Energy & Fuels 2009, 23, 4051–4058
4051
Material Properties Influencing the Oxidation and Ignition Reactivity of Activated Carbons: Thermal Analysis, HRTEM Study, and Statistical Modeling Thangavelu Jayabalan, Pascaline Pre,* and Vale´rie Hequet GEPEA UMR-CNRS 6144, Ecole des Mines de Nantes, BP-20722 4 Rue Alfred Kastler, Nantes-44307, France
Jean Noe¨l Rouzaud Ecole Normale Supe´rieure de Paris, Laboratoire de Ge´ologie, UMR 8538 CNRS-ENS, 24 Rue Lhomond, 75231 Paris Cedex 5, France
Pierre Le Cloirec Ecole Nationale Supe´rieure de Chimie de Rennes, CNRS, UMR 6226, AVenue du Ge´ne´ral Leclerc, UniVersite´ europe´enne de Bretagne CS 50837, 35708 Rennes Cedex 7, France ReceiVed February 13, 2009. ReVised Manuscript ReceiVed June 5, 2009
The aim of this work is to understand the influence of textural, chemical, and structural properties on the reactivity of activated carbons toward air. Multiscale organization of activated carbons was studied using highresolution transmission electron microscopy (HRTEM) and their quantitative structural data, like individual fringe length, interlayer spacing, and proportion of nonstacked layers, were extracted using a specific image analysis procedure. Intrinsic properties like the specific surface area SBET, pore volume, micropore width, and elementary composition were also measured. The reactivity of the carbon samples in air was quantified from the measurement of oxidation and ignition temperatures determined by thermogravimetry analysis coupled with a differential scanning calorimetry. The characteristics of the graphitic structures and the properties of the activated carbon materials were found to be dependent on the activation mode and the nature of the material. The results suggest that oxygen present in the form of surface oxygenated groups, the mineral content, and the dimensions of the basic structural units influence the reactivity of activated carbons. Highly stable carbons were found to contain less oxygen to carbon ratio, lower mineral content, and larger and better stacked polyaromatic layers.
1. Introduction Activated carbons are widely used in heterogeneous catalysis, personal protection equipment, vehicle filters, and for the removal of volatile organic compounds (VOC’s) and odors from industrial effluents. Stricter environmental policies and commitments have ensured lower emission standards and greater efficiency in the removal of air pollutants. The major outcome is the extensive use of pollution abatement techniques involving activated carbon materials particularly in the removal of VOC’s. However, these materials pose a safety problem because of oxidation and ignition risks that can occur under ambient temperature, in presence of oxygenated compounds, and result from exothermicity of adsorption and chemical reactions. Extensive literature sources are available relating incidents of fires and thermal runaways encountered with activated carbon beds during adsorption or at idle state, after saturation with organic gases and also during the handling and regeneration of spent carbon adsorbents.1-3 Fires * To whom correspondence should be addressed. E-mail: pascaline.pre@ emn.fr. Telephone: +33 251858268. Fax: +33 251858299. (1) Naujokas, A. A. Spontaneous combustion of carbon beds. Plant/ Oper. Prog. 1985, 4, 120–126.
were also reported during transit on boardships of chemically activated carbons.4,5 In practice, ignition risks are not well mastered because of lack of understanding of the oxidation processes that actually take place and the effect of the surrounding conditions on the acceleration of these processes. To get a better insight into these phenomena, a laboratory study was undertaken to evaluate how the textural, chemical, and structural properties contribute to the reactivity of the activated carbons. In the recent years, thanks to the advancement of the characterization techniques, many studies were accomplished in an effort to link the structural properties of carbon materials with their reactivity.6 Experi(2) Delage, F.; Pre´, P.; Tezel, H.; Le Cloirec, P. Mass transfer and warming during adsorption of high concentrations of VOC on an activated carbon bed: experimental and theoretical analysis. EnViron. Sci. Technol. 2000, 34, 4816–4821. (3) Zerbonia, R. A.; Brockman, C. M.; Peterson, P. R.; Housely, D. Carbon bed fires and the use of carbon canisters for air emissions control on fixed roof tanks. J. Air Waste Manage. Assoc. 2001, 51, 1617–1627. (4) Bowes, P. C.; Cameron, A. Self heating and ignition of chemically activated carbon. J. Appl. Chem. Biotechnol. 1971, 21, 244–250. (5) Jones, J. C. Towards an alternative criterion for the shipping safety of activated carbons. J. Loss PreV. Process Ind. 1998, 11, 407–411. (6) Russell, N. V.; Gibbins, J. R.; Williamson, J. Structural ordering in high temperature coal chars and the effect on reactivity. Fuel 1999, 78, 803–807.
10.1021/ef9001296 CCC: $40.75 2009 American Chemical Society Published on Web 07/07/2009
4052
Energy & Fuels, Vol. 23, 2009
Jayabalan et al.
Table 1. General Characteristics of Activated and Nonactivated Carbon Samples sample name
manufacturer
raw material
mode of activation
activation conditions
NC-50 NC-60 NC-100 RB-2 BPL GF-40 BC-120 PICABIOL CTP-A CTP-PAN-3:1-A CTP-PAN-1:1-A PAN-A PAN-C NC-C
PICA PICA PICA NORIT CHEMVIRON NORIT PICA PICA LCSM Nancy LCSM Nancy LCSM Nancy LCSM Nancy LCSM Nancy PICA
coconut shell coconut shell coconut shell peat coal olive stone wood wood coal tar pitch coal tar pitch and polyacrylonitrile fiber coal tar pitch and poyacrylonitrile fiber polyacrylonitrile fiber poyacrylonitrile fiber coconut shell
physical physical physical physical physical chemical chemical chemical physical physical physical physical nonactivated nonactivated
steam at 900 °C steam at 900 °C steam at 950 °C steam at 900 °C steam at 900 °C under vacuum H3PO4 at 600 °C H3PO4 at 450 °C H3PO4 at 450 °C steam at 900 °C steam at 900 °C steam at 900 °C steam at 900 °C carbonization at 520 °C carbonization at 900 °C
mental techniques like the HRTEM (high-resolution transmission electron microscopy), coupled with adsorption experiments with molecules of various sizes and shapes have confirmed that the nanostructure of porous carbons, such as the activated carbons consists of short, distorted, and weakly stacked aromatic sheets with spaces between the layers corresponding to the micropores between them.7,8 The structural organization of the activated carbons also plays an important role in the reactivity of the material and therefore it is important to take into account the effect of structural properties. Several test methods can be found in the literature for characterizing the oxidation reactivity of the activated carbons. These methods enable to determine different reactivity parameters representative of the tendency of the carbonaceous materials to react in the presence of oxygen and self-ignite. Oxidation and self-ignition processes of activated carbons at elevated temperatures were extensively studied by Suzin et al.9 Two temperature domains were defined corresponding to low and fast reaction processes. The limit of the first region was denoted by the point of initial oxidation (PIO). This point corresponds to the minimum temperature above which the heat released by the oxidation reaction becomes significant. The surface properties are modified as the advancement of the oxidation proceeds. In the second region, the material ignites in a self-sustaining manner. The region is denoted by the spontaneous ignition temperature (SIT). The PIO and SIT data are dependent upon the conditions applied (heating rate, oxygen concentration in the gas phase, air flow rate). They enable comparison of the reactivities of different activated carbons submitted to identical surrounding conditions. The objective of the present work therefore is to study the relationships among the chemical, textural, and structural properties of activated carbons and the reactivity parameters like PIO and SIT. The analysis is completed with quantitative statistical modeling approach aiming at weighting the influence of the material properties on their reactivity. This approach contributes to the understanding of the physical and chemical mechanisms going with the oxidation process of activated carbons in contact with air. 2. Experimental Section 2.1. Materials. Seventeen different activated carbons (Table 1), representative of a variety of raw materials, modes of activation, and porosity characteristics, were selected for this study. All of them were produced from the carbonization of various synthetic or natural raw materials (polyacrylonitrile fiber, coconut shell, peat, (7) Lillo-Rodenas, M. A.; Amoros, D. C.; Solano, A. L.; Be´guin, F.; Clinard, C.; Rouzaud, J. N. HRTEM study of activated carbons prepared by alkali hydroxide activation of anthracite. Carbon 2004, 42, 1305–1310.
coal, wood, olive stone) at 800-1000 °C in the absence of oxygen, followed by physical or chemical activation. Physical activation was carried out using steam, while the chemical activation was performed using concentrated H3PO4 solutions. Some samples having the same precursor, such as coconut-based ones, were activated to different extents, giving rise to development of various pore size distributions in the samples. The materials used to produce high nitrogen content activated carbon were polyacrylonitrile (PAN), the oxidized form of poly(4vinylpyridine) cross-linked with 25 wt % of divinylbenzene (PVPox), and coal tar pitch (CTP). The carbonization of the polymers was carried out for 2 h, and steam activation of the resultant chars was performed at 800 °C. The preparation of the CTP and PAN samples are documented in detail in other works.10,11 Some carbonized samples were also chosen for the study. These samples were not activated but carbonized in an inert gas. The different samples with their precursor and activation conditions are given in Table 1. 2.2. Structural Characterization. The structural characterization of the activated carbons was carried out using a Jeol 2011 HRTEM (acceleration voltage 200 kV, resolution in the lattice fringe mode 0.14 nm). The HRTEM allows the study of the multiscale organization of the carbon materials7,8,12 and quantification of them thanks to a lab made image analysis. The samples were first ground in a small agate mortar and dispersed in ethanol, and then, a drop of suspension was deposited on a classical TEM copper grid, previously covered by a holey amorphous carbon film. A specially developed HRTEM analysis software was used to obtain quantitative structural information.12 The raw HRTEM images were recorded on classical negatives, and a part of this image (16 nm × 16 nm) is sampled and digitalized (256 gray levels, resolution 4000 pixels per inch). With the help of an improved version of the visolog (Noesis) commercial software, an homogeneous image without background noise is first obtained, and then binarised and skeletonized; each fringe is now one pixel-large (the size of one pixel for this work: 0.017 nm). The skeletonized pixel based image is then transformed into a vectorial image, and each fringe is analyzed individually in relation to its neighbors, thanks to lab-made software. Careful analysis is done taking into account different criterions to avoid artifacts (for instance fringes smaller than 0.25 nm, that is, the size of a single aromatic ring, were considered without physical (8) Duber, S.; Rouzaud, J. N.; Clinard, C.; Pusz, S. Microporosity and optical properties of some activated chars. Fuel. Process. Technol. 2002, 77-78, 221–227. (9) Suzin, Y.; Buettner, L. C.; LeDuc, C. A. Characterizing the ignition process of activated carbon. Carbon 1999, 37, 335–346. (10) Grzyb, B.; Machnikowska, H.; Weber, J. V. Mechanism of copyrolysis of coal tar pitch with polyvinyl pyridine. J. Anal. Appl. Pyrolysis 2004, 72, 121–127. (11) Machnikowski, J.; Grzyb, B.; Weber, J. V.; Frackowiak, E.; Rouzaud, J. N.; Be´guin, F. Structural and electrochemical characterisation of nitrogen enriched carbons produced by the co-pyrolysis of coal tar pitch with polyacrylonitrile. Electrochim. Acta 2004, 49, 423–432. (12) Rouzaud, J. N.; Christian, C. Quantitative high-resolution transmission electron microscopy: a promising tool for carbon materials characterization. Carbon 2002, 77-78, 229–235.
ReactiVity of ActiVated Carbons
Figure 1. HRTEM image analysis: structural analysis showing coherent domain limits: L is the individual fringe length, La is the width of the coherent domain, Lc is its height related to d, which is the interlayer spacing, and N is the number of stacked layers within a domain.
sense and eliminated). Pairs of fringes were here considered as stacked layers to form coherent domains (i.e., Basic Structural Units or BSU), only if their angle is smaller than 15° and their interlayer spacing narrower than 0.6 nm (Figure 1). The software allows deaveraged data on fringe length, interlayer spacing, and number of stacked layers in a coherent domain, the proportion of non stacked layers (NSL) as shown in Table 2, to be obtained. Such HRTEM image analysis is a great advantage over the usual X-ray diffraction techniques. As previously indicated, HRTEM images measures individually each grapheme layer profile, length and interlayer spacing. Usually large distributions of values are obtained and can be quantified with histograms.4 In contrast, XRD usually gives only information averaged on a large volume of sample (about 1 mm3 to be compared with the 104 nm3 of the analyzed volumes by HRTEM). Only two structural parameters can be extracted from the position and the width at half-maximum of the 002 reflections (which are usually the only exploitable ones); d002 and LC can be thus calculated with the Bragg’s law and the Scherrer’s relation, respectively. 2.3. Chemical and Textural Characterization. 2.3.1. Elemental Analysis. The mass ratios over carbon for the elements such as oxygen, hydrogen, nitrogen and sulfur of the carbon samples were measured using Thermofinnigan Flash EA-112 elementary analyzer. The experimental error in the determination of the chemical composition of the activated carbons was found to be around 4% for three trials. The results of the elementary analysis can be found in Table 3. 2.3.2. Analysis of Mineral Content. The ash content was qualitatively analyzed using the SEM-X-ray technique. The elemental weight fractions of potassium can be seen in Table 4. We can observe obvious differences between the coconut shell based samples and peat and coal based. It may be seen that the coconut shell activated carbons namely NC-50 and NC-60 possess a higher
Energy & Fuels, Vol. 23, 2009 4053 K content. The presence of potassium in the coconut shell activated carbons is in the following order: NC-50 >NC-60 >NC-100. The higher percentage of K for the coconut-based carbons can be related to the activation or neutralization process during the manufacture where hydroxide or chloride of potassium is added to increase the activation reactivity and burn-offs, thereby increasing the microporosity in the carbons.13,14 The peat- and coal-based activated carbon samples show lower amounts of K. 2.3.4. Porosity Characteristics. The volume of large pores (>8 nm including macro and mesopores) of the AC samples was determined by mercury porosimetry (Micrometrics Auto pore IV 9500). The experimental error was found to be less than 10%. The microporous characteristics of the AC samples were determined from nitrogen adsorption isotherms at 77 K (Micrometrics ASAP 2010 adsorption instrument). Prior to the analysis, the carbon samples were degassed at 250 °C for about 48 h. The specific surface area was measured using the BET model from the adsorption of gases on the surface of the material. The volume of the micropores was measured from the t-plot or De Boer method that provides a simple mean of comparing the shape of a given adsorption isotherm with that of a standard nonporous solid.15,16 The average width of the micropore was evaluated from the Density Functional Theory (DFT).17 Repeatability tests showed that the specific surface area, as well as the volume and the width of the micropores were determined with an accuracy of about 10%. The results of the porosity characteristics and chemical composition can be found in Table 3. 2.4. Thermal Analysis. The instrument ATG-DSC, SETARAM111 was used for the determination of the Point of Initial Oxidation (PIO) and the Self Ignition Temperature (SIT). To eliminate the effect of external size of the AC particles on the properties measured, the samples to be tested were first crushed and finely sieved to have a uniform size (50 µm). About 3-5 mg of crushed sample was tested under a continuous flow of oxygen/helium mixture (21/79). Each sample was heated at 5 K/min from 30 to 650 °C, and an isotherm at 105 °C for 30 min was used to remove adsorbed moisture from the porous matrix. Further information on the methodology can be found from the work of Suzin et al.9 Figure 3 shows an example of the net heat flux and mass evolutions observed. The TG-DSC thermograms describe a continuous exothermic oxidation process ended by complete combustion of the sample. According to Suzin et al.,9 the SIT was determined from the net heat flux curve measured, and corresponds to the temperature where the tangent to the point of inflection intersects
Figure 2. HRTEM images of some of carbon samples used for the study (image size 16 nm × 16 nm).
4054
Energy & Fuels, Vol. 23, 2009
Jayabalan et al.
Table 2. Average Structural Parameters of Carbon Samples Obtained from HRTEM Image Analysis Procedure; L > 1 Ring, L > 2 Ring and L > 3 Rings Correspond to Mean Fringe Length of Fringes Longer Than One Aromatic Ring (0,246 nm), 2 and 3 Aromatic Rings Respectively sample
% nonstacked layers (NSL)
L > 1 ring (Å)
L > 2 rings (Å)
L > 3 rings (Å)
La (Å)
Lc (Å)
N
Dmoy (Å)
RB-2 PAN-A PAN-C CTP-A CTP-1:1-A NC-60 NC-100 NC-50 BPL BC-120 GF-40 Picabiol
57 59 47 14 60 53 51 50 58 65 74 66
5.8 6.4 6.3 9.7 5.8 6.3 6.1 5.6 6 5.5 5.4 4.8
8.4 9 8.8 12.3 8.4 8.9 9.1 8.4 8.6 7.7 7.7 7.1
14.5 14.7 14.4 18.1 13.8 14.4 16.5 14.5 15.0 13.3 12.6 15.2
4.6 3.9 4.3 7.2 3.8 4.4 4.7 4.2 3.7 3.5 3.0 3.4
6.0 5.3 6.9 10.3 5.6 5.9 5.4 6.0 5.4 4.9 4.7 4.5
2.5 2.4 2.8 3.7 2.5 2.5 3.9 2.6 2.5 2.2 2.2 2.1
4.2 3.9 3.8 3.8 3.9 4.0 3.9 3.8 3.8 3.9 4.0 4.1
Table 3. Chemical and Porosity Characteristics of Activated Carbon Samples Tested
a
sample
O/C (%)
volume of micropores (cm3/g)
average width of micropores (nm)
surface area (m2/g)
macro-meso porous volume (cm3/g)
NC-50 NC-60 NC-100 RB-2 BPL GF-40 BC-120 PICABIOL CTP-A CTP-PAN-3:1-A CTP-PAN-1:1-A PAN-A PAN-C CTP-PAN-3:1-C CTP-C CTP-PAN-1:1-C NC-C
1.72 3.60 3.30 5.90 4.10 34.60 35.40 40.60 1.72 3.10 7.20 13.40 6.30 2.40 0.70 3.30 12.40
0.36 0.32 0.27 0.35 0.30 0.29 0.33 0.24 0.04 0.20 0.21 0.26 a a a a a
1.35 0.97 1.11 0.92 0.93 1.15 1.12 1.38 1.30 1.32 1.11 1.15 a a a a a
1078 1220 1803 1012 1106 1718 1975 1534 102 468 482 515 a a a a a
1.28 0.35 0.47 0.34 0.40 0.80 1.50 1.34 0.07 0.25 0.27 0.27 a a a a a
Negligible.
Table 4. Results from Ash Content Elemental Analysis by X-ray Spectra from EDS Performed in a SEM peat
coal
sample
NC-50
coconut shell NC-60
NC-100
RB-2
BPL
K (Elt wt %)
22.70
15.5
10.60
4.60
1.40
the baseline of the DSC values. The SIT defines the limit beyond which the combustion of the sample takes place, and leads to the consumption of the carbon matrix as observed on the mass curve plotted in Figure 3. The PIO represents the point at which the exothermic reaction is significant: a low oxidation process takes place with partial oxidation of the carbon and the organic components. The PIO was extracted from the net heat flux curve by locating the initial point of deviation from the baseline (deviation of 2% from the running average of the prior 5 points). Using this method the relative errors for the experimental determination of the SIT and PIO were estimated respectively to be less than 5 and 10%.
3. Results and Discussions The analysis of the results of the different characterization techniques followed by establishing correlations between the reactivity parameters and the material properties are carried out in this section. This is done by first analyzing the qualitative trends, followed by quantitative correlations using statistical tools giving physical significance. The statistical correlations thus established relate the important material properties of activated carbons with the reactivity parameters assessing the safety of the activated carbons. 3.1. Influence of Oxygen to Carbon Ratio. The influence of oxygen to carbon ratio of activated carbons on the reactivity is shown in Figure 4 where the changes in the log (O/C) versus PIO and (O/C) versus SIT were plotted. First, letting apart the
points representative of the coconut-based samples, a linear trend is observed showing the decrease of the PIO and SIT data, meaning higher reactivity, with an increase in the oxygen over carbon ratio. Such results confirm the role of oxygenated surface functional groups associated to the carbon matrix in the acceleration of the oxidation reactions in air. Oxygen is present in the activated carbons in the form of surface oxygenated groups like carboxyl, hydroxyl, carbonyl, and ether groups.18 Some of the oxygen content in the material may come from its origin and others from its activation mode. When activated carbons are exposed to air and heated, these surface oxygenated groups are decomposed to form intermediate active complexes and desorbable products like CO2, CO, and H2O. Such reactions are accompanied with heat release which contributes to enhance the oxidation process.19,20 (13) Suhas, Carrott, P. J. M.; Ribeiro Carrott, M. M. L. Using alkali metals to control reactivity and porosity during physical activation of demineralised kraft lignin. Carbon 2009, 47, 1012–1017. (14) Amarasekera, G.; Scarlett, M. J.; Mainwaring, D. E. 1998. Development of microporosity in carbons derived from alkali digested coal. Carbon. 1998, 36, 1071–1078. (15) LeclouxA. Exploitation of adsorption and desorption isotherms of nitrogen for the study of textural properties of porous solids. Me´moires socie´te´ des sciences de Lie`ge, Belgium. Tome I, Fasc. 1971, 4, 169–209. (16) Olivier, J. P. Modelling physical adsorption on porous and nanoporous solids using density functional theory. J. Porous Mater. 1995, 2, 9–17. (17) Lastoskie, C.; Gubbins, K. E.; Quirke, N. Pore size distribution analysis of macroporous carbon: A density functional theory approach. J. Phys. Chem. 1993, 97, 4786–4796. (18) Boehm, H. P. Some aspects of the surface chemistry of carbon blacks and other carbons. Carbon 1994, 2, 759–769. (19) Hardman, J. S.; Lawman, C. J.; Street, P. J. Further studies of the spontaneous behaviour of activated carbon. Fuel 1983, 62, 632–638.
ReactiVity of ActiVated Carbons
Energy & Fuels, Vol. 23, 2009 4055
Figure 3. TG-DSC thermograms of a physically activated carbon sample NC-60 showing heat and mass curves.
Figure 4. Effect of O/C ratio on the oxidation and ignition characteristics of activated carbons.
Looking at Figure 4, one can notice deviations to the linear trends observed for all the activated coconut shell samples. These materials have lower PIO values than expected. Many reasons can be attributed for this behavior: higher affinity for the chemisorption of oxygen in air and especially high potassium content in these samples, acting as a catalyst of the oxidation reactions may explain the excessive reactivity of the materials.21 Compared to materials having far lower nitrogen content, nitrogen rich activated carbons produced from PAN fiber are found to be more stable and have higher PIO and SIT values. But nitrogen and oxygen are both present together particularly for the CTP and PAN based samples; it is difficult to show a tendency only for nitrogen to carbon ratio versus PIO and SIT. The main reason for the stability on these N rich activated carbons seems to be the high temperature treatment involved in the activated carbon preparation which leads to thermally stable nitrogen substituting for carbon atoms in the ring system itself. The effect of the O/C ratio on the reactivity is also demonstrated by comparing the carbons from the pyrolysis of coal tar pitch (CTP) and polyacrylonitrile (PAN) fiber blend samples. The carbon from the CTP sample has an O/C ratio of about 1.7% with a high PIO and SIT of 309 and 544 °C, (20) Jayabalan, T.; Pre´, P.; He´quet, V.; Le Cloirec, P. Statistical Quantification of the Influence of Material Properties on the Oxidation and Ignition of Activated Carbons. Adsorption 2008, 14 (4-5), 679–686. (21) van der Merwe, M. M.; Bandosz, J. A study of metal impregnated carbons: the influence of oxygen content in the activated carbon matrix. J. Colloid Interface Sci. 2005, 282, 102–108.
Figure 5. Influence of oxygen to carbon ratio on PIO and SIT values for the carbons from the blends of coal tar pitch and polyacrylonitrile.
respectively. On the other hand, the carbon from the PAN-A has nearly 13.4% O/C ratio approximately 8 times that of CTP based carbon with a lower PIO and SIT. The carbons from the blends CTP-PAN have PIO and SIT depending upon the amounts of CTP and PAN in the blend. For example CTP-PAN3:1-A resulting from the coking of 3 parts of CTP and one part of PAN has O/C of about 3.1% compared to the activated carbon sample having 1 part of CTP and 1 part of PAN having relatively higher O/C of about 7.2%. From Figure 5, it is obvious that the reactivity of CTP-PAN-1:1-A based carbon is higher than that of CTP-PAN-3:1-A which shows that the presence of large
4056
Energy & Fuels, Vol. 23, 2009
Jayabalan et al.
Figure 6. Oxygen content and PIO measured before and after the thermal treatment.
Figure 7. Influence of potassium content on the PIO and SIT.
proportion of CTP having lower oxygen content stabilizes the activated carbons from such blends. To verify that a decrease in the oxygen content of the activated carbon sample lead to a better stability of the material, simple thermal treatment tests were carried out. The activated carbon samples were heated under an inert atmosphere like helium to about 750 °C to remove a part of the surface oxygenated groups. The oxygen content of the samples was then measured by elementary analysis technique without exposing them to air after the 750 °C pyrolysis. This was followed by oxidation in an atmosphere of oxygen. The results reported in Figure 6 show that the thermal treatment reduces the oxygen content of the activated carbon materials with a considerable increase in their PIO temperature.20 This confirms the role of the surface oxygenated groups in the reactivity of the activated carbons. 3.2. Influence of Mineral Content. The influence of the potassium content on the PIO and SIT values is shown in Figure 7. An increase in the K content decreases the PIO and SIT of the activated carbons. The coconut shell activated carbons contain far higher potassium contents than the other materials tested in this study; hence the comparison is made for the coconut shell activated carbon samples. The influence of potassium on the reactivity of the coconut-based materials explains the deviation observed in Figure 4. In agreement with our results, the experimental work of Van der Merwe and Bandosz21 showed that the coconut shell activated carbons are susceptible to ignition reactions when compared to the peatbased activated carbons have a greater propensity to oxidation and self-ignition than peat based activated carbons. This is attributed to the highest amount of mineral matters present in the coconut-based samples resulting from the activation and neutralization process encountered during their manufacture.13,14
Mineral matters like potassium are supposed to provide a direct catalytic activity, which depends on the amount of minerals present, their chemical form and the degree of dispersion.21 3.3. Influence of Porosity Characteristics. Various porosity characteristics were considered in this study: the specific surface area (SBET), meso-macroporous volume (Vporous), the average width of the micropore (Wmicro), and volume of the micropores (Vmicro). The SBET of the carbon samples tested varies from 102 to 2000 m2 g-1 approximately. The chemically activated carbon samples have the highest surface area with an average of about 1700 m2 g-1, whereas it is about 865 m2 g-1 for the physically activated carbons. The carbon from the CTP-PAN blended activated samples has a low SBET with an average of 300 m2 g-1. The volume of micropores varies from 0.04 to 0.4 cm3 g-1, and their average width is evenly distributed with an average value of 1 nm. The average macro-mesoporous volume is about 0.60 cm3 g-1, the chemically activated carbons having the highest values. Despite the variety of these porous characteristics, no direct relationship could have been extracted with the PIO and SIT data set. One can suppose that if the porous properties play a role in the diffusion of oxygen into the interior of the activated carbon matrix as well as the presence of the active sites on their surface, the decomposition of the surface oxygenated groups has a predominant effect on the starting of the oxidation reactions. 3.3. Influence of Structural Properties. From the visual observation of the HRTEM images (Figure 2), we find obvious differences in the structural organization among the carbon samples. Three distinct characteristics have been observed in the carbon samples analyzed based on the fringe length, interlayer spacing, and the proportion of nonstacked layers (NSL). The first group of carbon samples have short fringe length and the graphitic layers are randomly arranged without appreciable stacking of the layers (NSL > 60%). This feature is mainly seen in the chemically activated carbons, namely, GF40, Picabiol, and BC-120. The main reason is being attributed to the mode of activation with a chemical agent and the low temperatures experienced during the process of manufacture. Aromatization, that is, formation of polyaromatic layers, and activation, responsible for an additional grafting of oxygenated groups on the layers boundaries, occur concomitantly. Consequently, small and highly cross-linked layers, are formed rending difficult the layer stacking. Therefore, in comparison with physically activated carbons, L and N are small, whereas d and % NSL are high. An important feature is that these samples have a high O/C ratio.
ReactiVity of ActiVated Carbons
Energy & Fuels, Vol. 23, 2009 4057 Table 5. Matrix of Correlation for Predictor Variables
SBET (m2/g) Vporous (cm3/g) Wmicro (nm) Vmicro (cm3/g) L > 2 rings (Å) L > 1 ring (Å)
Figure 8. Graphitic layer length of the carbon samples and their influence on PIO.
The second group of carbon samples show long graphitic sheets, weakly stacked (about 50% NSL) and slightly bent constituting spaces between them as seen in activated carbon samples like NC-60, BPL, NC-100, and RB-2. These samples were physically activated at around 800-1000 °C and exhibit low oxygen to carbon ratio. The third group of carbon samples contains mainly the activated carbons from coal tar pitch and polyacrylonitrile precursors. They exhibit a lamellar nanostructure of the graphitic layers which tend to be parallel to each other and better stacked compared to the first and second group of samples. The best example of this groups is the CTP-based activated carbon (2 rings (Å)
0.60 0.68 0.47 0.55 -0.62 -0.50
0.69 0.21 0.12 -0.68 -0.67
0.51 -0.17 -0.65 -0.61
-0.37 -0.21 -0.18
0.27 0.31
0.98
A nonlinear trend was observed pointing out the influence of the fraction of nonstacked layers on the reactivity parameters, PIO and SIT. Actually the most stable activated carbons are constituted with lower fractions of nonstacked layers. This effect may be explained by the ordered arrangement of the stacked layers which limit the accessibility of oxygen to the active sites. To quantitatively assess the most influent characteristics of the carbon materials on their reactivity and to determine which phenomena are actually the most limiting in the development of the oxidation reactions, a statistical modeling approach was carried out. 4. Quantitative Statistical Correlations The quantitative analysis of the relationships between the properties of the carbon materials and their reactivity was performed by establishing multiple linear regressions (MLR). Multiple linear regressions were constructed by following a stepwise multiple linear regression, consisting in the successive addition of the predictor variables so that an improvement in the prediction ability of the relationship is obtained at each stage. The prediction ability is assessed from statistical parameters (correlation coefficient, adjusted correlated coefficient, and error of prediction) at each stage. Such a procedure was applied by using Minitab software.20,22 In a first step, the set of samples and the types of predictor variables to be tested has to be defined. Eleven samples were retained for this part of the study. The nonactivated carbonized samples were left aside because of their negligible porosity characteristics. The selection of the predictor variable to be tested was based on the analysis of the matrix correlation of the physical and the chemical characteristics of the materials previously measured. This procedure allows to check that the variables are a priori not linearly dependent to each other.20,23 The multicollinearity character was tested by the use of a correlation matrix presented in Table 5. From Table 5, the predictor variables retained such oxygen to carbon ratio - (O/C) %, specific surface area - (SBET) m2 · g-1, macro-mesoporous volume - (Vporous) cm3 · g-1, micropore volume - (Vmicro) cm3 · g-1, the average width of the micropore(Wmicro) nm and the length of the graphitic layers (L> 2 ring Å) were not related to each other. The analysis of the residuals was also performed to check that the predictor variables were actually uniformly and randomly distributed and to avoid biases. Thus despite their demonstrated influence, the potassium content and the nitrogen over carbon ratio (N/C) were not considered along with the other variables because of the nonuniform distribution of the data within the variation range. (22) Cougnaud, A.; Brasquet, C. F.; Le Cloirec, P. Removal of pesticides from aqueous solution: Quantitative relationship between activated carbon characteristics and adsorption properties. EnViron. Technol. 2005, 26 (8), 857–866. (23) Brasquet, C. F.; Le Cloirec, P. Neural networks modelling of inorganics removal by activated carbon cloths. J. EnViron. Eng. 2001, 127, 889–894.
4058
Energy & Fuels, Vol. 23, 2009
Jayabalan et al.
From the modeling approach, the noninfluence of the other variables representative of the porosity characteristics was confirmed since the introduction of any of these variables in the equation deteriorates significantly the coefficient of determination. Regarding the self-ignition temperature SIT, the MLR equation points out the influence of the two predictor variables which are the O/C content and the macro-meso porous volume Vporous. The standardized equation obtained for the SIT was SIT °C ) -0.22, 6O/C(%) - 0.637Vporous (cm3g-1) with R2 ) 0.78, adj R2 ) 0.72, S ) 45 °C Figure 9. Result of the prediction ability of the best multiple linear regression for PIO.
To compare the relative influence of the predictor variables on the response, standardized data were computed and used to establish the MLE equation. The standardized data are derived from the following equation, and the coefficients were called beta coefficients.24 Regarding the prediction of the point of initial oxidation, the best multiple linear regression obtained involves the O/C ratio and the graphitic layer length. PIO ) -0, 6O/C (%) + 0.5L > 1 ring (Å) S ) 13° C, R2 ) 0.93, adj R2 ) 0.91 The adjusted determination coefficient R2 takes into account the increase in the number of predictor variables involved at each stage of the stepwise procedure. Its value was shown to increase from 0.60 to 0.91 with the introduction of the length of graphitic layer larger to 1 fused ring (in addition to the O/C ratio). The standard error of the estimate S was found to be 13 °C of the PIO. The values and signs of the “beta coefficients” give direct interpretation of the influence of each of the predictor variables on the PIO. Thus it can be concluded that the O/C ratio and length of the graphitic layer exert an opposite and rather equal weight influence on the value of the PIO. Such results are in agreement with the assumptions already advanced regarding the initial stage of the oxidation process. In the low temperature range, the oxidation reactions are mainly initiated by the surface oxygenated groups which decompose to form unstable species and radicals liberating heat and oxidation products. Short graphitic layer are favorable to the reaction progress providing easier accessibility to the oxygen and the active sites. Also, the short graphitic layers have higher amount of surface oxygenated groups at their edges. (24) Giraudet, S.; Pre´, P.; Tezel, H.; Le Cloirec, P. Estimation of the coupled influence of the porosity characteristics of activated carbons and the molecular properties of VOC’s on the adsorption energy. Carbon 2006, 44, 2413–2421.
One can notice that rather high standard error deviation and the value of the adjusted R2 is low showing insignificant trends. Hence the SIT was very clearly explained with the oxygen to carbon ratio with SIT ) 537 - 4.7 O/C (%) with R2 ) 0.96 and S ) 16 °C and all additional variables do not improve the predictive ability of the regression equation. 5. Conclusions The oxidation and ignition properties of a large number of activated carbons having diversified origins and characteristics were studied. The thermogravimetry analysis coupled with differential scanning calorimetry was used in order to determine the reactivity parameters like the point of initial oxidation (PIO) and self ignition temperature (SIT). The HRTEM observation coupled with sophisticated image analysis tools provided useful information about the structural organization of different carbon materials. This was completed with the measurement of chemical and textural characteristics. The influence of the physical and chemical characteristics of the carbon materials on their reactivity with air was quantitatively analyzed from the establishment of linear relationships. It was found that the graphitic layer length and the oxygen over carbon ratio are directly influent on the oxidation parameter while the only oxygen to carbon ratio is influent on the ignition parameter. No correlations were found with textural parameters such as the macromesoporous volume and the specific surface area. On the other hand the reactivity was better explained with the structural properties. Though the properties of oxidation and ignition may change with experimental conditions, the results obtained here enable us to assess the properties influencing the reactivity of activated carbons under a given condition. The results also give an insight on the mechanism of oxidation and ignition in activated carbons. Acknowledgment. The authors thank Professor Guy Furdin, Laboratoire de Chimie du Solide Mine´ral, University Henri Poincare´, Nancy, France for supplying nitrogenated activated carbon samples. EF9001296
