Influence of Sampling and Storage Protocol on Fractal Morphology of Soot Studied by Transmission Electron Microscopy

Abstract

The aim of this work was to compare the fractal characteristics, D_f and k_f, the primary particle diameter, D_pp, the gyration diameter of aggregates, D_g, and the overlap coefficient, C_ov, of carbon nanoparticle aggregates produced by an ethylene diffusion flame and sampled by means of four commonly used techniques. The first method involves a thermophoretic piston probe (TPP) which inserts a TEM grid into the flame. Three other methods were applied at the outlet of a dilution device, also inserted in the flame. The first of these used a nuclepore filtration sampler (NFS), and is based on filtration of particles onto a polycarbonate membrane. The second, post dilution method, the insertion particle sampler (IPS), inserts a TEM grid, perpendicular to the aerosol flow. Similar to TPP, the last method is a thermophoretic particle sampler (TPS) sampling directly onto a TEM grid. After collection, the samples are stored in the dark either, (1) in a nitrogen filled cell at low humidity or, (2) in ambient air for studying atmospheric ageing. Good agreement was observed between TPP, TPS, and IPS indicating that the dilution induced for TPS and IPS does not significantly change the morphology of soot. On the other hand, the NFS protocol tended to overestimate the overlap coefficient and the size of primary particles and aggregates. Finally, with regard to the aging effect, we found that k_f and D_pp evolve slowly during storage in the atmosphere while D_f, was insensitive to the storage conditions. However, the overlap coefficient increased and the gyration diameter decreased as a function of storage duration, while storage under nitrogen tended to reduce these changes.

NOMENCLATURE

	=	:   amplitude of the size distribution of primary particle diameters and aggregate gyration diameters
	=	:   projected overlap coefficient of two overlapped primary particles
	=	:   mean projected overlap coefficient of a population of aggregates
	=	:   overlap coefficient of a population of aggregates
	=	:   distance between the centers of mass of two overlapped primary particles (nm)
Df	=	:   fractal dimension
Dg	=	:   gyration diameter of the aggregate (nm)
	=	:   modal gyration diameter of the aggregate (nm)
Dpp	=	:   primary particle diameter (nm)
	=	:   mean primary particle diameter (nm)
IPS	=	:   insertion particle sampler
ka	=	:   empirical constant (1.095)
kf	=	:   fractal prefactor
N(Dpp)	=	:   primary particle diameters’ size distribution
Nbpixel	=	:   number of pixels composing the TEM micrograph of the aggregate
NFS	=	:   nuclepore filtration sampler
Npp	=	:   number of primary particles composing an aggregate
p(Dg)	=	:   probability density function of gyration diameters
Rg2D	=	:   projected radius of gyration of the aggregate (nm)
Sag	=	:   projected surface area of the aggregate (nm2)
Spp	=	:   projected surface area of a primary particle (nm2)
TEM	=	:   transmission electron microscopy
TPP	=	:   thermophoretic piston probe
TPS	=	:   thermophoretic particle sampler
()	=	:   coordinates of a given pixel on the TEM micrograph of the aggregate
	=	:   coordinates of the centre of mass of the aggregate

GREEK SYMBOLS

α	=	:   empirical constant (1.155)
ς1, ς2	=	:   empirical constants (ς1=1.1±0.1 and ς2=0.2±0.02)
σpp	=	:   standard deviation of the primary particle diameter size distribution (nm)
σgeo	=	:   geometric standard deviation of the aggregate gyration diameter size distribution

1. INTRODUCTION

Soot particles produced during combustion of hydrocarbons or organic matter are of strong interest in the fields of environmental science (global warming contribution, air pollution), fire science (radiative impact of soot particles on fire propagation), air filtration (clogging of high efficiency particulate air filters, Mocho and Ouf 2010), and combustion science (contribution of soot to the heat flux of flames, optimization of combustion processes). Since the pioneering work of Witten and Sander (1981) and the application of the fractal theory of Mandelbrot (1973) by Jullien and Botet (1987), many investigations have been reported which describe the specific morphology of particles produced by diffusion flames (Dobbins and Megaridis 1987; Megaridis and Dobbins 1990; Köylü and Faeth 1992; Sorensen and Feke 1996; Ouf et al. 2008), by automotive engines (Kittelson 1998; Sachdeva and Attri 2008) and by aircraft engines (Colbeck et al. 1997; Petzold et al. 1999; Popovitcheva et al. 2000; Barone et al. 2006). Because of their diffusion limited mode of aggregation, soot particles display well-known fractal morphology and several methods for analysis of their shape have been proposed in the literature (Köylü et al. 1995; Brasil et al. 1999; Xiong and Friedlander 2001; Bushell et al. 2002; Lee and Kramer 2004; Tian et al. 2006). Soot particles usually consist of small primary particles (diameter D_pp) that agglomerate to form irregular clusters. Their complex geometry can be described by a relationship, known as a fractal-law, between the number of primary particles in the agglomerate, N_pp, which is proportional to the mass, and one particular agglomerate parameter, R_g, the radius of gyration. Three other parameters are included in this relationship, the primary particle diameter D_pp, the fractal dimension D_f and the prefactor k_f.

Even though in-situ methods are available for measurement of the fractal dimension D_f, by measuring the angular light scattering for example (Sorensen et al. 1992; Sorensen 2001; Kim and Choi 2003), most studies take advantage of ex-situ methods which allow direct observation of the particles using Transmission Electron Microscope (TEM) projected images. In this case, the two-dimensional information can be extrapolated to determine three-dimensional properties by applying simple relationships (Medalia and Heckman 1969; Samson et al. 1987; Cai et al. 1993; Rogak et al. 1993; Köylü et al. 1995; Neimark et al. 1996; Park et al. 2004; Ishiguro et al. 1997; Wentzel et al. 2003; Chen et al. 2006; Tian et al. 2006; Chandler et al. 2007; Chung et al. 2008).

Various methods are used in order to deposit nanoparticles onto TEM grids. The thermophoretic method, based on a rapid insertion (as short as 30 ms) of a TEM grid in the flame, was first introduced by Dobbins and Megaridis (1987) and has since been widely used (Cai et al. 1993; Köylü et al. 1997; Tian et al. 2004; Lee et al. 2008). However, other sampling methods, involving insertion of a dilution probe into the flame or plume region, are also employed (Van Hulle et al. 2002; Matti Maricq et al. 2003; Zhao et al. 2003; Ouf et al. 2008; Teng et al. 2009). In these methods, a sampling probe is inserted in the vicinity of the flame and rapid quenching is carried out in order to prevent agglomeration, nucleation or water condensation at the surface of the soot. At the outlet of these dilution systems, particles can either be deposited directly onto a membrane (Spurny 1986; Colbeck et al. 1997; Ouf et al. 2008) or onto TEM grids using specific protocols primarily based on impaction, interception and diffusion processes (Sorensen and Feke 1996; Van Hulle et al. 2002) or thermophoresis (Bang et al. 2003; Thomassen et al. 2006; Wen and Wexler 2007).

FIG. 1 The burner and the experimental setup used to sample soot particles in the ethylene diffusion flame.

The main aim of this study is to investigate the possible influence of the sampling methods, currently widely used in the community, on soot morphological parameters (k_f, D_f, D_pp, D_g, C_ov) determined using TEM images. To this end, four methods for sampling and deposition of soot onto TEM grids have been applied simultaneously to an ethylene diffusion flame.

Several studies using high resolution TEM have shown that monomers are composed of planar nano-crystallites contained in an outer shell and an inner core; this core may incorporate several fine spherical particles (Ishiguro et al. 1997; Vander Wal 1997; Wentzel et al. 2003). All these structures show some graphite-like order. Variations of the inter-planar spacing have been observed (Shaddix et al. 2005; Kis et al. 2006) and can be attributed to the nature and number of the molecules that are absorbed on the planar graphitic nano-structure. Initially, the organic matter content of soot particles is high and there is a low oxygen concentration. During storage, exchange of molecules with the air is possible, which could give rise to variations in the primary particle structure. Moreover, according to recent reports in the literature (Yezerets et al. 2003; Kandas et al. 2005; Braun 2009), soot particle ageing may occur even under ambient conditions of temperature, pressure, humidity, and sunlight. More precisely, results have been reported on the ageing of soot, under ambient conditions, induced by ozone (Decesari et al. 2002), by desorption and oxidation (Cheng et al. 2006; Braun 2009; Cheng and Lehmann 2009), by water uptake (Popovitcheva et al. 2008), or by photolysis of polycyclic aromatic hydrocarbons (PAH, Kim et al. 2009) and all these phenomena are able to change the microstructure and composition of such particles (Ishiguro et al. 1991; Hurt et al. 1993; Popovitcheva et al. 2008; Kim et al. 2009). Hence, in the present work, the influence of storage duration, preservation conditions and sample preparation procedures, all of which could have influenced the soot morphological parameters, are also investigated by TEM analysis.

2. EXPERIMENTAL PROCEDURE

2.1. Diffusion Flame Burner

The soot production apparatus is shown in Figure 1. It consists of an atmospheric hybrid Holthuis burner (previously McKenna burner) equipped with a 60 mm diameter porous bronze sintered matrix around a central 6.35 mm diameter tube allowing the introduction of an injector (Lemaire et al. 2009) or gas flow. In the present study, a diffusion flame was obtained by injecting ethylene at a rate of 1 L/min into the central part of the burner. An air flow of 38 L/min passed through the annular porous component of the burner in order to stabilize the flame. The flame height obtained was around 10 cm.

As depicted in Figure 1, in-situ (Thermophoretic piston probe, TPP) and ex-situ sampling methods (dilution probe FPS 4000) were applied at the same location in the flame, at a height above the burner equal to 4 cm, in a region where the particle concentration is large. Figure 1 also shows three other sampling protocols, a thermophoretic particle sampler, TPS, an insertion particle sampler, IPS, and a Nuclepore filtration sampler, NFS, each implemented at the outlet of the dilution probe.

2.2. Sampling Procedures

Four sampling methods have been compared using this experimental device for soot production. The first involved inserting a TEM grid into the diffusion flame while the three other methods were applied at the outlet of the dilution probe. The techniques are mainly based on thermophoresis, or impaction, interception, and diffusion deposition on TEM grids, or filtration of soot particles by a polycarbonate membrane. Sampling protocols have been considered according to their distinct fields of application. The TPP device is widely used in the field of combustion science, in order to study soot particles inside flames. The three others devices (TPS, IPS, and NFS) are designed to sample particles under more diluted and moderate conditions, such as in flame plumes or in indoor or outdoor environments.

2.2.1. Thermophoretic Piston Probe (TPP)—Direct Flame Sampling

In the work reported here, a special Thermophoretic Piston Probe (TPP), similar to the one proposed by Dobbins and Megaridis (1987) and Köylü et al. (1997), was used to collect soot particles on a TEM grid (Cai et al. 1993; Köylü et al. 1997; Tian et al. 2004; Lee et al. 2008). A TEM grid is inserted parallel to the flame axis using a pneumatic piston. The grid immobilization time at the center of the flame (10 to 1000 ms) is optically controlled (Figure 1). Such an orientation promotes thermophoretic deposition of particles onto the grid, due to the temperature difference between the flame and the grid.

2.2.2. Dilution Device for Ex-Situ Sampling Systems

In contrast to the insertion of TEM grids using a thermophoretic piston, ex-situ techniques for measurements of size distribution (Matti Maricq et al. 2003; Zhao et al. 2003; Teng et al. 2009), mass concentration (Ouf et al. 2008), and optical properties (Zhu et al. 2000; Van-Hulle et al. 2002; Ouf et al. 2008), require continuous sampling at non-negligible flow rates. However, a rapid quenching of some physical and chemical phenomena (agglomeration, nucleation, and oxidation, for example) needs to be performed. In the present study, in order to compare the different methods used to deposit soot particles on TEM grids or membranes, soot particles were sampled in the flame using a two-stage dilution device (Figure 1) (DEKATI FPS 4000, Mikkanen et al. 2002; Ouf et al. 2008). The sampling probe was heated to 180°C to avoid condensation of volatile vapor and steam on its inner surfaces. The first dilution was performed with heated filtered air (180°C), passing through a porous tube to decrease the dew point of the sample. Then, a second dilution was applied to the sampled volume using cold air passed into an ejector-type dilution device. Therefore, the sample was at ambient temperature and pressure after leaving the probe and a dilution ratio of approximately 100 was used. In this study, three different methods were used to deposit soot particles onto TEM grids, as shown in Figure 1, and these are detailed below.

2.2.3. Thermophoretic Particle Sampler (TPS)

The Thermophoretic Particle Sampler is a system specifically developed to collect particles at ambient conditions; in this study, from the exhaust of the dilution system. This system was developed by the Fraunhofer Institute of Toxicology (Germany) and it is composed of a heated (150°C) sampling tube in which a TEM grid is inserted parallel to the aerosol flow (3 L/min) and cooled using a Peltier module to 20°C. This device has been recently used and described by Thomassen et al. (2006). Using the TPS, soot particles are deposited directly onto TEM grids without any treatment. The main physical phenomenon responsible for particle deposition on the TEM grid is thermophoresis as previously discussed. Comparison of the two techniques (TPP and TPS) allows the effects of sampling and dilution to be investigated independently of other possible effects.

TABLE 1 Gaussian fit parameters for the four sampling protocols TPP, TPS, IPS, and NFS, with standard deviation in parentheses

Sampling method	Number of primary particles	Dpp (nm)	Number of pairs of overlapped primary particles	Overlap coefficient Cov
TPP	814	24.4 (5.2)	111	0.18 (0.05)
TPS	764	26.9 (5.9)	113	0.20 (0.04)
IPS	958	26.8 (5.8)	90	0.17 (0.05)
NFS	904	37.1 (6.6)	229	0.23 (0.03)

2.2.4. Insertion Particle Sampler (IPS)

The second device was developed by Van Hulle (2002) for sampling onto TEM grids in the exhaust of a dilution probe inserted in a diffusion flame. This method, with a grid inserted perpendicularly to the aerosol flow, is based, depending on the particle size, on interception, impaction or diffusion of particles directly onto TEM grids. No information is available on its collection efficiency and its size dependence, nevertheless this very simple device has been implemented in order to demonstrate its potential suitability for sampling soot particles in such situations.

2.2.5. Nuclepore Filtration Sampler (NFS)

In parallel to the TPS and IPS, a filter holder was used for sampling soot particles at the outlet of the particle dilution device (Figure 1). Polycarbonate membranes with a pore size of 0.2 μm (Whatman Nuclepore) were used and a post-treatment was applied in order to transfer the soot particles from the membrane to the TEM grid.

In this experiment, TEM grids were observed without any particular preparation in the case of the TPP, TPS, and IPS procedures. However, a specific treatment needed to be applied to the Nuclepore membranes of the NFS procedure, following Spurny (1986). After sampling, particles were transferred onto TEM grids (200 mesh standard copper grids, Agar Scientific) according to the following protocol. First, the membrane was coated with a graphite film by vacuum evaporation (Jeol JEE 400). Next, a piece of coated polycarbonate membrane was placed on the TEM grid and then partially dissolved using chloroform. In this way particles were sandwiched between polycarbonate membrane remnants and evaporated graphite film. The influence of such a treatment on the morphology and size of soot particles (primary particles and aggregates) will be discussed below.

2.3. Storage Procedure

During the experimental procedure, two batches of samples (grids or membranes) were collected for each sampling protocol. The first batch of samples was stored away from light to avoid photolysis as reported by Kim et al. (2009), in a chamber filled with nitrogen (100 mbar) at a relative humidity of 15% and at a mean temperature of 20°C. The second batch of samples was stored under ambient conditions in the laboratory (20°C, 65% RH and also away from light). Moreover, to investigate soot ageing effects, morphology was studied as a function of storage times up to 10 months in ambient air.

3. MORPHOLOGICAL ANALYSIS

3.1. TEM Image Production

After sampling onto TEM grids (200 mesh copper grid with Formvar/carbon film, Agar Scientific) and on polycarbonate membranes (Whatman Nuclepore, with a pore size of 0.2 μm), transmission electron microscopy was used to obtain soot images for determination of the primary particle size and the fractal dimension of the aggregates. Soot particles were studied using a JEOL 100CXII TEM at magnifications from × 72.000 to × 320.000 with an accelerating voltage of 100 kV. For each magnification, polystyrene latex spheres of known diameter, 102 ± 3 nm (Duke Scientific), observed under the same conditions as the soot, were used to calibrate the images and convert dimensions from pixel numbers to a nanometer scale. A Dualvision 3000W model 780 CCD camera (Gatan Inc.) coupled to the microscope generated digital images for analysis.

Approximately 100 to 200 micrographs of individual soot particles were collected for each experimental condition. These were taken randomly at various locations across the grid. The primary particle diameters were manually determined from these micrographs.

3.2. Image Analysis Procedure to Determine the Morphological Parameter

This section describes the morphological analysis procedure based on the TEM micrograph treatment. Two analyses were performed by two different operators, each using different analysis software, since it was anticipated that the results of some steps in the image analysis might be influenced by contributions from the software.

For spherule diameters, the differences were always less than the observed dispersion, which is shown below. No systematic biases were observed for the fractal dimension and prefactor. For this reason, the results presented have been obtained after merging the two data sets.

3.2.1. Determination of Primary Particle Diameters

Many authors have reported primary particle diameter measurements for a large variety of fuels (Dobbins and Megaridis 1987; Megaridis and Dobbins 1990; Köylü and Faeth 1992; Sorensen and Feke 1996; Kittelson 1998; Van Hulle et al. 2002; Barone et al. 2006; Ouf et al. 2008). The analysis procedures are based on ImageJ (Rasband 1997–2009) and Saisam (Microvision Instruments) software and the diameters of the primary particles are measured by superposing a circular shape on the apparent primary particles. From 700 to 1000 primary particles (cf. Table 1) were analyzed for each experimental condition and a histogram was generated from these measurements using Sigmaplot (Systat software) or Saisam software. Finally a normal size distribution fit is applied to the experimental results according to Equation 1, where is the most probable diameter, σ _pp is the standard deviation and A is the amplitude of the distribution (A is not presented hereafter):

An average experimental uncertainty of 1 pixel has been estimated for all particle images reported here, corresponding to an uncertainty range of 0.4–1.5 nm for magnifications ranging from × 72,000 to × 300,000.

3.2.2. Determination of Fractal Parameters

The second step for fractal analysis of aggregates consists of converting their TEM projected images into binary, black/white, images of the aggregates and background respectively, for a given threshold setting. In this study, TEM image segmentations were processed using Adobe Photoshop and Adobe Photoshop Elements software (Adobe Systems). Then, the binary images obtained were analyzed, with software developed in the Matlab environment, to determine the morphological characteristics on a projected image of the aggregates such as the maximum length, maximum height, width, projected area, perimeter, and Feret diameter.

Following the pioneering work of Witten and Sanders (1981), the fractal nature of soot particles has been widely confirmed (Jullien and Botet 1987; Megaridis and Dobbins 1990; Köylü and Faeth 1992; Köylü et al. 1995). A soot particle is an aggregate of spherical primary particles. The number of primary particles, N_pp, composing this aggregate is related to the two characteristic diameters, the gyration diameter, D_g, and the primary particle diameter, D_pp, by:

where k_f and D_f are the fractal prefactor and the fractal dimension, respectively.

It should be noted that various procedures are available in the literature to determine the fractal dimension and prefactor from projected images. For example, one procedure is based on the maximum length L of the aggregates (Köylü et al. 1995; Brasil et al. 1999) and another is based on the box-counting method (Xiong and Friedlander 2001). Nevertheless, the aim of the present work is not to compare these different procedures but, for a given procedure, to compare the effects of the sampling and storage methods. Hence, it was decided to use the approach, based on the gyration diameter D_g, as described by Köylü et al. (1995) which is widely used in the literature (Sorensen and Feke 1996; Hu et al. 2003; Park et al. 2004; Ouf et al. 2008) and by the combustion research community.

First the primary particle projected surface area, S _pp , and the aggregate projected surface area, S_ag, were directly determined from TEM micrographs, in pixels. The ratio S_ag/S_pp defines a two-dimensional number representative of the primary particles composing the aggregate. To obtain the three-dimensional number of primary particles that compose the aggregate, N_pp, the following empirical relationship was used:

Here, k _a = 1.155 and α= 1.095 are empirical constants established by Köylü et al. (1995) for soot aggregates with a fractal dimension less than 2. The gyration diameter is also directly determined from TEM micrographs by computing the centre of mass of the projected aggregate. If (x, y) _i are the coordinates of the Nb_pixel pixels composing the aggregates, the coordinates of the center of mass (x _B , y _B ) are defined as:

The projected gyration radius of the aggregate can then be defined as the mean distance between the pixels composing the aggregate and its centre of mass:

This two-dimensional gyration diameter, D_g,2D, was then related to the three-dimensional gyration diameter, D_g, according to the empirical relationship introduced by Sorensen and Feke (1996):

N_pp and D _g were determined for several soot aggregates. The results were then plotted in log-log coordinates and a linear regression applied to the experimental results gave the fractal dimension and the fractal prefactor, which are respectively the slope and the exponential of the intercept of the linear regression.

3.2.3. Determination of the Overlap Coefficient

A significant parameter for describing soot particles as agglomerate or aggregate is the overlap coefficient. By definition, an agglomerate is composed of primary particles bonded together by weak, Van der Waals type, forces and exhibiting a low degree of overlap. Conversely, an aggregate is composed of strongly bonded primary particles (covalent type bonds) with a high degree of overlap. Brasil et al. (1999) defined the projected overlap coefficient C_ov,P of two touching primary particles, with a mean diameter of and a distance between the centers of mass of , as:

According to the work of Brasil et al. (1999), this projected overlap coefficient may be converted to the real, three-dimensional, overlap coefficient C_ov, taking into account the bias induced by an additional monomer superposition due to the 2D projection:

where ς₁ and ς₂ are proposed empirical constants. For D _f ≈ 1.78:

Although this method has recently been used and validated by Bau et al. (2010) for measuring the specific surface area of nanoparticle aggregates, the overlap coefficient may be subject to measurement artefacts. Therefore, in the present work, this parameter will only be used to report qualitative information.

4. EXPERIMENTAL RESULTS

Figure 2 shows TEM micrographs of soot obtained using the four different sampling protocols. Note that the holes observed on the NFS micrograph are the 0.2 μm pores of the polycarbonate membrane. The density of deposited particles cannot be compared because the exposure times and physical processes involved are not the same in each case. Nevertheless, qualitative observation shows that the size of deposited particles depends on the chosen protocol. The TPP method presents the highest number of weakly agglomerated particles because these are taken directly from the flame. TPS and IPS exhibit medium sized particles which may be due to agglomeration in the dilution probe. Finally, the NFS micrograph clearly shows that the NFS protocol enables observation of the largest particles, which cannot be collected by the other techniques (TPP, TPS, and IPS).

FIG. 2 TEM micrographs of samples collected by TPP, TPS, IPS, and NFS and stored in a N₂ atmosphere in the dark.

4.1. The Influences of Sampling Protocols

In order to compare morphological properties of the soot particles obtained from the four distinct procedures, the samples were stored in a chamber filled with nitrogen at a relative humidity of 15% and in the dark. For these samples, TEM micrographs were obtained after 60 to 100 days of storage in nitrogen.

4.1.1. Primary Particle Diameter and Overlap Coefficient

Following earlier authors (Hu et al. 2003; Tian et al. 2004; Ouf et al. 2008), the size distributions have been fitted by a Gaussian curve, and the parameters obtained, mean values, are given in Table 1.  From consideration of the direct sampling of soot particles on TEM grids (TPP, TPS, IPS), the most probable primary particle diameter lies in the range from 24.4 to 26.9 nm and this is in good agreement with the values of 19 to 35 nm and 28.3 nm reported respectively for turbulent and laminar non-premixed ethylene/air flames by Hu et al. (2003) and Tian et al. (2004). However, it is clear that the NFS protocol tends to over-estimate primary particle diameter by more than 10 nm.

Few values are available in the literature for the overlap coefficient. However, the present results are in good agreement with C_ov values reported by Wentzel et al. (2003) for diesel soot (ranging from 0.10 to 0.29) and with the values considered by Brasil et al. (1999) in their numerical study (ranging from 0 to 0.33). As with the primary particle diameter, the NFS protocol over-estimates the overlap coefficient by a factor of 1.2 to 1.4.

Table 1 shows that, for the primary particle diameter and overlap coefficient, the differences between TPP, TPS, and IPS results are small. Since no differences are observed, this suggests that no transformation process occurred which resulted in variation of the primary particle diameter during the early stages of the sampling in the dilution probe FPS.

FIG. 3 Probability density function of the gyration diameter obtained for the four sampling protocols (Number of aggregates analyzed: 211 for TPP, 141 for TPS, 202 for IPS, and 225 for NFS).

4.1.2. Aggregate Size, Prefactor, and Fractal Dimension

Turning now to the aggregate size; Figure 3 presents the probability density function p(D _g ) in terms of gyration diameter obtained for the different sampling protocols and their associated log-normal fitting (Equation (10)):

where is the modal diameter and σ_geo is the geometric standard deviation.

For the TPP, good agreement is observed between present results and published results of Megaridis and Dobbins (1990) and Hu et al. (2003) which, using the same sampling protocol, reported mean gyration diameters for soot from ethylene/air diffusion flames in the ranges 94–178 nm and 34–106 nm, respectively. From comparing modal diameters between thermophoretic sampler TPP and TPS (see Figure 3), it could be concluded that an agglomeration occurs in the dilution system prior to the TPS. However, as the temperature gradients and sampler characteristics are significantly different between these two methods, this discrepancy in diameter may also be explained by different collection efficiencies. For sampling at the outlet of the dilution device, the IPS size distribution returns lower diameters than TPS whereas the value from NFS is significantly larger than both of these (Figure 3). Furthermore, the IPS result is in good agreement with the TPP measurement which is considered to be the reference method in the field of combustion aerosol science. Since NFS is a filtration based method, its collection efficiency is not as size dependent as the other methods. Hence, one may assume that it is the most representative sampler in terms of aggregate size distribution but, as will be reported in the next section, the post-treatment induced by the NFS, overestimates the primary particle diameter and is suspected to displacing the size distribution to larger diameters.

The conclusion on the effect of sampling method on size distribution is that modal gyration diameters depend on the sampling method and range from 79 to 267 nm. The discrepancies observed between gyration diameters could be due to the different collection efficiencies of the sampling devices for which few data are available in the literature. However, as the number of aggregates used for determining the gyration diameter size distribution is statistically low (around 200), the quantitative interpretation of the size distribution established by TEM must be considered cautiously.

FIG. 4 The number of primary particles in the aggregates as a function of the ratio between gyration and primary particle diameters.

In a similar way the influence of sampling protocols on the prefactor, k_f, and on the fractal dimension, D_f, has been studied. As explained above, these parameters can be determined by plotting the logarithm of the number of primary particles composing the aggregate, N_pp, as a function of the logarithm of the ratio of the aggregate gyration diameter and the primary particle diameter. For convenience, N_pp is shown below as a function of D_g/D_pp (Figure 4) on a log-log scale. Even taking into account the scatter in the data points, it appears that the fractal parameters determined by the four sampling protocols are in good agreement and are not dependent on the sampling procedure. Linear fits, presented in Figure 4, give the values obtained for D_f and k_f following the analysis procedure described above. The uncertainties in D_f and k_f are computed according to the statistical analysis described by Ouf et al. (2008).

From Figure 4, it is evident that, when normalized according to the primary particle diameter, the gyration diameter for particles collected with the NFS tends to be larger than those collected with other samplers. This is consistent with the size distributions that were shown in Figure 3. Figure 4 also shows that the fractal dimensions range from 1.69 to 1.80 and the fractal prefactors from 2.09 to 2.84. The NFS method provides the lowest value for D_f and the highest value of k_f. This is mainly due to applying the linear regression to larger aggregates for the NFS method than for the other protocols. These results are in good agreement with previous results of Köylü et al. (1997) and Hu et al. (2003) who reported D_f= 1.72 with k_f= 2.8 and D_f= 1.74 with k_f= 2.2, respectively. There is a controversy in the literature regarding k_f and, while our present results are in agreement with previous works of Köylü et al. (1997), Hu et al. (2003), Wentzel et al. (2003), Tian et al. (2006), Ouf et al. (2008), some numerical simulations (Sorensen and Roberts 1997; Oh and Sorensen 1997; Brasil et al. 1999) and experimental characterization (Cai et al. 1993; Cai et al. 1995; Sorensen and Feke 1996) refer to k_f values less than 2. For numerical simulations the discrepancy observed with our results could be associated to the overlap coefficient which, according to Brasil et al. (1999), tends to increase k_f as it increases. Our measurements involved only one fuel (ethylene) and one method for calculating D_f and k_f. They are therefore not sufficiently comprehensive to resolve this discrepancy. Nevertheless, D_f and k_f were similar for the sampling methods that we used.

4.1.3. Analysis of the NFS Sampling Artefact

Discrepancies observed between direct sampling protocols (TPP, TPS, IPS) and Nuclepore filtration sampling (NFS) may be due to the transfer procedure from the Nuclepore to the TEM grid which consists of: (1) carbon deposition on the membrane surface and (2) dissolution of the membrane in chloroform. To separate the effects of these two treatments, a carbon deposition was applied to some TPP and TPS grids, and then the TPS grid was dissolved in chloroform. Morphology (D_f, k_f, D_pp, D_g) and mean distances between centers of mass of primary particles used for the C_ov measurement, are reported in Table 2.

It is clear that the carbon deposition significantly influences the primary particle diameter and the mean distance between the centers of mass of the primary particles. For the TPP and the TPS methods, an increase in D_pp of 28 and 46% was identified whereas the influence of CHCl₃ was not as significant, with less than 4% increase in primary particle diameter. The same effect has been observed for the overlap coefficient with an increase in this parameter of between 13 and 35% for the carbon deposition and of more than 21% for the CHCl₃ dissolution. This means that carbon deposition and chloroform dissolution both have an impact on the observed size of monomers and the distances between them. In a same way a growth of 20 to 30 nm is observed in the gyration diameter of the aggregates. Simultaneously, an increase of 15 nm was observed in the diameter of calibrated polystyrene latex sphere (102 nm) after carbon deposition. The present study shows that the carbon coating and chloroform dissolution involved in the NFS method is responsible for the growth in apparent size of the particle. This cannot be corrected for since it seems to be dependent of the particle's nature or shape.

TABLE 2 Influence of the NFS transfer procedure on morphological parameters. This influence is analyzed by applying carbon deposition (C) to TPP and TPS grids (standard deviation for D_pp and geometric standard for D_g in parentheses)

Protocol	Treatment	Df	kf	Cov	Dpp (nm)	(nm)	(nm)
TPP	Before C	1.7 (0.22)	2.37 (1.38)	0.16 (0.04)	26.5 (5.0)	19.0 (4.4)	100.1 (2.2)
	After C	1.82 (0.22)	2.33 (1.34)	0.18 (0.05)	38.6 (6.1)	25.9 (5.6)	128.4 (2.0)
TPS	Before C	1.76 (0.16)	2.28 (1.08)	0.17 (0.06)	33.4 (6.1)	22.7 (6.3)	122.9 (2.2)
	After C	1.82 (0.21)	2.43 (1.42)	0.23 (0.06)	42.9 (8.0)	26.3 (6.5)	151.6 (2.0)
	After CHCl3	1.77 (0.19)	2.45 (1.08)	0.28 (0.04)	44.6 (9.7)	25.4 (7.2)	139.5 (1.8)

TABLE 3 Evolution of morphological parameters as a function of storage duration in air and nitrogen; with standard deviation for D_pp and geometric standard deviation for D_g in parentheses. The initial values for TPP with nitrogen correspond to results analyzed in the previous section (Table 1; Figure 4)

Protocol	Storage duration (days)	Df	kf	Cov	Dpp (nm)	(nm)
TPP (air)	33	1.68 (0.18)	2.69 (1.39)	0.16 (0.05)	28.2 (4.8)	154 (1.9)
	109	1.70 (0.18)	2.54 (1.42)	0.18 (0.07)	22.9 (3.7)	149 (1.9)
	211	1.70 (0.20)	2.70 (1.41)	0.23 (0.03)	27.1 (4.5)	132 (1.9)
	336	1.65 (0.24)	2.88 (1.48)	0.27 (0.03)	29.0 (4.7)	126 (1.7)
TPP (nitrogen)	110	1.76 (0.09)	2.09 (1.14)	0.18 (0.05)	24.4 (5.2)	79 (2.1)
	372	1.83 (0.14)	2.07 (1.07)	0.21 (0.04)	26.4 (4.7)	97 (2.4)
TPS (nitrogen)	81	1.76 (0.10)	2.17 (1.18)	0.20 (0.04)	26.9 (5.9)	123 (2.3)
	372	1.76 (0.16)	2.28 (1.08)	0.17 (0.06)	33.4 (6.1)	123 (2.2)
IPS (nitrogen)	60	1.80 (0.09)	2.23 (1.13)	0.17 (0.05)	26.8 (5.8)	86 (2.4)
	348	1.84 (0.12)	2.17 (1.07)	0.22 (0.02)	27.7 (6.2)	80 (2.3)
NFS (air)	39	1.74 (0.12)	2.32 (1.25)	0.20 (0.03)	36.0 (6.5)	221 (2.4)
	110	1.74 (0.16)	2.66 (1.30)	0.24 (0.04)	35.1 (5.8)	171 (2.1)
	214	1.79 (0.18)	2.51 (1.31)	0.28 (0.06)	42.7 (6.9)	174 (1.8)
	337	1.76 (0.19)	2.52 (1.34)	0.30 (0.04)	40.6 (7.1)	167 (1.8)

4.2. Storage Protocols Artefacts

The evolution of the morphological parameters (D_g, D_pp, D_f, C_ov, and k_f) of soot particles, during long periods of storage at ambient conditions, was also investigated. Sampling was performed in similar ways to the previous analysis, “Sampling protocols influence,” however, two series of samplings (for both TPP and NFS) were undertaken and one set was stored in air, while the other was stored under nitrogen. Particle morphology was analyzed after approximately one, three, six, and ten months of storage in air under ambient temperature, pressure and relative humidity, while for nitrogen only two storage durations were investigated. Then, the determination of the morphological parameters was carried out as a function of storage duration in air; the results are presented in Table 3. The modal gyration diameter and its geometric standard deviation have been determined by log-normal fitting (Equation (10)) applied to the probability density functions.

The results presented in Table 3 show that the gyration diameter is sensitive to the storage environment. The observation that the gyration diameter decreases with increasing storage duration suggests that this evolution is more likely to be due to contraction of the primary particles than to collapse of the agglomerates. Nonetheless, no significant variations and no particular trends are observed for D_f as a function of storage duration and corresponding values, measured under ambient conditions, are in agreement with the initial values obtained for samples stored under nitrogen. Therefore, the decrease in the gyration diameter may not be assumed to be due to a collapse of the aggregate alone but is more strongly related to a contraction of the primary particles.

For D_pp, because of the associated uncertainties and standard deviation, the trend of its variation is not clear; however, a slight decrease is observed during the first 100 days, followed by an increase. On the other hand, as shown in Figure 5, a constant increase of the overlap coefficient as a function of the storage duration is observed for both samples in air. This increase appears to be a linear function of the storage duration and the slope is similar for the two different sampling methods, TPP and NFS. This indicates that the phenomenon is not mitigated by the specific treatment applied in the case of the NFS method (carbon deposition and chloroform dissolution). As the overlap coefficient increases, the primary particles coalesce with one another and this phenomenon explains the decrease in terms of gyration diameter. According to numerical studies (Sorensen and Roberts 1997; Brasil et al. 1999) as the overlap coefficient increases, the prefactor k_f is expected to increase. Nevertheless, for the present results, within the experimental uncertainties, the link between C_ov and k_f is not obvious and only a slight increase of the prefactor can be observed in Table 3.

FIG. 5 Evolution of the overlap coefficient as a function of the storage duration in air and nitrogen (three TEM micrographs are shown as an illustration of the increase in the overlap coefficient).

In parallel to the evolution of soot in ambient air, Table 3 and Figure 5 also present the changes in the morphological parameters as a function of the storage duration in the low humidity, nitrogen filled chamber. Under such inert conditions, the changes observed over time in ambient air are significantly mitigated, demonstrating that humidity, oxygen and others gases present in the atmosphere may have a major influence on the morphology and the inner microstructure of soot particles and monomers. Nevertheless, we should note that for storage under nitrogen, the first analysis was carried out after more than 100 days, so no conclusion can be drawn on any potential evolution of morphological parameters during this first period.

To our knowledge, only a few workers have reported such an evolution of the morphological parameters of soot particles for long duration storage under ambient conditions. Recently, Braun (2009) proposed a general model of ageing of soot in the atmosphere and suggested that, during the first weeks of soot ageing, PAH trapped on primary particles may be released and lost through desorption, decreasing the volume of primary particles. After this first step, Braun (2009) suggested that decomposition of the remainder of the soot particle may occur through chemical reaction with the air and humidity, forming a thin graphitic layer at the surface of the primary particles. This two-step model of desorption followed by oxidation of soot, may explain the initial decrease of the gyration and the primary particle diameters as observed in our study (Table 3). Furthermore, Popovitcheva et al. (2008) have described a “swelling” of the microstructure of soot due to water uptake under high relative humidity (more than 80%) and, as a consequence, an increase in the primary particle diameter. This “swelling” phenomenon is also observed in the present work for long storage duration (greater than 100 days) for which the primary particle diameter and the overlap coefficient increase.

5. CONCLUSIONS

The aim of the present work was to investigate the influence of the sampling methods and storage conditions on the morphological parameters (D_pp, D_g, D_f, k_f, and C_ov) of soot produced by a laminar ethylene diffusion flame and sampled using four different devices (TPP, TPS, IPS, NFS).

• Good agreement has been observed in the values from TPP, TPS, and IPS, which are based on sampling particles directly onto TEM grids. This indicates that the dilution induced upstream of TPS and IPS does not significantly change the morphological parameters of soot.

• Results obtained from direct sampling of soot onto TEM grids (TPP, TPS and IPS) show good agreement for D_pp, D_f, k_f, when compared to previous work.

• These results allow comparison of the reliability of these different procedures for sampling soot particles in different conditions (TPP for sampling in flames and TPS, IPS in more diluted conditions). However, a significant difference in aggregate size distributions was demonstrated, revealing different collection efficiencies.

• The transfer of soot from a Nuclepore membrane (NFS) to a TEM grid induces a significant overestimate of primary particle diameters by more than 10 nm and a corresponding increase in gyration diameters. This non negligible increase is due to the evaporated graphite film. At the same time, the overlap coefficient also seems to be affected by both the evaporated graphite film and the chloroform dissolution.

• The effects of storage in air or under inert gas and the effects of storage duration on morphological parameters have also been investigated. Further, the evolution of morphological parameters as a function of storage duration in air was determined for TPP and NFS procedures. With regards to the ageing of particles, a significant increase of the overlap coefficient was observed for ambient conditions, while D_pp only changed slightly. This effect was clearly reduced when particles were kept in a chamber filled with nitrogen, at a relative humidity of 15 % and a mean temperature of 20°C. On the other hand, the method used to measure D_f and k_f from TEM images does not reveal any significant dependency of the storage conditions on fractal law parameters. The observed soot contraction, due to ageing in ambient air, demonstrates the physical and chemical phenomena that occur under ambient conditions. This could be explained by PAH and transfer of O₂ and H₂O within the internal structure of the soot, in agreement with several reports in the literature. Nevertheless, further investigations (microstructure characterization by HRTEM and measurement of adsorption/desorption of PAH) must be undertaken in order to describe more precisely the evolution of morphological and structural properties of soot in the atmosphere.

In conclusion, in view of the results of this present work, it can be asserted that it is better to sample soot particles directly onto TEM grids, by thermophoresis or by diffusion or impaction/interception, rather than by filtration onto polycarbonate membranes. The samples must be stored in a cell filled with inert gas at low humidity and in the absence of light. Finally, TEM analysis of samples should be performed as soon as possible and not after more than two months of storage in ambient air. Lastly, it can be concluded that the different protocols are suitable for sampling particles produced by different combustion processes (gaseous burner, aircraft engines, diesel, fire, etc.). Nevertheless, with regards to atmospheric ageing, the present results must be applied with caution to other soot sources especially considering the large variability in physico-chemical composition of soot produced by different processes and conditions. Moreover, since the present work only aimed to determine sampling and storage influences on the morphological parameters, further investigations will now be conducted using high resolution TEM of the microstructure in the inner core and outer shell of soot particles, together with chemical analyses (EDX, FTIR), in order to improve our understanding of soot particle ageing under ambient conditions.

This work has been supported by the ANR program BLAN-06-346 (SOOT). The authors would like to thank M. Maugendre and L. Lucas from CORIA for their assistance during the sampling of soot particles and S. Ducourtieux from LNE for his participation in carbon deposit analysis.