You are on page 1of 7

Heat and Mass Transfer 40 (2004) 793799

DOI 10.1007/s00231-003-0474-4

Turbulent flow and convective heat transfer in a wavy wall channel

A.Z. Dellil, A. Azzi, B.A. Jubran

Abstract This paper reports the numerical modeling of
turbulent flow and convective heat transfer over a wavy wall
using a two equations eddy viscosity turbulence model. The
wall boundary conditions were applied by using a new
zonal modeling strategy based on DNS data and combining
the standard ke turbulence model in the outer core flow
with a one equation model to resolve the near-wall region.
It was found that the two-layer model is successful in
capturing most of the important physical features of a
turbulent flow over a wavy wall with reasonable amount of
memory storage and computer time. The predicted results
show the shortcomings of the standard law of the wall for
predicting such type of flows and consequently suggest
that direct integrations to the wall must be used instead.
Moreover, Comparison of the predicted results of a wavy
wall with that of a straight channel, indicates that the
averaged Nusselt number increases until a critical value is
reached where the amplitude wave is increased. However,
this heat transfer enhancement is accompanied by an increase in the pressure drop.

general non-orthogonal coordinate system
Jacobien of the coordinate transformation
time averaged variable
convection term
D/ i
diffusion term
source term
Kronecker delta
ui uj
Reynolds stress tensor
u0j h
turbulent heat flux
turbulent kinetic energy
turbulent transport coefficient
turbulent Prandtl number
mean rate of strain
Received: 21 October 2002
Published online: 17 October 2003
Springer-Verlag 2003
A.Z. Dellil
Faculte des sciences, Universite dOran Es-senia, Algerie
A. Azzi
Faculte de Genie-Mecanique, Universite des Sciences et de la
Technologie dOran, Algerie
B.A. Jubran (&)
Sultan Qaboos University, Department of Mechanical
and Industrial Engineering, Muscat, Sultanate of Oman

ll, le

isotropic eddy viscosity

model constant, =0.082
rate of dissipation of k
length scales, eqs (7) and f(9)
von Karman constant, = 0.4
damping function, eq. (8)
turbulent Reynolds number, eq. (10)
inter wall spacing
wave amplitude
wave length
friction velocity, ( sw =q)
Reynolds number, (HU/m)

Wavy wall flows occur under a wide variety of engineering
applications and have, consequently, received considerable
attention [12]. One of the most important applications is
the heat transfer enhancement in heat exchangers. The
physical process of enhancing heat transfer in such
application is to introduce some geometrical modifications
on the wall in question in order to break the boundary
layer that forms on the exchanger wall and replace it by a
fresh fluid from the free stream flow [3]. In real applications, engineers are also interested in the additional
pressure drop caused by such techniques. So, the best
solution is that provides the least pressure drop and the
largest heat transfer rate. Other parameters such as simplicity, manufacturability, maintenance, etc., are also
important parameters in the design phase [3].
Wavy wall heat transfer enhancement technique has
been used extensively in the design of compact heat
exchangers as can be seen from the numerous experimental and numerical investigations reported in the literature. One of the important observations in this field is
that in the laminar regime, where wavy passages provide
significant heat transfer enhancement when the flow is
unsteady. However, for steady flow, the enhancement is
insignificant. Wang and Vanka [3] conducted a numerical
investigation for laminar steady and unsteady cases with
wavy wall. They reported a heat transfer enhancement
factor of about 2.5 for the unsteady case compared with
that for a parallel-plate channel case. However, for the
steady case with wavy wall the heat transfer rate is only
slightly improved. The transition occurs at a Reynolds
number of around 180 for the geometrical configuration
studied by Wang and Vanka [3].


In most industrial process the flow is always turbulent.

Consequently, wavy wall flows are receiving renewed
attention. Recently, Hudson et al. [4] studied experimentally a rectangular water channel where the lower wall was
constructed with removable Plexiglas plates on which the
waves were milled. To insure that a fully periodic flow had
developed, the LDV measurements were made above the
31st of 36 waves. The wavelength and the amplitude were
set to (k = H) and (am = 0.05 H), respectively, where H is
the mean height of the channel. This experimental work is
one of the few studies which provides extensive measurements of the Reynolds stresses. It was followed by
numerical simulation of a fully developed turbulent flow
over a wavy wall [5] and [6]. This type of turbulent flow is
numerically very challenging, since it is characterized by
periodic changes of the pressure gradient, curved
streamlines, and for important wave amplitude, separation
and reattachment can occur. The work reported by Maa
and Schumann [6] is largely documented, and hence it was
selected as a benchmark to validate the dynamic part of
the present computation. In addition to the dynamic field,
the present paper reports a numerical parametric-investigation on forced convective heat transfer in a wavy wall
channel. Detailed flow and thermal fields were obtained by
solving RANS (Reynolds Averaged Navier Stokes equations) equations and a two-equation eddy viscosity turbulence model. The wall boundary conditions were applied
by using a new zonal modeling strategy based on DNS data
and combining the standard ke turbulence model in the
outer core flow and a one equation model resolving the
near-wall region. This strategy was initialized by Rodi et al.
[7] and tested with success in previous computations [8].

and S/ is the source term. Table 1 shows the above terms

for each considered dependent variable; yi denotes a
reference Cartesian coordinate system while xi is a general
non-orthogonal coordinate system.
The Reynolds-stress tensor and the turbulent heat
flux are approximated within the context of the ke
turbulence model and the near-wall viscosity-affected
region is resolved with a one-equation model [7]. The
two-layer approach represents an intermediate modeling
strategy between wall function and pure low-Reynolds
number model. It consists of resolving the viscosity-affected regions close to the wall with a one-equation
model, while the outer core flow is treated with the
standard ke model. In the outer core flow, the usual
eddy-viscosity hypothesis is used, applying a linear
relation of the Reynold-stress tensor to the velocity
gradient as follows:
u0i u0j dij k  2Cij Sij
u0j h 

Cij @ T
Prt @xj


where dij is the Kronecker delta, k  u0i u0j =2 is the turbulent

kinetic energy, and Sij is the rate of strain tensor. For highRe flows, the turbulent transport coefficient Gij is conventionally made isotropic and proportional to a velocity
scale (3k) and a time scale (k/e), characterizing the local
rate of turbulence and is given by:
Cij  mt  Cl k2 =e

where mt is known as the isotropic eddy viscosity, e represents the rate of dissipation of k, and Cl stand for a
model constant. The distributions of k and e are determined from the conventional model transport equations of
Jones and Launder [9], and standard values can be
assigned to the model constants. In this flow region the
turbulent Prandtl number is usually fixed at 0.9. In the one
equation model, the eddy viscosity is made proportional to
a velocity scale determined by solving the k-equation, and
a length scale ll prescribed algebraically. The dissipation
rate e is related to the same velocity scale and a dissipation
length scale le, also prescribed algebraically [7]. Such
Where / is the considered time-averaged variable, J is the model has the advantage of requiring considerably fewer
Jacobian of the coordinate transformation, Ci, Di/ repre- grid points in the viscous sub-layer than any pure Low
Reynolds scheme. Also, because of the fixed length-scale
sent respectively the convection and the diffusion terms
distribution near the wall, these models have been found
to give better prediction for adverse pressure gradient
Table 1. Form of terms in the individual equations
boundary layer than pure ke models.
The present two-layer model is a re-formulated version
of the so-called v2 velocity-scale based model (TLV)
b jqvj

proposed by Rodi et al. [7]. In a recent study Azzi and
i j
 1J @x@ j pbk
bij qvj
 lJ Bij @v
@xj bj xk
Lakehal [8] re-incorporate k1/2 as a velocity scale instead

i @k
PK qe
 rk J Bj @xj
b jqvj
v2 1=2 and ll and le are re-scaled on the basis of the same

DNS data of Kim et al. [10]. This model will be call
 rlek J Bij @x
Ce1 ke Pk  Ce2 q ek
hereafter as TLV model and defined as follows:
Bij cofactor of J @yi =@xj ; Bij bil bl

Cij  mt  Cl kll
l @mj
n @ml
m @mj
n @ml

The mathematical model
The mathematical model consists of the RANS, the twoequation eddy viscosity ke turbulence model and the
energy equation. The governing equations for steady,
turbulent, incompressible flows in non-orthogonal coordinates using Cartesian velocity components can be
written in a generalized form as follows:

1 @ 
Ci / D/i S/ ;
J @xi

xk bk @xl ; Pk







l ll lt ; l=rk  ll lt =rk ; l=re  ll lt =re

e k3=2 =le

are 4k and 2k in length respectively. For the inlet conditions, a similar procedure to that used by Patel et al. [2] is

applied here. The distributions of velocity and turbulence
0:116  R2y Ry
8 parameters in a fully developed flow at sufficiently large
distance from the entry of 80H straight channel, which are
independent of the initial conditions and invariant with
jCl3=4 y

9 distance, are used as inlet conditions for wavy-wall calle
2 17:28= fl Ry jCl
culations. The Reynolds number was set as in DNS computations [6] at Re(HU/m) = 6760. The turbulence intensity
10 is assigned a value of 5% and the turbulence dissipation is
Ry q ky=l
calculated based on a turbulent viscosity equal 50 times
Where j = 0.4 and Cl = 0.082. The outer and the near-wall the laminar viscosity. At the outflow boundary, zero-gramodel are matched at the location where fl = 0.95, indi- dient conditions are imposed for all dependent variables.
cating that viscous effects become negligible. More details
can be found in [8], where the model is tested for a fully 2.3
developed channel and applied for a film cooling config- Grid mesh
The quality of a computational solution is strongly linked to
the quality of the grid mesh. So a highly orthogonalized,
nonuniform, fine grid mesh was generated with grid nodes
considerably refined in the near-wall region. The normalNumerical procedure
The numerical procedure used to calculate the test case is ized y+ values at the near wall node are less than unity, and
based on a finite-volume approach for implicitly solving care is taken so that the stretching factors are kept close to
the incompressible Reynolds Averaged Navier Stokes
unity. Figure 1 shows a close-up of the computational grid
equations (RANS) on arbitrary non-orthogonal grids,
used for the case with am = 0.10 H. The grid adopted for the
employing a cell-centred grid arrangement. The momen- computations was obtained after a series of tests and contum-interpolation technique of Rhie and Chow [11] is used sists of 42,100 grid nodes (disposed on a global array 421
to prevent pressure-field oscillations and the pressure100 nodes in x, and y directions). For the parametric study,
velocity coupling is achieved using the SIMPLEC algoseveral grids were generated, where the wave amplitude was
rithm of Van Doormal and Raithby [12]. The resulting
varied from 0.0 to 0.10 by a step of 0.02.
system of the algebraic difference equations is solved using
the Strongly Implicit Procedure (SIP) of Stone [13]. The 3
convection fluxes are approximated by employing the
Results and discussion
QUICK scheme for all variables applied in a scalar form by
means of a deferred-correction procedure and bounded by 3.1
the Van-Leer Harmonic function as limiter. The diffusive Comparison with DNS data
fluxes are, however, approximated using second-order
In order to validate the code the first run was done with
central differences. Convergence was in all cases deterflow characteristics that are the same as those used by
mined based on a drop in normalized mass and momen- Maa and Schumann [6]. Figures 2 and 3 compare the
tum residuals by four orders of magnitude. A global mass normalized mean velocity and turbulent kinetic energy
balance algorithm was employed after each iteration to
profiles with DNS data. The comparison is done until the
correct the mass fluxes at the outflow. More details of the middle of the channel and at four streamwise locations of
mathematical formulation, which are now a standard
the seventh wave that correspond to the divergent, trough,
material and well known to most investigators, can be
convergent and crest positions, respectively (x/H = 0.304,
found in the following references [1415].
0.492, 0.804 and 0.992). The figures show reasonable
agreement for the mean velocity profiles. The reverse flow
occurring in the first half of the wave (x/H = 0.304 and
0.492) is clearly captured by the computations. However,
Computational domain and boundary conditions
The computational domain consists of 10 waves preceded the turbulent kinetic energy is globally underpredicted. As
and followed by upstream and downstream flat sections
for flow adjustment and recovery, respectively. This geometrical configuration was used previously by Patel et al.
[2] and seems to be more exact than using one wave length
of channel and applying periodic conditions. According to
Patel et al. [2] the periodic boundary conditions cannot be
established without prior knowledge of the flow, particularly with regard to the turbulent quantities and calculations of the type performed here.
The wavy and the flat plate are placed with a mean
spacing (H = 1). The amplitude and the wavelength of the
lower sinusoidal wavy wall are (am = 0.05H) and (k = H), Fig. 1. Close-up of the computational grid for the case with 0.1 H
respectively. The upstream and downstream flat sections of amplitude
ll jyCl3=4 fl


Fig. 2. Comparison of normalized mean streamwise velocity with

DNS data 6: x/H=0.304: divergent; x/H=0.492: trough; x/H=0.804:

convergent; x/H=0.992: crest

Fig. 3. Comparison of normalized turbulent kinetic energy with

DNS data 6: x/H=0.304: divergent; x/H=0.492: trough; x/H=0.804:

convergent; x/H=0.992: crest

Fig. 4. Law of wall for Velocity and temperature at trough and

crest of the wavy wall, am=0.05

it was observed in an earlier test on flow over flat plate [8],

this underprediction is a consequence of an overpredicted
value of the turbulence dissipation in the region where k+
reaches its peak value. However, the underprediction is
more important in this wavy wall case than it was for the
flat plate. It can also be observed that turbulent kinetic
energy is represented by very different plots along the
wave and until approximately 0.25 H normal to the wall.
The maximum peak is observed at x/H = 0.492 in the
streamwise direction and roughly at 0.05H normal to the
Figure 4 shows the normalized logarithmic axial
velocity and temperature profiles at trough and crest of the
seventh wave. In the figure, U+ and y+ are computed with
the friction velocity Us based on the local wall shear stress.
The standard velocity and temperature logarithmic laws
are also plotted in the figure. For comparison, the computed profiles for the flat plate and for the trough and crest
of wavy wall (0.05 H) are presented. As it was found by
Patel et al. [2], the most important observation is that a
standard law is clearly not applicable in such complicated
configuration. The strong adverse and favorable pressure
gradients are responsible for the new trend shown by the
plots. At the crest station, which corresponds to mildly
favorable pressure gradient, the velocity and temperature
distributions are underpredicted compared with the logarithmic law. At the second station, which corresponds to
the trough, the strong adverse pressure gradient is

responsible for the overprediction of the velocity profile

compared with that of the logarithmic law. Between the
two stations, it can be seen and as expected that there is a
zone where the profiles agree with the logarithmic law due
to the near zero pressure gradient.

Influence of the wave amplitude
The effect of geometric parameters is investigated by
varying the amplitude wave from zero (flat plate) to 0.1 H
by a step of 0.02 H. The friction and the pressure coefficients distributions for trough one wave in the fully
developed region are plotted in figures 5 and 6, respectively. Comparing the results for flat plate (am = 0) with
that obtained for the wavy wall, it can be seen that there is
a decrease in the values of friction coefficient in the trough
and an increase in the values in the crest. Except for very
small amplitude (am = 0.02 H), the shape of the distribution curves is highly deformed with a tendency to
increase values even in the trough. The separation and the
reattachment points where the friction coefficient vanishes
are plotted in figure 7. Separation and reattachment points
correspond to locations where the friction coefficient
vanishes. For (am = 0.5 H) they are at 0.13 k and 0.58 k,
respectively. In DNS computations [6], they were located
at 0.14 k and 0.59 k, respectively. The agreement is completely satisfying. It is also obvious from the figure that the


Fig. 5. Friction coefficient versus amplitude wave

Fig. 8. Stream function and vector plot

Fig. 6. Pressure coefficient versus amplitude wave

Fig. 7. Separation and reattachment points versus amplitude


reattachment point is more sensitive to amplitude waves

than the separation point. The pressure coefficient shows
also an undulated character with maximum in the vicinity
of the trough and minimum nearly in the crest (opposite
to that observed for the friction coefficient). It is obvious
that the pressure drop is proportional to the amplitude
wave (Cp is zero at the inlet of the channel for all cases).
Figures 8(a) and 8(b) show vector plots and computed
streamlines for wave amplitude of 0.05 H and 0.10 H,

respectively. One can see that separation is observed for

both cases. From figure 5, which shows the friction coefficient distribution, one can predict that separation occurs
first, roughly at am = 0.03 H, since for am = 0.2 H the
friction coefficient is always positive and reaches negative
values for am = 0.4 H. We can also see that the recirculation zone increases in size and its centre moves in the
downstream direction when the amplitude wave increases.
The turbulence intensity contours for the corresponding
geometry are plotted in figures 9(a) and 9(b). The maximum turbulence zone, which is located near the wavy wall,
increases in intensity and moves in the downstream
direction when the amplitude wave increases. This feature
corresponds to what is expected from the wavy wall in
perturbing the boundary layer and enhancing turbulent
heat transfer.
The temperature contours are presented for the same
two amplitude waves in figures 10(a) and 10(b). Having in
mind that zero value is assigned to the wall and unity value
for the inflow, one can see that reattachment points are
always recognized by high temperature gradient. We can
also see that temperature contours, which are straight in
the flat-plate channel, are distorted in the recirculation
region when increasing the amplitude waves. This supports the fact that the boundary layer is perturbed by
eddies. The variation of the local Nusselt number through
one wave in the fully developed region versus the amplitude wave is presented on figure 11. The minimum and the
maximum Nusselt numbers are located near the separation and the reattachment points, respectively. The difference between the maximum and the minimum values,
increases with the amplitude wave. However, as shown in
figure 12 the averaged Nusselt number increases with the


Fig. 11. Variation of local Nusselt number for various amplitude


Fig. 9. Turbulence intensity contours (in %)

Fig. 12. Variation of averaged Nusselt number versus amplitude


presented. The wavy channel studied corresponds to the

geometry of Maa & Schumann [6]. Special attention has
been focused on periodic, fully developed thermal and flow
fields. Computations have been performed by a finite
volume method, solving flow and energy equations. The
Reynolds-stress tensor and the turbulent heat flux are
approximated within the context of the ke turbulence
model. The near-wall viscosity-affected region is resolved
with a one-equation model. The predicted results are in
good agreement with published DNS data. The two-layer
model is found to be successful in capturing most of the
important physical features of such a flow with reasonable
amount of memory storage and computer time. The predicted results show the shortcomings of the standard law
of the wall for predicting this type of flows and conseFig. 10. Contours of a dimensional temperature
quently suggest that direct integrations to the wall must be
used instead. A geometrical parametric study was conducted by changing the amplitude-to-wavelength ratio.
amplitude wave until 0.06H, after which the Nusselt
Comparison of predicted results of a wavy wall with that of
number is approximately remains constant.
a straight channel indicates that the averaged Nusselt
number increases until a critical value is reached where the
amplitude wave is increased. However, this heat transfer
Numerical results for the turbulent flow and heat transfer enhancement is accompanied by an increase in the pressure drop.
in a two-dimensional channel with wavy wall are

1. Vijay KG; Maji PK (1988) Laminar flow and heat transfer in a
periodically converging diverging channel. Int J Numer Meth
Fluids 8: 579597
2. Patel VC; Tyndall Chon J; Yoon JY (1991) Turbulent flow in a
channel with a wavy wall. J Fluids Eng 113: 579586
3. Wang G; Vanka SP (1995) Convective heat transfer in periodic
wavy passages. Int J Heat Mass Transfer 38(17): 32193230
4. Hudson JD; Dykhno L; Nanratty TJ (1996) Turbulent production
in flow over a wavy wall. Exp Fluids 20: 27265
5. Cherukat P; Na Y; Hanratty TJ (1998) Direct numerical simulation
of a fully developed turbulent flow over a wavy wall. Theor
Comput Fluid Dynam 11: 109134
6. Maa C; Schumann U (1996) Direct numerical simulation of
separated turbulent flow over a wavy boundary. In: Hirschel EH
(ed) Flow simulation with high performance computers notes on
numerical fluid mechanics, 52: 227241
7. Rodi W; Mansour NN; Michelassi V (1993) One equation nearwall turbulence modelling with the aide of direct simulation data.
J Fluids Eng 115: 196205
8. Azzi A; Lakehal D (2002) Perspectives in modelling film-cooling of
turbine blades by transcending conventional two-equation turbulence models. ASME J Turbomach 124: 472484

9. Jones WP; Launder BE (1972) Prediction of relaminarisation with

a two-equation turbulence model. Int J Heat Mass Transfer 15:
10. Kim J; Moin P; Moser R (1987) Turbulence statistic in fully
developed channel flow at low Reynolds number. J Fluid Mech
177: 133166
11. Rhie CM; Chow WL (1983) A numerical study of the turbulent
flow past an isolated airfoil with trailing edge separation. AIAA-J
21: 12251532
12. Van Doormal JP; Raithby GD (1984) Upstream to elliptic problems
involving fluid flow. Comput Fluids 2: 191220
13. Stone HL (1968) Iterative solution of implicit approximation of
multidimensional partial differential equations. SIAM J Num Anal
5: 53
14. Majumdar S; Rodi W; Zhu J (1992) Three-dimensional finitevolume method for incompressible flows with complex boundaries. J Fluids Eng 114: 496503
15. Zhu J (1992) An introduction and guide to the computer program
FAST3D. Report No. 691, Institute for Hydromechanics, University of Karlsruhe