A hybrid quantification model and its application for coal analysis using laser induced breakdown spectroscopy

Abstract

The low sample-to-sample reproducibility of laser induced breakdown spectroscopy (LIBS) is the most critical obstacle to wide commercialization of the technology. Averaging multi-pulse normalized spectra is the only applied method to improve the sample-to-sample reproducibility; however, the method can only improve the reproducibility to some extent because the averaging process can only eliminate random noise and has no effect on non-random noise, and because normalization can only partially and indirectly remove the signal fluctuations. In addition, the measurement accuracy of LIBS is also limited due to matrix effects as well as signal uncertainty. In this work, we propose a set method to improve both precision (sample-to-sample reproducibility) and accuracy for LIBS quantification. The method includes three steps: (1) the intensities of all spectral lines from every single pulse are converted to a standard state value using a “spectrum standardization” method to reduce the measurement uncertainties to acceptable levels; (2) the standardized spectra are compared with a large spectral database to check whether the current sample is a new sample or is already in the database with known composition/property information; and (3) if the sample is found to be a new sample, a dominant factor based partial least square (PLS) model is applied to provide quantitative analytical results, and the new standardized spectral information and analytical results are inserted into the spectrum database, making the database self-adaptable for future measurements; while if the sample is found to be already in the database, the analytical results are directly obtained from the database. The proposed method was applied for coal analysis. The results showed that the relative standard deviation (RSD) of carbon for different measurements of the same sample is 0.3%, proving that LIBS is able to provide high reproducibility, at least for coal analysis applications. The average measurement errors for carbon, hydrogen, volatiles, ash and heat values are 0.42%, 0.05%, 0.07%, 0.17% and 0.07 MJ kg⁻¹, respectively, and all of these measurement accuracies fully meet the requirements of the national standard for coal analyses using traditional chemical processing methods. This is the first quantitative application of LIBS with real industrial requirements, and the present work proves the feasibility of LIBS for accurate quantification from a technical point of view.