Vous êtes sur la page 1sur 26

Numerical Heat Transfer, Part A: Applications

ISSN: 1040-7782 (Print) 1521-0634 (Online) Journal homepage: http://www.tandfonline.com/loi/unht20

Large-Eddy Simulations and Heat-Flux Modeling in


a Turbulent Impinging Jet

Naseem Uddin , Sven Olaf Neumann , Bernhard Weigand & Bassam A.


Younis

To cite this article: Naseem Uddin , Sven Olaf Neumann , Bernhard Weigand & Bassam A. Younis
(2009) Large-Eddy Simulations and Heat-Flux Modeling in a Turbulent Impinging Jet, Numerical
Heat Transfer, Part A: Applications, 55:10, 906-930, DOI: 10.1080/10407780902959324

To link to this article: https://doi.org/10.1080/10407780902959324

Published online: 12 May 2009.

Submit your article to this journal

Article views: 217

View related articles

Citing articles: 18 View citing articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=unht20

Download by: [Indian Institute of Technology - Delhi] Date: 05 December 2017, At: 23:26
Numerical Heat Transfer, Part A, 55: 906–930, 2009
Copyright # Taylor & Francis Group, LLC
ISSN: 1040-7782 print=1521-0634 online
DOI: 10.1080/10407780902959324

LARGE-EDDY SIMULATIONS AND HEAT-FLUX


MODELING IN A TURBULENT IMPINGING JET

Naseem Uddin1, Sven Olaf Neumann1,


Bernhard Weigand1, and Bassam A. Younis2
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

1
Institut für Thermodynamik der Luft- und Raumfahrt, Universität Stuttgart,
Stuttgart, Germany
2
Department of Civil & Environmental Engineering, University of California,
Davis, California, USA

This article documents the results of an investigation into aspects of the simulation and
modeling of turbulent jets that impinge orthogonally on a target surface. The focus is on
the case of a jet which issues from a circular pipe into stagnant surrounding at the relatively
high value of Reynolds number of 23,000 (based on nozzle diameter and bulk velocity) for
which experimental data are available. Large-eddy simulations were performed to obtain
details of the mean flows and the turbulence fields including distributions of all components
of the turbulent heat fluxes. The outcome of these simulations were used to assess three
alternative models for the turbulent heat fluxes which differ from the conventional Fourier’s
Law by not being based on the assumption of proportionality between the eddy and thermal
diffusivities via a constant Prandtl number. It was found that only one of the models
considered succeeds in representing the effects on the heat fluxes of the complex strain
field associated with the stagnation region and the subsequent development into the wall-
jet region. The reasons for this outcome are discussed.

1. INTRODUCTION
Turbulent flows of engineering interest often occur at high Reynolds numbers
and involve transport of heat or mass. Due to limitations of available computer
memory and time, the computation of these flows is still largely done by the solution
of the Reynolds-averaged Navier-Stokes (RANS) equations with appropriate
models for the resulting unknown turbulence correlations. Good progress has been
achieved in modeling the turbulent transport processes in thin shear layers where the
flow is unidirectional and the only significant transport by turbulence occurs in the
direction normal to the mean-flow direction. In these flows, turbulence is in approxi-
mate local equilibrium and where the mean deformation rates do vary, they do so at
a sufficiently modest rate for the turbulence to adjust to new equilibrium state.
A corollary of the local-equilibrium condition is that the principal directions of
the Reynolds-stress tensor tend to be aligned with those of mean deformation such

Received 23 August 2008; accepted 2 April 2009.


Address correspondence to Naseem Uddin, Institut für Thermodynamik der Luft- und Raumfahrt,
Universität Stuttgart, Stuttgart 70569, Germany. E-mail: naseem.uddin@itlr.uni-stuttgart.de

906
LES AND MODELING OF IMPINGING JET 907

NOMENCLATURE
C1, C2, . . . Model constants, a Thermal diffusivity, m=s
cp Spesific heat capacity at constant m Dynamic viscosity, Pa  s
pressure, J=(kgk) n Kinematic viscosity, m2=s
D Nozzle diameter, m q Density, kg=m3
H Distance to target wall, m E Dissipation of turbulent kinetic
k Turbulent Kinetic energy, energy, m2=s3
m2=s2 D Local grid spacing, m
Nu Nusselt number, C Eddy thermal diffusivity
Pr Prandtl number,
Subscripts and Superscripts
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

q Heat-flux, W=m2
Re Reynolds number, þ Wall coordinates
u0i u0j Turbulent stress, m2=s2 b Bulk quantities
ui Velocity vector, m=s s Subgrid scale
uT Friction velocity, m=s t Turbulent
U Mean velocity, m=s w Wall values

that an eddy-viscosity representation of turbulence transport would be adequate.


However, in most practically-relevant flows, changes in the mean deformation rates
are far too rapid for local-equilibrium to be established and hence the principal axes
of the Reynolds-stress tensor are not aligned with those of the mean deformation
tensor. In these cases, eddy-viscosity based models do not provide an acceptable level
of accuracy and the alternative is to use complete Reynolds-stress transport closures
where transport of the Reynolds stresses by the mean flow (advection) and by the
turbulence itself (diffusion) are directly accounted for in the model. Almost all
commercially-available computational fluid dynamics (CFD) software now provide
the option for using Reynolds-stress transport models and experiences elsewhere
show that the use of these models yield distinct improvements in the prediction of
complex engineering flows. While the stress-transport models yield improvements
in the prediction of the momentum field that justify the added effort involved in
using them, the same does not generally hold for the transport models for the heat
or mass fluxes. These models, which have largely been developed by analogy with the
Reynolds-stress closures, have not been shown to yield worthwhile improvements
and are thus rarely used. Instead, the prediction of heat and mass transfer is still
based on classical approaches, such as Fourier’s law, which relates the turbulent heat
fluxes to the gradients of mean temperature.

qT
u0i t0 ¼ Ct ð1Þ
qxi

where, Ct is the eddy thermal diffusivity, which is related to the eddy viscosity nt via
the turbulent Prandtl number.

Ct ¼ nt =Prt ð2Þ

The turbulent Prandtl number is usually assigned a constant value which


depends on whether the flow develops remotely from a solid wall or adjacent to one.
908 N. UDDIN ET AL.

However, experiments both physical and numerical via direct numerical simulations
(DNS) show this quantity to depend both on location within the shear layer and on
the direction of mixing. Analysis later in this article will show that this parameter is
in fact a function of the Reynolds stresses, the turbulent heat flux, and the gradients
of both mean temperature and velocity. In the DNS of Kasagi et al. [1] of passive
scalar field in turbulent channel flow, the turbulent Prandtl number was found to vary
strongly with distance from the wall. Moreover, Jischa and Rieke [2] have also found
that the turbulent Prandtl number increases rapidly as the molecular Prandtl number
reduces below unity.
Thus, it would appear that improvements in the prediction of the mean flow
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

field that can be obtained using Reynolds-stress transport closures are not fully
matched by improvements in the prediction of heat or mass transport. Some benefits
will arise from better prediction of the turbulence parameters that enter into the
definition of the eddy viscosity, but the scalar fluxes themselves will still be obtained
from a model which does not correctly represent the physical processes involved.
An alternative to both scalar-flux transport models and the conventional
gradient-transport closures, and one which can potentially be used in routine engin-
eering computations, is provided by a class of closures referred to as nonlinear
algebraic scalar-flux models. These models represent more of the physical processes
that determine the scalar fluxes than is reflected by Fourier’s law and hence stand a
better chance of capturing the effects on these fluxes due to a complex deformation
field. At the same time, these models are algebraic and are therefore more efficient in
terms of computing effort than the full scalar-flux transport closures. They are also
more straightforward to implement especially if they are explicit in the scalar fluxes
since no iterations would then be required. The vast majority of previous testing of
these models has been done with reference to data from relatively simple shear flows.
One exception being the study of Dietz et al. [3] who systematically assessed the per-
formance of some of these models for the case of a heated flow over a backward-
facing step. In this work, we extend the assessment to an even more complex flow;
namely that of a cold jet which impinges normally onto a flat plate. The complexity
of this flow stems from a number of factors which pose a challenging test to the
scalar-flux models. Amongst these factors is the fact that both free as well as wall-
affected flow regimes are present and the model for the scalar fluxes is expected to
correctly respond to the different levels of turbulence anisotropies found in each case.
Moreover, the contributions to the turbulence field made by the normal strain rates
are significant in the region around the stagnation point but then these become less
important as the shear strains become more dominant in the subsequent wall-jet flow.
Here, too, a properly formulated model is expected to obtain the correct response.
In order to provide data usable for models assessment, large-eddy simulations
(LES) of an impinging jet have been performed as part of this study. Previous LES stu-
dies of impinging jets have been reported by Gao et al. [4], Gao and Voke [5], Olsson
and Fuchs [6], Cizesla et al. [7, 8], Tsubokura et al. [9], and Hadiabdić and Hällqvist
[10, 11]. Recently, Abe [39] has used LES data for a plane impinging jet, with flow
Reynolds number of 6,000 and nozzle-to-plate spacing of 10, for the calibration of
the coefficients of a quadratic model. In this work, we carry out large eddy simulation
of an impinging jet at the relatively high Reynolds number of 23,000 (based on bulk
velocity and nozzle diameter) and at the relatively small target-to-wall distance of
LES AND MODELING OF IMPINGING JET 909

two nozzle diameters. These conditions emphasize many of the features that make
impinging cold jets a challenging test flow for model evaluation.

2. LARGE-EDDY SIMULATIONS
The procedure for deriving the appropriate equations for use for LES is well
known [12]. The outcome is a set of equations that describe the conservation of mass,
momentum, and thermal energy of the form.
qq qðqubi Þ
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

þ ¼0 ð3Þ
qt qxi
  
qðqubi Þ qðqubi ubj Þ qbp q ui qubj
qb qssgs
ij
þ ¼ þ m þ  ð4Þ
qt qxj qxi qxj qxj qxi qxj
" #
b Þ qðqcp ubj T
qðqcp T bÞ q b
qT qqsgs
j
þ ¼ Ccp  ð5Þ
qt qxj qxj qxj qxj

where sgs denotes subgrid scale and â denotes a quantity which is resolved directly in
the simulations. The quantity ssgs
ij is given by
0 1
B C
ssgs ui uj  ubi ubj Þ ¼ q@ud
ij ¼ qðd bi ubj  ubi ubj þ ud
bi uej þ ud
ei ubj þ ud
ei uej A ð6Þ
|fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflfflfflffl{zfflfflfflfflfflffl} |{z}
Lij Cij Rij

In Eq. (6), Lij are the Leonard terms, Cij are the cross terms, and Rij are the subgrid
or unresolved Reynolds stresses.
A large class of subgrid scale closure models for momentum are based on a
gradient-diffusion hypothesis. It can be expressed as a relation between the
anisotropic stress and (large-scale) strain rate tensor.
 
sgs 1 sgs c 1 c
sij  dij skk ¼ 2ms Sij  dij Skk ð7Þ
3 3
where ms is the subgrid viscosity, which needs to be specified by an appropriate model.
The simplest model for this is proposed by Smagorinsky [13], according to
which the subgrid viscosity can be prescribed as

b
ms ¼ qc1 D2 S ð8Þ

where
  qffiffiffiffiffiffiffiffiffiffiffiffiffi
Sij  ¼ S
b ¼ 2S cij Scij ð9Þ

hence,
qffiffiffiffiffiffiffiffiffiffiffiffiffi
b ¼ qðcs DÞ2
ms ¼ qðcs DÞ2 S 2S cij Scij ð10Þ
910 N. UDDIN ET AL.

The cs is referred to as the Smagorinsky constant. The main uncertainty associated


with LES arises from the choice of value for this constant.
Germano et al. [14] proposed a procedure which allows the determination of the
Smagorinsky constant dynamically. By this procedure the Smagorinsky constant is no
longer taken as a constant, but rather as a function of space and time: cs ¼ cs(x, t).
The dynamical computation of cs is accomplished by defining a test filter with width
larger than the grid filter width D. In the present article, we have used the Germano
dynamic model [14] in preference to assigning a constant value for cs. Details of this
model can be found in the original reference.
The data for determining cs obtained from the simulation are first averaged in
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

time and then averaged at the radial locations in azimuthal direction. The data is
clipped to positive values after the contraction procedure for stability reasons.
For the energy equation, also by using the gradient diffusion assumption, one
can write

b
qT
qsgs
ij ¼ Cs cp ð11Þ
qxi
Here, the Tb is the resolved temperature and Cs is the subgrid thermal diffusivity. By
substituting Eq. (11) into Eq. (5), the following form of energy equation is
" #
b Þ qðqcp ubj T
qðqcp T bÞ q b
qT
þ ¼ ðC þ Cs Þcp ð12Þ
qt qxj qxj qxj

The subgrid thermal diffusivity Cs is expressed by

Cs ¼ ms =Prs ð13Þ

where Prs is the subgrid Prandtl number. The value of subgrid Prandtl number is set
to 0.9.
The computations were performed with the computer code FASTEST-3D [15].
The code is based on a finite-volume discretization with pressure-velocity coupling
done through the SIMPLE algorithm. The second-order implicit Crank-Nicolson
method was used for time discretization. Discretization of the spatial gradients in
the convective terms was achieved with the second-order accurate central-differencing
scheme. The resulting set of algebraic equations were solved using a SIP solver [16].
The focus of this work is on the case of an orthogonally impinging round jet.
The jet, which is issued into stagnant surroundings, becomes fully turbulent when the
Reynolds number (based on the bulk velocity Ub and jet’s diameter D) is greater than
3,000. In the present case, the Reynolds number is 23,000. This value was chosen as it
was investigated experimentally by Baughn et al. [17] and Cooper et al. [18], and
because it has become a standard test case for assessment of turbulence models.
A schematic representation of the flow is shown in Figure 1. The flow field of an
impinging jet can be thought of as being composed of three different zones [19]: a
free jet zone which is comprised of a shear layer and the potential core; a stagnation
flow=impingement zone where the radial velocity increases with approach to the wall
and part of the jet is deflected; and a wall jet zone where the flow accelerates radially
LES AND MODELING OF IMPINGING JET 911
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 1. Schematic diagram of flow physics of impinging jet.

outwards in the form of a wall jet composed of an inner, boundary-layer like region
and an outer region which resembles a free mixing layer. The interactions between
these two regions exert a strong effect on the heat transfer rates. Each of these three
zones have associated with them features that test certain aspects of the scalar-flux
models. These will be discussed in the next section.
The computational domain, which is shown in Figure 2, consists of a circular
impingement zone and a circular pipe of length 6D. The domain is confined at the
top by an adiabatic, no-slip wall. The pipe walls were also taken to be adiabatic with
no slip. The radial extent of the computational domain was taken as 16D. A constant
heat flux of 1000 W= m2 is applied at the target wall. The jet temperature at inlet is
set to 293 K and the fluid Prandtl number at inlet is 0.7.
An O-grid was used to mesh the computational domain, with a hexahedral
structured cells. The grid was generated using ICEM-CFD of ANSYS, Inc. [20].
In direct numerical simulation (DNS) the complete flow field is resolved and the

Figure 2. Schematic diagram of computational domain used for the impinging jet LES simulation. The
origin is fixed at the geometrical stagnation point of the jet.
912 N. UDDIN ET AL.

grid-spacing must be of the order of the Kolmogorov length scales, defined as, g ¼
(n3=E)1=4. In LES, the requirements in grid resolution are not as stringent and a rather
coarser mesh can be used as only the large energy containing scales are resolved.
The most demanding region of the flow in terms of grid resolution require-
ments is in between the stagnation point and the wall jet region up to r=D  4. In this
þ þ
region, yþmin is close to unity, Dr is 36.3 and rDh 20. The mean grid spacing
þ þ þ
inside the jet is (D =g ) 6.6, where, D is the dimensionless grid spacing and g is
the length of the Kolmogorov scale. The length of the Kolmogorov scale is estimated
from the dissipation of turbulent kinetic energy, which is estimated through the
budgets of the turbulent kinetic energy transport equation. The grid resolution
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

was choosen by experience in such a way that based on a estimation for the turbulent
length scale, grid restrictions were met in order to resolve approximately 80% of the
kinetic energy in the near wall region as well as in the rest of the domain [12].
The evolution of the turbulent jet downstream of the exit nozzle is strongly influ-
enced by the nature of the perturbations imposed on the exit flow. Often, deterministic
inflow perturbations are used [9]. However, such periodic excitations are quite differ-
ent from the type of perturbations found in a real jet. An alternative approach to gen-
erating the inlet conditions is to use a precursor simulation of the pipe flow, but this
approach places heavy demands on computer resources. Recently, there has been a
growing interest in the generation of synthetic turbulence that can be used for the
simulations. We have adopted this approach and used the procedure proposed by
Klein et al. [21]. By this approach, it is possible to generate pseudo-turbulent velocity
data with a prescribed Reynolds-stress-tensor and a locally defined autocorrelation
function. Briefly, the generated inflow turbulent velocity is based on the relation
ui ¼ huii þ aij u0j ð14Þ

where, huii is the mean axial velocity, uj0 are velocity fluctuations, and aij is a tensor
related to the Reynolds stress tensor [21]. The velocity fluctuations generated are based
on digital filtering of random data. An autocorrelation function Ruu for homogeneous
turbulence in late stage is used to describe the turbulence.
!
p^r2
Ruu ð^rÞ ¼ exp  ð15Þ
4LðtÞ2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where ^r is the position vector and LðtÞ ¼ 2pnðt  to Þ, is the integral length scale at
the inflow plane. The fluctuation level is specified by the DNS data.
The mean velocity profile inside the pipe is specified according to the proposals
of Kays et al. [22].
" #
þ 1:5yþ ð1 þ r=rw Þ
u ¼ 2:5 ln þ 5:5 ð16Þ
1 þ 2ðr=rw Þ2

where yþ is the pipe wall normal direction and r is the radius of the pipe.
At the outlet, nonreflective boundary conditions were used for all dependent
variables. The use of this type of boundary condition significantly reduces the
computational effort associated with obtaining accurate temporal simulation in a
finite-sized domain.
LES AND MODELING OF IMPINGING JET 913

The total simulation time is equivalent to 20 cycles, where one cycle corresponds
to the preferred frequency of the jet. The flow becomes statistically stationary after
about 8 cycles. The simulations were constrained in such a way that the CFL number
remained less than unity. The dimensionless time step DtD=Ub was set equal to 7  108.
The computations were performed on the CRAY Opteron Cluster,
Höchstleistungs-rechenzentrum (HLRS), Stuttgart, Germany.

3. ALGEBRAIC HEAT-FLUX MODELS


The basis for these models is evident from inspection of the exact transport
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

equation for the turbulent scalar fluxes [23].



Du0i t0 0 0 qT 0 0
qUi p0 qt0 1 qu0i qt0
¼ ui uj  uj t  n 1 þ
Dt qxj qxj q qxi Pr qxj qxj
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflffl{zfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
PTi PTi ETi
! ð17Þ
q qu0 qt0 t0 p0
 nt0 i  cu0i þ u0i u0j t0 þ dij
qxj qxj qxj q
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
DTi

where DTi is the combined diffusion by molecular, turbulent, and pressure fluctua-
tions, PTi is the production term, PTi is pressure scalar gradient correlation, and
ETi is dissipation rate tensor.
It is immediately apparent from inspection of this equation that the turbulent
scalar fluxes should depend on gradients of the scalar in all three directions and, also,
on gradients of the mean velocity and on the Reynolds stresses themselves. Several
proposals for accounting for the various dependencies in Eq. (17) have been reported
in the literature. These include the proposals of Daly and Harlow [24], Rogers et al.
[25], So and Sommer [26], Suga and Abe [27], Abe and Suga [28], and Younis et al. [23].
The Daly and Harlow [24] model is the simplest of the proposals which
introduces a dependence on the Reynolds stresses and thereby allow for the thermal
diffusivity to be anisotropic. This model is given as

k qT
u0i t0 ¼ Ch u0i u0j ð18Þ
E qxj

While this model introduces the dependence on the Reynolds stresses that is required
by the exact equation, it is lacking a dependence on the mean deformation rate which
is required by the same equation. This is also the case with the model by Abe and
Suga [28] which reads
k qT
u0i t0 ¼ Ch u0i u0k Akj ð19Þ
E qxj
where
u0k u0j
Akj ¼ ð20Þ
k
914 N. UDDIN ET AL.

The dependence on the mean deformation rate is obtained in the model by Younis
et al. [23] which was derived by tensor representation theory. This model is given as

k2 qT k qT k3 qUi qT
u0i t0 ¼ C1 þ C2 u0i u0j þ C3 2
E qxi E qxj E qxj qxj
2

k qUj qUi qT
þ C4 2 u0i u0k þ u0j u0k ð21Þ
E qxk qxk qxj

The coefficients of the Younis et al. model were determined using homogeneous
turbulence data. The coefficients thus obtained are C1 ¼ 0.0455, C2 ¼ 0.373, C3 ¼
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

0.00373, and C4 ¼ 0.0235. To account for wall effects, one of the model coeffi-
cients was made dependent on the turbulent Reynolds number (Ret), the turbulent
Peclet number (Pet), the second and third invariants of the Reynolds stress anisotropy
tensor (Aij), and the stress flatness parameter (A). Ret is a scalar quantity which repre-
sents the ratio of the turbulent to molecular shear stresses—its value tends to zero at
the wall. Near the wall, where the turbulence anisotropy is most significant, its effect
on the scalar fluxes is accounted for via Aij. The coefficient C1 is modified near the wall
via a damping function [29].

C1 ¼ C1 ½1  expðAbPeat Þ ð22Þ

where

Pet ¼ Ret Pr
k2
Ret ¼
nE
u0i u0j 2
Aij ¼  dij
k 3
A2 ¼ Aij Aij
A3 ¼ Aij Ajk Aki
9ðA2  A3 Þ
A¼1
8
The coefficients a and b are assigned the values 0.1 and 1.5, respectively, arrived at
by matching DNS profiles of scalar fluxes in a fully-developed heated channel.

4. LES RESULTS
We first compare the present LES results with experimental data and previous
numerical simulations. A distinctive feature of the impinging jet flow is the presence
of a ring-shaped toroidal vortex which spreads across the target wall. This can be
visualized by plotting iso-pressure surfaces, as is done in Figure 3. The ring vortex
is immediately apparent as are its effects on the velocity, as seen by the plot of the
velocity vectors shown in the same figure.
The simulations were performed in Cartesian coordinates where x, y, and z are
the coordinate directions and u, v, and w are the corresponding velocity components
LES AND MODELING OF IMPINGING JET 915
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 3. Impinging jet with toroidal (ring) vortex at the target wall.

(see Figure 4). For convenience, however, the results are presented with reference to
a cylindrical coordinate system with the radial, azimuthal, and vertical components
of velocity denoted by ur, uh, and uy, respectively. The mean velocity is represented
as U. The data are those of Cooper et al. [18] obtained at three radial locations
r=D ¼ 1, 2, and 3. The results are shown in Figure 5. The agreement is generally good
though some differences are apparent in the maximum values which are greater in
the LES though the location, in terms of distance from the wall, where the maximum
velocity occurs is quite similar in the LES and measurements. The discrepancy
between the experimental data and the numerical simulations are due to uncertain-
ties in the inlet conditions. The turbulent jet is very sensitive to the prescription of
both the mean velocity distribution and the turbulent fluctuations. The jet does

Figure 4. Cartesian coordinate system used in simulation. Cylindrical coordinate system used for the
presentation of the results.
916 N. UDDIN ET AL.
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 5. Velocity distribution in wall jet at r=D ¼ 1,2, and 3. U is the mean velocity and Ub is the bulk
velocity at the jet’s exit.

not attain the self-similar state at small jet outlet-to-target wall distances. However,
to check the development of jet in the wall jet zone, the scaling log laws are used. The
wall jet developing after impinging jet does not exhibit the conventional law of the
wall behavior. The following semi-logarithmic relation can be used to model the
inner layer of the wall jet.

U 1
yus
uþ ¼ ¼ ln þB ð23Þ
us j n

where us is the friction velocity and j is the von Karman constant.


Wygnanski et al. [30] have found that a logarithmic profile might be fitted to the
data but the constant (B) present in the law of the wall must be adjusted. Özdemir
and Whitelaw [31] have found that in case of an impinging jet, if the outer edge of
the equilibrium layer is attached to the point of maximum velocity (Um), which
occurs close to wall, then this maximum velocity can be used as an appropriate
scaling parameter. They also found that B cannot be taken as a constant. Instead,
the following function was proposed.

Um
B¼f ð24Þ
us

Using the approach proposed by Özdemir et al. [31] and Guerra et al. [32] the
following relationship is obtained.

Um
B ¼ 1:122  10:53 ð25Þ
us

The first term in the above equation is termed the profile shift parameter. If this
shift parameter is subtracted from the law of the wall, the resulting curve shows
the presence of an equilibrium layer that extends up to the point of maximum
velocity. Figures 6a and b shows the radial velocity profiles with and without sub-
tracting the shift parameter on the semi-logarithmic axes.
LES AND MODELING OF IMPINGING JET 917
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 6. (a) Semilogarithmic plot of the radial velocity profiles without subtracting the shift parameter;
and (b) velocity profile with subtraction of profile shift parameter. The line shows the characteristic of the
equilibrium layer that it extends until the location of maximum velocity. r=D ¼ 1, þr=D ¼ 1.5,  r=D ¼
2,  r=D ¼ 2.5, 3 r=D ¼ 3, " r=D ¼ 3.5, 4 r=D ¼ 4.

The present results thus confirm the experimental findings of refs. [31] and [32]
that the inner layer of the wall jet, which is formed after the jet impingement on the
wall, can be scaled using a log-law formulation.
In Figures 7 and 8, the distribution of radial velocity fluctuations u0r u0r and
shear stress u0r u0y in the wall jet zone are compared with the experimental data of
the Cooper et al. According to Cooper et al. [18], the root mean square fluctuating
velocities have the experimental measurement uncertainties of 4% (ur0 ), 6% (uy0 )
and shear stresses (u0r u0y ) have uncertainties of 9%. Close to the wall the agreement
between the shear stresses and wall normal velocity fluctuations is very good. The
radial velocity fluctuations agree with the experimental data as the jet develops.
Contours of the turbulence kinetic energy (k) are shown in Figure 9. Note the
emergence of high levels of k in the free shear layer which develops from the pipe’s
exit. These high levels are sustained in the region of high shear rates in the deflected
flow and are then sustained in the wall jets region. Around the stagnation point, k is
at a low level since the flow there is essentially a remnant of the potential core. A well-
known defect in many turbulence closures is the prediction of maximum levels of k at
the stagnation point—caused by over-emphsise of the production of k by normal

Figure 7. Distribution of u0r u0r in wall jet at r=D ¼ 1, 1.5, 2, 2.5, and 3, Ub is the bulk velocity at the jet’s exit.
918 N. UDDIN ET AL.

Figure 8. Distribution of u0r u0y in wall jet at r=D ¼ 1,2.5, and 3. Um is the maximum mean velocity in the
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

wall jet zone.

strains. In contrast, the LES results show that the levels of k close to the stagnation
point are quite small being close to the levels obtained in the potential core.
Nishino et al. [33] have utilized the anisotropy invariant map (AIM) to
investigate the nature of turbulence in the stagnation zone of an impinging jet. They
found that near a wall, the turbulent flow follows the axisymmetric contraction state.
This is contrary to what is known about channel flows. Namely, that the flow region
near wall is characterized by a turbulence structure that is in a two-component state.
In Figure 10, the nature of turbulent flow of an impinging jet at r=D ¼ 2 is compared
to flow at r=D ¼ 7 using anisotropy invariant mapping. It is found through
anisotropy invariant mapping that the flow has not fully recovered from the non-
equilibrium effects till its exit from the computational domain. However, the flow
approaches the well known anisotropy behavior of near wall channel flows.
Finally, a comparison is made with the experimental data for the radial distri-
bution of Nusselt number. The results are plotted in Figure 11. The experimental

Figure 9. Turbulent kinetic energy distribution in wall jet developed after an impinging jet.
LES AND MODELING OF IMPINGING JET 919
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 10. The anisotropy invariant map of an impinging jet at r=D ¼ 2 and 7.

results are those of Giovannini and Kim (Re ¼ 23,000) [34], Fenot (Re ¼ 23,000) [35],
Baughn et al. (Re ¼ 23,750) [17], Lee and Lee (Re ¼ 20,000) [36], Baughn and
Shimizu (Re ¼ 23,300) [37], and Yan and Saniei (Re ¼ 23,000) [38]. Here, some
important differences are apparent. The LES does not succeed in capturing the
initial rise in Nu which some of the earlier experiments show to occur at r=D ¼ 0.5.
0.5. This peak is absent from the later sets of measurement [34]. The extent and
location of the secondary peak is well matched by the LES and the measurements.

Figure 11. The radial distribution of Nusselt number normalized by Re2=3.


920 N. UDDIN ET AL.

The wall-normal distribution of the LES turbulent heat fluxes are plotted at
two radial locations in Figure 12. The heat fluxes were made nondimensional using
the computed friction velocity and friction temperature. The two radial locations
selected are r=D ¼ 0.5 and 2, i.e., one close to the stagnation zone and the other
in the wall jet zone close to the secondary maximum in the radial distribution of
the Nusselt number. Close to the wall, the radial turbulent heat flux is the largest
of the three components. Away from the wall, at y=D  0.05, its magnitude
becomes of the same order as that of the tangential component. At y=D  0.2,
all the turbulent heat fluxes become roughly of the same order of magnitude. It
is interesting to note that the wall temperature increases as the wall jet region
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

develops.
Attention is turned now to consideration of the turbulent Prandtl number, a
quantity of interest in conventional modeling approaches. In a two-dimensional
flow, this quantity is defined as

u0r u0y qT
qy
Prt
ð26Þ
u0y t0 qU
qy

Using the LES results, the turbulent Prandtl number is computed at three
different radial locations in the wall jet zone. Figure 13 shows the distribution
of the turbulent Prandtl number in the wall-normal direction. It is immediately
clear that the regions of the flow where Prt is constant are adjacent to the
wall and are very limited in extent. Even there, its value is seen to be a strong
function of distance away from the stagnation point. Across the wall jet, Prt
varies quite significantly especially at the early stages of development. The
assumption of a constant Prt is therefore not justified in this complex flow
and it would be reasonable to expect that models utilizing this assumption might
not perform well.

Figure 12. Distribution of u0r t0 , u0h t0 , and u0y t0 obtained from LES at (a) r=D ¼ 0.5, and (b) 2. The
distributions are normalized by frictional velocity at pipe wall and temperature qw=qcpus.
LES AND MODELING OF IMPINGING JET 921
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 13. LES results of turbulent Prandtl number distribution at r=D ¼ 3, 4, and 6.

5. ASSESSMENT OF HEAT-FLUX MODELS


In this section, we use the LES results presented above to assess the perform-
ance of the heat-flux models. In a departure from usual practice, we focus on the
values assigned by the models originators to the various coefficients that appear in
their models. These values were determined by reference to simple, homogeneous,
and inhomogeneous shear flows, and re-evaluating them here for a much more
complex flow will serve to highlight their validity in a fully three-dimensional
strain field.
We consider first the Daly and Harlow model [24] which is based on general-
ization of the gradient diffusion hypothesis (GDH). The model contains a single
coefficient whose value, as deduced from the LES results, is plotted in Figures 14a
and 14b. The value assigned to this coefficient by the model originators was 0.3 while
the maximum value implied by the LES data analysis is around 0.15. Close to the
wall, the value deduced from LES is lower by two orders of magnitude. This large
difference is attributed to the fact that this model was not adapted for application
in the low Reynolds number near-wall flows.
In the Abe and Suga model [28], a single coefficient is again used and is
assigned the value of 0.6. The LES results suggest a much lower value, as can be seen
from Figures 15a and 15b where the maximum value obtained turns out to be around
0.2. It is interesting to note that both the Daly and Harlow and the Abe and Suga
models are essentially variations on the conventional gradient diffusion hypothesis
differing from it by allowing for the anisotropy of eddy diffusivity. Not surprisingly,
therefore, the trend of the coefficient distribution is similar in both models. This
behavior is an indication of the dominant role of GDH in the Abe and Suga model.
Recently, Abe [39] has stressed the importance of GDH for the scalar-flux modeling
in the impinging jet flow.
922 N. UDDIN ET AL.
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 14. Daly and Harlow model [24] coefficient distribution close to target wall from (a) u0y t0 at
y=D ¼ 1  103 and from (b) u0r t0 at y=D ¼ 0.02. The symbol shows the model coefficients obtained from
LES data.

The model of Younis et al. [23] contains four independent coefficients which
are evaluated here using both the radial (u0r t0 ) and wall-normal (u0y t0 ) components.
The outcome is shown in Figure 16. Using a standard mathamatical library (Matlab)
a polynomial is fitted to the values obtained and the mean, the standard deviation,
and the coefficient of determination are listed in Table 1. The closer the coefficient of
determination is to unity, the closer the fit between the model and LES. The coeffi-
cients in wall normal scalar-flux distributions are found to be more sensitive then the
radial scalar flux distributions. Because of large oscillations in the optimized coef-
ficient values, it is important to analyze the role of each term in detail.
In this model, the C3 and C4 terms represent the effect of the production terms
and scalar gradients. The need for these terms is evident from the theoretical analysis
LES AND MODELING OF IMPINGING JET 923
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 15. Abe and Suga model [28] coefficient distribution from (a) u0y t0 at y=D ¼ 1  103 and from (b) u0r t0
at y=D ¼ 0.02. The symbol shows the model coefficients obtained from LES data.

of Dakos and Gibson (1987) of the form of the term in Eq. (17) which represents the
correlations between fluctuating pressure and scalar-gradients [40]. Furthermore,
Kataoka et al. [41] have found that large scale eddies of surface pressure turbulence
play an important role in an impinging jet and their effect is also correlated by the
cross-product of temperature and velocity gradients. Therefore the inclusion of these
terms in the scalar-flux model is justified. However the analysis through LES data
shows that the model coefficients for these terms needs to be adjusted. Figures 17c
and 17d shows that the Younis et al. model underpredicts the effect of this term
at the stagnation point. To compensate for this outcome, the values of C3 and C4
are found through optimization.
The predictions of uþ þ þ þ
y t and ur t from various scalar flux models is investi-
gated using the LES data. Like turbulent kinetic energy, the turbulent scalar-flux
fluctuations are large in the developing wall jet zone. However, as the flow spreads
924 N. UDDIN ET AL.
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 16. Younis et al. model [23] coefficient distribution at (a) y=D ¼ 1  103 computed from u0y t0 and
(b) at y=D ¼ 0.02 computed from u0r t0 .

Table 1. Coefficients of Younis et al. model [23]

Coefficient of
Coeff. Original value Mean Standard Dev. determination

C1 0.0455 0.0481 0.008 0.468


C2 þ0.373 þ0.3723 0.0017 0.3424
C3 0.00373 þ0.0028 0.0084 0.432
C4 0.0235 0.0195 0.0087 0.784
LES AND MODELING OF IMPINGING JET 925
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 17. Curve fitting of the coefficient distribution at y=D ¼ 1  103 (Younis et al. model) computed
from u0y t0 .

radially, the amount of scalar-flux fluctuations is reduced and becomes comparable


to the flow through channel or duct.
Figure 18 shows the distribution of scalar-fluxes obtained from the LES and
the various heat-flux models presented above. In the stagnation zone, the fluctua-
tions in the temperature and velocities are small. All models give satisfactory results.
Near the jet’s stagnation point, it was found that the first term of the Younis
et al. model plays a dominant role and is thus important for modeling the heat trans-
fer in the stagnation zone. This term is actually of the same form as the gradient
diffusion hypothesis—a feature it shares with both the Abe and Suga and Daly
and Harlow models. Consequently, all three models exhibit roughly the same
behavior there.
As the wall jet develops, the predictions from the Younis et al. model are more
in-line with the LES than the other scalar-flux models, as can be seen in Figures 19
and 20. The predictions of the Abe and Suga and Daly and Harlow models are more
like the flow in a channel. Although the entrainment is large near the jet emergence
zone, its influence on the boundary layer is small. However, the influence of the
external entraining fluid becomes significant as the boundary layer becomes thinner,
especially in the free-stream region of the boundary layer. For distribution of the
radial scalar flux, the agreement between the LES results and the Younis et al. model
926 N. UDDIN ET AL.
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 18. Distribution of uþ þ þ þ


r t and vy t at r=D ¼ 0.25. Behavior of scalar flux models is investigated
using LES data.

are much better than the other models. It is found that terms C3 and C4 play an
important role in this connection. When C3 and C4 terms are taken as zero, the
Younis model gives the same trend of scalar-fluxes as that of other models. This
indicates that the inclusion of velocity gradients in the scalar-flux models helps in
the prediction of complex flows. However, this is invalid for the wall-normal
scalar-flux distribution.

6. CONCLUSION
Large-eddy Simulations of an impinging jet with heat transfer were performed
in order to better understand the complex dynamical processes involved, and to
provide data suitable for validation of mathematical models for the turbulent heat
fluxes. Three such models were considered. The models contain coefficients whose
LES AND MODELING OF IMPINGING JET 927
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 19. Distribution of uþ þ þ þ


r t and vy t at r=D ¼ 3. Behavior of scalar flux models is investigated using
LES data.

values were determined by reference to data from homogeneous free shear flows and
from simple wall-bounded flows, such as Couette flows and flows in straight heated
channels. A priori testing of these models using the LES results showed that the coef-
ficients associated with two of these models are significantly different from the values
deduced by the models originators from these simpler flows. This outcome is taken
here as indication of the inability of these models to correctly capture the effects on
the heat fluxes of a rapidly evolving strain field, such as that which occurs in the
vicinity of the stagnation point. The results obtained with the Younis et al. model
were relatively better though even with this model, which fully incorporates a direct
dependence on the strain field, differences with the LES results were apparent. The
same model, which is explicit in the heat fluxes and can therefore conveniently be
incorporated into practical prediction methods, also yielded the closest correspon-
dence with experimental data in a companion study of heated turbulent flow over
928 N. UDDIN ET AL.
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Figure 20. Distribution of uþ þ þ þ


r t and vy t at r=D ¼ 6.2. Behavior of scalar flux models is investigated using
LES data.

a backward-facing step [3]. As the performance of this model is strongly dependent


on the accuracy of the mean-flow and turbulence parameters that are required as
inputs, its use in actual computations can only be justified when employed in con-
junction with an accurate model for the Reynolds stresses. This is currently a subject
under active consideration by many researchers, though definite recommendations
are yet to emerge.

REFERENCES
1. N. Kasagi, Y. Tomita, and A. Kuroda, Direct Numerical Simulation of Passive Scalar
Field in a Turbulent Channel Flow, J. Heat Transfer, vol. 114, pp. 598–606, 1992.
2. M. Jischa and H. B. Rieke, About the Prediction of Turbulent Prandtl and Schmidt
Numbers from Modified Transport Equations, Int. J. of Heat and Mass Transfer,
vol. 22, p. 1547, 1979.
LES AND MODELING OF IMPINGING JET 929

3. C. Dietz, Computational Heat Transfer of Internal Flows with Vortex–Inducing Elements,


Universität Stuttgart, Germany, 2008.
4. S. Gao, D. C. Leslie, and P. R. Voke, Large-Eddy Simulation of Thermal Impinging
Jets, Rep ME-FD=91.02 Dept. Mech. Eng., University of Surrey, Guildford, UK,
1991.
5. S. Gao and P. R. Voke, Large-Eddy Simulation of Turbulent Heat Transport in Enclosed
Impinging Jets, Int. J. Heat and Fluid Flow, vol. 16, pp. 349–356, 1995.
6. M. Olsson and L. Fuchs, Large-Eddy Simulations of a Forced Semi-Confined Circular
Impinging Jet, Phys. of Fluids, vol. 10, no. 2, pp. 476–486, 1998.
7. T. Cziesla, E. Tandogan, and N. K. Mitra, Large-Eddy Simulation of Heat Transfer from
Impinging Slot Jets, Numer. Heat Transfer A, vol. 32, pp. 1–17, 1997.
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

8. T. Cziesla, G. Biswas, H. Chattopadhyay, and N. K. Mitra, Large-Eddy Simulation of


Flow and Heat Transfer in an Impinging Slot Jet, Int. J. Heat and Fluid Flow, vol. 22,
pp. 500–508, 2001.
9. M. Tsubokura, A. T. Kobayashi, N. Taniguchi, and W. P. Jones, A Numerical Study on
the Eddy Structures of Impinging Jets Excited at the Inlet, Inter. J. of Heat and Fluid Flow,
vol. 24, pp. 500–511, 2003.
10. M. Hadz̆iabdić, LES, RANS and Combined Simulation of Impinging Flows and Heat
Transfer, Ph.d. Thesis, TU Delft, 2005.
11. T. Hällqvist, Large-Eddy Simulation of Impinging Jets with Heat Transfer, Ph.d Thesis,
Royal Institute of Technology, Department of Mechanics, Sweden, 2006.
12. S. B. Pope, Turbulent Flows, Cambridge University Press, UK, 2005.
13. J. S. Smagorinsky, General Circulation Experiments with the Primitive Equations: 1. The
Basic Experiment, Mon. Weather. Rev., vol. 91, no. 3, pp. 99–164, 1963.
14. M. Germano, U. Piomelli, P. Moin, and W. H. Cabot, A Dynamic Subgrid-Scale Eddy
Viscosity Model, Phys. Fluid A, vol. 3, no. 7, pp. 1760–1765, 1991.
15. Fachgebiet Numerische Berechnungsverfahren im Maschinenbau, http://www.fnb.
tu-darmstadt.de/
16. H. L. Stone, Iterative Solution of Implicit Approximations of Multidimensional Partial
Differential Equations, SIAM J. Numer. Anal., vol. 5, no. 3, pp. 530–558, 1988.
17. J. W. Baughn, A. E. Hechanova, and X. Yan, An Experimental Study of Entrainment
Effects on the Heat Transfer from a Flat Surface to a Heated Circular Impinging Jet,
J. of Heat Transfer, vol. 113, pp. 1023–1025, 1991.
18. D. Cooper, D. C. Jackson, B. E. Launder, and G. X. Liao, Impinging Jet Studies for
Turbulence Model Assessment, I. Flow Field Experiments, Int. J. Heat Mass Transfer,
vol. 36, no. 10, pp. 2675–2684, 1993.
19. R. Viskanta, Heat Transfer to Impinging Isothermal Gas and Flame Jets, Exper. Thermal
and Fluid Sci., vol. 6, pp. 111–134, 1993.
20. ANSYS ICEM CFD, http://www.ansys.com/products/icemcfd.asp. Accessed August
15, 2008.
21. M. Klein, A. Sadiki, and J. Janicka, A Digital Filter based Generation of Inflow Data for
Spatially Direct Numerical or Large Eddy Simulations, J. of Computational Phys., vol. 18,
pp. 652–665, 2003.
22. W. M. Kays, M. E. Crawford, and B. Weigand, Convective Heat and Mass Transfer,
McGraw-Hill, 2004.
23. B. A. Younis, C. G. Speziale, and T. T. Clark, A Rationale Model for Turbulent Scalar
Fluxes, Proc. of the Royal Society A, vol. 461, pp. 575–594, 2005.
24. B. J. Daly and F. H. Harlow, Transport Equations in Turbulence, Phy. Fluids, vol. 13,
no. 11, 2634–2649, 1970.
25. M. M. Rogers, N. Mansour, and W. C. Reynolds, An Algebraic Model for Turbulent
Flux of a Passive Scalar, J. Fluid Mechanics, vol. 203, pp. 77–101, 1989.
930 N. UDDIN ET AL.

26. R. M. So and T. P. Sommer, An Explicit Heat Flux Model for the Temperature Field,
Int. J. Heat Mass Transfer, vol. 39, pp. 455–465, 1996.
27. K. Suga and K. Abe, Nonlinear Eddy Viscosity Modelling for Turbulence and Heat
Transfer Near wall and Shear-Free Boundaries, Inter. J. of Heat and Fluid Flow, vol. 21,
pp. 37–48, 1999.
28. K. Abe and K. Suga, Towards the Development of a Reynolds-Averaged Algebraic
Turbulent Scalar-Flux Model, Int. J. Heat and Fluid Flow, vol. 22, pp. 19–29, 2001.
29. B. A. Younis, B. Weigand, and S. Spring, An Explicit Algebraic Model for Turbulent
Heat Transfer in Wall-Bounded Flow with Streamline Curvature, J. of Heat Transfer,
vol. 127, pp. 425–433, 2007.
30. I. Wygnanski, Y. Katz, E. Horev, On the Aplicability of Various Scaling Laws to the
Downloaded by [Indian Institute of Technology - Delhi] at 23:26 05 December 2017

Turbulent Wall Jet, J. Fluid Mech., vol. 234, pp. 669–690, 1992.
31. I. B. Özdemir and J. H. Whitelaw, Impingement of an Axis-Symmetric Jet on Unheated
and Heated Flat Plates, J. Fluid Mech., vol. 240, pp. 503–532, 1992.
32. D. R. S. Guerra, J. Su, and A. P. S. Feire, The Near Wall Behaviour of an Impinging Jet,
Int. J. Heat and Mass Transfer, vol. 48, pp. 2829–2840, 2005.
33. K. Nishino, M. Samada, K. Kasuya, and K. Torii, Turbulence Statistics in the Stagnation
Region of an Axisymmetric Impinging Jet Flow, Int. J. Heat and Fluid Flow, vol. 17,
pp. 193–201, 1996.
34. A. Giovannini and N. S. Kim, Impinging Jet: Experimental Analysis of Flow Field and
Heat Transfer for Assessment of Turbulence Models, Ann. of the Assembly for Inter. Heat
Transfer Conf., vol. 13, 2006.
35. M. Fenot, Etude du refroidissement par impact de jets, application aux aubes de turbines,
Université de Poitiers, 2004.
36. J. Lee and S. Lee, Stagnation Region Heat Transfer of a Turbulent Axisymmetric Jet
Impingement, Exper. Heat Transfer, vol. 12, pp. 137–156, 1999.
37. J. W. Baughn and S. Shimizu, Heat Transfer Measurement from a Surface with Uniform
Heat Flux and an Impinging Jet, Int. J. of Heat Transfer, vol. 111, pp. 1096–1098, 1989.
38. X. Yan and N. Saniei, Heat Transfer Measurements from a Flat Plate to a Swirling
Impinging Jets, Proc. of 11th Inter. Heat Transfer Conf., Kyonju, Korea, 1998.
39. K. Abe, Performance of Reynolds-Averaged Turbulence and Scalar Flux Models in
Complex Turbulence with Flow Impingement, Progress in Comp. Fluid Dynamics, vol. 6,
no. 1–3, pp. 79–88, 2006.
40. T. Dakos and M. M. Gibson, On Modelling the Pressure Terms of the Scalar Flux
Equations, in Durst et al. (eds.), Turbulent Shear Flows, vol. 5, pp. 7–18, Springer,
Germany, 1987.
41. K. Kataoka, Y. Kamiyama, S. Hashimoto, and T. Komi, Mass Transfer Between a Plane
Surface and an Impinging Turbulent Jet: The Influence of Surface-Pressure Fluctuations,
J. of Fluid Mech., vol. 119, pp. 91–105, 1982.

Vous aimerez peut-être aussi