, Volume 400, Issue 10, pp 3261-3271
Date: 18 Mar 2011

A PLS model based on dominant factor for coal analysis using laser-induced breakdown spectroscopy

Rent the article at a discount

Rent now

* Final gross prices may vary according to local VAT.

Get Access

Abstract

Thirty-three bituminous coal samples were utilized to test the application of laser-induced breakdown spectroscopy technique for coal elemental concentration measurement in the air. The heterogeneity of the samples and the pyrolysis or combustion of coal during the laser–sample interaction processes were analyzed to be the main reason for large fluctuation of detected spectra and low calibration quality. Compared with the generally applied normalization with the whole spectral area, normalization with segmental spectral area was found to largely improve the measurement precision and accuracy. The concentrations of major element C in coal were determined by a novel partial least squares (PLS) model based on dominant factor. Dominant C concentration information was taken from the carbon characteristic line intensity since it contains the most-related information, even if not accurately. This dominant factor model was further improved by inducting non-linear relation by partially modeling the inter-element interference effect. The residuals were further corrected by PLS with the full spectrum information. With the physical-principle-based dominant factor to calculate the main quantitative information and to partially explicitly include the non-linear relation, the proposed PLS model avoids the overuse of unrelated noise to some extent and becomes more robust over a wider C concentration range. Results show that RMSEP in the proposed PLS model decreased to 4.47% from 5.52% for the conventional PLS with full spectrum input, while R 2 remained as high as 0.999, and RMSEC&P was reduced from 3.60% to 2.92%, showing the overall improvement of the proposed PLS model.

Published in the special issue Laser-Induced Breakdown Spectroscopy with Guest Editors Jagdish P. Singh, Jose Almirall, Mohamad Sabsabi, and Andrzej Miziolek.