Analysis of mid-infrared spectral characteristics and recognition models of coal with different particle sizes

As a core energy source, the accurate identification of coal particle size is a critical part in the quality control of raw coal mining, separation and processing, storage and transportation, and quality upgrading. The differences in physical and chemical properties of coal with different particle sizes are reflected through the vibration characteristics of molecular groups. The mid-infrared spectroscopy technology, with its real-time and non-destructive advantages, can accurately capture the characteristic difference. However, the spectral response patterns of coal with different particle sizes remain poorly understood to date. Hence, three typical coal samples, namely anthracite, coking coal and fat coal, were selected and eight particle size fractions, namely < 1 mm, 1–2 mm, 2–3 mm, 3–4 mm, 4–5 mm, 5–6 mm, 6–7 mm and 7–8 mm, were prepared. Subsequently, the Fourier transform mid-infrared spectroscopy (MIR) was employed to collect spectral data, which was then preprocessed using mean centering (MC), multiplicative scatter correction (MSC), db4 wavelet transform, standard normal variate (SNV) transformation, and the adaptive iteratively reweighted penalized least squares (airPLS) algorithm. Then, recognition models were constructed for random forest (RF), BP neural network and the Mamba algorithm based on the preprocessed feature data, respectively. Afterward, by comparing the models, the results demonstrate that the coefficient of determination (R²) of the Mamba model recognition system for particle size identification across all coal samples reaches as high as 0.95, enabling efficient and accurate determination of coal particle sizes. This research provides a scientific foundation and effective technical support for optimizing coal separation processes, enhancing the comprehensive utilization rate of coal resources, and safeguarding production safety.