Evolution of In-Cylinder Diesel Engine Soot and Emission Characteristics Investigated with Online Aerosol Mass Spectrometry

To design diesel engines with low environmental impact, it is important to link health and climaterelevant soot (black carbon) emission characteristics to specific combustion conditions. The in-cylinder evolution of soot properties over the combustion cycle and as a function of exhaust gas recirculation (EGR) was investigated in a modern heavy-duty diesel engine. A novel combination of a fast gassampling valve and a soot particle aerosol mass spectrometer (SP-AMS) enabled online measurements of the in-cylinder soot chemistry. The results show that EGR reduced the soot formation rate. However, the late cycle soot oxidation rate (soot removal) was reduced even more, and the net effect was increased soot emissions. EGR resulted in an accumulation of polycyclic aromatic hydrocarbons (PAHs) during combustion, and led to increased PAH emissions. We show that mass spectral and optical signatures of the in-cylinder soot and associated low volatility organics change dramatically from the soot formation dominated phase to the soot oxidation dominated phase. These signatures include a class of fullerene carbon clusters that we hypothesize represent less graphitized, C5-containing fullerenic (high tortuosity or curved) soot nanostructures arising from decreased combustion temperatures and increased premixing of air and fuel with EGR. Altered soot properties are of key importance when designing emission control strategies such as diesel particulate filters and when introducing novel biofuels.


* S Supporting Information
ABSTRACT: To design diesel engines with low environmental impact, it is important to link health and climaterelevant soot (black carbon) emission characteristics to specific combustion conditions. The in-cylinder evolution of soot properties over the combustion cycle and as a function of exhaust gas recirculation (EGR) was investigated in a modern heavy-duty diesel engine. A novel combination of a fast gassampling valve and a soot particle aerosol mass spectrometer (SP-AMS) enabled online measurements of the in-cylinder soot chemistry. The results show that EGR reduced the soot formation rate. However, the late cycle soot oxidation rate (soot removal) was reduced even more, and the net effect was increased soot emissions. EGR resulted in an accumulation of polycyclic aromatic hydrocarbons (PAHs) during combustion, and led to increased PAH emissions. We show that mass spectral and optical signatures of the in-cylinder soot and associated low volatility organics change dramatically from the soot formation dominated phase to the soot oxidation dominated phase. These signatures include a class of fullerene carbon clusters that we hypothesize represent less graphitized, C 5 -containing fullerenic (high tortuosity or curved) soot nanostructures arising from decreased combustion temperatures and increased premixing of air and fuel with EGR. Altered soot properties are of key importance when designing emission control strategies such as diesel particulate filters and when introducing novel biofuels.

■ INTRODUCTION
Diesel engines are a major source of air pollution as they emit large amounts of fine particulate matter and nitrogen oxides (NO x ) to the atmosphere. Diesel exhaust and soot particles can cause acute inflammatory responses in airways and peripheral blood, 1 decreased lung function, 1 asthmatic symptoms 2 and are classified as carcinogenic. 3 Soot emissions (often referred to as black carbon; BC) also contribute to global warming through short-lived climate forcing. 4 Consequently, reducing diesel engine NO x and soot emissions is an essential measure to improve air quality and reduce the anthropogenic climate impact. For these reasons stringent diesel engine emission regulations of fine particulates and NO x have been introduced in the European Union and the United States.
Modern diesel engines commonly apply exhaust gas recirculation (EGR) to reduce NO x emissions. With EGR, a fraction of the engine inlet air is replaced by exhaust gases. This reduces the oxygen concentration and combustion temperature and consequently the rate at which NO x forms. 5 However, soot emissions increase rapidly with the decreasing inlet oxygen (O 2 ) concentration. 6,7 Soot emissions from diesel engines are determined by two competing in-cylinder processes: soot formation and soot oxidation. 8 The soot formation rate is determined by the availability of acetylene, the formation of polycyclic aromatic hydrocarbons (PAHs) and the inception of soot particles, 9 all of which are processes that depend strongly on temperature and local mixing of fuel and air. Soot oxidation (removal) depends strongly on the availability of hydroxyl radicals, O 2 and the temperature. The soot reactivity toward oxidation also depends strongly on the soot nanostructure 10,11 and surface oxygen content. 11 Soot reactivity is of key importance when designing and optimizing diesel particulate filters (DPFs), the most important and efficient soot particle emission mitigation technique in modern diesel vehicles.
Studying in-cylinder soot properties can provide fundamental information on the relationship between combustion con-ditions and soot properties in order to increase our understanding of the processes governing soot emissions. These properties include particle size, chemical composition (e.g., elemental carbon and organic components, including PAHs), microstructure (e.g., primary particle size, aggregate sizes, and morphology) and nanostructure (e.g., fringe length, tortuosity, and separation distance). A recently emerging topic is the identification of high curvature (high tortuosity) soot nanostructures found in biodiesel emissions 12 and in laboratory studies of partially premixed flames. 13 These properties may influence soot formation and oxidation rates, as well as soot reactivity and thus the design and efficiency of DPFs. Moreover, combining in-cylinder particle characterization and exhaust characterization can improve our knowledge of the relationship between combustion conditions and the eventual toxicity of emitted soot particles, which in a complex way depends on soot size, surface area and reactivity, surface coating material (e.g., PAHs), and metal content. 14 There are a number of established techniques for studying incylinder particle concentrations and properties. Engines modified to allow optical access 15 have enabled noninvasive highly time-resolved studies of in-cylinder gases (oxidants, PAHs, etc.) using laser-induced fluorescence, 16,17 and studies of soot concentration using laser-induced incandescence 18,19 and laser extinction. 20 Recently, TEM grids were mounted inside a cylinder allowing detailed studies of particle nanostructure using high resolution transmission electron microscopy 21 (HR-TEM). A total cylinder sampling system 22 can be used for rapid (1 ms) extraction of the full cylinder volume from a single combustion cycle and to study soot properties using an array of off-line techniques including HR-TEM and Raman microscopy. 23,24 Off-line techniques involve particle sampling, handling and often chemical processing that could lead to unintended changes in particle properties. Online aerosol measurement techniques are commonly used in ambient measurements to characterize particle properties directly in the aerosol phase, for example to investigate the transformation of climate-relevant soot properties upon aging in the atmosphere. 25 The direct radiative forcing of BC can be described by knowledge of soot morphology and coating thickness, 26 for which online characterization is essential.
Fast sampling valves 27−29 (FSVs) are in-cylinder gas extraction techniques that enable particle characterization using online aerosol instruments. FSVs repeatedly extract a small volume of in-cylinder gases at a well-defined position in the combustion process. The repeated extraction produces a semicontinuous flow of in-cylinder gases, and because of the small volume sampled in each extraction, the FSV has a negligible impact on combustion. 29 Additionally, the aerosol extracted is averaged over a large number of fired cycles which decreases the sensitivity to cycle-to-cycle variations compared to other in-cylinder particle extraction techniques.
The soot particle aerosol mass spectrometer (SP-AMS) was recently developed 30 to enable online chemical analysis of soot particles. It enables chemical analysis of the soot core by overlapping a focused Nd:YAG laser beam (λ = 1064 nm) with a beam of particles, permitting decoupled vaporization and electron ionization (70 eV) and thus detection of highly refractory species such as elemental carbon and heteroatoms (e.g., oxygen in the solid carbon core) in a high resolution timeof-flight (HR-ToF) mass spectrometer. The SP-AMS also allows the composition of nonrefractory material (e.g., organic material coating the soot core) to be characterized by vaporizing particles on a heated porous-tungsten vaporizer at 600°C. 31 Probing in-cylinder particles with the SP-AMS can therefore provide more detailed online information on the composition of the refractory soot core and of condensed nonrefractory material on soot particles than present laser diagnostic techniques, and without the potential artifacts associated with off-line particle analysis.
We combined an FSV 29 and the SP-AMS to directly and continuously probe the evolution of in-cylinder particle composition and concentration as combustion progressed in a modern heavy duty diesel engine. We specifically addressed the implementation of lowered combustion temperatures and increased premixing using EGR, and its effect on soot characteristics during combustion.
Our results show that in-cylinder soot processes exhibit two distinct phases: a soot formation dominated phase and a soot oxidation dominated phase. These soot phases are well described in the literature, 24,32,33 which indicates that representative in-cylinder conditions were probed. We show that EGR decreases both soot oxidation and soot formation rates which results in increased soot and PAH emissions. Finally, at high EGR we identified fullerene-like carbon cluster signals in the SP-AMS mass spectra. These large carbon clusters are hypothesized to represent high tortuosity (curved) soot nanostructures, linking combustion with EGR to increased C 5 chemistry also found in selected premixed flames. 13 ■ MATERIALS AND METHODS Engine Operation. A heavy-duty Scania D13 engine was operated at low load (5.5 bar IMEPg) and 1200 rpm. The common rail fuel injection pressure was set to 2000 bar, the air intake temperature and pressure were set to 337 K and 1.65 bar, respectively. The fuel used was Swedish diesel MK1 with ultralow sulfur content, a cetane number of 56.8, an aromatic content of 4.4 vol % and an alkane content of 95.6 vol %. A synthetic lubrication oil (Powerway GE 40, Statoil) was used. Engine conditions were altered by running the engine at no EGR (0%), 56% and 64% EGR. These three EGR levels corresponded to engine inlet air O 2 concentrations of 21%, 15%, and 13%, respectively. When introducing EGR, the start of fuel injection (SOI) was initiated earlier in order for CA50 (the crank angle position where 50% of the heat has been released) to remain approximately constant at 7.5 crank angle degrees after the top dead center (CAD ATDC). Regulated gas phase emissions were monitored for the three EGR concentrations, corresponding to 21%, 15%, and 13% inlet O 2, and were (in g/kWh): hydrocarbons (HC) 0.113, 0.054, 0.046; carbon monoxide (CO) 0.41, 0.57, 1.99; nitrogen oxides (NO x ) 11.2, 0.27, 0.08, respectively. For a complete description of the engine and the operating conditions, the reader is referred to Shen et al. 29 Aerosol Sampling and Dilution. To extract in-cylinder particles, a fast sampling valve 29 (FSV) was mounted on the cylinder head by replacing one of the exhaust ports, with the sampling position located between two adjacent fuel jets. The valve is actuated by a solenoid hammer and in-cylinder gases were repeatedly extracted at a precise timing in the combustion process. The reported sampling times represent the opening of the FSV and the sampling resolution of the FSV (i.e., the duration that the valve is open) was estimated to less than 0.5− 1 ms or 4−6 CAD, see Supporting Information (SI). The NO x concentrations increased rapidly during the combustion and reached a plateau concentration in the late combustion cycle

Environmental Science & Technology
Article (SI Figure S1) consistent with the exhaust NO x concentration (within 30%). The FSV working principles and an estimation of the sampling resolution and particle losses are presented in the SI. After exiting the FSV, the extracted aerosol was diluted in three stages with a nominal dilution factor of 700 times and transported to the instruments. The dilution system is elaborated in detail in the SI together with a schematic (SI Figure S2).
Aerosol Characterization. Particle chemical composition was investigated with a soot-particle aerosol mass spectrometer 30 (Aerodyne Inc. Billerica, MA). The SP-AMS was run in single or dual vaporizer mode. In the single vaporizer mode, particles are flash vaporized upon impaction on a heated (600°C ) tungsten surface. In the dual vaporizer mode, particles containing rBC are vaporized using an intracavity Nd:YAG laser (1064 nm). The vapors are then ionized (70 eV electron ionization) and detected in a HR-ToF mass spectrometer. The SP-AMS allows efficient sampling of particles in the diameter range ∼70−500 nm 34,35 (vacuum aerodynamic diameter). The SP-AMS setup and calibration is further detailed in the SI.
The equivalent BC mass concentration and absorption angstrom exponent (AAE) were analyzed using an aethalometer 36 (model AE33, Magee Scientific). A scanning mobility particle sizer (SMPS, classifier model 3071, CPC 3010, TSI Inc.) was used to measure the particle mobility size distribution. The in-cylinder diesel soot concentration was measured simultaneously using the SP-AMS and the aethalometer. The rBC mass concentration correlated well with the equivalent BC mass measured with the aethalometer (r 2 = 0.99) with an rBC to BC ratio of 0.18 (SI Figure S9). This difference between the two instruments was likely caused by lower SP-AMS collection efficiency 37 for diesel soot compared to the calibrant (Regal Black) and differences between optical detection (aethalometer) and chemical detection (SP-AMS). Reported rBC concentrations have been scaled to the BC concentrations by the factor 1/0.18.
To quantify the contribution of particle phase in-cylinder PAHs to the total organic aerosol (OA), molecular masses corresponding to nine parent peaks of common PAH isomers between MW 202 and MW 300 (C 16 H 10 and C 24 H 12 ) were included. Smaller PAHs (with 2−3 rings, MW < 202) were not included in the analysis due to their high vapor pressure and lack of AMS analysis of pure lab standards. We accounted for the relative abundance of the parent peaks in the mass spectra (∼0.25) found in previous laboratory calibrations for 7 PAH isomers. 38

■ RESULTS AND DISCUSSION
In-Cylinder Soot Formation and Oxidation. The engine was operated at three levels of EGR corresponding to 21% (no EGR), 15% and 13% inlet O 2 concentrations. At each level of EGR, the FSV was used to extract in-cylinder gas and particles at a number of positions during the combustion cycle. The reported in-cylinder concentrations were converted to standard temperature and pressure and reported in mg m −3 . The positions in the combustion cycle were chosen so that the soot formation dominated phase (increasing rBC concentration), the soot peak and the soot oxidation dominated phase (decreasing rBC concentration) were distinguishable.
The apparent rate of heat release (aRoHR) derived from the in-cylinder pressure data is presented in Figure 1a. The start of injection (SOI) occurred earlier when EGR was applied and the start of combustion (SOC) was delayed, which increased the time allowed for premixing of air and fuel by approximately 0.8 CAD when the inlet O 2 concentration was changed from 21% to 13%. The aRoHR was characterized by two well-defined peaks at 21% inlet O 2 concentration. The two peaks indicate premixed or partially premixed combustion (first peak) and spray driven combustion (second peak) which transitioned into diffusion controlled combustion when the fuel injection stopped. As a result of lowered temperatures and slower kinetics when EGR was introduced, the two peaks were poorly defined at 15% inlet O 2 concentration and only one peak was observed at 13% inlet O 2 concentration.
Due to the high in-cylinder temperatures, the in-cylinder soot mass was principally determined by the concentration of refractory black carbon (rBC). The in-cylinder rBC concentrations measured with the SP-AMS are presented in Figure 1b. The rBC concentration increased rapidly after SOC ("soot formation phase"), it reached a maximum shortly after CA50 ("soot peak"), it then decreased rapidly ("soot oxidation phase"). The soot peak concentration decreased strongly with increasing EGR which shows that soot formation rates were reduced by lower inlet O 2 concentrations. Most of the soot formation and oxidation observed in Figure 1b occurred within approximately 2 ms (14 CAD).

Environmental Science & Technology
Article oxidation was the dominant in-cylinder process that governed soot emissions in the raw engine exhaust. Lower soot oxidation rates with decreasing inlet O 2 concentrations have previously been shown to be the main cause of increased soot exhaust emissions with increasing EGR. 33,39 In-cylinder soot particle size distributions from the current measurements have been reported previously. 29 The exhaust soot size distributions for the three levels of EGR had the following geometrical mean mobility diameters and standard deviations: 54.8 ± 0.3 nm, 1.69 ± 0.01 at 21% inlet O 2 concentration; 46.9 ± 1.4 nm, 1.78 ± 0.02 at 15% inlet O 2 concentration; 70.3 ± 3.4 nm, 1.64 ± 0.01 at 13% inlet O 2 concentration. Figure 1c shows the ratio of particle phase PAHs (4−6 ring PAHs, MW 202−300) relative to the rBC concentration. At 21% inlet O 2 concentration, the PAH to rBC ratio was low throughout the combustion cycle. At 13% and 15% inlet O 2 concentrations, elevated ratios were observed during the soot formation dominated phase that decreased as the combustion proceeded. These observations are consistent with large PAHs (in the size range from Pyrene, MW 202, to Coronene, MW 300) being involved in the formation and growth of soot particles. It also shows that lowered flame temperatures with EGR reduced the soot formation rates, and allowed the concentration of larger PAHs to accumulate. The particle phase PAH emissions in the exhaust (MW 202-300) increased with EGR and were 0.1 μg m −3 at 21% inlet O 2 concentration; 0.7 μg m −3 at 15% inlet O 2 concentration; and 4.8 μg m −3 at 13% inlet O 2 concentration. Thus, a fraction of the PAHs that accumulated during combustion with EGR survived, and the process led to increased emissions.
The soot trends presented in Figure 1b are generally consistent with in-cylinder soot trends obtained with highly time-resolved laser extinction measurements for similar EGR levels in a similar engine with optical access. 33,40 Thus, the combination of the FSV with online aerosol instruments is complementary to noninvasive optical techniques such as laser extinction.
Aerosol Mass Spectra of Soot in the Formation and Oxidation Dominated Phases. Figure 2 shows aerosol mass spectra of particles sampled from the soot formation dominated and oxidation dominated phases at 13% inlet O 2 (high EGR). The sizes of carbon clusters resulting from IR laser vaporization followed by electron ionization (70 eV) of the soot cores are shown in Figures 2a-b. The carbon clusters were grouped into low-carbons (C 1 + −C 5 + ), midcarbons (C 6 + −C 29 + ) and fullerenecarbons (C 30 + −C 58 + ). During both soot formation and oxidation, rBC was dominated by C 1 + −C 3 + ion fragments. Signals from larger mid-and fullerene-carbon clusters were pronounced in the soot formation phase (Figure 2a) and low in the oxidation phase (Figure 2b).
Low volatility organic species condense onto the soot particles and form the organic aerosol (OA) primarily when gases cool during extraction with the FSV and in the subsequent piping. The OA observed can be considered as the low volatility in-cylinder vapor species that will ultimately condense and form a soot coating in the engine exhaust. Aerosol mass spectra with the laser turned off (Figures 2c−d) were used to identify organic ion fragments (m/z 10−115) and the main PAH parent peaks (m/z 128−400). The organic mass spectra showed strong contributions from the hydrocarbon peaks C 3 H 5 + (m/z 41), C 3 H 7 + (m/z 43), C 4 H 7 + (m/z 55), and C 4 H 9 + (m/z 57). High signal intensity was also found at m/z 67, 69, 71 and m/z 81, 83, and 85. The origin of these hydrocarbon fragments are related to straight-chain alkanes (nalkanes), branched alkanes (C n H 2n+1 + ) and cycloalkanes (C n H 2n-1 + and C n H 2n-3 + ) and they commonly dominate diesel exhaust OA. 41−43 Ratios larger than 1 of the signal intensities at m/z 69 and 71, and at m/z 83 and 85 are likely only for an OA containing more than 95% lubrication oil. 41,44 These ratios are

Environmental Science & Technology
Article shown in SI Figure S10. OA was dominated by unburnt lubrication oil throughout the combustion cycle. Aliphatic-like fragments originating mainly from saturated and unsaturated (or cyclic) compounds were grouped according to C x H y>x + . Other strong nonrefractory components included C 2 H 2 + , C 3 H 3 + , C 5 H 3 + and C 6 H 5 + . These fragments and ions fulfilling the formula C x H y≤x + originate mainly from aromatic or highly unsaturated aliphatic compounds and were assigned their own class in this study. Nonrefractory CO 2 + and other oxidized fragments were also assigned their own class (C x H y O z + ). The contribution of these two latter classes to OA decreased from the soot formation to the soot oxidation phase, with the mass spectra in the soot oxidation dominated phase being more similar to diesel exhaust OA signatures in the literature. 41 Figures 2c−d show that PAHs were significantly more abundant in particles during soot formation. The larger PAHs decreased substantially in the soot oxidation phase, consistent with a large fraction of these PAHs having been converted to soot. Wang et al. 32 found that the naphthalene concentration accounted for 26−84% of the total PAH mass throughout the combustion cycle. Thus, the signal detected with the AMS in the particle phase for the volatile PAHs (2−3 rings) may be only a fraction of their total concentrations. The reason we observed a signal from naphthalene, given its high vapor pressure, may be due to strong adsorption to the soot core surfaces and to a small extent fragmentation of larger PAHs during electron ionization.
Evolution of In-Cylinder Soot Characteristics. Figure 3 shows the evolution of soot properties between the soot formation and soot oxidation dominated phases at 21%, 15% and 13% inlet O 2 concentrations using the subgroups of rBC and OA described above. A clear transition of soot properties took place between the soot formation phase and soot oxidation phase at 15% and 13% inlet O 2 concentrations, but much smaller differences were observed at 21% O 2 . In the refractory mass spectra of the soot cores, a transition in the composition of rBC was observed at 15% and 13% inlet O 2 concentrations. The mass spectra had relatively high fractions of mid-and fullerene-carbons during the soot formation phase, and were completely dominated by low-carbons in the soot oxidation phase. During the soot formation phase, nonrefractory mass spectra of low volatility organics at 15% and 13% inlet O 2 concentrations were, in addition to aliphatic-like fragments (C x H y>x + ), composed of substantial fractions of aromatic-like fragments (C x H y≤x + ), oxidized organic fragments (C x H y O z + ), and PAHs. Figure 4(a−f) shows a more detailed analysis of the incylinder soot evolution and comparison with the exhaust. Normalized rBC concentrations in Figure 4f are shown for reference to the evolution of in-cylinder soot mass. The fullerene-carbons (C 30−58 + ) shown in Figure 4a accounted for several percent of the rBC signals in the soot formation dominated phase at 15% and 13% inlet O 2 concentration. The fullerene-carbon fraction of rBC decreased gradually as the combustion proceeded. At 21% inlet O 2 concentration, fullerene-carbons represented a negligible fraction of rBC. As shown in SI, fullerene-carbon signals from the exhaust soot at 21% inlet O 2 concentration were only slightly higher than the nonrefractory organic background signal obtained with the laser turned off. At 13% inlet O 2 concentration, the signals from fullerene-carbon clusters in the exhaust were clearly higher than the organic background (SI Figure S12), although they were very low compared to the total signal from all refractory carbon clusters.
Recently, SP-AMS mass spectra of refractory carbon clusters from 12 different soot and manufactured carbon black sources were characterized. 45 Fullerene signals were not observed in mass spectra from graphitic (mature) soot, nor in amorphous carbon but they were detectable in biomass burning, ethylene flame soot and Nano-C fullerene black. 45 The authors concluded that fullerene signals in the SP-AMS may arise from carbon nanostructures that can form fullerenes upon heating. 45 Figure 3. Soot chemical composition during the soot formation dominated phase, soot peak and soot oxidation dominated phase for 21%, 15% and 13% inlet O 2 concentrations. Left columns: grouped ion fragments originating from nonrefractory low volatility organics. Right columns: refractory carbon clusters from the soot core. Refractory soot core carbon clusters: low-carbons C 1 It was recently shown 13 that partial premixing of oxygen in benzene (and ethylene) flames introduced high curvature, C 5containing fullerenic nanostructures in the soot formed. The C 5 production was suggested to proceed through partial benzene oxidation yielding the phenoxy radical followed by CO loss to produce C 5 . EGR, when used in diesel engines as in this study, allows a higher degree of premixing of O 2 in the soot formation zone similar to the model flame systems discussed above. Thus, we hypothesize that the fullerene-carbon signals correlate with high curvature, C 5 -containing fullerenic-soot nanostructures. 46 EGR has also been shown to affect exhaust soot properties resulting in a more reactive soot with a higher disorder of the soot nanostructure, 47 consistent with fullerene-carbon signals in the mass spectra of exhaust soot only being present during combustion with EGR. In previous SP-AMS studies of diesel exhaust, fullerene-carbon signals were commonly not observed. 45,48 The evolution of soot nanostructures during the diesel combustion cycle has previously been investigated using HR-TEM 23 where it was found that the soot rapidly became more graphitic as the combustion proceeded, and that after 50 CAD, the gradual increase in graphitic structure was small. Following the discussion of Onasch et al., 45 we interpreted the decay in fullerene-carbon signal at 15% and 13% inlet O 2 concentrations to be due to a gradual ordering of the soot nanostructure, analogous to the results of Li et al. 23 The low signal from fullerene-carbon clusters in the late combustion cycle and exhaust is consistent with graphitic soot and reduced content of fullerenic nanostructures.
The importance of the soot nanostructure for reactivity and oxidation rates has been well documented. 10,49 Shorter graphene layer planes increase the reactive carbon layer edge sites and fullerenic nanostructures have weaker C−C bonds due to the curvature of the molecules which increases the accessibility for oxidation by O 2 and OH. 49 We therefore hypothesize that the soot associated with the fullerene-carbon signals in the SP-AMS was oxidized (removed) more rapidly than soot that did not produce a fullerene-carbon signal. This In addition to rBC, the refractory part of the mass spectrum included considerable signal intensity from CO 2 + ions. The refractory CO 2 + (rCO 2 + ) intensity to that of the refractory C x + signal intensity varied only weakly over the combustion cycle (Figure 4b), with slightly higher ratios when EGR was applied to the engine. We also observed refractory C 3 O 2 + ions. The signal from rC 3 O 2 + ions was approximately ten times lower than the ion signal from rCO 2 + . However, and more importantly, the rC 3 O 2 + and rCO 2 + signals showed trends in the combustion cycle that were similar to those of X-ray photoelectron spectroscopy analyses of oxygenated surface functional groups reported in the literature. 24 We hypothesize that these refractory oxygen-containing ion fragments observed in the SP-AMS mass spectra arose as a result of the partial oxidation of soot cores, and that the ions originated from oxygenated functional groups in the soot core carbon nanostructure and on the soot core surface as suggested previously in laboratory 50 as well as ambient 51 measurements.
When EGR was applied, the fraction of nonrefractory oxidized organic fragments (C x H y O z + ) to the total OA decayed from an initially high fraction (∼30%) down to ∼15% as the combustion proceeded (Figure 4c). This may have been a result of the higher availability of O 2 during the early combustion cycle due to the increased premixing with EGR. However, because this decay was correlated with the decay in the fullerene-carbon signal, it could be associated with decreased reactivity of the soot surfaces as the combustion proceeded. With no EGR, the initial contribution from oxidized organics to the total OA was low and instead, a gradual increase was observed as the combustion proceeded. This resulted in a small increase of the oxidized organic fraction of OA in the exhaust at 21% inlet O 2 concentration relative to 15% and 13% inlet O 2 concentrations. These observations indicates that the mechanism responsible for the higher fraction of nonrefractory oxidized organics in the exhaust at 21% inlet O 2 concentration compared to the EGR cases could be the higher availability of O 2 during the late cycle. A small fraction of the signal originating from surface oxides as described above for rCO x + may have been volatile enough to be vaporized at 600°C. This would explain the part of the signal in the nonrefractory spectra that varied only weakly with CAD. Figure 4d shows the quantified mass fraction of all selected PAHs to the total OA mass. With no EGR, PAHs made up only a small fraction of the total in-cylinder OA. With EGR, PAHs made up a considerable fraction of OA in the soot formation dominated phase. The PAH fraction of OA was 18.1% at 13% inlet O 2 concentration and 8.9% at 15% inlet O 2 concentration in the soot formation phase. In the soot oxidation phase and exhaust, the PAH fraction of OA was low at all three EGR levels but increased with EGR.
The in-cylinder PAH fraction of OA was strongly correlated with the fullerene-carbon signal fraction of rBC (r = 0.91) which suggests that these form from similar in-cylinder combustion processes. Soot formation is linked to the formation of moderately sized PAHs. 9 Considering only the growth of PAHs and soot particles, we propose that the low concentration of PAHs (MW 202-300) at 21% inlet O 2 concentration in the soot formation phase was a result of faster particle inception and surface reactions rates (i.e., a much more rapid soot formation) than PAH formation rates. On the other hand, we propose that at 15% and 13% inlet O 2 concentrations, the elevated PAH concentrations were the direct result of lower combustion temperatures, which allowed the PAH concentration to accumulate due to slower particle inception and reactions with soot surfaces. A previous incylinder PAH diagnostics study using laser-induced fluorescence showed that the PAH residence time in the flame-jet before the onset of soot formation increased drastically with increasing EGR, from tens of microseconds with no EGR to milliseconds with high EGR (12.7% O 2 ). 16 Extending the argumentation to include soot properties, when introducing EGR, longer residence times of PAHs and soot precursors coupled with lower flame temperatures would also be important for the formation of fullerenic nanostructures.
Finally, we observed variations in the optical properties of the in-cylinder soot (Figure 4e) by analyzing the absorption angstrom exponent (AAE), a measure of the absorption wavelength dependency of the particles. The AAE was highest in the early stages of combustion and progressively became smaller as the combustion proceeded. Remarkable transitions in AAE were observed when reducing the inlet O 2 concentration. In the soot formation dominated phase, the AAE was 1.7 at 13% inlet O 2 concentration, 1.3 at 15% inlet O 2 concentration, and 1.0 at 21% inlet O 2 concentration. In the soot oxidation dominated phase, AAE was 1.1, 1.0, and close to 0.8 at 13%, 15%, and 21% inlet O 2 concentrations, respectively. The order was similar in the exhaust but the differences were smaller. At 21% inlet O 2 concentration in the late combustion cycle, the particle concentrations were very low and the AAE measurement was most likely affected by the background particles (see SI) that were released from the sampling valve without combustion in the motored mode. These background particles had a larger size in the SMPS and appeared to be associated with a lower AAE than the particles in the exhaust.
We conclude that the low AAE found in the diesel exhaust soot was due to a gradual decrease in AAE as the combustion proceeded. These variations in AAE resemble the trends associated with increasing graphitization of the soot. Correlating the observed in-cylinder soot AAE with SP-AMS data, we concluded that the overall OA to BC ratio of the in-cylinder diesel soot showed a low correlation (r = 0.08) with the observed changes in AAE (OA originates mostly from lubrication oil). However, when grouped into their respective OA families, the fraction of OA belonging to aliphatic-like fragments (C x H y>x + ) was inversely correlated (r = −0.82) with AAE, while the fraction of PAHs (r = 0.93), aromatic-like (C x H y≤x + ) fragments (r = 0.84), and oxidized organic (C x H y O z + ) fragments (r = 0.68) were positively correlated with AAE. In addition to nonrefractory components, a strong correlation between AAE and the fullerene-carbon fraction (r = 0.91) and midcarbon fraction of rBC (r = 0.90) was observed.
It has previously been shown that variations in AAE can be related to nanostructural differences, with reported values of AAE ≈ 2 for fullerene soot 52 and AAE ≈ 2 for spark discharge soot with high curvature nanostructure. 53 Fullerene-carbons were exclusively observed during IR laser vaporization with the SP-AMS. The correlation between fullerene-carbons and AAE also suggests that the intrinsic properties of the soot core could have caused the variations in AAE, in addition to variations caused by UV-absorbing organics such as PAHs.
Implications. Evidence of fullerenic soot signatures in SP-AMS mass spectra from ambient air have been recently reported. 54 This stresses the importance of understanding the mechanisms behind the formation of fullerenic soot and

Environmental Science & Technology
Article associated organics such as PAHs. In addition, it merits further studies on the removal efficiency and transformation of fullerenic soot in DPFs, as well as on the toxicological responses, atmospheric transformation and climate relevance of such emissions.
Previous studies with EGR have highlighted an increased reactivity toward oxidation in exhaust soot. 55,56 The presence of fullerene-carbon clusters in our SP-AMS mass spectra implies that EGR leads to the formation of high tortuosity (high curvature) soot nanostructures. To elucidate the formation mechanisms of fullerenic soot in diesel engines, studies on soot formation using well-defined premixed flames 13 and shock tubes 57 may thus prove highly relevant. We found that the fullerene signals were reduced as combustion proceeded. Future studies should evaluate if this is due to preferential removal by oxidation of these structures or to a gradual conversion toward more graphitized soot nanostructures.
The results have important implications for emission control systems in modern vehicles. The oxidation reactivity of soot is a key input parameter when designing DPFs which can effectively reduce BC emissions. 58 It is well-known that the soot nanostructure affects the oxidation reactivity. 10,56 Diesel oxidation catalysts will reduce the PAH concentration. However, DPFs can increase the emissions of some nitro-PAHs. 59,60 The influence of EGR on soot nanostructure and concentration of PAHs merits further studies on how exhaust after-treatment can impact the final emitted aerosol properties. We also found altered optical properties of diesel soot with high fullerene-carbon signals. If these can be definitively linked to fullerenic nanostructures, it will affect the direct radiative forcing and the accuracy of BC source apportionment based on AAE for such emissions.
The blending of increasing fractions of oxygen containing FAME (fatty acid methyl esters) biodiesel into fossil diesel is currently encouraged to mitigate CO 2 emissions. FAME is associated with reduced particle emissions compared to fossil diesel 61 but was recently found to be associated with increased fractions of high tortuosity nanostructures 12 and increased in vitro toxicity. 62 The formation of high tortuosity fullerenic soot nanostructures has been linked to the premixing of fuel and oxygen in the soot formation zone. 13 EGR results in increased premixing of fuel and air, while biodiesel provides oxygen from the fuel. If diesel soot reactivity increases with EGR and biodiesel content, it may be possible to increase the efficiency of soot removal in DPFs.
We propose that future research should include in-cylinder sampling with the FSV and sampling at different stages in the exhaust aftertreatment system, combined with the SP-AMS and the mass-spectrometric signatures developed here. Such studies would involve novel engine concepts (e.g., low temperature combustion) and biodiesel fuels, and would have the potential to provide novel information on in-cylinder chemistry. This information can aid in the design of combustion systems, which ultimately influences emissions, climate effects and particle toxicity. In addition, adding online tandem mobility-mass measurements 63 downstream the FSV would allow simultaneous measurements of the evolution of particle composition and morphology over the combustion cycle.

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.6b03391. NO x vs CAD; Dilution system and experimental setup; The fast gas-sampling valve (FSV); SP-AMS setup, calibration and data analysis; SMPS mass weighted size distributions in late cycle and exhaust; rC x + and BC correlation; Adjusting for gas phase CO 2 and CHO; Origin of the organic aerosol: lube oil vs diesel; Refractory fullerene-carbon signal vs nonrefractory background signal (PDF) ■ AUTHOR INFORMATION Corresponding Author *Phone: +46 462220892; fax: +46 462224431; e-mail: vilhelm. malmborg@design.lth.se.

Notes
The authors declare no competing financial interest.

■ ACKNOWLEDGMENTS
This research was financed by Metalund, the FORTE Centre for Medicine and Technology for Working Life and Society, GenDies at the Competence Center for the Combustion Processes, Lund University and the Swedish Energy Agency (Project number 10738150289), and The Swedish Research Council VR project nr. 2013-5021. We acknowledge Volvo AB for providing the FSV hardware, Scania CV AB for providing the engine and Jan Eismark (Volvo AB), Timothy Benham (Chalmers University of Technology) and Mats Bengtsson (Lund University) for their technical support, Tim Onasch for discussions on the SP-AMS data, and Bengt Johansson for his contribution to the experimental procedure.