You are on page 1of 17

Reynolds number effects in the outer layer of the turbulent flow in a

channel with rough walls


Ole Martin Bakken, Per-Åge Krogstad, Alireza Ashrafian, and Helge I. Andersson

Citation: Phys. Fluids 17, 065101 (2005); doi: 10.1063/1.1900146


View online: http://dx.doi.org/10.1063/1.1900146
View Table of Contents: http://pof.aip.org/resource/1/PHFLE6/v17/i6
Published by the American Institute of Physics.

Related Articles
Simulation of the flow around an upstream transversely oscillating cylinder and a stationary cylinder in tandem
Phys. Fluids 24, 023603 (2012)
Stability and breakup of confined threads
Phys. Fluids 24, 012102 (2012)
Sedimentation of a charged colloidal sphere in a charged cavity
J. Chem. Phys. 135, 214706 (2011)
Non-axisymmetric impact creates pineapple-shaped cavity
Phys. Fluids 23, 091106 (2011)
Effect of central slotted screen with a high solidity ratio on the secondary resonance phenomenon for liquid
sloshing in a rectangular tank
Phys. Fluids 23, 062106 (2011)

Additional information on Phys. Fluids


Journal Homepage: http://pof.aip.org/
Journal Information: http://pof.aip.org/about/about_the_journal
Top downloads: http://pof.aip.org/features/most_downloaded
Information for Authors: http://pof.aip.org/authors

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
PHYSICS OF FLUIDS 17, 065101 共2005兲

Reynolds number effects in the outer layer of the turbulent flow


in a channel with rough walls
Ole Martin Bakkena兲
Department of Mechanical Equipment, Statoil ASA, N-4035 Stavanger, Norway
Per-Åge Krogstad,b兲 Alireza Ashrafian,c兲 and Helge I. Anderssond兲
Department of Energy and Process Engineering, Norwegian University of Science and Technology,
Kolbjoern Hejes vei 2, N-7491 Trondheim, Norway
共Received 12 April 2004; accepted 21 February 2005; published online 12 May 2005兲
Turbulent channel flow measurements for two different rough surfaces have been compared with a
smooth reference case. A range of Reynolds numbers, Re␶ 苸 具360,6000典, has been investigated using
hot-wire anemometry. Reynolds stresses and third-order moments are shown to be very little
affected by the substantially different wall conditions outside 5k, where k is the characteristic length
scale of the roughness. In this region, a reasonably good collapse with Reynolds number is
demonstrated when scaling with friction velocity is used. This contrasts some of the rough-wall
investigations previously published for boundary layers and channels with only one rough wall. It
is believed that the differences observed are due to the differences in boundary conditions and that
symmetrically roughened channel flows and flows in rough-wall pipes may be better candidates for
the Townsend’s wall similarity hypothesis than asymmetrical flows. © 2005 American Institute of
Physics. 关DOI: 10.1063/1.1900146兴

I. INTRODUCTION layer it is assumed that the flow does not know how the
turbulence was generated and, suitably scaled, will collapse
Due to its engineering significance, flows over rough
for all types of surfaces. Hence Townsend’s3 hypothesis im-
surfaces have received considerable attention over the years.
plies that the flow over smooth and rough surfaces are simi-
Experiments have traditionally formed the basis of this re-
lar outside the roughness sublayer.
search, since a sufficiently detailed description of the rough-
ness geometry to allow exact prediction of the flow within In an experimental investigation of a turbulent boundary
the roughness layer has until now been impossible. The ex- layer over a mesh roughened surface, Krogstad et al.5
periments have therefore formed the basis for simple predic- claimed that if the boundary layer is constantly fed by infor-
tion models with the primary aim to reproduce the effects of mation from the wall region which is uniquely connected to
surface roughness on the mean velocity profiles above the the surface geometry, this information will eventually affect
elements. Since considerable information was available for the outer layer to such a degree that the flow will be modified
flows over surfaces covered by sieved sand quite early,1 it compared to the flow over a smooth surface. Differences
became common in the CFD community to link the flows were found in the strength of the outer layer wake, in the
over arbitrary roughness geometries to an equivalent flow spectra and in the contributions to the shear stresses revealed
over a wall covered by sand roughness 共originally suggested by quadrant analysis. These results were followed up by an
by Schlichting2兲, thereby ignoring any possible dependency investigation where the turbulent stresses in the outer layer
in the turbulent flow structure on the specific surface geom- of the boundary layer were shown to depend strongly on the
etry. Implicit in this treatment of surface roughness is the surface geometry.6 Hence, both investigations challenge
assumption that the flow, only a short distance away from the Townsend’s hypothesis. Unfortunately, boundary layer inves-
roughness elements, becomes universal and looses the char- tigations over rough surfaces are considerably more compli-
acteristic footprints of the surface geometry that must exist cated to perform than for smooth walls since the exact loca-
for the flow between the individual roughness elements. This tion of the virtual wall is unknown and the wall shear stress
school of thoughts forms the basis for the “wall similarity is very difficult to estimate. Therefore, rough-wall boundary
hypothesis,” originally proposed by Townsend,3 and later ex- layer data are generally less reliable than those taken over a
tended by Perry and Abell.4 If k is a characteristic length smooth surface.
scale for the rough surface, a sublayer is assumed to exist up In order to reduce the uncertainty in the wall stress, the
to about 5k, where the flow is uniquely dependent on the experiments have therefore been repeated here in a fully de-
actual geometry of the roughness elements. Outside this veloped channel flow. Then the friction velocity, u␶ = 冑␶w / ␳
共␶w is the wall shear stress and ␳ is the density兲, can be
a兲
b兲
Electronic mail: olbak@statoil.com deduced directly from the mean pressure gradient dP / dx,
Author to whom correspondence should be addressed. Electronic mail: i.e., u␶2 = h / ␳共dP / dx兲 共h is the channel half-height兲, or from
per.a.krogstad@ntnu.no
c兲
Electronic mail: alireza.ashrafian@ntnu.no the shear stress distribution in the outer region, which has an
d兲
Electronic mail: helge.i.andersson@ntnu.no exact analytical solution. However, it is not obvious that the

1070-6631/2005/17共6兲/065101/16/$22.50 17, 065101-1 © 2005 American Institute of Physics

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-2 Bakken et al. Phys. Fluids 17, 065101 共2005兲

outer layer of the channel flow will be sensitive to surface


roughness in the same way as a boundary layer. The most
1
U+ = ln
␬ k
y
+C 冉冊 共2兲
obvious reason for this is the differences in the outer bound-
ary conditions. The boundary layer is allowed to expand, using k. The constants ␬ ⯝ 0.41 and B ⯝ 5.5 were determined
which means that the rate of entrainment into the outer layer experimentally in the smooth-wall case. The roughness func-
produces a much more dominant wake region than is found tion ⌬U+ causes a downward shift in the log law due to the
for channel flows. The growth rate of this region is dictated change in boundary conditions. The constant C is dependent
by the wall shear stress and therefore the amount of surface on the specific type of roughness; for sand-grain roughness it
roughness, as may be seen from the von Karman momentum is about 8.5. Combining Eqs. 共1兲 and 共2兲 gives the following
dependency of ⌬U+ on k+ = ku␶ / ␯, the roughness Reynolds
equation. Also the ratio of the boundary layer thickness to
number:
the relevant roughness length scale increases with stream-
wise distance. This is not the case for a channel flow where 1
⌬U+ = ln k+ + B − C. 共3兲
all length scales are independent of streamwise position. ␬
Therefore the channel flow reaches an equilibrium condition
Surface geometries that produce flows that follow this shift
which is never achieved in a boundary layer.
are known as k-type roughnesses. The equivalent sand-grain
In addition to the improved accuracy in determining ␶w,
roughness Reynolds number, ks+, is calculated using B − C
another reason for choosing to study roughness effects in a
= −3.0. For ks+ ⬍ 70, the flow is transitionally rough and for
channel flow is that it is much more suited for direct numeri-
ks+ ⬎ 70 the roughness elements extend into the logarithmic
cal simulations 共DNS兲 than a boundary layer due to its peri-
region, so the smooth-wall viscous sublayer is assumed to
odicity. In recent years some rough-wall DNS have been have been completely destroyed and the surface is consid-
published.7–9 Due to grid resolution problems, most of these ered to be fully rough.
simulations have had roughness elements on only one sur- In this paper the following similarity definitions will be
face and the evolving flow is therefore asymmetric. Some used. Surface similarity is the collapse of profiles from dif-
support for the outer layer effects observed in the boundary ferent surfaces at the same Reynolds number. Reynolds num-
layer by Krogstad et al.5 have been provided in these simu- ber similarity is the collapse of profiles from different Rey-
lations 共e.g., by Leonardi et al.10兲. However, it is unclear nolds numbers and complete similarity will be used for the
whether these observations are due to the rather large rough- simultaneous surface and Reynolds number similarity.
ness elements used 共of the order of 20% of the channel half-
height兲 or the asymmetry of the flow. Therefore, in this in-
vestigation much smaller elements were used and the II. SCALING CONSIDERATIONS
roughness geometry was the same on both surfaces. Two Scaling in the outer layer is dependent on the global
different roughness geometries were investigated, a rod properties of the flow. In a channel flow, the only obvious
roughness and a mesh roughness 共see Sec. III兲. In a compan- outer length scale is the channel half-height h. A proper ve-
ion DNS to the rod roughness experiment, the flow was also locity scale is more difficult to define and various proposi-
investigated numerically. These results have been published tions will be considered here. The friction velocity u␶ is the
by Ashrafian et al.11 and will be included here for most commonly used quantity. Zagarola and Smits13 have
comparison. argued that the mass flux deficit, UCl − Ub, is a true outer
The mean velocity distribution U共y兲 can be divided into velocity scale, where UCl is the centerline velocity and Ub is
an outer and inner region. For the inner layer, the appropriate the bulk velocity. It should be noted that both u␶ and UCl
velocity scale is u␶. The proper length scale is the viscous − Ub are frame invariant.14 Results from their measurements
length scale ␯ / u␶ 共␯ is the kinematic viscosity兲 in the in the “Superpipe” facility at Princeton University show that
smooth-wall case and/or the roughness length scale k in the UCl − Ub = ⌫ ⫻ u␶, where ⌫ is constant for high and varies for
rough-wall case.5 For an arbitrary rough surface, k will not low Reynolds number. In the low Reynolds number range,
be well defined by only one length scale. However, for the scaling with UCl − Ub is shown to improve the collapse of the
particular surface geometries in this study, k is taken to be velocity defect and the streamwise stress compared to using
the thickness of the rods or the plate.12 This implies that the u␶. Recent results from the same facility14 show that for the
inner region is limited by y = 5k or y / h 艋 0.15– 0.17 and what second and higher order moments of the streamwise velocity,
is outside this limit is taken to be the outer region. In the UCl − Ub is a better velocity scale for y / R ⬎ 0.35 than u␶, but
fully rough case, the mean velocity profile in the logarithmic along two distinct distributions, one when Red ⬍ 6 ⫻ 104, the
overlap region may be written as other when Red ⬎ 7 ⫻ 104 共Red is based on the pipe
diameter兲.
Figure 1 shows UCl − Ub vs u␶ from the present study for
all three surfaces for the range 6.0⫻ 103 艋 Re艋 7.6
1 ⫻ 104 共Re⬅ Ubh / ␯兲, which overlaps that of Zagarola and
U+ = ln y + + B − ⌬U+ 共1兲
␬ Smits13 in the lower end. The gradient of the linear fit 共solid
line兲 is ⌫ = 2.64 with 4.3% rms error and a correlation coef-
ficient of Rc = 0.9993. It is apparent that within experimental
using ␯ / u␶ as length scale or uncertainty, no Reynolds number dependence may be ob-

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-3 Reynolds number effects in the outer layer Phys. Fluids 17, 065101 共2005兲

FIG. 1. UCl − Ub vs u␶ . 䊊 Smooth; 䊐 rods; 䉭 mesh,


solid line, linear fit.

served. The implication of the data, all falling onto a straight layer momentum thickness.18 For internal flows, Wosnik et
line, is that there is no advantage in scaling with UCl − Ub, al.19 have proposed to use the bulk velocity Ub. Morrison et
compared to using u␶, for the present study. A possible dif- al.14 remark that there is a problem with frame dependency
ference between boundary layers and channel flow is sug- choosing a velocity scale involving any mean velocity. Their
gested in the Introduction. The apparent inconsistency with results also suggest that the rate of energy transfer from the
the Superpipe results suggests that there may be differences outer part of the boundary layer to the inner region, which is
also between pipe and channel flows, as suggested by a possible justification for the mixed scaling, scales with u3␶
Nieuwstadt and Bradshaw.15 rather than u2␶ UCl as proposed by DeGraaf and Eaton.16 In
In relation to the streamwise velocity component for a addition, DeGraaf and Eaton16 do not explain why the geo-
boundary layer, DeGraaf and Eaton16 have suggested an em- metrical mean of u␶ and UCl is to be used. This implies that
pirical mixed scaling 共u␶UCl兲共1/2兲. This improves the collapse even though the results are convincing, the physical basis for
of u2 data for y / ␦ ⬎ 0.15 in the range 600⬍ Re␶ ⬍ 10 000, scaling with 共u␶UCl兲共1/2兲 is somewhat weakened.
where ␦ is the boundary layer thickness. Additionally, a scal- Figure 2 shows UCl vs u␶ for the present study. The
ing with UCl has been proposed17,18 and it has been shown correlation coefficient is better than Rc = 0.999 for all the
that very good collapse could be obtained in the range 600 three linear regressions. Thus, the linear relationship between
⬍ Re␶ ⬍ 6000 for u2 / UCl2
vs y / ␪, where ␪ is the boundary UCl and u␶ is clear for all three surfaces. However, the gra-

FIG. 2. UCl vs u␶ . 䊊 Smooth; 䊐 rods; 䉭 mesh.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-4 Bakken et al. Phys. Fluids 17, 065101 共2005兲

TABLE I. Parameters for the three surfaces. The symbols used for the plots
have similar outer patterns for similar surfaces and similar inner patterns for
similar Re␶. For the mesh roughness k / h = 0.03 and for the rod roughness
k / h = 0.034.

Surface type Re Re␶ k+ Symbol

Smooth 6 300 360 䊊


Smooth 12 800 670 쎲
Smooth 33 000 1550 丢
FIG. 3. Geometry of the rod experiment setup.
Smooth 76 000 3300 䉺
Mesh 6 200 450 15 䉭
Mesh 11 600 950 32 䉱
Mesh 31 000 2500 83 䊗 performed at Re␶ ⬅ u␶h / ␯ = 400 共for details, see Ref. 11兲. Fig-
Mesh 68 000 5600 187 䉭
⫹ ure 3 shows the geometry of the rod roughness experiment.
Rod 6 000 600 20 䊏 In the case of the second rough surface, 1.5 mm thick
Rod 11 500 1200 40  perforated plates 共mesh surface兲 were used. This corresponds
Rod 30 000 3200 107 ⱓ to k / h = 0.03. The perforation consisted of 10⫻ 10 mm2
Rod 56 000 6000 200  square holes arrayed in a squared pattern with a center spac-
Smooth, DNSa 5 400 395 … ing of 12 mm giving an openness ratio of about 69%. For the
Smooth, DNSa 10 700 590 — rough surfaces, all the measurements were taken above the
Rod, DNSb 4 100 400 14 --- crest of the roughness elements to avoid the regions of very
a
Moser et al. 共Ref. 26兲. high turbulence intensity and reversed flow over the cavities.
b
Ashrafian et al. 共Ref. 11兲. This geometry is shown in Fig. 4.
The hot wires used were in-house made single- and
X-wire probes, partly etched from 2.5 ␮m Wollaston
dients are very different. A qualitatively similar trend is seen Pt-10%Rh wire with an active length lw of about 0.5 mm.
for Ub vs u␶ 共not shown兲. This is related to the very different The length-to-diameter ratio was close to 200 and the sepa-
C f = 2共u␶ / Ub兲2 for the various surfaces 共Fig. 7兲. Conse- ration between the wires was also ⬇0.5 mm. The included
quently, for each of the three separate surfaces, the Reynolds angle of the two wires was nominally 100°. The apex angle
number similarity will be equally good using either of the has been made larger than 90°, as suggested by the works of
above mentioned scalings as using u␶. However, Fig. 2 indi- Perry et al.22 and Krogstad et al.,5 because the rough surfaces
cates that the surface similarity properties of the various scal- need a larger apex angle than the smooth surface to capture
ing quantities suggested will be quite different. This will be the increased turbulence intensity near the wall. This is sup-
discussed further in the subsequent sections. ported by probability density functions of the angular distri-
bution 共not shown兲. It may be argued 共see, e.g., Tagawa et
III. EXPERIMENTAL DETAILS al.23兲 that the increased angle reduces the sensitivity to the v
fluctuations and therefore increases the measurement uncer-
Measurements were performed in a closed return wind tainty, but comparisons with other smooth-wall channel flow
tunnel with a working section consisting of a rectangular data using an apex angle of 90° have only shown differences
channel. The test section is 5 m long, with an inlet area of within experimental uncertainty.
1.35⫻ 0.10 m2. Measurements of velocity profiles at several For the low-velocity cases 共Ub ⬍ 5 m / s兲, a low-velocity
spanwise locations were performed to verify that the aspect
calibration jet was applied for the calibration of the hot
ratio 共⬇14兲 was sufficiently large for the mean flow to be
wires. The calibration rig was positioned on top of the chan-
two dimensional. The bulk Reynolds number 共Re⬅ Ubh / ␯兲 nel and adjusted so that the jet flow was coaxial to the chan-
can be varied by a factor of 10 共see Table I兲. The measure- nel flow. A high precision computer controlled traversing
ment station was about 5 cm upstream of the channel exit. system was used for calibration and measurements. This sys-
The flow along the ceiling and floor was tripped at the tem also enabled movement of the probes between the cali-
inlet using a 3 mm diameter circular rod followed by a 12 cm bration rig and the channel flow, so that the probes were
strip of No. 40 grit sandpaper, both spanning the width of the
section. In the case of rough surfaces, both the ceiling and
floor were covered. In the first rough case, square rods mea-
suring 1.7⫻ 1.7 mm2 were used, giving k / h = 0.034. The rods
spanned the entire width of the section. The center-to-center
rod spacing was eight times the width. According to experi-
ments by Furuya et al.20 in a boundary layer, this roughness
spacing creates the largest shift in the mean velocity profile
and hence the largest effective hydrodynamic roughness. Re-
cently, this has been verified for channel flows with rough-
ness on one side using DNS 共Ref. 7兲 and LES.21 For com-
parison with the rod roughness experiment, a DNS has been FIG. 4. Geometry of the mesh experiment setup.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-5 Reynolds number effects in the outer layer Phys. Fluids 17, 065101 共2005兲

never removed or disconnected between calibration and to within 1%. The latter method was therefore believed to be
measurement. accurate also for the lower Re␶ data. Using this method in the
The X wires were calibrated using a velocity dependent smooth-wall case, mean velocity profiles compared with re-
effective angle method. The method has the same benefits as sults from DNS 共Ref. 26兲 to within 2%. For single- and
a full velocity vs yaw-angle calibration routine, but requires UW-wire smooth-wall measurements, the confirmation of u␶
considerably less effort. The velocity calibration involves fit- was therefore done by inspecting the mean velocity profiles.
ting a least-squares fourth order polynomial to the velocity Because of the shift in the mean velocity profile in the rough-
vs voltage data obtained when the flow is directed along the wall case, another method had to be used in the low Re␶
probe axis. For angular response a cosine law is assumed rough cases for single- and UW-wire measurements. Method
using an “effective” rather than the geometrical angle. The 共iii兲: Results from the smooth case showed very good col-
2
effective angle has been found to be strongly dependent on lapse of the u+ profiles for y / h ⬎ 0.2 for all probes. This
the streamwise velocity in the low-velocity range. A power result was therefore used to estimate u␶ for the rough case
law is thus fitted to the effective angle vs velocity data de- low Re␶ single- and UW-wire measurements. After determin-
termined for the full range of velocities in the experiment. ing u␶ from the shear stress profile measured by the UV
2
During data reduction the velocity dependent effective angle probe, the corresponding u+ profile was used to find u␶ for
is determined on a sample-by-sample basis and the system of the single- and UW-wire measurements.
equations must be solved iteratively. Further details of this
method may be found in Bakken and Krogstad.24
The hot wires were operated using in-house made con- IV. RESULTS AND DISCUSSION
stant temperature anemometers at an overheat ratio of 1.5. A. Streamwise mean velocity
The filter frequencies of the low-pass filters were adjusted to
match the highest Kolmogorov frequency in the flow, after a
preliminary spectral survey had been made. The frequency Mean velocity profiles are shown in inner coordinates in
response of the hot wires was always better than the cutoff Fig. 5. For the smooth surface there is good agreement with
frequency. Sampling to a PC was performed using a National DNS of Moser et al.26 Approaching the wall, the usual over-
Instruments 16 bit A/D converter at sampling frequencies prediction of the hot wire caused by convection to the wall
twice the cutoff frequency. can be observed. For the rough surfaces, the error in origin
In a study of the measurement uncertainty based on an was determined through a fitting procedure of the mean ve-
estimate of errors traceable back to the calibration equipment locity profile to the log law.27 Towards the wall, the low Re␶
used and a repeatability study of the measured quantities in profiles have the same downward dip as the DNS of Ashra-
the outer layer of the channel, the uncertainties for fian et al.11 The rough-wall profiles also exhibit the expected
U , u2 , v2 , w2, and uv were estimated to be ±1.5% , downward shift ⌬U+, which is found to follow the charac-
± 4 % , ± 10% , ± 8%, and ±8%, respectively. Uncertainties teristic logarithmic dependency with respect to k+. This is
near the wall are somewhat higher. For the third order mo- consistent with a k-type behavior 关Fig. 6; Eq. 共3兲兴. The data
ments presented we have no good estimates, but the uncer- for the lowest Re␶ for the mesh surface sit slightly lower than
tainties will be increased compared to the second order mo- the logarithmic line. Calculations of the equivalent sand
ments. It was found that the inaccuracy increased rapidly roughness ks+ show that this case is in the transitional regime.
with the power of v, so that v3 is expected to have the high- The same is the case for the rod roughness DNS data.11 In
est uncertainty. the fully rough regime, the ratio ks / k is about 7.8 and 3.3 for
The temperatures in the calibration facility and in the the rod and mesh roughness, respectively, indicating a much
wind tunnel were constantly monitored with thermocouples. stronger roughness effect from the rod roughness than from
The effect of temperature drift was corrected for using the the mesh surface.
method of Bearman.25 The variation of the friction factor, C f = 2共u␶ / Ub兲2, with
For the determination of u␶, three different methods were Re is shown in Fig. 7. Comparison is made in the smooth
used. Method 共i兲: Measurement of the static pressure gradi- case with the empirical correlation of Dean,28 C f
−0.025
ent in the streamwise direction dP / dx. For measurements of = 0.073 Re2h . The rough-wall data show a constant C f in
dP / dx, the test section is fitted with pressure taps along the the region of fully rough flow and displays a downward dip
centerline of the sidewalls to eliminate disturbances from the in the transitionally rough regime as expected 共see, e.g.,
roughness elements. Additional taps are fitted along the cen- Pope29兲.
terline of the floor as a verification. This method was used Scaling with u␶ in the outer layer results in the velocity
for the higher Re␶ investigated. For the lower Re␶ the pres- defect law
sure drop was too low to obtain reliable estimates of dP / dx.
Method 共ii兲: Calculation of u␶ from a least-squares fit of the
Reynolds shear stress, −u+v+, to the estimated outer layer
UCl − U 1
u␶
= ln
␬ h
y
+ D, 冉冊 共4兲

turbulent shear stress. In a fully developed channel flow, the where D is a constant. The smooth-wall profiles for Re␶
total shear stress, ␶total
+
= 1 − 共y + / h+兲 = 共dU+ / dy +兲 − u+v+ is a ⯝ 360 and Re␶ ⯝ 670 共Fig. 8兲 are seen to follow the corre-
+ +
straight line; dU / dy is the mean velocity gradient, and su- sponding lines from DNS 共Ref. 26兲 at Re␶ = 395 and 590,
perscript + denotes normalization with viscous wall units u␶ respectively, down to y / h = 0.01. Both rough surfaces col-
or ␯ / u␶. For the high Re␶ data, methods 共i兲 and 共ii兲 compared lapse with the smooth wall results all the way down to the

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-6 Bakken et al. Phys. Fluids 17, 065101 共2005兲

FIG. 5. Mean velocity U+ in inner coordinates. For the


smooth surface, only data for Re␶ ⯝ 360 and ⯝3300 are
included. Symbols in Table I.

top of the roughness elements. This is a first indication that rough surfaces, only the data from the lowest Reynolds num-
the outer layer flow is unaffected by the surface roughness. bers capture the near-wall peak. These are the data of the
mesh surface at k+ ⯝ 15 共ks+ ⯝ 30, Re␶ = 450兲 and the DNS
B. Reynolds stresses data11 of the rod roughness at k+ ⯝ 14 共ks+ ⯝ 60, Re␶ = 400兲.
1. Streamwise normal stress Compared to the smooth-wall data, both of these have an
2 attenuated peak value down to about 4.2–4.4. The location of
The streamwise stress, u+ was measured with single
the peak is shifted to y peak +
⯝ 30, which corresponds roughly
wires. The spatial resolution in the wall region is adequate,
as indicated by the close agreement with the DNS. However, to y peak ⯝ y peak,smooth + k . The profiles of the next higher Re␶
+ + +

for the highest smooth-wall Reynolds number 共Re␶ = 3300, both reside in the fully rough regime 共ks,rod +
⯝ 130, k+ ⯝ 40,
lw+ ⬇ 33兲, spatial resolution starts to affect the data,30 which Re␶ = 600 and ks,mesh ⯝ 100, k ⯝ 32, Re␶ = 950兲. For these
+ +

are slightly attenuated near the peak 关Fig. 9共a兲兴. The average profiles the near-wall peaks are even more attenuated down
peak value is 7.27± 3%, which is similar to what was found to about 3–3.5 and have moved out to y peak +
⯝ y peak,smooth
+

by Mochizuki and Nieuwstadt30 in a large survey of both + k ⯝ 50– 60; y / h ⯝ 0.06– 0.09.
+
2
experimental and numerical data 共7.29± 6.5% 兲. This is In the inner region, the peak value of u+ is seen to
somewhat lower than the DNS of Moser et al.26 For the gradually decrease with increasing k+ 共or ks+兲 from the high

FIG. 6. Shift in mean velocity ⌬U vs k+. The lines are


fitted to the fully rough data using Eq. 共3兲. 䊐 Rods,
experiments; 䊏 rods, DNS; 䉭 mesh.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-7 Reynolds number effects in the outer layer Phys. Fluids 17, 065101 共2005兲

FIG. 7. Friction factor C f . 䊐 Rods; 䉭 mesh; 䊊 smooth;


solid line, Dean’s correlation: C f = 0.073 Re−0.025
2h .

level of the smooth-wall data to a much lower level for the the rough-wall profiles discussed above. This is caused by a
fully rough data. The peak in the smooth-wall case is mainly small region of local separation at the top of the roughness
due to viscous effects. In the rough-wall cases, the position elements and is obviously a speciality of these kinds of
of the peak is moved outwards from the smooth wall roughness geometries.
+
y peak,smooth ⯝ 15 by a value corresponding approximately to 2
+ In the outer part of Fig. 9共a兲, the magnitude of u+ over
k . In the wall region, there is a transition from a strong
the smooth wall increases with increasing Re␶. For all the
effect of viscous forces for low k+ to a dominance of pressure
forces caused by the drag on the roughness elements when k+ rough-wall profiles with Re␶ ⬎ 1000, the inner region peak is
increases.31 At low k+, the y + value of the top of the rough- absent, and the magnitude of the peak in the outer region is
ness elements is lower than the outer limit of the buffer layer. increasing with increasing Re␶ and k+, which is consistent
However, for increasing k+, the pressure forces are increas- with the smooth-wall behavior. However, the outer region
ingly important and in the fully rough regime, ks+ ⬎ 70, vis- peak is always inside y = 5k. The difference in peak values
cous forces are expected to be negligible. for the two rough surfaces at approximately similar Re␶ or k+
Very close to the wall, there is a small local peak for all is therefore as expected. In contrast to the low Re␶ profiles

+
FIG. 8. Velocity defect UCl − U+ vs y / h. Symbols in
Table I.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-8 Bakken et al. Phys. Fluids 17, 065101 共2005兲

2
FIG. 9. Streamwise normal stress u+ vs y + and y / h.
Symbols in Table I.

over the rough surfaces, all the high Re␶ profiles have a vortices.34 From the present data it seems plausible that there
downward dip towards the wall, indicating that the near-wall is a link between the gradually decreasing influence of vis-
structure over rough surfaces is very different from those cous forces and the increasing isotropization of the large-
found over smooth walls. scale vortices in the wall region. At the same time, these
It has been suggested that for smooth surfaces, large structures are moved outwards a distance corresponding ap-
Reynolds stresses and turbulence production in the viscous proximately to the roughness height. Even though the near-
wall region are associated with highly organized streamwise wall production mechanisms for the rough surface may have
vortices.32 In the present study the integral length scale in the been significantly modified compared to the smooth-wall
streamwise direction was found to be shorter than for the 2
flow, the outer layer u+ level appears to be little affected.
smooth surface near the wall for both types of roughnesses,33 2
indicating a break-up of the near-wall streamwise vortices Figure 9共b兲 shows u+ vs y / h. For similar Re␶ the data
over the rough walls. This is supported by means of the ␭2 from the different surfaces collapse outside y / h
vortex criterion in the companion DNS study.11 At the same ⯝ 0.12– 0.15. There is, however, a distinct difference be-
time, a quadrant analysis shows that the ejection/sweep pro- tween the lower 共400⬍ Re␶ ⬍ 1500兲 and the higher 共2500
cesses are very similar for the three surfaces.12 This mecha- ⬍ Re␶ ⬍ 6000兲 Re␶ profiles for 0.15⬍ y / h ⬍ 0.5, which is
nism is known to be important in the turbulence production outside of the log region. For y / h ⬎ 0.5, there are no system-
and in the wall region it is connected to the large-scale atic differences between the profiles.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-9 Reynolds number effects in the outer layer Phys. Fluids 17, 065101 共2005兲

2
FIG. 10. Wall-normal normal stress v+ vs y + and y / h.
Symbols in Table I.

The Re␶-dependent high and low levels of u2 when scal- flow14,35 suggest that a good collapse when scaling with u␶ is
ing with u␶, are consistent with observations in a smooth- obtained only in the region y / h ⬎ 0.4. Our results thus cor-
wall boundary layer.16 This has also been observed in a roborate earlier findings for smooth-wall internal flows. The
smooth-wall pipe flow when scaling with UCl − Ub.14 The re- present results extend these findings to be valid also for
sults of Fig. 1 show that the same effect would be obtained if rough-wall channel flows.
UCl − Ub was chosen for scaling. Since surface similarity is
well demonstrated when scaling with either u␶ or UCl − Ub,
the results of Sec. II and Fig. 2, in particular, imply that there
2. Wall-normal normal stress
is no possibility of surface similarity using either of the ve-
locity scales UCl, 共u␶UCl兲共1/2兲, or Ub. However, the Reynolds The fluctuating velocity component in the wall-normal
number similarity of each separate surface will be just as direction v is important in the diffusion of turbulent energy
good as in Fig. 9共b兲, but it will not be improved. across the channel. Differences in v between the different
The lack of collapse of the u2 profiles scaling with u␶ has surfaces may therefore indicate different transport mecha-
been suggested by Townsend3 to be due to the “inactive” nisms. The data points closest to the wall have been dis-
motion, which is described as a large-scale swirling motion carded, since they were obviously affected by measurement
in planes parallel to the wall. It has no vertical motion and errors. Following the discussion in the previous sections, the
2 2
therefore only contributes to u+ and w+ . Results from pipe scaling velocity used for v2 will be u␶2.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-10 Bakken et al. Phys. Fluids 17, 065101 共2005兲

2
FIG. 11. Spanwise normal stress w+ vs y / h. Symbols
in Table I.

2
The peak value of v+ in smooth-wall flows is found in effects than to the boundary condition. This is a strong indi-
the logarithmic region and is therefore little affected by vis- cation of similar transport mechanisms in the smooth- and
cous forces. There seems to be a general agreement about a rough-wall cases. 共It should, however, be pointed out that
Re␶ dependency of the peak value for Re␶ below except for the lowest Reynolds number smooth-wall data, all
1000.16,26,36,37 At higher Re␶, this dependency is at least less profiles irrespective of Re and surface condition, lie in a
important if not nonexistent.16,38 An implication of these re- rather narrow band with a spread of less than ±10%. This is
sults is that the “active” motion is Re␶ dependent at low about the same order of magnitude as the expected uncer-
Re␶.37 However, the amount of Re␶ dependency appears not tainty in v2 when using hot-wire anemometry, so consider-
to be settled. In the outer region of a smooth-wall boundary able caution should be shown in making too definitive con-
layer, the collapse has been shown to be very good in outer clusions about the effects of surface roughness on v2 in the
coordinates for Re␶ 艌 1000.16,39 In contrast to these results outer layer.兲
are those of Krogstad et al.5 in a boundary layer with circular
2
rods and mesh roughness. They found that v+ was generally
much higher over both types of roughnesses than over the
3. Spanwise normal stress
smooth surface. However, other results obtained for internal
2
flows suggest that v+ is attenuated for the entire cross sec- Measurements of the spanwise stress component w are
tion by the surface roughness.40,41 scarce and the uncertainties associated with near-wall data of
2
Results for v+ are shown in Figs. 10共a兲 and 10共b兲 in w2 from hot wires are larger than for v. Consequently, data of
inner and outer coordinates, respectively. The attenuation of w2 will be shown in outer variables only. The velocity scale
the peak value is considerable for Re␶ 艋 450 关Fig. 10共a兲兴. used is still u2␶ .
This is particularly distinct for the smooth 共Re␶ = 360兲 and For low Reynolds numbers 共Re␶ 艋 590兲, DNS for
rod-roughened 共DNS,11 Re␶ = 400兲 surfaces. The peak for the smooth-wall channel flow26 show a strong Re␶ dependency
2
mesh surface at Re␶ = 450 is much less influenced; there is a of the peak value of w+ . The Re␶ dependency primarily ap-
fairly small difference between the Re␶ ⯝ 450 and Re␶ plies to the near-wall region and the level outside y / h ⯝ 0.2
2
⯝ 950 profiles. The peak values of v+ for both the smooth is fairly constant for Re␶ = 395 and 590. This is consistent
and the mesh surfaces increase to a constant value for Re␶ with results from boundary layers for 1100⬍ Re␪ ⬍ 4400.38
2
⬎ 2500. For the rod-roughened surface the peak value is con- For rough-wall internal flows, it has been found that w+ was
stant for Re␶ ⬎ 630. In the outer region 关Fig. 10共b兲兴, there is attenuated for the entire cross section.40,41
good agreement with the smooth-wall Re␶ = 590 DNS data26
2
Figure 11 shows the distributions of w+ . In the outer
for all surfaces outside y / h ⯝ 0.1– 0.15 for Re␶ ⯝ 600– 1200. layer, there seems to be fairly good agreement between the
For the smooth and mesh surfaces, there is a general increase different surfaces and Reynolds numbers, except the high
in the level from the low Re␶ data to the data for Re␶ Re␶ data for the rod roughness, which are somewhat higher
⬎ 2500 for the entire cross section. However, the trend for towards the center of the channel. This is believed to be an
the rod roughness is different from the other flows, as the artifact, since it is hard to accept that the largest differences
peak level appears to be independent of Re␶. between the various surfaces reside in the region outside
2
It appears that the peak value and the level of v+ in the y / h ⯝ 0.4. 共As for the v2 data, the difference in level is
outer layer is much more connected to Reynolds number within the expected measurement uncertainty for w2.兲

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-11 Reynolds number effects in the outer layer Phys. Fluids 17, 065101 共2005兲

FIG. 12. Shear stress −u+v+ vs y + and y / h. Symbols in


Table I. 共a兲 The two lines added for Re␶ ⯝ 3000 are
calculated from Eq. 共5兲, solid line for smooth wall and
dash-dotted line for rods.

4. Shear stress gion the viscous term 关last term of Eq. 共5兲兴 becomes negli-
In the smooth-wall case, −uv may be calculated from the gible compared to U+V+. The major contribution to U+V+
streamwise mean momentum equation and the measured comes from the acceleration of the fluid just ahead of and
mean velocity profile directly. The nondimensional shear above the roughness element. A relatively long cavity will
stress can then be expressed as therefore diminish the plane-averaged effect. A consequence
of this for the present investigation is that the shear stress
y + ⳵ U+
− u +v + = 1 − − . 共5兲 profile measured above the rough surface will reach a peak
h+ ⳵ y + level that is reduced compared to the smooth-wall
This equation is also valid for a rough-wall case with random distribution.
roughness. There is an additional contribution in the present The measured shear stresses are shown in Figs. 12共a兲
wavy rough cases from the U+V+ correlation,42 which is dif- and 12共b兲 in inner and outer variables, respectively. To avoid
ficult to quantify experimentally using hot-wire anemometry, crowding in Fig. 12共a兲, only a few data sets are included to
since V is very small relative to U. From the rod roughness show the trend. The two lines added for the smooth and
DNS 共Ref. 11兲 共Re␶ = 400兲, the plane averaged contribution rod-roughened surfaces for Re␶ ⯝ 3000 in Fig. 12共a兲 are
of UV has been estimated to be 0.08 uv at the wall distance −u+v+ calculated from Eq. 共5兲 共without the U+V+ term兲. The
corresponding to the crest of the rods. Above the roughness almost complete collapse of the two calculated profiles is a
element the contribution is considerably higher. In this re- confirmation of the similarity of the gradients in the mean

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-12 Bakken et al. Phys. Fluids 17, 065101 共2005兲

3
FIG. 13. Third order streamwise moment u+ vs y + and
y / h. Symbols in Table I.

velocity profiles. For the outer region, the calculated and increasing with increasing Re␶. This is consistent with Eq.
measured profiles are seen to follow each other closely, 共5兲, as well as with data from both smooth-wall boundary
which is encouraging for the accuracy of the estimates of u␶. layers16 and channel flow.43
This is also seen in Fig. 12共b兲 where all profiles follow the
straight line 1 − y + / h+ in the outer region. For the smooth- C. Third order moments
wall profiles, the effect of viscosity is seen as a drop towards
Figures 13–16 show the third order moments
the wall. In the measurements only the lower Reynolds num- +3 2 2 3 3
ber data exhibit this feature 关Fig. 12共a兲兴, since for the high u , u+ v+ , u+v+ , and v+ . Except for in the u+ vs y + plot
Re␶ spatial resolution limitations prevented measurements 关Fig. 13共a兲兴, some of the data sets are omitted in this section
3
down to the viscous region. For the rough wall, all the pro- to avoid crowding in the figures. Note that u+ is measured
files, independent of Re␶, have a dip towards the wall. Fol- with single wires, so the near-wall resolution and accuracy of
lowing the discussion above, this is due to the effect of the this moment is expected to be considerably better than for
U+V+ term. The peak values at the same Re␶ are only slightly the other moments. For the mixed moments, no comparison
higher for the smooth-wall profiles than for the rough. There is made with smooth-wall DNS, since these are not part of
is, however, a trend for all the surfaces that the peak value is the database of Moser et al.26

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-13 Reynolds number effects in the outer layer Phys. Fluids 17, 065101 共2005兲

2
FIG. 14. Third order mixed moment u+ v+. Symbols in
Table I.

Since the ejection and sweep events are significantly vortices. Thus, in the transitional rough cases there are indi-
stronger than the outward and inward interaction terms over cations of a similar type of organization as over the smooth
both the smooth and rough walls, the sign of the streamwise wall, suggesting that viscosity is still having an important
turbulent transport of u2 关Figs. 13共a兲 and 13共b兲兴 can be influence in these flows. The strength of these structures does
loosely related to ejection and sweep events.12 Very close to however appear to have been reduced by the interaction of
3
the wall, u+ on the smooth surface is positive and therefore the roughness elements as conjectured by Krogstad et al.5
characterized by a domination of sweep type motions. Out- The fully rough profiles become positive further from
3
side y + ⯝ 15, there is a dominance of negative u+ events the wall. This is consistent with the quadrant analysis, which
throughout most of the layer, indicating strong ejections of shows that sweep type events are much more important to-
low speed fluid away from the wall. This is expected, since wards the wall over the fully rough surfaces than over the
there is a general energy transport away from the location of smooth surface.12 This is believed to be caused by a reduced
maximum turbulence production. Both the transitional damping of the wall-normal part of the sweeping motion
roughness data sets 关mesh roughness Re␶ = 450 and rod over the rough surfaces, due to the open nature of the sur-
roughness DNS 共Ref. 11兲 Re␶ = 400兴 follow each other quite face. The zero crossing point at y / h ⯝ 0.15 关Fig. 13共b兲兴
well. The positive near-wall peak is reduced and moved out- seems to be independent of Re␶ for the fully rough flows,
wards roughly by k+ and the negative peak has practically supporting the concept of a Re␶-independent near-wall layer
vanished because of the partial break-up of the streamwise that is affected by the surface roughness. Outside y / h

FIG. 15. Third order mixed moment u+v+2. Symbols in


Table I.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-14 Bakken et al. Phys. Fluids 17, 065101 共2005兲

3
FIG. 16. Third order moment v+ . Symbols in Table I.

⯝ 0.2– 0.25, the smooth- and rough-wall profiles follow each somewhat further out than for the Reynolds stresses. The
other for similar Re␶, indicating negligible structural differ- Reynolds number dependency in the outer region is increas-
ences in the outer part, again consistent with the quadrant ing with the power of u, indicating that this primarily affects
3
analysis.12 In the outer layer, u+ is clearly more negatively the inactive motion.
skewed with increasing Re␶, corresponding to the increase in
2
inactive motion observed for u+ . V. CONCLUSIONS
2
The vertical turbulent flux of u2 , u+ v+, represents most
of the outward diffusion of turbulent energy 共Fig. 14兲. The The turbulent flows in a two-dimensional channel where
significant positive peak which is seen for the smooth-wall the surface has been roughened by two-dimensional rods or
Re␶ = 670 profile is attenuated and moved outwards for the three-dimensional mesh roughness have been compared with
3 that of a smooth surface. The measurements have been per-
transitional roughness profiles, as found for u+ . The fully
formed with hot-wire anemometry. The results indicate that
rough profiles are much less skewed near the wall supporting
the boundary conditions have little influence on the turbulent
the findings of Krogstad and Antonia6 that surface roughness
properties in the outer region, y / h ⬎ 0.15– 0.17. Different
reduces anisotropy. The different surfaces show the same
scaling velocities have been examined and the best collapse
trends for y / h ⬎ 0.2– 0.3. In the outer region there seems to
is achieved when scaling with u␶.
be a tendency towards more positively skewed profiles for
3 In the fully rough regime, the roughness function ⌬U+,
increasing Re␶. This supports the findings from the u+ the vertical shift in the log law, displays the characteristic
profiles. logarithmic dependency on the roughness Reynolds number,
The wall-normal turbulent flux of −uv and v2, i.e., the k+. The velocity defect plots indicate a high degree of simi-
2 3
profiles of u+v+ and v+ , respectively, are shown in Figs. 15 larity in the outer layer.
and 16. The uncertainty in the estimates close to the wall is For the streamwise normal Reynolds stress there is a
growing with increasing powers of v. Thus, no conclusions substantial sensitivity to the wall condition in the roughness
are made for this region. In the outer region, the same sort of sublayer, y ⬍ 5k. As k+ increases, there is a transfer from a
3
similarity as observed for the u+ profiles is seen, except for domination of viscous forces to drag induced pressure forces,
the Reynolds number trend, which is much less pronounced, which causes a decrease in the magnitude of the near-wall
3
especially for v+ . These results compare well with the find- peak. In the outer layer, the effect of surface roughness is
ings of Raupach et al.44 in rough-wall boundary layers. Other small, supporting the Townsend wall similarity hypothesis.3
investigations from boundary layers, where differences in There is, however, a Reynolds number dependence common
Reynolds stresses between smooth and rough surfaces are to all three surfaces at low Reynolds numbers. For the other
found to be important, have also found large differences for stresses the collapse between the data from the different sur-
the third order moments.6,45 face geometries is good, but it is difficult to conclude about
As is observed for the Reynolds stresses, it may there- the Reynolds number dependency.
fore be concluded that the turbulent transport terms are Transport properties, as reflected by the third order mo-
modified by the surface roughness in the inner region. The ments, are very similar outside y ⯝ 7k – 8k for the three sur-
3
wall-normal extent of this modification is moderate for the face geometries. Observations of u+ confirm the conclusions
transitional roughness. For the fully rough profiles it extends drawn from a quadrant analysis of the same flows.12 Ap-

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-15 Reynolds number effects in the outer layer Phys. Fluids 17, 065101 共2005兲

13
proaching the wall, sweeps are enhanced over the rough sur- M. V. Zagarola and A. J. Smits, “Mean-flow scaling of turbulent pipe
flow,” J. Fluid Mech. 373, 33 共1998兲.
faces because of the reduced damping of the wall-normal 14
J. F. Morrison, W. Jiang, B. J. McKeon, and A. J. Smits, “Scaling of the
motion. streamwise velocity component in turbulent pipe flow,” J. Fluid Mech.
Previous investigations of flows in channels which have 508, 99 共2004兲.
15
been roughened on one side only, and of rough-wall bound- F. T. M. Nieuwstadt and P. Bradshaw, “Similarities and differences of
turbulent boundary layer, pipe and channel flow,” in Boundary-Layer
ary layers, have suggested strong roughness effects in the
Separation in Aircraft Aerodynamics, edited by R. A. W. M. Henkes and P.
outer layer. This strong interaction between the inner and G. Bakker 共Delft University Press, Delft, 1997兲.
16
outer layers has been found to mainly cause an increase in D. B. DeGraaf and J. K. Eaton, “Reynolds-number scaling of the flat-plate
2
turbulent boundary layer,” J. Fluid Mech. 422, 319 共2000兲.
v+ . This was not found to be the case in the present inves- 17
W. K. George and L. Castillo, “Zero-pressure-gradient turbulent boundary
tigation. We believe the main reason for this to be due to the layer,” Appl. Mech. Rev. 50, 689 共1997兲.
differences in boundary conditions. In the boundary layer the 18
D. B. DeGraaf, D. R. Webster, and J. K. Eaton, “The effect of Reynolds
increased growth rate over rough surfaces will cause in- number on boundary layer turbulence,” Exp. Therm. Fluid Sci. 18, 341
creased entrainment which will affect the outer layer diffu- 共1999兲.
19
M. Wosnik, L. Castillo, and W. K. George, “A theory for turbulent pipe
sion and therefore also to some extent the stress profiles. For and channel flows,” J. Fluid Mech. 421, 115 共2000兲.
flows in channels with only one rough surface, the difference 20
Y. Furuya, M. Miyata, and H. Fujita, “Turbulent boundary layer and flow
in drag between the rough and the smooth surfaces causes an resistance on plates roughened by wires,” J. Fluids Eng. 98, 635 共1976兲.
21
asymmetry that offsets the outer layer towards the smooth J. Cui, V. C. Patel, and C.-L. Lin, “Large-eddy simulation of turbulent
flow in a channel with rib roughness,” Int. J. Heat Fluid Flow 24, 372
side. It appears reasonable that this asymmetry will affect the 共2003兲.
outer layer structure. As a concluding remark, the symmetri- 22
A. E. Perry, K. L. Lim, and S. M. Henbest, “An experimental study of the
cally roughened channel flow and flows in rough-wall pipes turbulence structure in smooth- and rough-wall boundary layers,” J. Fluid
appear to be better candidates for the wall similarity hypoth- Mech. 177, 437 共1987兲.
23
M. Tagawa, T. Tsuji, and Y. Nagano, “Evaluation of x-probe response to
esis of Townsend3 than, e.g., boundary layers, suggesting wire separation for wall turbulence measurements,” Exp. Fluids 12, 413
that outer layer roughness effects may depend on the flow 共1992兲.
24
type. O. M. Bakken and P.-Å. Krogstad, “A velocity dependent effective angle
method for calibration of X-probes at low velocities,” Exp. Fluids 37, 146
共2004兲.
ACKNOWLEDGMENT 25
P. W. Bearman, “Corrections for the effect of ambient temperature drift on
hot-wire measurements in incompressible flow,” DISA Inf. 11, 25 共1971兲.
Financial support by the Norwegian Research Council 26
R. D. Moser, J. Kim, and N. N. Mansour, “Direct numerical simulation of
for A.A. is greatly acknowledged. turbulent channel flow up to Re␶ = 590,” Phys. Fluids 11, 943 共1999兲.
27
L. F. Moore, “An experimental investigation of the boundary layer devel-
1
J. Nikuradse, “Strömungsgesetze in rauhen Rohren,” VDI-Forsch. 361 opment along a rough surface,” Ph.D. thesis, State University of Iowa,
共1933兲 共English translation, “Laws of flow in rough pipes,” NACA TM 1951.
28
1292, 1950兲. R. B. Dean, “Reynolds number dependence of skin friction and other bulk
2
H. Schlichting, Boundary Layer Theory, 6th ed. 共McGraw-Hill, New York, flow variables in two-dimensional rectangular duct flow,” J. Fluids Eng.
1968兲. 100, 215 共1978兲.
3
A. A. Townsend, The Structure of Turbulent Shear Flow 共Cambridge Uni-
29
S. B. Pope, Turbulent Flows 共Cambridge University Press, Cambridge,
versity Press, Cambridge, 1976兲. 2000兲.
4 30
A. E. Perry and C. J. Abell, “Asymptotic similarity of turbulence struc- S. Mochizuki and F. T. M. Nieuwstadt, “Reynolds-number-dependence of
tures in smooth- and rough-walled pipes,” J. Fluid Mech. 79, 785 共1977兲. the maximum in the streamwise velocity fluctuations in wall turbulence,”
5
P.-Å. Krogstad, R. A. Antonia, and L. W. B. Browne, “Comparison be- Exp. Fluids 21, 218 共1996兲.
31
tween rough- and smooth-wall turbulent boundary layers,” J. Fluid Mech. P. M. Ligrani and R. J. Moffat, “Structure of transitionally rough and fully
245, 599 共1992兲. rough turbulent boundary layers,” J. Fluid Mech. 162, 69 共1986兲.
6 32
P.-Å. Krogstad and R. A. Antonia, “Surface roughness effects in turbulent J. Jimenez and P. Moin, “The minimal flow unit in near wall turbulence,”
boundary layers,” Exp. Fluids 27, 450 共1999兲. J. Fluid Mech. 225, 221 共1991兲.
7 33
S. Leonardi, P. Orlandi, R. J. Smalley, L. Djenidi, and R. A. Antonia, O. M. Bakken and P.-Å. Krogstad, “Small-scale intermittency in a turbu-
“Direct numerical simulations of turbulent channel flow with transverse lent channel flow and the effect of boundary conditions,” in Advances in
square bars on one wall,” J. Fluid Mech. 491, 229 共2003兲. Turbulence X, edited by H. I. Andersson and P.-Å. Krogstad 共CIMNE,
8 Barcelona, 2004兲, pp. 401–404.
Y. Miyake, K. Tsujimoto, and M. Nakaji, “Direct numerical simulation of
34
rough-wall heat transfer in a turbulent channel flow,” Int. J. Heat Fluid J. Jeong, F. Hussain, W. Schoppa, and J. Kim, “Coherent structures near
Flow 22, 1237 共2001兲. the wall in a turbulent channel flow,” J. Fluid Mech. 332, 185 共1997兲.
9 35
Y. Nagano, H. Hattori, S. Yasui, and T. Houra, “DNS of velocity and A. E. Perry and C. J. Abell, “Scaling laws for pipe-flow turbulence,” J.
thermal fields in turbulent channel flow with transverse rib roughness,” Fluid Mech. 67, 257 共1975兲.
36
Proceedings of Turbulence and Shear Flow Phenomena—3, edited by N. L. P. Erm, “Low-Reynolds-number turbulent boundary layers,” Ph.D. the-
Kasagi, J. K. Eaton, R. Friedrich, J. A. C. Humphrey, M. A. Leschziner, sis, University of Melbourne, Australia, 1988.
37
and T. Miyauchi 共Sendai, Japan, 2003兲, Vol. 3, pp. 1055–1060. R. A. Antonia, M. Teitel, J. Kim, and L. W. B. Browne, “Low-Reynolds-
10
S. Leonardi, P. Orlandi, L. Djenidi, and R. A. Antonia, “Structure of number effects in a fully developed turbulent channel flow,” J. Fluid
turbulent channel flow with square bars on one wall,” Proceedings of Mech. 236, 579 共1992兲.
38
Turbulence and Shear Flow Phenomena—3, edited by N. Kasagi, J. K. A. E. Perry and J. D. Li, “Experimental support for the attached-eddy
Eaton, R. Friedrich, J. A. C. Humphrey, M. A. Leschziner, and T. Miyau- hypothesis in zero-pressure-gradient turbulent boundary layers,” J. Fluid
chi 共Sendai, Japan, 2001兲, Vol. 1, pp. 123–128. Mech. 218, 405 共1990兲.
11 39
A. Ashrafian, H. I. Andersson, and M. Manhart, “DNS of turbulent flow in H. H. Fernholz and P. J. Finley, “The incompressible zero-pressure-
a rod-roughened channel,” Int. J. Heat Fluid Flow 25, 373 共2004兲. gradient turbulent boundary layer: An assessment of the data,” Prog.
12
O. M. Bakken and P.-Å. Krogstad, “Stress measurements in a rough wall Aerosp. Sci. 32, 245 共1996兲.
40
channel flow using a variable angle method of calibration for X-probes,” J. Sabot, I. Saleh, and G. Compte-Bellot, “Effects of roughness on the
Proceedings of Turbulence and Shear Flow Phenomena—3, edited by N. intermittent maintenance of Reynolds shear stress in pipe flow,” Phys.
Kasagi, J. K. Eaton, R. Friedrich, J. A. C. Humphrey, M. A. Leschziner, Fluids 20, S150 共1977兲.
and T. Miyauchi 共Sendai, Japan, 2003兲, Vol. 1, pp. 105–110. 41
A. Mazouz, L. Labraga, and C. Tournier, “Anisotropy invariants of Rey-

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
065101-16 Bakken et al. Phys. Fluids 17, 065101 共2005兲

44
nolds stress tensor in a duct flow and turbulent boundary layer,” J. Fluids M. R. Raupach, R. A. Antonia, and S. Rajagopalan, “Rough-wall turbulent
Eng. 120, 280 共1998兲. boundary layers,” Appl. Mech. Rev. 44, 1 共1991兲.
42 45
L. Djenidi, R. Elavarasan, and R. A. Antonia, “The turbulent boundary L. Keirsbulck, L. Labraga, A. Mazouz, and C. Tournier, “Surface rough-
layer over transverse cavities,” J. Fluid Mech. 395, 271 共1999兲. ness effects on turbulent boundary layer structures,” J. Fluids Eng. 124,
43
T. Wei and W. W. Willmarth, “Reynolds-number effects on the structure of 127 共2002兲.
a turbulent channel flow,” J. Fluid Mech. 204, 57 共1989兲.

Downloaded 10 May 2012 to 193.194.76.5. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions

You might also like