A recent study presented an AI-enhanced NIRS-XRF fusion spectroscopy method that significantly improves coal classification and quality prediction for coking enterprises.
The coal industry, despite its current economic and labor challenges, remains important for energy production in the United States and worldwide. Coal is primarily used as fuel, generating electrical power to many households (1). As recently as 2019, more than 23% of all electricity in the United States was generated by coal (1). Coal is also used to fuel several important industries, including the coal chemical and coking industries.
In a recent study, a team of researchers from Romania explored how artificial intelligence (AI) can be applied to analyze coal quality in industry environments. This study, which was published in the journal Optics & Laser Technology, demonstrates how integrating near-infrared (NIR) spectroscopy with X-ray fluorescence (XRF) spectroscopy can dramatically improve coal classification and quality prediction (2).
Coking coal is the process of softening the coal for burning, and it is primarily done to produce coke in the steel industry (3). Currently in the United States, coking coal is primarily done in Alabama, West Virginia, Arkansas, Pennsylvania, and Virginia (3). However, coking coal does come with its own set of unique challenges because of its varied sources, diverse types, and broad range of physicochemical properties (1). Coking enterprises typically rely on multiple coal varieties, including gas coal, fat coal, coking coal, and lean coal, each differing in metamorphic degree and chemical composition (1). This variability makes rapid and precise quality analysis difficult, yet it is critical for maintaining production stability and ensuring consistent coke quality.
The study investigated whether spectroscopy, when integrated with AI, can resolve this challenge. The research team developed a fusion spectroscopy approach that combines NIR spectroscopy and XRF measurements to enable a more comprehensive chemical and structural analysis of coal samples (1). The data generated was then processed using advanced machine learning (ML) and statistical methods, including principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE), to reduce dimensionality and enhance visualization of the differences among coal types (1).
Then the researchers incorporated AI to automate coal type classification. In their study, the researchers chose to apply a support vector machine (SVM) model, which allowed them to build specific regression models for each coal type using partial least squares regression (PLSR) (1). These targeted models significantly outperformed a single, unclassified model when predicting key quality indicators, such as ash content, volatile matter, and sulfur content (1).
The researchers demonstrated that for the classified regression model, the coefficient of determination (R²) reached 0.9987 for ash content, 0.9955 for volatile matter, and 0.9997 for sulfur content (1). They also discovered that prediction errors were low. They achieved root mean square error for prediction (RMSEP) values of just 0.31% for ash, 1.34% for volatile matter, and 0.05% for sulfur (1). Meanwhile, the mean absolute relative deviation for prediction (MARDP) was similarly minimal, at 2.48%, 3.58%, and 3.57%, respectively (1).
The results show how classified models improve significantly over unclassified ones. This finding has several important implications for coking coal enterprises. Coal input variability tends to high in this space, so having a method that could handle the variability and complexities of mixed-coal types is important. By integrating NIRS-XRF fusion spectroscopy with advanced data analysis, the researchers showed that their method can deliver reliable assessments while saving time (1)
Beyond the immediate benefits for coking coal quality assessment, the study’s methodology offers potential applications in other coal-dependent industries, such as coal chemical engineering and power generation. The ability to classify raw materials accurately before processing can help optimize feedstock blending, reduce waste, and improve the overall efficiency of production systems (1).
Going forward, this study highlights the growing importance of utilizing AI in industrial applications. An ongoing trend in the energy industry is to shift away from the more manual tasks by automating specific processes. The method developed in this study is one way that automated, data-driven solutions are driving industries such as coal to innovate while improving quality assessment protocols (1).
