Evaluation of Raman Parameters Using Visible Raman Microscopy for

Feb 11, 2013 - graphite planes in sp2 carbon materials.14−17 The D peak has been proven to ... related to sp3 carbon or impurities,22−24 amorphous...
0 downloads 0 Views 2MB Size
Article pubs.acs.org/EF

Evaluation of Raman Parameters Using Visible Raman Microscopy for Soot Oxidative Reactivity Hee Je Seong† and André L. Boehman*,‡ EMS Energy Institute, The Pennsylvania State University, 405 Academic Activities Building, University Park, Pennsylvania 16802, United States S Supporting Information *

ABSTRACT: Soot crystalline structure was evaluated for four soot samples and one carbon black using Raman parameters obtained from first-order Raman spectra. For this work, different numbers of peaks from three to as many as five curves with combined Lorentzian and Gaussian bands were fitted for the spectra, and Raman parameters were compared with soot oxidative reactivity, in order to investigate the correlation between the Raman parameters and the oxidative reactivity for each curve-fitting method. Among these methods, the combination of three Lorentzian-shaped bands at about 1200 (D4), 1360 (D1), and 1580 cm−1 (G), and one Gaussian-shaped band at about 1500 cm−1 (D3), which is designated as 3L1G, shows the most consistent results for various Raman parameters with respect to soot oxidative reactivity; amorphous carbon fraction, crystallite size, and distribution of crystallite size are in good agreement with soot oxidative reactivity. Also, crystallite sizes from several empirical formulas using Raman spectroscopy were evaluated for the soot samples investigated, with respect to XRD results. In conclusion, a modified Knight and White’s formula using the 3L1G fitting method, La = 4.4(AD1/AG)−1, is proposed to evaluate crystallite sizes of soot samples, which are more comparable to fringe lengths obtained by HR-TEM image analysis than to crystallite sizes by XRD. characteristic peaks appearing at ∼1360 cm−1 (D peak) and ∼1590 cm−1 (G peak) for first-order Raman spectra. The G peak is a stretching mode of E2g symmetry at sp2 sites,13,14 but the origin of the D peak has been debated.15,16 It is traditionally known as a breathing mode of A1g symmetry at the edges of graphite planes in sp2 carbon materials.14−17 The D peak has been proven to be due to the breakdown in the k = 0 selection rule of large crystals from the existence of edge sites and the relative position of the laser spot with respect to the edge.18−21 Some carbonaceous materials, such as polycrystalline graphite, show only sharp D and G peaks,22,23 but many disordered and amorphous carbons also indicate additional peaks appearing at ∼1180 (D4), ∼1500 (D3), and ∼1620 cm−1 (D2), which are related to sp3 carbon or impurities,22−24 amorphous carbon,25,26 and disordered carbon,22,27,28 respectively. Thereafter, many analytical studies have proposed a relationship between these peaks and particular carbon structural information.22−24,29−31 Since edge sites are closely related to active sites on carbon surfaces, where oxygen is adsorbed in the oxidation process,32,33 our previous result clearly showed that the active surface area measured by the amount of chemisorbed oxygen in soot is in good agreement with soot oxidative reactivity.34 Likewise, Raman spectroscopy can provide valuable information about edge sites with respect to carbon oxidation if the Raman parameter is carefully selected. In many examples, height and area ratios of the D to the G band, and widths and positions in the D and G bands have been used to describe the carbon structural information.10,12,14−17,21−24,29−31

1. INTRODUCTION Stricter regulation of diesel particulate emissions has required the application of diesel particulate filters (DPF) and motivated research on soot nanostructure and reactivity. According to recent studies, the soot crystalline structure, O and H content, and the graphene layer length in the primary particles can affect oxidation rates.1 Since burnoff of the particulate matter, particularly diesel soot, is a critical process in DPF systems, the effect of the physical and chemical properties of diesel soot on reactivity has been of interest.2 There have been many efforts made to evaluate the crystalline structure of carbons. In many cases, Raman spectroscopy, X-ray diffraction (XRD), and transmission electron microscopy (TEM) provide useful information about carbon structure. XRD is popularly used in the characterization of the crystalline structure of carbonaceous materials. In addition to their diffraction patterns, interlayer spacing (d002) and crystallite size, which is represented by crystallite height (Lc) and width (La), are used as indices of crystallite shape and disorder.3−5 TEM, which also provides structural information about d002, Lc, and La of various materials, enables a statistical analysis of the distributions of carbon layer length and its tortuosity through image analysis.1,6−10 In addition, since the near-edge structure of the carbon-K-ionization edge can be deduced from electron energy loss spectroscopy (EELS), the transition to unoccupied antibonding (π*)/the transition to antibonding (σ*) has been investigated to measure the ratio of sp2 to sp3.1,10−12 In particular, Raman spectroscopy has been employed widely to study carbonaceous materials, because it is nondestructive and sensitive to intrinsic defects of the graphitic lattice, as well as, chemical doping. Many carbonaceous materials show two © 2013 American Chemical Society

Received: September 15, 2012 Revised: February 7, 2013 Published: February 11, 2013 1613

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

Figure 1. Curve-fitted shapes of SCE soot: (a) 2 Lorentzian and 1 Gaussian, (b) 3 Lorentzian, (c) 3 Lorentzian and 1 Gaussian, (d) 4 Lorentzian, (e) 4 Lorentzian and 1 Gaussian, (f) 5 Lorentzian.

results suggest the best fitting method in analyzing soot crystalline structure with respect to soot oxidative reactivity. In addition, Knight and White’s study has been popularly applied in evaluating carbon crystallite sizes using Raman spectroscopy, comparable to those measured by XRD.35 Their empirical formula was based on many different carbon materials with crystallite sizes between 2.5 and 250 nm. This study examines the validity of Raman spectroscopy in evaluating crystallite size of soot samples, in comparison with XRD results.

To interpret the Raman spectra of carbonaceous materials, the spectral information is typically curve-fitted for quantitative spectral features. Different curve-fitting methods have been examined for the characterization of Raman spectra using Lorentzian, Gaussian, Voigt, and Breit−Wigner−Fano (BWF) functions with no theoretical justification. In particular, Sadezky et al. thoroughly summarized different fitting methods comprising Lorentzian and Gaussian bands, which are described in Figure 1.23 Despite a good attempt to compare these methods for various materials, their evaluation on the structural information of several soot materials is based only on 4 Lorentzian and 1 Gaussian fitting method (4L1G). When they performed multipeak fitting with 4L1G, they found that Raman parameters determined from curve-fitted spectra were quite variable at different positions on one sample. They attributed the variability in fitted results to the heterogeneity of soot particles, but did not discuss potential uncertainties of the fitting method they used. Accordingly, this work revisits Sadezky et al.’s curve-fitting approaches for soot samples and examines reproducibility and uncertainties of various multipeak fitting methods by conducting measurements at 10−14 different positions for each sample. Furthermore, based on several Raman parameters from different fitting methods, the

2. EXPERIMENTAL SECTION 2.1. Samples. Four different soot samples and a carbon black with no metallic species present were used in this study in order to exclude the catalytic effect of metallic species. One flame soot was obtained from a laminar diffusion flame burner with n-heptane, as described elsewhere.34 Two diesel soot samples (referred to as DDC 30 and DDC 75) were collected from a four cylinder 2.5L DDC/VM Motori turbo-charged common rail diesel engine at 30% and 75% of full load and 1800 rpm, respectively, for which additional details are found elsewhere.36 Another soot sample (referred to as SCE soot) was obtained from a single cylinder engine (SCE) in the General Electric Global Research Center, for which additional experimental details are available elsewhere,37 at a notch-8 condition (full load) with a 10% exhaust gas recirculation (EGR) ratio. The carbon black investigated in this study is obtained from Alfa Aesar (Lot no.: L10M08). 1614

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

2.2. TGA Instrument. An SDT Q600 thermogravimetric analyzer from TA Instruments was employed to evaluate soot oxidative reactivity. Each diesel PM sample was placed in a 90 μL alumina sample cup and was heated to 500 °C and kept for 60 min to drive off volatile compounds under N2 gas at 100 cc/min. After thermal treatment, soot oxidative reactivity was evaluated at 500, 550, and 600 °C with ultrazero air (99.0% purity) at 100 cc/min for isothermal experiments. The mass loss of each sample during isothermal oxidation was normalized with respect to the weight after thermal treatment, and the normalized mass loss was compared for each sample to evaluate soot oxidative reactivity. 2.3. Raman Spectroscopy. A WITec Confocal Raman Microscope CRM 200 was focused on each sample with the white light source using a 100× objective lens, and the light was switched to the laser beam with an Ar-ion laser source (514.5 nm). To avoid degradation of samples, the incident laser power on samples was performed at ∼0.1 mW, which produced highly reproducible spectra with low noise levels, as seen in Figure s1 of the Supporting Information. As Sadezky et al. observed,23 curve-fitted shapes were significantly dependent on the position of each sample. Accordingly, in order to evaluate Raman parameters statistically, Raman spectra were recorded at 10−14 different positions for each sample, where the integration time is 10 s and hardware accumulation is 10 times to reduce noise in the spectra. Three-, four-, and five-peak curve-fitting methods were employed for first-order Raman spectra obtained at each position via IGOR Pro 6.10 software (Wavemetrics Inc.), as summarized by Sadezky et al.,23 where 2 Lorentzian (L) (D1:1360 cm−1 and G:1590 cm−1) and 1 Gaussian (G) (D3:1500 cm−1), 3L (D1, D3, and G), 3L (D1, D4:1180 cm−1, and G) and 1G (D3), 4G (D1, D3, D4, and G), 4L (D1, D2:1620 cm−1, D4, and G) and 1G (D3), and 5L (D1, D2, D3, D4, and G) were fitted as indicated in Figure 1. Four Raman parameters were then investigated to interpret soot crystalline structure; the crystallite dimension from the height ratio of D1 to G (ID1/IG) and the area ratio of D1 to G (AD1/AG), distribution of crystallite sizes from D1 full width at half-maximum (D1 fwhm), and bond length and angle distortion in sp2 from G full width at half-maximum (G fwhm) were investigated for each Raman parameter. In addition, the D3 band was further studied as a way to evaluate amorphous carbon of soot samples. Each Raman parameter was averaged for all the examined positions with standard deviations in order to compare the results statistically. 2.4. X-ray Diffraction. XRD patterns of soot samples were collected using a PANalytical X’Pert Pro MPD θ/θ goniometer with Cu Kα radiation, fixed slit incidence (0.5° divergence; 1.0° antiscatter; specimen length, 10 mm) and diffracted (0.5° antiscatter, 0.02 mm nickel filter) optics. Resulting patterns were corrected for both 2θ position and instrumental peak broadening using NIST 640c silicon and analyzed with Jade+9 software by MDI of Livermore, CA. Using Jade+9 software, the full width at half-maximum (fwhm) was manually postprocessed for (002) and (10) peaks with a linear background for five times, and they were averaged. The crystallite height (Lc) and crystallite width (La) were calculated from (002) and (10) peaks, respectively, from their fwhm’s using Scherrer’s equations.

Figure 2. TGA results of soot samples at 550 °C.

DDC 30 soot > SCE soot > DDC 75 soot > Carbon black. Because the results were similar at 500, 550 and 600 °C, 1/t50% at 550 °C was introduced in this study as a parameter representing soot oxidation rate. 3.2. XRD Results. Before the crystalline structures of the soot samples were investigated by Raman spectroscopy, the Xray diffraction patterns of these samples were examined, as shown in Figure 3. All the samples show broad peak patterns at 27, 42, and 77°, which arise from (002), (10), and (110), respectively. In comparison to soot samples, carbon black has sharper (002), (10), and (110) peaks. To find the relationship between soot oxidative reactivity and crystallite size, crystallite sizes of the samples were further investigated and plotted against 1/t50% in Figure 4. Overall, both Lc and La show similar trends against the soot oxidative rate: flame soot has the smallest crystallite size, and carbon black has the largest. However, the crystallite sizes of SCE soot, DDC 30 soot, and DDC 75 soot are not much different despite their reactivity differences. Although crystallite size has been investigated to indicate the degree of crystalline order of carbon materials with increasing temperature39,40 and with gasification,41 the examples of its application for soot in terms of reactivity differences are few.12 Theoretically, the reactivity difference should be reflected in the crystallite size difference, if the degree of crystalline order is measurable in this range of crystallite size. However, the crystallite size by XRD does not seem to be sensitive enough to indicate oxidative reactivity differences when the reactivity differences are not large as shown in this study. Therefore, care should be taken especially for highly disordered carbon, such as soot, in the evaluation of soot oxidative reactivity via XRD analysis. 3.3. Raman Analysis. 3.3.1. Raw Spectra. First- and second-order Raman spectra of the samples are compared to determine if any structural trends exist in the Raman spectra of the samples. The representative first-order Raman spectra for each sample are compared for the different samples in Figure 5a. As observed for many disordered carbonaceous materials, all the samples show two broad peaks at ∼1360 and ∼1590 cm−1. Although there is no appreciable variation in the positions of the D and G bands among the samples, there are two trends observed in the patterns with respect to oxidative reactivity. With decreasing oxidative reactivity, as observed in section 3.1, one trend is the decrease in the height of the valley between the D and G bands, and the other is the decreasing width in the D band. However, there is no apparent difference observed in the heights of D and G bands. Since the valley and the D bandwidth account for the D3 band and D1 width, respectively, in this assignment, their variation may have a significant effect

3. RESULTS AND DISCUSSION 3.1. TGA Analysis. To compare soot oxidative reactivity with the crystalline structure of soot, soot oxidation was carried out in the TGA. The oxidative trends for the five samples were compared at 550 °C, as shown in Figure 2. Flame soot is the most reactive among the samples, and carbon black is the least reactive. SCE soot and DDC 30 soot were completely oxidized at almost identical times. However, DDC 30 soot is shown to oxidize faster than SCE soot during most of the oxidation process, although the difference is small. The oxidative trend in this study is well represented by the 50% burnoff time, as indicated by t50% in Table 1, which was also used by Dong et al.;38 greater 1/t50% indicates more reactive soot. Accordingly, the soot sample is more reactive in the order of flame soot > 1615

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

Table 1. Time to be Half-Oxidized at Different Temperatures 1/t50% (/min) operating temp (°C) 500 550 600 650 a

flame soot −3

7.31 × 10 3.72 × 10−2 1.02 × 10−1

SCE soot

DDC 30 soot

−3

−3

5.11 × 10 2.11 × 10−2 6.61 × 10−2

5.94 × 10 2.30 × 10−2 7.11 × 10−2

DDC 75 soot −3

4.49 × 10 1.92 × 10−2 5.94 × 10−2

carbon black 0a 1.93 × 10−3 8.10 × 10−3 2.80 × 10−2

1/∞ because there was no reaction at this temperature.

Figure 3. XRD patterns of soot samples.

Figure 5. Comparison of representative Raman spectra of soot samples: (a) first-order Raman spectra, (b) second-order Raman spectra. Figure 4. Crystallite sizes of soot samples against 1/t50%.

in order to evaluate how closely each fitting method fits the original spectra of the samples. As examined by Sadezky et al.,23 the 4L1G fitting method shows the least χ2 error among the given fitting methods for all the samples, as shown in Table 2. Although the 5L fitting method is the second best fitting method in terms of the χ2 error, according to Sadezky et al., the present work shows that 3L1G fits better than 5L. Overall, the χ2 error is smaller for the same peak number, when the D3 band is a Gaussian pattern. Accordingly, the D3 band seems to be more appropriate with a Gaussian pattern in terms of quality of fit to the original spectra. Since the D3 band has been studied to evaluate the amorphous portion of the carbon,23,30,31,43 it is given a significant emphasis here using the different peak-fitting methods, before continuing the discussion of other Raman parameters. As evident in Table 3, none of the D3 parameters from 2L1G, 3L, and 5L curve-fitting methods correlate well with the oxidative reactivity. In the case of 4L, the D3 height is in good agreement with the reactivity, but the differences in the D3 area are not statistically significant. In addition, there is no trend observed for D3 fwhm from any fitting method. In contrast, the D3 height and area from 3L1G and 4L1G indicate clear trends consistent with those in the reactivity. Despite the physical characteristic of the D3 band, Cuesta et al. found out that the contribution of ID3/ID1+G is poorly related with the

on Raman parameters depending upon the curve-fitting method applied. Indeed, the D3 band does not appear for ordered crystallites like graphite.22,23 Liu et al. indicated that the peaks at around 1060 and 1750 cm−1 appeared after many hours of exposure to O3, which corresponds to C−O and C O, respectively.42 As shown in Figure 5a, the examined samples do not have additional peaks at those Raman shifts, possibly because they are fresh samples and/or the contribution of these surface oxygen functional groups is not significant for the examined samples. In addition, the second-order Raman spectra of the samples are also compared in Figure 5b. The four soot samples show similarly broad patterns in the range of 2000− 4000 cm−1, but carbon black shows four distinct peaks appearing at 2450, 2700, 2900, and 3200 cm−1, which are assigned as 2*D4, 2*D1 overtones, G + D1 combination, and 2*D2 overtone, respectively.23 The peaks at 2700, 2900, and 3200 cm−1 are shown to become more distinct with increasing crystalline order,22 so the broad peaks for the four soot samples indicate that the soot samples are less ordered in their structure than carbon black, which is consistent with the first-order Raman spectra and XRD patterns. 3.3.2. Raman Parameters Using Curve-Fitted Peaks. The χ2 errors were calculated for the fitting methods of each sample 1616

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

Table 2. Comparison of χ2 Error at Different Curve-Fitting Methods curve-fitting method sample

2L1G

flame SCE DDC 30 DDC 75 carbon black

0.092 0.147 0.099 0.135 0.142

± ± ± ± ±

3L

0.005 0.007 0.007 0.005 0.024

0.123 0.182 0.140 0.171 0.136

± ± ± ± ±

3L1G 0.005 0.009 0.007 0.006 0.037

0.043 0.049 0.043 0.044 0.050

± ± ± ± ±

4L

0.005 0.006 0.005 0.004 0.005

0.064 0.074 0.072 0.071 0.059

± ± ± ± ±

4L1G 0.006 0.009 0.009 0.006 0.009

0.035 0.034 0.029 0.026 0.022

± ± ± ± ±

0.005 0.004 0.003 0.002 0.003

5L 0.050 0.051 0.044 0.040 0.030

± ± ± ± ±

0.005 0.004 0.005 0.004 0.005

Table 3. Height, Area, and fwhm of D3 Band at Different Curve-Fitting Methods function 2L1G

3L

3L1G

4L

4L1G

5L

variable D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3 D3

height area fwhm height area fwhm height area fwhm height area fwhm height area fwhm height area fwhm

flame soot 0.379 54.14 134.1 0.224 34.99 99.3 0.298 49.77 156.9 0.253 50.67 127.4 0.329 56.24 160.6 0.278 57.23 131.2

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.014 2.09 1.08 0.012 2.23 1.74 0.011 2.10 2.19 0.008 1.83 1.01 0.014 2.63 2.01 0.011 2.27 0.66

SCE soot 0.310 44.23 134.2 0.207 32.41 99.8 0.252 42.15 157.3 0.233 47.38 129.7 0.268 46.22 161.8 0.252 53.04 133.77

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.020 2.81 1.94 0.011 2.29 2.92 0.013 2.14 2.12 0.010 2.13 2.68 0.015 2.65 1.54 0.011 2.61 2.89

DDC 30 soot 0.398 53.92 127.4 0.273 42.77 100.2 0.288 47.05 153.5 0.245 49.37 128.4 0.286 49.72 163.2 0.227 45.52 127.4

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.029 2.64 3.84 0.053 7.38 2.39 0.017 1.55 5.35 0.016 2.30 2.85 0.011 1.63 3.18 0.010 2.14 2.58

DDC 75 soot 0.309 43.32 131.6 0.222 36.99 106.0 0.236 40.09 159.6 0.219 47.04 136.7 0.237 42.17 167.4 0.213 45.56 135.9

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.011 1.22 1.68 0.015 2.67 0.94 0.009 1.12 3.31 0.008 1.85 3.53 0.007 1.19 3.63 0.010 3.57 5.90

carbon black 0.084 14.32 159.8 0.079 16.93 134.7 0.106 24.36 213.3 0.097 22.02 141.3 0.109 22.32 190.3 0.111 28.05 158.7

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

0.029 4.78 3.59 0.024 6.54 14.03 0.025 6.66 17.67 0.026 8.04 12.38 0.026 6.57 9.58 0.026 8.25 9.35

Figure 6. Comparison of Raman parameters from 2L1G fitting: (a) height ratio of D1 to G (ID1/IG), (b) area ratio of D1 to G (AD1/AG), (c) D1 fwhm, (d) G fwhm.

degree of disorder for various carbonaceous materials,22 although it is not clear which curve-fitting method was applied in their study. Zaida et al. investigated the variation of ID3/IG with 4L fitting with increasing temperature for chars.29 Because they did not examine the relation of this parameter and char

reactivity, however, it is not possible to determine whether ID3/ IG from 4L fitting is also informative in correlating char reactivity. Although there are few reports about the relation of the D3 band to carbon oxidative reactivity, Ivleva and coworkers found that AD3/AG and AD3/AG+D2+D3 from 4L1G 1617

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

Figure 7. Comparison of Raman parameters from 3L fitting: (a) height ratio of D1 to G (ID1/IG), (b) area ratio of D1 to G (AD1/AG), (c) D1 fwhm, (d) G fwhm.

Figure 8. Comparison of Raman parameters from 3L1G fitting: (a) height ratio of D1 to G (ID1/IG), (b) area ratio of D1 to G (AD1/AG), (c) D1 fwhm, (d) G fwhm.

reactivity for 2L1G fitting, although diffusion flame soot and DDC 30 soot have similar values, despite a significant reactivity difference. The G fwhm is found to decrease with an increase in reactivity. However, since it is a typical observation that the G width decreases with increasing temperature,44 the decreasing G fwhm in this work may not imply any physical meaning regarding the relation between crystalline order and reactivity. Similar trends in the Raman parameters, except for G fwhm, are also observed for 3L fitting, with less consistent results with

provide good structural information for soot and other carbonaceous materials.30,31,43 Therefore, depending upon the fitting method, the D3 band, which accounts for the portion of amorphous carbon, seems to be a useful Raman parameter in the study of crystalline structure. The curve-fitting methods assuming three peaks in the deconvolution, which are 2L1G and 3L fitting methods, were investigated for several Raman parameters. As shown in Figure 6, ID1/IG, AD1/AG, and D1 fwhm increase with the increase in 1618

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

Figure 9. Comparison of Raman parameters from 4L fitting: (a) height ratio of D1 to G (ID1/IG), (b) area ratio of D1 to G (AD1/AG), (c) D1 fwhm, (d) G fwhm.

Figure 10. Comparison of Raman parameters from 4L1G fitting: (a) height ratio of D1 to G (ID1/IG), (b) area ratio of D1 to G (AD1/AG), (c) D1 fwhm, (d) G fwhm.

size than the XE Printex does. Although AD1/AG and D1 fwhm indicate different crystalline information, such as the distribution of crystalline size and defects of the graphitic lattice,14,37,44 the structural variations in these two parameters seem to correlate well for the soot samples used in this study with respect to the oxidative reactivity. Four Raman parameters were also investigated using 3L1G and 4L fitting methods, which use four peaks in the deconvolution of the raw Raman spectra. As evident in Figure 8, ID1/IG, AD1/AG, and D1 fwhm using 3L1G fitting show a

respect to the oxidative reactivity in Figure 7. With both 3L and 2L1G fitting methods, D1 fwhm appears to be quite reproducible at different positions of each sample, having the smallest standard deviation among the examined Raman parameters. Jawhari et al. also investigated AD1/AG and D1 width for carbon black and microcrystalline graphite using 2L1G fitting.45 They observed that microcrystalline graphite has a wider D peak than XE Printex, despite significantly reduced ID1/IG for the microcrystalline graphite, because the microcrystalline graphite has a wider distribution of crystallite 1619

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

Figure 11. Comparison of Raman parameters from 5L fitting: (a) height ratio of D1 to G (ID1/IG), (b) area ratio of D1 to G (AD1/AG), (c) D1 fwhm, (d) G fwhm.

same sample to heterogeneous properties of soot samples,23 it rather indicates the sensitivity of these fitting methods. In particular, it was noted during the fitting process that the ratio of G to D2 varied significantly. As a consequence, most Raman parameters, related to the G band, show large deviations, regardless of sample, in comparison to three and four curvefitting methods. Indeed, D1 fwhm for 4L1G fitting is shown to have small deviations, as observed in other fitting methods. Since the deconvolution approach with G and D2 bands at about 1580 and 1620 cm−1, respectively, has a minor effect on the D1 band at about 1360 cm−1, there is no appreciable change in D1 fwhm from 4L1G compared to those from the four curve-fitting methods. Poor trends in ID1/IG, AD1/AG, and G fwhm are also observed for 5L fitting in Figure 11. Despite the poor trend in AD1/AG in this study, Al-Qurashi and Boehman found that AD1/AG provides useful information in the structural difference between non-EGR and EGR soot samples.12 Since the present samples were obtained from various combustion sources, there are significant differences among the samples in their physical and chemical properties, such as crystalline structure, pore structure, volatile fraction, and oxygen functional groups. Although it is presumed that soot oxidative reactivity is closely related to the degree of structural disorder, ID1/IG and AD1/AG from 4L1G and 5L fitting methods may not fully indicate these structural changes for some materials. Indeed, Sadezky et al. also observed that the AD1/AG ratios from 4L1G for GfG soot (spark discharged soot) were not larger than those for other types of soot, although GfG soot is known to have relatively high oxidative reactivity.23 However, Sheng obtained a positive result for coal chars with the 4L1G method.46 Although it is hard to determine which material is suitable for 4L1G and 5L fitting methods, some materials seem to not be appropriate for Raman analyses using ID1/IG and AD1/AG from these fitting methods. The above results show that D1 fwhm correlates well with the oxidative reactivity trends using the four and five curve-

better correlation with the oxidative reactivity than those from the 2L1G and 3L fitting methods. The ID1/IG of SCE soot is similar to that of the DDC 30 soot, and the AD1/AG shows an increasing trend with respect to the reactivity, although the standard deviation of the DDC 30 soot is much higher than those of the other samples. In comparison, SCE soot and DDC 75 soot have similar D1 fwhm values. However, there is no trend observed in the G fwhm like those in Figures 6 and 7. The trends in AD1/AG and D1 fwhm from the 4L fitting method are also similar to those from 3L1G fitting, as shown in Figure 9b,c. However, there are poor correlations shown for ID1/IG and G fwhm, as evidenced in Figure 9a,d, respectively. Accordingly, only AD1/AG and D1 fwhm from 4L fitting seem to provide structural information, while ID1/IG as well as AD1/ AG and D1 fwhm from 3L1G fitting can be useful. Zaida et al.’s study implies that, when the effect of heat treatment was investigated using cellulose chars, there was a gradual increase in ID1/IG from 600 to 1900 °C and a decrease in D1 fwhm from 4L fitting with increasing temperature.29 They speculated that the small ID1/IG at low temperatures reflects the small crystallites that do not resonate with a visible Raman source. However, it is possibly because this parameter is not suitable for the study of crystalline structure in comparison to AD1/AG, as Dong et al. also observed a huge discrepancy between ID1/IG and AD1/AG.37 Although they did not investigate the variation of D1 fwhm with respect to char oxidation, this parameter seems to be more useful in the study of carbon crystalline structure. The results using five deconvoluted peaks of the raw Raman spectra, referred to here as five curve-fitting methods, are shown in Figures 10 and 11. ID1/IG, AD1/AG, and G fwhm from 4L1G fitting correlate poorly with the oxidative reactivity, whereas D1 fwhm shows a trend consistent with the reactivity. In particular, DDC 30 soot and DDC 75 soot have large standard deviations for ID1/IG and AD1/AG. Although Sadezky et al. attributed a large variation in Raman parameters for the 1620

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

fitting methods, and AD1/AG shows an excellent trend for the four peak curve-fitting method. Accordingly, 3L1G and 4L fitting methods from the four peak curve-fitting approach seem to provide good Raman parameters for the soot samples from this study. Although D1 fwhm using the five peak curve-fitting method is a good parameter, as Ivleva and co-workers have demonstrated,30,31,43 the same trend can be obtained from that for the four peak curve-fitting methods. Despite the usefulness of G fwhm in determining the crystalline order of carbon,47 no fitting method provides a satisfactory result for G fwhm as a means of measuring soot oxidative reactivity. Therefore, it is speculated that these disordered soots and carbon blacks may have similar bond lengths and angle distortions in the sp2 structure, despite their distinct reactivity differences. Since the trends in both intensity and area of the D3 band from 3L1G fitting are also shown to correlate well with oxidative reactivity in this investigation, 3L1G is found to be the best choice among the examined fitting methods. Some researchers have reported that the D2 band is related with disordered carbon. Indeed, there are several carbon materials in the literature showing a unique peak at about 1620 cm−1.48 However, as shown in Figure 5a, all the samples examined do not have any particular peak at that range. Despite the fact that the G band with the 3L1G method includes disordered carbon, this method provides a good estimation of crystalline structure because uncertainties on G and D2 bands with 4L1G and 5L methods are significant for the examined soot samples. Consequently, it is concluded that the relation between soot oxidative reactivity and crystalline structure can be interpreted more reliably and consistently when all of these parameters are analyzed. In this case, 3L1G fitting provides a reasonable explanation about the effect of crystalline structure on soot oxidative reactivity. The soot oxidative reactivity is closely related to the abundance of edge sites from ID1/IG and AD1/AG, and amorphous carbon from ID3 and AD3. Correspondingly, more reactive soot shows a wider distribution of crystallite sizes from D1 fwhm with increasing disorder. 3.3.3. Extension of Knight and White’s Formula. In Table 4, ID/IG from raw spectra is compared to ID1/IG and AD1/AG

However, Cançado et al. proved that eq 1 is only valid for 514.5 nm, because AD/AG is dependent upon the laser energy.49 Accordingly, they provided a universal equation considering all the laser energies, as given in eq 2, from the studies of XRD and scanning tunneling microscopy (STM).

La =

Figure 12. Comparison of crystallite width: Knight and White is obtained from eq 1 using ID/IG of raw spectra, and Cançado et al. is obtained from eq 2 using AD1/AG of 3L1G fitting method.

there is difficulty in applying the same method for soot samples because the D and G bands are not discrete. Therefore, AD1/AG from 3L1G was used for the calculation of eq 2 in this work. As evident in Figure 12, La is shown to be similar regardless of soot sample, when it is calculated using eq 1. Therefore, it is evident that the typical Knight and White formula using ID/IG is no longer useful in evaluating crystallite size of the soot sample. In comparison, the La values calculated from eq 2 provide similar trends in soot reactivity, although the values are significantly larger than those from XRD. Indeed, the materials analyzed in Cançado et al.’s study are diamond-like carbon (DLC) films with La from 20 to 490 nm, and the D and G bands are discrete with narrow D bands, unlike soot samples. Therefore, Cançado et al.’s empirical formula also does not seem to apply to disordered carbon like soot and carbon black with wide D bands. Instead of using ID/IG from raw Raman spectra, Knight and White’s formula was further investigated for the 3L1G fitting method. The calculated La values from eqs 3 and 4 are also compared to those from XRD, as shown in Figure 4.

Raman parameter sample flame soot SCE soot DDC 30 soot DDC 75 soot carbon black a

0.960 1.032 0.965 1.028 1.047

± ± ± ± ±

0.009 0.038 0.035 0.022 0.075

ID1/IGa (with 3L1G) 1.381 1.283 1.294 1.174 0.959

± ± ± ± ±

0.035 0.030 0.040 0.031 0.093

AD1/AGa (with 3L1G) 3.680 2.606 3.119 2.256 0.925

± ± ± ± ±

0.137 0.104 0.279 0.103 0.105

I indicates height intensity, and A indicates area intensity.

from 3L1G. As noted earlier, ID1/IG and AD1/AG from 3L1G show consistent trends with soot oxidative reactivity, but ID/IG from raw spectra is in poor agreement with the reactivity. According to Tuinstra and Koenig,17 ID/IG is inversely correlated to crystallite width (La), and Knight and White proposed the following equation covering from amorphous carbon to graphite.35 La = 4.4(ID/IG)−1

(2)

where El is laser energy (eV), AD/AG is area ratio of D to G, and El is 2.41 eV for 514.5 nm. In these equations, it should be noted that Knight and White used the intensity ratio of D to G, whereas Cançado et al. used the integrated ratio (area ratio) of D to G. Using eqs 1 and 2, crystallite width was calculated for soot samples, and it is compared to the values calculated by XRD, as shown in Figure 12. Although Cançado et al. obtained AD/AG from raw spectra,

Table 4. Height and Area Ratios of D to G for Raw Spectra and the 3L1G Fitting Method ID/IGa (with no fitting)

560 (AD /A G)−1 E l4

La = 4.4(ID1/IG)−1

(3)

La = 4.4(AD1/A G)−1

(4)

where ID1/IG and AD1/AG are obtained from 3L1G fitting method. Figure 13 indicates that the crystallite width is overestimated with ID1/IG for soot samples compared to the XRD result except for carbon black, and it is underestimated with AD1/AG

(1) 1621

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

(distribution of crystallite size), and width of the G band (bond length and angle distortion in sp2 crystallite). From the consistent results obtained for all the Raman parameters employed, the 3L1G fitting method appears to be the best fitting method, indicating that soot oxidative reactivity is closely related to crystalline disorder. However, this tendency is not clearly indicated with other fitting methods. In addition, when the 3L1G fitting method was extended to estimate crystallite width using Knight and White’s formula, the 3L1G fitting method provides more similar values to fringe lengths given by TEM image analysis. Finally, this fitting method is suggested to provide a useful analytical approach for evaluating soot crystalline structure.



Figure 13. Comparison of crystallite width: Knight and White1 is obtained from eq 3 using ID1/IG, and Knight and White2 is obtained from eq 4 using AD1/AG of the 3L1G fitting method.

ASSOCIATED CONTENT

S Supporting Information *

for all the samples examined. However, eqs 3 and 4 seem to provide more reasonable results than eq 1, when results from XRD are considered. However, it is still difficult to judge if La determined by Raman spectroscopy should be similar to that by XRD in the case of soot with small crystallites, because XRD is known to be more sensitive to larger crystallites.14 In comparison, TEM image analysis provides direct information about the distribution of crystallite width in carbon crystallites. According to Vander Wal et al.,50 the crystallite width of soot treated at 1350 °C, which is referred to as the fringe length, is distributed from 0.5 to 3.5 nm with the median value of about 1 nm. Yehliu et al. reported a similar result.51 Since fringe length from HR-TEM should be comparable to crystallite width, the present modified Knight and White’s formula gives more reasonable values than XRD does. In conclusion, while the original Knight and White’s formula is limited in its application for disordered carbon materials, the modified Knight and White’s formula using 3L1G fitting developed in this study may be usefully employed in the estimation of crystallite width in the case of soot samples. Ferrari and Robertson proposed that, when La is smaller than 2 nm, La2 is proportional to the intensity of D to G, which decreases with the increased amorphization.14 In this light, however, our present result that the intensity of D1 to G is closely related to soot oxidative reactivity cannot be explained. Therefore, it is understood that soot samples are different from carbon materials, including nanocrystalline graphite and amorphous carbon, although soot samples are thought to contain amorphous carbon. Nevertheless, Atribak et al. showed that Ferrari and Robertson’s proposed equation is applicable for their soot samples.52 Since they used a different laser source (1064 nm) from the laser source in this work (514.5 nm), it is presumed that Raman responses of crystallites for different laser sources provided opposite results, assuming that their analytical approach is comparable to that used in the present work.

Raman spectra and TGA results. This material is available free of charge via the Internet at http://pubs.acs.org.



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Telephone: +1-734-7646995. Present Addresses †

H.J.S.: 9700 South Cass Avenue, Argonne National Laboratory, Energy Systems Division, Argonne, IL 60439-4815, USA. ‡ A.L.B.: 1231 Beal Avenue, University of Michigan, Mechanical Engineering, 2007 WE Lay Auto Lab, Ann Arbor, MI 481092133, USA. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS The authors thank General Electric Global Research Center and General Electric Transportation, in particular, David Walker, Omowoleola Akinyemi, Roy Primus, Adam Klingbeil, David Watson, Raj Rajiyah, and David Komoroske for their support of this work. This work was a part of the “Clean and Efficient Diesel Locomotive” program sponsored by the U.S. Department of Energy under No. 08NT002788.



REFERENCES

(1) Müller, J. O.; Su, D. S.; Wild, U.; Schlögl, R. Bulk and surface structural investigations of diesel engine soot and carbon black. Phys. Chem. Chem. Phys. 2007, 9, 4018−4025. (2) van Setten, B. A. A. A.; Makkee, M.; Moulijn, J. A. Science and technology of catalytic diesel particulate filters. Catal. Rev. 2001, 43, 489−564. (3) Babu, V. S.; Seehra, M. S. Modeling of disorder and X-ray diffraction in coal-based graphitic carbons. Carbon 1996, 34, 1259− 1265. (4) Menndez, J. A.; Xia, B.; Phillips, J.; Radovic, L. R. On the modification and characterization of chemical surface properties of activated carbon: Microcalorimetric, electrochemical, and thermal desorption probes. Langmuir 1997, 13, 3414−3421. (5) French, B. L.; Wang, J. J.; Zhu, M. Y.; Holloway, B. C. Structural characterization of carbon nanosheets via X-ray scattering. J. Appl. Phys. 2005, 97, 114317. (6) Sharma, A.; Kyotani, T.; Tomita, A. Comparison of structural parameters of PF carbon from XRD and HRTEM techniques. Carbon 2000, 38, 1977−1984. (7) Sharma, A.; Kyotani, T.; Tomita, A. A new quantitative approach for microstructural analysis of coal char using HRTEM images. Fuel 1999, 78, 1203−1212.

4. CONCLUSIONS Crystalline structures of four soot samples and one carbon black sample showing different oxidative reactivities were investigated for first-order Raman spectra using visible Raman spectroscopy. For this study, broad first-order Raman spectra were curve-fitted between 800 and 2000 cm−1 with three, four, and five peaks, where the D3 band at ∼1500 cm−1 is fitted with the Gaussian or the Lorentzian band along with deconvoluted Lorentzian bands. For this work, there were comprehensive investigations of the D3 band (amorphous portion), intensity ratio of the D1 to G (crystallite size), width of the D1 band 1622

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

(8) Palotas, A. B.; Rainey, L. C.; Feldermann, C. J.; Sarofim, A. F.; Vander Sande, J. B. Soot morphology: An application of image analysis in high-resolution transmission electron microscopy. Microsc. Res. Tech. 1996, 33, 266−278. (9) Vander Wal, R. L.; Tomasek, A. J.; Pamphlet, M. I.; Taylor, C. D.; Thompson, W. K. Analysis of HRTEM images for carbon nanostructure quantification. J. Nano Res. 2004, 6, 555−568. (10) Song, J.; Alam, M.; Boehman, A. L.; Kim, U. Examination of the oxidation behavior of biodiesel soot. Combust. Flame 2006, 146, 589− 604. (11) McCulloch, D. G.; Hoffman, P. A. Structural investigation of xenon-ion-beam-irradiated glassy carbon. Phys. Rev. B 1994, 50, 5905− 5917. (12) Al-Qurashi, K.; Boehman, A. L. Impact of exhaust gas recirculation (EGR) on the oxidative reactivity of diesel engine soot. Combust. Flame 2008, 155, 675−695. (13) Nemanich, R. J.; Solin, S. A. First- and second-order Raman scattering from finite-size crystals of graphite. Phys. Rev. B 1979, 20, 392−401. (14) Ferrari, A. C.; Robertson, J. Interpretation of Raman spectra of disordered and amorphous carbon. Phys. Rev. B 2000, 61, 14095− 14107. (15) Ferrari, A. C. Raman spectroscopy of graphene and graphite: Disorder, electron−phonon coupling, doping and nonadiabatic effects. Solid State Commun. 2007, 143, 47−57. (16) Escribano, R.; Sloan, J. J.; Siddique, N.; Sze, N.; Dudev, T. Raman spectroscopy of carbon-containing particles. Vib. Spectrosc. 2001, 26, 179−186. (17) Tuinstra, F.; Koenig, J. L. Characterization of graphite fiber surfaces with Raman spectroscopy. J. Compos. Mater. 1970, 4, 492− 499. (18) Katagiri, G.; Ishida, H.; Ishitani, A. Raman spectra of graphite edge planes. Carbon 1988, 26, 565−571. (19) Bowling, R. J.; Packard, R. T.; McCreery, R. L. Activation of highly ordered pyrolytic graphite for heterogeneous electron transfer: Relationship between electrochemical performance and carbon microstructure. J. Am. Chem. Soc. 1989, 111, 1217−1223. (20) Wang, Y.; Alsmeyer, D. C.; McCreery, R. L. Raman spectroscopy of carbon materials: Structural basis of observed spectra. Chem. Mater. 1990, 2, 557−563. (21) Casiraghi, C.; Hartschuh, A.; Qian, H.; Piscanec, S.; Georgi, C.; Fasoli, A.; Novoselov, K. S.; Basko, D. M.; Ferrari, A. C. Raman spectroscopy of graphene edges. Nano Lett. 2009, 9, 1433−1441. (22) Cuesta, A.; Dhamelincourt, P.; Laureyns, J.; Martínez-alonso, A.; Tascón, J. M. D. Raman microprobe studies on carbon materials. Carbon 1994, 32, 1523−1532. (23) Sadezky, A.; Muckenhuber, H.; Grothe, H.; Niessner, R.; Pöschl, U. Raman microspectroscopy of soot and related carbonaceous materials: Spectral analysis and structural information. Carbon 2005, 43, 1731−1742. (24) Schwan, J.; Ulrich, S.; Batori, V.; Ehrhardt, H.; Silva, S. R. P. Raman spectroscopy on amorphous carbon films. J. Appl. Phys. 1996, 80, 440−447. (25) Robertson, J. Amorphous-carbon. Adv. Phys. 1986, 35, 317−374. (26) Nemanich, R. J.; Glass, J. T.; Lucovsky, G.; Shroder, R. E. Raman scattering characterization of carbon bonding in diamond and diamondlike thin films. J. Vac. Sci. Technol., A 1988, 6, 1783−1787. (27) Vidano, R.; Fischbach, D. B. New lines in the Raman spectra of carbons and graphite. J. Am. Ceram. Soc. 1978, 61, 13−17. (28) Lespade, P.; Al-Jishi, R.; Dresselhaus, M. S. Model for Raman scattering from incompletely graphitized carbons. Carbon 1982, 20, 427−431. (29) Zaida, A.; Bar-Ziv, E.; Radovic, L. R.; Lee, Y. J. Further development of Raman microprobe spectroscopy for characterization of char reactivity. Proc. Combust. Inst. 2007, 31, 1881−1887. (30) Ivleva, N. P.; McKeon, U.; Niessner, R.; Pöschl, U. Raman microspectroscopic analysis of size-resolved atmospheric aerosol particle samples collected with an ELPI: Soot, humic-like substances, and inorganic compounds. Aerosol Sci. Technol. 2007, 41, 655−671.

(31) Knauer, M. K.; Carrarar, M.; Rothe, D.; Niessner, R.; Ivleva, N. P. Changes in structure and reactivity of soot during oxidation and gasification by oxygen, studied by micro-Raman spectroscopy and temperature programmed oxidation. Aerosol Sci. Technol. 2009, 43, 1− 8. (32) Xu, Y. J.; Li, J. Q. The interaction of molecular oxygen with active sites of graphite: A theoretical study. Chem. Phys. Lett. 2004, 400, 406−412. (33) Radovic, L. R. Active sites in graphene and the mechanism of CO2 formation in carbon oxidation. J. Am. Chem. Soc. 2009, 131, 17166−17175. (34) Seong, H. J.; Boehman, A. L. Studies of soot oxidative reactivity using a diffusion flame burner. Combust. Flame 2012, 159, 1864−1875. (35) Knight, D. S.; White, W. B. Characterization of diamond films by Raman spectroscopy. J. Mater. Res. 1989, 4, 385−393. (36) Seong, H. J.; Boehman, A. L. Impact of intake oxygen enrichment on oxidative reactivity and properties of diesel soot. Energy Fuels 2011, 25, 602−616. (37) Topinka, J. A.; Hibshman, J. R.; Glenn, W. D. Correlation of single-cylinder to multi-cylinder performance for a medium speed diesel engine. Proceedings of 2006 Spring Technical Conference, ASME IC Engine Div., Aachen, Germany, May 8−10, 2006; ASME: New York, 2006; ICES2006-1331. (38) Dong, S.; Paterson, N.; Kazarian, S. G.; Dugwell, D. R.; Kandiyoti, R. Characterization of tuyere-level core-drill coke samples from blast furnace operation. Energy Fuels 2007, 21, 3446−3454. (39) Zickler, G. A.; Smarsly, B.; Gierlinger, N.; Peterlik, H.; Paris, O. A reconsideration of the relationship between the crystallite size La of carbons determined by X-ray diffraction and Raman spectroscopy. Carbon 2006, 44, 3239−3246. (40) Lu, L.; Sahajwalla, V.; Harris, D. Characteristics of chars prepared from various pulverized coals at different temperatures using drop-tube furnace. Energy Fuels 2000, 14, 869−876. (41) Tran, K. N.; Berkovich, A. J.; Tomsett, A.; Bhatia, S. K. Crystalline structure transformation of carbon anodes during gasification. Energy Fuels 2008, 22, 1902−1910. (42) Liu, Y.; Liu, C.; Ma, J.; Ma, Q.; He, H. Structural and hygroscopic changes of soot during heterogeneous reaction with O3. Phys. Chem. Chem. Phys. 2010, 12, 10896−10903. (43) Knauer, M.; Schuster, M. E.; Su, D.; Schlögl, R.; Niessner, R.; Ivleva, N. P. Soot structure and reactivity analysis by Raman microspectroscopy, temperature-programmed oxidation, and highresolution transmission electron microscopy. J. Phys. Chem. A 2009, 113, 13871−13880. (44) Gruber, T.; Zerda, T. W.; Gerspacher, M. Raman studies of heat-treated carbon blacks. Carbon 1994, 32, 1377−1382. (45) Jawhari, T.; Roid, A.; Casado, J. Raman spectroscopic characterization of some commercially available carbon black materials. Carbon 1995, 33, 1561−1565. (46) Sheng, C. Char structure characterised by Raman spectroscopy and its correlations with combustion reactivity. Fuel 2007, 86, 2316− 2324. (47) Ferrari, C.; Rodil, S. E.; Robertson, J. Interpretation of infrared and Raman spectra of amorphous carbon nitrides. Phys. Rev. B 2003, 67, 155306. (48) Dongil, A. B.; Bachiller-Baeza, B.; Guerrero-Ruiz, A.; RodríguezRamos, I.; Martínez-Alonso, A.; Tascón, J. M. D. Surface chemical modification induced on high surface area graphite and carbon nanofibers using different oxidation and functionalization treatments. J. Colloid Interface Sci. 2011, 355, 179−189. (49) Cançado, L. G.; Takai, K.; Enoki, T.; Endo, M.; Kim, Y. A.; Mizusaki, H.; Jorio, A.; Coelho, L. N.; Magalhães-Paniago, R.; Pimenta, M. A. General equation for the determination of the crystallite size La of nanographite by Raman spectroscopy. Appl. Phys. Lett. 2006, 88, 163106. (50) Vander Wal, R. L.; Tomasek, A. J.; Street, K.; Hull, D. R.; Thompson, W. K. Carbon nanostructure examined by lattice fringe analysis of high-resolution transmission electron microscopy images. Appl. Spectrosc. 2004, 58, 230−237. 1623

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624

Energy & Fuels

Article

(51) Yehliu, K.; Vander Wal, R. L.; Boehman, A. L. A comparison of soot nanostructure obtained using two high resolution transmission electron microscopy image analysis algorithms. Carbon 2011, 49, 4256−4268. (52) Atribak, I.; Bueno-López, A.; García-García, A. Uncatalysed and catalysed soot combustion under NOx + O2: Real diesel versus model soots. Combust. Flame 2010, 157, 2086−2094.

1624

dx.doi.org/10.1021/ef301520y | Energy Fuels 2013, 27, 1613−1624