Density measurement of fine aerosol fractions from wood combustion sources using ELPI distributions and image processing techniques

Fuel 88 (2009) 947–954

Contents lists available at ScienceDirect

Fuel journal homepage: www.elsevier.com/locate/fuel

Density measurement of fine aerosol fractions from wood combustion sources using ELPI distributions and image processing techniques Nicolas Coudray a,1, Alain Dieterlen a,1, Estelle Roth b,2, Gwénaëlle Trouvé c,* a

Laboratoire MIPS – Groupe Lab.El, I.U.T. de Mulhouse, 61, Rue Albert Camus, 68093 Mulhouse Cedex, France Groupe de Spectrométrie Moléculaire Atmosphérique (GSMA), Moulin de la Housse, BP 1039, 51687 Reims Cedex 2, France c Laboratoire de Gestion des Risques et Environnement, 25 Rue de Chemnitz, 68200 Mulhouse, France b

a r t i c l e

i n f o

Article history: Received 22 June 2007 Received in revised form 7 November 2008 Accepted 9 December 2008 Available online 4 January 2009 Keywords: ELPI Scanning electron microscopy Density Aerosol

a b s t r a c t Aerosols from combustion sources are of high concern since they present a risk for health and environment. Particle size distribution of aerosols and in particular number size distribution are easily and quickly obtained using an Electrical Low Pressure Impactor (ELPI). However, this technique is depending of aerosol density; q, which may lead to biased particle size distributions. Aerosol density from combustion sources is usually not well known and depends on several parameters. Aerosol density cannot be measured with usual methods since there is generally not enough matter collected on each stage of the ELPI. Our approach uses electronic microscopy to evaluate q at each impaction stage in order to increase the accuracy of the number size distributions resulting from the ELPI measurements. Particles were collected on glass substrates deposited on each impaction stages. Images were obtained using a scanning electron microscope and image processing tools were applied. This method was first tested with silica particles resulting from a combustion process which have a constant density found to be comprised between 2.2 and 2.4 g cm3 for stages 2 (57 and 95 nm) and 3 (95 and 158 nm), respectively. Once validated, this method was used to determine the density of wood combustion aerosols. The results match well for fly ashes from wood combustion with densities varying from 1.1 to 3.0 g cm3 for particles of mean equivalent diameter ranging from 69 to 157 nm, respectively. Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction Fine particles are likely to present a risk to environment and health because they may penetrate deeply into the respiratory tract. Particularly, very small particles with size below 1 lm in diameter are present in a large proportion in the particle size distributions and are thought to be responsible for adverse health effects associated to air-pollutant exposure [1]. These ultra fine particles get deposited in the alveolar parts of the lung where the absorption efficiency for trace elements is up to 60–80% [2– 4]. Therefore, in a context of environmental policy, it is useful to identify and characterize the sources of ultra fine particles. Cascade impactors are widely used for this purpose since they allow the particles collection and their measurement in the number size distribution. These impactors are used in various fields, e.g. particles emitted from diesel engines, ambient aerosols or other aerosols from combustion sources [5–11]. * Corresponding author. Tel.: +33 3 89 32 76 63; fax: +33 3 89 32 76 61. E-mail addresses: [email protected] (N. Coudray), [email protected] (A. Dieterlen), [email protected] (E. Roth), [email protected] (G. Trouvé). 1 Tel.: +33 3 89 32 76 65; fax: +33 3 89 42 32 82. 2 Tel.: +33 3 26 91 32 31; fax: +33 3 26 91 31 47. 0016-2361/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.fuel.2008.12.013

In this study, an Electrical Low Pressure Impactor (ELPI) manufactured by Dekati Ltd. (Tampere, Finland) was used to collect particles from 29 nm to 10 lm into 12 size fractions. Within the ELPI, the particles are first electrically charged according to their Stokes diameter, and then impacted on different stages according to their inertia related to their aerodynamic diameter. Finally, the induced current is measured. The number of particles depends on the induced current related both to the aerodynamic and the Stokes diameters. As will be shown later the conversion of the Stokes or equivalent diameter into the aerodynamic diameter needs to know the particle density. This parameter is needed for processing the data with the ELPI software. However, the value of the aerosol density is usually not precisely known so that no indication about density is generally provided by the ELPI technique [8–9] and/or a default value equal to 1 g cm3 is admitted [10–12]. Hence, it is not surprising that several authors draw attention on the fact that the number size distribution data have to be considered with some caution due to the lack of precise density values [12]. Particle density depends on several parameters. Parametric conditions in combustion processes (i.e., temperature, gas flow rate, treatments of flue gas and oxygen fraction) have a strong influence on the nature of particles emitted (mineral fly and bottom ashes resulting from char or tar combustion and/or pyrolysis)

948

N. Coudray et al. / Fuel 88 (2009) 947–954

and therefore on their density. Also, the type of fuel (municipal solid waste, hazardous waste, biomass, industrial sludge and coal) also influences the chemical composition of aerosols and thus their density. For instance, Rau showed that in the case of a wood stove, the contribution of the organic carbon to the total particle mass is equal to 14% for burning conditions in an excess of air supply whereas it reaches 57% in a shortage of air supply [13]. The losson-ignition, LOI, which measures the amount of the unburned carbon remaining in the fly ash, is one of the most significant chemical properties of ash samples. Compared to the combustion of coal which gives fly ashes with LOI values in the range 10–50%, fly ashes resulting from pure lignite and wood chips combustion exhibit lower LOI values close to 2% [14]. Schneider demonstrated that the dynamic behaviour of compacted particles from biomass combustion (oak) suggested limits of bulk density in the range of 1.4– 2.0 g cm3 [15]. An upper limiting value equal to 2.0 g cm3 was previously reported for carbon black and might be representative for the density of particles resulting from biomass burning [16]. Bottom ashes from Municipal Solid Waste Incineration are mainly composed of minerals like silica, calcite, tringite and contain alkaline, alkaline-earth and heavy metals. This mineral composition of bottom ash increases the particle density up to 2.0–2.5 g cm3 [17– 19]. The presence of a soluble organic fraction containing aliphatic and aromatic hydrocarbons up to 20% in diesel particles leads to lower density values [20–22]. The density also changes according to the particle diameter [23–25]. As an example, Mamakos et al. found that the effective density varies from 1.2 to 0.2 g cm3 for Euro 3 engine exhaust with mobility diameters ranging from 30 to 300 nm [25]. To our knowledge density vs. particle diameter of aerosols resulting from wood combustion was not established as yet. Indeed, the density q of an aerosol collected via an ELPI on each impaction stage cannot be easily measured with classical experimental tools (like pycnometer), since there is not enough matter collected on each stage of the impactor in the exhaust during the combustion process. Some previous studies combine two steady flow instruments either in series or in parallel in order to obtain a density value. Ristimäki et al. [26] or more recently Maricq and Xu [27] provide a good review of such combinations [26]. In particular, Kelly and McMurry were the first ones who proposed a differential mobility analyzer (DMA) followed by single stages of the MOUDI impactor [28]. The DMA–LPI (Low Pressure Impactor) or ELPI series were also proposed by some authors [27,29,30]. The effective density and fractal dimension can also be measured using a DMA and an aerosol particle mass analyzer [31]. Recently, a new particle mass classifier was developed and was called the ‘Couette centrifugal particle mass analyzer’ [32]. Also, the density can be obtained by fitting ELPI and Differential Mobility Particle Sizer (DMPS) [23] simultaneous measurements or ELPI and SMPS (Scanning Mobility Particle Sizer) [18,26]. Most of these works were devoted to soot characterization, i.e. effective density and fractal dimension. An overall density equal to 2.0 g cm3 was obtained over the whole range of particle size of an aerosol resulting from wood combustion by using in parallel SMPS and ELPI techniques [18]. These methods imply the knowledge of calibration curves [33] and, as a matter of fact, to have access to both technologies. SEM or TEM image processing is widely used to study the particle morphology and in particular for determining the fractal dimension of soot aggregates [34,35]. TEM image processing was also used by Park et al. for the determination of the volume of agglomerated diesel particles which masses were preselected by using an aerosol particle mass analyzer. Density of diesel exhaust particles increases from 1.27 to 1.78 g. cm-3 as particle mobility size increases from 20 to 220 nm. When particles were preheated in order to remove the volatile organic fraction, their density was found to be close to 1.78 g cm3, whatever their size. This density

measured after heating correspond to inherent material density do diesel soot [16]. Only few images of wood combustion particles are available [30,36,37]. For determining the density, SEM was previously coupled with particle settling velocity measurements or with a cascade impactor [38–39]. The work presented herein combines ELPI data with SEM images processing to evaluate the particle density. This method is based on the relation between the density, the Stokes (or equivalent) diameter and the aerodynamic diameter. The aerodynamic diameter is known from the ELPI since each stage is characterized by its lower cut-off aerodynamic diameter. As for the equivalent diameter, it is estimated by SEM image analysis. In a first part, the ELPI limitations are depicted. Then, the protocol for determining the density using image processing tools is set up using a silica aerosol of well known and constant density all over the size range. Finally, the validity of the method is discussed and critically applied to the determination of the density of tar and fly ashes from biomass and wood waste combustion. 2. Experimental section 2.1. Formation and collection of aerosols Two types of aerosols were collected on glass substrates (22 mm in diameter) purchased from ROTH:  A mineral polydispersed aerosol was produced by flowing 20 mg of SiO2 from SOVITEC in a tubular reactor. The electrical impactor was connected to the end of the reactor and the silica aerosol was collected for 30 min. This experiment was done twice and the resulting samples will be called ‘experiment 1’ and ‘experiment 2’, respectively. The density of silica was previously measured by pycnometry using 2-propanol and was found equal to 2.5 ± 0.4 g cm3.  An aerosol from natural beech wood combustion was also collected for a period of 1 h from a residential wood fireplace purchased from FONDIS SA with the ELPI connected on the chimney. 2.2. Electrical low-pressure impactor (ELPI) limitations The ELPI is made of three parts: the corona charger in which the particles are electrically charged, the stages on which the particles impact according to their inertia and the pump which samples the aerosol at constant flow rate of 9.82 l min1. Finally, the size distribution is evaluated from the induced current of the pre-charged particles [40,41]. The charged particles induce an electrical current when impacted on a stage. Each stage is electrically insulated with Teflon and a real-time measurement of the current I is obtained by electrometers. This current is then converted into a size distribution of the aerosol. The distributions (number size, mass size, diameter size, etc.) depend on the conversion of the measured current to a concentration of particles. The relation between the concentration Ci of particles per stage i and the current I is given by the following Eq. (1):

Ci ¼

I PneQ

ð1Þ

where Ci is the number concentration of particles per second of sampling (cm3 s1), P the penetration expressed through the charger, n the number of charges per particle, e is the charge of an electron and Q is the flow rate (9.82 l min1). The factor P  n characterizes the charger efficiency and depends on the Stokes or

949

N. Coudray et al. / Fuel 88 (2009) 947–954

equivalent diameter deq which in the ELPI is related to Eq. (2) [28,42,43]:

dae deq ¼ rffiffiffiffiffiffiffiffiffiffiffi

3.E+07

ð2Þ

in which dae is the diameter of a particle having a density of one and the same velocity than the real particle and q0 is the standard density (1 g cm3). The aerodynamic cut-off diameter of the particles for a considered stage is characteristic of the ELPI (Table 1). Cdx (x = eq or ae) are the Cunningham slip-correction factors [41,43]. DeCarlo [42] proposed the following equation between the volume equivalent diameter dve and the aerodynamic diameter dae (Eq. (3)):

dae

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 q  C dve ¼ dve v q0  C dae

2.3. Method and correction protocol The goal of the control process is to enhance the determination of the particle size distribution analysis by estimating q on each

ρ=1.1 ρ=1.7 ρ=3

2.E+07

corrected distribution

1.E+07

5.E+06

0.E+00 0.01

ð3Þ

in which v is a dynamic shape factor for taking into account a correction for non-spherical particles. This factor is equal to 1 for spherical particles and for compact aggregates and larger than 1 in other cases [42]. For spherical particles, Eq. (3) becomes equivalent to Eq. (2) with dve corresponding to deq. As a result, deq can be directly determined by measuring the diameter of spherical particles on the images of impacted particles on a glass substrate. In our case the studied nanoparticles are mostly spherical (at least 70% of the particles have a form factor above 0.7). Consequently, Eq. (2) is applicable for silica particles. Taking into account the value of the measured density (2.5 ± 0.4 g cm3) one may determine the effect of the density on the values of dae and deq and the results are shown in Table 1. The ELPI technology also implies that while the aerosol separation into size fractions depends on aerodynamic characteristics, the measured current depends on density q which is usually unknown. The number size distributions of aerosol from natural wood beech combustion tests were performed in our laboratory. The Fig. 1 shows the evolution of the number size distributions with the density. The number size distribution shown in Fig. 1 is typical of an aerosol emitted from combustion sources with at least 99% of the number of particles having a diameter below 1 lm [18]. The curves in Fig. 1 indicate that the increase of the particle density from 1 to 1.1 and 3 g cm3 enhances the number of particles by 9% and 380%, respectively. The density is therefore a relevant parameter but, it cannot be easily measured with classical experimental tools (like pycnometer) since not enough matter is collected on each stage of the impactor during the combustion process. Another limitation of the method arises from the fact that a constant density value over the whole range is considered in the software of the equipment [43,43]. Indeed, q may vary with the size according to the sampled aerosol and its chemical compounds. The imprecise and global knowledge of the density q will induce errors in the evaluation of deq and consequently in the characterization of the aerosol and number size distribution (Fig. 1).

ρ=1

2.E+07 N particles/Dlogdeq

qCdeq q0 Cdae

0.1 deq (µm)

1

Fig. 1. Influence of the density on number size distributions for natural beech wood combustion aerosol.

ELPI stage. The aerosol is sampled by the ELPI and the particles are collected on glass substrates. The glass substrates avoid inhomogeneous background in the images. This problem occurs with other substrates such as quartz or Teflon. In addition, particles are more likely to conserve their initial shape. On the contrary, the particles are likely to condense around fibres when impacted on fibrous substrates. Images were obtained using a scanning electron microscope which resolution allows a precise measurement of sizes. The gold-covered substrates are then placed in a SEM for image acquisition. The SEM used is an FEI model Quanta 400 with a resolution close to 10 nm in standard conditions. Finally, the measure of the equivalent diameter deq is obtained via the image processing tools of ImageJ software [44]. The results are combined with the aerodynamic diameter dae of the studied stage to calculate the density q. This value is then input into the ELPI software to correct the size distribution according to Eq. (2). 3. Results and discussion 3.1. Image processing – measure of the equivalent diameter The diversity and the complexity of the images led us to set up adapted processes to treat two kinds of images: (i) piled-up particles and (ii) spread-out particles (Fig. 2). In the second case, an automated process made up of the four following steps and illustrated in Fig. 3 was set-up:     o

pre-treatment to reduce the noise, segmentation: discrimination of the particles post-treatment: separation of close particles. Measures: Equivalent diameter deq

deq ¼ 2 

rffiffiffiffiffiffiffiffiffiffiffi Area

ð4Þ

p

Table 1 Equivalent diameter versus apparent density for silica aerosol.

Cut-off diameter for 50% efficiency q = 2.1 q = 2.5 q = 2.9

Stage

1

2

3

4

5

6

7

8

9

10

dae (lm) deq (lm) deq (lm) deq (lm)

0.029 0.014 0.012 0.010

0.057 0.029 0.025 0.022

0.095 0.050 0.044 0.038

0.158 0.089 0.078 0.070

0.264 0.159 0.141 0.127

0.384 0.240 0.215 0.196

0.616 0.398 0.361 0.332

0.953 0.627 0.573 0.529

1.610 1.075 0.985 0.914

2.400 1.615 1.482 1.377

950

N. Coudray et al. / Fuel 88 (2009) 947–954

Fig. 2. Piled-up silica particles on stages 2 (on the left) and 8 (on the right).

Fig. 3. Example of an automated treatment (from left to right): pre-treated image; segmented image; post-treated image; labels of the measured particles.

Considering aliasing phenomena, the uncertainty of the measurements was estimated to be equal to 6% for deq. It leads to a 12% error on the estimated value of q. o Circularity FF A form factor of the particle, FF, is used to characterize the sphericity of an object and is defined as follows:

FF ¼

4  p  Area 2

Perimeter

ð5Þ

For FF = 1, the particle is spherical. The lower the FF value is, the more elongated the particle is. If the 2D image of the particle is circular then the particle in 3D is likely to be spherical and the equivalent diameter measured on the 2D image of the particle may be considered as the volume equivalent diameter (dve in Eq. (3)). If the segmented object is not spherical, then the deq is not considered for the calculation of q. Only objects with a FF value equal or larger than 0.8 were considered for the calculation of the mean equivalent diameter deq and its standard deviation. Other objects are likely to be poorly segmented in particular non circular or piled-up particles. As stated before, non circular particles represent a small amount of impacted particles. Piled-up particles are supposed to be the result of particles impacted on the same area, rather than aggregates formed before impaction. For instance, Fig. 3 shows the image processing protocol for:  Two hundred and thirteen measured particles (particles with less than 50 pixels were rejected since they either originate from noise artefacts or are too small in size) and 174 have an FF value lower than 0.8.  The average equivalent diameter of the 39 remaining particles consists in 6 pixels (34 nm) with a standard deviation of 3.4 pixels (9 nm).  For piled-up particles, the segmentation and the post-treatment were granted by a semi-automated procedure based on the active contour segmentation called ‘‘Snake” [45].

The benefit of this procedure is to test a new substrate: glass substrates allow the impaction of the particles with the ELPI and suit well for SEM observations. For a given plate, according to this procedure, the area of the particles was measured and the equivalent diameter was deduced. A size distribution of the particles impacted on each ELPI stage was established. The distribution is then fitted to a Gaussian distribution from which the mean value was calculated. 3.2. Impaction/experimental protocol and results: silica, tar and fly ashes particles from wood combustion The method was first tested on polydispersed Silica particles of known density (q = 2.5 ± 0.4 g cm3). Two experiments with Silica were performed: in the first one, small particles were collected (up to stage 4, Fig. 4) whereas in the second one larger particles were collected (from stages 4 to 10, Fig. 5). 3.2.1. Calculation of the expected values of the diameter of silica particles impacted on each stage (deq) Knowing the silica particles density (q = 2.5 ± 0.4 g cm3), the aerodynamic lower cut-off diameter, dae, which is specific for the ELPI (Table 1), the Cunningham coefficient and the expected deq are calculated using Eq. (2). A comparison between the aerodynamic cut-off diameters and the calculated equivalent diameters is shown in Table 2. For instance, the diameter of the silica particles impacted on stage 2 should vary from 22 up to 50 nm if their density stands within the interval 2.5 ± 0.4 g cm3. However, since the cut-off diameter is given for 50% efficiency, some particles collected are bound to be slightly below and above these values. Accordingly, the size distribution of silica particles on a stage will be established using SEM. 3.2.2. Treatment of experience 1 The number distribution given by the ELPI for a density q equal to 2.5 g cm3, is shown in Fig. 4 in which each bin corresponds to a

951

N. Coudray et al. / Fuel 88 (2009) 947–954 5.00E+11

% of measured particules

35%

4.50E+11

N particles/Dlogdae

4.00E+11

Stage 1

3.50E+11 3.00E+11 2.50E+11 2.00E+11 1.50E+11

Stage 2

1.00E+11

ρ=2,5 g/cm3

ρ -15%

30% 25% 20% 15% 10% 5% 0%

5.00E+10

10 15 20 25 31 38 44 48 52 56 60 64 68 72 76 80

0.00E+00 0.10

1

1.0

Fig. 4. Number size distribution of silica from ELPI measurements – experience 1.

3.00E+08 2.50E+08

dae ¼

2.00E+08 1.50E+08 1.00E+08 5.00E+07 0.00E+00 0.01

0.1

1

10

dae (μm) Fig. 5. Number size distribution of silica from ELPI measurements – experience 2.

Table 2 Results of the analysis on silica particles collected during experiences 1 and 2.

dae (nm) deq (nm) Correlation factor (%) q (g cm3)

Fig. 6. Number size distribution of silica particles impacted on stage 2 (1098 measures).

stage 2, deq = 38 nm, is used to compute the density for Eq. (2). The aerodynamic diameter dae is taken as the mean of the upper and lower cut-off aerodynamic diameters of the considered stage as follows:

3.50E+08

Stage

deq (nm)

10

μ dae (μm)

N particles /DlogDp

ρ à +15%

Experience 1

Experience 2

2

3

6

7

8

9

10

76 38 99 2.2

127 62.4 94 2.4

500 800 85 0.4

785 858 81 0.8

1282 834 97 2.2

2005 1118 93 2.9

3205 1771 95 2.2

stage (from left to right: stage 1, stage 2. . .stage 10, stage 11 and stage 12). In this experience, particles were mainly collected on stages 1, 2 and 3. However, only particles on stages 2 and 3 were studied since the resolution of the SEM is not sufficient for stage 1 (the value of deq being expected less than 25 nm). The SEM image corresponding to stage 2 enables to distinct 1098 particles. They are classified into 13 size fractions according to the precision measurement of diameters with ImageJ tools (Fig. 6). The three more important bins with horizontal lines represent the particles of size within the expected range for q = 2.5 g cm3 (from 25 to 44 nm – see Table 1), corresponding to 74% of the measured particles. Considering a density variation equal to 15%, 86% of the particles are within the expected equivalent diameter range (22– 50 nm, Table 1). Using a reverse approach, the distribution of Fig. 6 may be used to determine the mean density q of particles impacted on stage 2. The fitting of this repartition using a Gaussian distribution leads to a good approximation of the mean diameter, 38 nm. Thus, the average equivalent diameter of the particles impacted on

dc2 þ dc3 57 þ 95 ¼ 76 nm ¼ 2 2

ð6Þ

in which dc2 is the lower cut-off diameter of stage 2 and dc3 is the lower cut-off diameter of stage 3 (Table 1). It comes from Eq. (2) that the average density of the particles is equal to 2.20 g cm3. This value stands in the range defined by the experimental pycnometer measures (2.5 ± 0.4 g cm3). The same reasoning is applied to the particles measured on stage 3 where 94% of the particles stand in the expected range, from 39 to 80 nm (Table 1). The equivalent diameter of the associated Gaussian distribution is equal to 62.4 nm. Since the cut-off aerodynamic diameters of stage 3 are equal to 95 and 158 nm (mean dae = 127 nm), the density calculated using Eq. (2) is equal to 2.4 g cm3. This value confirms the potential of the method since it is in agreement with the value found by pycnometry (2.5 ± 0.4 g cm3). 3.2.3. Treatment of experience 2 The second experiment done with Silica permitted an analysis of the distribution of larger particles (Fig. 5). The images of the particles impacted on the upper stages were acquired with a SEM. The size range of the particles impacted on stages 6–10 is extended from nanometric particles up to micrometric ones. This result substantiates the loss of fine particles occurring on the upper stages [46]. The developed method was also used to extract deq and evaluate q for the particles of the upper stages. The results are shown in Table 2. The density values corresponding to the expected values are in bold. For stage 9, the distribution looks like a lognormal one and is shifted toward lower sizes, while deq is still the mean value of a Gaussian distribution. This may explain why the density is slightly higher (2.9 g cm3). Moreover, the correlation factor is slightly lower than for the first experiment. The fact that these distributions are so wide probably means that the stages 6 and 7 are affected by the loss of fine particle and by the rebounds. Stages 8–10 are probably as well affected by these losses too, but in that case the particles were too small as compared to the measured ones. Consequently, the test done on silica particles demonstrates that this method enables the determination of the density of the particles when the Pearson correlation factor r is higher than 94%. 3.2.4. Application to wood combustion experiments Once validated, the method was applied to an aerosol collected on the ELPI stages during the combustion of natural beech wood

952

N. Coudray et al. / Fuel 88 (2009) 947–954

logs. About 99.9% of particle numbers have diameters below 1 lm which is the diameter of inhalable particles (Fig. 1). Thus, this work particularly focused on stages 2, 3 and 5 for SEM analyses and image processing of particles. (see Figs. 7 and 8 for stage 2). As shown in Table 3, the density of the aerosol q1 resulting from image processing and calculated according to Eq. (2) exhibits a large variation all along the impaction stages. This result underlines that the formation of particles in the exhaust during a combustion process occurs with different sizes related to different densities. The density increases with increasing diameter, since small particles result from nucleation of unburned organic carbon with low densities and larger particles are mostly constituted by mineral fly ash [30]. Number (N1) and mass (M1) of particles were calculated with the software of the ELPI and taking values for the density q1 extracted from image processing for stages 2, 3 and 5, respectively. They were compared to those obtained with the usually reference density q2 of 1.0 g cm3 (N2 and M2). As shown in Table 3, the influence of density on the mass is low with a relative difference between M1 and M2 below the precision of the technique. Indeed,

Table 3 Influence of the density on the number and mass of particles calculated by the ELPI.

Mean dae (nm) Mean deq (nm) q1a (g cm3) N1: number of particles for q1 N2: number of particles for q2b = 1.0 g cm3 Relative difference between q1and q2 (%) Relative difference between N1 and N2 (%) M1: mass of particles collected for q1 (mg) M2: mass of particles collected for q2 (mg) Relative difference between M1 and M2 (%) a b

Stage 2

Stage 3

Stage 5

76 69 1.1 3.0  1012 2.5  1012 11 16 0.53 0.52 2

126 82 1.7 6.9  1012 3.5  1012 43 50 3.2 3.3 3

324 157 3.0 1.1  1012 4.3  1012 67 63 6.7 7.2 7

q1 was calculated according the Eq. (2) resulting from image processing. q2 is related to a spherical particle of density 1 g cm3.

the increase of the density by a factor 3 will lead to an overestimation of the mass of the particles is close to 7%. On the opposite, the influence of the density on the number of particles calculated by the software on each stage is significant: the relative difference between q1 and q2 is of the same order of magnitude than the relative one between N1 and N2 for the three impaction stages, respectively. Finally, the number size distribution is corrected (bold line in Fig. 1) according to the density determinations. The distribution is worth to be corrected since the maximum is twice higher with the corrected density than with the default value of 1 g cm3.

4. Conclusions

Fig. 7. SEM images of fly ash particles impacted on stage 2 during wood combustion.

25.00%

% of particles

20.00%

15.00% Gaussian fitting 10.00%

5.00%

0.00% 15

30

45

60

75

90 105 120 135 150 165 180 195 210 225

deq (nm) Fig. 8. Number size distribution of fly ash particles on stage 2 produced by wood combustion.

In this paper a method is proposed to evaluate the density of size fractions of aerosols emitted by combustion sources and sampled with an Electrical Low Pressure Impactor. When the size distributions are performed with the ELPI in real life-time measurements which is of considerable advantage in aerosols studies they may however be biased if the density of each size fraction is not well known. Indeed, the chemical compositions of different size fractions may induce a significant change in density. Hence, it is essential to determine the precise density value for the determination of the real size distributions of an aerosol (mass, number, etc.) from impactor measurements. A new substrate for collecting the aerosol was used in this study, i.e. a glass substrate. This substrate allows the impaction of the particles with the ELPI and is well suited for SEM observations. Treatment of the SEM images was easily performed with classical image processing tools. Several images could be treated to extract an equivalent diameter distribution. The average equivalent diameter was extracted by fitting a Gaussian function to the distribution for one studied stage, and the correlation factor validated the Gaussian model. The protocol to determine the effective density was successfully tested with spherical silica particles. The main benefit concerns the evaluation of the density of the aerosol depending of its size. According to particle circularity observed by SEM and previously determined fractal dimension [47], wood combustion particles analyzed with an ELPI does not present restriction encountered in soot particles studying [48]. Experiments from natural wood logs combustion in a domestic stove demonstrate a large variation of the density of the particles (1.1–3.0 g cm3) emitted in the exhaust related to their diameter. These results indicate that the mean density value of wood combustion particles is close to 2 g cm3 as determined previously [18,30]. With regards to the environmental regulations concerning pollutant emission factors, the number size distribution of particles

N. Coudray et al. / Fuel 88 (2009) 947–954

emitted in the exhaust could be now calculated with a better accuracy by using the real density for each impaction stage of the ELPI. These results are especially available for the lower and upper stages of the ELPI. However, when the size of the particles is widespread, all the particles could not be acquired on a same picture because of the lack of resolution of the microscope. This problem mostly occurs for the upper stages leading to a biased distribution. When a wide range could be observed and measured, the result was still biased if the differentiation between the lost and impacted particles could not be done. Hence, taking into account particle loss may lead to a better approximation of q. However, in the present case, this phenomenon is not occurring since the whole wood aerosol particles were located on the five first stages. Indeed, rebounds and fine loss particles are well known as being limitations of the ELPI. The distribution may then not only be wide, but also bimodal (especially for the intermediate stages), which is inappropriate for a Gaussian model. In other cases, a lognormal distribution was observed. Then, another model to fit the distribution has to be considered to enhance the accuracy of the method. Nevertheless, the developed method permits an evaluation of q, even though the variation for its value may be quite large. This variation may probably be reduced by taking into account the particle loss and by considering other models for fitting the distribution curves.

[15]

[16]

[17] [18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

[26]

Acknowledgements Authors whish to thank FONDIS companies for their kind technical contributions.

[27]

[28]

References

[29]

[1] Becker S, Mundandhara S, Devli RB, Madden M. Regulation of cytokine production in human alveolar macrophages and airway epithelial cells in response to ambient air pollution particles: further mechanistic studies. Toxicol Appl Pharmacol 2005;207:269–75. [2] Linlay WH, Stapleton KW, Zuberbulher P. Fine particle fraction as measure of mass depositing in the lung during inhalation of nearly isotonic nebulized aerosols. J Aerosol Sci 1997;28:1301–9. [3] Renwick LC, Donaldson K, Clouter A. Impairment of alveolar macrophage phagocytosis by ultrafine particles. Toxicol Appl Pharmacol 2001;172:119–27. [4] Becker S, Devli RB, Madden M. Regulation of cytokine production in human alveolar macrophages and airway epithelial cells in response to ambient air pollution particles: further mechanistic studies. Toxicol Appl Pharmacol 2005;207:269–75. [5] Tsukamoto, Y., Goto, Y., Odaka, M., 2000. Continuous measurement of diesel particulate emissions by a electrical low-pressure impactor. SAE Technical Paper Series 2000-01-1138. [6] Shi JP, Khan AA, Harrison RM. Measurements of ultrafine particle concentration and size distribution in the urban atmosphere. The Science of the Total Environment 1999;235:51–64. [7] Lavta-Somppi J, Moisio M, Kauppinen E, Valmari T, Ahonen P, Keskinen J. Aerosol formation in fluidized bed incineration with waste sludge. J Aerosol Sci 1998;29:461–80. [8] Xiaohong Y, Ngai TL, Chak KC, Ming F. Size distributions and condensation growth of submicron particles in on-road vehicle plumes in Hong Kong. Atmos Environ 2007;41:3328–38. [9] See SW, Balasubramanian R. Risk assessment of exposure to indoor aerosols associated with Chinese cooking. Environ Res 2006;102:197–204. [10] Wierzbicka A, Lillieblad L, Pagels J, Strand M, Gudmundsson A, Gharibi A, et al. Particle emissions from district heating units operating on three commonly used biofuels. Atmos Environ 2005;39:139–50. [11] Geresdi I, Meszaros E, Molnar A. The effect of chemical composition and size distribution of aerosol particles on droplet formation and albedo of stratocumulus clouds. Atmos Environ 2006;40:1845–55. [12] Kinsey JS, Mitchell WA, Squier WC, Linna K, King FG, Logan R, et al. Evaluation of methods for the determination of diesel-generated fine particulate matter: physical characterization results. J Aerosol Sci 2006;37:63–87. [13] Rau JA. Composition and size distribution of residential wood smoke particles. Aerosol Sci Technol 1989;10:181–92. [14] Koukouzas N, Hämäläinen J, Papanikolaou D, Tourunen A, Jänti T. Mineralogical and elemental composition of fly ash from pilot scale fluidised

[30]

[31] [32]

[33] [34]

[35]

[36]

[37]

[38] [39] [40] [41] [42]

[43] [44] [45]

953

bed combustion of lignite, bituminous coal, wood chips and their blends. Fuel 2007;86:2186–93. Schneider J, Weimer S, Drewnick F, Borrmann S, Helas G, Gwaze P, et al. Mass spectrometric analysis and aerodynamic properties of various types of combustion related-aerosol particles. Int J Mass Spectr 2006;258:37–49. Park KD, Kittelson B, Zachariah MR, Mac Curry PH. Measurement of inherent material density of nanoparticle agglomerates. J Nanoparticles Res 2004;6: 267–72. Matsunuga T, Kim JK, Hardcastle S, Rohatgi PK. Crystallinity and selected properties of fly ash particles. Mater Sci Eng 2002;A325:333–43. Johansson, L.S., 2002. Characterisation of particle emissions from small-scale biomass combustion. Thesis of Degree of Licentiate of Engineering, Chalmers University of Sweden. Forteza R, Far M, Segui C, Cerda V. Characterization of bottom ash in municipal solid waste incinerators for its use in road base. Waste Manage 2004;24: 899–909. Collura S, Chaoui N, Azambre B, Finqueneisel G, Heintz O, Krzton A, et al. Influence of the soluble fraction on the thermal behaviour, texture and surface chemistry of diesel exhaust soot. Carbon 2005;43:605–13. Kittelson DB, Watts WF, Johnson JP. On-road and laboratory evaluation of combustion aerosols—Part 1: Summary of diesel engine results. J Aerosol Sci 2006;37:913–30. Kittelson DB, Watts WF, Johnson JP, Schauer JJ, Lawson DR. On-road and laboratory evaluation of combustion aerosols—Part 2: Summary of spark ignition engine results. J Aerosol Sci 2006;37:931–49. Ahlvik, P., Ntziachristos, L., Keskinen, J., Virtanen, A., 1998. Real time measurements of diesel particle size distribution with an electrical low impactor. Society of Automotive Engineers, SAE Technical, Paper No. 980410. Maricq MM, Podsiadlik DH, Chase RE. Size distributions of motor vehicle exhaust PM: a comparison between ELPI and SMPS Measurements. Aerosol Sci Technol 2000;33:239–60. Mamakos A, Ntziachristos L, Samaras Z. Evaluation of the Dekati mass monitor for the measurement of exhaust particle mass emissions. Environ Sci Technol 2006;10:4739–45. Ristimäki J, Virtanen A, Marjamäki M, Rostedt A, Keskinen J. On-line measurement of size distribution and effective density of submicron aerosol particles. J Aerosol Sci 2002;33:1541–57. Maricq MM, Xu N. The effective density and fractal dimension of soot particles from premixed flames and motor vehicle exhaust. J Aerosol Sci 2004;35:1251–74. Kelly WP, McMurry PH. Measurement of particle density by inertial classification of DMA-generated monodisperse aerosols. Aerosol Sci Technol 1992;17:199–221. Schleicher B, Künzel S, Burtscher H. In-situ measurement of size and density of submicron aerosol particles. J Appl Phys 1995;78:4416–22. Hueglin CH, Gaegauf CH, Künzel S, Burtscher H. Characterization of wood combustion particles: morphology, mobility, and photoelectric activity. Environ Sci Technol 1997;31:3439–47. Park K, Cao F, Kittelson DB, McMurry PH. Relationship between particle mass and mobility for diesel exhaust particles. Environ Sci Technol 2003;37:577–83. Olfert JS, Symonds JPR, Collings N. The effective density and fractal dimension of particles emitted from a light-duty diesel vehicle with a diesel oxidation catalyst. Aerosol Sci 2007;38:69–82. Marjamäki M, Lemmetty M, Keskinen J. ELPI response and data reduction I: Response functions. Aerosol Sci Technol 2005;39:575–82. Sachdeva K, Attri ak. Morphological characterization of carbonaceous aggregates in soot and free fall aerosol samples. Atmos Environ 2008;42: 1025–34. Van Gulijk C, Marijnissen JCM, Makkee M, Moulijn JA, Schmidt-Ott A. Measuring diesel soot with a scanning mobility particle sizer and an electrical low-pressure impactor: performance assessment with a model for fractal-like agglomerates. Aerosol Sci 2004;35:633–55. Lind T, Valmari T, Kauppinen EI, Sfiris G, Nilsson K, Maenhaut W. Volatilization of heavy metals during circulating fluidized bed combustion of forest residue. Environ Sci Technol 1999;33:496–502. Gwaze P, Schmid O, Annegarn HJ, Andreae MO, Huth J, Helas G. Comparison of three methods of fractal analysis applied to soot aggregates from wood combustion. J Aerosol Sci 2006;37:820–38. Van Hout R, Katz JA. Method for measuring the density of irregularly shaped biological aerosols such as pollen. Aerosol Sci 2004;35:1369–84. Barthel H, Heinemann M, Stintz M, Wessely B. Particle sizes of fumed silica. Chem Eng Technol 1998;21:745–52. Virtanen A, Marjamäki M, Ristimäki J, Keskinen J. Fine particle losses in electrical low-pressure impactor. J Aerosol Sci 2001;32:389–401. Marjamäki M, Keskinen J, Chen DR, Pui DYH. Performance evaluation of the electrical low-pressure impactor (ELPI). J Aerosol Sci 2000;31:249–61. DeCarlo PF, Slowik JG, Worsnop DR, Davidovits P, Jimenez L. Particle morphology and density characterization by combined mobility and aerodynamic diameter measurements. Aerosol Sci Technol Part 1: Theory 2004;38:1185–205. Moisio, M., 1999. Real time size distribution measurement of combustion aerosols. PhD of the Tampere University, Publications 279, Finland. Image Processing and Analysis in Java. . Boudier T. Elaboration d’un modèle de déformation pour la détection de contours aux formes complexes. Innov Technol Biol Med 1997;18:1–18.

954

N. Coudray et al. / Fuel 88 (2009) 947–954

[46] Virtanen A, Ristimäki J, Vaaraslahti KA, Keskinen J. Effect of engine load on diesel soot particles. Environ Sci Technol 2004;38:2551–6. [47] Slowik JG, Stainken K, Davidovits P, Williams LR, Jayne JT, Kolb CE, Worsnop DR, Rudich Y, DeCarlo PF, Jimenez JL. Particle morphology and density characterization by combined mobility and aerodynamic diameter

measurements. Part 2: Application to combustion-generated soot aerosols as a function of fuel equivalence ratio. Aerosol Sci Technol 2004;38:1206–22. [48] Van Gulijk C, Schouten JM, Marijnissen JCM, Makkee M, Moulijn JA. Restriction for the ELPI in diesel particulate measurements. J Aerosol Sci 2001;32: 1117–30.