High accuracy raw coal analysis by LIBS via fusing feature aware and score-space anomaly detection

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Materials Science

Background:

• Online analysis of heterogeneous solids like coal presents challenges due to matrix effects and anomalous samples.
• Laser-induced breakdown spectroscopy (LIBS) is a promising technique but requires robust methods to handle complex spectral data.
• Accurate ash quantification is crucial for coal quality control and process optimization.

Purpose of the Study:

• To develop a robust ash quantification model for raw coal trucks using LIBS.
• To address challenges of matrix effects and anomalous sample interference in LIBS analysis of coal.
• To achieve high accuracy and industrial robustness for real-time coal quality assurance.

Main Methods:

• Integrated analytical strategy combining feature-aware spectral classification and Principal Component Analysis (PCA)-based anomaly detection.
• Classification of coal spectra into three ash grades (ULA, LA, MA) using cyano radical (CN) and diatomic carbon radical (C2) emission bands.
• Anomaly detection in principal component space to identify and eliminate spectral outliers.

Main Results:

• Grade-specific partial least squares regression models achieved high prediction accuracy (R² > 0.92) with low root mean square errors of test (RMSET) across all ash grades.
• The method successfully filtered non-coal impurities and identified anomalous spectra.
• Field validation on moving coal trucks showed excellent agreement with laboratory methods (measurement error < 1.0 wt%).

Conclusions:

• The developed LIBS analysis paradigm offers a generalizable approach for reliable spectral analysis of heterogeneous materials.
• The method provides a potent solution for real-time coal quality assurance.
• This technique can be extended for online compositional monitoring of other complex industrial materials.