You are on page 1of 9

Atmospheric Pollution Research 10 (2019) 294–302

HOSTED BY Contents lists available at ScienceDirect

Atmospheric Pollution Research


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

Numerical study of ultrafine particles dispersion in the wake of a cylinder T


a,d,∗ a b c c d
N.S. Keita , A. Mehel , F. Murzyn , A. Taniere , B. Arcen , B. Diourte
a
Department of Mechanical and Environmental Engineering, Air Quality and Depollution Group, ESTACA, Paris Saclay Campus, 78180, Montigny-le-Bretonneux, France
b
Department of Mechanical and Environmental Engineering, Air Quality and Depollution Group, ESTACA, West Campus, 53000, Laval, France
c
Université de Lorraine, CNRS, LEMTA, F-54000, Nancy, France
d
Physics Department, Faculty of Science and Technology Bamako, USTTB, Mali

A R T I C LE I N FO A B S T R A C T

Keywords: The dynamics of ultrafine particles are strongly dependent to the air flow topology. In this study, the dispersion
Ultrafine particles dispersion of carbon nanoparticles in the wake of a cylinder placed in an air flow at Reynolds number of 9300 is nu-
Cylinder wake flow merically investigated. An Eulerian/Lagrangian approach is used to compute the two-phase flow; the carrier
Turbulence phase is predicted through a RANS (Reynolds Averaged Navier-Stokes) model while a particle Lagrangian
Reynolds stress model
tracking method provides the trajectory of each nanoparticle. The present numerical investigation allows
Brownian diffusion
studying the influence of Brownian and turbulence effects on the nanoparticles dispersion. The main results
reveal that: (1) The Brownian diffusion tends to concentrate the carbon nanoparticles in the form of a filament at
the periphery of the vortices that are generated in the cylinder wake. (2) The turbulence increases the lateral
dispersion of the nanoparticles. It disperses them from the edge to the core of the vortices. (3) The highest
concentration levels decrease from the near wake of the cylinder to the far wake region while the lateral dis-
persion increases.

1. Introduction concentration in the urban environment (Johansson et al., 2007; Pey


et al., 2009). Furthermore, 99% of these particles in the atmospheric
Ultrafine particles have attracted attention from the scientific urban environment have a size which is less than 300 nm (Kumar et al.,
community due to their negative impacts on human health (Donaldson 2008a, 2008b, 2008c, and 2009). It is for this reason that the most
et al., 2005), urban visibility (Jacobson, 2005) and global climate important metric to assess the effects of nanoparticles on human health,
(IPCC, 2007). Indeed, some recent studies have reported high con- besides the exposition duration, may be the number concentration of
centrations of ultrafine particles (UFP, diameter below 100 nm) in the nanoparticles rather than their mass concentration. However, it is re-
vicinity of roadways and urban streets (Zhu et al., 2002; Kittelson et al., levant to get a better understanding of their dynamics when released
2004). It is worthwhile to note that harmfulness of particles is linked into the atmosphere in order to characterize their ability to accumulate
with their properties and in particular with their size, the smallest ones in certain regions. This should be helpful to provide solutions to reduce
being the most dangerous. It has been demonstrated that inhaled UFP the risk of human exposure. Some studies have been conducted to assess
can penetrate into the blood system as well as in cells (Oberdörster the effect of street canyon and wind on UFP dispersion where they
et al., 2004) increasing significantly the risk of lung cancer and cardi- showed that the increase of the aspect ratio (building height H/street
ovascular diseases (Pope et al., 2011). The study of Pope et al. (2002) width W) increase the UFP concentration accumulation in the bottom
regarding the health impact of such pollutant analyzed the mortality (Scungio et al., 2013; Hassan and Crowther, 1998; Kumar et al., 2008).
data over 500 000 individuals. It shows that every increase of 10 μg/m3 Indeed, their dynamics strongly depends on the structure and properties
in PM2.5 (particles with a diameter below 2.5 μm) causes hikes of 4%, of the flow (Kanda et al., 2006; Carpentieri et al., 2012), particularly
6% and 8% for the risk of all-causes, cardiopulmonary and lung cancer the vortices (Scungio et al., 2013; Hassan and Crowther, 1998; Kumar
mortality, respectively (Pope et al., 2002). et al., 2008; Mehel and Murzyn, 2015) and the turbulence (Mehel et al.,
One of the major sources of nanoparticles (or UFP) is the road ve- 2010; Keita et al., 2016). Since the wake flows of the emitting ground
hicles. They contribute up to 86% of the total particle number vehicles are strongly turbulent with the presence of vortices, they have

Peer review under responsibility of Turkish National Committee for Air Pollution Research and Control.

Corresponding author. Department of Mechanical and Environmental Engineering, Air Quality and Depollution Group, ESTACA, Paris Saclay Campus, 78180,
Montigny-le-Bretonneux, France.
E-mail address: namamoudou-sidiki.keita@estaca.fr (N.S. Keita).

https://doi.org/10.1016/j.apr.2018.08.006
Received 29 May 2018; Received in revised form 30 July 2018; Accepted 7 August 2018
Available online 10 August 2018
1309-1042/ © 2019 Turkish National Committee for Air Pollution Research and Control. Production and hosting by Elsevier B.V. All rights reserved.
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

strong influence on nanoparticles dispersion. Thus, in order to get a under the influence of the Brownian and the turbulence effects? And
better understanding of the interaction between nanoparticles and what are the influences of the distance to the cylinder (near and far
turbulent vortical structures, we focus our study on the wake flow de- wake)? The present paper aims to answering these questions. Therefore,
veloping downstream of a circular cylinder. It is a well-known flow (the we study the evolution in time of nanoparticles lateral dispersion
single phase flow, i.e. without particles), which the topology has been function and the concentrations profiles in the cylinder wake at dif-
widely investigated in the past decades both numerically and experi- ferent dimensionless positions (near wake x/d < 10 and far wake x/
mentally. Well-organized vortices appearing in the wake of the cylinder d > 10). Section 2 depicts the numerical methods and results are dis-
are named von Karman Street (Norberg, 1987; Beaudan and Moin, cussed in section 3. Conclusions and future works are suggested in the
1996). The main goal of the numerical studies was to characterize the last section of this paper.
different flow topology as a function of the Reynolds number based on
the cylinder diameter and fluid velocity. 2. Numerical methods
To describe the flow stability in the wake of a cylinder, three ranges
have been identified depending on the Reynolds number (Roshko, 2.1. Fluid phase
1954). This latter is defined by Re = U∞ d/ ν , where U∞ is the un-
disturbed inlet velocity, d is the diameter of a cylinder and ν is the The carrier phase is modeled using 2D unsteady Reynolds-Averaged
kinematic viscosity of the fluid. In the first one (Re < 49), two vortices Navier-Stokes (RANS) simulations. The RANS equations were solved
appear in the recirculation zone. In the second, two stages are identi- with the commercial software ANSYS FLUENT (2014) in order to pro-
fied: for the first one (49 < Re < 150), the flow becomes unsteady vide the mean fluid flow characteristics. Considering the fluid as
due to the separated shear layer and asymmetry. This corresponds to Newtonian and isothermal, the system of equations involves the con-
the von Karman structures (Williamson, 1996a,b). For the second stage tinuity equation Eq. (1) and the momentum equations Eq. (2):
(150 < Re < 300), successive passages from a 2D to a 3D mode are
observed. In the last stage (Re > 300), the flow becomes more chaotic. ∂ρf ui
= 0,
Furthermore for Re > 3900, some experimental studies (Norberg, ∂x i (1)
1987; Roshko, 1954; Parnaudeau et al., 2008; Lourenco and Shih, 1994;
Ong and Wallace, 1996) classified the wake flow into two zones: the ∂ρf ui ∂ρf ui uj ∂p ∂ ⎡ ⎛ ∂ui ∂uj ⎞ ⎤ ∂
+ =− + μ⎜ + ⎟ + (−ρf ui′ u′j ),
near wake (x/d < 10) and the far wake (x/d > 10), x being the ∂t ∂x j ∂x i ∂x j ⎢ ⎝ ∂x j ∂x i ⎠ ⎥ ∂x j (2)
⎣ ⎦
streamwise (longitudinal) distance. Although a large number of papers
were devoted to the flow developing downstream of a cylinder, few of where, ui (i = 1, 2) are the components of the mean velocity, x i are
them aimed at describing the particle dispersion. Recently, some nu- coordinates, ρf is the fluid density, μ is the dynamic viscosity, p is the
merical studies have been conducted to investigate particle-laden gas mean pressure. The term ui′ u′j is unknown and represents the Reynolds
flows past single and multiple circular cylinders (Liu et al., 2004; Huang stresses. This term has to be closed in order to solve the momentum
et al., 2006; Chen et al., 2009; Zhou et al., 2010). The majority of these equations and it is usually modeled using an eddy-viscosity approach
studies that were focused on particle dispersion were conducted using a (Greifzu et al., 2015; Launder and Spalding, 1974). Equations (1) and
Lagrangian approach for the particles while the air flow was predicted (2) are solved using the least squares cell based method, with the
from different methods (DNS, Discrete Vortex Method, Lattice Boltz- second order upwind method for the momentum equations. The pres-
mann method). Huang et al. (2014) conducted a 2D numerical study on sure-velocity coupling problem is solved with the Simple algorithm.
particle dispersion released from the surface of a circular cylinder Concerning the time discretization, a second order implicit approach is
placed in a gas flow at Reynolds number Re = 200000. They in- used.
vestigated the lateral dispersion functions for particles with various In the present study, we explore two turbulence models. The first
Stokes numbers (St) using Discrete Vortex Method to predict the air one is the standard k-ε where k and ε denote the turbulent kinetic en-
flow. The Stokes number is defined as the ratio of the particle response ergy and the dissipation rate, respectively. The second one is the
time ( ρp dp2/18μ ) to the main time scale of the fluid (d/ U∞) , where dp and Reynolds Stress Model (RSM). Both predict the Reynolds stresses in a
ρp are respectively the diameter and density of the particle and μ the completely different way. Indeed, the k-ε approach uses the Boussinesq
dynamic viscosity of the fluid. They took into account the drag, virtual hypothesis and a model of the turbulent viscosity to close the equations.
mass, pressure gradient, Saffman's lift, and Brownian forces. It was For the RSM, the goal is to solve transport equations for each Reynolds
shown that particles with St = 0.001, 0.0038, 0.01 can be found both in stress tensor components. The RSM model is clearly more accurate than
the vortex core and around whereas particles with St = 0.1, 1.0 cannot the k-ε model when the anisotropy of turbulence has a dominant effect
enter the vortex core and mainly accumulate around the vortex per- on the mean flow (highly swirling flows and stress-driven secondary
iphery. They also mentioned that the intensity of particle lateral dis- flows). This is particularly true for the near wall region. RSM model is
persion function decreases significantly as the Stokes numbers St in- potentially the most general and complete model amongst the classical
creases from 0.001 to 1.0. The lateral dispersion of particles with turbulence models. However, it remains very expensive and the nu-
St = 0.001 being nearly twice that of particles with St = 1.0. For merical solution can be difficult to reach. But it has been selected for its
Re = 140–260, Luo et al. (2009) conducted 3D numerical study using ability to reproduce turbulence anisotropy which may have a sig-
an Eulerian-Lagrangian approach with only the drag force taken into nificant effect on particles dispersion in the near-wall region, especially
account. They showed that the particles lateral dispersion function in- the vicinity of the cylinder is an inhomogeneous and anisotropic region.
creases with the Stokes number (St = 0.01, 0.1, 1, 10, 100) during a RANS models require then wall corrections to properly account for
first stage. Then, for the particles with smaller Stokes numbers of 0.01 cylinder near-wall behavior and its influence on vortices generation. In
and 0.1, the lateral dispersion decreases to reach an almost constant this sense, this is why such a turbulence model (RSM) is associated with
value. For the other particles, the lateral dispersion still increasing and the near-wall turbulence model (that is the Enhanced Wall Treatment,
larger Stokes number particles lead to higher lateral dispersion. EWT) (Lourenco and Shih, 1994). In this way, the simulation of the flow
To date, a question remains: which contribution has the most sig- as well as the nano- and micro particles dispersion in wall-bounded
nificant effect on the particle dynamics between the Brownian diffusion turbulent flows is expected to be predicted with accuracy (Mehel et al.,
and the interaction between nanoparticles and turbulence (turbulent 2010).
dispersion)? In this sense, particular attention must be paid to the link In order to get a full vision of the capabilities of these two turbu-
between the wake flow characteristics and the nanoparticles spatio- lence models, it is required to compare them with reference data
temporal distribution. How do von Karman's vortices trap nanoparticles available in the literature. The comparison is based on two parameters

295
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

Fig. 1. Geometry and the mesh of the simulation domain.

of the cylinder wake flow topology: the flow separation angle θ (the general topology (based on the mean velocity in the wake of the cy-
angle between the line connecting the origin with the point on cylin- linder) using the RSM agrees better with the experimental results
der's surface where the wall shear stress is zero and the longitudinal (Norberg, 1987) compared to that given by the k-ε model (Fig. 1). In
axis) and the dimensionless recirculation length (Lr/d), where Lr is the particular, the vortices that form the von Karman Street are more ac-
recirculation length (defined as the distance between the point down- curately depicted by the RSM model compared to the experimental
stream of the cylinder wall and the point where the centerline long- results of (Norberg et al., 1993). Indeed, the anisotropy induced by the
itudinal velocity is zero). For the continuous fluid carrier phase, the RSM model induces a wake flow that is less smooth compared to the k-ε
inlet air velocity (velocity-inlet) is U∞ = 5.56 m/s corresponding to a results. This is confirmed when we compare the recirculation length
Reynolds number Re= 9300 , where d = 25 mm. The relative outlet provided by both models with the experimental results (Table 1). The
pressure (pressure-outlet) is set at the standard atmospheric pressure RSM model has definitively a better ability to reproduce the cylinder
(P = 101325 Pa). A sketch of the computational domain is given in wake. In Table 1, we provide and compare results regarding the re-
Fig. 1. The cylinder is centered (Fig. 1a), the mesh is provided in Fig. 1b circulation length and flow separation angle given by both turbulence
(left), with a zoom in the vicinity of the cylinder (Fig. 1b, below). The models with those obtained by (Kravchenko and Moin, 2000) using
dimensional parameters of the study domain have been selected ac- Large Eddy Simulation. As it can be noticed, the relative error assessed
cording the study of (Nishino et al., 2008). The lengths between the from the literature results clearly shows better results given by the RSM
cylinder center and the domain entrance and exit are respectively model.
Lo = 10d and Li = 20d. The height of the domain is H = 10d. On the other hand, for Re > 9300, a good agreement with Norberg
No slip and stationary wall boundary conditions were applied at the et al. (1993) is found when comparing the flow topology (Fig. 2). The
wall cylinder. The top (wall-up) and down (wall-down) of the study recirculation length given by (Norberg et al., 1993) ranged from 1.8 to
domain are respectively assumed symmetry and wall conditions. All 0.8 for 1500 < Re < 15000 while we found 0.9 in the present study
particles are trapped on wall-down. At the outflow boundary, non-re- for Re = 9300. Thus, the RSM will be used in this study to properly
flecting conditions are employed. To capture the boundary layer around reproduce the cylinder wake flow.
the wall cylinder, 10 layers are created with a progression rate equal to
1.2 from the size of the first layer (y = 0.004d). Up to a radius of 2.8d,
2.2. Particle phase
the size of the elements is 0.08d. Far from the wall cylinder, the size of
the mesh elements is set to 0.2d.The total number of elements of the
The simulation of the carbon nanoparticles motion in an unsteady
mesh grid is 28402. Notice, that the mesh in the boundary condition
flow is achieved using a Lagrangian approach. We assume that the
was set to have a first layer with y+ = yu τ / ν < 5 with u τ being the
dispersed phase is sufficiently dilute to neglect any effect on the con-
friction velocity, in the present study the first layer is at y+≤ 3.5.
tinuous flow field and inter-particles collisions. The two-phase flow is
The topology of the flow is presented in Fig. 2 using the mean ve-
thus computed under the one-way coupling assumption. For the dif-
locity vector. Our numerical predictions are compared to the experi-
ferential equations related to particle location and velocity, it is as-
mental results of (Norberg et al., 1993) for the same Reynolds number.
sumed that the carbon nanoparticles are spherical while heat and mass
The color scale depends on the values of the mean fluid velocity vector
transfer are neglected. The motion of a particle is described by
magnitude (red is the higher and blue is the lower). We notice that the
Equations (3) and (4):

296
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

Fig. 2. Topology of mean velocity vector for Re = 9300.

Table 1 T is the absolute temperature of the fluid (K) (T = 300 K in the


Comparison of the recirculation length and flow separation angle with litera- present study) and kb is the Boltzmann constant. CC is the Stokes-
ture experimental results at the Reynolds Re = 3900. Cunningham slip correction factor which depends on the Knudsen
Studies Lr/d Relative error θ (°) Relative error (%) number (the ratio of the gas molecular mean free path λ to the particle
(%) diameter, Kn = λ / dp ) according to:

Kravchenko and Moin 1.35 88 2λ


(2000) Cc = 1 +
dp
(1.257 + 0.4e−0.55dp/ λ)
Present study (k-Ɛ) 1.11 18 81 8
(10)
Present study (RSM) 1.30 4 80 9
The components of the Saffman lift force are computed according
to:
dXp
= up
dt (3) 2ρf Kν1/2Sij
FL = (u͠ − up)
ρp dp (Slk Skl )1/4 (11)
dup
mp = ΣFi
dt (4)
where K = 2.594 and Sij are the components of the mean strain tensor.
where xp is the particle position vector, uP is the particle velocity The expressions of the lift and drag forces are function of the in-
vector, mP is the particle mass, and ΣFi represents the sum of all re- stantaneous fluid velocity at the particle location, u͠ . For turbulent
levant forces. In the present study, the particle to fluid density ratio is flows, a strategy to reconstruct this instantaneous velocity is decom-
high enough to assume that the main forces acting on each particle are: posed into average and fluctuating parts, u͠ = u + u′. The averaged part
ΣFi = FD + FB + FL (5) is given by the URANS computation, and the fluctuation is modeled
along each particle trajectory using a stochastic dispersion model. In
with FD the drag force, FB the Brownian force and FL the Saffman lift the present study, the turbulent velocity fluctuation is predicted by the
force. The expression of the drag force is given by: Eddy-Interaction Model (EIM). It generates the fluctuating velocity
3 ρf mp component of the carrier fluid with respect to the local average tur-
FD = CD (u͠ − up) u͠ − up
4 ρp dp (6) bulent intensity. This fluctuation is kept constant for a given time
(Morsi and Alexander, 1972). The components of the fluctuating velo-
where dp is the particle diameter, ρp is the particle density, CD is the city of the fluid seen by the particle are generated through:
drag coefficient, and u͠ = u (xp , t ) is the instantaneous fluid velocity
vector at the particle location. CD depends on the particle Reynolds u͠ i = ς ui′ 2 (12)
number Rep = ρf dp u͠ − up / μ . For the present computation, CD is given
by (ANSYS FLUENT, 2014):
with ui′ 2 is the standard deviation of the i-th velocity component, and
a a
CD = a1 + 2 + 23 ς is a normally distributed random number.
R ep R ep (7) Regarding the particles properties and injection, the injector simu-
where the constants a1, a2 and a3 are given by (Morsi and Alexander, lating the exhaust pipe is located below the cylinder at a dimensionless
1972). position (x/d = 0 and y/d = −0.62). The origin is taken at the cylinder
The Brownian force and the extended Saffman's lift force are mod- center, x being the streamwise direction positive downstream and y
eled according to the expressions proposed by (Li and Ahmadi, 1992). being the vertical direction positive upward. We inject 1000 parcels of
Each component of the Brownian force FB is expressed as a Gaussian carbon nanoparticles with a diameter dP = 10nm (St = 3⋅10−6) and a
white noise process: density ρp = 2000 kg/m3 at each time step Δt = 10−4s. The particles
injection velocity is up, i = 6.875 m/s with a mass flow rate of carbon
πS0 nanoparticles dm = 0.0625 mg/s. This flow rate was calculated ac-
FB = G dt
Δt (8) cording to the standard EURO6 concerning the limitation levels of
where G is the vector of independent Gaussian random numbers with particles emitted by vehicles (4,5 mg/km) and for an urban velocity of
zero-mean and unit variance, Δt is the time step used in the calculation. (50 km/h). Finally, each parcel represents 5.97⋅106 carbon nano-
S0 is called the spectral intensity and is given by: particles. .

216ρf νkb T
S0 =
π ²ρp2 dp5 Cc (9)

297
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

3. Results and discussion

3.1. Influence of Brownian force and the turbulence on nanoparticles


dispersion

Before investigating the influence of the turbulence and the


Brownian motion on particle dispersion, we compare our simulations
with the numerical results of (Huang et al., 2014) for the same particles
size (St) and flow conditions. This aims at validating our numerical
model. In that simulation the Reynolds number was Re = 200000, the
diameter of their cylinder was 11 mm. The air flow was solved using the
Discret Vortex Method (described in Huang et al., 2014) while the
Stokes number of the injected particles was St = 0.001. They used a
Lagrangian approach based on one-way coupling between the con-
tinuous gas phase and dispersed particles to simulate the particle mo-
tion in unsteady gas flow fields. The comparison is made on the lateral
dispersion function Dy(t) of nanoparticles in the vertical (y) direction.
This function is defined according to (Huang et al., 2014):

1
NP 2
⎧ 1 ⎫
Dy (t ) =
⎨ Np
∑ [Yi (t ) − Ym (t )]2

⎩ i=1 ⎭ (13)

where NP is the total number of particles in the simulation domain at


time t, Yi (t ) is the displacement of the ith particle in the y-direction from
time t − Δt to t, Ym (t ) is the mean value of particle displacement in the
y-direction from time t − Δt to t. Results are presented in Fig. 3, t* re-
presents the dimensionless time given by t ∗ = t∗/ d . Our results are in
good agreement with those of (Huang et al., 2014), except during the
transient regime that is for t* < 1332. The approach used in the pre-
sent study being validated, then we investigate the Brownian force and
turbulence impact on the dispersion of carbon nanoparticles
(St = 3⋅10−6) with a time step of Δt = 10−4s and at Re = 9300, in the
wake flow of a cylinder.
The assessment of the turbulence influence on nanoparticles dis-
persion is achieved by comparing the results provided when the EIM
model is activated with those obtained when it is not activated. In the
same way, the Brownian force impact on the dispersion of carbon na-
noparticles is assessed by comparing simulations in which, i.e. when the
Brownian force is considered or not. Hence in the next results, three
cases are presented: the reference case (the Brownian force and the
dispersion by turbulence EIM model, are not taken into account), the
Brownian impact case, where only Brownian force is considered, and
the turbulence impact case where only the EIM is activated. For all
these cases, the drag and lift forces are taken into account. The results Fig. 4. Nanoparticles distribution colored by their residence time in the wake
are presented and discussed below. flow of the cylinder.

3.1.1. Particles tracks


Fig. 4a-b-c show the nanoparticles dispersion (distribution of par-
ticle positions in the wake flow) at the dimensionless time t∗ = 55.
Particles are colored by their residence time. In this study, the residence
time is defined as the lifetime of carbon nanoparticles in the compu-
tational domain. It is highly variable and depends mostly, for a given
particle size, on the particle interaction with the turbulent vortex
structures. Overall, Fig. 4 show that the flow structure has a strong
influence on the dynamics of the carbon nanoparticles in the wake flow
of the cylinder. In Fig. 4a, the Brownian force and the EIM are not take
into account. As a consequence, the nanoparticles are dispersed solely
by the impact of the vortex structures that develop in the wake flow of
the cylinder. This corresponds to our reference case. In Fig. 4b, the
Brownian force is considered whereas EIM is not. We observe that
particles tend to accumulate mostly in the form of a filament at the
periphery of the vortex structures forming the well-known von Karman
vortex street with a small diffusion in the core of theses vortices. By
Fig. 3. Comparison of the lateral dispersion coefficient for particles stokes activating the EIM model (Fig. 4c), we notice that the carbon nano-
(St = 0.001). particles diffuse more significantly in the wake of the cylinder.

298
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

Fig. 5. Evolution of nanoparticles instantaneous non-dimensional concentration profiles in the near wake (left) and the far wake (right) of the cylinder at time
t* = 55.

Furthermore, particles with a high residence time are found in the core 3.1.2. Concentration profiles
of the vortices. This trends show that vortices trap carbon nanoparticles We are also interested in the carbon nanoparticles concentration
first at the periphery of vortices and then in their core. fields. Vertical instantaneous concentration profiles were studied at
At this stage, it is obvious that the vortical structures influence different dimensionless distances (1 < x/d < 18) downstream of the
dispersion since particles are trapped by these structures. The Brownian cylinder at a dimensionless time t∗ = 55. These positions are depicted
force and the turbulence in a more significant way make them diffusing by the vertical black dotted lines in Fig. 4. The concentration profiles
largely in the von Karman street structures. In particular, they tend to are presented as dimensionless concentrations (C*). C* is given by Eq
make them migrating from the periphery to the core of the vortices. The (14):
lateral dispersion function is enhanced when moving from near wake to
the far wake zone. This point will be discussed in the last part of the C∗ = C / Cmax (14)
present paper.
where C (kg/m3) is the instantaneous concentration of particles and

299
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

instantaneous profiles exhibit the strong correlation between vortices


positions and the peaks of instantaneous profiles. Depending on the
dimensionless position x/d downstream the cylinder, the position of the
peak is either on the right (y/d > 0) or left (y/d < 0) of the centerline
but always on the same side of the vortices.
Otherwise, the lateral dispersion in the y-direction is characterized
by the widening of the concentration profile at its base and the decrease
in the peak level as x/d increases. Since the profiles are instantaneous
(Fig. 5), the location of the peaks on both sides (y/d > 0 and y/d < 0)
corresponds to the location of the vortices in the upper or lower part.
The position of these vortices is strictly the same when comparing the
three cases (a, b, c) since the Strouhal number (St = f ∗d/ U∞) is iden-
tical for these cases (St = 0.2). This corresponds to a time duration of
Fig. 6. Contours of nanoparticles positions colored by their residence time at tr = 0.0225s meaning that the vortical structures are released at a fre-
t∗ = 85. quency f = 1 tr = 45Hz .
When comparing the different effects, it appears that the Brownian
force enhances the lateral dispersion with larger peaks of concentration.
This is particularly true in the far wake in comparison to the reference
case (lines 1d, 2d) where −2.5 < y/d < 2.5 in comparison to the
reference case where it was −1.5 < y/d < 1.5 in the far wake region.
The comparison between the reference case and the EIM model
(interactions particles/turbulence) shows that the turbulence influences
the dispersion of the injected nanoparticles. When EIM is activated, the
nanoparticles dispersion becomes more homogeneous and more ex-
tended (Fig. 5c) compared to the case for which EIM is not activated
(Fig. 5a). Indeed, the concentrations profiles are wider (looking at their
bases) than that of the reference case, especially for x/d > 5. At x/
d = 18, we find that particles can be observed from y/d = −3 to y/
d = 3 (Fig. 5c) whereas it was from y/d = −1.5 to y/d = 1.5 for the
reference case (Fig. 5a). In the vicinity of the cylinder, at x/d = 1,
Fig. 5a and c shows that the nanoparticles are simply driven by the flow
with a reduced dispersion field. This can be explained by the closeness
of the injector where the particles do not have enough time to disperse.
Fig. 7. Evolution of nanoparticles cumulated non-dimensional concentration in From the instantaneous concentration profiles, we can notice that
the near wake (line-1d, line-2d, line-5d) and in the far wake (line-10d, line-15d,
the peak positions are time dependent. The nanoparticles are trapped in
line-18d) of a cylinder.
the vicinity of the vortical structures (at the periphery or in the core
region) and these vortices are periodically found on both sides of the
central line (y/d = 0). This is still true in the far wake for the reference
case but tends to gradually disappear for the Brownian case and in most
of situations when the EIM is activated. This is due to the enhancement
of the diffusion process with this latter.
To assess the evolution of the lateral dispersion of the nanoparticles
from the near wake to the far wake, regardless this time-dependency, it
is then important to analyze the cumulated concentration profiles in the
next paragraph.

3.2. Dispersion of nanoparticles under the combined effects of Brownian


force and turbulence

In that part, our attention is focused on the simulation of the dis-


persion of carbon nanoparticles taking into account both effects. In
Fig. 6 we present the cumulated concentration given by cumulating the
whole amount of particles that pass through these vertical lines for the
total duration of the simulation. From the above results, it appears that
Fig. 8. Evolution of the lateral dispersion function in the near wake (line-1d, the Brownian force and the turbulence have a strong influence on the
line-2d, line-5d) and in the far wake (line-10d, line-15d, line-18d) of a cylinder. dispersion of carbon nanoparticles in the wake of a circular cylinder.
We first determine the time for which the cumulated concentration
profiles converge and reach a steady state. This is reached for t* = 85.
Cmax is the maximum corresponding value of C in the domain obtained
Fig. 6 shows the contours of the particles distribution colored by their
for each case (a, b, c), these latter are the same cases than described
residence time at this time (t* = 85). We see that the nanoparticles are
above for Fig. 4. Results of C* are presented in Fig. 5a-b-c.
firstly found at the vortices periphery before diffusing in their core. The
Our results show the decrease of C* with the dimensionless distance
dispersion is thus achieved everywhere in the von Karman alley. In-
x/d from the near wake (x/d = 1, 2, 5) to far wake (x/d = 10, 15, 18),
deed, the particles that have just been trapped appear in blue color at
whatever the configurations (a, b, c). This is due to particles dispersion
the periphery, while those that have a large residence time are found
with vortical structures enlargement. Note that particles number is in-
not only in the center of the vortices but also in some particular region
creasing since we continuously inject particles at each time step. The
between the core and the periphery forming a kind of arch. On the other

300
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

hand, the evolution of the cumulated concentration profiles (Fig. 7) trap them from their periphery before they diffuse into their core
show a decrease of the peak as the distance to the cylinder increases causing an increase of their residence time. This study shows the im-
and hence from the near to the far wake. At the same time, we clearly portance of the interactions between vortical structures and the nano-
observe a widening of the dispersion area which is depicted by the particles as well as the predominance of turbulence and Brownian
enlargement of the cumulated concentration profiles at their base forces for the dispersion of such small particles.
(Fig. 7). The values of the cumulated concentration peaks for lines x/ Future works will focus on simulations of nanoparticles dispersion
d = 1 and x/d = 18 are 6.39⋅10−7 and 1.84⋅10−7 kg/m3, respectively. in the wake of a car model (Ahmed body). Particles will be injected
The dispersion of particles is homogeneous around the center line (y/ from the tailpipe and comparisons in terms of velocities and con-
d = 0). The particle lateral dispersion is wider in the far wake compared centrations will be made with experimental data provided by wind
to the near wake of the cylinder. Indeed, this region expands from y/ tunnel investigations. The final goal will be to identify accumulation
d = ± 0.9 at x/d = 1 to y/d = ± 3.5 at x/d = 18. regions for such small and low-inertia particles emitted by ground ve-
Fig. 8 presents the lateral dispersion function. The Strouhal number hicles.
being equal to 0.2, we chose the time step of 0.0225s to compute the
particles lateral dispersion function. The global analysis shows that the Acknowledgement
dimensionless particle lateral dispersion function (Dy (t )/ d ) increases
during the particle passage from the near wake area (from lines 1d to University of Technical sciences and Technology of Bamako (Mali),
5d) to the far wake area (from lines 10d to 18d). Two stages can be ESTACA (France), University of Lorraine (France) are acknowledged for
observed for the evolution of this function: their financial and technical support.

• The first one is from t ∗


= 5 to t∗ = 40: Dy (t )/ d increases to reach a References
maximum and then decreases. This maximum increases with the
distance to the cylinder. If we compare the time gap that separates ANSYS FLUENT, 14.5, 2014. User's and Theory Guide. ANSYS, Inc, Canonsburg,
these peaks we find dt* = 13.34 between x/d = 5 and 18. This Pennsylvania, USA.
Beaudan, P., Moin, P., 1996. Numerical Experiments on the Flow Past a Circular Cylinder
corresponds roughly to the ratio of the distance between these two at Sub-critical Reynolds Number. Report No. TF-62. Department of Mechanical
lines (13d) to the mean velocity. This may be explained by the fact Engineering, Stanford University 1994. 5C.
that during this transient stage, most of the early injected nano- Carpentieri, M., Kumar, P., Robins, A., 2012. Wind tunnel measurements for dispersion
modelling of vehicle wakes. Atmos. Environ. 62, 9–25.
particles are trapped by the vortices first at their periphery. Then, Chen, B., Wang, C., Wang, Z., 2009. Investigation of gas-solid two phase flow across
they are carried at the flow mean velocity (U∞=5.56 m/s). The first circular cylinders with discrete vortex method. Appl. Therm. Eng. 29 (8–9),
particles being trapped mostly from the periphery rather than from 1457–1466.
Donaldson, K., Tran, L., Albert Jimenez, L.A., Duffin, R., Newby, D.E., Mills, N., MacNee,
the core of the vortices, the dispersion function is then larger. This is
W., Stone, V., 2005. Combustion-derived nanoparticles: a review of their toxicology
denoted by the blue points in Fig. 6. A small part starts to diffuse in following inhalation exposure. Part. Fibre Toxicol. 5/6, 553–560.
the vortices cores (green points); Greifzu, F., Kratzsch, C., Forgber, T., Lindner, F., Schwarze, R., 2015. Assessment of

• The second is for t* > 40. The particles are well mixed and Dy (t )/ d particle-tracking models for dispersed particle-laden flows implemented in
OpenFOAM and ANSYS FLUENT. Eng. Appl. Comput. Fluid Mech. 10 (1), 30–43.
becomes nearly constant (Fig. 8). https://doi.org/10.1080/19942060.2015.1104266.
Hassan, A.A., Crowther, J.M., 1998. Modelling of fluid flow and pollutant dispersion in a
It has been shown that the nanoparticle dispersion in the flow of a street canyon. Environ. Monit. Assess. 52, 281–297.
Huang, Y., Wu, W., Zhang, H., 2006. Numerical study of particle dispersion in the wake of
cylinder depends strongly on the vortices structures generated in the gas-particle flows past a circular cylinder using discrete vortex method. Powder
wake flow. Hence, to assess the UFP dispersion in the wake flow of a Technol. 162 (1), 73–81.
vehicle one should characterize accurately the vortical structures in Huang, Y., He, W., Kim, C., 2014. A numerical study on dispersion of particles from the
surface of a circular cylinder placed in a gas flow using discrete vortex method. J.
that particular wake flow. Hydrodyn. 26 (3), 384–393.
IPCC, 2007. In: Soloman, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B.,
4. Conclusion and future works Tignor, M., Miller, H.L. (Eds.), Climate Change 2007: the Physical Science Basis.
Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change. Cambridge University Press,
The dispersion of carbon ultrafine particles has been numerically Cambridge, United Kingdom.
studied using the Reynolds Stress Model (RSM) to describe the turbu- Jacobson, M.Z., 2005. Fundamentals of Atmospheric Modeling, second ed. Cambridge
University Press.
lent carrier phase. For the dispersed phase, a particle Lagrangian Johansson, C., Norman, M., Gidhagen, L., 2007. Spatial & temporal variations of PM10
tracking has been used. These particles (St = 3⋅10−6) are injected from and particle number concentrations in urban air. Environ. Monit. Assess. 127,
the bottom of a cylinder in an air flow at a Reynolds number of 9300. 477–487.
Kanda, I., Uehara, K., Yamao, Y., Yoshikawa, Y., Morikawa, T., 2006. A wind-tunnel study
The influences of the Brownian force and turbulence on the particle
on exhaust-gas dispersion from road vehicles—Part I: velocity and concentration
dispersion have been investigated. The results show that the dynamics fields behind single vehicles. 94 (9), 639–658.
of the nanoparticles is strongly influenced by the Brownian force and Keita, N.S., Mehel, A., Murzyn, F., Diourté, B., Tanière, A., 4th – 9th September 2016.
turbulence. Indeed, the vortical structures affect the particle dispersion Numerical study of the dispersion of carbon nanoparticles in the near wake of a
cylinder. In: The Proceedings of the 22nd European Aerosol Conference, (Tours,
by trapping them at their periphery. The Brownian and the turbulence France).
make them diffusing largely in the von Karman street structures. In Kittelson, D.B., Watts, W.F., Johnson, J.P., 2004. Nanoparticles emissions on Minnesota
particular, they tend to make them migrating from the periphery to the highways. Atmos. Environ. 38, 9–19.
Kravchenko, A.G., Moin, P., 2000. Numerical studies of flow over a circular cylinder at
vortices cores. The turbulence diffusion is more important than the ReD=3900. Phys. Fluids 12 (2), 403–417.
Brownian diffusion for such particles size even if we are dealing with Kumar, P., Fennell, P., Langley, D., Britter, R., 2008c. Pseudo-simultaneous measurements
nanoparticles. On the one side, it appears that the maximum di- for the vertical variation of coarse, fine and ultra fine particles in an urban street
canyon. Atmos. Environ. 42, 4304–4319.
mensionless concentrations decrease during the transit from the near to Kumar, P., Paul Fennell, P., Britter, R., 2008. Effect of wind direction and speed on the
the far wake regions due to the lateral dispersion. On the other side, the dispersion of nucleation and accumulation mode particles in an urban street canyon.
lateral dispersion function increases as we move in the streamwise di- Sci. Total Environ. 402 (1), 82–94.
Kumar, P., Fennell, P., Britter, R., 2008a. Effect of wind direction and speed on the dis-
rection. This is mainly due to the vortices that are moving away from persion of nucleation and accumulation mode particles in an urban street canyon. Sci.
each other. Furthermore, the distribution of the nanoparticles is almost Total Environ. 402, 82–94.
homogeneous around the central y-axis in the wake of the cylinder. This Kumar, P., Fennell, P., Britter, R., 2008b. Measurements of particles in the 5-1000 nm
range close to road level in an urban street canyon. Sci. Total Environ. 390, 437–447.
has been put in evidence from the cumulated concentration profiles.
Kumar, P., Fennell, P., Hayhurst, A., Britter, R.E., 2009. Street versus rooftop level
The vortices play an important role in the particle dynamics since they

301
N.S. Keita et al. Atmospheric Pollution Research 10 (2019) 294–302

concentrations of fine particles in a Cambridge street canyon. Boundary-Layer 437–445.


Meteorol. 131, 3–18. Ong, J., Wallace, L., 1996. The velocity field of the turbulent very near wake of a circular
Launder, B., Spalding, D., 1974. The numerical computation of turbulent flows. Comput. cylinder. Exp. Fluid. 20, 441.
Meth. Appl. Mech. Eng. 3 (2), 269–289. Parnaudeau, P., Carlier, J., Heitz, D., Lamballais, E., 2008. Experimental and numerical
Li, A., Ahmadi, G., 1992. Dispersion and deposition of spherical particles from point studies of the flow over a circular cylinder at Reynolds number 3900. Phys. Fluid. 20,
sources in a turbulent channel flow. Aerosol. Sci. Technol. 16, 209–226. 085–101.
Liu, L., Ji, F., Fan, J., 2004. Direct numerical simulation of particle dispersion in the flow Pey, J., Querol, X., Alastuey, A., Rodríguez, S., Putaud, J.P., Van Dingenen, R., 2009.
around a circular cylinder. J. Therm. Sci. 13 (4), 344–349. Source apportionment of urban fine and ultra-fine particle number concentration in a
Lourenco, L.M., Shih, C., 1994. Characteristics of the Plane Turbulent Near Wake of a Western Mediterranean city. Atmos. Environ. 43, 4407–4415.
Circular Cylinder, a Particle Image Velocimetry Study, vol. 3 published in Reference. Pope, C.A., Burnett, R.T., Thun, M.J., Calle, E.E., Krewski, D., Ito, K., Thurston, G.D.,
Luo, K., Fan, J., Li, W., Cen, K., 2009. Transient, three-dimensional simulation of particle 2002. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine par-
dispersion in flows around a circular cylinder (Re=140-260). Fuel 88, 1294–1301. ticulate air pollution. J. Am. Med. Assoc. 287, 1132–1141.
Mehel, A., Murzyn, F., July 2015. Effect of air velocity on nanoparticles dispersion in the Pope, C.A., Burnett, R.T., Turner, M.C., Cohen, A., Krewski, D., Jerrett, M., Gapstur, S.M.,
wake of a vehicle model: wind tunnel experiments. Atmos. Pollut. Res. 6, 612–617. Thun, M.J., 2011. Lung cancer and cardiovascular disease mortality associated with
Mehel, A., Tanière, A., Oesterlé, B., Fontaine, J.R., 2010. The influence of an anisotropic ambient air pollution and cigarette smoke: shape of the exposure-response relation-
Langevin dispersion model on the prediction of micro- and nanoparticles deposition ships. Environ. Health Perspect. 119, 1616–1621.
in wall-bounded turbulent flows. J. Aerosol Sci. https://doi.org/10.1016/j.jaerosci. Roshko, A., 1954. On the Development of Turbulent Wakes from Vortex Streets. NACA
2010.04.011. Transactions Report 1191.
Morsi, S.A., Alexander, A.J., 1972. An investigation of particle trajectories in two-phase Scungio, M., Arpino, F., Stabile, L., Buonanno, G., 2013. Numerical simulation of ultrafine
flow systems. J. Fluid Mech. 55 (2), 193–208. particle dispersion in urban street canyons with the spalart-allmaras turbulence
Nishino, T., Roberts, G.T., Zhang, X., 2008. Unsteady RANS and detached-eddy simula- model. Aerosol Air Qual. Res. 13, 1423–1437.
tions of flow around a circular cylinder in ground effect. J. Fluid Struct. 24, 18–33. Williamson, C.H.K., 1996a. Vortex dynamics in the cylinder wake. J. Fluid Mech. 328,
Norberg, C., 1987. Effects of Reynolds Number and Low-intensity Free-stream Turbulence 345–407.
on the Flow Around a Circular Cylinder. Department of Applied Thermodynamics and Williamson, H.K., 1996b. Vortex dynamics in the cylinder wake. Annu. Rev. Fluid Mech.
Fluid Mechanics, Chalmers University of Technology, Sweden Publication No. 87/2. 28, 477.
Norberg, C., 1993. Pressures forces on a circular cylinder in cross flow. In: Eckelmann, H., Zhou, H., Mo, G., Cen, K., 2010. Numerical investigation of dispersed gas’solid two-phase
Graham, J.M.R., Huerre, P., Monkewitz, P.A. (Eds.), IUTAM Symposium Bluff-Body flow around a circular cylinder using lattice Boltzmann method. Comput. Fluids 52,
Wakes, Dynamics and Instabilities 7-11 September 1992. Spinger-Verlag, gottingen, 130–138.
Germany. Zhu, Y., Hinds, W.C., Kim, S., Sioutas, C., 2002. Concentration and size distribution of
Oberdörster, G., Sharp, Z., Atudorei, V., Grelein, R., Kreyling, W., Cox, C., Jun, 2004. ultrafine particles near a major highway. J. Air Waste Manag. Assoc. 52, 1032–1042.
Translocation of inhaled ultrafine particles to the brain. Inhal. Toxicol. 16 (6–7),

302

You might also like