods were evaluated comparatively on their ability to predict unknown iron ore samples. RF exhibited better predictions of classification. The study presented here demonstrates that LIBS-RF is a useful technique for the identification and discrimination of iron ore samples, and is promising for automatic real-time, fast, and robust measurements.Two parameters of SVM-the penalty factor, kernel function parameters directly affect the classification results. Based on two parameters, it was proposed optimization based on support vector machine approach-genetic algorithm optimization and particle swarm optimization. GA-SVM and PSO-SVM were respectively applied to the classification of iron ores. These two methods improve the classification results of support vector machines, which PSO method shows better performance.