Development and application of a coal quality intelligent inspection system based on NIRS-XRF technology

Abstract

As an important component of industrialization, coking plants require high-quality coking coal. Traditional coal quality analysis methods are cumbersome and inefficient, allowing inferior coal to enter production. In this study, a coal quality intelligent inspection system is developed by combining a fully automatic sampling unit and an analysis control platform, realizing a closed-loop, unmanned process for coal collection, preparation, measurement, and storage, enabling rapid detection of industrial indicators of coal, ensuring every truck's coal quality, distinguishing genuine from fake, and enhancing the ability of coking plants to identify inferior coal. Technologically, we adopt the NIRS-XRF dual-spectral coal quality analysis technique, which combines near-infrared spectroscopy (NIRS) and X-ray fluorescence spectroscopy (XRF). This technique uses NIRS to efficiently and reliably detect organic groups in coal and XRF to accurately and reliably measure inorganic ash-forming components and sulfur elements; consequently, it enables the simultaneous and precise measurement of key indicators such as ash content, volatile matter, and sulfur content in coal. In terms of modeling strategy, we employ a multi-modeling approach to address the complex relationships between different coal quality indicators and spectra, as well as the matrix effects among different coal types. Through analysis and research on different partial least squares regression (PLSR) modeling strategies, we selected the optimal prediction models for each indicator to ensure the accuracy and reliability of the system monitoring results. Specifically, we achieved favorable prediction results for ash content and volatile matter through subtype modeling, while sulfur content attained high accuracy through holistic-segmented modeling, with coefficients of determination (R²) of 0.97, 0.94, and 0.97, respectively, root mean square errors of prediction (RMSEPs) of 0.29%, 0.92%, and 0.06%, respectively, average absolute errors (AAEs) of 0.24%, 0.76%, and 0.05%, and average relative errors (AREs) of 2.59%, 3.02%, and 3.38%. In terms of industrial applications, the system operates fully automatically, demonstrating high accuracy and repeatability, meeting the requirements of practical industrial applications. This system provides an efficient and feasible solution for rapid coal quality detection, contributing to the stability and sustainability of coking plant production and promoting the development of the entire coal-based energy industry towards intelligence and efficiency.