Non-Destructive Coal Type Classification Using FT-NIR Spectroscopy and Machine Learning

Reliable and rapid coal quality assessment is critical for power generation, steel, cement and other coal-dependent industries, yet traditional laboratory-based methods such as proximate/ultimate analysis and calorific value determination are slow, destructive, and sample-intensive. This paper presents a non-destructive coal classification system that uses Fourier Transform Near-Infrared (FT-NIR) spectroscopy combined with machine learning to distinguish among anthracite, bituminous and lignite coals and provide quality-related information directly from spectral data. Diffuse reflectance FT-NIR spectra are acquired in the 900–2500 nm range, preprocessed using first-derivative and scaling operations, and compressed via principal component analysis (PCA). A supervised classifier (support vector machine and random forest variants) is then trained on the reduced feature space using stratified cross-validation. The resulting model is encapsulated in a deployable pipeline that accepts spectral files in CSV format and returns coal type predictions to a Flutter-based front-end through a Python backend API. The proposed approach demonstrates how modern spectroscopic techniques and machine learning can be integrated into a practical, end-to-end workflow for intelligent, non-destructive coal classification and forms a basis for future real-time coal quality monitoring solutions.