Online analysis of coal particle flow by laser-induced breakdown spectroscopy based on pelletized coal calibration samples and feature-based transfer learning

Abstract

The application of laser-induced breakdown spectroscopy (LIBS) for directly measuring coal particle flow is an optimal choice for the actual industrial operations. With the objective of facilitating the detection of particle flow, we established a LIBS detection system coupled with the coal particle circulation bench, which can continuously and automatically provide particle flow samples for laser ablation. A quantitative analysis method for particle flow combining feature-based transfer learning was proposed, so a dual-mode optical LIBS module was designed and integrated into this system to obtain the spectral signals from different forms of coal samples (pellet and particle flow) through the same optical configuration. The spectral characteristics and the correlation between pellet and particle flow were firstly analyzed. Then a spectral correction method based on polynomial fitting was proposed to enhance the correlation between the pellet spectra and particle flow spectra. Finally, the feature space mapping method was introduced for improving the effect of feature transfer, and the model was trained on highly stable pellet spectra to perform a direct quantitative analysis of coal particle flow. The results demonstrated that the root mean square error (RMSE) for the analysis of calorific value, volatile matter, and ash content of particle flow was 0.757 MJ/kg, 2.630 %, and 3.034 %, respectively. This work provides a practical application scheme for on-line analysis of coal particle flow.