lable at ScienceDirect

Atmospheric Environment 101 (2015) 34e40

Contents lists avai

Atmospheric Environment

journal homepage: www.elsevier .com/locate/atmosenv

Metal nanoparticles in diesel exhaust derived by in-cylinder meltingof detached engine fragments

Anthi Liati a, *, Sushant Sunil Pandurangi b, Konstantinos Boulouchos b, Daniel Schreiber a,Yadira Arroyo Rojas Dasilva c

a EMPA Material Science and Technology, Laboratory of Internal Combustion Engines, CH-8600 Dübendorf, Switzerlandb ETH Zürich, Aerothermochemistry and Combustion Systems Laboratory, 8092 Zürich, Switzerlandc EMPA, Electron Microscopy Center, CH-8600 Dübendorf, Switzerland

h i g h l i g h t s

� Traffic-related metal nanoparticle emissions can pose environmental and health concerns.� Fe3O4 nanoparticle agglomerates occur in diesel exhaust.� Steel fragments detached from cylinder engine parts can melt in the combustion chamber.� A hitherto neglected mechanism of metal nanoparticle formation from internal combustion engines is suggested.

a r t i c l e i n f o

Article history:Received 23 August 2014Received in revised form4 November 2014Accepted 6 November 2014Available online 7 November 2014

Keywords:Metal nanoparticlesReacting flow CFDCombustion-generated PMDiesel PMTraffic-related atmospheric pollutantsParticle heat transfer

* Corresponding author.E-mail address: anthi.liati@empa.ch (A. Liati).

http://dx.doi.org/10.1016/j.atmosenv.2014.11.0141352-2310/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

A wide range of environmental and health effects are linked to combustion-generated pollutants relatedto traffic. Nanoparticles, in particular, are a major concern for humans since they can be inhaled and havepotentially toxic effects. The variability and sources of combustion-related nanoparticle pollutantsremain inadequately investigated. Here we report the presence of ca. 5e100 nm large Fe3O4 nano-particles, in form of agglomerates, in diesel exhaust. The mode of occurrence of these nanoparticles, incombination with their chemical composition matching that of steel indicate that they derive by meltingof engine fragments in the combustion chamber and subsequent crystallization during cooling. Toevaluate this hypothesis, we applied CFD simulations of material transport in the cylinder of a dieselengine, assuming detachment of steel fragments from various sites of the cylinder. The CFD results showthat fragments �20 mm in size dislodged from the piston surface or from the fuel nozzle interior can beindeed transported to such hot areas of the combustion chamber where they can melt. The simulationresults concur with the experimental observations and point out that metal nanoparticle formation byin-cylinder melting of engine fragments can occur in diesel engines. The present study proposes ahitherto neglected formation mechanism of metal nanoparticle emissions from internal combustionengines raising possible environmental and health concerns, especially in urban areas.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

There is compelling evidence that ambient particulate matter(PM) related to traffic poses environmental and health risks and isassociated with an increase in respiratory-related mortality andmorbidity (e.g., Barregard et al., 2006; McCreanor et al., 2007; Popeet al., 1995). The role of the metal-bearing fraction of ambient airparticles is of particular importance. Transition metals (e.g. Fe, Ni,

Cu), in particular, are most likely to have adverse health effectswhen they are inhaled, i.e. of sub-micrometer size, due to theirpotential to produce reactive oxygen species in biological tissues(Kelly and Fussell, 2012; Smith et al., 2000). Sub-micrometer largeparticles have a greater surface area per unit mass, a longer resi-dence time in the atmosphere and can penetrate deeper into thelungs. In addition, metal elements can be adsorbed to atmosphericPM and be deposited to soil, water, and plant leaves via wet and drydeposition.

The exhaust of diesel combustion engines is known to contain aminor fraction of metal-bearing compounds (the so-called ash),apart from abundant carbonaceous components (soot). Ash forms

Fig. 1. (A) Schematic illustration of coupling between the flow-field solver (STAR-CD)and the in-house combustion and reacting flow solver based on conditional momentclosure; (B) Illustration of the mesh used for numerical simulations.

A. Liati et al. / Atmospheric Environment 101 (2015) 34e40 35

less than 1% of the total PM emissions (e.g., Sharma et al., 2005) andoriginates primarily from combustion of low amounts of metals(e.g., Zn, Ca, Mg) added into the lubricating oil to protect enginecomponents. Furthermore, metal fragments (e.g. Fe-, Ni-, Cu-, Cr-,Sb-bearing) abraded mechanically from diverse parts of the engineand/or the exhaust after-treatment systems are also found in theexhaust of diesel engines (e.g., Sappok et al., 2012). Importantly, ashoccurs in the exhaust of all types of internal combustion engines,thus increasing its total absolute amount in the atmosphere andtherefore its contribution to air pollution and human health effects.The nature, morphology and variability of the metallic fraction ofcombustion-generated PM remain poorly understood.

During an experimental study dealing with particle emissions inthe exhaust of diesel engines at Empa Laboratories, we found bymeans of transmission electron microscopy, among other metallicphases (CaePeZneMgeS-bearing ones related to lube oil andCreNieFe-bearing mechanical fragments related to engine wear),branch-like metallic agglomerates of newly formed Fe-oxidenanoparticles. Formation of the above nanoparticles requires veryhigh temperatures that can cause melting of Fe-bearing enginefragment precursors such as steel. Such high temperatures (�ca.1800 K; approx. steel melting point) are encountered only in thecombustion chamber of the engine.

Within the framework of the present study we (a) describe thedetailed electron microscopic characteristics of the Fe-oxidenanoparticles sampled from the diesel exhaust; (b) apply compu-tational fluid dynamic (CFD) simulations to examine whether steelfragments detached from possible sites of the cylinder engine canbe indeed transported to areas of the combustion chamber wherethey can melt, and (c) constrain the possible sources and sizes ofthe steel fragment precursors that can melt to produce theobserved Fe-oxide nanoparticles. For this purpose, we simulatedthe trajectories of steel particle fragments of different sizes undervarious scenarios.

2. Experimental

2.1. Sampling

The experiments were carried out on a chassis dynamometerwith an Iveco Daily vehicle (MY 2003; Euro IV emission limits)equipped with a 2.3-L 4-cylinder (F1A) common rail diesel enginewith turbocharger using commercial fuel (S content <10 ppm) andcommercial lubricating oil and fitted with a diesel oxidation cata-lyst and a diesel particulate filter. For details on the experimentalsetup the reader is referred to (Liati et al., 2013). Sampling of un-diluted exhaust gas was performed under normal operating con-ditions during steady state operation at 2000 rpm, 13 kW enginespeed (70 km/h) and output, respectively using a pump at 0.5 L/min. The temperature of the exhaust gas was approximately 260 �C.

The PMwas collected from the exhaust stream directly on 3mmcopper-supported carbon-coated TEM grids by means of an elec-trostatic particle sampler. The sampling site near the TEM grids washeated to prevent condensation. The sampled material was sub-sequently analysed by TEM.

2.2. Analytical techniques

For TEM-imaging, a JEOL 2200FS microscope with an Omegafilter, a Schottky field emission gun at 200 kV and a point to pointresolution of 0.23 nm at the Electron Microscopy Center of EMPAwas used. The TEM instrument is equipped with an EDX detectorfor elemental analysis. TEM analysis was performed using STEMmode (bright field (BF) and dark field (DF)), high resolution(HRTEM) and X-Ray diffraction.

2.3. Simulations e CFD of reacting flow

A CFD simulation was carried out to determine the flow andtemperature distribution in time, on the example of a single-cylinder MTU engine at ETH Zürich, for which more details onthe overall experimental setup can be found in e.g. (Brueckner,2014). The commercial CFD solver STAR-CD from CD-adapco Inc.was employed coupled with an in-house code for the turbulentcombustion model; the coupling is described schematically inFig. 1A, while the computational geometry and CFD mesh areshown in Fig. 1B. STAR-CD is a finite-volume based code thatcomputes the NaviereStokes equations for momentum and energy.The standard PISO solution scheme is employed with the pro-prietary MARS discretization scheme built into STAR-CD for solvingthe momentum equations. CHEMKIN is used to obtain chemicalreaction rates (ReactionDesign, 2014), which are fed to the CMCcode along with pressure and velocity information from STAR-CD,in order to compute the transport and evolution of all chemicalspecies; these in turn are fed back to STAR-CD to continue theiteration cycle. The CMC code adopts a mix of central differencingand TVD schemes for stability. This finite-difference code, detailedin Wright et al. (2005), operates on a coarser CMC-grid synchro-nized with the CFD mesh. The CFD mesh consists of hexahedralcells with a resolution of 1 mm3 in the piston bowl resulting inapproximately 50,000 cells at top-dead-centre position; at bottom-

A. Liati et al. / Atmospheric Environment 101 (2015) 34e4036

dead-centre the mesh grows to approximately 240,000 cells. Thepolar configuration of the CFD mesh follows the same approach asin Bolla et al. (2014) where a mesh sensitivity analysis has beenperformed.

The operating point considered was 10 bar bmep with inletconditions 18 �C, 1.3 bar. Running at 1000 RPM, 6� of crank-anglemeasurement equates to 1.0 ms of time. Most importantly, theengine is equipped with a state-of-the-art common rail fuel injec-tion system and is operated at speeds and loads representative ofthe operating maps of typical commercial engines. Therefore, thecombustion and overall thermodynamic behaviour reflect actualusage conditions to a large extent.

The methodology followed in general for engine simulations iselucidated in other reports (Bolla et al., 2013a; Farrace et al., 2013)and is only briefly repeated here. In the present study, the fuel spraybegins at �5.7ºCA and ends at 11.3ºCA, lasting slightly over 2.0 ms.The interaction of turbulence and thermochemistry is crucial foraccurate prediction of heat release rates from the combustion, andits modelling has been the subject of much research (Bilger, 2000).The conditional moment closure (CMC) approach is used in thepresent study with an in-house code (Wright, 2005) that hasrelatively modest computational requirements and very good ac-curacy. The CMC approach has been demonstrated and validatedsuccessfully for different problems, such as single-phase turbulentreactive flows (non-premixed attached (Roomina and Bilger, 2001),piloted (Woolley and Fairweather, 2007) and lifted (Kim andMastorakos, 2005) turbulent flames), two-phase flows at condi-tions relevant for diesel engine applications (the Sandia combus-tion cell (Borghesi et al., 2011), ETH cell (Wright et al., 2010), Aachencell (Wright, 2005)) and in diesel engines (Bolla et al., 2013a;Farrace et al., 2013; Wright et al., 2009) also with multiple in-jections (Pandurangi et al., 2014). As CMC is a spatially-resolvedapproach, it can be used to predict spatial characteristics such asthe ignition location and has shown very good agreement withexperiments (Wright et al., 2010). In the present study, a simplifiedreactionmechanism of 22 chemical species in 18 reaction steps wasapplied (Liu et al., 2004) based on much previous success (Bollaet al., 2013a, 2013b; Farrace et al., 2013).

The simulation provides temperature, pressure, gas properties,and velocity, from �100� crank angle (CA) to þ120� CA (exhaustvalve opening); the primary validation against experiments is doneby means of the pressure which shows very good agreement asshown in Fig. 2A. Crank angles are measured with respect to thetop-dead-center (TDC) position which corresponds to 0� CA. Whileit is not possible to obtain the evolving temperature with currentexperimental methods, to aid the reader the average temperature

Fig. 2. (A) Comparison of pressure obtained by simulation (solid line) with measured presscombustion. The average temperature is close to the typical steel melting point; near the flameither side of the average temperature line.

in the cylinder obtained from the simulation is shown in Fig. 2B.Ignition delay is slightly over-predicted by the simulation: 7 ºCAcompared to 5 ºCA in the experiment, a commonly observed limi-tation of the compact (reduced) chemical reaction mechanism. Thechemical mechanism is obtained by Liu et al. by reduction of adetailed n-heptane reaction mechanism. For a discussion on thereduction approach and details on the elementary species and re-actions, the reader is referred to (Liu et al., 2004).

2.4. Simulations e particle tracking

After the base simulation was performed, the trajectory offragments under various scenarios was obtained by following theirmotion through the transient relative velocity field. The time oforigin of the fragment, and initial velocity were chosen according tothe considered scenario. The various scenarios are described inmore detail in Section 4 (Results of CFD Simulations). Beyond theassumptions made for the introduction of the fragments, no furtherboundary conditions are imposed on the fragments and their mo-tion is governed by drag forces depending entirely on the flow fieldwithin the engine relative to the fragment's own motion. Gasthermo-physical properties are obtained from CHEMKIN poly-nomials (ReactionDesign, 2014). For calculating heat transfer, theNusselt number is obtained by the following correlation (Bergmanet al., 2007):

Nu ¼ 2þ�0:4Re1=2 þ 0:06Re2=3

�� m

mS

�1=4Pr0:4

which holds as long as: 0.7 < Pr < 380; 3.5 < Re < 7.6 � 104; 1.0 < m/mS < 3.2

Pr ¼ mCp/k is the Prandtl number of the gas; m and mS are thedynamic gas viscosities at the local and the saturation temperaturerespectively; Cp is the specific heat capacity, and k the thermalconductivity. Convection and conduction are related by Nu ¼ hD/k;this relationship can be used in conjunction with the above cor-relation to evaluate h. The heat added is then dq ¼ hA(TP � T∞) dt,where TP is the particle temperature and T∞ the ambient. The role ofdq is twofold: (i) it raises the temperature until the melting tem-perature and (ii) it contributes towards the latent heat of melting ifthe melting temperature is reached.

The Boltzmann number indicates the ratio of radiative andconvective heat transfer by Bo ¼ sT3/rCPVR where s is the Ste-faneBoltzmann constant. In all scenarios in this study, Bo wasfound to be less than 0.1% and thus radiation heat transfer wasneglected. After the melting point is reached, an approximate

ure (dashed line); (B) plot of the temperature obtained by simulation during and aftere, temperatures are expected to be much higher. One standard deviation is marked on

A. Liati et al. / Atmospheric Environment 101 (2015) 34e40 37

melted fraction is found by assuming that melting is linear with anyfurther added heat. Specific heat for steel at the relevant hightemperatures was taken from measurements in literature (Umino,1935) as also were the melting point temperature and the latentheats of melting and solidification (Smetana et al., 2011) (1%carbon-steel assumed).

3. Results of TEM-imaging

The agglomerates observed under TEM are branch-like andconsist of nano-sized primary particles with spherical (ca.10e100 nm, mostly ca. 40e60 nm in diameter), square (ca.7e20 nm, rarely up to 40 nm) and irregular shapes (ca. 7e13 nm,rarely down to 5 nm) (Fig. 3AeC).

The square and irregular particles (strongly rounded squares;Fig. 3C) exhibit rounded corners and edges interpreted as me-chanical degradation (erosion) due to interaction of the newlyformed particle with a hot surrounding (probably fast flowing gas)after initial formation. The irregular shape of some of the particlesmay thus be a secondary feature. Alternatively, the lack of well-defined crystal edges may be a result of extremely short resi-dence time in the melted phase, in combination with minimisationof surface energy, which applies for very small particles (Li et al.,2007), comparable to those encountered here.

High resolution TEM images (Fig. 3C) reveal that all of the par-ticle types exhibit lattice fringes, which suggests that they arecrystalline. Energy dispersive X-ray TEM-analyses show that the

Fig. 3. (AeC): TEM images of Fe3O4 nanoparticle agglomerates collected from the exhaustsentative irregularly shaped particles with more or less rounded corners are indicated in (pattern obtained from the Fe-oxide nanoparticles. The bright dots correspond to reflectionmagnetite, giving a perfect fit. The numbers denote the lattice planes from the simulated a

nanoparticles of all shapes and sizes are composedmainly of Fe andO with small amounts of Mn, some Cr and sometimes traces of Sb.TEM-diffraction analyses indicate that the analysed particles, irre-spective of their shape, match perfectly with magnetite (Fe3O4).Magnetite belongs to the cubic system (space group Fd-3m). Theidentification of the polycrystalline diffraction pattern (Fig. 3D) wasperformed by applying the JEMS software (http://cimewww.epfl.ch/people/stadelmann/jemsWebSite/jems.html).

The Fe3O4 agglomerates observed here are different frommetallic particle types previously described in the literature fordiesel exhaust emissions in that (a) they exhibit morphologicalfeatures indicating new formation resembling nanoparticle ag-glomerates produced synthetically for commercial purposes byindustrial aerosol processes involving flame spray pyrolysis (e.g., Liet al., 2007; Strobel and Pratsinis, 2009). The combustion chamberof internal combustion engines probably represents an analogousformation environment; (b) their morphological features, in com-bination with their good fit with the composition of steel (Fe withsome Mn and Cr), show that they derive from melted enginefragments and not from metal additives in the lubricating oil, as isusually the case for chemically formed metallic diesel emissions(e.g., Sappok and Wong, 2007).

4. Results of CFD simulations

CFD simulations can reveal probable scenarios for the origin ofmetallic fragment precursors of the Fe3O4 nanoparticles, and their

stream showing spherical, square and irregularly shaped particle constituents; repre-C) by the arrows and shown in the inset image of (C); (D): Polycrystalline diffractions defining lattice planes. The circles correspond to the simulated diffraction pattern ofnd experimental diffraction patterns.

Table 1Summary of location and timing of detachment of the steel fragments.

Location Timing(s) Total scenarios

(1) Start of injectionNozzle (2) Midway during injection 3

(3) End of injectionLocation of flame impingementTop-most point (4) At time of impingement 2Bottom-most point (5) At time of impingement

Inlet valve (6) At time of shutting of valve 1(7) When piston is at top-dead-center

Cylinder walls (8) When piston is at 10� before TDC 3(9) When piston is at 20� before TDC

Fig. 5. Graphical illustration summarising all scenarios tested for eventual steelmelting of fragments detached from different engine parts. The black shaded areasshow the time in milliseconds spent in conditions conducive to Fe3O4 formation:oxygen-rich, T � 1800 K (melting point). *Scenarios for inlet valve and cylinder wall (atthree different initiation times) are depicted together. Numbers on the extreme leftindicate the considered fragment size in micrometers.

A. Liati et al. / Atmospheric Environment 101 (2015) 34e4038

likely fate inside the combustion chamber. The trajectory of such afragment is traced in the reactive turbulent flow inside a typicaldiesel engine cylinder (Fig. 4), and its temperature and the localoxygen concentration along this path are determined (see above,Section 2.3.).

The different scenarios examined for the source of the metallicfragments are: impingement of the flame on the piston surface;abrasion on the cylinder inner walls; friction or cavitation in thefuel nozzle; and shutting impact of the inlet valve (Fig. 4). Theoutlet valve (opening or shutting) is not considered explicitly;fragments from the outlet valve may only be affected by the nextcombustion cycle and will, even in that case, behave very similarlyto the inlet valve fragments. The simulated fragments are consid-ered to be spherical. The following locations and timings ofdetachment of the fragments are examined: three scenarios for thefuel nozzle (at the start of injection, midway during injection and atthe end of injection); two scenarios involving the flame hitting thepiston (top and bottom extreme locations as shown in Fig. 4); onescenario from the inlet valve; and three scenarios from the cylinderwalls considering fragments originating at different times. The totalduration of the individual scenarios (until exhaust begins) isslightly less than 20 ms. Details on the location and timing ofdetachment of the fragments are summarized in Table 1.

The results (summarised in Fig. 5) reveal that steel fragmentprecursors of the observed nanoparticles can originate from (i) thefuel nozzle, at either the exact start of injection or shortly after, and/or (ii) the piston-bowl, at either of the two extreme points on thesurface hit by the flame (Fig. 6).

Note that the actual detachment/chipping process of the metalis not simulated, but assumed to occur instantly and yield a frag-ment. A fragment from the nozzle undergoes complete melting;especially so if it is conceived somewhat later than the exact start ofinjection (Fig. 6) since the maximum flow rate is reached after asmall time lag, which is the case for most. typical diesel injectors.The short residence times at high temperature (a fewmilliseconds)seen in both the scenarios (i) and (ii) are conducive to the formationof Fe3O4, and not to the thermodynamically more stable Fe2O3which requires rather longer residence times in similar oxygenavailability to form (Strobel and Pratsinis, 2009). Fragments from(iii) the cylinder inner wall, at all considered initiation times, and(iv) the inlet valve are not likely tomelt, because from these sourcesthere is a lack of transport into hot regions and therefore insuffi-cient heating for melting. The size (diameter) of the detachedfragments is not calculated a priori; rather a wide range is testedand the sizes not resulting in melting are ruled out, to arrive at awindow of 10e20 mm as the possible size range. Fragments above20 mmdiameter are toomassive to be heated sufficiently. Theymay,however, melt only partially at their outer surface, evenwithout theentire fragment reaching the melting temperature. Those below10 mm are expected to melt completely because of low mass andeasy displacement. However, due to lack of heat transfer correla-tions in that size range, they have not been further considered here.

Fig. 4. Illustration of a cross-section of a typical internal combustion engine (right) with a mmay be detached and eventually melt, shown as used for the simulations.

Correlation of the fragment's own temperature with the equiv-alence ratio in the immediate vicinity, a measure of the locallyavailable ambient oxygen, shows that in the favourable scenarios,oxygen availability coincides with the fragment being at or abovethe melting temperature, and thus in a partially or fully meltedstate. ‘Oxygen-rich’ implies that there is, locally, higher oxygenconcentration than required for the stoichiometric combustion offuel at the same location, i.e. excess oxygen is present which is thenpresumed to be available for Fe3O4 formation.

5. Discussion

The type, mode of occurrence and way of formation of the Fe3O4nanoparticles described in the present work are, to the best of our

agnified view of a typical fuel nozzle (left) exhibiting candidate sites where fragments

Fig. 6. Evolution of the fragment temperature (solid black lines) and fragment meltingfraction (shaded grey). Time (x-axis) is measured from the moment the piston reachesits top extremity (top-dead-center) i.e. when expansion begins. Sources of steel frag-ment: (a) fuel nozzle at start of injection; (b) fuel nozzle midway during injection; (c)piston-bowl at bottom point of flame hit; (d) piston-bowl at top point of flame hit. Foreach source, two considered fragment sizes are shown (10 mm, left side diagrams;15 mm, right side diagrams).

A. Liati et al. / Atmospheric Environment 101 (2015) 34e40 39

knowledge, not reported in the literature, neither for diesel exhaustnor for the exhaust of any other type of internal combustion engine.The different scenarios examined in the CFD simulations carriedout on the trajectories of steel particles in the engine cylinder revealthat steel fragments �20 mm in size dislodged from (a) the pistonsurface by a repeatedly hitting flame (flame impingement), and (b)the fuel nozzle can indeed pass through high-temperature andoxygen-rich areas during the combustion phase, where they canpartly or fully melt to form, during cooling, the observed Fe3O4nanoparticles which may finally escape into ambient air.

The occurrence of Fe3O4 nanoparticles of different shapes in thecollected samples shows that not all particles formed under iden-tical conditions. Very rapid cooling of the melted material can leadto spherical shapes (Li et al., 2007; Rudin and Pratsinis, 2012), whilelonger residence times under high temperatures can result in themore stable (cubic) structure. An analogous phenomenon isobserved during the evolution of particle shapes from near-spherical to hexagonal/octagonal platelets, traced by the gradualincrease in flame temperature and prolonged residence time dur-ing particle synthesis by flame reactors (Li et al., 2007). Part of theFe3O4 cubic particles observed here may have spent sufficient timethrough the reacting flame to interact mechanically with the hotsurrounding air and be partly eroded. This procedure may beresponsible for the rounding of the corners and edges observed forsome particles. In the case of the smallest (5e13 nm large) particlesof that kind, shorter residence time and minimum surface energymay also account for their rounded shape.

Both the presence of this type of metal nanoparticles in theexhaust of vehicles, as well as the mechanism of formation were asyet overlooked when considering air pollution studies related totraffic. The abundance of such particles or their proportion to the

total of emitted diesel particles cannot be determined by theelectron microscopy method applied in this study, as this methodfocuses on crucial small scale details, down to the nanometer range,on cost of rigorous statistics. It can be, however, ascertained thatsuch particles are subordinate compared to other types of ashparticles in the exhaust (lube-oil derived, fuel-derived or me-chanically abraded -non-melted- metal fragments).

Importantly, the nature of the formed nanoparticles may not beconstrained to Fe3O4 but include also other material used for thefabrication of combustion engine components, such as Al (highlyenriched in the so-called Y-alloy consisting of 92% Al), fragments ofwhich can experience greater heating due to lower densitycompared to steel. It is worth mentioning also that although thepercentage of Fe-oxide nanoparticles among particle emissions indiesel vehicles is probably quite low, such particles (or equivalentones with a composition corresponding to that of the enginecomponents) may be produced by vehicles applying other internalcombustion engine technologies (e.g. gasoline cars). In that case,the absolute contribution of metal oxides formed by in-cylindermelting of engine fragments to the total vehicle emissions wouldbe higher than hitherto presumed. Metal oxide emissions are ex-pected, for instance, to be higher in urban areas due to excessiveabrasion (low lubrication effect) attributed to stop-and-go drivingconditions, short driving distances and/or cold start. Detailedstudies on metal emissions from vehicles are generally restrictedbut they all stress the necessity for further investigations on metalparticle emissions from vehicles (Liati et al., 2013; Lim et al., 2009;Mayer et al., 2012). Regarding the health effects of Fe-oxides,although not many studies emphasise their presence and/orimportance in diesel exhaust, potential concerns for human healthhave been raised for such particles (Lee et al., 2006; Miller et al.,2007).

Last but not least, it should be noted that the recognition of theFe3O4 particles and the identification of their detailed characteris-tics, as described in the present work, would not have been possiblewithout the electron microscopic methods applied in the presentstudy. This combined experimental and simulation study empha-sises the necessity of electron microscopy in the research of vehicleexhaust, as well as the support and importance of simulations,which substantiate the inference that melting of engine wear in thecylinder is indeed possible.

6. Conclusions

The experimental and CFD simulation results of the presentwork lead to the following conclusions:

1. Fe3O4 nanoparticles forming branch-like agglomerates wereidentified in diesel exhaust.

2. Based on CFD simulations, steel fragments dislodged from thepiston surface and/or the fuel nozzle can enter hot areas of thecombustion chamber where they can melt. Subsequent coolingand crystallization in oxygen-rich surroundings can result in theformation of Fe3O4 nanoparticles.

3. Metal nanoparticles formed by in-cylinder melting of enginefragments could be an additional source of air pollution.Although generally subordinate compared to other types of ashparticles in the exhaust, such nanoparticles may be morefrequent in urban areas, due to expected excessive abrasionowing to the driving style and probably need to be taken intoaccount in studies dealing with air pollution related to traffic.

The experimental findings of this work hold true, in a strictsense, for the engine investigated. However, the fact that thesimulation results strongly support and validate the TEM findings

A. Liati et al. / Atmospheric Environment 101 (2015) 34e4040

strengthens the significance of the experimental results and em-phasizes the need for further testing and more accurate evaluationof metal particle emissions. Such studies would be even morerequired for vehicles not equipped with particle filters, such asgasoline cars.

Acknowledgements

Part of the work of AL and DS was supported by the CompetenceCenter for Energy and Mobility (CCEM) Project 51-NADiP: ‘NOxAbatement in Diesels’. SP and KB would like to acknowledge thesupport of CCEM project 711 ‘In-Cylinder Emission Reduction in LargeDiesel Engines’. AL wishes to thank Jesper Knijnenburg, ParticleTechnology Laboratory, ETH Zurich, for discussions on iron nano-particle formation and characteristics.

References

Barregard, L., S€allsten, G., Gustafson, P., Andersson, L., Johansson, L., Basu, S.,Stigendal, L., 2006. Experimental exposure to wood-smoke particles in healthyhumans: effects on markers of inflammation, coagulation, and lipid peroxida-tion. Inhal. Toxicol. 18, 845e853.

Bergman, Lavine, Incropera, DeWitt, 2007. Fundamentals of Heat and Mass Transfer.Wiley.

Bilger, R.W., 2000. Future progress in turbulent combustion research. Prog. EnergyCombust. Sci. 26, 367e380.

Bolla, M., Farrace, D., Wright, Y.M., Boulouchos, K., 2014. Modeling of soot formationin a heavy-duty diesel engine with conditional moment closure. Fuel 117,309e325.

Bolla, M., Gudmundsson, T., Wright, Y., Boulouchos, K., 2013a. Simulations of dieselsprays using the conditional moment closure model. SAE Int. J. Engines 6.

Bolla, M., Wright, Y., Boulouchos, K., Borghesi, G., Mastorakos, E., 2013b. Soot for-mation modeling of n-heptane sprays under diesel engine conditions using theconditional moment closure approach. Combust. Sci. Technol. 185, 766e793.

Borghesi, G., Mastorakos, E., Devaud, C.B., Bilger, R.W., 2011. Modeling evaporatingeffects in the conditional moment closure for spray autoignition. Combust.Theory Model. 15, 725e752.

Brueckner, C., Kyrtatos, P., Boulouchos, K., 2014. Extending the NOx reduction po-tential with miller valve timing using pilot fuel injection on a heavy-duty dieselengine. SAE Int. J. Engines 7 (4), 1838e1850. http://dx.doi.org/10.4271/2014-01-2632.

Farrace, D., Bolla, M., Wright, Y.M., Boulouchos, K., 2013. Predicting in-Cylinder sootin a heavy-duty diesel engine for variations in SOI and TDC temperature usingthe conditional moment closure model. SAE Int. J. Engines 6, 1580e1593.

Kelly, F.J., Fussell, J.C., 2012. Size, source and chemical composition as determinantsof toxicity attributable to ambient particulate matter. Atmos. Environ. 60,504e526.

Kim, I.S., Mastorakos, E., 2005. Simulations of turbulent lifted jet flames with two-dimensional conditional moment closure. Proc. Combust. Inst. 30, 911e918.

Lee, D., Miller, A., Kittelson, D., Zachariah, M.R., 2006. Characterization of metal-bearing diesel nanoparticles using single-particle mass spectrometry.J. Aerosol Sci. 37, 88e110.

Li, D., Teoh, W.Y., Selomulya, C., Woodward, R.C., Munroe, P., Amal, R., 2007. Insightinto microstructural and magnetic properties of flame-made [gamma]-Fe2O3nanoparticles. J. Mater. Chem. 17, 4876e4884.

Liati, A., Schreiber, D., Dimopoulos Eggenschwiler, P., Arroyo Rojas Dasilva, Y., 2013.Metal particle emissions in the exhaust stream of diesel engines: an electronmicroscope study. Environ. Sci. Technol. 47, 14495e14501.

Lim, J., Lim, C., Yu, L.E., 2009. Composition and size distribution of metals in dieselexhaust particulates. J. Environ. Monit. 11, 1614e1621.

Liu, S., Hewson, J.C., Chen, J.H., Pitsch, H., 2004. Effects of strain rate on high-pressure nonpremixed n-heptane autoignition in counterflow. Combust.Flame 137, 320e339.

Mayer, A., Czerwinski, J., Kasper, M., Ulrich, A., Mooney, J.J., 2012. Metal OxideParticle Emissions from Diesel and Petrol Engines. SAE International.

McCreanor, J., Cullinan, P., Nieuwenhuijsen, M.J., Stewart-Evans, J., Malliarou, E.,Jarup, L., Harrington, R., Svartengren, M., Han, I.-K., Ohman-Strickland, P.,Chung, K.F., Zhang, J., 2007. Respiratory effects of exposure to diesel traffic inpersons with asthma. N. Engl. J. Med. 357, 2348e2358.

Miller, A., Ahlstrand, G., Kittelson, D., Zachariah, M., 2007. The fate of metal (Fe)during diesel combustion: morphology, chemistry, and formation pathways ofnanoparticles. Combust. Flame 149, 129e143.

Pandurangi, S.S., Frapolli, N., Bolla, M., Boulouchos, K., Wright, Y.M., 2014. Influenceof EGR on post-injection effectiveness in a heavy-duty diesel engine fuelledwith n-heptane. SAE Int. J. Engines 7.

Pope, C.A. 3rd, Bates, D.V., Raizenne, M.E., 1995. Health effects of particulate airpollution: time for reassessment? Environ. Health Perspect. 103, 472e480.

ReactionDesign, 2014. CHEMKIN III.Roomina, M.R., Bilger, R.W., 2001. Conditional moment closure (CMC) predictions of

a turbulent methane-air jet flame. Combust. Flame 125, 1176e1195.Rudin, T., Pratsinis, S.E., 2012. Homogeneous iron phosphate nanoparticles by

combustion of sprays. Ind. Eng. Chem. Res. 51, 7891e7900.Sappok, A., Kamp, C., Wong, V., 2012. Sensitivity analysis of ash packing and dis-

tribution in diesel particulate filters to transient changes in exhaust conditions.SAE Int. J. Fuels Lubr. 5, 733e750.

Sappok, A.G., Wong, V.W., 2007. Detailed Chemical and Physical Characterization ofAsh Species in Diesel Exhaust Entering After Treatment Systems. SAEInternational.

Sharma, M., Agarwal, A.K., Bharathi, K.V.L., 2005. Characterization of exhaust par-ticulates from diesel engine. Atmos. Environ. 39, 3023e3028.

Smetana, B., �Zaludova, M., Zla, S., Matekja, V., Dobrovska, J., Gryc, K.,Tkadleckova, M., Sikora, V., 2011. Latent heats of phase transitions of Fe-C basedmetallic systems in high temperature region. In: 20th Anniversary InternationalConference on Metallurgy and Materials, Brno, Czech Republic.

Smith, K.R., Veranth, J.M., Hu, A.A., Lighty, J.S., Aust, A.E., 2000. Interleukin-8 levelsin human lung epithelial cells are increased in response to coal fly ash and varywith the bioavailability of iron, as a function of particle size and source of coal.Chem. Res. Toxicol. 13, 118e125.

Strobel, R., Pratsinis, S.E., 2009. Direct synthesis of maghemite, magnetite andwustite nanoparticles by flame spray pyrolysis. Adv. Powder Technol. 20,190e194.

Umino, S., 1935. On the Specific Heat of Iron-carbon System at High Temperatures,and the Heat Changes Accompanying Those of Phase. The Science Reports of theTohoku University, pp. 665e793.

Woolley, R.M., Fairweather, M., 2007. First- and second-order elliptic conditionalmoment closure calculations of piloted methane diffusion flames. Combust.Flame 150, 92e107.

Wright, Y.M., 2005. Numerical Investigation of Turbulent Spray Combustion withConditional Moment Closure. Swiss Federal Institute of Technology, Zurich,Switzerland.

Wright, Y.M., Boulouchos, K., De Paola, G., Mastorakos, E., 2009. Multi-dimensionalConditional Moment Closure Modelling Applied to a Heavy-duty Common-railDiesel Engine. SAE, 2009-01-07.

Wright, Y.M., De Paola, G., Boulouchos, K., Mastorakos, E., 2005. Simulations of sprayautoignition and flame establishment with two-dimensional CMC. Combust.Flame 143, 402e419.

Wright, Y.M., et al., 2010. Experiments and simulations of n-heptane spray auto-ignition in a closed combustion chamber at diesel engine conditions. FlowTurbul. Combust. 84.