You are on page 1of 13

International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow


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

Turbulence statistics in a rotating ribbed channel


Vagesh D. Narasimhamurthy , Helge I. Andersson
Fluids Engineering Division, Department of Energy and Process Engineering, Norwegian University of Science and Technology (NTNU), NO-7491 Trondheim, Norway

a r t i c l e

i n f o

Article history:
Available online xxxx
Keywords:
Surface roughness
System rotation
Square rods
Reynolds stress budget
Coriolis forces

a b s t r a c t
The combined effects of system rotation and rib-roughness on turbulent channel ow have been investigated by means of direct numerical simulations. The 40 wall-units high square ribs were placed on both
walls in a non-staggered arrangement with pitch-to-height ratio 8. Mean ow elds and turbulence statistics for rotation numbers Ro 2 and 6 were presented and compared with corresponding data for
Ro 0. The ow eld in the vicinity of the ribs was affected differently on the two sides of the rotating
channel. The separated ow region behind the pressure-side ribs shrinked with increasing Ro and the
originally d-type roughness turned into a k-type roughness. At Ro 6, a pressure-loss reduction of about
20% was found. In spite of the 10% blockage due to the ribs, the ow eld exhibited a statistical homogeneity in the streamwise direction over more than half of the channel. The turbulence statistics were
substantially affected by the system rotation with enhanced and reduced turbulence levels along the
pressure and suction sides, respectively. Among other things, we also learned that the combined inuences of rib-roughness and system rotation on the turbulent ow eld in rotating channels cannot be
foreseen by straightforward superposition.
2014 Elsevier Inc. All rights reserved.

1. Introduction
The turbulent ow of a liquid or a gas through plane channels
and straight rectangular ducts represents classical prototype ows
in uid mechanics. Internal cooling ducts in gas turbines are often
equipped with ribs aimed to enhance the turbulent mixing and
thereby augment the heat transfer rate and the cooling efciency.
The ow and temperature elds in a cooling channel are inevitably
also affected by the blades rotation. In this paper we aim to
explore in detail the combined effects of spanwise ribs and system
rotation on the ow eld in a rotating channel. Although this constitutes a problem of generic interest, the outcome might also be
benecial for the turbulence treatment in computational uid
dynamics (CFD) software aimed at turbomachinery applications;
see e.g. Johnston (1998) and Tucker (2013).
A common feature of almost all predictions of ribbed channel
ows, be it either direct or large-eddy simulations, is that only
one of the walls is roughened by means of transverse ribs while
the other wall remains smooth. In some of these studies the onesided roughness was partly motivated by the laboratory experiments of Hanjalic and Launder (1972), in which only one channel
wall was roughened with the intention to produce an asymmetric
mean velocity prole. The inuence of surface-mounted ribs on the
Corresponding author.
E-mail addresses: vagesh@alumni.ntnu.no (V.D. Narasimhamurthy), helge.i.
andersson@ntnu.no (H.I. Andersson).

mean ow and the turbulence eld depends on the size of the ribs,
typically the rib height k, and the inter-rib spacing k  w, where w
is the rib width and k is the pitch. It is common practice to distinguish between k-type and d-type behavior. k-type roughness
implies that the length of the cavity between two consecutive ribs
is large compared to the rib height, i.e. k  w  k. Physically this
implies that the shear layer which separates from the corner of a
rib reattaches to the bottom of the cavity. On the contrary, if the
bulk ow is skimming the uid trapped between two consecutive
ribs, the conguration is classied as d-type and the effect of the
ribs are not felt all across the channel; see e.g. Perry et al. (1969),
Jimnez (2004) and Leonardi et al. (2007). A number of direct
numerical simulation (DNS) studies of turbulent ow in rib-roughened channels have appeared after the turn of the century, as summarized by Jimnez (2004) and Krogstad et al. (2005). A
comprehensive investigation on the effect of the pitch-to-height
ratio k=k of ribs on one channel wall was presented by Leonardi
et al. (2003), whereas Ashraan et al. (2004) and Ashraan and
Andersson (2006a,b) studied the effects on the ow eld when
both walls were equipped with k 8k ribs. In laboratory ows,
as well as in DNS and LES, we believe that the ow eld in the
vicinity of a ribbed wall is fairly independent of whether the other
wall is smooth or ribbed. In the core region of the channel, however, the inuence of both walls is felt and the ow eld in the center region is thus different in channels with one-sided roughness
from those with two-sided roughness.

http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008
0142-727X/ 2014 Elsevier Inc. All rights reserved.

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

Pressure-driven turbulent ow in a plane channel conguration


is a prototype problem in uid mechanics and the DNS data provided by Moser et al. (1999) and Abe et al. (2001) are frequently
used for benchmarking. The statistical symmetry about the midplane of the channel is immediately broken if the channel is set
in orthogonal-mode rotation about a spanwise axis. This phenomenon was reported more than four decades ago by Johnston et al.
(1972). Their illuminating ow measurements and visualizations
showed that the mean velocity prole became asymmetric and this
was ascribed to the stabilizing effect of the Coriolis force at the suction side and the destabilizing effect at the pressure side of the
rotating channel. These striking effects of system rotation on an
originally symmetric plane channel ow were reproduced by
Launder et al. (1987) and Launder and Tselepidakis (1994) by solving the Reynolds-averaged NavierStokes equations (RANS) with
second-moment closures. More recent experiments have conrmed and supplemented these ndings, e.g. the hot-wire anemometry measurements by Nakabayashi and Kitoh (1996, 2005)
and the particle-image velocimetry (PIV) measurements by
Visscher et al. (2011). Turbulent ows through a rotating channel
has also been simulated numerically by means of DNS. The results
thus obtained by Kristoffersen and Andersson (1993) and
Lamballais et al. (1996) are fully consistent with the early observations and measurements by Johnston et al. (1972).
The inuence of the Coriolis force which arises from the
imposed rotation depends on the magnitude of the angular
velocity X as well as on the orientation of the rotation axis relative
to the mean ow vorticity vector. It is common practice to dene a
local rotation number S 2X=dU=dy for plane channel ows
where dU=dy is the vorticity of the mean ow (y is the coordinate
normal to the channel walls). The local rotation number S is positive on one side of the Poiseuille ow channel where the ow is
exposed to cyclonic rotation and negative at the other side which
is exposed to anti-cyclonic rotation. The notion of cyclonic and
anti-cyclonic rotation stems from geophysical uid dynamics and
refers to whether the uid rotation tends to be in the same (cyclonic) or opposite (anti-cyclonic) direction as the system rotation;
see e.g. Andersson (2010). Bradshaw (1969) demonstrated that a
uniform shear ow subjected to constant system rotation is
unstable when N SS 1 < 0 and otherwise stable. Both S 0
and S 1 are neutral situations whereas S  12 is the most
unstable situation. Cambon et al. (1994) emphasized that
although linear analysis suggested neutral stability for S 0 and
S 1, the ow eld will be fundamentally different in these
two situations.
A rotating plane Poiseuille ow is thus simultaneously affected
by cyclonic rotation S > 0 at the suction side and anti-cyclonic
rotation S < 0 at the pressure side of the channel since the mean
velocity Uy attains a maximum value somewhere in the center
region and obeys no-slip at both walls. At low and moderate rates
of rotation the following observations were made in the aforementioned studies:
 The mean velocity prole Uy became asymmetric and the
position of the maximum velocity was shifted away from the
mid-plane towards the suction side.
 The mean velocity seemed to develop a region with a linear variation such that dU=dy  2X. In this region S  1, i.e. neutral
stability.
 The turbulence intensity was reduced and sometimes even suppressed near the suction side and enhanced at the pressure side
of the channel.
 Streamwise-oriented counter-rotating roll cells occurred at the
pressure side where 1 < S < 0; i.e. the so-called Bradshaw
number N is negative.

The ndings summarized above apply for plane channel ows


where the mean velocity vector and thus the streamlines are parallel with the channel walls. Further complexing ow features
occur in the presence of a sudden expansion of the channel, either
a one-sided expansion as the so called backward-facing step (BFS)
ow or a double-sided expansion (Lamballais, 2014). With the
view to investigate the effects of system rotation on a turbulent
shear layer Rothe and Johnston (1979) examined the free shear
layer which arises when a ow separates from the corner of a
BFS. This particular ow conguration has been widely used for
verication of NavierStokes solvers against benchmark data and
for testing and validation of turbulence models in RANS-based
computer codes. Rothe and Johnston (1979) used the same rotating
apparatus as in the earlier investigation of rotating plane channel
ow (Johnston et al., 1972). Their ow visualisations showed that
the shear layer which separated from the corner of the BFS reattached earlier to the wall downstream of the step when the channel was subjected to anti-cyclonic rotation S < 0, i.e. if the step
was on the pressure side, than if the rotation was cyclonic
S > 0 with the step on the suction side. These primary effects of
rotation on BFS ow were mimicked by RANS calculations by
Nilsen and Andersson (1990) using a second-moment closure.
More recently, DNS studies by Barri and Andersson (2010) and
PIV measurements by Visscher and Andersson (2011) conrmed
the early ow visualizations by Rothe and Johnston (1979) and
provided in addition detailed mean ow and turbulence statistics
on separated ows affected by a Coriolis force due to system
rotation.
Surface-mounted ribs and system rotation both affect the turbulent ow eld in plane channels and ducts. RANS-based computations of the combined effects of transverse ribs and spanwise
rotation on the ow and heat transfer have been reported by
Iacovides (1998) and Raisee et al. (2009). Saha and Acharya
(2005) considered a duct with ribs in a staggered arrangement
on the pressure and suction side with the aim to compare largeeddy simulations (LES) with unsteady RANS-predictions. Detailed
ow eld measurements using PIV have more recently been presented by Coletti and Arts (2011) and Coletti et al. (2012, 2013).
They considered an almost square duct with transverse ribs
mounted on one of the four walls. Fransen et al. (2013) carried
out LES of the same ow conguration. Their simulations compared favorably with the PIV measurements for the stabilizing case
when the ribbed wall was exposed to cyclonic rotation whereas
substantial differences were observed when the duct was subjected to destabilizing anti-cyclonic rotation. Abdel-Wahab and
Tafti (2004a,b) and Viswanathan and Tafti (2006) predicted the
ow and heat transfer in a rotating square duct with surfacemounted ribs by means of large-eddy and detached eddy simulations (DES), respectively. Whereas Fransen et al. (2013) used a
computational domain which comprised 8 rib pitches k, the
domain used by Abdel-Wahab and Tafti (2004b) and
Viswanathan and Tafti (2006) covered only a single rib pitch.
Table 1 provides an overview of the most relevant computational
studies of turbulent ows in rib-roughened channels and ducts
with and without system rotation.
The effects of system rotation on shear ows with almost parallel streamlines are well understood and can be explained in terms
of the mean rotation rate vector and the system rotation vector. In
the presence of ribs, however, the mean ow separates from a rib
and forms a zone of recirculating ow just behind the rib. The associated uid rotation may either be in the same or opposite sense of
the system rotation, e.g. cyclonic or anti-cyclonic, and the magnitude of this localized uid rotation is typically smaller than the
uid rotation associated with the mean ow gradient along a
smooth wall.

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

Table 1
Overview of the most relevant simulations of turbulent ows in square-ribbed channels. The entries are arranged in chronological order. The rib height k and the pitch k are the
essential geometrical parameters. The table does not distinguish between Reynolds numbers Re and rotation numbers Ro based on h (half of the distance between the walls) or
half of the hydraulic diameter. The rough approximation us  0:05U b has been used for the entry by Fransen et al. (2013).

Author(s)

Method

Re

ribs

k=h

k=k

Romax

Leonardi et al. (2003)

DNS

180

Ashraan et al. (2004)


Abdel-Wahab and Tafti (2004a,b)a
Viswanathan and Tafti (2006)a
Narasimhamurthy and Andersson (2009)
Fransen et al. (2013)a
Present study

DNS
LES
DES
LES
LES
DNS

400
3300
3300
400
390
400

1  30
12
2  24
21
21
28
18
28

0.2
0.2
0.034
0.2
0.2
0.1
0.2
0.1

1.33
20
8
10
10
8
10
8

2.0
2.0
6.3
8.0
6.0

Duct ow simulations with aspect ratio 1.

Table 2
Computational parameters in some relevant plane channel ow simulations arranged in chronological order. The length Lx and width Lz of the computational domain are given in
terms of the channel half-height h.

Author(s)

Lx

Lz

Nx  Ny  Nz

Dx

Dz

Moser et al. (1999)


Abe et al. (2001)
Leonardi et al. (2003)
Ashraan et al. (2004)
Abdel-Wahab and Tafti (2004a,b)a
Viswanathan and Tafti (2006)a
Narasimhamurthy and Andersson (2009)
Fransen et al. (2013)a
Present study

2p
6.4
8
6.53
2.0
2.0
6.4
18.3
6.4

256  193  192


256  192  256
400  140  97
768  160  160
128  128  128
64  64  64
256  128  128
Unstructured
512  320  200

10
9.88
3.6
3.2

10

6.5
4.94
5.83
7.85

10

6.4

3.2

p
p
2.0
2.0
3.2
1.8
3.2

Duct ow simulations with aspect ratio 1 and therefore variable grid spacings.

The rst computer simulations of turbulent ow in an orthogonal-mode rotating plane channel with transverse ribs on both walls
were presented by Narasimhamurthy and Andersson (2009). These
early results were obtained using a fairly coarse mesh on which the
tiniest scales of the ow eld were not adequately resolved. These
simulations can therefore be considered as large-eddy simulations
without an explicit subgrid-scale model. Nevertheless, the results
from this coarse-mesh simulations were found to compare reasonably well with results obtained on a rened grid (Narasimhamurthy
and Andersson, 2011). The aim of the present study is to provide a
comprehensive coverage of the turbulence statistics in a rotating
ribbed channel for three different rotation numbers Ro = 0, 2, and
6 both at mid-cavity and mid-rib locations. In our earlier paper
(Narasimhamurthy and Andersson, 2011) only results at mid-cavity
were presented and only for Ro 6.
2. Flow conguration and numerical method
The ow conguration is as shown in Fig. 1. Roughness elements of square cross-section are considered with height
k=h 0:1 and pitch k=k 8. The present choice of pitch value corresponds to the so-called k-type laboratory roughness (Perry et al.,
1969). The computational domain comprised 8 square ribs on each
channel wall. The reason for having 8 ribs was motivated by the
need to make the length Lx of the computational domain longer
than 6h and thereby allow the two-point correlations of the velocity components to decay to zero for streamwise separations of Lx =2.
The driving pressure-gradient dP=dx is prescribed such that the
Reynolds number based on the channel half-width h and the
1=2
wall-friction velocity us q1 hdP=dx
is equal to 400. This is
essentially the same Reynolds number as the medium Re case
reported by Moser et al. (1999) and Abe et al. (2001) for smooth
channel ows and by Ashraan et al. (2004) for a rod-roughened
channel ow (see Table 2). The roughness height k is therefore
equal to 40 times the viscous length scale m=us . The denition of

Fig. 1. Flow conguration (not to scale). The actual computational conguration


comprises eight square ribs on each side of the channel with k w 0:1h and pitch
k 8k 0:8h. The system is rotating with constant angular velocity X about the
spanwise z-axis.

the wall-friction velocity us is routinely used in DNS and LES studies of smooth channel ows, e.g. Moser et al. (1999) and Abe et al.
(2001). In a ribbed channel ow, however, the relative contribution
of viscous wall friction to the overall pressure drop depends on the
rib height k=h, the pitch k=w (k=h), and the Reynolds number Re
(and in the present case also on the imposed system rotation). Nevertheless, us can be adopted as a velocity scale representative of
the driving pressure gradient dP=dx also for ribbed channel ows
even though most of the pressure drop is caused by form drag
(Ashraan et al., 2004).
The NavierStokes equations for an incompressible and isothermal ow in a constantly rotating frame of reference can be written
as,

~i
~
~
~i
@u
@u
@p
1 @2u
~j i 
~k
u

 ei3k Rou
@t
@xj
@xi Re @x2j

~i
@u
0
@xi

~ i and xi have been non-dimensionalized by us and h. u


~ i is the
where u
instantaneous velocity component in the xi -direction in a Cartesian
coordinate system (see Fig. 1) which rotates along with the channel
with a constant angular velocity X about the z-axis. The last term on

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

1.5

and represents an order-of-magnitude ratio between the Coriolis


force and inertia. The centrifugal effects are absorbed in the effec~p
~s  18 Ro2 r2 , where p
~s is the normalized static prestive pressure p
sure and r denotes the dimensionless distance from the axis of
rotation. Periodic boundary conditions are employed in the streamwise and spanwise directions. No-slip and impermeability conditions are imposed on all the rigid surfaces. In the present study
three different rotational numbers are considered, i.e. Ro = 0, 2
and 6.
The governing equations are solved in three-dimensional space
and time using the parallel nite volume code MGLET (Manhart,
2004). The code uses staggered Cartesian grid arrangements. Spatial discretizations of the convective and diffusive uxes are carried
out using a second-order central-differencing scheme. The
momentum equations are advanced in time by a fractional-step
method using a second-order explicit AdamsBashforth scheme.
The Poisson equation for the pressure is solved by a full multi-grid
method based on pointwise velocitypressure iterations. The computational grid is divided into an arbitrary number of subgrids that
are treated as dependent grid blocks in parallel processing. In the
present study, the size of the computational domain in each coordinate direction Lx  Ly  Lz is 6:4h  2h  3:2h, i.e. practically the
same as in the DNS case of Moser et al. (1999), Abe et al. (2001)
and Ashraan et al. (2004) (see Table 2). In wall-units it will be

L
x  Ly  Lz equal to 2560  800  1280, respectively. The twopoint correlation data from Abe et al. (2001) indicate that this
domain size is sufcient for the Reynolds number considered. In
addition, spanwise two-point correlation data from the present
study also support their ndings.
The number of grid points in each coordinate direction
N x  N y  N z is shown in Table 2. Uniform grid spacing is adopted
in the streamwise and the spanwise directions, while a non-uniform mesh is used in the wall-normal direction. The actual grid
is chosen such that grid lines coincide with all sides of the roughness elements and no-slip and impermeability are enforced in the
same manner as at the channel walls without the need of an
immersed boundary method. The square-shaped cross-section of
a rib comprises 8 grid points in the streamwise direction and 40
grid points in the wall-normal direction. The grid resolution close

to the wall is xed to Dy


w 1 and kept constant until y 40,
i.e. up to the crest of the roughness elements. Onwards from there
the mesh is gradually stretched towards the center line to achieve
Dyc 6:6. The ne grid up to y  40 is required in order to adequately resolve the extreme gradients of the velocity variables in
the boundary layer which develops along the upper surface of each
rib (as we will see in the results sections). It is worth mentioning
here that we have also carried out a grid sensitivity test (see
Narasimhamurthy and Andersson (2011)). The total consumption
of CPU time for one case was nearly 60,000 h on an IBM P575+ parallel computer.
3. Results
The direct numerical simulations result in time-varying threedimensional ow elds. After the ow eld has reached a statistically steady state, statistics were obtained by averaging both in
time and in the homogeneous z-direction. The time-averaging
was performed over 400 samples separated 0:05h=us in time, i.e.
over 20 large-eddy time units. The roughness elements induced a

1
0.5
1

3
x/h

3
x/h

3
x/h

(b)
1.5
y/h

2Xh
Ro
us

(a)
y/h

the right hand side of Eq. (1) represents the Coriolis force due to
system rotation and eijk is the alternating unit tensor. The rotation
number Ro is dened as,

1
0.5

(c)
1.5
y/h

1
0.5

Fig. 2. Instantaneous turbulent kinetic energy u2 v 2 w2 =2 for various rotation


numbers: (a) Ro 0; (b) Ro 2; and (c) Ro 6. Contour values are normalized by
u2s and they range from 0.0 (white) to 8.0 (blue). The two black circles in panel (c)
highlights the energetic layers referred to in the text. (For interpretation of the
references to colour in this gure legend, the reader is referred to the web version of
this article.)

streamwise periodicity of the averaged ow eld, i.e. a statistical


equivalence between two points x; y; z and x nk; y; z where k
is the pitch and n is an integer. A pitch-averaging was also utilized
to further improve the quality of the nal statistics. Mean values
thus obtained will be denoted by capital letters and uctuations
~ V v is the velocity component in
by lower case letters, e.g. v
the wall-normal y-direction. A variety of ow statistics will be presented in this paper. Since all statistics vary over the x; y-plane,
we concentrate on wall-to-wall proles at the midpoint of the cavity between two subsequent ribs x=k 0:43 and at the midpoint
of a rib x=k 0:93. It should be noted that these two locations are
chosen to match with the grid points closest to the exact mid-cavity and mid-rib locations x 0:4375k and x 0:9375k.
Let us rst of all consider examples of the instantaneous ow
eld. The kinetic energy of the uctuating ow eld
1
u2 v 2 w2 is shown in Fig. 2. It is readily observed that the
2
energy level is higher in the vicinity of the ribbed wall than in
the center region of the channel. In the lee of each rib, however,
a rather low energy level can be seen (the ow is from left to right).
The instantaneous ow eld in the vicinity of one rib is generally
rather different from the ow eld around a neighboring rib, as
one should expect. Highly spatially and temporally resolved PIV
measurements by Coletti et al. (2013) showed strong rib-to-rib
interactions, i.e. ow features triggered by the rib-induced separation affect the velocity eld up to the following rib.
It is intuitively obvious that the wall-mounted ribs tend to
enhance the turbulence level as compared with a smooth wall.
Although the present ribs are about three times bigger than those
of Ashraan and Andersson (2006b), the ow eld in their
Fig. 12(b) closely resembles the present Fig. 2(a). Since the Reynolds number in the DNS study by Ashraan and Andersson
(2006a,b) was the same as here, the pronounced difference in rib
height remains the same in global and wall units. In the presence
of rotation, however, a more vigorous ow eld is observed along
the lower wall in Fig. 2(b and c) while an almost quiescent ow is

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

seen at the upper wall of the channel. These qualitative observations are in accordance with the well-known effects of system rotation in smooth-walled channels, namely a destabilization of the
ow along the anti-cyclonic (pressure) side and a stabilization of
the uid motion at the cyclonic (suction) side (Johnston et al.,
1972; Kristoffersen and Andersson, 1993). At the highest rotation
number Ro 6, distinct high-energy layers can be seen to emerge
upstream of almost all ribs along the upper wall in Fig. 2(c). These
noteworthy ow structures are probably cyclonic vortex sheets
which appear more pronounced when the surrounding turbulence
is damped by the stabilizing rotation. Since the shear layers are
essentially two-dimensional, these shear layers will not be suppressed by stabilizing cyclonic rotation. This situation compares
qualitatively with the cyclonically rotating mixing layer studied
by Bidokhti and Tritton (1992).
Let us denote the streamwise and wall-normal components of
the mean velocity vector Ux; y and Vx; y, respectively, whereas
the mean velocity W in the z-direction is zero due to spanwise
homogeneity. The variation of the mean streamwise velocity U is
shown in Fig. 3. The proles are distinctly different in the near-wall
region. Nevertheless, the proles collapse in the central part of the
channel, at least in the center region from y=h 0:5 to 1.5. Without
rotation, the prole is symmetric about the centerline at y=h 1:0
and exhibits a modest backow U < 0 along the cavity walls. The

16
14
12

Ro = 6

10
Ro = 2

8
6

Ro = 0

4
2
0
-2

0.5

1.5

y/h

Fig. 3. Mean streamwise velocity proles Uy non-dimensionalized by the wall


friction velocity us for various rotation numbers. Proles are taken at two locations:
() mid-cavity position x=k 0:43; (. . ..) mid-rib position x=k 0:93. The broken
straight line has slope 2X.

y/h

(c)

y/h

0.1
0
0
0.2

0.2

0.4

0.6

0.8

0.1
0
0
0.2

0.2

0.4

0.6

0.8

0.1
0
0

(e)
y/h

y/h

(b)

(d)

0.2

(f)
y/h

y/h

(a)

peak value of U ( U=us ) is 8.78. This is considerably lower than


the peak value 20.13 in a smooth channel ow at the same Reynolds number; Moser et al. (1999). This substantial decrease is
obviously caused by the presence of the ribs. The present decrease
DU 11:35 is much larger than DU  7:0 reported by Ashraan
et al. (2004) for the same pitch-to-height ratio at the same Re but
with smaller ribs. Leonardi et al. (2003) reported DU  13:2 for a
case with twice as high ribs but at a signicantly lower Reynolds
number Re 180.
The distinctly different mean velocity proles in Fig. 3 suggest
that the bulk ow rate through the channel may be changed due
to the imposed rotation. Indeed, the Reynolds number Reb based
on the bulk ow velocity decreases modestly from 5056 to 4904
when the rotation number increases from 0 to 2. At Ro 6, however, Reb 6064 which amounts to a 20% increase of the ow rate.
Since the driving mean pressure gradient is the same in all three
cases, the enhanced ow rate is equivalent with pressure loss
reduction.
When the rib-roughened channel is subjected to rotation about
a spanwise axis, the symmetry of the mean velocity prole is broken. Somewhat away from the pressure side of the channel, the
mean velocity U seems to increase almost linearly with y and the

slope dU =dy is higher for Ro 6 than for Ro 2. However, the


position at which the maximum velocity is observed is offset about
0:4h from the channel center and the effect seems to be independent of Ro. Once again, the mean velocity prole midway between
two ribs is indistinguishable from the prole observed in the center region of the channel at the rib location.
Streamlines deduced from the mean ow are presented in
Fig. 4. The results for the non-rotating case are shown in the two
upper panels and exactly the same topology is observed at both
sides of the channel. The mean ow separates from the corner of
a rib and forms a large separation bubble which lls about 2/3 of
the cavity. The separated shear layer seems to reattach at the bottom of the cavity. However, inspection of the wall shear stress (not
shown here) shows that sw remains negative almost all along the
bottom of the cavity between two consecutive ribs. Here, a saddle
point is formed slightly above the wall at x  0:65h and separates
the large separation bubble from the minor recirculation zone
formed just upstream of the next rib. The circulation in both these
recirculation zones are in the clockwise direction at the lower wall,
whereas a small anti-clockwise bubble can be seen in the lee of the
upstream rib. The latter secondary bubble arises when the return
ow U < 0 along the bottom of the cavity separates due to a local
adverse pressure gradient.

0.2

0.4

0.6

0.8

2
1.9
1.8
0
2

0.2

0.4

0.6

0.8

0.2

0.4

0.6

0.8

0.2

0.4

0.6

0.8

1.9
1.8
0
2
1.9
1.8
0

Fig. 4. Streamlines illustrating the recirculation zones for various rotation numbers: (ac) pressure side of the channel; (df) suction side of the channel. (a, d) Ro 0; (b, e)
Ro 2; (c, f) Ro 6. Here x-axis corresponds to a single pitch x=k.

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

(a)

(b)

=
Ro

Ro
Ro

0
0

(c)

0.5

Ro = 2

=6

=2
1.5

y/h

Ro = 6

0
0

Ro = 0
0.5

y/h

1.5

1.5

(d) 1.5

0.5
3

Ro

0
Ro = 6

Ro

-0.5

Ro

-1
Ro = 2

0.5

=2
=6

-1.5

Ro = 0

0
0

=0

1.5

y/h

-2

0.5

y/h

Fig. 5. Reynolds stress proles normalized by u2s for various rotation numbers. Symbol (+) in uv plot corresponds to non-rotating smooth-channel data of Moser et al. (1999).
See Fig. 3 for details.

(b)

(a)
0.5

2.5 3
2.5

1.8
2

1.6

(c)
0.5

1.5

1.5

1.8
1.5

1.5

1.6

1.6
1

1.4

1.4

0.5

1.5

1.8

1.4

y/h

1.2

1.2

0.8

0.8

0.8

0.6
0.4

0.4

2
2.5

0.2
0.5

1.5

0.6

1.5

2.5 3

0.2 0.4 0.6 0.8


x/

1.2

0.2
0.5

2
2.5

0.6
0.4

2
2.5

1.5

3.5
4
3.5

0.2 0.4 0.6 0.8


x/

0.2
0.5

3.5
3.5

4.5

0.2 0.4 0.6 0.8


x/

Fig. 6. Contours of TKE normalized by u2s : (a) Ro 0; (b) Ro 2; and (c) Ro 6.

In the presence of system rotation, the topology of the streamlines becomes distinctly different at the two sides of the channel.
At the suction side (right column) the saddle point which separated the major and minor anti-clockwise circulations is shifted

away from the cavity wall and the secondary separation bubble
seems to vanish so that a continuous backow exists all along
the cavity wall. The ow along the rib elements is now clearly of
the skimming type, i.e. d-type according to Perry et al. (1969). At

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

(d) 0.02

(a)
0.7

Ro = 2 Ro = 6

0.01
Ro = 0

0.6

-0.01

Ro = 0
+

0.4

Cuu

Puu

0.5

0.3

Ro = 2
-0.02
-0.03

0.2
0.1

-0.04

-0.05

-0.1

0.5

y/h

1.5

-0.06

Ro = 6

0.5

1.5

1.5

1.5

y/h

(e) 0.06

(b) 0.08
0.06

0.05
Ro = 6

0.04

0.04

Ro = 6

Ro = 2
0.02

0.03
+

Cvv

Pvv

Ro = 0
0

0.02
Ro = 2

-0.02

0.01

-0.04

-0.06

-0.01

Ro = 0
-0.08

(c)

0.5

y/h

1.5

-0.02

0.5

(f) 0.04

0.2

Ro = 6

0.15
0.02

0.1

Ro = 2
0

Ro = 0

Cuv

Puv

0.05
0

-0.02

-0.05
Ro = 0

-0.1

-0.04

-0.15
-0.2

y/h

Ro = 2 Ro = 6
0

0.5

y/h

1.5

-0.06

0.5

y/h

Fig. 7. Mean-strain production rate P ij (left panel) and rotational production rate C ij (right panel) terms normalized by u4s =m for various rotation numbers. Note that
P ww C ww 0. See Fig. 3 for details.

the pressure side of the rotating channel (left column), however,


the ow which separates from the rib does now reattach to the
bottom of the cavity at Ro 6 and the major recirculation zone
is clearly separated from the minor zone. The reduction of the
major recirculation zone downstream of the rib is in qualitative
agreement with earlier experimental studies of the ow over a
rotating backward-facing step by Rothe and Johnston (1979) and

Visscher and Andersson (2011) and DNS studies by Barri and


Andersson (2010). The originally d-type ow pattern has therefore
been turned into k-type roughness on the pressure side. In our earlier reports (Narasimhamurthy and Andersson, 2009, 2011) we
overlooked that the ow in the non-rotating case was of d-type
and therefore concluded that k-type roughness was turned into
d-type roughness at the suction side.

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

(a)

Ro = 2
Ro = 0, 6

0.2
0.1

0.2

(b)

x 10

10

-0.1

0.1

Ro = 0
0

0.05

0.1

0.15

0.2

Ro = 0

0.1

Ro = 2, 6

-0.1

Ro = 2, 6

vv

-0.2

D+

D+uu

-3

15

0
-0.1

-0.2

1.85

0.4
0.3

0.3

Ro = 6

0.2

Ro = 2

0.5

1.5

y/h

-3

x 10

(d)

Ro = 0

Ro = 2, 6

0.05

Ro = 2, 6

-0.1
0

ww

Ro = 0

0.1

D+

1.5

y/h

-5

Ro = 0

0.1

0.2

1.95

0.05

0.1

-0.05

1.9

1.95

0.1

+
Duv

(c)

0.5

1.9

-5

-10

-0.1
0

0.5

y/h

1.5

-15

0.5

1.5

y/h

Fig. 8. Viscous diffusion term Dij normalized by u4s =m for various rotation numbers (inset: magnied views of pressure- and suction-sides). See Fig. 3 for details.

In order to explain the preceding observations, the four nonzero components of the Reynolds stress tensor are presented in
Fig. 5. Without rotation the diagonal components vary symmetrically across the channel whereas the off-diagonal or shear-stress
component exhibits an anti-symmetric variation. The latter is compared with DNS data by Moser et al. (1999). As for the mean velocity proles in Fig. 3, the Reynolds stresses in the central region of
the channel are independent of the actual streamwise position
whereas substantial differences can be observed near the ribs. In
presence of rotation, the situation is dramatically changed. If the
behavior in the near-wall region is discarded, the streamwise uctuations are reduced at the pressure side whereas the uctuations
in the spanwise and wall-normal directions are increased with
increasing rates of rotation and so is the magnitude of the covariance uv . At the suction side of the channel, however, all the four
Reynolds stress components are reduced when the rib-roughened
channel is subjected to rotation.
The results for the diagonal components of the Reynolds stress
tensor presented in Fig. 5(ac) can be summarized in contour plots
of the mean turbulent kinetic energy (TKE) in Fig. 6. Here, TKE is
dened as the time and spatial average of 12 u2 v 2 w2 shown
in Fig. 2. These plots conrm that the Reynolds stresses and thus
TKE is essentially independent of streamwise position in the center
region of the channel. It should be recalled that the mean velocity
proles in Fig. 3 exhibited an almost linear variation over a substantial region from y  0:5h to about y 1:3h in the rotating
channel. The turbulent kinetic energy in this central region of the
ow is clearly increasing with the rate of rotation. The same

distinct trend was reported by Andersson and Kristoffersen


(1995) in a rotating channel with smooth walls. Here, TKE
increases monotonically with Ro not only in the core region but
also at the pressure side of the channel. The instantaneous ow
eld in the vicinity of the suction-side ribs in Fig. 2 is distinctly
altered when Ro is increased from 2 to 6. Nevertheless, the contour
lines of TKE near the ribs are surprisingly unaffected.
The striking effects of the system rotation can be explained by
means of the exact transport equation for the second-moments
of the velocity uctuations:

Dui uj
Pij C ij Dij Gij T ij Pij  eij
Dt

where the right-hand-side terms, namely, production due to mean


strain (P ij ), production due to rotation (C ij ), viscous diffusion (Dij ),
pressure diffusion (Gij ), turbulent diffusion (T ij ), pressurestrain
rate (Pij ) and the viscous dissipation (eij ) are dened as:

Pij ui uk

@U j
@U i
 uj uk
;
@xk
@xk

C ij 2Xd3k uj um eikm ui um ejkm


5

Dij m

@ 2 ui uj
1 @
@
; Gij 
pu d puj dik ; T ij 
ui uj uk
@xk
@xk @xk
q @xk i jk
6



p @ui @uj
;
Pij

q @xj @xi

eij 2m

@ui @uj
@xk @xk


7

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

0.1

Ro = 2
Ro = 6

0.05

Ro = 0

-0.05

-0.1
0

(b)

0.35
0.3
0.25

8
+

Gvv

0.5

1.5

y/h

0.3

Ro = 6

0.1

0.2

Ro = 2

0.05

0.1

Ro = 0

0.15

0.1

-0.05
1.85

0.2

1.9

1.95

0.1
0.05
0
-0.05

0.5

1.5

y/h

(c) 0.2
0.1
0
uv

where the rightmost expression applies in the core region of the


channel where V  0. The mean strain production Pvv is zero in a fully
developed plane channel ow in which V 0. In absence of ribs, no
production of wall-normal velocity uctuations takes place without
rotation. The energy associated with vv is redirected from the
streamwise velocity component by means of pressurestrain interactions Pvv . In the present case, however, the bulging of the streamlines
in the vicinity of the ribs in Fig. 4 shows that V 0 in the near-wall
region, at least within 23 rib heights from each of the walls. The
mean ow is typically directed towards the wall in between the ribs
and away from the wall above the ribs, as explicitly shown by
Ashraan et al. (2004) in their Fig. 4(a). The strain rates @V=@x and
@V=@y give rise to modest sources Pvv > 0 or sinks Pvv < 0 just
above the ribs as well as in the cavities between the ribs.
When the ribbed channel is subjected to rotation about the
spanwise axis, also the rotational source term C vv comes into play.
With anti-clockwise rotation X > 0; C vv > 0 at the pressure side
and C vv < 0 along the suction side of the channel. The changeover
from being a rotational source to becoming a sink term is colocated with the change-of-sign of the Reynolds shear stress uv
in Fig. 5d. It is noteworthy that this particular position is shifted
from the channel center towards the suction side in the presence
of rotation, just as in a rotating plane channel ow (Johnston
et al., 1972; Kristoffersen and Andersson, 1993). In the central part
of the channel C vv attains an appreciable level which increases
with the rate of rotation. Since P vv is practically zero, the positive
C vv gives rise to a rather extreme increase of the wall-normal
velocity uctuations, as already observed in Fig. 5b. This phenomenon is well-known from studies of rotational effects in plane
channel ow (Johnston et al., 1972; Kristoffersen and Andersson,
1993). In the present case, however, the production of vv due to
mean strain P vv is also affected by system rotation. This is, however, an indirect effect either due to the rotational-induced
changes of the mean ow eld in Fig. 4 or most likely due to the
enhanced magnitude of uv and vv on the pressure side and the
corresponding reduction along the suction side as shown in
Fig. 5d and b, respectively.
The mean strain P uu is crucial for the turbulence generation in
turbulent ows in a non-rotating plane channel. In that canonical
ow, turbulence is produced due to interactions between the

0.2

-0.1

0.2

G+



@V
@V
 4Xuv  4Xuv
Pvv C vv 2 uv
vv
@x
@y

(a)

+
Guu

This differential equation can be derived from the governing


Eqs. (1) and (2) without any approximations; see e.g. Launder
et al. (1987) and Hanjalic and Launder (2011). It is noteworthy that
the left-hand side comprises not only the local variation @ui uj =@t
but also the advective change U k @ui uj =@xk . The latter contribution
arises due to the presence of the ribs, whereas the former is zero
in the present statistically steady ow. For the sake of completeness, all the different terms on the right-hand side of Eq. (4) are
shown in Figs. 712. We will see that some of the terms are by
far more important than others. Within the framework of a second-moment closure of the Reynolds-averaged NavierStokes
equations, however, all terms are required in a model.
The key to explain the dramatic changes in the turbulence eld
in the presence of system rotation is the production terms associated with mean strain P and rotation C. These terms are shown in
Fig. 7 for the streamwise and wall-normal directions as well as for
the off-diagonal component. The corresponding terms in the spanwise direction, i.e. P ww and C ww , are identically zero in the present
ow case since the two-dimensional mean ow in the x; y-plane
is statistically homogeneous in the spanwise z-direction.
It is tempting to rst look at the production terms for wall-normal uctuations:

-0.2

-0.3

-0.2

-0.4

-0.4

0.1

Ro = 0
Ro = 2
Ro = 6
0

-0.5

0.05

0.5

0.1

0.15

y/h

-0.1
1.85

1.5

1.9

1.95

Fig. 9. Pressure diffusion term Gij normalized by u4s =m for various rotation numbers
(inset: magnied views of pressure- and suction-sides). Note that Gww 0. See
Fig. 3 for details.

Reynolds shear stress and the mean ow gradient, i.e.


Puu 2uv @U=@y P 0, and the source term is unconditionally
positive since the shear stress changes sign at the symmetry plane
where the mean velocity U peaks. This fact motivated Hanjalic and
Launder (1972) to add roughness elements on only one side of
their channel in order to allow P uu to locally attain negative values.
The presence of ribs also induces a streamwise variation of U so
that the total production becomes:

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

10

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

(a)

0.2

0.1

0.15

(b) 0.05
0

-0.1

0.1

0.1

0.2

0.3

-0.05

0
0.05

-0.05

-0.1

-0.1

0.5

-0.1
-0.15

-0.05

-0.15
-0.2

Ro = 0

Tvv

0.05

T+uu

Ro = 2, 6
Ro = 0

1.8

y/h

1.9

-0.2

1.5

Ro = 2

1.5

1.5

y/h

(d) 0.1

(c) 0.05

Ro = 2
Ro = 6

Ro = 0

0.05

-0.05

Tuv

Ro = 0

Tww

Ro = 6
0.5

-0.1
Ro = 2

-0.15
Ro = 6
-0.2
0

0.5

y/h

1.5

-0.05

0.5

y/h

Fig. 10. Turbulent diffusion term T ij normalized by u4s =m for various rotation numbers (inset: magnied views of pressure- and suction-sides). See Fig. 3 for details.





@U
@U
@U
4Xuv  2uv
Puu C uu 2 uu
uv
 2X
@x
@y
@y

Ashraan et al. (2004) argued that the ow is decelerated in the


immediate vicinity of a rib so that the secondary production
2uu@U=@x becomes positive around the rib. This is indeed conrmed in the present case where highly localized peaks of the dotted lines are observed in Fig. 7(a).
The rotational production of uu is exactly opposite to that of vv
since C uu C vv , as evident from Figs. 7d and e. The extra production due to the system rotation therefore tends to reduce the
streamwise uctuations at the pressure side and generate higher
uctuations on the suction side of the channel. The proles in
Fig. 5a show that uu is indeed reduced at the pressure side, at least
outside of the rib-affected region. However, uu is reduced also on
the suction side of the channel. In the wide region where
@U=@y  2X, the positive P uu is practically outweighed by the negative C uu , see the rightmost part of Eq. (9), so that the total source
of streamwise uctuations almost vanishes. On the suction side of
the mean velocity peak, however, the observed decay of uu is
caused by the rotational-induced reduction of uv in Fig. 5d rather
than by the slightly positive C uu . The modications of the streamwise velocity uctuations on the suction side is therefore not a
direct Coriolis force effect on u but rather an indirect effect through
the covariance of u and v. The turbulence eld is affected directly
by the system rotation through the rotational source term C ij

dened in Eq. (5). This is the only explicit appearance of the angular velocity X in the exact second-moment Eq. (4). These direct
changes lead, in turn, to indirect changes in the mean strain production Pij and other terms in the second-moment Eq. (4).
The covariance of u and v is determined by the mean strain and
rotational production terms:



@V
@U
@U
 2Xuu  vv  vv
Puv C uv  uu
vv
@x
@y
@y
 2Xuu  vv

10

It should be noticed that we have used the mean of the mass conservation equation to show that:

uv



@U @V
0

@x @y

11

and thereby to simplify the above expression for Puv .


The primary mean strain production P uv exhibits different signs
on the two sides of the channel and this is directly reected in the
change-of-sign of uv in Fig. 5d. In absence of rotation, both Puv and
uv vary anti-symmetrically from one side to the other. Contrary to
Puv , the rotational production C uv is independent on the mean ow
eld and remains negative as long as the turbulent ow eld
exhibits the usual anisotropy with the streamwise velocity uctuations exceeding the wall-normal uctuations. This is the case in
the near-wall regions on both sides of the rotating channel, but

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

11

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

(a)

(b)

0.15

0.05

Ro = 6
0.1

Ro = 2

-0.05
Ro = 0

Ro = 0

-0.1

vv

uu

0.05
0
-0.05

-0.15
Ro = 2

-0.2
-0.25

-0.1

-0.3
0

(c)

0.5

y/h

1.5

(d)

0.2

Ro = 6

0.5

Ro = 2

0.05

Ro = 0

uv

+
ww

0.1

0.4

0.2

0.3

0
-0.2

0.2

1.5

y/h

Ro = 6

0.4

Ro = 6
0.15

0.5

0.1

Ro = 2
Ro = 0

0
-0.1

0.05

0.1

0.15

-0.2
1.85

1.9

1.95

0.1
0

-0.05

-0.1
0

0.5

y/h

1.5

-0.2

0.5

y/h

1.5

Fig. 11. Pressurestrain rate term Pij normalized by u4s =m for various rotation numbers (inset: magnied views of pressure- and suction-sides). See Fig. 3 for details.

not in the core region where vv > uu. Here, C uv changes sign and
accordingly tends to oppose P uv .
Diffusive transport of second-moments ui uj may arise from viscous effects (Dij ) as well as from turbulent pressure (Gij ) and velocity (T ij ) uctuations. In channel-ow turbulence these terms are
non-negligible only in the near-wall regions. Since the source
terms P ij and C ij both tend to zero at a solid wall, the various diffusion terms become increasingly important as the wall is
approached even though they are practically zero in the core
region of the channel. For the sake of completeness, however, the
terms representing viscous, pressure, and turbulent diffusion are
shown in Figs. 810, respectively. In contrast with a smooth-wall
channel ow, e.g. Moser et al. (1999), all components of the various
diffusion terms except Gww are non-zero in the rib-roughened
channel since the presence of the ribs destroys the statistical
homogeneity in the streamwise direction. The various diffusion
terms are generally enhanced near the pressure side and reduced
at the suction side when the ribbed channel is subjected to rotation. Aside from the turbulent diffusion T ij (see Fig. 10b and d)
the different terms remain negligible in the core region of the
channel. The nite T vv > 0 over most of the region where
@U=@y  2X suggests that the velocity uctuations provide fairly
efcient diffusive transport of vv from the pressure side towards
the suction side. This happens to occur in the region where the
conventional mean-strain production Puu is practically outweighed
by the rotational source C uu .

The components of the pressurestrain correlation tensor are


presented in Fig. 11. In conventional wall turbulence
Puu Pvv Pww < 0 since turbulent energy is transferred
from streamwise to spanwise and wall-normal uctuations. This
route of inter-component energy transfer can also be observed
above the cavity center. Near the bottom of the cavity, however,
turbulence energy is instead transferred from the wall-normal
direction (Pvv < 0) to the streamwise and spanwise directions.
This reversal of the conventional inter-component energy transfer
has also been reported by Ashraan and Andersson (2006a) for
substantially lower ribs. These effects are more accentuated at
the pressure side of the rotating channel than in a non-rotating
channel, whereas reduced energy transfer rates are seen at the suction side.
The nal term in the second-moment transport Eq. (4) is the
viscous energy dissipation rate tensor eij ; see Fig. 12. The highest
values of the individual components of eij are found in the immediate vicinity of the cavity wall and at the top surface of the ribs.
The streamwise component euu in Fig. 12a exhibits its maximum
value at the wall, but a secondary peak can be seen ush with
the top of the ribs. The dissipation evv of the wall-normal velocity
uctuations is substantially increased with increasing rotation and
seems to meet the enhancement of vv in Fig. 5b caused by the
rotational production C vv in Fig. 7e. The off-diagonal component
euv in Fig. 12d attains non-negligible levels, especially in the
near-wall regions. The results in Fig. 12 show beyond any doubt

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

12

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

(a)

(b) 0.03

Ro = 2
Ro = 0, 6

0.2

0.25
0.025

0.1

0.05

0.15

0.02

0.15

0.2

0.1

0.015
0.01

0.1

Ro = 6

0.05
0

0.1

+
vv

0
1.85

(c) 0.5

0.5

1.5

1.5

y/h

Ro = 6
Ro = 2

0.02

0.05

0.5

Ro = 0

Ro = 0

0.1
0

0.05

0.1

0
0.15 1.85

1.9

1.95

0.01

+ww

1.5

Ro = 0

(d) 0.03

Ro = 2

0.2

y/h

0.1

0.3

0.3

1.95

Ro = 6

0.4

0.4

1.9

Ro = 2

0.005

+
uv

uu

0.2

0.2

0.1

-0.01

0.5

Fig. 12. Dissipation rate term

y/h

1.5

-0.02

0.5

y/h

eij normalized by u4s =m for various rotation numbers (inset: magnied views of pressure- and suction-sides). See Fig. 3 for details.

that the turbulent energy dissipation is anisotropic in the channel


center as well as in the near-wall regions. The qualitative trends of
the proles of the different dissipation rate components seem to be
unaffected by rotation.
The data presented in Figs. 712 are meant to give a complete
coverage of how the different terms in the second-moment (or
Reynolds stress) Eq. (4) vary from the pressure to the suction side
of the channel at different rates of rotation. The data has been presented as proles across the channel at two characteristic streamwise locations and may readily be compared with results from
RANS-based predictions. Alternatively, the data can be used to
compare a typical modeled term with the data for the exact term
obtained from our DNSs.

4. Conclusions
We have been able to study in some detail the combined effects of
system rotation and rib-roughness on the turbulent ow in a plane
channel by means of DNS. Even at a rotation number Ro 2, the
mean ow eld and the turbulence statistics became distinctly different from the corresponding results for the non-rotating channel.
The effects of rotation are stronger, and in some aspects different, at
the highest rotation number Ro 6 considered in this paper.
In spite of the 10% blockage due to the ribs, the ow eld exhibited a statistical homogeneity in the streamwise direction over
more than half of the channel. However, the ow eld in this

homogeneous core region was also affected by the Coriolis force


due to the imposed system rotation. The mean velocity prole
Uy became asymmetric with the peak shifted towards the suction
side of the channel and at the same time tended to a linear variation U  y at the pressure side, similarly as in a smooth-wall channel subjected to orthogonal-mode rotation; see e.g. Johnston et al.
(1972) and Kristoffersen and Andersson (1993). The typical anisotropy of the Reynolds-stress tensor in turbulent shear ows were
inverted so that wall-normal velocity uctuations became more
energetic than streamwise uctuations.
The turbulence statistics in the vicinity of the wall-mounted
square ribs were substantially affected by the system rotation
and enhanced and reduced turbulence levels were observed along
the pressure and suction sides, respectively. At the rotation rates
considered the most apparent effects on the turbulence eld can
be interpreted in terms of alterations in the mean-strain and rotational production terms (5) in the budget Eq. (4) for the individual
second-moments. The statistical data provided demonstrate the
substantial inuence of the Coriolis force on the individual components of the Reynolds-stress tensor when the ribbed channel ow
is rotated about the spanwise axis. The rotational or Coriolis terms
C ij dened in Eq. (5) do not appear naturally in the transport equation for the TKE since the contraction C ii is identically zero. Turbulence closures based on the assumption of an isotropic eddy
viscosity, like the k  e and the k  x models, do therefore not
respond adequately to system rotation. This calls for turbulence
closures at the second-moment level as pioneered by Launder

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

V.D. Narasimhamurthy, H.I. Andersson / International Journal of Heat and Fluid Flow xxx (2014) xxxxxx

et al. (1987) and Launder and Tselepidakis (1994); see also Jakirlic
et al. (2002) and the recent textbook by Hanjalic and Launder
(2011).
The vast majority of DNS and LES investigations of smooth-wall
and rough-wall channel ows, with and without rotation, use periodic boundary conditions in the streamwise direction. The present
length of the computational domain, Lx 8k, is sufcient to allow
the two-point velocity correlations to decay to zero for locations
Lx =2 apart in the streamwise direction. If rib-to-rib interactions
occur, as recently reported in the PIV measurements of Coletti
et al. (2013), we believe that the present choice of Lx is also sufcient to eliminate any adverse effects from the enforced periodicity.
We have learned that the combined inuences of rib-roughness
and the Coriolis force on the turbulence eld in rotating channels
cannot be foreseen by straightforward superposition. Let us, for
instance, consider the dramatic damping of the velocity uctuations in the streamwise direction at the suction side of a rotating
smooth-walled channel, as reported by Johnston et al. (1972) and
Kristoffersen and Andersson (1993). In the present case, uu is also
reduced, but the damping observed in Fig. 5(a) is much more modest and an appreciable level of uu still persists in the near-wall
region in the presence of stabilizing rotation. This is apparently
due to conventional production associated with a signicant mean
shear-rate where the fast bulk ow is skimming the slowly recirculating ow in the rib cavity; see Fig. 4(f). In turbomachinery applications the aerothermal eld is also of major importance for the
performance of the device or the cooling system. The temperature
eld and the heat transfer rate are, however, crucially dependent
on alterations in the mean ow eld in general and on the turbulent ow characteristics in particular. A reliable heat transfer analysis therefore depends crucially on a faithfully reproduced or
predicted ow eld.
Acknowledgments
The work reported herein has been supported nancially by the
Research Council of Norway (RCN) through the project Roughness
and rotating uid turbulence (Project No. 171725/V30). Computing time has been granted by RCN (Programme for Supercomputing) and by the Program in Computational Science and
Visualization (BVV) at The Norwegian University of Science and
Technology (NTNU).
References
Abdel-Wahab, S., Tafti, D.K., 2004a. Large eddy simulation of ow and heat transfer
in a 90 ribbed duct with rotation effect of coriolis forces. In: ASME Turbo
Expo 2004. GT2004-53796.
Abdel-Wahab, S., Tafti, D.K., 2004b. Large eddy simulation of ow and heat transfer
in a 90 ribbed duct with rotation: effect of Coriolis and centrifugal buoyancy
forces. ASME J. Turbomach. 126, 627636.
Abe, H., Kawamura, H., Matsuo, Y., 2001. Direct numerical simulation of a fully
developed turbulent channel ow with respect to the Reynolds number
dependence. ASME J. Fluids Eng. 123, 382393.
Andersson, H.I., 2010. Introduction to the effects of rotation on turbulence. In: Bilka,
M., Ramboud, P. (Eds.), VKI Lecture Series 2010-08 on Effect of System Rotation
on Turbulence with Applications to Turbomachinery. von Karman Institute for
Fluid Dynamics, Belgium, ISBN-13 978-2-87516-010-2.
Andersson, H.I., Kristoffersen, R., 1995. Turbulence statistics of rotating channel
ow. Turbulent Shear Flows, 9. Springer Verlag, Berlin, pp. 5370.
Ashraan, A., Andersson, H.I., 2006a. Roughness effects in turbulent channel ow.
Prog. Comput. Fluid Dyn. 6, 120.
Ashraan, A., Andersson, H.I., 2006b. The structure of turbulence in a rodroughened channel. Int. J. Heat Fluid Flow 27, 6579.
Ashraan, A., Andersson, H.I., Manhart, M., 2004. DNS of turbulent ow in a rodroughened channel. Int. J. Heat Fluid Flow 25, 373383.
Barri, M., Andersson, H.I., 2010. Turbulent ow over a backward-facing step. Part 1.
Effects of anti-cyclonic system rotation. J. Fluid Mech. 665, 382417.
Bidokhti, A.A., Tritton, D.J., 1992. The structure of a turbulent free shear layer in a
rotating uid. J. Fluid Mech. 241, 469502.
Bradshaw, P., 1969. The analogy between streamline curvature and buoyancy in
turbulent shear ows. J. Fluid Mech. 36, 177191.

13

Cambon, C., Benoit, J.P., Shao, L., Jacquin, L., 1994. Stability analysis and large-eddy
simulation of rotating turbulence with organized eddies. J. Fluid Mech. 278,
175200.
Coletti, F., Arts, T., 2011. Aerodynamic investigation of a rotating rib-roughened
channel by time-resolved particle image velocimetry. Proc. IMechE Part A: J.
Power Energy 225, 975984.
Coletti, F., Maurer, T., Arts, T., Di Sante, A., 2012. Flow eld investigation in rotating
rib-roughened channel by means of particle image velocimetry. Exp. Fluids 52,
10431061.
Coletti, F., Cresci, I., Arts, T., 2013. Spatio-temporal analysis of the turbulent ow in a
ribbed channel. Int. J. Heat Fluid Flow 44, 181196.
Fransen, R., Vial, R., Gicquel, L.Y.M., 2013. Large eddy simulation of rotating ribbed
channel. In: Proc. ASME Turbo Expo 2013. GT2013-95076.
Hanjalic, K., Launder, B.E., 1972. Fully developed asymmetric ow in a plane
channel. J. Fluid Mech. 51, 301335.
Hanjalic, K., Launder, B., 2011. Modelling Turbulence in Engineering and the
Environment: Second-Moment Routes to Closure. Cambridge University Press.
Iacovides, H., 1998. Computation of ow and heat transfer through rotating ribbed
passages. Int. J. Heat Fluid Flow 19, 393400.
Jakirlic, S., Hanjalic, K., Tropea, C., 2002. Modeling rotating and swirling turbulent
ows: a perpetual challenge. AIAA J. 40, 19841996.
Jimnez, J., 2004. Turbulent ows over rough walls. Annu. Rev. Fluid Mech. 36, 173
196.
Johnston, J.P., 1998. Effects of system rotation on turbulence structure: a review
relevant to turbomachinery ows. Int. J. Rotating Mach. 4, 97112.
Johnston, J.P., Halleen, R.M., Lezius, D.K., 1972. Effects of spanwise rotation on the
structure of two-dimensional fully developed turbulent channel ow. J. Fluid
Mech. 56, 533557.
Kristoffersen, R., Andersson, H.I., 1993. Direct simulations of low-Reynolds-number
turbulent ow in a rotating channel. J. Fluid Mech. 256, 163197.
Krogstad, P.., Andersson, H.I., Bakken, O.M., Ashraan, A., 2005. An experimental
and numerical study of channel ow with rough walls. J. Fluid Mech. 530, 327
352.
Lamballais, E., 2014. Direct numerical simulation of a turbulent ow in a rotating
channel with a sudden expansion. J. Fluid Mech. 745, 92131.
Lamballais, E., Lesieur, M., Mtais, O., 1996. Effects of spanwise rotation on the
vorticity stretching in transitional and turbulent channel ow. Int. J. Heat Fluid
Flow 17, 324332.
Launder, B.E., Tselepidakis, D.P., 1994. Application of a new second-moment closure
to turbulent channel ow rotating in orthogonal mode. Int. J. Heat Fluid Flow
15, 210.
Launder, B.E., Tselepidakis, D.P., Younis, B.A., 1987. A second-moment closure study
of rotating channel ow. J. Fluid Mech. 183, 6375.
Leonardi, S., Orlandi, P., Smalley, R.J., Djenidi, L., Antonia, R.A., 2003. Direct
numerical simulations of turbulent channel ow with transverse square bars
on one wall. J. Fluid Mech. 491, 229238.
Leonardi, S., Orlandi, P., Antonia, R.A., 2007. Properties of d- and k-type roughness in
a turbulent channel ow. Phys. Fluids 19, 125101.
Manhart, M., 2004. A zonal grid algorithm for DNS of turbulent boundary layers.
Comput. Fluids 33, 435461.
Moser, R.D., Kim, J., Mansour, N.N., 1999. Direct numerical simulation of turbulent
channel ow up to Res 590. Phys. Fluids 11, 943945.
Nakabayashi, K., Kitoh, O., 1996. Low Reynolds number fully developed twodimensional turbulent channel ow with system rotation. J. Fluid Mech. 315, 1
29.
Nakabayashi, K., Kitoh, O., 2005. Turbulence characteristics of two-dimensional
channel ow with system rotation. J. Fluid Mech. 528, 355?377.
Narasimhamurthy, V.D., Andersson, H.I., 2009. Roughness effects in a rotating
turbulent channel. In: Advances in Turbulence XII. Proceedings in Physics, vol.
132. Springer, pp. 665668.
Narasimhamurthy, V.D., Andersson, H.I., 2011. DNS of turbulent ow in a rotating
rough channel. In: Direct and Large-Eddy Simulation VIII. ERCOFTAC Series, vol.
15. Springer, pp. 413418.
Nilsen, P.J., Andersson, H.I., 1990. Rotational effects on sudden-expansion ows. In:
Rodi, W., Ganic, E. (Eds.), Engineering Turbulence Modelling and Experiments.
Elsevier Science, Amsterdam, pp. 6572.
Perry, A.E., Schoeld, W.H., Joubert, P.N., 1969. Rough wall turbulent boundary
layers. J. Fluid Mech. 37, 383413.
Raisee, M., Naemi, H., Alizadeh, M., Iacovides, H., 2009. Prediction of ow and heat
transfer through stationary and rotating ribbed ducts using non-linear ke
model. Flow Turb. Comb. 82, 121153.
Rothe, P.H., Johnston, J.P., 1979. Free shear layer behavior in rotating systems. ASME
J. Fluids Eng. 101, 117120.
Saha, A.K., Acharya, S., 2005. Flow and heat transfer in an internally ribbed duct
with rotation: an assessment of large eddy simulations and unsteady Reynoldsaveraged NavierStokes simulations. ASME J. Turbomach. 127, 306320.
Tucker, P.G., 2013. Trends in turbomachinery turbulence treatments. Prog. Aero. Sci.
63, 132.
Visscher, J., Andersson, H.I., 2011. Particle image velocimetry measurements of
massively separated turbulent ows with rotation. Phys. Fluids 23, 075108.
Visscher, J., Andersson, H.I., Barri, M., Didelle, H., Viboud, S., Sous, D., Sommeria, J.,
2011. A new set-up for PIV measurements in rotating turbulent duct ows. Flow
Meas. Instrum. 22, 7180.
Viswanathan, A.K., Tafti, D.K., 2006. Detached eddy simulation of ow and heat
transfer in fully developed rotating internal cooling channel with normal ribs.
Int. J. Heat Fluid Flow 27, 351?370.

Please cite this article in press as: Narasimhamurthy, V.D., Andersson, H.I. Turbulence statistics in a rotating ribbed channel. Int. J. Heat Fluid Flow (2014),
http://dx.doi.org/10.1016/j.ijheatuidow.2014.10.008

You might also like