Unsupervised classification of iron ores based on elemental and isotopic signatures from atomic and molecular emissions of laser-induced plasma

Abstract

Background

Iron ore classification by grade and origin is critical for quality assurance in steel manufacturing. However, traditional analytical methods can be time-consuming and may not provide comprehensive information for accurate classification.

Results

The present study introduces a novel classification approach utilizing laser-induced breakdown spectroscopy (LIBS) to simultaneously discriminate iron ore grades (fine, lump, and concentrate) and origins (Australia, Brazil, and South Africa), without necessitating prior knowledge of the samples. A key aspect of this approach is to leverage multiple spectral features from laser-induced plasma: (i) atomic lines representing elemental composition, and (ii) molecular bands reflecting isotopic signatures. Based on this comprehensive spectral data, principal component analysis (PCA) was employed to visualize the correlations among the datasets, thereby illustrating how elemental composition and isotopic abundance correlate with grades and origins. Subsequently, the k-means clustering algorithm was utilized as an unsupervised classifier, eliminating the need for previously labeled training data. Consequently, the present approach could achieve an accuracy exceeding 94 % in determining the grade and origin of ores.

Significance

This study successfully established the rapid and versatile classification method for unknown iron ores, offering significant benefits for various industrial applications.