Quantitative Analysis of Carbon Content in Bituminous Coal by Laser-Induced Breakdown Spectroscopy Using UV Laser Radiation LI Xiongwei ( ) 1,3, MAO Xianglei ( ) 2, WANG Zhe ( ) 1, Richard E. RUSSO 2 1 State Key Lab of Power Systems, Department of Thermal Engineering, Tsinghua-BP Clean Energy Center, Tsinghua University, Beijing 100084, China 2 Lawrence Berkeley National Laboratory, University of California, Berkeley 94720, United States 3 Guodian New Energy Technology Research Institute, Beijing 100084, China Abstract The carbon content of bituminous coal samples was analyzed by laser-induced breakdown spectroscopy. The 266 nm laser radiation was utilized for laser ablation and plasma generation in air. The partial least square method and the dominant factor based PLS method were used to improve the measurement accuracy of the carbon content of coal. The results showed that the PLS model could achieve good measurement accuracy, and the dominant factor based PLS model could further improve the measurement accuracy. The coefficient of determination and the root-mean-square error of prediction of the PLS model were 0.97 and 2.19%, respectively; and those values for the dominant factor based PLS model were 0.99 and 1.51%, respectively. The results demonstrated that the 266 nm wavelength could accurately measure the carbon content of bituminous coal. Keywords: LIBS, coal, carbon content, PLS, quantitative measurement PACS: 42.62.Fi, 52.70.Kz DOI: 10.1088/1009-0630/17/11/07 (Some figures may appear in colour only in the online journal) 1 Introduction Bituminous coal is the main fuel for coal-fired power plants. As the heat value can be quickly evaluated from the carbon content for bituminous coal, the on-line or fast measurement of carbon content of bituminous coal is very helpful for power plants to realize combustion optimization and coal pricing in real time [1]. Laserinduced breakdown spectroscopy (LIBS) has great potential for the on-line or fast measurement of carbon content of coal because of its merits, including nearly no sample preparation, rapid analysis, simultaneous multielement measurement, and so on [2 4]. Up to now, the analysis of carbon content of bituminous coal by LIBS has been performed in several studies [5 8]. The commonly used partial least square (PLS) method has been utilized for quantitatively measuring the carbon content of bituminous coal [7,8], but the results are not satisfactory because of matrix effects, variations in the experimental condition, and so on [9 11]. Other data processing methods, such as the spectrum standardization method and the dominant factor based PLS method, have also been proposed to improve the measurement precision and accuracy of carbon content of bituminous coal [12 14]. Laser irradiance at 532 nm wavelength was mainly utilized for measuring the carbon content of bituminous coal in previous studies [7,8,12 14]. Their results showed that the atomic carbon emission line intensity for coal samples that have high volatile content is obviously reduced by the matrix effect, resulting in a poor linearity between the carbon content and the atomic carbon emission line intensity [12]. LIBS measurements can be influenced by the laser wavelength, which has close relationship with the processes of laser ablation and plasma generation [15]. As shown in previous studies, the ultraviolet (UV) wavelength can improve the coupling efficiency and, therefore, it has better performance than the longer wavelengths (i.e., 532 nm and 1064 nm), in the qualitative or quantitative analysis by LIBS [16,17]. Besides, in nanosecond laser-induced breakdown, the plasma shielding that is caused by the interaction of laser radiation with expanding plume, can be better reduced by the UV wavelength compared to the longer supported by National Natural Science Foundation of China (No. 51276100) and the National Basic Research Program of China (973 Program) (No. 2013CB228501). The authors also thank the financial funding from the U. S. Department of Energy, Office of Basic Energy Sciences, Chemical Science Division at Lawrence Berkeley National Laboratory (No. 2013CB228501) 928
LI Xiongwei et al.: Quantitative Analysis of Carbon Content in Bituminous Coal by LIBS wavelengths [18]. In this study, the 266 nm laser irradiance was used for measuring the carbon content of bituminous coal to investigate whether the UV laser ablation can achieve good quantitative results. The PLS method and the dominant factor based PLS method were used to establish the calibration model. 2 Description of the dominant factor based PLS model A previously established model that combines the spectrum standardization method and the dominant factor based PLS method was utilized for the measurement of carbon content in coal [14]. As the combination model utilized the spectrum standardization method to build the dominant factor model, it could still be regarded as a dominant factor based PLS model, which will be described in this section. In the previously proposed dominant factor based PLS model [7], most related spectral information was explicitly extracted to express the main part of elemental content. As the explicitly extracted spectral information expressed the dominant part in the model, it was called the dominant factor. There was still a difference between the true value of the elemental concentration and the value calculated by the dominant factor, which was caused by the fluctuations in plasma temperature and electron number density, inter-element interference and so on. The whole spectrum, containing some useful information about the sources of the difference, was utilized to further reduce the difference using PLS. It is essential to find the most related spectral line intensities to the carbon content to establish an accurate dominant factor model. Besides the atomic carbon emission line, the molecular emission lines of C 2 and CN can also be found in the coal s LIBS spectrum. C 2 can be produced either from the direct laser ablation of the coal or from the recombination of carbon atomics in the plasma [19,20], and CN can be produced either from the direct laser ablation of coal or from the reaction between C and N if the coal is measured in air. The production of C 2 and CN shows that a portion of the ablated carbon cannot radiate the atomic carbon emission, so the molecular emission of C 2 and CN should also be used to express the carbon content. A previous study has shown that the two molecular emission line intensities could be utilized to compensate the atomic carbon emission line intensity for those coal samples that have high volatile content, and the compensated carbon emission line intensity had better linearity with the carbon content compared with the atomic carbon emission line intensity [12]. Therefore, the atomic carbon emission line intensity is compensated by the two molecular emission line intensities. The carbon content can be calculated as follows: C = l 1 I C + l 2 I C2 + l 3 I CN + l 4, (1) where C is the carbon content, I C is the atomic carbon emission line intensity, I C2 is the emission line intensity of C 2 (470-474 nm ), I CN is the emission line intensity of CN (385-390 nm), and l 1, l 2, l 3, and l 4 are the regression coefficients calculated by PLS. The compensated carbon intensity can be expressed as follows: I ij = I C + l 2 l 1 I C2 + l 3 l 1 I CN, (2) where I ij is compensated carbon line intensity. The compensated carbon intensity can be influenced by the variations of plasma parameters, including the plasma temperature, the plasma electron number density, and the total number density of the measured species. The spectrum standardization method is utilized to further correct the compensated carbon line intensity [12]. In the spectrum standardization method, it is assumed that there exists a standard plasma state, in which the plasma parameters are constants. The plasma state in the measurement is regarded as a state deviated from the assumed standard state. Therefore, the deviation of the compensated carbon intensity from its value at the standard state results from the deviations of the plasma parameters from their values at the standard state. The standardized carbon intensity is calculated by Taylor expansion as follows: I ij (n s0, T 0, n e0 ) I ij (n s, T, n e ) (k 1 dn s + k 2 CdT + k 3 Cdn e ), (3) where n s0, T 0, n e0 are the standard plasma parameters; I ij (n s0, T 0, n e0 ) is the standardized carbon intensity; and, k 1, k 2, and k 3 are the coefficients. The deviations in the plasma parameters are further expressed by the measured spectral information [12]. The deviation in the total carbon number density, dn s, is calculated using the segmental spectral areas, which is k dn s = n s n s0 = k 1i I T i C + k 21 C, (4) i=1 where I T i is the segmental spectral area, k 1i and k 21 are constants. The deviation in the plasma temperature, dt, is related to the intensity ratio of two atomic emission lines according to the Boltzmann distribution principle. The full width of half maximum (FWHM) of a spectral line can be assumed to be proportional to the plasma electron number density, since for typical LIBS measurements the Stark broadening is the main cause of the spectral line broadening. The deviation in the plasma electron number density, dn e, can be calculated by the FWHM of the H α spectral line. The standardized carbon line intensity is calculated as follows: I ij (n s0, T 0, n e0 ) = I ij + +b 3 {ln k b 1i I T i C + b 2 C i=1 ( I2 I 1 ) [ ln ( I2 I 1 )] 0 } C +b 4 [ λ stark ( λ stark ) 0 ] C, (5) 929
where I ij (n s0, T 0, n e0 ) is the standardized carbon line intensity, I 2 /I 1 is the intensity ratio of two silicon atomic lines (212.412 nm and 250.689 nm), λ stark is the FWHM of the H α spectral line through Stark broadening, both [ln(i 2 /I 1 )] 0 and ( λ stark ) 0 are calculated from all the measured spectra averages, which are utilized to express their standard state values, and b 1i, b 2, b 3, and b 4 are the coefficients determined by an iterative regression process [12]. The main portion of the carbon content is expressed by the standardized carbon line intensity. By establishing a relationship between the carbon content and the standardized carbon line intensity, the dominant factor model can be established as follows: C i = ki ij (n s0, T 0, n e0 ) + b, (6) where C i is the carbon concentration calculated from the dominant factor; I ij (n s0, T 0, n e0 ) is the standard carbon line intensity; and, k and b are regressed coefficients. The deviation between the true value of the carbon content and the value calculated from Eq. (6) was compensated by the whole spectrum information using PLS to further improve the inadequacies of the dominant factor, inter-element interference, and other unknown factors. The final expression of the combined model is C = ki ij (n s0, T 0, n e0 ) + b + e 0 + e 1 x 1 + e 2 x 2... + e n x n, (7) where C is the calculated carbon concentration of the combination model; x 1, x 2,..., x n are the spectral intensities at different wave lengths; and, e 0, e 1, e 2,..., e n are the regression coefficients calculated by PLS. 3 Experimental setup The carbon content of bituminous coal was measured by the RT100-EC LIBS system (ASI Inc., USA). The 266 nm laser irradiance with a 5 ns pulse duration was emitted by a Q-switched Nd:YAG laser. The laser energy was 9 mj/pulse. The diameter of the laser spot on the sample surface was 100 µm. Six Czerny-Turner spectrographs and charge coupled device (CCD) detectors in the LIBS system covered an overall range (nm) from 190 to 309, 309 to 460, 460 to 588, 588 to 692, 692 to 884, and 884 to 1041, respectively. The nominal spectral resolution was 0.07 nm. The gate delay time was 0.1 µs. The integration time was fixed to 1 ms. The samples used in the experiment were 24 bituminous coal samples, which were certified by the China Coal Research Institute. As shown in Table 1, the carbon content in these coal samples ranged from 42% to 82%, and the volatile matter content in these coal samples ranged from 11% to 35%. Coal powders were firstly placed in an aluminum pellet die with a diameter of 30 mm and a height of 3 mm, and then pressed into coal pellets at a pressure of 20 tons for subsequent measurement. The twenty-four bituminous coal samples were divided into a calibration set and a validation set to establish the calibration model and evaluating the performance of the calibration model. Sixteen samples were used for calibration, and eight samples were used for validation. The samples were arranged by the carbon content, and then one of every three samples was chosen for validation, ensuring an even and wide range distribution of the carbon content in both sets. Each coal pellet was measured at twenty-five locations on the pellet surface. The aerosol particles produced from each laser shot were blown off to eliminate the aerosol influence on the measurement. The emission line intensity was calculated by integrating the spectral intensities of an emission line after subtracting the background emission intensities. The system was warmed up for at least half an hour before the experiment. 4 Results and discussion A previous study has shown that the result of calibration between the carbon content and the atomic carbon emission line intensity at 247 nm for bituminous coal samples was not good, and the coefficient of determination (R 2 ) was only 0.46 [12]. As shown in Fig. 1, the R 2 of the calibration curve is 0.60 in this study. The improvement in R 2 shows that the linearity between the carbon content and the atomic carbon emission line intensity at 247 nm is improved compared with the previous study. When only those coal samples whose volatile matter content is less than 23% as the calibration samples were selected, the R 2 can be further increased to 0.80, as shown in Fig. 2. Yet, there is still a clear diminution of atomic carbon emission intensity for some of those coal samples that have a high volatile content, indicating that the matrix effect is not completely eliminated by using the UV laser for ablation. Table 1. Carbon content of 24 bituminous coal samples Calibration set Validation set No. C(%) Volatile matter (%) No. C(%) Volatile matter (%) No. C(%) Volatile matter (%) 1 47.12 11.31 9 70.45 14.41 17 53.42 25.58 2 52.61 23.23 10 74.7 33.4 18 55.67 19.11 3 53.77 14.03 11 76.69 33.41 19 59.91 28.9 4 54.72 13.1 12 77.28 32.22 20 72.71 30.91 5 58.12 30.43 13 78.64 33.9 21 75.96 32.94 6 59.84 28.65 14 79.02 11.42 22 78.58 32.41 7 67.18 18.21 15 79.98 31.92 23 79.7 15.3 8 67.77 34.46 16 81.54 12.43 24 81.45 11 930
LI Xiongwei et al.: Quantitative Analysis of Carbon Content in Bituminous Coal by LIBS Fig.1 Calibration plot of the atomic carbon emission line intensity at 247 nm versus the carbon content Fig.3 Calibration and validation results of the PLS model Fig.2 Calibration plot for those coal samples that have low volatile content To achieve a good quantitative analysis result, the PLS model and the dominant factor based PLS model were established. The PLS model was established using the whole spectral information, which included all the intensities at each wavelength in the whole spectrum. The number of principle components in the PLS model was evaluated by the leave-one-out cross validation (LOO-CV) method to avoid noise over-fitting. The calibration and validation results of the established PLS model are shown in Fig. 3, which shows that R 2 and root-mean-square error of prediction (RMSEP) of the PLS model are 0.97 and 2.19%, respectively. The results show that when the PLS method is utilized to build the calibration model, the 266 nm wavelength can accurately measure the carbon content of coal. Fig. 4 demonstrates the calibration and validation results of the dominant factor based PLS model. As shown in Fig. 4, R 2 is 0.99, and RMSEP is 1.51%. The R 2 is improved and the RMSEP is reduced compared with the PLS model, indicating that the prediction accuracy can be improved by explicitly extracting the most related spectral information and utilizing the whole spectrum information to further reduce the deviations. Fig. 4 also shows that the 266 nm wavelength is capable of good performance in measuring the carbon content of coal. Fig.4 Calibration and validation results of the dominant factor based PLS model 5 Conclusions The carbon content of bituminous coal samples was measured by LIBS using the 266 nm laser wavelength. Compared with the previous study, the linearity between the carbon content and the atomic carbon emission line intensity at 247 nm was improved. Yet, the calibration plot of the atomic carbon emission line intensity at 247 nm versus the carbon content still showed that the matrix effect in the measurement of carbon content of coal was not completely eliminated. The PLS model was established to measure the carbon content of bituminous coal. The dominant factor based PLS model was also established to further improve the measurement accuracy of the carbon content of coal. The results showed that when the two multivariate calibration models were utilized, the 266 nm wavelength was capable of good performance in measuring the carbon content of coal. References 1 Yuan T B, Wang Z, Lui S L, et al. 2013, J. Anal. At. Spectrom., 28: 1045 2 Russo R E, Mao X L, Liu H C, et al. 2002, Talanta, 57: 425 931
3 Wang Z, Yuan T B, Hou Z Y, et al. 2014, Front. Phys. 9: 419 4 Russo R E, Mao X L, Gonzalez J J, et al. 2013, Anal. Chem., 85: 6162 5 Yin W B, Zhang L, Dong L, et al. 2009 Appl. Spectrosc., 63: 865 6 Li J, Lu J D, Lin Z X, et al. 2009, Optics & Laser Technology, 41: 907 7 Feng J, Wang Z, West L, et al. 2011, Anal. Bioanal. Chem., 400: 3261 8 Wang Z, Yuan T B, Lui S L, et al. 2012, Front. Phys., 7: 708 9 Ma Q L, Motto-Ros V, Lei W Q, et al. 2010, Spectrochim. Acta B, 65: 896 10 Castle B C, Talabardon K, Smith B W, et al. 1998, Appl. Spectrosc., 52: 649 11 Yu J, Ma Q L, Motto-Ros V, et al. 2012, Front. Phys., 7: 649 12 Li X W, Wang Z, Fu Y T, et al. 2014, Appl. Spectrosc., 68: 955 13 Feng J, Wang Z, Li L, et al. 2013, Appl. Spectrosc., 67: 291 14 Li X W, Wang Z, Fu Y T, et al. 2014, Spectrochim. Acta B, 99: 82 15 Shaikh N M, Kalhoro M S, Hussain A, et al. 2013, Spectrochim. Acta B, 88: 198 16 Barnett C, Cahoon E, Almirall J R, et al. 2008, Spectrochim. Acta B, 63: 1016 17 Cahoon E M, Almirall J R. 2010, Appl. Opt., 49: 49 18 Ma Q L, Motto-Ros V, Laye F, et al. 2012, Appl. Phys., 111: 053301 19 Dong M R, Mao X L, Gonzalez J J, et al. 2013, Anal. Chem., 85: 2899 20 Dong M R, Chan George C-Y, Mao X L, et al. 2014, Spectrochim. Acta B, 100: 62 (Manuscript received 25 April 2015) (Manuscript accepted 12 June 2015) E-mail address of LI Xiongwei: lixiongwei@cgdc.com.cn 932