A study on the accurate classification of complex coal samples using Raman-XRF combined spectroscopy

Abstract

Coal is one of the most important energy resources globally, and its classification and analysis are crucial for improving utilization efficiency, optimizing process design, and achieving cleaner usage. However, due to the complex composition of coal and significant differences in its physicochemical properties, traditional single-spectral detection techniques often struggle to simultaneously consider the information on organic molecular structures and inorganic components in coal sample classification, resulting in limited classification accuracy. To address this issue, this study proposed a combined detection technique based on Raman spectroscopy and X-ray fluorescence (XRF) spectroscopy, integrated with machine learning algorithms, to achieve precise classification of complex coal samples. Raman spectroscopy provided high sensitivity to the organic components of coal samples, focusing on molecular structures and aromatic group characteristics, while XRF spectroscopy revealed inorganic components through its rapid quantitative capabilities for multiple elements. The combination of these two types of spectra achieved a complementarity of organic and inorganic information, effectively enhancing the accuracy and stability of the classification. This study first conducted standard deviation (SD) and coefficient of variation (CV)analyses on the collected Raman and XRF spectral data to assess their stability. The results indicated that the Raman and XRF spectral data sets possessed good stability, providing a solid foundation for the classification study. Subsequently, various machine learning algorithms, including Support Vector Classifier (SVC) and Random Forest (RF), were employed for coal sample classification, with model parameters optimized through cross-validation grid search. The research demonstrated that the SVC model based on Raman-XRF fusion data significantly improved classification accuracy compared to single-spectral techniques, with increases of 5.95 % (Raman) and 4.76 % (XRF), thereby validating the advantages of Raman spectroscopy in detecting organic molecular structures and the efficient complementarity of XRF spectroscopy in chemical element analysis. This study not only enhances the classification accuracy of complex coal samples but also provides theoretical and technical support for online real-time detection of coal in complex environments within the coal industry.