Effective Density and Volatility of Particles Emitted from Gasoline Direct Injection Vehicles and Implications for Particle Mass Measurement

Abstract

The effective density and volatility of particulate emissions from five gasoline direct injection (GDI) passenger vehicles were measured using a tandem differential mobility analyzer (DMA) and centrifugal particle mass analyzer (CPMA) system. The measurements were conducted on a chassis dynamometer at three steady-state operating conditions. A thermodenuder was employed to find the volatility and mixing state of the particles as well as the effective density of nascent and non-volatile particles (defined as particle phase remaining after denuding at 200°C). The mass–mobility exponent ranged between 2.4 and 2.7 for nascent (or undenuded) particles and between 2.5 and 2.7 for non-volatile particles; higher than typical diesel soot. The effective density function was 4278d_m^−0.438 ± 76.3 kg/m³ (for mobility diameter, d_m, in nm) for nascent particles and 3215d_m^−0.395 ± 37.9 kg/m³ for non-volatile particles. The effective density functions of the non-volatile particles were fairly similar for the conditions studied. The uncertainty in using the effective density and mixing state data to determine the mass concentration of the aerosol by integrating mobility size distributions was examined. The uncertainty in mass concentration is minimized when only the non-volatile component is measured. However, the uncertainty in the mass concentration increases substantially if nascent particles are measured due to uncertainties in the particle mixing state and their associated effective densities. Furthermore, transient vehicle operation (cold-starts, accelerations, and decelerations) would likely change the mixing state of the exhaust particles suggesting it is difficult to accurately measure the mass concentration of undenuded GDI exhaust particulate using integrated size distribution methods.

Copyright 2015 American Association for Aerosol Research

1. INTRODUCTION

Gasoline direct injection (GDI) engines are widely used because of their higher specific power output and fuel economy compared to conventional port injection gasoline engines (He et al. 2012). According to the European commission, 35% of vehicles will be gasoline direct injection vehicles by 2020 (Mamakos 2011). For some GDI engines, up to 20% improvement in fuel economy compared to port fuel injection (PFI) engines has been reported (Mamakos 2011). However, GDI engines produce more particulate emissions than PFI engines in terms of both particle number and mass (Zhao et al. 1999).

It has been shown that particle emissions can affect the climate by scattering and absorbing solar radiation (Boucher et al. 2013). Particles can also indirectly affect the climate by interacting with clouds (Albrecht 1989; Giordano et al. 2015). They can also penetrate deeply into body organs and remain in the body for a long time (Balasubramanian et al. 2010). It has been shown that the morphology of the soot agglomerates plays an important role in their effects on climate (Scarnato et al. 2013) and human health (Hassan and Lau 2009).

Particle emissions from combustion engines are often a combination of solid particles and condensed semi-volatile material. The solid particles that are mostly soot consist of small, nearly spherical primary particles that form polydisperse agglomerates by coagulation (Maricq and Xu 2004). Semi-volatile material may condense on the surface of the solid particles (i.e., semi-volatile material internally mixed with soot) and increase their mass and mobility, and/or they can nucleate to form new particles consisting of only semi-volatile material (i.e., semi-volatile particles externally mixed with soot). Studies have been carried out in the literature to measure the volatility of the particles from GDI engines using number-based (Mathis et al. 2005; Khalek et al. 2010) and mass-based (Storey et al. 2010) methods. Khalek et al. (2010) measured the total and solid particle size distribution for a GDI vehicle using three different fuels. Mathis et al. (2005) also employed a thermodenuder to measure the total and non-volatile number concentrations from a direct injection gasoline passenger vehicle. The ratio of the mass of organic to the elemental carbon was determined by Storey et al. (2010) for a GDI vehicle using gasoline fuel with a range of ethanol content. It should be noted that the volatility measured by organic carbon/elemental carbon (OC/EC) analysis determines the total semi-volatile mass internally and externally mixed with solid particles; however, the mass fraction of semi-volatile GDI particles on a size-segregated basis has not been studied.

The effective density function, defined as the mass of a particle divided by the volume of its mobility equivalent sphere, can describe the morphology of soot particles. It can also be used to convert particle size distributions to particle mass distributions from which total particle mass concentration can be determined. The integrated particle size distribution (IPSD) method is seen as an option for particle mass emission factor measurement for the modern vehicles (Liu et al. 2009, 2012), which produce very few particles and where gravimetric measurement is difficult due to the low mass collected and the significance of errors due to sampling artifacts (Chase et al. 2004). The effective density function for GDI particles has been measured by Maricq and Xu (2004), Symonds et al. (2008), and Quiros et al. (2015). However, the effect of volatility on particle effective density, and by extension the mass concentration measured by the IPSD method, has not been studied.

In this study, five GDI vehicles have been tested on a chassis dynamometer to measure the mass–mobility relationship (effective density) at three steady-state operating conditions. The mass–mobility relationship is required to estimate the mass concentration using the IPSD method, which may be one method that has the required sensitivity to measure the low emission levels from modern GDI vehicles where the traditional gravimetric method is potentially inaccurate. The feasibility and uncertainty of using an effective density function to estimate the particle mass emission factor using particle size distributions is examined.

2. EXPERIMENTAL METHODS

2.1. Vehicles and Test Conditions

Five gasoline direct injection passenger vehicles have been evaluated on a single roll chassis dynamometer (Clayton Industries, C-200). The specifications of the test vehicles are shown in Table S1 in the supporting information. These same vehicles were previously tested to determine particle number emission factors and the number fraction of volatile particles while operated on-road (Momenimovahed et al. 2015).

GDI engines are generally classified in three different categories depending on their injection strategies. In wall-guided engines, the fuel is guided toward the spark plug by the shape of the bowl on top of the piston whereas air-guided engines are those where cylinder internal airflow is used to guide the fuel flow. In the spray-guided system, the injector is placed in the center of the combustion chamber close to the spark plug. This latter case is shown to have a better fuel economy in comparison with the other two types (LEV III PM 2011). The engines used in this study are a combination of different fuel injection strategies so a comparison between different injection methods can be made. It is known that fuel aromatic content and volatility affect particle emission rates (Khalek et al 2010; Kim et al. 2013; Storey et al. 2014). The test fuel was randomly selected commercially available gasoline fuel, in order to assess the variability in the emissions of the in-use fleet.

Particle emissions were measured on the chassis dynamometer at three different steady-state operating conditions at 60 km/h vehicle speed. The chassis dynamometer was set to absorb tractive power at 0%, 5%, and 10% of the maximum rated engine power for each vehicle (e.g., 0 kW, 6.1 kW, and 12.2 kW for vehicle 1 that has a maximum engine power of 122 kW). Tractive powers of 5% and 10% are the powers approximately required for steady-state operation of the vehicles under normal conditions at speeds of 65 km/h and 85 km/h, respectively. For the case of 0% tractive power (a dynamometer setting of 0 kW), there is still a small load on the engine as the vehicle runs on the dynamometer as the engine provides power to overcome drivetrain friction and rolling resistance. Note that the evaluated steady-state operating conditions are not necessarily representative of real world transient driving conditions and they are chosen based on the capability of the test facilities. Furthermore, transient measurements were not made, as the aerosol measurement techniques used here require a steady-state aerosol source.

A pitot tube was installed in the tailpipe to directly measure the total exhaust flow rate that is used to calculate the particle mass emission factor in terms of mg/km. The exhaust gas temperature was monitored and was used to convert the flow rates measured by the pitot tube to the flow rates at the reference temperature (25ºC).

2.2. Aerosol Sampling System

Figure 1 shows a schematic of the experimental setup. Aerosol sample was drawn at the end of the tail pipe. To prevent condensation of semi-volatile material and water vapor, the sample was immediately diluted by a factor of 5–6 and a heated sample line, with a temperature of 80°C, was used to transfer the particle emissions to the particle instruments. The heated sample line might potentially affect the volatile fraction since there may be some volatile material in the sample that can evaporate at 80°C. The dilution system (indicated with a dashed box) provides high-efficiency particulate air (HEPA)-filtered air to dilute the sample. The vacuum pump in the top line is used to provide air for dilution and the other vacuum pump is employed to draw the diluted sample from the tailpipe to the measurement devices. A cyclone was used upstream of the measurement devices to remove particles larger than 1 µm (aerodynamic diameter).

FIG. 1. Schematic of the experimental setup.

The conditioned aerosol was measured using two systems: (i) a scanning mobility particle sizer (SMPS) with thermodenuder and (ii) a differential mobility analyzer (DMA), thermodenuder, and centrifugal particle mass analyzer (CPMA) system.

2.2.1. SMPS–Thermodenuder System

The SMPS (TSI Inc., Shoreview, MN, USA, Model 3081 and Model 3776, DMA3 and CPC3 in Figure 1; CPC = condensation particle counter) and thermodenuder were used to determine the nascent and non-volatile particle size distributions. The aerosol flow controlled by the CPC was set to 0.3 LPM and the DMA sheath flow was set to 3 LPM (this results in a resolution, or the inverse of the normalized full-width half-maximum (FWHM) of the transfer function, of 10 in terms of mobility). To measure the non-volatile size distribution, the thermodenuder was placed upstream the DMA3-CPC3 to remove the semi-volatile particles. The thermodenuder heated the sample to 200°C to remove semi-volatile material from the particles and had an inner diameter of 9.5 mm and length of 665 mm. “Semi-volatile material” has an operational definition and here it is defined as the material removed from the particle phase with the denuder at 200°C.

To measure the nascent particle size distributions, the aerosol bypassed the thermodenuder. A bypass line with the same length as the denuder line was used to minimize the systematic errors caused by thermophoretic and diffusional particle loss in the thermodenuder. NaCl particles were used to measure the thermophoretic and diffusional loss in the thermodenuder line relative to the bypass line. The ratio of number concentration out of thermodenuder to the number concentration out of the bypass line is shown in Figure S1 in the online supplementary information (SI). The non-volatile size distributions are then corrected for the losses to ensure that the difference between nascent and non-volatile size distributions is only due to the removal of semi-volatile particles.

2.2.2. DMA-Thermodenuder-CPMA/DMA System

A DMA-thermodenuder-CPMA/DMA system was used to determine the effective density of nascent and non-volatile particles and also to determine the mass of semi-volatile material condensed on the non-volatile particles. In general, the effective density can be found by knowing two out of the following three parameters: particle relaxation time (or aerodynamic-equivalent diameter), particle mobility, or particle mass. Experimentally, several methods have been used to measure the particle effective density from vehicle exhaust. The effective density has been found using a differential mobility analyzer (DMA) in series with an electrical low-pressure impactor (ELPI) to measure the aerodynamic-equivalent diameter of mobility-classified particles (Maricq and Xu 2004). Another method is to measure aerodynamic and mobility size distributions simultaneously and minimize the difference between the two size distributions using an effective density function (Virtanen et al. 2002). Alternatively, particle mass can be measured using an aerosol particle mass analyzer (APM; Ehara et al. 1996) or Couette centrifugal particle mass analyzer (CPMA; Olfert and Collings 2005). Park et al. (2003) and Rissler et al. (2013) have used a DMA in series with APM to measure the effective density of the soot particles from different sources such as a diesel engine, flame, and candle. Barone et al. (2011) also used the same method to compare the particle effective density from a premixed charge compression ignition engine with a conventional diesel engine. A CPMA along with a DMA was employed to measure the effective density of the particle emissions from a light-duty diesel vehicle at several operating conditions (Olfert et al. 2007).

In the current study, a DMA-thermodenuder-CPMA/DMA system was used. The addition of the thermodenuder to the conventional DMA-CPMA system allows for the measurement of the effective density of nascent particles, the effective density of non-volatile particles, and the mass of semi-volatile material condensed on the non-volatile particles as a function of the particle size.

In the experiment, particles were selected by mobility with DMA1 (TSI Inc., Model 3081) and were passed through the thermodenuder or through the thermodenuder's bypass (the same thermodenuder described above operating at the same temperature). Then the particle mass of the size-classified particles was measured with the CPMA. The CPMA consists of two rotating cylindrical electrodes and classifies particles based on their mass to charge ratio (Olfert and Collings 2005). The mass of size-classified particles was measured by stepping the CPMA over a wide mass range and measuring the particle concentration with CPC1 (TSI Inc., Model 3776, Product Information: Model 3776 Condensation Particle Counter). The spectrum of the particle mass distribution measured by the CPMA-CPC typically contains a single peak corresponding to the mass of the dominant species by number. The spectrum will also contain particle counts due to multiply charged particles exiting DMA1 and potentially charged particles with other densities than the dominant species (e.g., internally and externally mixed particles may have different effective densities). Fitting the data with a lognormal distribution is used to locate the peak. If there was evidence of significant multiply charged particles (typically seen as a skewed distribution to higher particle masses), then a bimodal lognormal distribution was used to isolate the single and multiply charged peaks. Furthermore, when the thermodenuder is bypassed between the DMA and CPMA, it might be expected that two peaks would appear in the mass spectrum due to potentially different effective densities of the internally and externally mixed particles. However, this was not the case for two reasons: (i) the mass spectrum is relatively broad because soot is being measured. Soot is an agglomerate of primary particles and the agglomerates will have a wide range of shapes for a given particle mass. Or stated another way, particles with the same mobility will have wide range of particle masses. The broadness of the mass spectrum is not limited by the resolution of the DMA or CPMA but by the distribution of particle masses for a given particle mobility. And (ii) the number fraction of externally mixed particles (those that are only comprised of semi-volatile material) is relatively small (∼15%) as shown in the experimental results below. The broadness of the mass spectrum, and the fact that the vast majority of particles are internally mixed, means that only one peak is seen in the mass spectrum, which corresponds to the average mass of the internally mixed particles for experiments when the thermodenuder was bypassed. In the experiments where the particles passed through the thermodenuder, all externally mixed semi-volatile particles are removed and the peak of the mass spectrum corresponds to the average mass of the non-volatile core of the internally mixed particles.

The CPMA measures the mass of the non-volatile size-selected particles when the aerosol is passed through the thermodenuder. When the thermodenuder is bypassed, the CPMA measures the total mass (both non-volatile and semi-volatile components) of the size-selected particles. Therefore, the difference between the total particle mass and the non-volatile particle mass is the mass of volatile material. The volatile mass fraction is defined as the mass of semi-volatile material on a particle divided by the total mass of the particle.

The sample flow rate through the CPMA was 1.5 LPM controlled by CPC1. The resolution of the CPMA is controlled by the rotational speed and voltage of the classifier, and was set to approximately 5 in terms of mass (i.e., the inverse of the normalized FWHM of the transfer function, was 5). This is equivalent to a resolution of 15 in terms of particle size (for spheres).

As volatile material is removed from the particles in the thermodenuder, the mobility of the particles will decrease. Therefore, another DMA (TSI Inc., Model 3081, DMA2) was used in a tandem DMA system (DMA1-thermodenuder-DMA2-CPC2) to find the mobility-equivalent diameter of the denuded particles. The diameter of the non-volatile particles was measured by stepping the voltage on DMA2, measuring the particle concentration with CPC2 (TSI Inc., Model 3776), and determining the diameter corresponding to the maximum particle concentration, similar to the method described by Rader and McMurry (1986).

It should be noted that the loss in the sampling lines in the DMA-thermodenuder-CPMA/DMA system does not affect the mass and size measurements as the location of the peaks in the mass spectrum (CPMA-CPC1) and size spectrum (DMA2-CPC2) is not dependent on the magnitude of the particle concentration.

3. EXPERIMENTAL RESULTS AND DISCUSSION

3.1. Volatility of the Particle Emissions

The fraction of externally mixed volatile particles can be found on a particle number basis by comparing denuded and undenuded size distributions. The size distributions for all vehicles are shown in Figure 2. In all cases, the size distributions are unimodal and there is no distinct nucleation mode, which is consistent with the results shown by Zhang et al. (2014) and Gu et al. (2012). In most cases, the distributions are not log-normal, rather the distributions are skewed (on a log-scale) toward smaller particle sizes. This asymmetric size distribution was also previously reported in the literature (Maricq et al. 1999; Harris and Maricq 2001; Storey et al. 2010). It has been shown that the size distributions of GDI particles for both homogenous charge operation (Maricq et al. 1999) and stratified charge operation (Harris and Maricq 2001) can be asymmetric with higher particle concentrations for smaller particles.

FIG. 2. Nascent and non-volatile particle size distributions for (a) vehicle 1, (b) vehicle 2, (c) vehicle 3, (d) vehicle 4, and (e) vehicle 5.

The size distributions also show that there is very little change in the count median diameter (CMD) between the nascent and non-volatile size distributions. The CMD for all the vehicles tested ranges between 55 and 73 nm with an average of 65 nm for the nascent particles and between 51 and 71 nm with an average of 63 nm for the non-volatile size distributions.

In most cases, a small fraction of the total particles in terms of number was removed by the denuder. On average, the volatile fraction for the steady-state operating conditions is 0.15 ± 0.14 (mean ± one standard deviation) for the steady-state conditions measured in the present study. Previously, Momenimovahed et al. (2015) tested the same vehicles in real-world, transient operation and showed that the number fraction of externally mixed semi-volatile particles was 0.20 ± 0.15, which is similar to the steady-state measurements made in this study. These results are consistent with the results reported by Khalek et al. (2010); using a GDI vehicle on two driving cycles, they found denuding the particles resulted in a decrease in the number concentration but a very small change in the CMD. Hedge et al. (2011) also observed the same behavior from a GDI engine at some steady-state operating conditions, although there was no aftertreatment installed on their engine. Ntziachristos et al. (2013) also found that a small fraction (16%) of the total particle concentration is externally mixed volatile material from a GDI vehicle operated on hot start and cold start New European Driving Cycles (NEDC).

The ratio of the semi-volatile mass to the total mass for internally mixed particles as a function of particle mobility size is shown in Figure 3. The average semi-volatile mass fraction for the five vehicles tested is shown in the plots and the error bars represent one standard deviation. The variability in these results is mostly due to the variability between vehicles as the repeatability of the DMA-CPMA system is very high (i.e., the standard deviation of measurements of a stable aerosol source is typically within 3% of the mean value). The figure shows there is relatively more condensed material at smaller particle sizes. This is consistent with observations on particles emitted from a diesel engine (Sakurai et al. 2003; Ristimaki et al. 2007), a compression-ignition natural-gas direct-injection engine (Graves et al. 2015), and a premixed ethylene flame (Ghazi et al. 2013).

FIG. 3. Average ratio of the mass of internally mixed semi-volatile material to the nascent particle mass for all vehicles. Error bars represent one standard deviation.

The figure also shows that the internally mixed semi-volatile mass fraction increases with tractive power. At 0% tractive power, for example, the ratio of the semi-volatile mass to the total mass is less than 0.15 while this ratio is generally greater than 0.2 at 10% tractive power. The increase in semi-volatile emissions at higher load could possibly be due to a reduced rate of fuel droplet evaporation or decreased fuel spray penetration at higher cylinder pressures (Stone 2012). Although size-segregated internally mixed volatility measurements have not previously been reported in the literature, qualitatively they can be compared to thermo-gravimetric or EC/OC analysis, which reports the total amount of volatile (or organic material) and elemental carbon independent of mixing state and size. Using EC/OC analysis, Storey et al. (2010) reported a range of 0.17–0.3 for the ratio of the organic carbon (presumably semi-volatile) to total carbon (OC/TC) for post-catalyst particulate at two steady-state operating conditions for a GDI vehicle operating on three different gasoline-ethanol blends (0%–20%). These OC/TC ratios are roughly comparable to the volatile mass fractions measured here, which were also measured post-catalyst. It is important to note that their pre-catalyst results show that the majority of the particulate is organic, which was also found by Price et al. (2007) and Chen et al. (2010). Therefore, during cold-starts, when the catalyst performance is poor, the internally mixed size-segregated semi-volatile mass fraction would likely be higher than what is reported here.

3.2. Particle Effective Density

Particle mass and mobility diameter are often shown to scale through a power law relationship (Park et al. 2003),[1] where d_m is the particle mobility diameter, D_m is the mass–mobility exponent, l is the unit length of one nanometer, and C is a constant. The mobility diameter, d_m, is divided by 1 nm to make the dimension of constant value, C, independent of D_m. Using the mass–mobility relationship, the effective density, which is the ratio of the mass to the volume of the mobility equivalent sphere, is[2]

The measured effective densities at three steady-state operating conditions are shown in Figure 4 for nascent (Figure 4a) and non-volatile particles (Figure 4b). The effective density decreases for increasing mobility diameter for all operating conditions. This is expected since there are more voids between primary particles in relatively larger aggregates and consequently the ratio of the mass to the mobility-equivalent volume drops as the aggregate size increases. The solid line represents the fit to all effective density values and the shaded area shows the fit standard error. Figure 4b shows a small degree of variability (a smaller standard error) for the non-volatile particles, which suggests that the effective density of the soot particles without semi-volatile material is relatively independent of the vehicle. The nascent effective density function (Figure 4a) shows a larger degree of variability in the effective density. Therefore, it is the semi-volatile material, and the varying amounts of it, that causes a greater degree of variability in the effective density. Figure 4a also shows the effective density functions reported by Maricq and Xu (2004), Symonds et al. (2008), and Quiros et al. (2015). As can be seen in the figure, the density functions measured by Symonds et al. (2008) and Quiros et al. (2015) are almost the same as the upper and lower bound limits, respectively, as measured in the present study. Maricq and Xu (2004), using a DMA-ELPI system, measured higher densities for particles smaller than ∼90 nm. Besides measurement uncertainty, the higher effective densities measured by Maricq and Xu (2004) could be a result of increased semi-volatile material condensed on the particles, which would generally increase their effective density.

FIG. 4. Effective density functions for (a) nascent and (b) non-volatile particles for all vehicles. The solid line represents the fit to all effective density values and the shaded area shows the standard error of the fit.

Figure 5a shows that the average nascent particle effective density slightly increases with tractive power, although it is difficult to quantify the significance of this trend due to the variation between the vehicles. However, this increasing trend is expected by noting that the mass fraction of internally mixed semi-volatile material is larger at higher tractive powers (Figure 3). The condensed semi-volatile material fills the voids in the agglomerate particles, consequently increasing the mass of the particles while causing only a slight increase in its mobility. Therefore, the effective density is larger at higher tractive powers where more semi-volatile material is available for condensation.

FIG. 5. Average effective density for all vehicles as a function of tractive power for four particle sizes for (a) nascent and (b) non-volatile particles. Error bars represent one standard deviation.

Figure 5b shows that for the non-volatile particles, the effective density is approximately constant at all tractive powers and the small change in effective density is not statistically significant. Therefore, considering only non-volatile particles, a unique effective density function may be representative for different tractive powers.

The mass–mobility exponent is 2.56 and 2.60 for the nascent and non-volatile particles, respectively; based on a fit of the data of all the vehicles. A higher mass–mobility exponent for non-volatile particles was also found for the vast majority of the individual vehicles and test conditions as shown in Table S2 in the SI. This is expected as there is relatively more semi-volatile mass at lower particle sizes (Figure 3), so the effective density of the small particles is increased by a larger degree compared with the larger sized particles. As a result, the difference between the effective density of the small and large particles is higher for the nascent particles and consequently the mass–mobility exponent will be lower.

The measured mass–mobility exponents for the GDI vehicles in the present study are higher than the reported values for diesel soot (Park et al. 2003; Olfert et al. 2007) or flame-generated soot (Maricq and Xu 2004) with little or no semi-volatile material condensed on the particle (typically 2.2–2.3). This suggests that the GDI particles may have a fundamentally different structure than diesel soot (Seong et al. 2014). Ghazi et al. (2013) have shown that the mass–mobility exponent will increase if the primary particle size increases with the soot aggregate size. Although transmission electron microscopy (TEM) images were not collected in this study, Barone et al. (2012), Lee et al. (2013), Seong et al. (2014), and Dastanpour and Rogak (2014) have all observed that primary particle size increases with soot aggregate size in emissions from GDI engines.

3.3. Implications for Mass Concentration Measurements

The particle effective density functions were used to convert the particle size distributions to mass distributions, which were then integrated to obtain the particle mass concentration. To find the particle mass distribution, the size distributions are simply multiplied by Equation (1). Since the particle size distributions measured by the SMPS were limited to particles smaller than approximately 700 nm, the size distributions were extrapolated by fitting lognormal functions, and for particles within the range of 700–1200 nm the fit was used. The upper limit of 1200 nm was chosen based on the cut point of the cyclone. Although a fit was used, on average the calculated mass concentration above 700 nm was only 0.6% of the total mass concentration. To find the emission factor, mass concentration is multiplied by the exhaust flow rate. Since the vehicle speed is constant, the emission factor is found by dividing the emission rate by the vehicle speed. The mass concentrations, mass emission factors, and corresponding uncertainties for the five vehicles are shown in Figure 6. Note that the mass emission factors are less than 1.7 mg/km for all vehicles, which is a few times lower than the limit in the Euro 6 emission standard. This can be explained by noting that the mass concentrations in the emission standards are measured at transient operating conditions that generally increase the emission factor because of higher emission rates during acceleration and deceleration.

FIG. 6. Particle mass concentrations for nascent and non-volatile particles. Mass concentrations for nascent particles assuming that all particles are internally mixed are shown as points.

In terms of uncertainty, the propagation of uncertainty can be used to estimate the uncertainty in the calculated mass concentration from the IPSD method. Assuming the number of size bins in the SMPS size distribution is p, the mass concentration (M) can be calculated from[3] where n_i is the number concentration in each size bin. The uncertainty for the mass concentration for each size bin is[4]

The uncertainty in the mass concentration for each size bin are dependent on each other, therefore, the uncertainty in the total mass concentration is simply the sum of the uncertainties for all size bins . It is assumed that the uncertainty is 10% for the number concentration (TSI, CPC specifications) and 3% for the particle mobility diameter (Kinney et al. 1991) both with a 95% confidence interval. The uncertainty in the effective density including the nascent and non-volatile effective densities ranges between 3% and 40% (for particle sizes between 15 and 1200 nm), which includes the uncertainty in the DMA-CPMA system (Johnson et al. 2013) and the variability in the effective density measurements shown in the present study. It should be noted that the effective density functions measured in the present study are limited to particles larger than 40 nm since the number concentration of particles smaller than 40 nm was not sufficient for the DMA-CPMA system to accurately measure. It has been shown that there might be solid particles as small as 6 nm in the GDI exhaust (Barone et al. 2012); however, small particles have a negligible effect on the mass concentration since the particle mass is dependent on the cube of the particle mobility size. For a typical non-volatile size distribution measurement in this study, the total uncertainty in mass concentration is approximately 16% (with 95% confidence).

In regulation testing on a transient driving cycle, one would be required to measure the size distribution with a real-time mobility spectrometer such as the differential mobility spectrometer (DMS; Cambustion Ltd., UK) or engine exhaust particle sizer (EEPS, TSI Ltd.). However, this comes with increased uncertainty in particle sizing and concentration measurement. By simultaneously calibrating the aerosol charger and electrometers in the mobility spectrometer using morphologically relevant particles (as currently done in the DMS), it is possible to achieve uncertainties of 10% in mobility size and 20% in number concentration (Symonds 2010). If this calibration is not done, the uncertainties will be much higher (Aswathi 2013). Therefore, assuming an uncertainty in the number concentration and mobility diameter to be 20% and 10%, respectively; the uncertainty in measured mass concentration would be 37% for real-time spectrometer. Other uncertainties, including uncertainties in the dilution ratio, would increase this estimate.

The uncertainty in mass concentrations in the case of nascent particles will depend on the mixing state of the semi-volatile material. In most of the emissions measured here, the semi-volatile material is both internally and externally mixed with solid particles. In this case, the effective density and the size distribution of the two types of particles must be known to calculate the mass concentration. In diesel engines, it is typical to have a nucleation mode comprised mostly of semi-volatile particles and an accumulation mode comprised mostly of soot, which are readily distinguishable based on the size distribution. These modes can be fit with log-normal distributions, assigned separate effective density functions, and the mass concentration can be calculated. This method is used by the differential mobility spectrometer (DMS) for mass concentration measurements (Symonds et al. 2007). In the DMS software, the nucleation mode is assumed to have an effective density of 1 g/cm³; however, the uncertainty in this density has very little effect on the total mass concentration because the size (and mass) of the nucleation mode particles tends to be very small (approximately less than 20 to 30 nm). However, as shown by Khalek et al. (2010) and here (in Section 3.1), GDI particle emissions contain semi-volatile particles that are not readily distinguished from the internally mixed particles in the size distribution and they can have significant mass. This is major source of uncertainty in calculating mass concentration of nascent GDI emissions using the IPSD method and it is examined presently.

In this study, the semi-volatile material condensed on the soot particles does not significantly increase its mobility diameter. This is seen in Figure 7, which shows the average ratio of the non-volatile mobility diameter to the nascent mobility diameter for DMA-selected particles is 0.97. Therefore, assuming the non-volatile size distribution is equal to the size distribution of the internally mixed particles, the difference between the nascent and non-volatile size distributions will be the distribution of the purely semi-volatile particles. For this case, since the size distribution of externally mixed semi-volatile particles is not directly measured by SMPS, the uncertainty in the semi-volatile mobility diameters is assumed to be 8% instead of 3% (a conservative estimate based on the data in Figure 7). Here we have assumed purely semi-volatile particles have a constant effective density of 1 g/cm³ with an uncertainty of 20%. The total mass of nascent particles is found by adding the mass of the internally mixed particles with the mass of purely semi-volatile particles. The mass concentrations and uncertainties for the nascent particle measurements for the five vehicles are also shown in Figure 6.

FIG. 7. Ratio of non-volatile to nascent mobility diameters as a function of the nascent particle mobility diameter.

As can be seen in Figure 6, the mass concentrations and also uncertainties are both higher for nascent particles due to the semi-volatile material. The estimated mass concentrations for nascent particles have an uncertainty of up to 46%; significantly higher than the uncertainty in the non-volatile mass concentrations. Furthermore, it would be expensive from a regulatory viewpoint to measure the nascent and non-volatile size distributions independently as two real-time spectrometers would be required (one equipped with a denuder). Alternatively, one could make the simplifying assumption that all the aerosol is internally mixed. Therefore, the mass concentration can be calculated using the nascent size distribution (requiring only one spectrometer) and using the effective density function shown in Figure 4a. Using this assumption, the mass concentration was calculated and shown in Figure 6 as a point. This underestimated the mass by 7% to 34%. It should be noted that the percentage of underestimation is highly dependent on the concentration of pure volatile particles, i.e., having more purely volatile particles in the exhaust causes a higher degree of underestimation. Since different gasoline direct injection vehicles might produce various amounts of purely volatile particles at different operating conditions, this last method would be less accurate for mass emission determination for all GDI vehicles.

4. CONCLUSION

The effective density function for five gasoline direct injection vehicles were measured at three steady-state operating conditions for both nascent and non-volatile particles. The results reveal that the ratio of volatile mass to the total mass is higher for relatively smaller particles and also at higher tractive powers. The non-volatile effective density functions showed an independent behavior with respect to the vehicle and tractive power suggesting that a unique density function is representative for a wide range of vehicles at different operating conditions. For nascent particles, however, the variation between effective density values from different vehicles is higher, which is likely a result of different amounts of semi-volatile material in the particle phase in the exhaust of different vehicles. It should be noted that the thermodenuder temperature was set to 200ºC and there may be other material denuded from the particle phase at higher temperatures. Therefore, the non-volatile density functions might be somewhat different than the measured values in the present study if the denuder temperature is higher than 200ºC.

The uncertainty in the mass concentration calculated from IPSD method for non-volatile particles is approximately 16% using an SMPS system. For the case of nascent mass concentrations, the uncertainty of the IPSD method is higher (∼46%) due to uncertainties in the particle mixing state and their associated effective densities. The uncertainty in the IPSD method is even potentially higher if the method is applied to transient test cycles since the number of externally mixed semi-volatile particles maybe higher during acceleration compared to steady-state operating conditions. Therefore, to minimize uncertainty it would be prudent to only use the IPSD method to estimate the mass emission factor for only non-volatile particles since the accuracy of the mass emission factor for nascent particles is strongly dependent on the mixing state of the particles. Furthermore, if this method were to be used in regulatory applications, a fast response instrument such as DMS or EEPS would be required to measure transient driving cycles. However, the uncertainty in the size distribution measured by these instruments is higher than the SMPS and therefore the uncertainty in the mass concentration can be up to 37% for non-volatile mass concentrations.

SUPPLEMENTAL MATERIAL

Supplemental data for this article can be accessed on the publisher's website.

ACKNOWLEDGMENTS

The authors would like to thank Dallin Bullock for his help in operating the vehicles.

Funding

This project was undertaken with the financial support of the Government of Canada through the Federal Department of the Environment. The author would like to thank the Natural Sciences and Engineering Research Council of Canada (NSERC) and AUTO21 for their funding support.

Notes

¹ Upon denuding, particles will decrease in mobility size if sufficient semi-volatile material is contained within the particle. Therefore, the measured mobility size of the denuded particles was slightly different at each engine load. To plot lines of constant mobility diameter for the non-volatile particles (Figure 5b), the effective densities and mobility diameters were calculated based on the equation of fit of the effective density data.

² It should be noted that the uncertainty in the inversion method used by the SMPS to invert the particle counts to the size distribution is not included in the uncertainty analysis.