Sei sulla pagina 1di 11

Traffic Induced Pollution Dispersions in Urban Street Canyon

Aldo Malvin, 005381

MM4 CFD, Computational Fluid Dynamics


Faculty of Engineering, School of Chemical Engineering
Supervisor: Prof. Andy Chan

Submitted 27 December 2008

Abstract
Three-dimensional numerical model based on Reynolds-averaged Navier-Stokes equations
coupled with Renormalization Group (RNG), Reynolds Stress Models (RSM) and Large Eddy
Simulation (LES) was developed to investigate the air flow development and pollutant dispersion
within an urban street canyon using the FLUENT code. Model generated using three different
turbulence models showed a good agreement with the literature data. The use of LES model
suggested that the vortex formed within the street canyon is not stationary but precesses around
the canyon. Three dimensional movement of individual pollutant was observed when using LES but
not in the RNG k-ε and RSM turbulence models. Instead of one side pollutant distribution as shown
by RNG k-ε model, RSM model predicted evenly distributed pollutants within the street canyon.
Low numbers of pollutants escaping the canyon were observed. Possible alteration of building
configuration and roof design might improve the rate of pollutants escaping from the canyon. The
use of LES model resulted in simulations which correspond better to the real situation.
1. Introduction
Poorly ventilated street canyon often results in high concentrations of car-exhaust gases.
This issue is typically associated with low wind velocities or situations when the wind blows
perpendicular to the canyon axis. The past few years saw a growing interest in utilizing
computational modeling to investigate the development of flow behavior and pollutant dispersion
in an urban street canyon [1,2,3,4,5]. The use of turbulence model such as standard and
Renormalization Group (RNG) k-ε turbulence models had been proven reliable for simulating two
dimensional pollutant dispersion and air flow development within an urban street canyon [1].
However, actual three dimensional flows within street canyons are always characterized by flow
separation and recirculation, conditions that are known to pose difficulties for traditional
turbulence models such as k-ε model mentioned earlier [2].
With the advancement of computer technology and performance, it is decided to use large-
eddy simulation (LES), model which takes into account the unsteadiness of the flow and provides
detail information on the flow structure including turbulence statistics [3]. However, it has to be
noted that LES requires the most computational effort due to the fact that all simulations being the
transient type calculations performed on fine, three dimensional grids.
Therefore, the main aim of this study is to develop a three-dimensional numerical model
based on the Reynolds Average Navier-Stokes equations coupled with RNG k-ε turbulence model,
Reynolds Stress Model (RSM) and LES model for simulating the air flow development as well as
pollutant dispersion mechanisms. Comparison on the performance of each model will be carried
out based on computational effort. In addition, validation by means of comparison with the
experimental results obtained from Walton and Cheng [3] and Garmory et al. [4] is carried out.

2. Street Canyon Simulation


The studied three dimensional computational grids consist of 78,000 sub-map hexagonal
nodes with a section of the grid shown in Figure 1(A). Meanwhile, the 3-D schematic diagram of the
studied urban street canyon and the type of boundary condition applied are shown in Figure 1(B).
The total height of the building was 24 m and it extended 26 m above the building height with a
plane of symmetry specified at the upper boundary. Periodic boundaries were specified in the
streamwise direction so as to simulate an infinite series of canyons under perpendicular flow
conditions which means, the outlet velocities are fed back to the inlet plane at every time step
allowing a realistic flow profile to develop [3]. All simulations were carried out with the aid of
computer with the following specifications:
 CPU: Intel® Core™ 2 Duo E8300 @ 2.83 GHz
 Memory (RAM): 4.00 GB
 Operating System: Windows Vista™ Service Pack 1
 System Type: 32-bit Operating System

2
Periodic Boundary

Symmetry Plane

Pressure
Outlet 26 m

Velocity
Inlet
Wall

Wall 20 m

Wall
20 m Wall 24 m

Wall
30 m

20 m

(A) (B)

Figure 1. (A) A section of street canyon computational grid. (B) Street canyon geometry and boundary condition.

2.1. Model Definition


Three types of 3-D Reynolds Average Navier-Stoke (RANS) methods, namely steady-state
RNG k-ε turbulence model, steady-state Reynolds Stress model (RSM) and unsteady-state Large
Eddy Simulation (LES), were used to calculate the velocity and turbulence fields in FLUENT. Air
was used as material with inlet velocity of 5 m/s. The default constants in FLUENT were used as the
parameters governing these three equations. Exception applies on the RNG k-ε turbulence model
whereby instead of “Standard” pressure discretisation, “PRESTO” was utilized. In LES, time step
size of 2s with 100 time steps were taken to obtain a fully converged simulation, thus ensuring a
reliable modeling of air flow development in the street canyon. As suggested by Walton and Cheng
[3], the LES was conducted in two stages; the first to allow the flow to reach the statistically steady
state, and the second to collect turbulence statistics on the flow. Meanwhile, using the RNG k-ε
results as an initial guess could reduce the flow development time required, especially for the
concentration of the pollutant.

2.2. Pollutant Injections


Discrete phase model (DPM) was introduced, after a fully converged simulation was
obtained, to allow for the tracking of pollutants within the studied street canyon. The emission
source was defined at the centre point of the canyon (x=30, y=0, z=15), using CO 2 as pollutants.
This was intended to simulate a row of vehicles in the centre of the canyon mimicking lines of
moving traffic. 20 particles of CO 2 were injected along the canyon at mass flow rate of 0.1 kg/s at
each time step. The resulting particle velocity magnitude and particle ID were collected for
validation purposes.

3
3. Governing Equations

Air is incompressible with constant density and kinematic viscosity.


Continuity equation:

(1)

Momentum equation:

(2)

k and ε transport equation in the RNG k-ε turbulence model:

(3)

(4)

Where,
α k = inverse effective Prandtl number for k η O = constant, 4.38
α ε = inverse effective Prandtl number for ε β = constant, 0.012

η = scalar measure for the deformation tensor µ eff = effective turbulent viscosity.

Reynolds Stress Models (RSM):

∂u i' u 'j ∂u i' u 'j ∂Tkij


+ ui + = Pij + Rij − ε ij (5)
∂t ∂xi ∂x k
Where,
∂Tkij
= Reynolds stress flux
∂x k

Pij = production tensor

Rij = pressure rate of strain tensor

ε ij = dissipation tensor

Large Eddy Simulation (LES) [4]:


Reynolds tensor is used to relates the non linear term in the momentum equation with the filtered
variables,

τ ijR = u i u j − u i u j (6)

(7)

4
which gives

(8)

u i : Filtered velocity components in tensorial form

p : Filtered pressure field

τ ijR : Residual stress tensor analogous to the Reynolds stress tensor

Smagorinsky subgrid model:

1 ∂u i ∂u j (9)
S ij = ( + )
2 ∂x j ∂xi

τ ijr = −2vt S ij (10)

vt = C S2 ∆2 ( S ij S ij )1 / 2 (11)

τ ijr : isotropic residual tensor

S ij : strain rate of the resolved field


vt : eddy viscosity

C S : the Smagorinsky constant found experimentally ≈0.1-0.2


The finalized filtered momentum equation is,

(12)

Where,
Re = global Reynolds number
Re ti = local Reynolds number

Pollutant Transport Equations

(13)

Where,
Pe = global Peclet number, UL/D
Pe ti = local Peclet number, UL/D t ; D t = Eddy mass diffusivity.

These governing equations (1) – (13) were performed using the FLUENT code and SIMPLE
algorithm was utilized to solve the equations.

5
4. Results and Discussions

4.1. Air Flow Development


Velocity magnitude vectors for a section of the canyon are shown in Figure 2. Generally,
figure 1A, 1B and 1C show a similar flow pattern with largest velocity magnitude of 5.1 m/s
dominating the upper domain (above the building). This is expected as the upper domain consists
of the least resistance in term of frictional force to air flow as compare to region within the street
canyon. On the other hand, significant reduction in velocity magnitude could be observed in the
street canyon region. This is due to the existence of building walls and also objects of the streets
which potentially act as strong resistance, such as large frictional forces to the flow of air.

(A) (B)

(C) (D)

Figure 2. Vectors of velocity magnitude within the street canyon under different studied turbulence models: (A) steady
state RNG k-ε model, (B) steady state RSM, (C) unsteady-state LES. (D) is the result obtained from Garmory et al. [4]
under standard k-ε turbulence model.
Despite the identical vertical velocity profiles at the inlet plane, RNG and RSM models show
a slightly weaker velocity magnitude within the canyon as compare to LES turbulence model.
Another interesting observation is the random movement of vortex vectors found within the
canyon under LES turbulence model. This observation indicates that the vortex is unlikely to be
stationary but precesses around the canyon. Similar observation was obtained by Walton and

6
Cheng [3], which directly suggests that the current studied model is reliable and the flow within the
studied geometry had fully developed. As shown in Figure 3, to some extent, residual curves could
also serve as indication on the degree of convergence of a model. Further comparison between
Figure 1A to 1C with Figure 1D results in a good agreement in term of flow development within the
canyon. These facts serve as important supporting factors for the validation of the studied model.

(I)

(II)

(III)

Figure 3. Residual curves obtained under three different turbulence models: (I) RNG k-ε model, (II) RSM model, (III) LES
model.
7
4.2. Pollutants Dispersion
The particle ID and particle velocity magnitude (m/s) are shown in Figure 4. Out of 20 CO 2
particles injected, only few managed to escape from the canyon whereas the rest of the particles
are circulated continuously within the canyon. Similar flow patterns of particle were observed
between RNG and RSM turbulence model. However, the latter shows an even distribution of
particles along the canyon which corresponds better to real situation. This observation suggests
that for a similar steady state condition, RSM results in a better prediction than k-ε turbulence
model.

(A1) (A2)

(B1) (B2)

(C1) (C2)

Figure 4. Residual curve under different studied turbulence model: (a) standard k-ε model, (b) RNG k-ε model, (c) RSM
model.

8
The results shown in Figure 4 were validated by mean of comparison with contours of
mean concentration in and above the canyon as shown in Figure 5. Difficulty in obtaining the
contour of mean concentration in this study had lead to the use of particle track as subject for
comparison. In term of velocity magnitude, Figure (A2), (B2) and (C2) show the lowest magnitude
at the bottom of leeward side. This implies that most of the pollutant would settle down and
concentrate at this point due to the absence of sufficient driving force (loss of momentum).
Meanwhile, Figure 5A and 5B both indicate the largest concentration of pollutants at the bottom of
the leeward side. This observation is encouraging as it provides an absolute proof for the validation
of the studied model.

(A) (B)

(C)

Figure 5. Mean concentration of passive scalar (gm-3) under (A) LES and (B) k-ε model obtained from Walton and Chang
[3]. (C) Contour of mixture fraction in and above canyon (root mean square) obtained from Garmory A. et al. [4].

Major difference was observed in the pollutant dispersion within the canyon under LES as
compare to other steady state RNG and RSM turbulence model. The use of LES results in a random
dispersion of pollutants which occupied the whole region of the canyon. Under LES model, the
movement of individual pollutant was not limited to uniform x-y direction as shown in Figure 4A1
and 4B1, instead, each pollutant shows a three dimensional movement which involve x-y-z
directions as shown in Figure 4C1. This observation is definitely encouraging as it shows a better
correspondence to the real situation.

9
The use of current studied building configuration sees only a few number of pollutant
particle escaping from the canyon. This is of course not desirable as such pollutants are harmful to
human health. Alteration in building configuration such as introduction of height difference
between two opposite buildings and the use of slanted or sloped roof instead of planar roof might
be a good start for further works in the effort to improve pollutants escaping from the canyon.

4.3. Turbulence Model Performance


Based on the observation of air flow development and pollutant dispersion, it can be
deduced that Large Eddy Simulation (LES) turbulence model results in a more promising result
which correspond better to the real situation. However, it involves a significantly huge
computational effort as shown in Table 1. RSM required almost 5 times the computational time
required by RNG k-ε model but LES required even more, approximately 2 times, the computational
effort required by RSM. This is perhaps the only drawback which prevents the wide use of LES
model for case studies and simulations especially those which involved complicated geometry and
required large number of fine meshes.

Table 1. Performance of 3 different types of turbulence model based on the number of iterations and computational time.

Performance Criteria Time step No. of


Viscous Model
Iterations Time size (s) time step
RNG k-ε 440 6 min 50 sec - -
Reynolds Stress Model 1209 28 min 42 sec - -
Large Eddy Simulation 3623 56 min 2 100

5. Conclusions
Three turbulence models namely, steady-state RNG k-ε model and RSM model also
unsteady-state LES model, were utilized to investigate the air flow development and pollutant
dispersion within a street canyon. Validation of studied model was carried out by means of
comparison with literature data obtained from Walton and Cheng [3] and Garmory A. et al. [4]. All
of the simulation results show a good agreement with literature data. The use of LES model
suggested that the vortex formed within the street canyon is not stationary but precesses around
the canyon which corresponds exactly to the one obtained by Walton and Cheng [3]. In term of
pollutant dispersion, LES model showed that each pollutant exhibits a three dimensional
movement within the canyon. This observation, however, was not observed in both RNG k-ε and
RSM models. Even distribution of pollutants was also observed in the use of LES. Further
investigations on building configuration as well as the design of building roof was recommended
for further work to improve the rate of pollutants escaping from the canyon. Despite its powerful
prediction, LES required the most computational effort especially simulations involving
complicated geometry and fine meshes.

10
Reference
[1]. Chan T. L., Dong G., Leung C. W., Cheung C. S., Hung W. T., Validation of a two-dimensional
pollutant dispersion model in an isolated street canyon, Atmospheric Environment, 36 (2002)
861-872.
[2]. Kastner-Klein P., Fedorovich E., Rotach M. W., A wind tunnel study of organized and
turbulent air motions in urban street canyons, Journal of Wind Engineering and Industrial
Aerodynamics, 89 (2001) 849-861.
[3]. Walton A., Cheng A. Y. S., Large-eddy simulation of pollution dispersion in an urban street
canyon-Part II: idealized canyon simulation, Atmospheric Environment, 36 (2002) 3615-3627.
[4]. Garmory A., Kim, I. S., Britter R. E., Mastorakos E., Simulations of the dispersion of reactive
pollutants in a street canyon, considering different chemical mechanisms and micromixing,
Atmospheric Environment (2008), doi:10.1016/j.atmosenv.2008.07.033.
[5]. So E. S. P., Chan A. T. Y., Wong A. Y. T., Large-eddy simulations of wind flow and pollutant
dispersion in a street canyon, Atmospheric Environment, 39 (2005) 3573-3582.

11

Potrebbero piacerti anche