Fuel 258 (2019) 116150
Contents lists available at ScienceDirect
Fuel journal homepage: www.elsevier.com/locate/fuel
Full Length Article
Feasibility study of gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel using laser-induced breakdown spectroscopy
T
⁎
Zhimin Lua,b,c, Xiaoxuan Chena,b,c, Shunchun Yaoa,b,c, , Huaiqing Qina,b,c, Lifeng Zhanga,b,c, Xiayang Yaoa,b,c, Ziyu Yua,b,c, Jidong Lua,b,c a b c
School of Electric Power, South China University of Technology, Guangzhou, Guangdong 510640, China Guangdong Province Key Laboratory of Efficient and Clean Energy Utilization, Guangzhou, Guangdong 510640, China Guangdong Province Engineering Research Center of High Efficient and Low Pollution Energy Conversion, Guangzhou, Guangdong 510640, China
A R T I C LE I N FO
A B S T R A C T
Keywords: Laser-induced breakdown spectroscopy (LIBS) Solid biomass fuel Gross calorific value Carbon content Volatile matter Ash content
Rapid determination of the solid biomass fuel properties is essential for optimizing the combustion process of biomass. In this work, a feasibility study on using laser-induced breakdown spectroscopy (LIBS) in conjunction with partial least squares (PLS) for simultaneous measurement of gross calorific value, carbon content, volatile matter content and ash content was carried out for 66 wood pellet samples. The best quantitative analysis results were obtained with the PLS model based on spectra that combined baseline correction with Z-score standardization. The root mean square error of prediction (RMSEP) of the gross calorific value, carbon content, volatile matter content and ash content were 0.33 MJ/kg, 0.65%, 1.11% and 0.38% respectively, while the average standard deviation (ASD) were 0.08 MJ/kg, 0.15%, 0.43% and 0.16% respectively.
1. Introduction Biomass composition changes significantly during biomass processing and utilization, which is especially true in China, where there is no mature supply chain for the solid biomass fuel. In China, grid connected dedicated biomass power installed capacity has reached 17.81 GW till 2018 [1], and this figure is planned to increase to 30 GW by 2030 [2]. Other from the dedicated biomass firing, co-firing biomass with coal for power generation and CHP (combined heat and power) generation has gained great momentum since 2017, when National Energy Administration and Ministry of Ecology and Environment of China established a coal-biomass coupling power demonstration program, granting 89 projects for cofiring of coal with agricultural and forestry wastes (especially straw) and sludge [3]. However, highly heterogeneous nature of the lignocellulosic biomass feedstock remains a major problem in many processes [4], for example, combustion, which is the most widely used thermochemical processes [5]. The design and operation of biomass combustion systems depend substantially on several biomass property indexes, e.g., gross calorific value, carbon content, volatile matter content and ash content. Traditionally, the contents of property indexes in biomass are measured in accordance with different national and international
⁎
normative, such as GB/T 28731-2012 [6] for gross caloric value; DL/T 568-2013 [7] for carbon content; ASTM E-1755 [8] and GB/T 307272014 [9] for ash content; ASTM E-872 [10] and GB/T 30727-2014 [9] for volatile matter content determination. All those methods are timeconsuming and tedious, which normally required several hours. A fast, simple, reliable and accurate method of the solid biomass fuel properties would be desirable, and would certainly help the biomass processing and utilization. A number of spectroscopic techniques are available for on-line characterization of biomass feedstock properties. For example, near infrared (NIR) spectroscopy has been shown to be an effective method in the prediction of the gross calorific value, carbon content, volatile matter content, ash content and moisture content of biomass fuel [11–13]. X-ray fluorescence (XRF) spectroscopy has been applied in the analysis of the heating value and ash content of biomass fuel [14]. However, there are also disadvantages for these techniques, such as overlapping spectral peaks and high requirement of the sample homogeneity (for NIR); difficulties in analyzing light elements such as C and O, requiring high maintenance cost and having potential radiation hazard (for XRF). Laser induced breakdown spectroscopy (LIBS) is a relatively new analytical spectroscopy technology, with many attractive advantages, such as no radiation hazard, less or no sample preparation
Corresponding author at: School of Electric Power, South China University of Technology, Guangzhou, Guangdong 510640, China. E-mail address:
[email protected] (S. Yao).
https://doi.org/10.1016/j.fuel.2019.116150 Received 26 May 2019; Received in revised form 22 July 2019; Accepted 4 September 2019 Available online 10 September 2019 0016-2361/ © 2019 Elsevier Ltd. All rights reserved.
Fuel 258 (2019) 116150
Z. Lu, et al.
Table 1 Descriptive statistics of reference values of property indexes of biomass fuel. Fuel property indexes
Minimum
Maximum
Mean
Standard deviation
Gross calorific value (ad) (MJ/kg) Carbon content (d) (%) Volatile matter content (ad) (%) Ash content (ad) (%)
16.00 43.96 70.39 0.58
20.35 50.12 83.92 9.71
18.95 47.43 77.93 2.80
1.15 1.75 3.39 2.20
Notes: ad = air dry basis; d = dry basis.
1.05 ms. Pulsed laser was focused 2 mm below the sample surface. The laser interacts with the biomass sample to form a plasma, and the resulting spectral signal is then transmitted through optical fibers to a four-channel spectrometer. The spectrometer scans a spectral coverage of 178–354 nm, 342–499 nm, 487–620 nm and 607–827 nm simultaneously with a spectral resolution of 0.3–0.4 nm. In order to reduce the influences of sample composition inhomogeneity and shot-to-shot spectral fluctuations, 15 points on each sample surface were randomly selected by manually adjusting the mobile platform, each point was hit by 20 pulsed lasers continuously, and a total of 300 spectral data were collected for each sample.
requirements, and ability of simultaneous rapid analysis of multiple elements, which make it used in more and more various fields [15–17]. Therefore, LIBS is one of the most promising candidates for the characterization of biomass feedstock properties in harsh industrial environments. At present, in biomass field, LIBS technology is mainly used for the detection of alkali metal elements in the biomass thermal transformation process [18,19]. In addition, to the best of our knowledge, there have been two published papers on the study of biomass properties using LIBS, one of which discusses the usage of LIBS technology for the high heating value and ash fusion temperature of biomass combined with coal [20], and the other discusses the ash related analysis in biomass utilization [21]. Nevertheless, the feasibility of applying LIBS to the analysis of solid fuel properties has been proved, mostly through the studies on coal property analysis, and methods for simultaneous measurement of multiple properties parameters, including gross calorific value [22,23], carbon content [24,25], volatile matter content [23,26] and ash content [26,27] were also developed and optimized over the years. Compared to coal, solid biomass fuel feeding to a power plant generally has a significant higher oxygen and volatile content, and a lower carbon and ash content [28], and the ash composition are also significantly different to that of the coal ash [29]. Thus, it’s expected that the relative intensities of LIBS emission lines are significantly different between the biomass and coal samples, because of their variation in chemical composition and physical characteristics, as known as the “Matrix effect” [30]. In the present work, feasibility study on using laser-induced breakdown spectroscopy (LIBS) for simultaneous measurement of the gross calorific value, carbon content, volatile matter content and ash content was carried out for 66 wood pellet samples. PLS was used to establish quantitative analysis models of these property indexes, and then the validation set was used to evaluate the performance of the model. Moreover, three prediction models were constructed with three normalization methods plus baseline correction respectively and the best one was chosen to obtain the optimal predictive performance of the models. Lastly, the repeatability of LIBS was evaluated by comparing its average standard deviation with the repeatability limit of national standards and general rule.
2.2. Biomass fuel samples 66 samples of solid biomass fuels used in this study were collected from various power plants, mostly in the form of 6–8 mm pellets. The proximate analysis, gross calorific value and carbon content were determined in accordance with the China national standards GB/T 307272014 [9], GB/T 28731-2012 [6] and general rule for elemental analyzer DL/T 568-2013 [7] respectively. The detailed values can be found in the supplementary materials. For simplicity, Table 1 lists the descriptive statistics of the gross calorific value, carbon content, volatile matter content and ash content for these 66 solid biomass sample. All the samples were ground into powder by a grinder (Tianjin Taisite, FW 135) to particle size of less than 125 μm. Then the pulverized samples were put into a loft drier, holding at 105 °C for 24 h air-drying to ensure complete drying [32]. Weighed 1.50 g dried sample was put into an automatic tableting press (Pike Technologies Crushir), being pressed for 2 min to make a pellet sample with a diameter of 24 mm. The optimal pressing pressure of 9 t was selected based on the orthogonal experiment method, and the optimization of pressing pressure can be found in the supplementary materials. 66 solid biomass fuel samples were randomly divided into two groups: calibration set and validation set. Fiftysix samples in the calibration set were used to establish PLS models with different property indexes. Ten samples in the validation set were used to evaluate the prediction performance of the established model. 2.3. Data processing Data processing is one of the necessary procedures for LIBS measurement, which connects the original spectra and measurement results [33]. Lignocellulose biomass belongs to inhomogeneous substance, and its spectrum is highly complicated and full of redundant and interfered information. Therefore, it is substantial to process the spectral data. The flow chart of data processing in the present study is shown in Fig. 1, which can be divided into two parts, i.e., data preprocessing and quantitative analysis, principles of which method are described as follows.
2. Material and methods 2.1. LIBS apparatus An integrated LIBS experimental set-up was used in the present study and details can be found elsewhere [31]. Briefly, it consists of a pulsed Nd:YAG Laser (Beamtech Optronics, E-lite200, China), a multichannel spectrometer (Avantes, AvaSpec-2048FT, Holland), a programmable pulse generator (DG535, Stanford Research Systems, America), and some optical components and so on. Samples were placed on a three-dimensional mobile platform and interacted with laser under atmospheric pressure. By orthogonal experimental parameters optimization, the laser energy was determined to be 55 mJ, and the delay time of the spectrometer was determined to be 1.0 μs. The optimization of experimental parameters can be found in the supplementary materials. Integration time was the default minimum of
2.3.1. Baseline correction Before the quantitative analysis, the baseline noise and signal of the spectra were modified to enhance the spectral signal and increase the stability of the spectral signal, so as to meet the requirement of quantitative analysis. The Baseline Estimation And Denoising using Sparsity (BEADS) algorithm [34] is generally used to process signals with 2
Fuel 258 (2019) 116150
Z. Lu, et al.
Fig. 3. The variation curve of RSD with the average number of spectral lines C Ⅰ 247.87 and Ca Ⅰ 396.85.
Fig. 3 shows the variation curve of RSD for C Ⅰ 247.87 and Ca Ⅰ 396.85 with the average number of spectral lines. From Fig. 3, it can be seen that with the increase of average number, the RSD of analytical spectral lines decreases gradually, which means the spectra is stabilized gradually. However, the further increase in the averaging number would have little effect on the improvement of signal repeatability. From the Fig. 3, we can see that the RSD of the analytical spectral lines increased when the average number is more than 100. Therefore, the average number of spectra selected in this study was 100.
Fig. 1. Flow chart of data processing.
2.3.3. Normalization processing In order to avoid large prediction errors caused by order of magnitude differences between spectrum, the normalization processing is required. For different samples, the results obtained by normalization processing methods are different, so it is necessary to compare and choose the normalization processing method with the best performance. This study selected three common normalization methods to process the original spectral data, and their improvements of the accuracy of LIBS analysis were compared. The three methods are: Internal standard method, Total intensity normalization and Z-score standardization. Their principles are described respectively as follows.
Fig. 2. The spectra of the third channel (487–620 nm) of C1 sample before and after the baseline correction.
2.3.3.1. Internal standard method. Internal standard method is commonly used to improve the accuracy of LIBS analysis [37]. The principle is as follows: Firstly, selecting an elemental characteristic spectral line as the internal standard spectral line. Then, dividing the intensity of original spectral data by the intensity of the internal standard spectral line. In this study, the C atomic line at 247.86 nm was selected as the characteristic spectral line. The spectral intensity after the Internal standard method can be calculated by Eq. (1) [37].
baseline interference, which can self-adaptively deduct the continuous background and noise, avoid the influence of parameter selection on the results. In this study, we used BEADS algorithm to make the baseline correction for the spectrum of third channel (487–620 nm) of the spectrometer because the spectrum of third channel is uplifted slightly. The results of spectra of the third channel of C1 sample before and after the baseline correction are shown in Fig. 2.
Iin =
2.3.2. Data average processing In many reported LIBS investigations, data average processing is mostly often used to reduce measurement errors [35]. In this study, the characteristic spectral lines C Ⅰ 247.87 and Ca Ⅰ 396.85 of the main elements C and Ca in solid biomass fuels were selected as the analytical spectral lines, and their relative standard deviation (RSD) were selected as the indexes of average number selection. The RSD of spectral line can be defined as [36]:
⎡∑ RSD = ⎣
(2)
where, Iin is the spectral intensity after the Internal standard method, Ij is the absolute intensity of the spectrum, and Ic is the spectral intensity of C Ⅰ 247.86. 2.3.3.2. Total intensity normalization. Total intensity normalization is another common spectral data processing method in LIBS. Pulse-topulse fluctuations in experimental conditions and matrix effects can be corrected after using total intensity normalization [38]. The spectral intensity after the total intensity normalization can be calculated by Eq. (2) [35]:
(xi − M )2 ⎤ n−1 ⎦
M
Ij IC
(1)
where n is the number of a set of measurements; xi is the absolute line intensity of element i in each measurement; and, M is the arithmetic mean line intensity of the repeated measurements.
Inorm = 3
Ij Iall
(3)
Fuel 258 (2019) 116150
Z. Lu, et al.
where, Inorm is the spectrum after the total intensity normalization, Ij is the absolute intensity of the spectrum, and Iall is the sum of all spectral intensities. 2.3.3.3. Z-score standardization. The data processed by Z-score standardization conforms to the standard normal distribution, that is, the mean value is 0 and the standard deviation is 1. The spectral intensity after the Z-score Standardization can be calculated by Eq. (3) [39]:
Xs =
Xi, j − μ σ
(4)
where Xs is the standardized spectral intensity, Xi,j is the absolute intensity of the original spectrum of the j variable of the i sample, μ is the average value of the spectrum of the j variable calibration set, and σ is the standard deviation of the spectrum of the j variable calibration set. The steps of data preprocessing using Z-score standardization are as follows:
Fig. 4. Spectra of sample C1.
3.1.2. Carbon content For the LIBS measurement of carbon content in solid biomass fuel, part of the carbon cannot generate the atomic carbon emission due to the formation of molecular carbon, such as C2 and CN [44]. Therefore, when quantitatively analyzing carbon content in solid biomass fuel, spectral intensity of C2 and CN molecular bands should be considered in the calibration model. In addition, the solid biomass fuel contains a lot of mineral elements, such as Ca, Si, Mg, Al, Fe, etc. As these metallic elements are easily ionized substances, when the C element is ionized, these matrix elements will also be ionized, which affect the intensity of the spectral lines [45], so the spectral lines of these metallic elements should be considered in the analysis of carbon content. Therefore, the spectral bands of C, C2, CN and metallic elements Ca, Si, Mg, Al, Fe, etc. were selected as the input variables of PLS quantitative analysis model for carbon content of solid biomass fuel.
(1) Firstly, the spectral mean and standard deviation of all calibrated samples of each variable are calculated. (2) Then the standardized spectral data of the calibration set are obtained by using Eq. (3). (3) The spectral data of the unknown samples and the spectral mean and standard deviation of the calibration set are substituted into the above Eq. (3) to calculate the standardized spectral data of the validation set. 2.4. Quantitative analysis model Partial least squares (PLS) method is used as a quantitative analysis model for gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel. The combination of LIBS and PLS has been proved to be an effective quantitative analysis method [27]. Detailed principles of PLS can be found in reference [40].
3.1.3. Volatile matter content The main components in volatile matter are: (1) combustible components: light hydrocarbons dominated by CH4, C2H2, CO, H2, etc.; (2) non-combustible components: namely CO2, H2O, N2, NH3, NOX (NO, NO2), N2O, etc. [46]. It can be seen that the volatile matter of biomass is mainly composed of C, H, O and N elements. However, considering the complex molecular structure of biomass and the existence of a large number of easily ionized elements, such as Ca, Al and Fe, these mineral elements are easy to combine with O to form oxides. Therefore, the analysis of volatile matter content can’t be expressed only by the concentration of organic elements. In addition to the content of organic elements, the effect of matrix effect should also be considered. The characteristic spectrum of mineral elements contains some matrix information. Therefore, we select the spectral bands of C, H, O, N and mineral elements Ca, Mg, Al, Fe, etc. as the input variables of PLS quantitative analysis model for volatile matter content of solid biomass fuel.
3. Results and discussion 3.1. Input variable selection 3.1.1. Gross calorific value Gross Calorific Value is also known as Higher Heating Value (HHV) [41]. The HHV of biomass on dry basis can be estimated from the contents of C, H, S, N, O and ash in biomass according to the Eq. (4) [42].
HHVDB = 0.3491ECC,DB + 1.1783ECH,DB + 0.1005ECS,DB − 0.0151 ECN,DB − 0.1034ECO,DB − 0.0211ADB
(5)
where HHVDB is the HHV on dry basis (MJ/kg), ECC,DB is C content on dry basis, ECH,DB is H content on dry basis, ECS,DB is S content on dry basis, ECN,DB is N content on dry basis, ECO,DB is O content on dry basis, ADB is ash content on dry basis. From Eq. (4), it can be seen that the gross calorific value of biomass is mainly related to C, H, S, N, O and ash, while ash content is mainly composed of oxides of metal elements [29]. Fig. 4 shows the spectra of sample C1 as an example, in which the atomic lines were identified using the NIST Atomic Spectra Database [43]. From Fig. 4, we can see that the spectral lines of C are mainly located in the first channel (178–354 nm), while the spectral lines of H and N are mainly located in the fourth channel (607–827 nm), and the spectral lines of metal elements are distributed in each channel. Therefore, the full spectrum was selected as the input variables of PLS quantitative analysis model for gross calorific value of solid biomass fuel.
3.1.4. Ash content Ash in biomass is mainly composed of oxides of metal elements Ca, Si, Mg, Al, Fe, Na, K, Ti, etc. [29]. The ash content of biomass should be linearly related to the content of these metal elements. As shown in Fig. 4 above, we can see that the main elements of biomass ash are distributed in each channel, so we chose the full spectrum as the input variable for the PLS model. The selection of model input variables for gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel are summarized in Table 2. 4
Fuel 258 (2019) 116150
Z. Lu, et al.
association. The closer R2 is to 1, the stronger the correlation. RMSEP is used to evaluate the prediction accuracy of calibration model, where lower values indicate higher prediction accuracy of the calibration model. In addition, ASD is used to evaluate the repeatability of the experiment, where lower values indicate better reproducibility of the experiment.
Table 2 Input variables for modeling different fuel property indexes of solid biomass. Fuel property indexes
Input variables
Gross calorific value Carbon content
Full spectrum (178–828 nm) Spectral bands (178–321 nm, 355–499 nm, 516–517 nm) Spectral bands (246–249 nm, 372–455 nm, 588–590 nm, 611–672 nm, 714–778 nm) Full spectrum (178–828 nm)
Volatile matter content Ash content
3.4. Quantitative analysis results The input variables of gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel were pretreated separately, and then the PLS components for them were chosen to be 12, 5, 6 and 4 to build models respectively. Finally, the samples in the validation set were used to evaluate the predictive performance of the models. Three prediction models were constructed with baseline correction plus three normalization methods (internal standard method, total intensity normalization and Z-score standardization), respectively. The quantitative analysis results with different normalization methods were shown in Table 3. For gross calorific value of solid biomass fuel, after Zscore standardization plus baseline correction, its R2 increased from 0.981 to 0.998, ASD decreased from 0.12% to 0.08%, and RMSEP slightly increased from 0.26% to 0.33%. However, the overall prediction accuracy of gross calorific value of solid biomass fuel after C internal standard and total intensity normalization method were both inferior to the original spectra. For carbon content of solid biomass fuel, after Z-score standardization plus baseline correction, its R2 increased from 0.720 to 0.988, RMSEP decreased from 0.69% to 0.65%, ASD decreased from 0.26% to 0.15%. However, the overall prediction accuracy of carbon content of solid biomass fuel after total intensity normalization method was slightly inferior to the original spectra. For volatile matter content, after Z-score standardization plus baseline correction, its R2 value increased from 0.832 to 0.963, RMSEP decreased from 1.38% to 1.11%, ASD decreased from 0.56% to 0.43%. However, the overall prediction accuracy of volatile matter content of solid biomass fuel after C internal standard and total intensity normalization method were both inferior to the original spectra. For ash content of solid biomass fuel, after Z-score standardization plus baseline correction, its R2 increased from 0.961 to 0.995, RMSEP decreased from 0.66% to 0.38%, ASD decreased from 0.19% to 0.16%. Although the prediction accuracy is also improved after the C internal standard, the overall improvement was not as good as the Z-score standardization. What’s more, the overall prediction accuracy of ash content of solid biomass fuel after total intensity normalization method was inferior to the original spectra. In summary, on the basis of baseline correction, compared with the other two normalization methods, i.e., Internal standard method and Total intensity normalization method, the Z-score
3.2. Calibration model building In this study, Leave-one-out Cross Validation [47] was used to test the robustness and fitting performance of the model. The appropriate number of PLS components for the calibration models was determined by considering the root mean square error of cross validation (RMSECV). The RMSECV was calculated by Eq. (5) [48]. n
∑i = 1 (yi − yi)2
RMSECV =
(6)
n
where n is the number of samples of calibration set, yi is the reference value of samples, and yi is the predicted value of samples. 3.3. Model evaluation In this study, model performance was evaluated in terms of the coefficient of determination (R2), root mean square error of prediction (RMSEP), and average standard deviation (ASD), calculated with Eqs. (6), (7) and (8) respectively [49–51]. n
R2 = 1 −
yi )2 ∑i (yi − −
n
∑i (yi − y )2
(7)
m
RMSEP =
ASD =
m ∑i
yi )2 ∑i (yi − (8)
m −
(yi − y )2 (9)
m
where yi, yi are the reference and predicted values of solid biomass fuel − property, y is the mean reference value of solid biomass fuel property, n, m denotes the number of the calibration set of solid biomass fuel samples and the number of the validation set of solid biomass fuel samples, respectively. Coefficient of determination (R2) is a measure of total variance between measured and predicted values that can be modeled by linear
Table 3 Quantitative analysis results of four property indexes of biomass fuel by different normalization methods. Evaluation index
Original spectra
Baseline correction Baseline correction only
+Z-score standardization
+C internal standardization
+Total intensity normalization
Gross caloric value
R2 RMSEP (MJ/kg) ASD (MJ/kg)
0.981 0.26 0.12
0.980 0.27 0.11
0.998 0.33 0.08
0.985 0.34 0.10
0.980 0.39 0.16
Carbon content
R2 RMSEP (%) ASD (%)
0.720 0.69 0.26
0.720 0.68 0.26
0.988 0.65 0.15
– – –
0.716 0.70 0.27
Volatile matter content
R2 RMSEP (%) ASD (%) R2 RMSEP (%) ASD (%)
0.832 1.38 0.56 0.961 0.66 0.19
0.831 1.37 0.57 0.952 0.58 0.22
0.963 1.11 0.43 0.995 0.38 0.16
0.877 1.47 0.62 0.976 0.42 0.16
0.830 1.50 0.48 0.924 0.48 0.18
Ash content
5
Fuel 258 (2019) 116150
Z. Lu, et al.
Fig. 5. Quantitative analysis results of four property indexes of solid biomass fuel based on PLS and baseline correction plus Z-score standardization: (a) gross caloric value (b) carbon content (c) volatile matter content (d) ash content. Table 4 Comparison of repeatability of LIBS application results with national standards and general rule.
Gross caloric value (MJ/kg) Carbon content (%) Volatile matter content (%) Ash content (%)
ASD from present study
Repeatability requirement in national standard
Note
0.08 0.15 0.43 0.16
0.12 0.51 0.60 0.15
GB/T30727-2014 DL/T 568-2013 GB/T28731-2012 GB/T28731-2012
repeated for 3 times. It can be seen from the Fig. 5 that the four quantitative analysis models have good predictive performance, which demonstrated the feasibility of applying LIBS to determine the gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel.
standardization method has the best improvement on the overall prediction accuracy of the PLS model. Therefore, the PLS model constructed with baseline correction plus Z-score standardization was selected in the study and the final results for gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel are shown in the Fig. 5. The error bar in Fig. 5 represents the standard deviation (SD) of a single sample
6
Fuel 258 (2019) 116150
Z. Lu, et al.
3.5. Comparison of LIBS application with the present national standards and general rule
Particulate Wood Fuels. ASTM Int; 1998. doi:10.1520/E0872-82R06.2. [11] Fagan CC, Everard CD, McDonnell K. Prediction of moisture, calorific value, ash and carbon content of two dedicated bioenergy crops using near-infrared spectroscopy. Bioresour Technol 2011;102:5200–6. https://doi.org/10.1016/j.biortech.2011.01. 087. [12] Labbé N, Lee SH, Cho HW, Jeong MK, André N. Enhanced discrimination and calibration of biomass NIR spectral data using non-linear kernel methods. Bioresour Technol 2008;99:8445–52. https://doi.org/10.1016/j.biortech.2008.02.052. [13] Allison GG, Morris C, Hodgson E, Jones J, Kubacki M, Barraclough T, et al. Measurement of key compositional parameters in two species of energy grass by Fourier transform infrared spectroscopy. Bioresour Technol 2009;100:6428–33. https://doi.org/10.1016/j.biortech.2009.07.015. [14] Torgrip RJO, Fernández-Cano V. Rapid X-ray based determination of moisture-, ash content and heating value of three biofuel assortments. Biomass Bioenergy 2017;98:161–71. https://doi.org/10.1016/j.biombioe.2017.01.005. [15] Yao S, Lu J, Li J, Chen K, Li J, Dong M. Multi-elemental analysis of fertilizer using laser-induced breakdown spectroscopy coupled with partial least squares regression. J Anal At Spectrom 2010. https://doi.org/10.1039/c0ja00027b. [16] Wang Z, Zhou Y, Whiddon R, He Y, Cen K, Li Z. Investigation of NO formation in premixed adiabatic laminar flames of H2/CO syngas and air by saturated laserinduced fluorescence and kinetic modeling. Combust Flame 2016;164:283–93. https://doi.org/10.1016/j.combustflame.2015.11.027. [17] He Y, Zhu J, Li B, Wang Z, Li Z, Aldén M, et al. In-situ measurement of sodium and potassium release during oxy-fuel combustion of lignite using laser-induced breakdown spectroscopy: effects of O2 and CO2 concentration. Energy Fuels 2013. https://doi.org/10.1021/ef301750h. [18] Fatehi H, Li ZS, Bai XS, Aldén M. Modeling of alkali metal release during biomass pyrolysis. Proc Combust Inst 2017;36:2243–51. https://doi.org/10.1016/j.proci. 2016.06.079. [19] Hsu LJ, Alwahabi ZT, Nathan GJ, Li Y, Li ZS, Aldén M. Sodium and potassium released from burning particles of brown coal and pine wood in a laminar premixed methane flame using quantitative laser-induced breakdown spectroscopy. Appl Spectrosc 2011;65:684–91. https://doi.org/10.1366/10-06108. [20] Zhu T. Using LIBS and advanced data processing to analyze biomass and coal feedstock for utility. Boiler Appl 2013. [21] Westover TL. Rapid analysis of ash composition using breakdown spectroscopy (LIBS) rapid analysis of ash composition using laser- induced breakdown spectroscopy (LIBS); 2013. [22] Lu Z, Mo J, Yao S, Zhao J, Lu J. Rapid determination of the gross calorific value of coal using laser-induced breakdown spectroscopy coupled with artificial neural networks and genetic algorithm. Energy Fuels 2017;31:3849–55. https://doi.org/ 10.1021/acs.energyfuels.7b00025. [23] Zhang L, Gong Y, Li Y, Wang X, Fan J, Dong L, et al. Development of a coal quality analyzer for application to power plants based on laser-induced breakdown spectroscopy. Spectrochim Acta – Part B At Spectrosc 2015;113:167–73. https://doi. org/10.1016/j.sab.2015.09.021. [24] Feng J, Wang Z, Li L, Li Z, Ni W. A nonlinearized multivariate dominant factorbased partial least squares (PLS) model for coal analysis by using laser-induced breakdown spectroscopy. Appl Spectrosc 2013;67:291–300. https://doi.org/10. 1366/11-06393. [25] Yao S, Zhao J, Xu J, Lu Z, Lu J. Optimizing the binder percentage to reduce matrix effects for the LIBS analysis of carbon in coal. J Anal At Spectrom 2017;32:766–72. https://doi.org/10.1039/c6ja00458j. [26] Yuan T, Wang Z, Lui SL, Fu Y, Li Z, Liu J, et al. Coal property analysis using laserinduced breakdown spectroscopy. J Anal At Spectrom 2013;28:1045–53. https:// doi.org/10.1039/c3ja50097g. [27] Yao S, Lu J, Dong M, Chen K, Li J, Li J. Extracting coal ash content from laserinduced breakdown spectroscopy (LIBS) spectra by multivariate analysis. Appl Spectrosc 2011;65:1197–201. https://doi.org/10.1366/10-06190. [28] Cuiping L, Chuangzhi W, Yanyongjie Haitao H. Chemical elemental characteristics of biomass fuels in China. Biomass Bioenergy 2004. https://doi.org/10.1016/j. biombioe.2004.01.002. [29] Vassilev SV, Vassileva CG, Song YC, Li WY, Feng J. Ash contents and ash-forming elements of biomass and their significance for solid biofuel combustion. Fuel 2017;208:377–409. https://doi.org/10.1016/j.fuel.2017.07.036. [30] Aguilera JA, Aragón C, Madurga V, Manrique J. Study of matrix effects in laser induced breakdown spectroscopy on metallic samples using plasma characterization by emission spectroscopy. Spectrochim Acta – Part B At Spectrosc 2009;64:993–8. https://doi.org/10.1016/j.sab.2009.07.007. [31] Bai K, Yao S, Lu J, Zhao J, Xu J, Lu Z. Correction of C-Fe line interference for the measurement of unburned carbon in fly ash by LIBS. J Anal At Spectrom 2016. https://doi.org/10.1039/c6ja00307a. [32] Bach QV, Tran KQ, Khalil RA, Skreiberg Ø, Seisenbaeva G. Comparative assessment of wet torrefaction. Energy Fuels 2013. https://doi.org/10.1021/ef401295w. [33] Sheta S, Afgan MS, Hou Z, Yao S-C, Zhang L, Li Z, et al. Coal analysis by laserinduced breakdown spectroscopy: a tutorial review. J Anal At Spectrom 2019. https://doi.org/10.1039/c9ja00016j. [34] Ning X, Selesnick IW, Duval L. Chromatogram baseline estimation and denoising using sparsity (BEADS). Chemom Intell Lab Syst 2014;139:156–67. https://doi.org/ 10.1016/j.chemolab.2014.09.014. [35] Body D, Chadwick BLU. Optimization of the spectral data processing in a LIBS. Spec Issue Spectrochim Acta Part B 2001:725–36. [36] Wang X, Zhang L, Fan J, Li Y, Gong Y, Dong L, et al. Parameters optimization of laser-induced breakdown spectroscopy experimental setup for the case with beam expander. Plasma Sci Technol 2015. https://doi.org/10.1088/1009-0630/17/ 11/04.
China national standards (GB/T 28731-2012 [6] and GB/T 307272014 [9]) and general rule (DL/T 568-2013 [7]) have established repeatability requirements for gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel respectively. Table 4 compares the repeatability limits regulated by the national standards and general rule with the average standard deviation (ASD) by LIBS measurement. As it can be seen from Table 4, the ASD measured by LIBS technology of gross calorific value, carbon content and volatile matter content of solid biomass fuel meet the repeatability limit required by the national standards and general rule, while the repeatability of ash content needs to be improved. Although the measurement of national standard methods can achieve a high accuracy and good repeatability, rapid measurement is more important for field applications. Therefore, the application of LIBS technology to solid biomass fuel has its practical value, which is conducive to the control of solid biomass fuel quality and the timely grasp of fuel information to optimize combustion control. 4. Conclusion This work demonstrated the feasibility of using the LIBS technique to rapidly determine the gross calorific value, carbon content, volatile matter content and ash content of solid biomass fuel. The best quantitative analysis results were obtained with the PLS model based on spectra that combined baseline correction with Z-score standardization, where the prediction accuracy and repeatability of models are acceptable, demonstrating the potential for industrial application in timely monitoring of solid biomass fuel quality. Acknowledgements This work was supported by the National Natural Science Fund of China (51876068 & 51676073), Nature Science Foundation of Guangdong Province (2018A030313800), Guangdong Province train high-level personnel special support program (2014TQ01N334) and Science and Technology Project of Guangdong Province (2015A02021500). Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.fuel.2019.116150. References [1] National Energy Administration. Introduction to the operation of renewable energy grid connection in 2018 2019. http://www.nea.gov.cn/2019-01/28/c_137780519. htm. [2] Xingang Z, Zhongfu T, Pingkuo L. Development goal of 30 GW for China’s biomass power generation: will it be achieved? Renew Sustain Energy Rev 2013;25:310–7. https://doi.org/10.1016/j.rser.2013.04.008. [3] Mao J. Co-firing biomass with coal for power generation. Distrib Energy 2017;2:47–54. [4] Williams CL, Westover TL, Emerson RM, Tumuluru JS, Li C. Sources of biomass feedstock variability and the potential impact on biofuels production. Bioenergy Res 2016. https://doi.org/10.1007/s12155-015-9694-y. [5] European Environmental Agency. Energy efficiency and energy consumption in the household sector 2011; October 2011. [6] Standardization Administration of the People’s Republic of China (SAC). GB/T 28731-2012, Proximate analysis of solid biofuels; 2012. [7] National Energy Administration. DL/T 568-2013, Test methods for instrumental determination of carbon, hydrogen and nitrogen in laboratory samples of fuel; 2013. [8] ASTM. Standard test method for ash in biomass E1755 – 01. 2015. doi:10.1520/ E1755-01R07.2. [9] Standardization Administration of the People’s Republic of China (SAC). GB/T 30727-2014, Determination of calorific value for solid biofuels; 2014. [10] E872 – 82 ASTM. Standard Test Method for Volatile Matter in the Analysis of
7
Fuel 258 (2019) 116150
Z. Lu, et al.
[45] Yao S, Zhang L, Xu J, Yu Z, Lu Z. Data processing method for the measurement of unburned carbon in fly ash by PF-SIBS. Energy Fuels 2017;31:12093–9. https://doi. org/10.1021/acs.energyfuels.7b02692. [46] Vassilev SV, Vassileva CG, Vassilev VS. Advantages and disadvantages of composition and properties of biomass in comparison with coal: an overview. Fuel 2015;158:330–50. https://doi.org/10.1016/j.fuel.2015.05.050. [47] Kearns M, Ron D. Algorithmic stability and sanity-check bounds for leave-one-out cross-validation. Neural Comput 1999. https://doi.org/10.1162/ 089976699300016304. [48] Moreira SA, Sarraguça J, Saraiva DF, Carvalho R, Lopes JA. Optimization of NIR spectroscopy based PLSR models for critical properties of vegetable oils used in biodiesel production. Fuel 2015. https://doi.org/10.1016/j.fuel.2015.02.082. [49] Canha N, Felizardo P, Menezes JC, Joana Neiva Correia M. Multivariate near infrared spectroscopy models for predicting the oxidative stability of biodiesel: effect of antioxidants addition. Fuel 2012. https://doi.org/10.1016/j.fuel.2012.02.017. [50] Balabin RM, Lomakina EI, Safieva RZ. Neural network (ANN) approach to biodiesel analysis: analysis of biodiesel density, kinematic viscosity, methanol and water contents using near infrared (NIR) spectroscopy. Fuel 2011. https://doi.org/10. 1016/j.fuel.2010.11.038. [51] Yao S, Mo J, Zhao J, Li Y, Zhang X, Lu W, et al. Development of a rapid coal analyzer using laser-induced breakdown spectroscopy (LIBS). Appl Spectrosc 2018;72:1225–33. https://doi.org/10.1177/0003702818772856.
[37] Yao S, Xu J, Bai K, Lu J. Improved measurement performance of inorganic elements in coal by laser-induced breakdown spectroscopy coupled with internal standardization. Plasma Sci Technol 2015;17:938–43. https://doi.org/10.1088/1009-0630/ 17/11/09. [38] Cremers D, Radziemski L. Handb Laser-Induced 2013. https://doi.org/10.1002/ 9781118567371. [39] Wang H, Zhang J. Analysis of different data standardization forms for fuzzy clustering evaluation results’ influence. 3rd Int Conf Bioinforma Biomed Eng ICBBE 2009 2009. p. 1–4. https://doi.org/10.1109/ICBBE.2009.5162346. [40] Brereton RG. Introduction to multivariate calibration in analytical chemistry. Analyst 2000. https://doi.org/10.1039/b003805i. [41] Telmo C, Lousada J, Moreira N. Proximate analysis, backwards stepwise regression between gross calorific value, ultimate and chemical analysis of wood. Bioresour Technol 2010. https://doi.org/10.1016/j.biortech.2010.01.021. [42] Cai J, He Y, Yu X, Banks SW, Yang Y, Zhang X, et al. Review of physicochemical properties and analytical characterization of lignocellulosic biomass. Renew Sustain Energy Rev 2017;76:309–22. https://doi.org/10.1016/j.rser.2017.03.072. [43] Kramida A, Ralchenko Y, Reader J, Team NA. NIST Atomic Spectra Database (version 5.5.6). NIST At Spectra Database (Version 54), [Available Online Http// PhysicsNistGov/Asd] Natl Inst Stand Technol Gaithersburg, MD; 2016. [44] Dong M, Mao X, Gonzalez JJ, Lu J, Russo RE. Time-resolved LIBS of atomic and molecular carbon from coal in air, argon and helium. J Anal At Spectrom 2012;27:2066–75. https://doi.org/10.1039/c2ja30222e.
8