Professional Documents
Culture Documents
edited by
E. COUSTOLS
ONERAICERT,
Toulouse Cedex,
France
Preface vii
Final programme ix
List of referees In
Preface
In the last decades, a lot of effort has been directed towards manipulation of
turbulent boundary layers by passive devices such as external manipulators (thin
flat plates or aerofoil section devices embedded in the outer layer) and/or internal
manipulators (small streamwise grooves acting directly on the inner region) for the
purpose of reducing viscous drag. The former are commonly referred to as LEI3U s or
BLADEs and the laHer riblets or grooves. Though the details of the mechanisms are
not firmly understood, world-wide experimenta.! data are available and consistent
enough in order to assert the potential of such devices for turbulent drag reduction.
It should be noted that following on from recent and successful flight tests, the
concept of using grooved surfaces is rather close to finding industrial applications.
During the last few years, in Europe, there has been considerable interest in
lookillg at the behaviour of such passi,'e turbulence manipulators. A lot of intense
research, concerning both experimental and theoretical studies. has been carried out
in some European research centres. For the last fi\'e years. informal gatherings. called
,.\ \'orking Pi\l'ty i\Ieetings" , have been set up, once a year; the aim of these meetings
is not only to bring together European researchers acti,'e in the field of turbulent
drag reduction by passi"e means and to hear about recent de\'c!opments but also to
o u t1 ine sui tit ble directions for future research or collaborative programmes.
Thus, follO\\'ing on from previous meetings at EPF Lausanne (September 1st-
:3rd, 1986), ONEllA/CEllT Toulouse (September lOth, 1987) and ONEnA/Chalillon
(September 20th-:30th, 1988), the fourth European Drag Reduction WorLing Party
:'Ieeting \\'itS held at EPF Lausanne on July 2 Hh, 1989 under the auspices of the
c
11COFTAC Pilot Centre by invitation of Professor 1.1. 11yhming and Doctor T.V.
Truong. T\\'ellty six participants from fi \'e European count ries (Fra nce, CermallY,
S \\' i tzerl a nd, The N etherla nels and U ni ted Kingdom), together \\'il h collaborating
colleagues from Canada, presented 18 contributions of \\'ork which eilher had heen
recently completed or was in progress. The purpose of this Proceedings Book is
to pro\'ide papers '\'hich reflect the contents of the talks given at that conference.
It should lw lloled that "A report on the 4 th European Drag neduction \\'orking
Party T\lccting" \\"as published in the EHCOFTAC Bulletin IV. December 19S'}
011 a personal note, I would like to say how grateful I am to all the rc1'erees
for reviewing some of the dozen or so papers offered after the meeting. Thanks
to their assist ance. this Proceedings Book should be published in time for the :5th
\\'orking Parly i\Ieeting which will be held at I3i\IT Fluid i\Iechanics Limited. Ted-
dington. Uniled Kingdom. I would also like to thank Doctor Nigel Hollillg\\'orth,
frolll KIll wer academic ))ll blishers, for all his va 1uable commen b dming the prepa-
ration of this Proceedings Book.
E. Coustols
vii
Final Programme
High speed LEBU device drag measurements. J.P. Bonnet &; J. Delville
Complementary drag reduction measurements. L.C. Squire &; A.:'.I. Savill
I-Combination of riblets and manipulators at low Re ; 2-Some observations of tllC
static pressure distributions in pipe flow perturbed by a LEEU ring. Y.D. I\guyen
&; J. Dickinson
Turbulent pipe flow manipulation and modelling. A. Pollard, A.'\1. Sa,'ill &; H.
Thomann
Sessioll /[ - Round Table. (E. Coustols, A.M. Savill &; LL. Ryhming)
ix
Balance of Turbulent Kinetic Energy
Downstream a Single Flat Plate Manipulator
Comparisons Between Detailed Experiments and
Modelling.
Summary
Previous numerical studies undertaken at ONERA/CERT /DERAT in Toulouse, ha\'e
shO\vn that usual closures, using three or five equations, were able to quite well repro-
duce the developpement of a manipulated boundary layer for the specific geometrical
configuration given from CERT. The goal sought after, in this study. is to check
the predictiw capabilities of the CERT code on an other specific configuration. given
from CEAT Poitiers, for which detailed experiments were available in the downstream
vicinity of an external manipulator of a rectangular section. In the .'\ avier-Stokes
equations, the unsteady and the diffusion terms in the streamwise direction are ne-
glcctecl. For the turbulent motion, a two layer model is adopted using a mixing length
scheme in the near wall region and a k, E, u'v' transport equations model everywhere
else. A brief description of the numerical procedure is first given. The experimental
apparatus and techniques are then described. The numerical results agree qualita-
tively \yell with the experimental mean velocity, turbulence and shear stress data
profiles. The experiments provide the terms of the energetic balance of the kinetic
energy equation ; they are, for further information, compared with the numerical
terms. Although the manipulated flow is strongly out of equilibrium, conventional
closure models appear to be able to predict rather correctly the measured quantities,
particularly the turbulent diffusion term and the dissipation behaviour.
2
1 Introduction
During the last decade, lot of experiments have been conducted in order to reduce skin
friction drag by the control of the eddy structures of turbulent boundary layers. For
airfim'>s, the techniques proposed can be divided into active and passive ones [2], [15].
One of the passive methods consists in immerging thin ribbons, having either rectan-
gular or aerofoil section, within the external part of a turbulent boundary layer. This
method, so called external turbulence manipulation, is studied in this article.
Experimental investigations have shown that the mechamisms leading to drag re-
duction are complex. Among several effects, one can notice that the externalmanipu-
lator induces large skin friction reductions and spectacular modifications on turbulent
profiles (excess of turbulence level in the device wake. important decreases belm\' the
mean axis of this wake, ... ). Furthermore, the manipulated boundary layer is highly
out of equilibrium O\'er a large dmvnstream extent.
Besides these experiments, numerical studies have been undertaken. The goal
was then to check usual turbulent closures, using several transport equations, for spe-
cific geometrical configurations. l"lost of the numerical codes, used up to now. '\Tre
based on boundary layer assumption and thereby started downstream of the device
with the initial data given from experiments. These numerical studies have been
reviewed by Coustols &: Savill [4]. Let us point out that, using appropriate initiali-
sations. authors were able to represent the mean and fiuctuating velocity profiles at
further stations downstream of the manipulator with a satisfactory agreement with
experiments. However, if one aims, for instance, at optimizing the device geometry,
it is necessary to start the computation upstream of the device and to follow the
whole of the fiow development. This second approach has been used in works pre-
viously undertaken at the Aerodynamic department of ONERA/CERT [3]. [13]. On
a given experimental configuration, check of several closure models have been done:
two, three and fiw transport equations, using cOlwentional turbulence models. The
conclusions ,'>as, though the k - E model was able to reproduce qualitati"ely the be-
haviour of such manipulated flow, better results have been obtained using three or
five transport equations. The mean and turbulence velocity profiles as well as the
skin friction coefficient have been calculated in better agreement with the experimen-
tal data [3] [13]. Let us point out that three or five equations models give rather
comparable results, except that the latter allows to get a better representation of the
C f evolution. A remaining question can be pointed out: Is these conclusions on the
CERT configuration stay valid for different device geometries?
The purpose of this study is then to check the predictive capabilities of the
C.E.R.T. numerical code on an other configuration given from the C.E.A.T. Poitiers,
for which detailed experiments were available in the downstream vicinity of an ex-
ternal manipulator having rectangular section [9], [10]. We decided to restrict the
present numerical approach by using only the three transport equations code. This
option gives a good compromise between computer time saving, previous prediction
3
2 Numerical approach.
2.1 Basic equations.
The equations of the mean flow are the Reynolds averaged Navier-Stokes equations.
The usual two-dimensional assumptions are used; x represents the streamwise direc-
tion and y the axis normal to the wall. The unsteady terms as well as the diffusion
ones in the :r direction are neglected. It follows:
aU aV
-+-=0. (1)
a:r ay
a ([J[j)
aol'
+ a(UV)
ay
= _~ aP
pool'
+.!!.- [~ (aU)] + .!!.- (-U'2) + .!!.- (-177l)
ay p ay ax ay (2)
a(0'V)
ax
+ a(11V)
ay
= _~ of +.!!.- [~(a1l)] +.!!.- (-u'v') +.!!.- (_v'2)
p ay ay p ay a:r ay (3)
For the turbulent motion, a two-layer model is adopted:
1. Although the flow is out of equilibrium even close to the ,Yall, a mixing length
scheme is applied, for simplicity, in the near wall regions (up to y+ = 60),
including the lower and upper sides of the device. This scheme is based upon
an eddy \'iscosity assumption:
The eddy viscosity coefficient {LI is expressed, using a mixing length formulation.
as follows:
(5)
where I is the mixing length and F the Van Driest type corrector function for
the viscous sublayer. The use of this scheme, close to the manipulator, can be
4
made if we assume that the boundary layers, developping on both sides of the
device, are turbulent. Let us notice that this assumption is not clearly evidenced
by the available experiments.
2. A three transport equations model is used everywhere else (y+ > GO). These
equations are :
---
It au: au:
p aXj aXj
(6)
C" and D" are respectively the convect.ion and molecular diffusion terms; they
do not require any modelisation. The turbulent diffusion term T" is modeled as
follows [8] :
(7)
Let 11S notice that this model is based upon a high turbllient Reynolds number
assumption (RI = pk 2 / w:). Close to the limit y+ = GO or in the deyice wake
the RI values could be moderate; therefore, we use a wall damping function
III :
[ -3.4 ] with
11' = exp (1 + Rt/50)2
The pressure-velocity correlation <Pk is neglected. c represents the dissipation
rate .
(8)
12 = 1. - 0.3exp (-Rn
5
Finally, the transport equation for the turbulent shear stress is used :
(9)
with
120U
P12 = -v -oU
D12 = -u,2-
-
oy oy
and
C 2 +8
a: = - - - -
11
f3 = 2 - 30C2
55 ,=- 8C112 - 2
In this three-equations model, one needs closure relationships for the diagonal
Reynolds stress components. From the transport equations for these components and
through a local equilibrium assumption, we can correlate the diagonal components
(u;u:, fori = j) to the turbulent kinetic energy; the ratios (U:llj / k, for i = j)
are function of constants C l , C 2 . The choice C l = 1.5 and C 2 = 0.4 leads to the
following values:
However, it has been experimentally evidenced [10] that, in the imediate \icinity
of the manipulator, these three ratios are only slightly affected. Then, we keep the
assumption of local equilibrium for the three normal components of the Reynolds
stress tensor. Furthermore, the - u'v' / k ratio is strongly modified just downstream
of the deyice [10]' justifying the use of the u'v' transport equation.
3 Experimental investigation.
Wind tunnel.
The experimental study is performed in a closed loop wind tunnel with a test section
2. III long and 300 x 300 mm 2 cross section. The settling chamber is equiped with
filters and grids: the contraction ratio is 10. Both the upper and lower walls are
adjusted in order to pre\'ent pressure gradient. The turbulence le\'el in the free stream
is less than 0.2%.
The experimental data proceed from hot wire anemometry measurements (TSI 1750
CTA). Both single and X wires are used in order to measure the different quanti-
ties involved in the transport equation of the turbulent kinetic energy. The spatial
gradients included in this equation are properly estimated via a finite difference mea-
surement grid, tightened in the regions of strong gradients (near the wall and in the
vicinity of the trailing edge). The measurements are however perfonned far enough
from very strong gradients in order to limit integration effects arising from spatial
extent of the X wire probes [10]' [l1J.
Single normal wire: modified miniature probe TSI 1260 T1.5 operating with
tungsten wire of 0.5 Tl7m long and 2.5 Ilm in diameter.
X wires: TSI 1248 Tl.5 probe with tungsten wires of l.25771m long and 4 I'm.
in diameter.
Acquisition system.
All the terms of the transport equation of k (high Reynolds number assumption)
are measured except 1>" and E. An estimation of 1>" + E is then obtained by balancing
this equation. If one assumes that the pressure-velocity correlation 1>" is negligeable,
the remainder of the k equation can be considered as an estimation of E.
i\Iean velocity
vVith this three equations model, a rather good overall agreement with experi-
mental data can be observed (Fig. 1). Howeyer, at the first location (X/Do = 0.43),
the predicted wake is too large when compared with experiments. This discrepancies
might come from the use of a mixing length scheme in the treatment of the near wall
region on both sides of the device. Further downstream, one can notice that the evo-
lution of the deficit pocket is slower than the one observed in the experiments. Similar
conclusions have previously been pointed out on the CERT numerical results. At the
farthest location (X/Do = 20.), the mean velocity profile is slightly overestimated over
the whole thickness of the boundary layer (Fig. 1) .
Turbulent fields
At the first downstream station (X/Do = 0.43), the excess of longitudinal turbu-
lence intensity is quite well predicted (Fig. 2). On the other hand, farther downstream
(X/Do = 7.5), the well known effect of the manipulator inducing the important de-
creases of 1l'2 component below the wake axis, is well reproduced by the calculation.
Unlike the comparisons with the ONERA/CERT data [3], [13], the streamwise tur-
bulence intensity is well predicted in the region near the solid wall where a mixing
length assumption is made.
The results concerning the turbulent shear stress profiles are plotted on figure 3.
Close to the trailing edge of the manipulator, in the outer region located above the
mean axis of the wake, the calculation overpredicts the u'v' level. These discre-
pancies might be due, one more time, to the use of a mixing length scheme for the
9
calculation of the turbulent motion on both sides of the device. This overestima-
tion of u'v' decreases as the downstream distance increases, to fit with the data at
about 7.5 boundary layer thicknesses (Fig. 3). At contrary, below the device height,
the important decrease of the turbulent shear stress profile, experimentally observed
downstream of the device, is rather well predicted by the model (Fig. 3). However,
close to the wall, the numerical code provides with too high u'v' values. At the last
measured station (X/Do = 20.), the Reynolds stress tensor component v'v' tends to
be overestimated in a large part of the boundary layer thickness.
The performance of the present model can also be evaluated on its ability to predict
the evolution of the skin friction coefficient in manipulated flow. Indeed, this effect
is the most important for pratical purposes and its correct prediction is crucial for
optimizations. The values of C f / C fre! obtained both by calculations (C f = l/;~U;'
Tp = (p~U)
uy
)
y=o
and data acquisitions (Preston tube), are plotted \"Cl"SUS X /6 0 on
figure (Fig. 4) ; C f,.e! denotes the skin friction coefficient corresponding to the natural
flow.
The calculations show a more important local skin friction reduction than the
experiments (Fig. 4). Once again, the wall region treatment employed (mixing length
scheme) might explain these discrepancies. Furthermore, the location of the maximum
of local reduction is located slightly upstream of the measured one. This behaviour
can be explained by the fact that the calculated deficit pocket of the mean wlocity
spreads faster towards the wall than the experimental one. Then, the calculated
wake reaches the wall at a point located slightly upstream when compared to the
experiments. This interpretation is supported by previous experiments : actually,
visualizations and thermal marking of the wakc have experimentally revealed that the
location of maximum C f reduction corresponds to the point whcrc the wakc reaches
the solid wall [1], [6], [10]. Finally, one can notice that the predicted relaxation of the
skin friction coefficient toward natural values is in good agrecment with experimental
data (Fig. 4).
From the analysis of this preliminary step, one can observe rcsults on mean and
fluctuating quantities profiles in rather good qualitative agreement with the CEAT
experimcnts. j\Ioreovcr, some discrepancies occur on shear stress profilcs in the im-
mediate downstream vicinity of the manipulator and \"Cry close to the wall. In fact,
the turbulence model, used in the wall region of both sides of the device, can be
responsible of these discrepancies. This application of a three equation model to the
flat plate manipulator expcriments of CEAT [9], [10] confirms thc results prc\iously
obtaincd \\"ith the experimental data of ONERA/CERT [3]' [13].
In ordcr to more preciscly analysc the different turbulence models and tcrms
involvcd in the present closurc, the behaviour of several terms of the equations can
be studicd. On one hand, as previously described, one original aspect of the available
experiments [10]' [11] is the estimation of every term of the transport equation of the
turbulent kinetic energy k. On the other hand, the present code allows to calculate
10
all the terms involved in the k equation. The purpose of thc next section is then to
compare expcrimental and numerical valucs of the term balance of the k equation.
(Fig. 7) compared to the experimental one. Just below the manipulator posItIon
the association of the wake velocity defect and the boundary layer profile imposes
~u = 0 ; the production term then vanishes whatever the turbulent shear stress is.
J~st aboYe the manipulator height, the production of f.;; is numerically O\'erestimated,
due to theu'v' overprediction already mentionned (see 4.1).
For locations dO\vnstream of the device, one observes that the numerical relaxation
of h: is slower than the experimental one in the lower part of the boundary layer.
In this region, theu'v' level is well predicted, the discrepancies on PI,; can only be
attributed to the mean velocity prediction. Even at the last station (X/[,o = 20.)
(Fig. 7), the production term given from calculation has not recovered its natural
value while, in the experiments, there is no difference between manipulated and non-
manipulated PI,; profiles [10].
The dissipation rate E, calculated through the transport equation( 8), is plotted on
figure (8). Though this equation involves modelisations developped for high Reynolds
numbers, an overall rather good agreement is achieved on the E profiles compared to
the estimated experimental values (Fig. 8), even in the near wake where the Reynolds
number is relatively low; then the low R t corrector functions seems to be sufficient.
Howe\'Cl", at the first location (X / [,0 = 2.), the predicted width of the E increase
is larger than the experimental results. These discrepancies might be related to
the owrprediction of Pk , since the production term of the dissipation equation 8 is
modeled through the use of PI,;.
5 Conclusion.
After the numerical studies undertaken at ONERA/CERT /DERAT in Toulouse, it
\vas raised that usual closures, using three or five equations, were able to reproduce
quite well the clewloppement of a manipulated boundary layer for a specific geomet-
rical configuration given from CERT. The goal sought after, in this study, is to check
the predictive capabilities of the CERT code on other experimental configuration. As
detailed experiments were available at CEAT Poitiers, in the downstream vicinity of
an external manipulator having a rectangular section, we ha\'e decided to applied the
CERT code to this specific CEAT geometrical configuration. For computer time sav-
ing, we decided to restrict the present study; the turbulent motion is then calculated
using a three transport equations model (k, c, u'v').
Numerical results have shown a rather good agreement on the evolution of the
mean and fluctuating velocity profiles with experiments, except close to the trailing
edge of the device where some discrepancies occur on the shear stress profiles in
the manipulator wake. As regards to the downstream evolution of the skin friction
coefficient, one can notice that the maximum of C f reduction is overestimated by the
calculations. However, the location of this maximum as well as the relaxation of C f
are rather well predicted. One more time, a remainding difficulty in the calculation
concerns how to model the wall region, including both the upper and lower sides of
12
the device. Let us add that rather same observations have been pointed out when
discussing the numerical results on the CERT-type configuration. A solution might
be to use more sophisticated closures using transport equations in the wall layer.
As experimental balance of the turbulent kinetic energy equation was available at
three locations downstream the manipulator, we have decided to undertake a com-
parison between experiments and modelling on each term of the k equation.
The molecular diffusion is negligible in front of the other terms. The numerical
estimations of convection and turbulent diffusion terms agree quite well with the cor-
responding experimental profiles; the downstream relaxation of these terms is pretty
well reproduced. As regards to production and dissipation profiles, rather good agree-
ment with experiments could also be observed. However, some slight overpredictions
occur at the device height, in the downstream vicinity of the manipulator; this might
be related to the overprediction of U'V ' at the same location. Furthermore, the relax-
ation of Pk seems to be slower than in the experiments; this can be attributed to the
prediction of the downstream evolution of the mean velocity.
At last, let us say that, although the manipulated flow is strongly out of equilib-
rium, conventional closure models are able to predict rather correctly the measured
quantities, particularly the turbulent diffusion term and the dissipation behaviour.
Acknowledgements
This work has benefited from an active collaboration between ONERA/ CERT /DERAT
Toulouse and CEAT Poitiers. The detailed experiments has been recorded at the
CEAT Poitiers while the calculations have been performed at the Aerodynamic De-
partment of the CERT Toulouse.
C. Tenaud thanks specially Professor Jean Cousteix, head manager of the Aerody-
namic Department of CERT, and Doctor Eric Coustols, research engineer in the same
group, for their numerous and valuable comments and dicussions on the modelling of
disturbed boundary layers through outer layer devices.
References
[1] J.P. Bonnet - J. Delville - J. Lemay Study of LEBUs Modified Turbulent
BOllndary Layer by Use of Passive Temperatllre Contamination. International
Conference on Turbulent Drag Reduction by Passive Means, The Royal Aero-
nautical Society, London: RAeS 1 pp.45-68, 1987
[5] B.J. Daly - F .H. Harlow Transport equations in turbulence. Physics of fluids
Vol. 16, p. 157 (1973)
[12] S.V. Patankar - D.B. Spalding A calculation procedure for heat, mass and
momcntum transfer in thrce dimensional parabolic flows. Int. J. Heat \1ass Trans-
fer, Vol. 15, p. 1787 (1972)
~---.~-~
2.
40 1------+---- ~- --
40 ~
----
4
~- - - --j------+----1
201-----+----, 20
')
0.4
o
o 0.4
lJ 0.8
y (mm)
t
7.5 20.
40 1- -~
401--~--~~--r_~
lL
20 20 f---f--1-----1
lJ~
o OL--L__~~__L_~
o 0.4 0.8 o 0.Lt 0.8
X/oo =0.43 2.
f.O - -~---I
20
~ ~
~~
o L--_~_'-----'
o 5 10 5 10
y (mm)
t
~u'2/Ue(%)
-.-
7.5 20.
40 40 --f-----,----I
20 \
--i
\
~
5 10 10
X/5 0 =0.43 2.
LI 0 -~f-------. 40
201----il-----+ 20 ~
\~
~
O~ __L-~L_~_=~ o ~
-.002 0 .002 ~002 o .002
y(mm)
-u-,-,/U
t
2
V e
~
7.5 20.
40 40
- -_ . f
4
20
~. 20 \
o
-.002 o
" b
.002
o
f------ - -
-.002 o
" :~ .002
1.
0.8
XI[)o
o. 10 20 30 40
-J
00
y/5
-
1.2 1.2 1.2
XI fJ o = 2, 7.5 20,
t I
0.8 0.8 0.8
I
!
, t !
i ,
~ :> ~ - -
0.4 0.4 0.4 I I
I I I
"-
Ii !
r--
t !
: I
o ~ o i<:LLLLc.L.LJ::.L..:L..L.
i o.
I-
-10 o 10 -10 0 10 -10 o 10
---I.~ Ck , 01 U~ x 1 0 4
1.2 ,--
1.2 1.2
X 10 0 =2. t 7.5 20.
0,8I 0.8 0.8
r. i ~
0.4It--f-d ~ 0,4 0.4
l
lr 't
)
/
o O. o '%
r "/// //hJ
\0
N
o
N
Manipulation and Modelling of Turbulent Pipe Flow:
Some Parametric Studies of Single and Tandem Ring Devices
A.Pollard H.Thomann
Professor, Professur fiir Str6mungslehre,
Department of Mechanical Engineering, Eidgen6ssiche Technische Hochschule,
Queen's University at Kingston, Sonnegstrasse 3,
Ontario K7L-3N6 Zurich CH-8092
CANADA SWITZERLAND
A.M.Savill
Rolls-Royce Senior Research Associate,
Engineering Department,
University of Cambridge,
Cambridge CB2 IPZ
ENGLAND
SUMMARY
The effect of both single and tandem plate manipulators on developing pipe flow
has been investigated both experimentally and computationally. Numerical calcula-
tions have been performed for nearly developed flow using a finite- volume, elliptic,
low-Re turbulence model code. The results are compared with measured wall pres-
sure, mean velocity, and turbulence intensity profiles, as well as earlier computational
results for similar devices mounted in the initial developing flow region. The numer-
ical results agree closely with experiment and both suggest that nett drag reduction
may only be possible in the fully developed case and over a longer development
length behind a more optimum tandem configuration than could be considered in
the present studies. Further calculations are now being performed to examine this
latter possibility prior to any further experimental investigations.
INTRODUCTION
23
24
M=1.2 [e.g. see Savill (1989)]. Whether similar benefits may be attainable through
the use of aerofoil manipulators is still a matter for debate and further study, but it
has been shown that these can be constructively combined with riblets and indeed
both devices have been successfully tested in conjunction with other established flow
control techniques such as polymers and blowing [e.g. see Savill (1989), Sellin &
Moses (1989)]. At the same time considerable interest has arisen in the possibili-
ties of using manipulators, and also riblets, for other purposes e.g. to reduce wall
pressure flnctuations, and hence noise, and to control heat transfer and mixing.
Initial attempts to model manipulated boundary layers with simple mixing length
and k-E models have led rapidly to more sophisticated treatments incorporating stress
transport closures [Savill (1989)] and even, most recently, some Large Eddy Simula-
tion results [Friedrich & Kline 1989,1990]. Parabolic stress transport computations
ha\'e been used successfully; first to post-dict experimental results for thin single
and tandem flat plate devices in zero and adverse pressure gradient boundary layers,
including the influence of free-stream turbulence; then to predict the effect on such
manipulated layers of larger free-stream turbulence intensities and length scales rep-
resentative of conditions within aero-engines [Savill 1987], and streamline curvature
[Savill (1988)], for which there are as yet no experimental data. At the same time
an elliptic scheme, starting upstream of the manipulators and employing k-E, or 3-
or 5-equation turbulence models has been developed for ONERA-CERT [Coustols
et al. (1987b)] and used, again successfully, to predict the optimum height and
spacing of thin tandem plate manipulators for low Reynolds number operating con-
ditions. The same approach is now being extended to curvilinear co-ordinates in
order to make parametric optimisation computations for the more robust aerofoil
devices which will be required in any practical application. Some extensive results
have already been obtained for N ACA 0009 and ONERA D profile devices in laminar
flow conditions [Djenidi et al.(1990a)], for which there is experimental evidence that
Tollmien-Schlicbting wave growth can be delayed by manipulators [Bardakhanov et
al. (1989)], and these are presently being compared with the only other available
'data' from simulations performed by Friedrich & Klein [1990]. Some initial tur-
bulent computations are also now also being performed by Djenidiet al. (1990b)
for comparison with the experimental results of Coustols et al. (1987a). It is clear
that there are still some deficiencies in the model approximations nsed, particularly
for the near-wall, near-device, and device-wake regions. To overcome these waIl
function or mixing length models, low-Re closures are being used. Indeed initial
low-Re k-E results have already been obtained by Johnston (1987) and Shenxi &
Zhou (1989), using the model of Chien (1982), as well as by Tenaud (1988) using
the Launder-Sharma (1974) model, and in future it seems likely the introduction of
newly-developed 2-dimensional, 2-component limit models [e.g. see Fu et ai. (1987)]
will lead to a further improvement in predictions, particularly since it has already
been demonstrated that these can also improve predictions for wakes.
The concept of using such passive devices to beneficially control external turbu-
25
lent boundary layers is thus well established, but surprisingly, particularly in view
of the possible applications to pipelines and ducting, considerably less attention has
been paid to internal flow manipulation. In fact riblet performance levels similar to
those found in external flows have recently been recorded in a fully developed duct
Lowson et al. (1989) and in a range of pipe flow experiments, Dinkelacker et al.
(1987), Reidy and Anderson (1988), Rohr et al. (1989), Liu et al. (1989); even for
the case of two-phase flow [Kelman et al. (1989)). By contrast, manipulator experi-
ments which have been conducted in fully developed channel flow at NAL Bangalore
[by Prabhu et al. (1987), Vasudevan et al. (1990)), and in developing pipe flow, at
ETH Zurich [Pompeo & Matievic (1987a,b), Lineton (1988), Samec (1989)) and at
Laval University [by Dickinson & Nguyen - see Coustols et al. (1990)), have revealed
no nett benefits. However only a limited number of device configuration parameters
have so far been investigated, and it is not clear what should be the optimum values
for these, or even how they should be scaled in the case of fully developed flow when
fj can no longer be used as a reference. So far it has generally been assumed that
this can simply be replaced by the duct/channel half-height or the pipe radius, but
the NAL group's finding of a best result for a device mounted at the duct centre-
line casts some doubts on this. Despite the fact that further improvements are still
required to near-wall/near-device modelling, there is clearly some advantage to be
gained from applying current computational schemes to internal flows in an attempt
to resolve this scaling question and to help circumvent the need for very extensive
and time-consuming parametric experimental optimisation studies. Indeed, prelimi-
nary low-Re k-e elliptic computations, performed previously by Pollard et al. (1989),
have already successfully post-dicted some of the ETH experimental results for single
ring manipulators mounted in the developing flow region of the same pipe.
The purpose of this paper is to present new low-Re k-e predictions for both
single and tandem ring manipulator configurations in nearly developed pipe flow
and to compare these with the earlier computational results for developing flow
and a wider range of experimental data obtained at ETH Zurich. The aim of the
present study has been to try to establish the criteria for optimising devices in pipe
flows, before conducting any further experimental studies, in order to minimise the
parametric measurements required to determine the maximum possible benefit that
can be achieved by such manipulation. As such, this work forms part of a much
wider collaborative program on turbulent drag reduction and modelling involving a
far larger number of research groups both within the European Research Community
on Flow Turbulence and Combustion (ERCOFTAC) and in Canada [see Lemay et
al. (1990)].
EXPERIMENTS
downstream from the inlet. The manipulator rings were of different radii (r) so as to
vary the distance (h=R-r) of these from the pipe wall, and were of metal construction,
with a cord length (1) of 20mm and a separation for the tandem arrangements of
135 or 35 mm; each ring being supported from the internal surface of the pipe
by three, thin, equally-spaced struts. The basic geometry considered in both the
experiments and the computations was thus as indicated by Figure 1, although in
the experiments the manipulators were tapered with their thickness (t) falling from
0.3mm at the leading (rounded) edge to 0.15mm at their trailing (sharp) edge.
The pipe walls were instrumented with pressure taps and mean velocity and
turbulence data were recorded by means of Pitot (employing a Macmillan correction
for shear and wall inflence effects) and hot-wire probes respectively. Data collected
included axial wall pressure distributions, and radial mean velocity and turbulence
intensity profiles at a circumfirential location mid-way between two of the support
vanes.
The range of parameters covered by the experiments are indicated in Table 1.
Note that for the case of manipulators introduced into the initial developing flow
region. where a wall layer thickness 8 could still be identified, these included the
apparent optimum values of: h;::::O.78, 1;::::1.58, and for the tandem case: s;::::108 sug-
gested for external flow applications. The corresponding device parameters for the
similar fully developed pipe flow experiments conducted subsequently at Laval, and
channel flow investigations performed at N AL, are also included in Table 1, together
with those adopted for the Large Eddy Simulations performed at Technische Uni-
versitiit l\hinchen. It can be seen that, a1though similar Reynolds numbers and wall
separations have been investigated in each case, the ETHZ manipulators were rather
shorter and more closely spaced than the devices studied at either Laval or N AL
(when all are scaled in terms of pipe or duct diameter for fully developed flow con-
ditions). The total development length of 120R, and hence the maximum available
reCO\'ery length, was also rather less for the ETHZ experiments than for the other
experimental studies, and the flow only attained a fully developed state at X=55D
(although the initial wall layers had merged by x;::::30D).
TURBULENCE MODELLING
The computations were performed using a low-Reynolds number k-t model, in-
corporated into a finite-volume code with a non-uniformly distributed staggered grid,
which has previously been validated for turbulent pipe flow by Pollard and Mart-
inuzzi (1989a,b). In view of their recommendations, based on comparative tests of
various k-t and stress transport models, the Lam & Bremhorst (1981) low-Re k-t
model was adopted. Since this low- Re model accounts for molecular viscosity effects
by introducing a damping function II' for the eddy viscosity which varies between
o at a solid surface and 1 in the fully turbulent region, the distance from any such
surface must be specified uniquely. For the manipulated flow configurations this dis-
27
tance was evaluated by taking the smallest distance between any point in the flow
and the nearest solid boundary. (The support struts were ignored).
The code utilised hybrid-differencing for the convective terms since Martinuzzi
and Pollard, and many others have shown that for pipe flow this results in negligible
false-diffusion. However placing a manipulator within the flow introduces a wake, the
spreading of which may be over-estimated by the use of such a one-sided differencing
scheme [e.g. see Pollard and Siu (1982)]. Fortunately this effect can be reduced by
using a grid sufficiently fine enough to ensure that central differencing is recovered
when the grid Peclet number is less than 2. Thus for the present computations the
number of grid points was increased to 297 (in x) by 98 (in 1') from the 147x68 mesh
used for the earlier developing pipe flow calculations. The grid \vas further refined
near the wall, 40% of the points being below y+=50 with the first at y+ :::;1. In
addition, for some of the present computations the number of iterations was about
1000 resulting in a factor 20 improvement in normalised residuals for axial and radial
momentum reported earlier. While these additional 400 iterations produced only
negligible differences in other flow quantities (much less than 1%), the pressure field,
particularly in the region downstream of the manipulators, was altered by about 5%.
This change occurred early in the extra iteration count, thus it is believed a fully
converged solution for all variables was obtained.
The manipulators were delineated by 6 control volumes across their thickness,
which was assumed constant at 0.3mm, and 20 control volumes along their length,
so that at both the leading and trailing edge the first grid point away from the device
was located at approximately one wall unit from its surface. In addition care was
taken to ensure that the control volumes bordering the manipulators were suitably
modified to account for the presence of the blockage. (Note that all estimates in wall
units are based on the value of U T for the undisturbed fully developed pipe flow).
The wall boundary conditions used in the present calculations were the same
as those defined by Martinuzzi and Pollard (1989a,b). The inlet conditions required
some modification in order to achieve conformity with the mean axial velocity profiles
measured for the initial plane, un-manipulated pipe flow. The "best fit" agreement
was found when a standard log-law (constants, A=2.,5, B=5.5) was specified from
the wall to y+=250, with a linear sublayer below y+=l1.5, and a constant velocity
between y+=250 and the pipe centre-line. In the absence of any data, the inlet
conditions for the turbulence kinetic energy and its dissipation rate were assigned
uniform values across the pipe radius: k=0.002 U~verage and f=Cl'k 3 / 2 /O.02R. Typical
computations for 1000 iterations on the 297x98 grid required 137 CPU minutes on
an IBM 3081G.
COMPUTATIONAL RESULTS
Increasing the grid resolution resulted in a clear improvement in the model per-
formance, even for the plane or unmanipulated pipe flow, as illustrated by the com-
parison of predicted and experimental Cp (== 1 f'j2- P , ) distributions for the initial
2 P average
28
developing flow region presented in Figure 2. The resulting Cp development, with and
without a single manipulator introduced into the nearly developed flow at x=28D, is
shown in Figure 3. These computations, performed over 600 iterations, reveal some
discrepancies with the experimental data for the more developed region of the plane
pipe flow. This was still evident, but somewhat diminished, after the additional
400 iterations were performed to ensure more complete convergence on the pressure
field. A further indication of the performance improvement with grid refinement
for the un-manipulated case is provided by the kinetic energy profiles at X=30D
plotted in Figure 4. It can be seen that the higher resolution computations are in
good agreement with experiment bearing in mind the uncertainty involved in both
of these.
'-Vith the same refined mesh the computed mean velocity defect (manipulated mi-
nus un-manipulated U) close behind a single manipulator introduced into the nearly
developed flow at X=28D from the pipe inlet (leading edge of device at X=27.3D)
were also in considerably better agreement with the experimental data than previous
lower-resolution computations of Pollard et al. (1989) - see Figure 5. The lack of
significant improvement in the calculations for X=10.4D is quite probably due to a
too rapid spreding of the wake; however, the error in the experimental data, obtained
with a Pi tot tube, can also be exagerating the discrepancy. The corresponding delta
Cp (manipulated minus un-manipulated Cp ) distribution is plotted on Figure 6 and
is also in respectable agreement with the experimental results (initial 147x68 grid
computations overpredicted the delta Cp to the extent that the results would lie just
above the domain covered by the figure). The dashed curve in Figure 6 indicates the
somewhat improved results obtained after 1000 iterations as compared with those
after only 600 iterations (solid curve). Again both computed curves fit the exper-
imental data within the experimental and numerical uncertainties (note that the
experiments also indicated that the Cp development was unchanged when ReD was
increased by a factor of 2-3). In addition the computations reproduced the trend of
all the experimental results up to X=50D, except in the initial region immediately
downstream of the device. A similar discrepancy is evident in high-Re k-E computa-
tions for external boundary layers performed by Coustols et al. (1987), and must be
attributed to defficiencies in the modelling of this region with such closure schemes.
The quality of the predictions obtained with the highest grid resolution/number
of iterations is better appreciated when these are replotted on a different scale for
comparison with similar experimental and computational results for an equivalent
tandem maipulator configuration (one with the first plate at the same location as
for the single device just considered and with the same parameters for both plates,
but s=2.7R), as in Figure 7. However it is clear that the tandem configuration is not
so well predicted, and that the computations indicate the effect of introducing the
second plate is simply to double the pressure drop due to the single plate manipula-
tor, presumably due to the increased blockage (in both cases only about half of the
pressure drop due to the device appears to be recovered), whereas the experiments
29
show that the penalty associated with the second device is less than for the first. In
fact the pressure was clearly still relaxing even at the end of the test section in the
tandem case (65R downstream of the device) where the .6.C p had fallen to the level
attained behind the single plate [see Pollard et al. (1989b)]. For the latter it ap-
peared recovery was completed within 40R, a similar recovery length to that found by
the N AL researchers. Unfortunately, storage limitations (8Mb) restricted the extent
of the flow development that could be considered in the computations and so it was
not possible to discover if the trend of the experimental results in the fully developed
section of the flow could be reproduced by these although the indication from the
shape of the curves presented in Figure 7 is that this was unlikely. The reason for
such a discrepancy is unclear at present, but is probably partly related to the well-
known defficiency of current transport models, based on local equilibrium closure
approximations, that they tend to recover too rapidly to equilibrium conditions and
thus underpredict the overall effect of disturbances imposed on flows [c.g. see Savill
(1987)]. However the fact that, for devices introduced into the initial developing flow
region, recovery seems to have been completed within 60R, or approximately 2008
(a distance comparable with that found for external boundary layer manipulation
by similar device configurations and reproduced by model computations - see Fig-
ure 8), together with fact that the expected shape of the turbulence profiles in the
immediate vicinity of the device were captured well by the calculations (see Figure
9), suggests there is some additional problem with the physical modelling of the
manipulator influence in the fully developed flow situation. It has previously been
generally thought that the lack of an entrainment surface under these conditions
might preclude any nett drag reduction (and also perhaps simplify the modelling),
because the important suppression of such outer layer effects seen in external bound-
ary layers would be missing. However, at least for fully developed duct flow, the
experimental studies of Antonia and Teitel (1989) and others have shown that this
external influence is replaced by an equally significant interaction between the flow
in the two halves of the duct; large scale motions from one side moving across the
centre-line to disturb the flow on the opposite side wall.
It is evident from an analysis of the initial low resolution 32x16x16 Large Eddy
Simulation results obtained at T.U. Miinchen by Friedrich & Klein (1989a) that ma-
nipulators not only influence the flow structures in the near wall region, but also
in the core flow; both reducing the scale of these and suppressing any cross-flow
interactions for some distance behind the devices [see Savill & Djenidi (1990)]. This
might help explain why the N AL researchers obtained their best result for a device
mounted on the centre-line of their duct. However it is significant that their next best
result was for two single plates mounted at h=0.38R. Considering all of the experi-
mental/LES data and computational results obtained so far it would seem that the
optimum height for essentially developed flow conditions may be of order h=0.25R.
It is also evident, from the different best results obtained at Laval (whole device
drag recovered), N AL (device drag almost recovered) and ETHZ (approximately half
30
device drag recovered) using similar height single plate devices, with progressively
smaller l/R, that a manipulator chord length of at least I.25R (the value adopted
for the most detailed LES study) is required for optimum performance, and further
model computations are now being performed to evaluate the benefits of such longer
devices [Pollard ct al. (1990)]. Regarding the question of the optimum spacing it
seems clear from the comparison of the tandem results from N AL (nett drag twice
that for single device) and ETHZ (minimum nett drag approximately less than or
equal to single device for s=2.7R, but greater than single for s=0.7R), that simply
re-scaling the optimum value found for external applications by the initial devel-
oping wall layer thickness, or by half the duct or pipe diameter, is insufficient to
maximise the benefits of tandem manipulators for internal flows. Instead a much
closer non-dimensional spacing of around 3R or 8 appears to be much more bene-
ficial. Howe\u if one studies the predicted C f (== 1 ~~'all ) distributions for both
2 P average
single and tandem de\'ices (which were not measured in the experiments) as plotted
in Figure 10 severa1 points emerge. First considering the whole C f development for
the case of a single manipulator it is evident that, while the downstream C f assymp-
totes to the Blasius profile va1ue as expected, this is actually a higher ("overshoot")
va1ue than that predicted immediately upstream of the device. Furthermore when
the C f distribution in the immediate downstream vicinity of the single manipulator
is examined in greater detail and the equivalent distribution computed when the
second plate of the tandem with s=I.35D is superimposed (see (0) curve on Figure
10, lower figure) it is clear that there is an upstream influence of this second plate
which pre\'ents the minimum C f due to the first plate alone from being attained
before the effect of the second plate (leading edge at X=28.65D) is felt by the flow.
Clearly the tandem configutaion could be improved if the second plate were shifted
downstream. To test this suggestion some additional computations have now been
performed for a similar plate configuration, but with the leading edge of the second
plate at first X=28.9D, and then X=29.4D. The results are presented in Figure II.
Assuming that minimising the overall C f is the most important criteria for achieving
optimum nett drag performance, these suggest that a spacing of s=2D (or 4R) can
be recommended. In the present case this corresponds to 101 which is similar to the
external flow value, but is not clear whether this is significant since this condition
was nearly satisfied for the N AL tandem experiments also. It seems more likely that
s should also be scaled on D or R to maintain a close physical presence of the two
plates particularly as the immediate effect of the device on Cp persists for only 0.6R
(as opposed to approximately 58 for an equivalent device: 1=0.48, h=0.28 t=0 ..5mm
in external flow) and the maximum C f reduction of 5% occurs only about 3R from
the de\'ice (as opposed to 1.5% at 88), although the recovery lengths of 40R and 508
are comparahle.
CONCLUDING REMARKS
The results of the present work can be summarised as follows.
31
New computational and experimental results have been presented for turbulent
pipe flow manipulated by both single and tandem ring devices. It appears that
the low-Re k-c model employed is sufficient to produce predictions in reasonable ac-
cord with the experiments or at least equivalent to the level of agreement achieved
previously in external flows. The results, when considered together with earlier com-
putational and experimental results including data obtained elsewhere, suggest it is
unlikely that any nett benefit could be obtained in developing pipe flow, but that
this may still be possible in fully developed flow given an optimum tandem manip-
ulator geometry and sufficient development length in which to recover the device
drag. In particular it would appear that (when scaled in terms of R rather than
8) the non-dimensional height and spacing of the devices both need to be reduced
relative to their optimum values for external flow operation, although a similar,
non-dimensional, chord length may be appropriate. For the best present case the
development length both upstream and downstream of the manipulators was in-
sufficient to determine whether any nett benefit could be obtained, but there was
little difference in the slope of either the measured or computed longitudinal pres-
sure gradients indicating that any drag reduction was less than the experimental or
numerical uncertainty. It is hoped that further improvements to the understanding,
modelling and optimisation of internal flow manipulation will come from a more
detailed analysis of higher resolution 64x32x32 and 192x32x32 LES 'data' currently
being performed by Savill it et al. (1990).
ACKNOWLEDGEMENTS
Professor Pollard acknowledges the financial support of the Natural Science and
Engineering Research Council of Canada, Queen's University Computing Centre and
The School of Graduate Studies of Queen's University. Dr.Savill acknowledges the
continuing support and interest of Rolls-Royce pIc. The experimental work was
unfunded and performed by students under the supervision of Professor Thomann
at ETH, Zurich. Additional support and computing resources were provided by the
EPFL ERCOFTAC Pilot Centre in Lausanne, Switzerland as part of a collaborative
ERCOFTAC project, and these are also gratefully acknowledged.
REFERENCES
R.A. Antonia and M. Teitel (1989): Organised motion in a fully developed tur-
bulent duct flow, Proceedings, 10th Australasian Fluid Mechanics Conference, Mel-
bourne, Decemder.
S.P. Bardakhanov, V.V. Kozlov and V.V. Larichkin (1989): Influence of an acous-
tic field on the flow structure behind a LEBU in a turbulent boundary layer. Poster
Paper at IUTAM Symposium on Turbulence Structure and Drag Reduction, Zurich.
F. Chen, Y-P. Tang and M-Z. Chen (1990): An experimental investigation of loss
reduction with riblets on cascade blade surfaces and isolated aerofoils. Proceedings
ASME Turbo Expo - Land, Sea & Air, Brussels.
32
K.-Y. Chien (1982): Predictions of channel and boundary layer flows with a low
Reynolds number turbulence model. AIAA J. 20, p.33-38.
E. Coustols, J. Cousteix and J. Belanger (1987a): Drag reduction performance on
riblet surfaces and through outer layer manipulators. Proceeding RAeS International
Conference on Turbulent Drag Reduction By Passive Means. 2., pp.250-289.
E. Coustols, C. Tenaud and J. Cousteix (1987b): Manipulation of turbulent
boundary layers in zero pressure-gradient flows: detailed experiments and modelling.
Proceedings 6th Symposium on Turbulent Shear Flows, Toulouse, Paper 11-5:1-5.
E. Coustols and A.M. Savill (1988): Resume of important new results presented at
the Third Turbulent Drag Reduction Working Party, ONERA/Chatillon, September
29-30. Appl.Sci.Res. 46, pp. 183-196.
E. Coustols, A.M. Savill and T.V.Truong (1990): Report on the 4th European
Drag Reduction Working Party, Lausanne PC. ERCOFTAC Bulletin IV, p.8-12.
J. Dickinson, V.D. Nguyen, Y. Chalifour, A. Smaili, A. Page and F. Paquet
(1989): Boundary layers over a ribleted surface using tandem turbulence manipula-
tors. Proceedings 1st Canadian Symposium on Aerodynamics, Ottawa.
A. Dinkelacker, P. Nitsche-Kowsky and W.-E Reif (1987): On the possibility of
drag reduction with the help of longitudinal ridges in the walls, Proceedings IUTAM
Symposium on Turbulence Management and Relaminarisation, Bangalore, (Springer-
Verlag) pp. 109-120.
L. Djenidi, A.lVI. Savill and L.C. Squire (1990): Parametric computational study
of aerofoil manipulator performance in laminar flow. For Proceedings Volume 5th
European Drag Reduction Working Party Meeting, BMT London (Kluwer Academic
Publishers; Ed: K.-S.Choi).
1. Djenidi, C. Tenaud, E. Coustols and A.M. Savill (1990): Comparison of com-
putations with experiments for aerofoil manipulators in turbulent flow. For Proceed-
ings Volume 5th European Drag Reduction Working Party Meeting, BMT London
(Kluwer Academic Publishers; Ed: K.-S.Choi).
R. Friedrich and H. Klein (1989a): Large scale turbulence structures in a manip-
ulated channel flow. Proceedings IUTAM Symposium on Turbulence Structure and
Drag Reduction, Zurich. Springer-Verlag.
R. Friedrich and H. Klein (1990): Manipulating large scale turbulence in a chan-
nel. Private communication.
S. Fu, B.E. Launder and D.P. Tselepidakis (1987): Accommodating the effects
of high strain rates in modelling the pressure-strain correlation. U11IST Mech. Eng.
Dept. Report TFD /87/5.
A. Johnston (1987): Private communication.
J.S. Kelman, 1. Shearer and J. Hall (1989): Riblet surface pressure losses in single
and two phase flow. See Sellin & Moses (1989).
H. Klein and R. Friedrich (1989b): Manipulating large-scale turbulence in a
channel and a boundary layer. Proceedings 7th Turbulent Shear Flows Symposium,
Stanford p.25.4.1-6.
33
C.K.G. Lam and K. Bremhorst (1981): A modified form of the k-E model for
predicting wall turbulence, ASME, Journal of Fluids Engineering, 103, pp. 456-460.
B.E. Launder and B.I. Sharma (1974): Application of the energy dissipation
model of turbulence to the calculation of flow near a spinning disc. Letters in Heat
and mass Transfer, 1, p.131-138.
J. Lemay, A. Pollard, J.P. Bonnet and A.M. Savill (1990): International co-
operative research programme in the area of turbulent boundary layer manipulation.
7th Canandian Conference on Engineering Education, Toronto, June 27-28.
B. Lineton (1988): Unpublished ETHZ data.
K.N. Liu, C. Christodoulou, O. Riccius and D.D. Joseph (1989): Drag reduction
in pipes lined with riblets. Proceedings, 2nd IUTAM Symposium on Structure of
Turbulence and Drag Reduction. Zurich, Switzerland, July (Springer- Verlag).
M.V. Lowson, A.D. Jewson and J.H.J. Bates (1989): Drag reduction by riblets
in a duct flow. See Sellin & Moses (1989).
R. Martinuzzi and A. Pollard (1989): A comparative study of turbulence models
in predicting turbulent pipe flow. Part I: algebraic stress and k-E models, Part II:
Reynolds stress and k-E models. AIAA Journal, 27, No.1, pp. 29-:36 and 27, No.
12, pp. 1714-172l.
A. Pollard, A.IVI. Savill and H. Thomann (1989a): Turbulent pipe flow manip-
ulation: some experimental and computational results for single manipulator rings,
Journal of Applied Scientific Research, 46, No.3, July, pp. 281- 290.
A. Pollard, A ..M. Savill and H. Thomann (1989b): Turbulent pipe flow manipu-
lation: some experimental and computational results for tandem manipulator rings.
Proceedings 10th Australasian Fluid Mechanics Conference, Melbourne. Vol. II paper
9C.3
A. Pollard and A.L-W Siu (1982): The calculation of some laminar flows us-
ing various discretisation schemes, Computer Methods in Applied l'vIechanics and
Engineering, 37i, pp. 293-313.
A. Pollard, H. Thomann and A.M. Savill (1990): Further computational results
for manipulated pipe flow. Proceedings Volume 5th European Drag Reduction \Nork-
ing Party Meeting, BMT London (Kluwer Academic Publishers; Ed: K.-S.Choi).
1. Pompeo and T. Matievic (1987a): Turbulenzbeeinflussung, Semesterarbeit in
Fluiddynamik, ETH Zurich, Institut fur Aerodynamik Report WS 1986/87.
L. Pompeo and T. Matievic (1987b): Investigation of drag reduction in manipu-
lated turbulent boundary layers. ETHZ Institut fur Aerodynamik Internal Report.
A. Prabhu, P. Kailas Nath, R.S. Kulkarni and R. Narasimha (1987): Blade ma-
nipulators in channel flows. Proceedings IUTAM Symposium on Turbulence Man-
agement and Relaminarisation. Bangalore, Springer-Verlag, pp. 97- 108.
L.W. Reidy and G.W. Anderson (1988): Drag reduction for external and internal
boudary layers using riblets and polymers, AIAA-88-0138, 26 Aerospace Sciences
Meeting, Reno, Nevada.
J.J. Rohr, L.W. Reidy and G.W. Anderson (1989): An experimental investigation
of the drag reducing benefits of riblets and polymers in pipes. See Sellin & Moses
(1989).
K. Samec (1989): Investigation of drag reduction in manipulated developing pipe
flow. ETHZ Institut fur Fluiddynamik Internal Report.
A.M. Savill (1987): Algebraic and Reynolds stress modelling of manipulated
boundary layers including effect of free-stream turbulence. RAeS International Con-
ference on Turbulent Drag Reduction by Passive Means, 1, pp. 89-143.
A.lIt Savill (1988): Structural refinements to Reynolds stress closures for plane,
manipulated and wake-boundary layer flows. Proceedings, 3rd International Sympo-
sium on Refined Flow 110delling and Turbulence Measurements, Tokyo (Universal
Academy Press Inc.).
A.:\I. Savill (1989): Recent developments in turbulent drag reduction by pas-
sive means. Rcview lecture paper for the 2nd. IUTAM Symposium on Structure
of Turbulence and Drag Reduction, Zurich, Switzerland, July, to be published by
Springer- Verlag.
A.M. Savill and L. Djenidi (1990): Lausanne ERCOFTAC Pilot Centre Visitors
Report.
A.:\1. Savill, T.V. Truong and 1.L. Rhyming (1988): Turbulent drag reduction by
passive means: a review and report on the first European drag reduction meeting.
Journal de lIIecanique 1 (4), pp. 353-378.
A.:\1. Savill, H. Klein and R. Friedrich (1990): Pattern recognition analysis of
structure in manipulated channel flow simulations. For Proceedings Volume 5th
European Drag Reduction Working Party Meeting, BMT London (Kluwer Academic
Publishers; Ed: K.-S. Choi).
R.J.Scllin and R.T. rvloses (1989): Drag Reduction In Fluid Flows: techniques for
friction control. Proceeding, IAHR International Conference DRAG REDUCTION
89 on the Fundamentals and Applications of Drag Reduction Techniques: Additives
and Passive Devices. Davos, Switzerland, (Ellis Horwood).
S. Shengxi and Zhou. 1!J.D. (1989): The influence of plate manipulator wake on
skin friction. (Unpublished Report from Nanjing Aeronautical Institute, China)
C. Tenaud (1988): Unpublished ONERAjCERT computational results.
B. Vasudevan, A. Prabhu and R. Narasimha (1990): Blade manipulators do not
reduce pressure losses in turbulent duct flow. To appear.
S.P. Wilkinson, J.B. Anders, B.S. Lazos and D.M. Bushnell (1987): Turbulent
drag reduction research at NASA-Langley: Progress and plans. In: Turbulent Drag
Reduction by Passive Means, Royal Aero. Society, Sept.
35
u R r
0t--------------------------
I,
G ';
-0,1
-0.2 \
\
.,
-0,
12 16 20 24 28
X/D
0.1
0-
C>
-0.1
-0.2
-0.5 o Experiment
-0.6
-0.7
+ Calculation Pipe Flow
Figure 3: Cp vs. X/D for pipe flow with single manipulator with leading edge at
X/D=27.3, r/R=O.8, Re=180,OOO.
+ +
"+
"+
"
"
o Experimental data (Laufer) (Fully developed)
o
o
<>
Calculation 68 * 147
+ Calculation 98 267
20 40 60 BO 100
0.03 , - - - - - - - - - - - - - - - - - - - - - - ,
I~ ~:~i ::t=:--+=::-=:::-:==-+-----~-_-_-
~ -OO~ +--'-------=--:...------~-\---.,~---.::j.
<l =~:~5
-0.04
-0.05
-0.06 X"/D~O.4
-0.07
-0.08
-0.09 +------::-'C,----~--~-~-~-+-_--1
0.2 0.4 0.6 0.8
0.025
0.02
0.015
-~--
0.01
0.005 f--- ----
0
~
-0.005
-0.01 X"/D~2.4
-0.015
-0.02
0.2 0.4 O.S 0.8
0.02
0.018
0.016
0.014
0.012
0.01
0.008
0.006
0.004
.
0.002
-0.002
-0.004
0
X"/D=10A
---- ~
-0.006
0.2 0.4 0.6 0.8
r/R
~
0.021
"'-
. ,,+
0.02 Experimental Data
0.019
~ 0.018 1 0.7
0.017 -+- l ' O.B
5 0.016
0.015 +
'\ c o l ' 0.9
~ 0.014 c \
0- 0.013
u 0.012
.3 0.011
0.01
'"" 0.009
0.008
+---r--r:--
l + .B3 Ii
0.007
0.006 Calculations (1 O.B)
0.005
0.004
0.003 600 Iterations
0.002
0.001 1000 Iterations
0
30 31 32.5 35.5 38.5 40.5 42.5 44.5 46.5 48.550.7554.75 57 60
X/D
Figure 6: Cp difference downstream of single manipulator with leading edge at
X/D=27.3, r /R=O. 7, O.S and 0.9. (-) Calculations 600 iterations, (- -) Calculations,
1000 iterations.
39
0.07
o +-
0.05
Single at X/D = 27.3 -
~
:5
0.05 !:. Tandem at X/D = 27.3 & 28.65 - 0 -
;;;
0.04-
:5
~ 0.03
A
u
0.02
-<l
0.01
28 32 36 40
44 X/D 48
0.0< , - - - - - - - - - - -_ _ _ _ _,
rlpHlmental
0.035 o hO.7
olxO.8
~
, 0.03 -+hO.9
1.21:0.8
~ 0.025 _ _ 1:rO.7Ca!cs
:5 0.02
GO.015
~ 0.01
0.005
1-1-00C25
Figure 8: Cp difference for single manipulator at either r/R=0.7, 0.8 or 0.9 and
tandem manipulators at r/R=0.8 (axial separation of 1.35D). Comparison of calcu-
lations with data for single manipulator at r/R=0.7
o X/D=4.900
o XjD::5.100
6. X/D=5.201
X X/D:::5.250
I
~ I -=i=i I ,I
1
j
0.6 0.7 0.8 0.9
r/R
Figure 9: Radial turbulence kinetic energy distribution prior to, on and after a
single manipulator with leading edge at X/D=5, r/R=0.7, Re=180,000.
40
4.50E-03
Cf 4.40E-03
4.30E-03
4.20E-03
4.10E-03
4.00E-03 Blasius Cf=0.00384
3.90E-03
3.80E-03
3.70E-03
3.60E-03
3.50E-03
HOE-03
20 40
4.50E-03
Cf 4.40E-03
4.30E-03
4.20E-03
4.10E-03
4.00E-03
3.90E-03
3.80E-03
3.70E-03
3.60E-03
3.50E-03
HOE-03
27 27.4 27.8 28.2 28.6 29 29.4 29.8
X/D
Figure 10: C J distribution vs. X/D. Top figure - single manipulator at X/D=27.3.
Bottom figure - Window of 27 ::; X/D ::; 30 (D) single manipulator, (< tandem
manipulator. Leading edges at X/D=27.3 and 28.65, r/R=0.8, Re=180,000.
0.0043
Cf
0.0042 I.
0.0041
0.004
0.0039
0.0038
0.0037
0.0036
Ii
0.0035
20 24 28 32 36 40 44 48
X/D
Figure 11: C J predictions for single and two different tandem arrangements. (-
+-) single, leading edge at X/D=27.3; (< Tandem, leading edges at X/D=27.3
and 28.9; (- - -) Tandem, leading edges at X/D=27.3 and 29.4. All for r /R=0.8,
Re=180,000.
Large-Eddy Simulation of Manipulated Boundary Layer and Channel
Flows
Abstract
The large-eddy simulation (LES) technique is used to study boundary layer and
channel turbulence at high Reynolds numbers altered with flat plate manipulators.
The simulations demonstrate the effects of such manipulators on the instanta-
neous turbulence structures as well as on the statistically averaged flow field. The
time-dependent and three-dimensional filtered :\a,"ier-Stokes equations for an in-
compressible fluid are solved. A two-part eddy-viscosity model takes into account
the non-resolved subgrid-scale effects. Characteristics of the numerical method are
a leapfrog-scheme for explicit time integration, the projective method and the ca-
pacitance matrix technique to solve the Poisson equation for the pressure. First,
the results of a zero-pressure-gradient boundary layer flow are discussed. The ma-
nipulated boundary layer (1\1BL) is compared with the standard boundary layer
(SBL) without any manipulator under the same inflow conditions. Secondly, the
comparison of a manipulated and standard channel flow, abbre,"iated :\ICF and
SCF, respecti,"ely, is presented. Both the results of the MBL and l\ICF exhibit a
decrease in the turbulence intensity. Correspondingly, coherent structures of the
instantaneous flow field are suppressed in both fiow types by the manipulators.
The results of the IvIBL and MeF indicate no fundamental differences concerning
the effects of fiat plate manipulators. This beha,"iour can be explained by the fact
that the manipulators act in regions where the turbulent structures in both flows
are wry similar.
NOlnenclature
Symbols
~A.j grid cell surface
cf = 1 ;;;:~ , skin-friction coefficient
Ci convection term
Di diffusion term
Dij deformation tensor
E;.o SGS kinetic energy (isotropic part)
Fr total drag force
41
42
h channel half-width
h characteristic mesh-width
hm wall-distance of manipulator
1m length of manipulator
linh characteristic length of inhomogeneous SGS model
lmix mixing length
p pressure
Re = v,,! I re ! Reynolds number
VreJ ' ...
t. !:::"t time, time step
bulk flow velocity
= ( <T~> ) ~ , friction velocity
free-stream velocity
grid volume
Vi,U,l'.W velocity components
Xi, X, y. Z coordinate directions (longitudinal, spanwise, vertical)
Zwall nearest wall-distance in z-direction
5 boundary layer thickness
52 momentum thickness
5ij Kronecker symbol
Tij = ~e D ij , shear-stress tensor
Tw \vall shear-stress
J' J.1iso isotropic SGS eddy-viscosity
ji J.1inh inhomogeneous SGS eddy-viscosity
~ distance downstream of manipulator trailing edge
Definitions
<9> statistical mean value
rjy" (Reynolds') fluctuating value
F-
9 grid volume average
j,
CJ surface average, taken oYer !:::"Aj
9' deviation from GS quantity
Subscripts
inh inhomogeneous part of the SGS model
~so isotropic part of the SGS model
Introduction
The dynamics of turbulent flows is not at all dominated by random motions.
In definition of the phenomenon "turbulence" coherent structures and coherent
motions play an important role. \Vhile studying wall bounded flows so-called wall
events can be observed. There is for example the burst-process involYed in the
breakup of a single longitudinal streak structure (Bogard et al. 1987). As a con-
sequence of a burst, low-speed fluid is carried away from the \\'all region \vhich is
called an ejection. In contrast, sweeps are characterized by the rapid transport of
high-speed fluid towards the wall. Ejections as well as sweeps contribute about
43
70% to the total Reynolds stress (Landahl et a1. 1986) and represent an impor-
tant mechanism of producing surface drag in turbulent boundary layers (Corke et
a1. 1980). Furthermore, free or wall bounded turbulent shear flows contain large-
scale coherent structures (Fiedler 1987). They dominate the turbulent transport,
contain most of the turbulence energy and consequently control the mixing rate,
noise-production and -emission and vibration phenomena. As far as boundary
layer flows are concerned, some of these large-scale structures propagate into the
outer part of the layer and cause the entrainment process (Coustols et a1. 1986.
Tennekes et a1. 19(2). The speed, at which the interface between the turbulent
and nonturbulent fluid with the large-scale structures moves into the irrotational
region, controls the entrainment (Tennekes et a1. 19(2). In addition, fluid from
the irrotational region penetrates into the boundary layer (Praturi et a1. 19(8),
which explains the intermittent character of the outer part of a boundary layer
flow. Coustols et a1. (1986) summarize that turbulent structures are related to a
cycle of movements featuring near wall events (e.g. ejections, sweeps) and outer
flow e,'ents (e.g. entrainment).
It is pointed out in several experimental studies (Plesniak et a1. 1985, Savill et
a1. 19S5, Anders et a1. 19S5) that flat plate manipulators are a means of reducing
the drag in turbulent boundary layers at low Reynolds numbers. A computational
investigation with various statistical turbulence models (Coustols et a1. 1989) pro-
duces similar results but is, of course, not able to show the modification of instan-
taneous flow structures. High Reynolds number experiments (Reo, >S, 000) on an
axisymmetric body in the NASA Langley Tow Tank resulted in no net drag reduc-
tion by positioning tandem airfoil manipulators in the outer part of the boundary
layer (\Valsh et a1. 19S9). Concerning the manipulation of turbulent internal flows
drag reduction is not found at all. Prabhu et a1. (19SS) found such results in the
case of a turbulent channel flow at low Reynolds number (Reh =24,500). Equally,
recent experimental and computational studies of Pollard et a1. (19S9) give the
conclusion that single ring manipulators do not reduce the drag in pipe flows at
high Reynolds number (ReD=IS0, 000).
Finally, the results of experimental studies (Savill et a1. 19S5, Savill 1987,
Guezennec et a1. 1990, Lemay et a1. 19S9) in boundary layers show that flat plate
manipulators change the flow field in several ways. The mechanisms of how ma-
nipulators work are classified into "plate" and "wake" effects, compare table l.
The "plate" effects include the alteration of the flow field in the immediate "icin-
ity of a plate. Due to the presence of the manipulator all velocity fluctuations
are suppressed, in particular wall normal fluctuations (pI). The destruction of
the large-scale structures is understood as the break-up of large-eddies (p2). In
consequence of the flow displacement in a boundary layer the circulation around
the plate produces lift and induces downward velocities in the wake according to
Biot-Savart's law. This phenomenon is called the "downwashing" of the plate \vake
(p3). The "wake" effects determine the flow modifications due to the wake of a ma-
nipulator. New energetic small scales are introduced into the base-flow immediate
downstream of the manipulators. These small scales promote an enhanced energy
cascade from existing larger scales (wI) (Savill et a1. 19S5). Furthermore, the wake
acts as a shield that prevents incursions of high-speed potential fluid and blocks
44
the interaction between the outer and near wall part of the flow (w2) (Coustols et
al. 1986, Savill et al. 1988). Both the "plate" and "wake" effects are responsible
for the skin-friction reduction (Savill 1987). With reference to the maximum drag
reduction, it is important to point out that the minimum local skin-friction occurs
close to the position where the wake reaches the wall. This explains why the dis-
tance between manipulator and wall is an important parameter. In a boundary
layer it should vary inversely with the Reynolds number (Savill et al. 1988).
The subgrid-scale (SGS) quantities represent the deviations from the GS quantities:
(2)
Using a cartesian grid the filtered equations haye the following non-dimensional
form:
(3)
aF'iji .. j-_.
-at- + oJ (J'ijJ'ij
J t
+ v.lv'
J t
+Jp-C -)~ .. ) -
U)l I JI -
0 (4)
45
where
j- j-
.
JJ'r- -Re
- D 1"
1 j- jD .. _ OVj + OVi
( 5a, b)
J' -
OXi OXj
(6)
j- j-,-,
Tji,SGS = - VjVi , (7)
(8)
111
linh
--~+--
C2.inh h Imix
, (10c)
(11 )
This model contains a fluctuating part (9) which considers the locally isotropic
SGS effects and a statistical part (10) that takes into account inhomogeneities
46
near walls and manipulators. The isotropic SGS energy of relation (9c) results
from equating the production and dissipation terms in the transport equation for
v-
E'isa. For more details of the SGS model see Schumann (1975) and Friedrich
(1988). The factors of the isotropic model (-Yl=0.93, 12=1.0, C31=0.74) are the
same as those used by Schmitt et al. (1987) to simulate the fiow O\-er a backward
facing step. The grid dependent constants are computed by assuming the the-
ory of locally isotropic turbulence and the validity of the Kolmogorov spectrum
(Schumann 1975, Schmitt 1988). For the grid used in this study the following
values are obtained: c2=0.069, c2o=0.83, c3=0.595, xC5 =Y C5 =0.785, zC5=0.939,
xX c=YY c=1.589, zZc=1.435, xY c=0.981, xZ c=yz c=0.747. The constant of the inho-
mogeneous model is chosen as, C2,inh=0.1.
The isotropic SGS energy amounts about 18% to the total turbulence energy
for the used grid and Reynolds number. The statistical part of the model becomes
effective near walls and manipulators and contributes to an enhanced momentum
transport.
The system of equations (3),(4) is solved on a staggered grid so that one may
assume vVi ~;v; and jp ~vp. All GS quantities in equation (4), which are not
defined, are approximated by algebraic averages, i.e.
(12)
(13 )
(14)
Fortin's version of the projection method is used in (13) which first neglects the
pressure-gradient. To update the velocity V;+l, P has to be determined from a
Poisson equation which results from equation (14) and (3):
solved directly. B is "similar" to A in the sense that these two n X n-matrices are
equal except for a few mn rows. These m equations belong to the mesh cells
with internal boundaries. Correspondingly, A, Band q are decomposed into
q= GJ
(AI,B I : mxn-matrices, A 2 : (n-m)xn-matrix, qI vector of length m, q2 vector of
length (n-m)). We get the solution of the A-problem with the help of the m x m-
capacitance matrix C and the nxm-'selection'-matrix lV
C = .h B- 1 W W= (~)
(I: mxm-unit matrix, 0: (n - m)xm-zero matrix) by applying the steps:
Bp=q (160)
Cv = w q=q+Wv (16b, c. d. e)
Bp=q (16f)
(U). w. v: ,'ectors of length m, p. q: vectors of length n). That means we obtain the
desired solution at the cost of solving two B-problems (16a,f) and a m x m algebraic
system of equations (16d). The capacity matrices (one for each decoupled plane)
depend only on the irregular geometry and are pre-computed at the start of a
simulation by solving m B-problems to determine one matrix.
Boundary Conditions
The specification of proper inflow boundary conditions is very costly in the
present case of spatially developing flows. At each grid point of the entrance plane
the LESs of ~dBL and MCF need the instantaneous GS velocity vector and the
SGS turbulence energy at each time step of the simulation process. Separate LESs
provide this data and allow storing of all relevant quantities from a plane normal to
the flow direction at each time step. In the case of a channel a fully dewloped flow
(FDCF) ,,ith periodic boundary conditions in the main flow direction and in the
case of a boundary layer a zero-pres sure-gradient flow (ZPGBL) with inflow/out-
flow boundary conditions, described in Richter et al. (1987), have to be simulated.
The Reynolds number, grid spacing, time step and SGS model are the same as in
the cases of the manipulated flow.
A special treatment of the "elocity wctor in the outflow plane has prowd suc-
cessfully. It consists in a linear extrapolation of the mean streamwise component
and a constant extrapolation of the remaining two mean components, but in soh'-
ing a conwction equation for all the velocity fluctuations using < u > as proper
transport velocity. For more details, see Richter et al. (1987).
The velocity component normal to solid walls is set to zero. Resolution restric-
tions require approximate boundary conditions tangential to walls as Piomelli et
48
al. (1989) propose for flows of engineering interest. By means of Schumann's (1975)
condition the instantaneous wall shear-stress is in phase with the instantaneous
velocity in the logarithmic layer. This assumption is supported by experimental
investigations (Piomelli et al. 1989). The same concept is applied at surfaces of
manipulators. The authors are aware that the validity of the logarithmic law is
questionable there. However, the inviscid damping of vertical velocity fluctuations
as the primary effect of manipulators is considered correct.
In the case of the boundary layer flow we specify Dirichlet conditions at the top
of the computational domain for the tangential velocity components and the SGS
turbulence energy. The normal velocity is obtained from the continuity equation.
'Ve use periodic boundary conditions in both flow types in the spanwise direc-
tion.
Results
Three different LESs were performed for boundary layer as well as for channel
flows.
l.) Two different LESs serve to produce inflow boundary condi tions for both flows.
a) Boundary layer flow:
LES of a zero-pres sure-gradient boundary layer flow (ZPGBL) with inflow /
outflow boundary conditions. Some sort of structural periodicity (Richter
et al. 1987) is assumed in this case to generate inflow conditions.
b) Channel flow:
LES of a fully developed channel flow (FDCF) with periodic boundary
conditions in the main flow direction.
2.) LESs of the manipulated boundary layer (MBL) and manipulated channel flow
(MCF) with inflow boundary conditions from l.)
3.) LESs of the standard boundary layer (SBL) and standard channel flow (SCF)
with the same inflow conditions as in 2.). "Standard" refers to unmanipulated
flows.
I
o
50
The ZPGBL was computed with 36,000 time steps. Inflow data of 12,000 time
steps for the MBL and SBL were stored on magnetic tape. Consequently the
simulations of the MBL and SBL covered 12,000 time steps out of which 6,000
are needed for statistics. Besides time averaging spatial averaging is used in the
homogeneous y-direction.
Figures 1a,b give a qualitative impression of how a single flat plate manipulator
affects a boundary layer flow. The contour surfaces are plotted for high \'alues of
the instantaneous (-u" w" )-correlation. The Reynolds stress mainly results from
these structures. In the SBL the (-u" w")-structures are uniformly distributed
over the whole computational domain (fig.1a). Figure 1b shows the suppressing
effect of the manipulator. Unlike the portion of the flow field upstream of the
manipulator, there is no similarity in the MBL and SBL downstream of the plate.
The manipulator destroys larger structures and breaks them up into smaller ones.
8
I I I10 12
SC"LEI - - - 8 -4.000E+OO
I
z I I
I
I I
)
e
'='
)
J
=
tc;
J
El
,
.::l
= ~
,
8 10 12 I ~ 16 18 20 22 X 24
The rms-values of the longitudinal velocity fluctuations clearly reflect the sup-
pressing-effect of the manipulator, compare fig.4. The solid line refers to the ma-
nipCllated and the dashed line to the standard boundary layer, respectively. \Vhile
a slight influence is found immediate downstream of the manipulator at x=48
(~=0.58), the turbulence intensity is considerably reduced at x=98 (~=5.58). It
is of interest that the velocity fluctuations are reduced mainly between z=hm and
the wall. The measurements of Coustols et al. (19S6,19S9) confirm these results.
Further downstream the urms-values recover gradually and reach almost the level
of the SBL at x=198 (~=15.58).
SCALE' - - - 6.000[1'00
2 10 12 14 16 1B 20 22 X 24
Q .... .. ..........
.. . . . . .. . .
. . . . . . . . .......... ..
. . . .. .. .. .. .. .. .. .. .. .
.... . ...... .
....
.. .. .. .. .. .. .. .. .. ...
,
...... . .........
~~~~~~~~ll~;ll~~~~~j~~
tttfHU1HU~llU}lll
2 3 5 6 7
x
t
o 2 3
x
t
2 3
1
< <
I
i
S'l s
SIH-Z
S I H-Z SiX\,-Z
part. The SGS kinetic energy is a measure of the absolute value of the fluctuating
deformation tensor. It shows that the edge of the boundary layer is not affected
by the manipulator, compare plate Id.
Figures 6a,b compare fluctuating quantities in a plane perpendicular to the main
flow direction. The plane lies slightly downstream of the manipulator trailing edge.
The contour-lines of the vertical velocity fluctuations show the break-up of large-
scale structures into smaller ones. The wake of the plate can be made out from
those fluctuating "elocity-vectors which are aligned with the y-direction. They are
located at the height of the manipulator, h m =O.31258.
The development of the momentum thickness 82 for the MBL compared to that
for the SBL in fig.7a reveals the typical characteristics. The increase in the range
of 2.2::; x::; 3.7 is induced by the presence of the manipulator. Further downstream.
for ~ ;:: 2.58, the slightly smaller slope for 82 in the Iv1BL indicates the tendency
to-wards a skin-friction reduction.
f
:':
~
;i.1
>=<
I
:::;
I
:: I'"
I~ ~
- SIlL
- ' ,illL
I
10 15 20 X
~ M I
b ~~~~~~_~
i-~L ~
i - . 1illL I
I
o :1 ========================================
10 15 20 x
Fig.ib Comparison of skin-friction coefficient in the SBL and MBL.
A true skin-friction reduction for ~;:: 2.58 is reflected in the shape of the skin-
friction coefficient C f in fig.7b. The reduction of C f immediately upstream of the
plate is a consequence of flow deceleration. Consequently, the following increase at
the manipulator location is due to flow acceleration. The maximum skin-friction
reduction of about 11.5% is found at ~ = 138. The cf-distribution shows a skin-
friction reduction until the end of the computational domain. The total drag
force F,. results from integrating the shear stress oYer the boundary layer wall and
the two manipulator surfaces. F,. does not giYe a net drag reduction by using a
computational length of 248 (~=20.58). The ratio of F,. in the ~1BL to that in the
55
SBL drops from the yalue 1.66 for the short computational domain down to 1.16
for the longer domain:
FTMBL = 1.66 ---+ 1.1 6 .
F TSBL
::
- SDL
- MDL
.---. MUL 2
X 8
M
o
~ /~ I
~ I
- SDL
- MDL
<>-----<> MOL 2
X 8
This fact gives support to the conclusion of Savill et a1. (1988) that the optimum
position of a manipulator should be close to the wall at high Reynolds numbers.
For comparison with the experimental data of Coustols et a1. (1988) and Tenaud
(1988) a manipulated boundary layer at low Reynolds number REb2 ::::2500 was
simulated. The length of the manipulator is 0.758 for the simulation and 0.78 in the
experiment. The downstream development of the skin-friction coefficient related
to the coefficient of the nonmanipulated boundary layer is in good agreement with
the measurements, compare fig.9. However, the maximum skin-friction reduction is
56
1. 05
1. 00 _._._.-._._._._._.-
-'-'-'-'
o
.95 o 0
o
o
.90
.e5~----------------------------~------------
-5 10 15 20 ~ 25
somewhat larger in the simulation. In addition, the simulation reflects the increase
of the skin-friction at the location of the manipulator.
Channel flow
The size of the computational domain in a (x,y,z )-coordinate system is agam
24x4x2 in the long channel case and 8x4x2 in the short channel case. It is
measured in terms of the channel half-width h. Just as for the boundary layer flow
a 192 x 32 x 32 and 64 x 32 x 32 equidistant grid has been choosen. The Reynolds
number based on h and the friction velocity Ur,l at the inflow plane is 3240, based
on the channel width 2h and the bulk velocity Ub is 150,000. All quantities are
non-dimensionalized with hand Ur,l' By imaging the channel flow as the result
of two merged boundary layer flows, it is natural to group two manipulator plates
symmetrically. They are fixed at a distance 2h downstream of the inflow plane.
The geometrical data of one plate are the same as in the boundary layer case. Thus
the wall-distance h m of such an infinitely thin plate amounts to 0.3125h and the
length lm to 1.5h.
The LES for the FDCF was performed with 38,000 time steps in order to
generate inflow data. The ~\'fCF and SCF were simulated with 17,000 time steps
out of which 12,000 were used for statistical averaging besides averagmg in the
homogeneous y-direction.
~ ---~ fI 4.U-4E+01
I Ii I I I
"
z
0 6 8 10 12 14 16 18 20 22 X 24
Fig.10 gi\'es an impression of the manipulated flow field in terms of the mean
longitudinal velocity component. One can make out the wakes of the manipulators.
57
which spread and weaken similarly to those of the MBL while developing down-
stream. The effect of the manipulators becomes more obvious from plotting the
momentum deficit 6.<U>=<U>MCF-<U>SCF, compare fig.lI. The momentum
deficit of the wakes shifts in the channel as well as in the boundary layer flow to-
wards the wall. This leads to the conclusion that the" downwashing" -phenomenon
exists also in the channel. As a consequence of continuity the flow is accelerated
in the core region. Thus, the velocity profile becomes peakier in the manipulated
case and leads to a reduced wall shear-stress.
o
11
8 10 12 1~ 16 18 20 22 X 24
~.
(
/
\, \,
~
o
~> '~
o 6 B 10 12 14 16 18 20 22 X 24
:.:::::1 ~U:::":~<J)~I
~r"::;f~'
I:~. ~~I
Ihl~~~:1
N N
..0
passed the manipulator while the fluctuating velocity-vectors have been aligned in
the direction of the plate. Again 100 time steps later the structure has travelled
further downstream, but the strong fluctuations in the vicinity of the ,vall have
quite completely disappeared. The effects of suppression and destruction of large-
scale structures by the manipulators become also evident from the structure present
in the lower channel half. \Vhile travelling downstream it is split into two parts by
the plate and remains devided.
The time-development of spanwise velocity fluctuations in terms of contour-
lines is shown in figures 14a,b. Among the three velocity components, v" most
perfectly reveals the typical inclination of large-scale structures with respect to
the walls. In fig.14a a structure which travells downstream in the SCF is marked
with arrows. By contrast, in the MCF the same structure has been deformed and
broken up by the lower manipulator (fig.14b).
J0 20 x
Fig.lS Comparison of wall shear-stress in the SCF and MCF.
Finally, the downstream development of the wall shear-stress Twin the l\1CF
is compared with that in the SCF, see fig.15. The reduced Tw slightly upstream
of the manipulators is a consequence of flow deceleration, as the following increase
in T w is due to the flow acceleration between manipulator and ,vall. The following
decrease in T w for ~ ;::: 2.5h leads to a maximum wall shear-stress reduction of
11 % at ~ = 13.5h. Although T w remains below its SCF -value until the end of the
computational domain (~= 20.5h) no net drag reduction is found. The ratio of the
total drag force Fr in the MCF to that in the SCF decreases from a value of 1.67
for the short to 1.16 for the long computational domain:
By comparing this result with that in the MBL no difference can be made out
concerning the drag behaviour of both manipulated flow types.
Summarizing discussion
The presented results show that flat plate manipulators act in similar ways in
boundary layers and in channels though there are differences between both flow
types. Table 2 contains correspondences and differences between these flows. Since
both are bounded by walls they produce wall ewnts, such as ejections and sweeps.
61
Both flows also contain large-scale coherent structures which are characteristic of
any shear layer. The entrainment process does not exist in a channel. The reason
is the missing free-stream boundary. Last not least the fully developed channel
flow exhibits no mean flow displacement.
Whitelaw,J.H. (eds.): Turbulent Shear Flows 6, Selected Papers from the 6th
Int. Symp. on Turb. Shear Flows, Toulouse, France, Sept. 7-9, 1987, Springer Ver-
lag, pp. 179-193, 1989.
Plesniak,M.W., Nagib,H.M., Net Drag Reduction in Turbulent Boundary Lay-
ers Resulting from Optimized Manipulation. AIAA-85-0518, AlA A Shear Flow
Control Conference, Boulder, Colorado, March 12-14, 1985.
Pollard,A., Savill,A.M., Thomann,H., Turbulent pipe flow manipulation: some
experimental and computational results for single manipulator rings. - In: Sav-
ill,A.M. (ed.): Applied Scientific Research, Vo1.46, No.3. Special issue: Drag Re-
duction Applications of Riblets and Manipulators, pp. 281-290, 1989.
Prabhu,A., Vasudevan,B., Kailasnath,P., Kulkarni,R.S., Narasimha,R., Blade
manipulators in channel flow. - In: Liepmann,H.W., Narasimha,R. (eds.): Turbu-
lence Management and Relaminarisation, IUTAM Symp. Bangalore, India, 1987,
Springer-Verlag, pp. 97-107, 1988.
Praturi,A.K., Brodkey,R.S., A stereoscopic visual study of coherent structures
in turbulent shear flow. J.Fluid Mech., vo1.89, part 2, pp. 251-272, 1978.
Richter,K., Friedrich,R., Schmitt,L., Large-eddy simulation of turbulent wall
boundary layers with pressure gradient. Proc. of the 6th Symp. on Turb. Shear
Flows, Toulouse, France, Sept. 7-9, 1987.
Savill,A.M., On the Manner in which Outer Layer Disturbances Affect Tur-
bulent Boundary Layer Skin Friction. - In: Comte-Bellot,G., Mathieu,J. (eds.):
Advances in Turbulence. Proc.of the First European Turbulence Conference, Lyon,
France, 1-4 July 1986. Springer-Verlag, pp. 533-545, 1987.
Savill,A.M., Mumford,J.C., Manipulation of turbulent boundary layers by outer-
layer devices: skin-friction and flow-visualization results. J.Fluid Mech., vol. 191,
pp. 389-418, 1988.
Schmitt,L., Grobstruktursimulation turbulenter Grenzschicht-, Kanal- und Stu-
fenstromungen. Diss., Lehrstuhl fur Stromungsmechanik, TU Munchen, 1988.
Schmitt,L., Friedrich,R., Application of the large eddy simulation technique to
turbulent backward facing step flow. Proc. of the 6th Symp. on Turb. Shear Flows,
Toulouse, France, Sept. 7-9, 1987.
Schmitt,L., Friedrich,R., Large-eddy simulation of turbulent backward facing
step flow. - In: Deville,M. (ed.): Proc. of the 7th GAMM-Conf. on Numerical
Methods in Fluid Mechanics, Louvain-La-Neuve, Belgium, Sept. 9-11, 1987. Notes
on Numerical Fluid Mechanics, Vol. 20, Vieweg, pp. 355-362, 1988.
Schumann,U., Subgrid scale model for finite difference simulations of turbulent
flows in plain channels and annuli. J.Comp. Phys.18, pp. 376-404, 1975.
Schumann,U., Fast elliptic solvers and their application in fluid dynamics. -
In: Kollmann,W. (ed.): Compo Fluid Dynamics. Hemisphere, pp. 402-430, 1980.
Tenaud,C., Simulation Numerique de l'Ecoulement autour d'un Manipulateur
Externe de Couche Limite. These Docteur, l'Ecole Nationale Superieure de l'Aero-
nautique et de l'Espace, Toulouse, 1988.
Tennekes,H., Lumley,J.L., A First Course in Turbulence. The MIT Press, 1972.
\Valsh,M.J., Anders,J.B., Jr., Riblet/LEBU research at NASA Langley. - In:
Savill,A.M. (ed.): Applied Scientific Research, Vo1.46 , No.3. Special issue: Drag
Reduction Applications of Riblets and Manipulators, pp. 255-263, 1989.
64
by
Arild Bertelrud
High Technology Corporation, Hampton, Virginia, U.S.A.
and
FFA, The Aeronautical Research Institute of Sweden, Bromma,
Sweden
SUMMARY
It is proposed that a well organized vortex street behind the device is a necessary
requirement for turbulent boundary layer drag reduction with outer-layer
manipulators.
The communication from the outer region to the wall is inhibited through the
existence of the vortex layer. The necessary wake structure is obtained with an
open laminar separation on the front device. At low Reynolds numbers most flat
plates and airfoils can provide this. At high Reynolds numbers different shapes are
required, since conventional airfoils with sharp trailing edges produce a turbulent
wake when the Reynolds numbers are increased. For tandem devices, the two
airfoils shed vortices at slightly different frequencies, causing development of a
strong, low-frequency component in the transitional wake.
OUTER-LAYER MANIPULATORS
At low Reynolds number several experiments have shown that local and overall
skin friction reduction is possible for turbulent boundary layers manipulated by
so-called LEBU's. These are flat plates or airfoils in single or tandem arrangement
in the outer part of the boundary layer. Low Reynolds number experiments
performed at 'host' boundary layer Rea of below 8000 gave 8-15 % net drag
reduction for an optimum tandem device at y/8=0.8, L 2 8, gap25 8. Due to the
high drag of a plate with finite thickness and its tendency to flutter, it was assumed
that an efficient device would be a minimum drag airfoil at practical Reynolds
numbers.
At high Reynolds numbers the limited experiments performed with airfoils so far,
have shown small local and no overall drag reduction 1. This has been attributed to
a scale mismatch between device and turbulence structure and/or a difference in
turbulence structure at high Reynolds number, Rea , of the 'host' boundary layer. A
sensitivity to minute geoemetric changes has also been noted by several authors.
In the present paper flow changes over the airfoil and in the wake due to Reynolds
number variation and device shape are emphasized.
67
68
ANALYSIS
Several mechanisms have been proposed for the influence outer-layer
manipulators 26 : for example a plate effect, i.e. reducing the overturning of the
large vortices in the outer part of the boundary layer, and a wake effect where the
downstream wake has beneficial effects on the turbulence structure. In the present
paper we observe that an increase in Reynolds number based on the 'host'
boundary layer momentum thickness also leads to a completely different flow
pattern ON the device AND IN ITS WAKE. The wake structure may have a large
influence on the downstream development of the turbulence structure.
There are several important aspects in the discussion of how and why a particular
outer-layer manipulator is suitable for high Reynolds number conditions:
- 'Host' boundary layer structure
- Organized, laminar vortex shedding
- Slightly different shedding frequency from the rear device
- Moderate device drag.
If a direct interaction between the LEBU wake and the boundary layer turbulence
is responsible for the drag reduction, the devices must be mounted close to the
wall at high Reynolds numbers. If the communication loss is the responsible
mechanism, the required changes in location inside the boundary layer with
Reynolds are very small. Up to y+=100 Andreopolous' data showed a bursting
period T B U / 8 independent of Reynolds number (i.e. outer scaling), while the
burst period more than doubled for y+> 100 and scaling with outer variables was
not possible.
To have the drag reducing effect lasting over any useful distance, the wake must
have a structure allowing it to survive far downstream. It is turbulent over the major
part of the downstream region, and longitudinal vortices dominate the structure of
the far wake. The role of the manipulator is to set up this vortex system with a
minimum of device drag.
A modified laminar flow airfoil would be suitable, but it is possible to obtain the
open laminar separation with a traditional NACA profile turned backwards, as
suggested in Ref. 6. The drag as function of angle of attack? for a NACA 0012
airfoil mounted traditionally and reversed are shown in Figure 2 at Re.c=321 ,000 ,
i.e. at Reynolds numbers appropriate for flight. The drag coefficient is higher for
the reversed airfoil, but similar to the drag coefficient at a somewhat lower
Reynolds number for the traditionally mounted airfoil. This has been illustrated in
Figure 3, where drag measurements with NACA0012 airfoils have been compared
with data from the literature 8 ,9. The NACA0012 reversed is comparable to a flat
plate of 2% thickness. There is a consistent shift in drag coefficient with free stream
turbulence level, and one may expect that locating an airfoil inside a boundary
layer has an effect also on the drag. Also an increase in velocity will have an effect
- simple compressibility rules yield that a NACA0012 in incompressible flow has a
pressure distribution equivalent to a NACA0009 in high subsonic flow (Mach = 0.6
- 0.7). At transonic Mach numbers one may expect that open laminar separations
can create sharp drag increases.
For LEBU applications flat plates and airfoils are of interest. The Strouhal number
selected by the flow depends on the thickness, t, of the fluid layer11 involved in
the vortex street formation, and will vary depending on the type of device flow:
-Airfoil with open laminar separation: t = 2 . ( h + 8*) if we neglect the trailing edge
thickness. The 'effective thickness' is determined exclusively by the displacement
thickness, 8*, and the separation height, h, at the trailing edge.The trailing edge
thickness is negligible. However local geometrical changes on the airfoil can
cause a change in separation position or transition pattern for the wake.
Figure 4 gives a variety of values when plates of rounded leading and trailing
edges 12 -14 are included. For airfoils the data falls in two categories: laminar and
turbulent as will be discussed later. Both Bauer16 and Liandrat 16 have found
Strouhal numbers in excess of 0.6 along the laminar curve when the maximum
thickness is used as basis. Similar numbers are found by Paterson et.aI. 18 , Arbey
and Bataille 19 and others along the turbulent curve.
S = xsepfL
The integral W is a result of the inciscid flow set up around the airfoil, and since it
depends on the pressure distribution and separation location, it is to a first
approximation independent of Reynolds number.
The shedding frequency fsh for flow around bluff bodies 11 is fsh t / U = 0.16. For
an airfoil the frequency of shedding fsh from each side is given by21:
fs h=0.08 . U / t (3)
Y= tan- 1( B / Res.s ep ) where we use the fit B= - 5.736 + 3.704 In(ReS.s ep ) (5)
The parameter B is usually assumed constant with a value of 15 to 20. The fit
chosen here is based on results by Bandyopadhyay17 and Liandrat 16 with two
different NACA0009 airfoils. Choice of a constant B would change the computed
curves slightly.
The displacement thickness is obtained directly from the momentum thickness at
separation using the shape parameter at separation, Hsep = 3.55 and assuming
that it does not change over the separated area. This is a reasonably assumption
since the pressure gradient is almost zero over the region back to the trailing edge.
(6)
j(
_ 2 1 - ~sep) B + Hsep )
- t- -
8 sep (-.lL).
Uinf sep
W
72
0.16 . H
<p= 3LL
(1 - ~sep) 8 H iW (7)
(-.1L)
Uinf. sep
YW + sep
Transitional/turbulent flow
From Figure 5, it is clear that the cases with open laminar separation are well
described with the simple equations. As the Reynolds number increases, the
shedding frequency drops and in a region it is possible to obtain several values of
the Strouhal number. The existence of a bifurcation point due to an absolute
acoustic instability in the wake has been predicted by Koch22 for flow behind flat
plates with blunt trailing edges. In the present case, it is reasonable to regard the
separated airfoil case as similar, since the separated area acts as a bluff body at
the trailing egde. This corresponds to the sudden jump in Bauer's result for the
NACA 0012 airfoil, and the knee in the drag polar of the airfoil.
*
following a correlation curve given by Paterson et.aL 18, which for our purpose is
expressed:
s = 0.011 YRed (8)
As discussed by Arbey and Bataille 19 , the actual shape of the curve is governed
by an acoustic feedback loop producing a ladder-type behaviour associated with
laminar flow on the pressure side of airfoils.
APPLICATION TO MANIPULATORS
If the manipulator geometry is sharp edged plates or ribbons, a laminar wake will
be formed immediately behind the front (or single) device as long as the Reynolds
number based on device length is well below the transition Reynolds number.
Airfoils tested so far have been limited to NACA0009 and related shapes. They
have an open laminar separation only at low Reynolds numbers, and will not work
as the Reynolds number is increased or if they are moved inward in the boundary
layer where the turbulence intensity and structure can cause premature transition
in the free shear layer of the airfoil.
73
Although the wake behind the front device initially is laminar, the first phase of
transition will develop rather rapidly as described by Sato and Kuriku 12 and
Miksad et.aI. 23 . In the first, short region, the linear instability region, higher
harmonics to the fundamental start to develop. The maximum amplitude is initially
concentrated to the vertical positions corresponding to the highest mean velocity
gradient. Rather soon the amplitude spreads more evenly through the wake, but
this does not mean the flow is turbulent. The transition region is rather long 12 , and
is still dominated by distinct 'carrier' frequencies and their sidebands. It is essential
that the rear airfoil is located in the deterministic, transitional part of the wake.
While the higher harmonics soon dissipate and lose their identity in the turbulent
wake, the low difference frequency, f2-f 1, is capable of setting up the strong vortex
pattern that is essential for the turbulence manipulation.
REFERENCES
4. Badri Narayanan, M.A., Raghu, K. and Poddar, K.: 'Wall shear fluctuations in
a turbulent boundary layer.' AIAA Journal ,Vo122 No, pp. 1336-1337
75
6. Bertelrud,A.: ' A profile family for use in boundary layers with large-eddy
breakup devices." DRAG REDUCTION '89, Davos, Switzerland, July 31-August 3,
1989
10. Taneda, S.: ' Oscillations of the wake behind a flat plate parallel to the flow. '
J. Phys. Soc. Japan, 13 (4) ,418-425 (1958)
11. Levi, E.: ' A universal Strouhal law.' J. Engineering Mechanics, ASCE, Vol.
109, No.3, June 1983, pp. 718-727.
12. Sato,H. & Kuriki,K.: ' The mechanism of transition in the wake of a thin flat
plate placed parallel to a uniform flow.' J. Fluid Mech, 11, 321-352 (1961)
15. Bauer,A.B.: ' Vortex shedding from thin flat plates parallel to the free stream. '
J. Aero. Sci., 28, 340-341 (1961)
18. Paterson, R.W., Vogt, P.G., Fink, M.R. and Munch, C.L.: ' Vortex noise of
isolated airfoils.' J. Aircraft, May 1973, pp. 296-302.
19. Arbey, H. and Bataille, J.: ' Noise generated by airfoil profiles placed in a
uniform laminar flow.' J. Fluid Mech, (1983), vol 134, pp. 34-37
24. Meier, G.E.A. and Timm, R.: 'Unsteady airfoil vortex interaction.' AGARD
CP-386. (1985)
Forced
ShC<lr~Layer Possible ~airing l::1
C____ --=-- ? ~? ~ ? ~ ? ~ C_____
----
~
????~$~~=- FT
~~--.:>-~~v_=_ ~
0.06
C NACA0012
NACA 0012 Reversed
0.05
l/
0.04
vt;
0.03 1/
CD l/
0.02
V
t--- ji....
"
1,..-1,..-
L.--
0.01
0.00
-5-4-3-2-1012345
Alpha [deg]
78
0.02 0
0 NACA0009 (Llandrat)
CD
1:..
0.01 o~
0.00
1 03
RE.c
Figure 4 5trouhal number 5=f dIU for NACA 0009 & 0012 airfoils.
Flat plate results are shown for comparison.
1.0
Comp.-Laminar
Comp.turbulent
0.8 I. D Bandyopadhyay
o Liandrat
I
Bauer
) Arbey & Batailie
0.6 I Lawaczeck & Kreplin
5TROUHAL
"
I,. Sharp T.E.
NUMBER ti
0.4
J1 .. !,: a.
RoundT.E.
I:.~ '&~
0.2
~~
./
0.0
100 1000 10000 100000 1000000
RE.d
BOUNDARY LAYER FLOW VISUALISATION PATTERNS ON A RIBLET
SURFACE
Abstract
Boundary layer flow visualisation methods, developed at
Queen Mary and Westfield College, have been applied to a
riblet surface. The results reveal cellular cross flows
developing in the grooves between the riblets. These local
flow regimes appear to have little direct effect on the flow
in the wall layers immediately adjacent to them.
Qualitatively, the behavior of the wall layers appears to be
that which would be expected if a virtual surface existed at
a level slightly above the riblet tops, but a tendency for
the origin of longitudinal eddy pairs to become anchored to
the top of a riblet is noted.
1. Introduction
Experiments with flow visualization in a turbulent
boundary layer have been in progress in the Department of
Aeronautical Engineering at Queen Mary and Westfield College
for a number of years (Refs. 1 & 2). Oil droplet smoke is
used, with plane beams to give cross sections of the
resulting smoke patterns. Such techniques are now familiar,
but the light beams used at Q.M.W. are not produced by the
more usual method of spreading a laser beam, but by shining
a focused beam, from either tungsten halogen projector lamps
or standard photographic flash units, through narrow slit
masks. The original aim of these experiments was to find a
less expensive alternative to laser light.
As a preliminary to a proposed programme of studies,
tests were undertaken to establish whether these lighting
methods could be successfully applied to a riblet surface.
Though the tests were only exploratory, the photographs
obtained were unexpectedly informative.
79
80
particularly clear.
Raising the smoke wire to 7mm above the riblet tops
produces a marked change in the type of pattern produced.
An example is shown in Fig. 14. The smoke tends to remain
clear of the riblets for some considerable distance. At
that level the direct effect of the rib lets is apparently
less important than the larger scale movements taking place
in the wall layers immediately above. The uneven nature of
the smoke sheet in this view is due to the oil on the smoke
wire collecting into globules, an effect which is difficult
to avoid with certainty.
Moving the view across, as before, to include part of
the plain surface, again shows the difference between the
two flow regimes. The patterns which form on the plain
surface have a wavelike appearance, rolling over to form
streamwise scrolls in some instances but by no means always.
Figs. 15 & 16 show a typical examples. A comparison with
the general form of the pattern in Fig. 14, in which the
smoke wire was approximately one riblet height above the
rib let tops reveals similarities , giving further support to
the idea that the normal flow patterns of the wall layers
are not greatly altered by the presence of the riblets, but
that they are simply displaced from the surface.
6. Conclusions
Considerable activity directly associated with the
presence of the riblets, and consisting of cellular
cross-flows, exists in the region within about 1.2-1.5
riblet heights from the surface. These movements, wi thin
what might be termed a "riblet sublayer", appear to have
little direct effect on the activity in the adjacent wall
layers, but they are strongly influenced by the nature of
the motions in the adjoining flow.
The patterns which appear tend to have the same general
form as those which can be seen over a plain surface, but
the level from which recognisable features appear to
originate, is displaced from the layer immediately adjacent
to the surface, to a region just above the riblet tops.
These results are consistent with the view that one of
the primary effects of the riblets is to prevent inrushes of
high speed fluid from spreading out laterally close the
surface. Such movement is displaced away from the wall by a
distance somewhat greater than the height of the riblets,
and the ability of the high speed fluid to penetrate close
to the basic surface is strongly inhibited.
88
7. Acknowledgements
7. Wilkinson, S.P. and Lazos, B.S. "Direct Drag and Hot Wire
Measurements on Thin Element Riblet Arrays"
Turbulence Management and Relaminarisation, IUTAM
symposium Bangalor 1987 (Springer-Verlag 1988)
8. Van Driest, E.R. "On turbulent flow near a wall"
J Aeronaut. Sci. 23, pp.1007-1011 & 1036.
9. Coles, D. and Hirst, E.A. "Computation of turbulent
boundary layers" AFOSR-IFP-Stanford Conf. 2 (1968)
89
Smofte '-.titre
/X/-COYr'\era
/'
~
Ri.blet Geometr-!:I
20~----------------~----------------~----------~------,
Wall Law
u* (Van Driest)
10~----------------~~--------------~----------------~
Pi tot tube A
Hot Wire +
oL----------------L________________L -______________ ~
o 1 2 3
Log(y*)
~'
, Fla.=e",,-_,.
Abstract
In this paper we present some results obtained from simultaneous LDA and flow
visualization measurements over several types of longitudinal, microgrooved
surfaces. No new structures were detected above the grooved walls apart from
those that exist over the smooth walls. The measurements indicate subtle
differences in flow over the smooth and grooved surfaces. In particular the shear
stress distribution in the low speed streak seems affected. These differences are of
great interest because the shape of the riblets used effects a decrease in drag
compared to the smooth wall drag (Walsh 1982).
1. Introduction.
97
98
activity above the grooved plate. One would expect from these results that the
turbulence intensity normal to the wall should decrease and this is not supported
by measurements of turbulent intensity mentioned earlier. However the effect of
reduced sweep activity on the turbulence intensity is not that clear.
Since the work presented in this paper was done, a very detailed study on the
structure of a turbulent boundary layer over a riblet surface was published by
Choi (1989). The work included measurements of velocity, turbulence intensities,
skin friction apart from fairly extensive flow visualization photographs over
smooth and riblet surfaces (Choi also provides an up-to-date list of references on
the subject). The VITA technique applied to the wall friction signals showed that
the frequency of the near-wall bursts over the riblet surface was nearly eight
times that over the smooth plate. However the duration of the burst was reduced
by a factor of two. In addition the turbulent statistics for the two surfaces were
different. The skewness and kurtosis of the skin-friction signal over the riblet
surface was 1.31 and 6.82 respectively compared to 0.42 and 3.12 over the smooth
surface.
In the water channel a pulsed hydrogen bubble flow visualization facility is used.
Images were recorded with a standard video recording system. Processing these
video images with picture processing techniques enables us to obtain quantitative
streamwise velocity measurements, i.e. the velocity of the hydrogen bubbles.
Additionally measurements of the instantaneous streamwise and normal velocity
components were taken with a laser doppler anemometer (LDA). The LDA
measurement volume was located at 0.7 0.1 mm below the hydrogen bubble
generating wire and 1.0 0.2 mm upstream of it. As the LDA, the wire and the
video camera were mounted on the same traversing mechanism their relative
positions remained fixed during the experiments.
Four surfaces were studied during this investigation: one smooth (UU), two
longitudinal grooved (SA and SS) and one transverse rough (GG) surface. The
dimensions and the shape of the surfaces are shown in fig 1. The measurements
~!MA ~ll
~j ~
. ,.
~~G ---:-::-15.7=-----
ss 5.0
The visualization experiments were made with the wire at right angles to the free
stream and parallel to the wall. The video images are obtained by photographing
a single hydrogen bubble line a few hundred milliseconds after its release by the
bubble generating wire. The position of the line is measured automatically by
computer processing and provides a measurement of the streamwise velocity
component as a function of the spanwise position. Comparison with the LDA data
showed a, good correlation between the flow velocity calculated from the position
of the bubble line and the flow velocity measured with the LDA (Pulles 1989). No
correlation exists between the local turbulent intensity or the local shear stress as
measured with the LDA and the velocity calculated from the position of the
bubble line.
The method is especially suited for measuring the spanwise correlation of the
streamwise velocity component and provides an estimate of the average spanwise
extent of flow structures. Since four released bubble lines are visible in the video
pictures one can observe the structure of the flow. The velocity measurements are
100
made on the bubble line closest to the wire (or the line that was released last in
the sequence of four). The signal of the LDA during this period gives additional
information, in particular on the vertical transport and the instantaneous
Reynolds shear stress which are not readily visible in the video frames. With this
setup two sets of data were obtained. One set consists of 500 velocity profile
measurements, recorded at four different heights above the four surfaces and two
main flow speeds. The total set thus comprises of 16000 velocity profiles.
Simultaneously with the profile measurements the two velocity components as
registered by the LDA were recorded. A second set consists of six times 500
velocity profile measurements, with the LDA measurements and six times 500
fragments of video film showing the movement of the bubble lines, during the
measurements. These were obtained on three different surfaces (the smooth, one
longitudinal grooved (SS) and the spanwise grooved surface) at two different
heights.
3. Results.
First the results derived from the first data set are presented and discussed. An
obvious quantity that can be calculated from the velocity profiles is the average
50~-------- __________________ ~
40
30
20
10
O~~-r~~4--~~6-.~~8--r-1~0-'~12
if
SO,---------------------------~b
40
30
20
10
OO~--c-~-4r-~-6r-~~8--~~10--~12
u+
FIGURE 2. Correlation lengths obtained by jitting an exponential curve to the
observerved average correlation function. a:Urn = 95 mm/s,. b: Urn = 140 mm/s. 0 :
smooth plate, UU,. + : longitudinal grooved plate, SA,. 0 : longitudinal grooved
plate, 55,. t, : spanwise grooved plate, GG.
101
The results suggest a decrease in length scale near the wall . This is in agreement
with the measurement of Gallagher & Thomas (1984). However an increase in
length scale is observed in the region of the higher velocities, presumably above
the buffer layer. This is to be expected since the lengthscale near the wall must be
50
-20 -10 o I. 10 20 30 mm
50
~~~
Jt;-~ss 0 J\~A A~~
___
v '\r'-JV :v SA
'---::-2~0---~,0'---O~z-.~'0-+~2L-O----L30:-m-'m -20 -,0 o z.'O 20 30mm
FIGURE 3.Average velpcity versus distance to the nearest low speed streak. U00 =
95 mm/s, Ylda = 1.0 mm ~ 5 y*. SS andSA: longitudinal grooves; UU: smooth,'
GG: transverse grooves.
strongly determined by the distance between the grooves (12 and 24 viscous units
at 95 mmls main flow speed, 17 and 34 at 140 mmls main flow speed), which is
smaller than the distance between the low speed streaks that determine the
lengthscale higher up in the boundary layer (100 viscous units).
Figure 3 shows the average streamwise and vertical velocity components versus
102
the distance to the nearest low speed streak. These can be extracted from
thepresent measurements as follows. Firstly the position of the most prominent
low speed streak in the velocity profile as measured by the hydrogen bubble line is
determined. Then the velocity components measured with the LDA at its fixed
position are filed against the distance from the low speed streak identified earlier.
This operation was carried out for the 500 profiles over the four surfaces for one
vertical position of the bubble-wire LDA assembly. The figure was obtained by
averaging over the 500 measurements. The figure shows that above the smooth
wall there exists an average upwards velocity component at the position of a low
speed streak, as is to be expected and as is also found by other authors (Kline et
al 1967). This feature of positive vertical velocity is absent above the longitudinal
grooved wall SA but does persist over the plate SS. However this is in consistent
with the observation that at the speed at which these measurements were done
plate SS shows very little drag reduction (Pulles et a1.1989)
T
T = fJ (U - U) . (V - V) dt (A)
o
Here U and V are the average velocity components averaged over the 500
sequences measured. The averaging time T is 240 ms, which is the period during
which video recordings of the flow were made. In figure 4 the 500 values of T
calculated from the expression (A) are ranked according to the value of their
10~-------------------------- ____________~
"j
..
s. -.: ... ,"",-
"' '. c
b . . . . ----..
.. ~ 20 sequences
min. stress \
max.. stress
-5
'.
-10L-------~------~------~------~-------"
o 100 200 J~O 400 500
N
Reynolds stress. The twenty sequences of the highest and the twenty sequences
with the lowest (most negative) Reynolds stress were selected for a more detailed
study. Figure 4 shows that these account for nearly 30% of the Reynolds stress on
the positive as well as on the negative side. Thus this set of sequences is highly
relevant, despite its small sample size. For comparison twenty random selected
sequences were also analyzed.
c f
FIGURE 5.Examples of structures. In each figure the flow direction is from bottom-
to top. a: longitudinal vortex; b: end of a low speed streak; c: a low speed streak; d:
a wide high speed region; e: an accelerating region; f: a pattern outside
classification.
11. Miscellaneous patterns. Includes patterns like vertical vortices,
105
fe~---------------r-----'----~---r------,
Maximum
I
'1+ =8 I
I
II IV IV
I
I
I
I
I
I
I
f~;:~~::::~::~~~~~::::~~~::::~
random
'1+ =8 I
I
II III IV IV
I
I
I
I
I
I
I
f~~~::~~~::~~~~~::::~~~~~~
mlnlrn..m
'1+ = 8
I
I II III IV
I
I
I
I
I
I
I
2 :I 4 e e 7 e It 10 1f 12
Category rn.mber
1e~---------------r----~------r--r------'
Maxirrun I
y" = 17 I
I I
II III IV IV
I
I
random
y" = 17
I I
I II III IV IV
I I
I I
I
minlmun I
y" = 17 I
I
, 10 :
II III IV IV
I I
I
I
I
I
2 3 4 e II 7 g 10 11 12
Category runber
~ longituQinal grooves
Higher in the boundary layer (y+ = 17) the classification becomes more difficult,
as is evident by the increasing numbers of events assigned to categories 11 and 12.
Above the smooth plate the high speed region is a much observed phenomena at
this height: it occurs in all three selections. So it has apparently no net
contribution to the shear stress. Above the transverse grooved wall the events
with negative shear stress are still the low speed streaks. The longitudinal grooved
wall shows no clear trend, but the events of high and low shear stress seem to be
more evenly distributed over the low speed streak and the high speed regions.
The authors made some further observations looking at the video sequences,
which were difficult to quantify but nevertheless interesting enough to mention
here.
i) The flow over the transverse grooved wall the low speed streaks are visually
dominating. It is tempting to relate this to the high shear stress in this
boundary layer.
ii) The low speed streaks appear to be shorter in length above the longitudinal
plate compared to the streaks above the smooth wall, indicating some rather
subtle interaction between the grooves and the flow.
iii) As the low speed streaks were clearly seen to meander across the crests of
the grooves any straight forward resonance effect is ruled out.
4. Discussion.
Visual observation of the flow revealed no new structures above the two grooved
walls in addition to the ones already found above the smooth wall. An important
question that is not answered during the course of the present work is the
statistical reliability of the results in figures 2 and 6. It is not at all clear that
merely increasing the sample size would provide the requisite information. However
if we accept the differences in counts in figure 6 as indicative of the possible
changes one can expect, i.t follows that the grooves affect both the vertical position
and/or size of the structures and their contribution to the turbulent shear stress.
The absence of vertical flow in the low speed streak above the longitudinal grooved
plate (fig. 3) also points to a different shear stress distribution in a low speed
streak above the grooved wall compared to the same feature above the smooth
wall. This mix of effects, together with the absence of great qualitative changes in
the inner boundary layer could explain the difficulties in reconciling the mean
velocity measurements with the quadrant analysis. In this context it is instructive
to quote Choi: liAs regards possible mechanisms of turbulent drag reduction with
riblets, there may be more than one involved" (1989). Our results show that it is
108
worthwhile to look at the shear stress distribution in the different flow structures
in greater detail.
Acknowledgements.
The research reported in this paper was partially supported by the Netherlands
Foundation for Technical Sciences (STW) as part of the programme of the
Foundation for Fundamental Research on Matter (FOM). Two of the authors
(CJAP & KKP) are grateful for the assistance of Ing. G. Trines, MI. J.C.
Stouthart, II. C. Nieuwvelt and Drs. A.M. Koppius during the course of the
investigation at Eindhoven.
References.
KWING-SO CHOI
British Maritime Technology
1 Walde grave Road
Teddington
Middlesex TWll 8LZ, UK.
Summary
A study of turbulent boundary layers was carried out using the hot-\vire
and film anemometry over smooth and riblet surfaces under different
pressure gradient conditions. Detailed measurements into the near-wall
turbulence structure indicated that the changes in time averaged
turbulence quantities, produced by the addition of riblets, were not
significantly altered by the presence of a longitudinal pressure
gradient. This is in line with the previously reported results on the
near-wall structure obtained by a conditional sampling technique.
1. Introduction
109
110
2. Experimental Set-Up
TABLE 1
H l. 36 l. 38 l. 65
(rnrn) 7.5 13 23
f3 -
Lilli. 0.16 0 3.1
TW dx
dU w
v 0.20 x 10. 6 0.57 x 10. 6
K - Vco 2 ~ 0
III
The mean velocity profiles over the smooth and riblet surface are shown
in Figure 2 in log-law format for three different pressure gradient
conditions (see Table 1). The top two profiles correspond to the
112
30,----~--~--~--
dpJdx < 0
3D
dp/dx:: 0
u
~" 30
dpldx. > 0
1
J
15
Present study !
Smooth socta"
" Albie! surface
i preVi.OUS dato (ChoI, 1989)-
\\\Smooth surface
IIIAlbie! surfan>
o --~---~~-----
o , 2 3 4
Log y.
+ +
u 5.5 log Y + 5.45 (1)
from which the friction velocity u * was derived. The velocity profiles
over the riblet surface were, however, fitted to the following log-law
obtained by Choi (1989) for the experiment using similar riblets:
u
+ U/u * (3)
and y+ y u * Iv (4)
113
It is clear from this figure that the general fit to the log-law is
good for both smooth and riblet cases in all the three pressure
gradient cases tested. It is also noticed that there is an upward
(positive) shift in the intercept of the log-law for the riblet cases,
indicating a drag reduction due to thickening of the viscous sublayer.
Similar shifts are also reported by Hooshmand et al. (1983) over a
riblet surface and with Large Eddy Breakup (LEBU) devices (see
Bandyopadhyay, 1986). This is also a common phenomenon for the
turbulent boundary layer with drag-reducing polymers (see Lumley,
1973). The extent of logarithmic region is, however, smaller in the
adverse pressure gradient case, with a larger extent of wake region.
Strictly speaking, the log-law is only valid in zero pressure gradient
conditions. As Mellor (1966) and McDonald (1969) pointed out, however,
the effects of longitudinal pressure gradients on the log-law are small
as long as the pressure gradient parameters a defined by
*3
a ~ (v/p.u ) . dp/dx (5 )
is small enough. Our value for this parameter in the adverse pressure
gradient case is 0.0196 (see Table 1), which is small enough not to
affect the constants of the log-law. The value for a is even smaller in
the favourable pressure gradient case. A small deviation from the log-
law is, however, expected at large y+ even with a small value of a
1.00 -------,--------,
100
dp/dx =0
11I1f/1I 11 "/l1f
/IIIIII~", '\ ,,,,, \ \\ \\ '\ \
.!.! I,ll, \ '\ '\ \ \ '\ \
U'" 0.75 II,..!.!'\\\\
"
""
050 dp/dx>O
Present study
Smooth surface
025 )( Riblet surface
Previous data (ChOI, 1989)
\\\ Smooth surface
/1/ Riblet surface
0 I
0 25 50 75
0.15,-----....,----r------,
0.15
Pre-sent study
0.05 Smooth surface-
)( Rjbl~t surface-
Previous dato(Choi.1989):
\\\Smooth surface
II/Riblet surface
OL-______l -_ _ _ _ _ _ ~ ______ ~
o 25 50 75
y'
(Mellor, 1966). The comparison with the previous results in the ze:o
pressure gradient condition by Choi (1989) is very good as shown ~n
Figure 2, although the extent of logarithmic region is slightly shorter
in the present case owing to its smaller Reynolds number of Re ~ 2.7 x
10 3 compared with Re ~ 4.6 x 10 3 in Choi (1989).
1. 0 ,---r---,-----,,-----,
Present study
Smooth surface
)( Riblet surface
0.5)(--- __ x dp/dx< 0
L----o~x
1) A , ~o~
~
Xl-~'A ..............
~I
PrevIous dotQ(d. p/dx = 0)
III 0 6.)(~ .0. Kreplin & Eclo:elmann
~ 0.5 0 Ueda & Hmze
.---- ...... a
~
lfl I ~X dp/dx=O
1.0 x I o~x
I"~ A 0
oI I 10--;'
o 5 10 15 20
Smooth surfac~
1
4 0~ )( A,ble-t surface-
30r-, "
I ' _ ..... l(",---
i
\ -e ____! = : - _
5 0' ~,---',-~---'---
~ 4.0 L',
o \ 6 Kreplin & Eckeolmann
DUedo & Hinze
.s ~ \ "- dpldx =0
~ 30
I -""0',D~ 1
0
",
~o ox~
;ol\~ ~-1
40
1\ j
J
dp/dx> 0
I "
o "-
3 0 r'~-~'::~-j-~
20~-~ I
o 10 15 20
y'
Deviation
Figure 7 Figure 9
00 ------------.---~-
dp/dx = 0
Smooth 5urfac~
y' 0 16 RlbI~t 5Urfac~
Probability densities of the
u-component velocity and the
I) 6 wall-skin friction in the
favourable (Figure 7) , zero
(Figure 8) and adverse (Figure 9)
pressure gradient cases. A circle
symbol denotes smooth surface, a
cross denotes riblet surface.
.is
Dashed line represents Gaussian
E
o probability density.
a:
'1 l""_",,,.. 20 40
Deviation
Figure 8
lIS
density are such that the skewness changes its sign from positive to
negative with an increase in the distance from the wall surface. This
is also shown in the skewness profiles (see Figure 5). The positive-
tail probability is largest at y+~7 (closest position to the wall
surface), indicating that large positive excursions of the u-component
velocity signal are the dominant feature in the near-wall region of the
turbulent boundary layer. This is in good agreement with the results of
the quadrant analysis by Wallace et al. (1972) and Willmarth and Lu
(1972). They both showed that the "sweeps", events associated with a
positive u-signal and a negative v-signal, are more important closer to
the wall surface than the "ejections" (u<O and v>O). It is believed
that "near-wall bursts", quasi-periodic events associated with a sharp
increase in wall-skin friction (Choi, 1989), are responsible for this
large positive-tail probability since they are caused by the downwash
of high-momentum fluid between pairs of counter-rotating longitudinal
vortices close to the wall surface. The comparison of the present
results with the probability densities by Choi (1989) for the wall-skin
friction signal in zero pressure gradient condition is excellent for
both smooth and riblet cases. As far as the effects of riblets on the
probability densities in the adverse pressure gradient case are
concerned, they are only noticeable within a near-wall region of the
boundary layer (say, y +<11) although the effects are noticeable up to
+
y ~16 for the other two pressure gradient cases. This is in line with
the results of the skewness and kurtosis (see Figures 5 and 6) as
discussed above.
4. Concluding Remarks
100 ~ y' ~ 11
! y' ~ 7
~ 10 ~ __
I10~'fi
if}
Wall- skin
fnctlOn
~4
10~4
,0
~6L
I
~6
10 10 - - - Smooth sur/eel'"
- Riblet surface
Figure 10 Figure 12
dpld, ~ a
I Spectral densities of the
l
u-component velocity and the
wall-skin friction in the
~
favourable (Figure 10) , zero
(Figure 11) and adverse (Figure
I 12) pressure gradient cases.
f Solid line indicates riblet data.
l Dashed line indicates smooth
I surface data. The sequence of
~
[
figures from
correspond
top to
respectively
bottom
to
+
y ~ 16, 11 and 7 with the bottom
figure being the wall-skin
10.'~! I
l
-- -
--
Smooth surface
Rlblet sur/acE' friction, y+ ~ O.
Drev,ous dato (ChOl.1989)
I \\\ Smooth surfaCE'
10~8L. __ ~ _/~~~~ SUr_f_Q~ __ _ _ ---1-___ _
~
~I
10 10' 10'
~'requency (Hz)
Figure 11
120
These results seem to indicate that the effects of riblets on the near-
wall turbulence structure would not, in general, be altered by the
longitudinal pressure gradients. This conclusion is also supported by
the previously reported results (Choi & Johnson, 1989) using a
conditional sampling technique that the wall-shear stress signatures
during "near-wall bursts" are essentially unaffected by the pressure
gradients. This does not necessarily mean, however, that the net drag
reduction by riblets is unaffected. This is because the performance of
riblets is determined by the balance between the turbulent skin
friction reduction and viscous skin-friction increase. What the present
results seem to suggest is that the turbulent skin friction reduction
due to riblets would not be altered by the pressure gradients.
Acknowledgements
The author would like to express his gratitude to Dr. R. Johnson for
his help in the execution of the experiment and data analysis. He would
also like to extend his acknowledgements to Prof. M. Gaster of
Cambridge University, Mr. M. Firmin and Mr. W. Sawyer of RAE, and
Dr. J.A.B. Wills and Dr. M.E. Davies of BMT for their encouragement and
valuable discussions. This work has been supported by British Aerospace
plc and the Department of Trade and Industry, U.K.
References
Choi, K.-S., Pearcey, H.H. and Savill, A.M. 1987. Test of drag reducing
riblets on a one-third scale racing yacht. Proc. International
Conference on Turbulent Drag Reduction by Passive Means, London.
Choi, K.-S. 1990. Drag-reduction test of riblets using ARE's high speed
buoyancy propelled vehicle - MOBY-D. Aeron J., March, 79.
121
Ueda, H. and Hinze, J.O. 1975. Fine structure turbulence in the wall
region of a turbulent boundary layer, J. Fluid Mech., 67, 125.
Wallace, J.M., Eckelmann, H. and Brodkey, R.S. 1972. The wall region in
turbulent shear flow. J. Fluid Mech., 54, 39.
Willmarth, W.W. and Lu, S.S. 1972. Structure of the Reynolds stress
near the wall. J. Fluid Mech., 55, 65.
Synthesis of ExperiInental Riblet Studies in
Transonic Conditions
SUl\1MARY - The present paper summarizes the status of the experimental research car-
ried ant both at O~ERA/CERT and ONERA/1Iodane, as regards internal manipulators.
commonly named, by a lot of researchers, riblets. That turbulent boundary layer ma-
nipulation program was begun at CERT in mid-1986. Emphasized will be drag reduction
performances of such passive devices, tested at transonic conditions, on a cylindrical body,
a CAST I aerofoil for two-dimensional turbulent boundary layers, and, at last. a complete
l/11th scale Airbus A320 model.
1 Introduction
The structure of a turbulent boundary layer has been studied by numerous investi-
gators and its description in terms of coherent structures is subject to much debate.
Though turbulence seems to be a very complex phenomenon, a specific set of fea-
tures ill the turbulent boundary layer has been identified: streamwise ,'ortices, low
speed (walllilyer) streaks, ejections, break-up, sweeps, outer layer motions and their
interactions with the walL entrainment of irrotational flows, ... The production stage
of l urhulcnce has usually been referred to as the "bursting phenomenon", [5]. One
approach to illterpreting turbulence-control experiments is "to view them as attempts
to interfere with some component of the turbulence production cycle", [6] : thus. one
could think of modifying either the large outer eddies or the wall layer events. or.
consequently, of changing the communication bet ween the ,'arious scales.
III the last decade, a lot of effort has been devoted towards manipulation of
tl11'bulelll boullClary layers by passi\'e devices such as internal manipulators (small
slrealllwise grooYE's acting straightly on the inner region), commonly referred to as
riblets. for tile purpose of reducing viscous drag. Though the details of the mecha-
nisms are not firmly understood, "such a concept of llsing grooved surfaces to obtain
skin-friction reduction is rather close to industrial applications", [7]. In fact. the
question which arises is to address the problem of how and to what extent the riblets
alter the turbulent 11m\l structure in reducing viscous drag. Though it is not the
purpose of this paper to explore the eJIects of these passive devices on the mean and
fluctuating quantities of a turbulent boundary layer, [16], one must recognize that
some studies showed up that grooves would restrict the spanwise motion of the lon-
123
124
gitudinal \'orLices [7], [17], would not lock the low speed streaks in a fixed spanwise
location [5] but \Vould increase their spanwise spacing [4],
At NASA Langley Research Center, a complete experimental investigation of
such passi\'e drag-reducing devices was performed by Walsh [24], and \\lalsh and
colleagues [2,j], [26], [27], They have tested many different groove geometries and
found that the optimum shapes for drag reduction have a sharp peak protruding
into t he flo\\' and have a height and spanwise spacing of the order of the viscous
sublayer thickness, l\ett drag reductions up to S% have been recorded on the most
popular symmetric V-shaped riblet. 1Ioreover, performances of grom'ed surfaces were
unaffected by the presence of moderate ach'erse and/or favorable pressure gradients.
and by misalignment up to 1.5 0
The first experiments at CERT, as regards performances of ribbed surfaces. were
carried out in mid-19S6, Thus. initial tests performed in zero- and acl\'erse pressure
gradient conditions for low subsonic speeds, confirmed the earlier findings of nett drag
reduct ions. The efficiency of grooved surfaces was essentially determined through
eiilier momentum balance technique or wake surveys, [9], [12], [15]. These preceding
"fundamental" experiments have allowed to verify, with laboratory measurements,
that carefully designed internal devices could provide nett drag reductions. In order
to go closer to flight applications, the effect of small streamwise grom'es on a slender
body (1 j:38 th scale Airbus-type fuselage) was checked in the F2-0NERA/Fauga wind
tunnel. [11]. Nett skin-friction drag reductions \yere measured by means of an intern al
one-componellt drag balance.
Tile following step was to pursue the evaluation of such a passive drag reduction
at transonic speeds, in order to provide data on riblet performances applicable to
transport aircraft fuselage and/or wings, This paper summarizes the experimental
research carried out in transonic wind tunnels at O:\ERA,
LcarhOrnndnIn hand
r 1'~halaJ-lce g -~-;==----===------+-)---'9) j
-,-----S-.----(_in
M[
1.Or t Moo=
',I /~~- _ _ _ _ _ _ _ _ _ _ _ 0.80
I(~ 0.70
IV
" -------- - 0.60
0.5j~ -0.55
~~ 0.42
II
- X/c (c=0.75m)
o 0.5 1.0
Figure 2: Local .lIach number distribvtions (cylinder)
the greater the amplitude of the excess of velocity is. Knowing these distrihutions.
boundary Inyer computations were performed in order to estimate the contribution of
the friction drag to the total measured drag. The code. developped at CERT by Arnal
et aL [:3], soh'es the local boundary layer equations in three-dimensional compressible
flows. By integrating the local shear stress along the curvilinear abscissa. one ends up
wi th t he computed skin-friction drag coefficient, Cd J : the chosen reference surface
has been based on the cylinder diameter (7r9 2 / -'). Boundary layer tripping induces
inevitably the problem of over-thickening due to roughness clrag. Computing the
step in momentum thickness, 0, at the transition location, [2], and including it in tlw
bounda ry byer code, one could deduce the change in 0 at the end of the cylindrical
body. The btler is rather small (0.6%), [13], which means that the penalty roughness
drag can be lleglected. Thus, comparing Cd J to the measured total drag coefficient
Cel. rewalecl that the contribution of the pressure drag was rather weak (:::::: :3-''5'0)
and almost independent of the Reynold" number, [1:3].
20
~~d If ("10) 0 .
10-
t B
..
0
0 h (mm) = 0.023
.r. 0.033
.. ~
0.051
.. . , 0.076
coo
0
08.
o&,' OP/Ox =0
8 I 0
9~. ~
,0 0
,",
-10 --' _h~
0 50 100
For the [our considered riblet models, the variations of the friction drag coefficient,
over 1, arc plotted versus h~ on figure 3. For the explored range of free-stream 1Iach
number and stagnation pressure or Reynolds number, this choice of grom'ed surfaces
allow to go through h~=7.7 to 77. There is some scatter on figure 3 due to the fact
that drag decrease is very sensitive to cross-section uniformity, surface finition, ... It
can be noticed that the biggest rib size (h=76pm) constitutes the lower branch of the
data bane!. For this specific groove height, the lower the value of 1C"J is. the smaller
the non-dimensionalised height is, but also the weaker the consistency of results is,
according to the afore-mentioned explanations (i.e. non adapted lower and upper
walls of the test section, for low values of ?lIex,). That scatter might also come from
the recorded dependency of percentage of drag variations on the free-stream i\Iach
number, for a given Reynolds number. Indeed, for a given manipulated Reynolds
number, IlL of about 13 10 6 , it was noticed that the greater :\lx is, the larger the
percentage of drag reduction is and/or the smaller the percentage of drag increase
is. [13] .
.\'evertllcless, maximum turbulent skin-friction reductions close to 7 - 8% have
been recorded for h~ close to 10-15 : however, when h~ :s: 20 (res]). 2': 30) drag reduc-
tions (resp. ill creases ) exist whatever riblet model is concerned. Let us mention that
on this same figure :3, the dotted line represents results obtained in incompressible
flows for the same kind of symmetric V-groove riblet, manufactured by the 3M Com-
pany ; aspect ratio equal to one, but h=0.152mm, [9]. These data were obtained from
meaSllremellts of the change in momentum thickness. There are a few difficulties in
using that teclmique for evaluating the efficiency of rib let models; in fact, the drag
128
o McLean et al 19
1.08 o Walsh et al 26
/':, Coustols 9
1.04
1.00
til
0 Squire et al 22
'-.....
0 0.96
reduction of riblets is small and the performance, i.e the le\'el of reduction, might be
affected either by the scatter in the momentum thickness measurements or in the (J
de\'elopment along the acti\'e riblet area exposed. From the reported experiments. it
seems that the "ariable h;; allows to gather data from high and low subsonic ]0\'"5.
Some authors argue that the upper limit of the reduction domain (the zero-drag
reduction crosSO\"Cr point, as quoted by Walsh et aI, [27]) be slightly higher in the
transonic regime. lIowe\"er, one has to be aware that for transonic conditions, the rib
height could be scaled with either the wall Yariables : h~= (hlv w ) . JTwl p"" or the
free-stream OWeS: h-:-= (hiVe) . JTwl Pe' The latter corresponds to the incompressible
\'alue and i~ used by some authors, for instance Squire et aI, [22]. Depending upon
the Reynolds and .'-Iach numbers, the difference between h~ and h-:- can be more
or less important. Thus, considering this cylinder-type body, when RL ~ 1:3 lOG
the ratio (h-:- I h~) is close to 7.2% (resp. 12.2%) for a free-stream Mach number
of 0.6 (resp. 0.82). This might be the reason of the differences between high and
low subsonic data since it is not always ob\'ious to guess what variable is used on
different plots ... Anyway, in their last paper. \\'alsh et aI, [11], summarized and cor-
related the <l,"ailable riblet film data. It appears that "significant riblet data are now
available tv firmly establish their drag reduction performance"" Collecting data from
Coustols, [9], ?dcLean et aI, [19], Squire et aI, [22] and Walsh et aI, [26], Walsh et
aL [27], found that all the transonic data, as for the low speed data, are essentially
contained in a 4-pcrcent data band as shown on figure 4. One has to be aware that
the rib let performances ha\'e been obtained with different techniques for estimating
the drag ,"ariations (changes in momentum thickness, direct drag measurements, ... ).
Let us add that all the considered riblet models had the worldwide tested symmetric
V-shape, with an aspect ratio equaJ to 1, explaining why any rib parameter (height,
h, or spacillg, s) can be used for plotting riblet performances. Let us recall that
experiments performed by Coustols, [9] or COllstols et aL [121, pointed out that the
129
a
6~d I (%)
t
10
o
f
o
o 0
O~--------~--------~~------~----L---~
o
o
o
o
.
o
00')
-10
s=h=0.023 mm -1{"w
10 15 20 25
Figure .5: Effect of free-stream flow angle
zero-drag reduction cross-over point was smaller when the aspect ratio (s/h) increases
from 1 to 3. The effect of varying the aspect ratio seems to affect the le\"el of skin-
friction reduction; one has to be aware that the greater this ratio is, the smaller the
increase of \\"etted area is.
M~ ~ 0,76
/
a=O
Hc=3.5 10(,
-..X/c
Ol-------
o
Pigure 6: Local Mach number distributions (CAST 7 Aerofoil)
The local 11ach number distributions, 1l" are plotted on figure 6 for three values of
the free-stream 11ach number, Moo = 0.65, 0.70 and 0.76. As it can be observed,
for an angle of attack of 0, the chosen increase in 1\100 allows a significant variation
of pressure gradients. On the pressure side, whatever the value of Moo is, the flow
is decelerating from xl c=35% to 90%, then accelerating towards the trailing edge.
On the other hand, on the suction side, all along the distance subject to future
manipulation (0.15 5::x/c5:: 1.00), the data exhibit a flat plate-type evolution for
11::0 5:: 0.7-1 ; above that value, a shock appears at about 40% chord length.
For the reference configuration, when the pressure and suction sides of the
aerofoil are covered with a smooth vinyl sheet, some of the 103 pressure taps have been
uncorked in order to check eventual modifications of the pressure distributions, [20].
\Vithin the experimental uncertainty, there was no obvious variation; furthermore,
the total drag coefficient estimated through wake surveys was almost identical to the
one measured with an uncovered aerofoil. Though grooved surfaces ha\'e a significant
effect on fricLion drag, none experiment clearly establishes they might also modify
the pressure field. Thus, one could believe that replacing part of the smooth film by
riblct film \\'ould not have any influence on the pressure drag.
KnO\\'ing the local Mach number distributions, the same boundary layer code,
as the one mentioned in the preceding sections, has been run in order to compute
the local skin-friction coefficient and to integrate it all along the aerofoil, so that one
can get an estimate of the friction contribution to the total measured drag. That
contribution slightly decreases as the free-stream Mach llumber increases: 58% at
l\Lxo=0.65 dowll to 48% at 1100=0.76.
From the boundary layer code, it is also possible to compute the streamwise
evolution of the non-dimensionalisecl rib height parameter, h~, on either the pressure
or suction side of the CAST 7 aerofoil. Because of the different pressure gradients, h~
132
1,3
I>
0 smooth
100 Cd
h=O.023mm
I>
1,2 >--- + h=O.033mm
r I> h=O.OSlmm
1>1>
I>
+
+
1,1
0
1/--
I> 0
+
;I- 0
[D
+ ~
1,0 -'1]
+
0,9
I ) 1=
0,63 0,67 0,71 0,75
Figure 7: Synthesis of drag measurements
does not behaxe as smoothly as, for instance, on the cylinder. Nevertheless, one can
end up with an average value, integrated all along the rib let model between 15% and
100% chord length. This value is not identical on both sides of the aerofoil : indeed,
for a free-stream I\Iach number of 0.72, on the pressure (resp. suction) side, ht=0.7
(resp. 0.6) for a given groove height h=lflm. As these average values are rather close,
a mean value of 0.6.5 per pm would characterize the riblet model, root of passive
turbulent manipulation on both sides of the aerofoil. That mean value h~=0.65 is
slightly modified by the variations of the free-stream Mach number.
[1
25
20
- I
/:::,. Cd
- (%) c CAST 7
~
Cd 15 -- - Cylinder
friction
- Cylinder V
over L
~
I
10
V ------
/ V/
Jr V
5
/ /
0
-5
- _h:
-B-
L-------'" -
-
-10
o 10 20 30 40 SO
Figure 8: Synthesis of drag data (with and without pressure gradient)
in the T2-wind tunneL As the level of recorded skin-friction reductions was relatively
low, it was decided to cover not only the pressure side but also the suction one in
order to increase the percentage of manipulated wetted area, and with that the drag
variations_ From these preliminary results, it was tried to estimate the variations of
the friction drag coefficient over the manipulated aerofoil surface, the percentage of
which. towards the total wetted area, is close to 85%. Let us recall that boundary
layer computations showed up that the contribution of the friction part in the total
drag balance represented between 58% and 48% depending upon the ,-alue of the
free-st ream :-1ach number. The variations of the friction drag coefficient. over L, are
plot t eel on figure S versus h1;;, mean value of both integrated h-:: parameters e,-aluated
along the manipulated surface on pressure and suction sides. Data are compared to
those obtained on the cylinder; because of the scatter in this laUer experiment the
upper and 10\\'er curves of the data domain are represented. R('sults obtained by
lTlanipulating the turbulent boundary layer which develops on the CAST 7 aerofoil
are plotted in the shape of rectangular boxes: the horizontal box side expresses the
variation of h~ with 1\L;;v, when the vertical one denotes the slight dependency of drag
varia tions \\'i tll 1\ I 0::' and the experimental scatter. The skin-friction drag variat ions
brought about b.y the two smallest riblet models (2:3 and 33{Lm) are in agreement \\"ith
those recorded on the cylinder without any pressure gradient. Howewr, the lewl of
cl rag increase obtained from the deeper rib (51pm) is higher. This difference could be
explained probably by a bad quality of the model surface finition or a cross-section
non uniforlllity, since the other two models do not seem to show up a noticeable effect
of pressure gradients on riblet performances.
Thus, when grooved surfaces cover about 8-5% of the aerofoil wetted area, fric-
tion drag reclLlctiollS have been recorded [or ht :::; 20. 1\Iaximum total drag decrease
of about 3.5% was obtained for the smaller rib hei~, which corresponds to max-
imum skin-friction reduction of almost 7.5 - 8% at h1;; close to 12 - 16. This result
is obtained by assuming thai grooves do not modify the pressure drag but act only
on the frictioll drag. For free-stream rvIach numbers less than the di\-ergence iV1ach
134
number, the average pressure gradient parameter estimated along the manipulated
area, {J - Clauser parameter - for instance, is weak on the suction side since the local
l\Jach number is almost constant over 35 - 40% chord length; on the pressure side,
this parameter is stronger but still moderate.
The performances of grooved surfaces under pressure gradient conditions have
been investigated by some researchers in incompressible flows by Choi et ai, [8], Cous-
tols, [9], Truong et ai, [23], in transonic flows by Squire et ai, [22] and also in flight
by ?I1cLean et ai, [19]. The data collected from all these experiments seem to be
consistent, though the drag variations were estimated through different measurement
techniques. In summary, significant riblet data are available, for flows subject to ad-
verse and/ or favourable pressure gradients, to suggest their drag reduction efficiencies
as the pressure gradient is moderate, which is the case for most of the wing surface,
where grooved surfaces might be applied.
2.2.4 Miscellaneous
For these experiments, the grooved surfaces were applied onto the trailing edge of
the CAST 7 aerofoil. Then, depending upon the considered vinyl sheet, smooth
ftlm or rib let one, the thickness of the aerofoil base val'ies from 0.2111111 (reference
case) up to 0.282mm (h=0.0.51mm). Compared to the aerofoil covered with smooth
film, changing the rib size from 2:3llm to 51pm would raise the base drag of about
25%. For the reference configuration, at 1\100=0.7, an estimate pointed out that
the base drag accounts for a little bit more than 1 %. As in all these experiments,
one is looking for small variations of the total drag coefficient, it would be better
to keep constant the thickness of the trailing edge, so that, when manipulating the
turbulent boundary layer, the only recorded changes ,\'attld be devoted to skin-friction
reductions or increases. Of course, this remark is useful for small models in wind
tunnel. but "'oldd be useless for aircraft applications.
alone", for which the maximum Reynolds number based on the fuselage length is
a pproximaiely equal to 40 10 6 . The angle of attack of the model could vary between
_2 and +3.
Boundary layers were tripped on the fuselage and on the wings, using carborun-
dum bands. Their locations had been defined by Aerospatiale during other sets of
experiments performed with this specific Airbus model.
Only one symmetric V-groove riblet, manufactured by the 31\11 Company in an
adhesive backed film, with an aspect ratio of one, has been tested on the fuselage and
wings. The knowledge of the flow field around the fuselage, obtained from the wing-
body configuration, allowed us to perform boundary layer calculations, based upon
the resolution of integral equations ; thus, for instance, at ]\1==0.7 and 0=3.7,
one could get the evolution of h~ along the upper and lower symmetric lines, the
lateral mid-lines and so forth. It appeared that this quantity was not varying too
much with either the streamwise abscissa or the peripheral co-ordinate; so, a mean
\"alue - though it might change with the angle of attack and the free-stream 11ach
number - would be representative of the rib let scale. Thus, in the case of ~roove
depth of 0.02:3mm, for the rib let material set on the fuselage, an optimized h;;, value
close to 8 - 0 was found at ~-Ioo =0. 7.
1Ieasuremenis have been performed for fuselage and \\"ing-body type arrange-
ments. Several configurations have been considered:
- Fuselage alone without riblets ;
- Fuselage coyered with riblets ;
- Wing-body configuration without riblets ;
- Wing-body configuration with riblets set only on the fuselage;
- Vving- body configuration covered with riblets.
Along 1he wings, the grooves were approximately aligned with the external
free-stream flow direction. No grooved surface was set in areas subject either to high
geometric curvature (nose or tail cone) or to important streamline curvature (more
than 1:5 from the ribs direction), i.e. wing-fuselage intersection, upper surface of the
wing near the trailing edge, ... The corresponding percentage of wetted areas co\"ered
with longitudinal grooves is approximately:
- Fuselage alone: 7:3% ;
- Wing- body configuration ,,ith riblets set only upon the fuselage: 47o/c :
- Wings and fairings: .57% ;
- Wing-body configuration (fuselage, wings and fairings covered with riblets) : 66%.
- for the wiug-body configuration, ~Cd < 1.0 10- 4 in all cases and < 0.5 10- 4 in
136
CL
test Mao
test Mo>
47-0.800 0,5
53 --- 0.800 104 - 0.699
2
63 -0.800 113 --- 0.700
73 0.800 114-0.699
1
( ) 0,3
6C"x10'
-2 0 2 0,2
b)
-1
a) 0,1
-2 6C Dx10'
~
-2 0 2
93% cases.
As tbis repeatability is of excellent quality, it was witb confidence that the
efficiency of grooved surfaces has been checked under very good conditions. Further-
more, let us point out that for both configurations, a rather good repeatability on
a.ngle of attack measurements was recorded since c.a < 0.010 in 00% cases.
A dozen of pressure taps was inserted within the fuselage (sting cavity) allowing
to check the homogeneity of these pressure measurements as well as their constancy
during all tIle drag measurements.
( 6 Cd I 1.9 % ( 6 Cd I 2.3 %
Cd )U:CI cruise Cd )0( CI cruise
2 2
o o
t.Cd=Cd ribleC Cd smooth
-1 -1 -1
-2 -2 -2
(~ Cd)
Cd Clcrulse
= 1.2 % ( ~CCdd) = 1.6 %
Clcrulse
Cl
0,4 0,4
0,3 0,3
0,2 0,2
0,1 0,1
with negligible changes in llCd over the Cl range: 0.1 - 0.6 (figure 11).
Thus, at these explored free-stream Mach numbers, the benefit of covering wings
with riblets was observed, though the percentage of grooved surface compared to the
wing wet ted area was rather small (~ 57%). Moreover, though the grooves were
aligned with the external free-stream velocity, the effect of behaviour of riblets in
three-dimensional flows might be different from the influence of misalignment in hYo-
dimensional flo,,-s. Indeed, the main difference comes from the feature that the yeloc-
ity vector varies very quickly in the wall vicinity; so, non negligible angle deflections
exist over distances within the turbulent boundary layer as large as groo,-e depths.
Let us mention that the performances of riblets are under investigation at CERT in
three-dimensional flows: riblets are covering parts of the pressure and suction sides
of an aerofoil, set at an angle of sweep close to that of an Airbus-type wing.
\Vhate,-er configuration is concerned, it is rather difficult to guess a correct
estima te of the skin friction reduction mainly because the sting caused pressure field
modification and, consequently, drag interferences. Anyway, at i\L;,,=0.7 and at cruise
level, though the model has no fin and no horizontal tail, if one assumes that friction
represents about 50% of the total drag, the average nett skin-friction drag reduction,
estimated m-er ille 66% manipulated wetted area, is close to 4.8.5%.
This benefit is not certainly the highest one, because:
- the percentage of wetted areas covered with ribbed surfaces was rather smalL espe-
cially on the wings due to boundary layer tripping;
- on the fuselage, close to wing-body junction, the grooves had not been re-aligned
because of streamline curvature. Compared to the fuselage configuration, tested at
first. several strips of riblet film had only been taken off ;
- the grom-es size had not been optimized on the wings; the same depth and geom-
etry as the ones used on the fuselage were considered, mainly because of material
availa bili ty.
In spite of these observations, substantial drag gains have been recorded on the
wing-body configuration. Furthermore, these results roughly agree with the friction
drag gains measured under laboratory conditions, for instance with data obtained at
CERT on either the cylinder or the CAST 7 aerofoil in the T2-wind tunnel.
4 Conclusions
All the experiments carried out under transonic conditions have alloweu to verify the
efficiency of internal manipulators, riblets or grooved surfaces, for the purpose of
reducing turbulent skin-friction drag. Thus, from different research groups, significant
I'iblet data are now available to firmly establish the potential of such devices for drag
reduction performances. It appears that maximum drag reduction is found \\' hen
groow geometries, such as the height or the spanwise spacing is t}-pically of the order
of 10 - 1;3 v JUT.
Furthermore, experiments performed on a complete A320 model, in the Sl-wind
tunneL showed up that important total drag coefficient reductions could be achieved,
at cruise conditions. Since the fuselage Reynolds number reaches around 40 10 6 , it
is expected that these results could be rather easily applied to practica.l flight test
conditions where the Reynolds number is only increased by a factor :3.
Although the mechanisms involved in such a drag reducing process have not,
139
yet, been understood, a couple of flight tests have already been performed with fuse-
lage, wings, fin, horizontal tail and nacelles equipped with riblets. Some information
concerning the flight tests, carried out by Airbus Industrie and its partners, as a
consequence of the promising Sl-wind tunnel results, on the Airbus A320 N1, could
be found in [21].
Conscquently to these useful results obtained in transonic conditions, the next
step would consist to look at the behaviour of grooved surfaces in supersonic regime.
Indeed, for a Concorde-type aircraft, the friction drag represents about 30 - 35% of
the total drag, which implies that improvement of aerodynamic performances of such
supersonic planes could be achieved by friction drag reduction. To our knowledge,
the only supersonic drag measurements of a riblet film were performed by Gaudet
at a Mach number of 1.25, [18]. \Valsh et aI, [27], reported that the supersonic data
falls within the transonic data band. Let us mention that experiments will be carried
out at Ol'\ERA, under zero-pressure gradient conditions, for higher values of Mach
numbers up to 2..5.
Acknowledgements
Financial support was provided by Airbus Industrie and the "Service Technique des
Programmcs Aeronautiques". Special thanks are due to F. l\'larentic from 3?11-USA
and A. Dcladwnal from 31\I-France, for providing us \yith all of the riblet material.
References
[1] Archambaud J.P., Blanchard A., Seraudie A. : 11th Congress on Instrumentation
in Aerospace Simulation Facilities - Stanford, (1985)
[2] Arnal D. : V.K.I. Lecture Series - AGARD Report W 709 (1984)
[3] Arnal D., Jelliti 1\-1. : CERT Internal Technical Report (1985)
[4] Bacher E.V., Smith C.R. : AIAA Journal, Vol. 24, W8, pp. 1:382-138.5, (1986)
[.5] Blackwelder R.F. : AIAA Paper 89-1009 (1989)
[6] Bushnell D.l\1., IVlcGinley C.B. : Ann. Rev. Fluid 1\1ech., Vol. 21, pp. 1-20, (1989)
[7] Choi K.S. : J. Fluid Mech., Vol. 208, pp. 417-458, (1989)
[8] Choi K.S., Pearcey H.H., Savill A.M. : Int. Conf. on Turbulent Drag Reduction
by Passive Means, London (1987)
[9] Coustols E. : AIAA Paper 89-0963 (1989)
[10] Coustols E. : 4th Int. Conf. on Drag Reduction, Davos (1989)
[11] Coustols E., Gleyzes C., Schmitt V., Berrue P. : 24ieme Colloque AAAF Poit-iers-
France (1987)
[12] Coustols E., Cousteix J. : 16th ICAS Congress, Jerusalem (1988)
[n] Coustob E., Seraudie A., 1\Iignosi A, Breil J.F. : CERT Internal Technical Report
(1988)
[11] Coustols E., Savill. A.1\1. : Applied Scientific Research, Vol. 46, pp. 183-196,
( 1989)
[1.5] Coustols E., Cousteix J. : 2nd IUTAM Symp., Zurich (1989)
[16] Coustols E., Cousteix J. : 7th Symp. on Turb. Shear Flows, Stanford (1989)
[17] Djenicli L., Liandrat J., Anselmet F., Fulachier 1. : 2nd European Turb. Conf.
(1988)
140
[18] Gaudet L. : Applied Scientific Research, Vol. 46, pp. 245-25'1, (1989)
[19] }\1cLean J.D., George-Falvy D.N., Sullivan P.P. : Int. Conf. on Turbulent Drag
Reduction by Passive Means, London (1987)
[20] Prudhomme S., Coustols E., Mignosi A., Dor J.B., Plazanet M. : CERT Internal
Technical Report (1989)
[21] Robert J.P. : to be published in that Proceedings Volume (1990)
[22] Squire L.e., Savill A.~1.: Int. ConL on Turbulent Drag Reduction by Passi\"e
!\leans, London (1987)
[2:3] Truong T.Y., Pulvin P. : Applied Scientific Research, Yolo 46, pp. 217-227, (1989)
[24] Walsh 1\1.3. : AIAA Paper 82-0169 (1982)
[25] \\"alsh :\1.3., Lindeman A.IV!. : AIAA Paper 84-0:347, (1984)
[26] Walsh :\1.3., Sellers III \V.L., McGinley C.B. : AlA A Paper 88-2.,)54, (1988)
[27] 'Walsh 11.J., Anders J1'. J.B. : Applied Scientific Research, Yol. c16, pp. 2.5,')-262.
( 1989)
Effect of Riblets on either Fully Developed
Boundary Layers or Internal Flows in Larninar
Reghne
J. LIANDRAT 1 , E. COUSTOLS 2 , L. DJEI'\IDIl,
F. ANSELMET 1 , X. de SAINT-VICTOR2 , F. FIOC 2 &
L. FULACHIER 1
1 Illstitut de Mecanique Statistique de la Turbulence, Unite Mixte [inivasite.
CNRS I\'" 38033, Marseille, France;
2 ONERA-CERT, Aerothernwdynamics Department. Toulouse. France.
SUMMARY - The present paper summarizes the status of the numerical research carried
out both at OKERA/CERT and IMST, as regards the efficiency of internal manipulators,
more commollly named by many researchers, riblets, in external as well as internal flows,
under laminar conditions. First results obtained at CERT during preliminary studies will
briefly be recalled. From a numerical point of view, many difficulties arose; most of them
stemmed from the singularity at the riblet edge. First of all, results issued from different
grid refinements will be presented. Then, emphasis will be put on analogies and differences
between manipulated fully developed internal and external flows in the presence of either
\'- or L-shaped riblets ; the latter corresponds in fact to half of a U-shaped groove. These
theoretical studies showed up that the wetted area increase did not induce extra skin friction
in laminar boundary layers over V- or L-shaped grooves.
1 Introduction
It is clear from various experiments, performed up to now in turbulent boundary
layers, that l'iblets of suitable dimensions can reduce wall friction by about 6 - S%
(see. for instance, Bushnell [:3], Coustols [.5] or Walsh et al [19]). In the case of
internal turbulent flows there are much fe\ver experimental results but it seems from
Liu et al (sec [:20]), Lowson et al [Ue] and Rohr et al [16], that similar reduction could
exist. This reduced skin-friction seems indeed to be related to an increased stability
of near-wall structures (i.e. Bacher and Smith [1], Choi [4]). But, what is the physic~l
mechanism responsible for this stabilisation? By considering the most usual situation
of Y-groO\'es with s=h, (s denotes the spacing between two adjacent grooves and h the
rib depth), the wetted a.rea increase can be greater than 100%. If one considers that
this specific increase tends to enhance skin friction, as pointed out experimentally
by Pulles et al [1.5] and numerically by Launder et al [13], it is difficult to accept
that turbulence Oll its own can compensate for and even overshoot this negative
effect. Thus, it is reasonable to think that a purely viscons mechanism is involved
and either entirely or partly counteracts this areil increase. This is the reason why
an experimenlal study on laminar boundary layers o\,er triangular l'iblets has bee;!
performed in the I.lVI.S.T water tunnel [11]. Results have shown that, over grooves
141
142
with s=2.3h, [10], [7], (32% wetted area increase) and riblets with s=l.2h (94 % wetted
area increase, unpublished results so far), the overall skin friction is practically the
same as that oyer the smooth wall: there is, thus, no straight relationship between
skin friction a11(1 wetted area.
In order to specify whether skin friction is slightly increased or eventually re-
ducecL a numerical study has been undertaken at Il\lST, in the same conditions as
in experiments, based on a code initially developed at CERT [6]. First results, ob-
tained at IMST, [8], showed a small drag reduction of about 3%. although the grid
mesh had to be refined in particular in the crest vicinity [9]. These calculations were
ca.rried out over V-shaped riblets and the rectangular grid was not orthogonal to
walls: this is, mainly, one of the reasons why computations over L-shaped grooyes
have been pursued at CERT and performed using different grid refinements. \' ever-
theless, these first Il\JST results were in agreement with Kahn's ones, [12] : indeed,
Kahn computed the boundary layer development over a V-groove riblet, s=2h, and
reported a 17c reduction in laminar regime.
As far as laminar internal flows are concerned, there seems to be only a single
experiment<d result relative to a fully developed pipe flow with triangular grom-es
with an aspect ratio (s/h) of 1, [16] : within the measurement accuracy, there was
no effect of riblets on skin friction. Preliminary calculations carried out at CERT in
a rectangular channel with V-shaped grooves [6], and computations run at Ul\IIST
in channel with L-shaped grooves [13], have shown a drag increase, especially in
Reference [13], where the skin friction variation is significant, about a 307c increase
for s=h and h/II=0.05 (II denotes the channel half-width).
It should be noted that a large amount of theoretical work has also been per-
formed by 13echert et ai, [2] ; indeed, he and his colleagues looked at several groove
cross-sections by considering transformal mapping and solving Laplace's equation.
Their results were analysed in terms of "protusion height", a measure of the influ-
ence of the riblets upon the boundary layer, rather than in terms of ability of such
surfaces to decrease shear stress.
In order to try to clarify some specific features, attention is focused, within this
paper, upon:
- analogies and/or differences between fully developed internal flows and boundary
layer flows in the presence of either V- or L-shaped riblets
- existence or lack of skin friction increase.
(2)
Though the diffusion terms, taken along the streamwise direction, are neglected, the
elliptic feature holds according to the streamwise pressure gradient. Thus, the preced-
ing system corresponds to what has been called "semi-elliptic case". If the pressure
is given, the clliptic behaviour disappears and the system to be solved provides us
with a "parabolic case".
This set of equations is solved using an x-marching method, where it finite
\'olume schemc is considered for discretising any deduced equation. The numerical
code is based on a Semi-Implicit Method for Pressure Linked Equations, [IS]. During
iterations, en'ry equation (momentum and pressure correction equations) arc solved
by either the Modified Strongly Implicit algorithm, [17], or t\\o sweeps in y- and
:;-clirections. The efficiency of one or another method depends strongly upon the
mesh size within the groove, the number of points within the computational domain,
the groove cross-section shape, and so forth.
Initial conditions arc given on the velocity components at the initial (Y. z) plane,
whereas the pressure or its derivative with respect to :r, is fixed at the last streamwise
(Y. z) plane. No-slip boundary conditions apply at all rigid surfaces and symmetry
conditions along the lateral boundaries within the fluid (aU/az = 0, aF/az = O.
TV = 0) : at the upper limit of the domain, which corresponds to approximately 20
boundar}' layer thicknesses, the first derivative with respect to y of the streamwise
velocity component is equal to zero.
First calculations were performed on a V-shaped groove such as s=211, 11/8 :::
0.125. [6]. whcre 8 denotes the thickness of the laminar boundary layer at the leading
edge of 1he riLlet model. Computations started on a smooth plate. with a Blasius t~'IW
profile. After 0.25m downstream distance, the laminar boundary layer dewlopcd o\'er
the riblet model. the length of which was 0.2.5m. Then. the wall geometry went back
to a smooth plate. A single configuration was only studied because of computational
economy due to the great number of nodes: 7.5 planes in the streamwise direction wit h
all * 19 mesh in the y * z directions \\'ith 10 point.s coyering the groo\'e depth. Doth
approaches, semi-elliptic and parabolic ones, prm'idec! local skin-friction reductions
oyer the riLlet modeL compared to a smooth surface. Hmveyer. some differences
su bsisted. mainh' in the vicinity of the geometric singularity. i.e junet ion of the riblet
configuration alld the downstream smooth onc, due to the downward step effect. for
installce. a lilrge skin-friction increase \\'a5 recorded upstream of this singularity, in
the semi-elliptic approach. III facL that same behayiour was also obsen'ecl when
computing the development of laminar flow nearby a downward step.
Besides these considerations, one of the most significant results has been that
the transyersc velocity component, TV, is negligible in all the computa.tional domain,
solving the system of equations with either a semi-elliptic or parabolic approach. This
144
has also been evidenced through experiments performed at IMST, [11], [7]. Neglecting
this transverse velocity component is, moreover, consistent with Kahn's work, [12].
Indeed, those last computations revealed, by solving the parabolized streamwise mo-
mentum and vorticity equations, the absence of secondary flows, which do not form
a vortex within the groove, in laminar flows. However, additional computations
performed by the same author but, in turbulent regime, indicated the presence of
secondary flows which "are found helping the viscous forces in the drag reduction
process" .
It should be noted that such an assumption is very useful since it allows to
simplify tremendously the system of equations and, consequently, to diminish the
computational time. Thus, the system of parabolic equations can be written as
follows:
o(UU) o(UF) 1 dP 02U 02U
-ox- + -oy- = ---
p dx
+ v( - +-)
oy2 OZ2
aU + oV = 0 (-)
ox oy
That system of equations is solved using an x-marching method; the strealmvise ve-
locity component, U, is obtained from equation (3), taking equation (4) into accounL
until convergence. Then, from the continuity equation the normal to the wall velocity
component, F, is deduced. The initial and boundary conditions are identical to those
used for soh'ing the complete three-dimensional case.
This "semi three-dimensional approach with Hi = 0", has been developed both
at IMST and CERT for looking at laminar boundary layer manipulation respectively
with V- and L-shaped grooves. Particular attention has been devoted to the estima-
tion of the mean local skin friction coefficient, since small variations of this quantity
are seeked [01'. Depending upon the groove cross-section, that estimation will be
more or less easy. The choice of L-shaped riblets made at CERT differs from that of
either earlier studies [6] or Il\IST work [7], [9], [10]. ;\Ievertheless, these specific ribs
gave similar drag reductions to ones of triangular section, as demonstrated experi-
mentally by Wilkinson et al [21] for turbulent boundary layer flows; it was expected
that it would be the same case in laminar external flows. That choice was made
in the interest of computational "easiness" since the rectangular grid mesh could be
easily defined. ]\loreover, the skin friction coefficient can be obtained directly from
streamwise velocity component derivative with respect to y or z co-ordinate. The
main resulls o[ these studies will be discussed in !1.
The efficiency of riblets in laminar internal flows had briefly been considered
at CEnT, [G]. To our knowledge, the only other numerical work concerning the ef-
fect of ribleted surfaces in internal flows has been performed, later on, at Ul\IIST
by Launder and Li [13]. Nevertheless, based on the results obtained from developing
turbulent boundary layers, these authors made the assumption that "the contribution
of the conVE'cti"e terms to the flow pattern in the vicinity of the riblets is negligi-
ble", since the groove depth does not exceed a few percent of the boundary layer
thickness. Thus, they performed a complete examination, from a numerical point of
view, of mani j)ulation of fully developed flow through a plane channel under laminar
conditions, [1:3] ; it should be noted that they considered knife-edge riblets in internal
turbulent flows.
The computations carried out at CERT were performed using equations (1)
and (2), soh'eel under the parabolic form. The main difference with the developing
boundary flow is that the mass flow will be kept constant ; that constraint allows
one to obtain the mean streamwise pressure gradient at each computed plane in the
streamwise direction. The numerical code was run for as many streamwise abscissa
as necessary, until convergence was reached. Since the geometry was symmetrical.
the flow \\'as studied for 0 ::; y ::; H, where H is the channel haH-width, i.e half
the distance bet\\'een the two facing grooves valleys. The number of grid points
was ,11 * 19 in the y * : ; directions, with 10 points covering the rib height. The
boundary conditions were similar to those used for external flows calculations. 1'\0-
slip boundary conditions applied at all rigid surfaces and symmetry conditions along
the lateral boundaries within the fluid (aU / az = 0, D1T / az = O. TV = 0) ; at the
upper limit of the domain (y=II), the first derivati"e with respect to y of all the
velocity components was equal to zero.
Three configurations were examined:
- Configuration A : h/5=0.:300 , h/1I=0.290 ;
- Configuration 13 : h/s=0.:360 , h/H=0.OS.5 ;
- Configuration C : h/5=0.0:36 . h/H=0.009 .
Once tlie convergence was reached, the computed streamwise pressure gradient
\\'as compared to the one corresponding to a smooth \\'all for the same flow rate and
the same chanllel cross-section. Thus. for configuratioll A. an increase of friction
drag close to 2G% was recorded. As the ratio h/lI was rather important. it \\'as
bclicH'cl thai numerical uncertainties could exist due to the small amount of grid
points used lo define the groO\e. It \\'as then decided to decrease the yaluc of this
height: channel width ratio, as performed in configurations 13 and C. Ne"ertheless,
drag increases were still obtained: 4% (resp. c::::: 0%) for h/1I=0.OS.5 (resp. 0.009).
Then, when the rib depth \\'as adapted to the half-channel width, let say h/E
less than 10 (!ci, skin friction drag increases were recorded for laminar internal flO\\s.
Though the mesh grids were not carefully defined (especially around the crest vicin-
ity). these results might be subject to caution. Nevertheless, when running the same
three-dimensional parabolic code on developing boundary layers, wilh the same type
of mesh size, one ended up with almost 4-5% skin friction drag reduction o\'er a
similar grooye geometry h/ 8=0.10 and h/s=0 ..50. Thus, because of non grid refine-
ment, the afore-mentioned interpretations have to be considered as a qualitative even
compa.rative (internal - external flow behaviour) approach and analysecl in such a
way.
This skin friction drag increase, for internal laminar flows, is consistent with
146
results obtained later by Launder and Li [13]. Their reported results concerned
manipulation of fully developed flows through plane channels with knife-edge riblets.
Different parameters were varied; for instance, h/s from 0.5 to 3.0 and h/I-I up to
0.08. The equation, they solved, is the Poisson equation which could be read in the
same co-ordillate system:
1 dP cPU fJ2U
0= - - -
p dx
+ v( - + -)
f} y 2 f} z 2
(5)
This equation could be in fact derived from the semi three-dimensional approach with
HT =0 by neglecting the convective terms in equation (3). Information concerning
computational details could be found in [13]. An interesting feature was that be-
cause of very small dimensions of the riblet, these authors applied a one-dimensional
treatment at heights greater than 2h above the rib crest. Then, this computational
economy allo"'ed Launder and Li to consider several geometry configurations. A
typical grid 111esh was 28 * 46 for the two-dimensional region, with an extra 22 lines
y=constant for the one-dimensional domain. In fact, that grid size \\"as quite compa-
ra ble to the one we used for triangular grom-es. It is unfortunate that they did not
appear to have investigated the effect of grid refinement since they themseh-es have
stated that "it is not clear that any of the studies has adopted a sufficiently fine grid
to provide a convincing demonstration that drag reduction is achieved, or. indeed is
achieva ble in laminar flow" .
Anyway, whatever geometry configuration was considered, the results did not
indicate a skin friction drag reduction compared with the case of a smooth channel.
for a constant aspect ratio his, the drag augmentation factor increases with hilI, (see
figure 10). nesults obtained at CEnT qualitatively agree with those from Launder
and Li, though the considered system of equations and numerical method were rather
different.
So, in conclusion, this preliminary numerical study, performed over micro-
grooves geometry with triangular cross-section, clearly showed up that the efficiency
of riblet appears to be different between internal and external flows. nesults, deduced
from the same numerical code with the same type of grid mesh for almost identical
aspect ratio, indicate a drag decrease (resp. increase) for developing boundary layer
flows (resp. fully developed plane channel flows).
I!
III I II III
I I II II
I I I I
II III ,I I
II I I i Iii II II I I i if
y
II I III I j II III Ii
II I III II II III II
II
II I
I III
I I!
I II
III
I III
I i 1111 Ii
Iii t
I, I I I I III I I1111 I I I
11111111 I I! I , l l II I III
Iii' I I III, I I' 1I11
I! I i I I !III I II I 1111/ I
I I~ I I I I: I I I 1'1 I :.11 I
Ii I r II 11111111111111111 I I I
II I I 1'-' I I' II 1111111 I I I '
1111 t " ,II Illl! 1 ' 1 1 1 1'1
Ii 1111 I II' III I II 1111 I
,
I I II II 11111 111111
I::!: ~ : :
I,' 11111 III 111111'
II!
111111 : : I I , : : : I ~ I:
111111 I I, ;111/)11 II III i
IIII! I 1I11 I III III" 1 I I
11I11 II 11II III '1111 I I
groove spacing. The mean skin friction coefficient is, then, obtained when dividing
that force by the considered wetted area. 1\'1oreovcr, b is the thickness of the laminar
boundary layer taken at the riblet leading edge.
('Jlj)
\JY
C
(~~)
II
o ~
LI __________________________________________ ~ ______ ~
10- 11 10- 9
6 Y (In)
c
Figure 2: InJlllencc of the size of the first step ovcr the V-grool'c crEst (h=lmm. 0
s/II=0.:3 . s/h=L3. Ue =6.7cms- 1 . grid mesh: 81*39)
number of grid points (figure 3) did not influence the computed overall skin friction
provided that it \\'as larger enough (more than 41 * 19).
The overall wall friction was inferred from the spanwise distribution of velocity
gradient along the line connecting two adjacent crests. Indeed, it was shown., ac-
cording to references [7] and [10], that friction along that line was equal to friction
along the \'-\\'all. since momentum fluxes corresponding to the ach'ectin' terms in the
elementary volmne delimited by this line and the wall were negligiblE'. ;"Ioreo\,e1', the
velocity gradient distribution along the V-wall exhibited much larger spanwise \'aria-
tions than that related to velocity gradients along the crest line: estimating the area
beneath the curve related to the second distribution is easier for the determination
of the overi,ll wall friction, especially in the crest vicinity. Three methods were used
to perform this integration. In the first one, the trapezia method was applied to the
gradient velocity values obtained at the grid mesh points: friction was overestimated
even if the number of grid points was increased and did not depend upon this number
(figure 3). Indeed, when the grid was refined, velocity gradients per unit surface were
getting larger since they were related to smaller surfaces. This integration technique
gave drag enhancements of about 0.3% with respect to tbe smooth wall. The second
method dealt with eye's smoothing of the data (i.e. drawing a best-fit curve by eye)
and integration with a large enough number of trapezia in order to produce an in-
tegration error as much negligible as possible. About 1..5% drag reduction was then
obtained: this was corroborated using the standard Newton-Cotes method.
For the parametric study of the riblet geometry [7], the grid mesh was 81 * :39
with ~YL=7.;j 10- 6 b. Keeping either s or h constant allowed one to let varying the
aspect ratio s/h. Then, a 1% up to 2% drag reduction was achieved \\'hen 0.0:3 <h/b <
OAO for s= 1111111 , (figure il).
149
L1FF ('!)
.,
a
0 0 0
0
Figure 3: II/fluence oj lhe number oj grid points on jriction jorce 1'arialion (h=l mm.
s=1.5mm. Uc =G.7 cms- 1 , a integmtion jl'OTn grid points only. _ integratioll llsing c:rtm
points j/'Um eye's smoothing jitting)
- 4
o
o
o
"T
o o
o
sill
o ~---------------L________________L -______________~______~
o
Figure cl: Influence oj aspect ratio on jridion jorce variatian in V-gl'Oa['( ajtEl' eye's
smaathingfiiting (a h=lml7l," s=lmm, Ue =6.7cms- 1 , grid mesh,' 81*39)
150
5,Oe-4 5,Oe-4
5,Oe-" """'~~=~~,r
y(m)
2,Se--<1
1- - ----
t 2,Se-4 2,5e-..:j
--- ---
-.-Z(m)
II III
2-
10 3 Cf
.2 .4 .6
Figure 6: Influence of the grid shape on the streamwise evolution of the mean Cf
(h=s=O.Smm)
2-
10 3 Cf
t Alnslus
<10' no
1-
OL1
o
__ I_ _ I_ _ I _ _ _ I_ _ _ I_ _ _ I_ _ _ I _ _ _ I ____ I. _______ I. _ _ I __
.2 .4 .6
7 _~_:J1~J
Figure 7:fllfl uence of the number of node points on the st7'eamwise evolution of the
mean Cf (h=s=O.Smm, grid type III)
153
le-7 .............................................[............................................................................................ ]
8.6e-7
o 6.4 e-7
:: o --+---+~y c (m)
o D ..
Oe+O+--~~~~~~r--~~~~~~--~~~~~~
10- 7 10. 6 10- 5 10- 4
recorded; moreover, after about 50 boundaTY layer thicknesses downstream the ri-
blet leading edge, this drag gain seemed to be independent of Reynolds number.
This observation justifies the fact that the value of Cf taken at the last computed
streamwise abscissa (x = 0.75m) will be retained for evaluating riblet performances.
\Vhen plotting the evolution of the average skin friction coefficient versus momen-
tum thickness Reynolds number, R() with () estimated from integration all along the
groove width, one ended up with the same amount of overall drag reduction at this
last computed streamwise abscissa. In fact, in order to judge of riblet performances,
one would have to compute the overall skin friction drag by integrating the local skin
friction over the streamwise rib surface. It should be noted that the evolution of
the skin friction coefficient just downstream of the leading edge of the riblet model
was strongly dependent upon the 6.x step; the latter was chosen very small at this
specific location and then became larger according to an expansion factor.
The smaller 6.Yc is, i.e. the smaller the mesh thickness just above the peak is,
the larger, at the groove peak, the velocity gradient per unit surface is. In fact, when
plotted the local shear force versus 6.Yc at a given streamwise location (:1' = 0.6m,
figure 8), one could notice that this force did not depend anymore upon !:::.Yc as long
as this latter "'as small enough. This minimum 6.Yc may vary with either the groove
height or the aspect ratio (s/h) ; for the considered groove, s/h=l and h=0.5111m,
(6.Yc)min ~ 1.28 10- 3 h ~ 1.10 10- 4 8, which is an order of magnitude higher than
the minimun1 value 6.Yc considered for V-shaped grooves. It should be noted that,
although both Il\IST and CERT calculations were performed with essentially the same
code, the estimations of the average skin friction coefficient were different. In these
preliminary calculations of laminar flows over L-grooves, the main purpose was in
fact to deal with grid independence rather than grid refinement. vVhen dealing with
parametric study, for h/s=0.5, h=0.25mm and ny = 80, then (6.Yc)m.;n ~ 1.76 10- 3 h.
154
-2 ~~~,.-.~~~~
. ~
Q
I
l1F
I i i
r
F(%)
0 I
0 I Q !I
i
0 I
T 0
I
-4
r
-~~n
hi 0= 0.042
! ...
hi 0= 0.083
,,
hi 0=0.170 i
i I
l
-6 I
s / h
I i~
-8
o 1 2 3 4
Figure 9: Parametric study for L-shaped grooves
\Vhen comparing the mean skin friction coefficient at the trailing edge of the riblet
model to the one obtained at the same location over a smooth plate, one could plot the
percentage of friction drag variations, 6.F /F, versus the aspect ratio s/h for different
values of b//5 (figure 9). For some values of this aspect ratio, a couple of symbols are
present if variations on the friction force have been recorded by increasing ny from 60
to 80 : COl1\'crgence might not be completely achieved but is not very far from being ...
For all the cxamined configurations, small friction drag decreases were obtained. It
seems that:
- for constant s/h. the level of reduction decreases as h/ /5 increases:
- for constant h//5. t.he level of reduction decreases as s/h increases.
Thus, the striking feature is that. weak drag reductions, or more exactly, no
drag increasc has been evidenced when manipulating a laminar boundary layer with
L-shaped grooves. though the increase of wetted area could be t.remendous in some
cases : '100S{ (res]). 200%) when s/h=0 ..5 (resp. l.0). This tendancy is completely
opposite to the one obtained, from a numerical point. of \'iew 1)" Launder and Li [1:3].
in int.ernal channel flows with the same device geometry.
5 Concluding Renlarks
Thus, according to those laminar boundary layer calculations over v- or L-shaped
groO\'es, the skin friction drag is not increased despite the important wetted area
increase; the latter can even be very large especially for L-shaped ribs. Tbe obtained
results hayc been recorded \\'ith various grids which \\'('re refined in regions of large
velocity gradients. Since it is rather difficult to compute these strong gradients with
accuracy, it is more reasonable to conclude only that there is. at least. no drag
155
--/.:?'I'----- ]
external flows
Cf V-groove h/s=O.50 1121
o 1,2 L-groove h/s=1.00
i .,,' j
L-groove h/s=O.50
V -groove h/s=O.67
V-groove h/s=O.40
1,1
: I
I
1,0 ..f.....~-n;::.:""'""'~=6+1~~.~~=.-..+~~==~.y.....=D........=-+==!::,.r==.........,
-+ h/~ = h/o
0,9+-~~~-+1~~~~+-~~~-T~~~~+-~~~--j
0,00 0,05 0,10 0,15 0,20 0,25
reduction. i.e. that riblets have practically no effect on friction, in the laminar regime.
It is beyond doubt that friction enhancements, similar to those computed in a laminar
channel flow, [13], could not be obtained. This is in agreement with experimental
results on V-groo\'es, [7]. The results collected on figure 10 for either L- or V-shaped
riblcts in both illternal and external flows support these conclusions. It should be
ITlentionecl that first results reported by de Saint- Victor, [6], are not plotted on that
graph. Although those indicated a trend similar to the U1IIST and Il\IST ones, the
results might be subject to caution because the mesh grids were not carefully defined
in the regions of high \'elocity gradients, i.e. in the vicinity of the l'iblet edge.
Results reported from 1:\ISl' and CERT computations have shown the numerical
difficult ies stemmed from the singularity at this groo\'e crest, whate\'er riblet-shape si
concerned. It should be noted that additional work is under consideration at Ul\IIST
and Camhriclge as regards greater refinement tests (orthogonal mesh, conformal map
ping ... ,) for various groove cross sections.
In addition, we belie\'e that the analogy, propounded by Launder and Li. [13].
cannot. in fact. demonstrate there is an obvious drag increase to support their con
tent ion that the drag must be increased in fully developed internal flows. This anal
ogy considers "the problem of heat conduction in an infinite plate of uniform thermal
condudi vity "'ithin which a uniform heat generation rate is occuring. The surface
of the plate is maintained at a uniform known temperature. In tbe steady state, ... ,
two situations have to be compared: in one, the plate surface is completely smooth
while, in the other, miniature slits are cut in the surface which are also kept at the
156
uniform cold surface temperature. Thus, for h/ s=l, the effective surface area is three
times that of the plate without slits. Suppose the same internal heat.-generation rate
applies in each case. In which of them will the average temperature be higher? "
Launder and Li, [13], thus argue that it will obviously be the case of the smooth
plate, since in the slit case, the cooling surface is increased. However, for h/s=3
(resp. 1), the cooling surface is larger by 600% (resp. 200%), whereas calculations
reveal 32% (resp. 27%) heat flux increase, when, for instance, h/H=0.05 (figure 10).
Thus. the heat flux increase is much weaker than the wetted area increase and. there
is no straight relationship between cooling and wetted area. It is, a priori, rather
difficult to know temperature or velocity gradient distributions and, even if it is not
easy to imagine there is maybe a small decrease in wall heat transfer or friction, it
is not obvious - and, certainly, not demonstrated - that there is an increase. In fact,
there is a large region where gradients are negligible, [10], and their distribution is
such that enlargement in the crest vicinity is practically balanced by strong damping
in the valley. III addition to that, experiments performed by Rohr et al [16] in a fully
de\'eloped pipe flow suggested that riblets (V-shaped grooves, s/h=l, h/H=O.Oll and
0.02-1 where II denotes the pipe radius) "while greatly increasing the wetted surface
area, redistribute wall shear stress to provide both: no drag increase in laminar flow
and drag reduction in certain regimes of turbulent flow".
Thus, because of the velocity gradient redistribution within V- or L-shapecl
riblets (which may be also true for other geometries), the wetted area increase does
not induce extra skin friction in laminar boundary layer; in fully developed internal
laminar flows, measurements and computations give conflicting results. In turbulent,
as well as laminar boundary layers, the increase of wetted area does not have to
he considered. However, one should not forget the basic difference between fully
developed internal flows and boundary layer flows : there is no acl\'ection in the
former ones. This feature could be essential in laminar regime since the skin friction
coefficient represents an integral effect and not a local one. Looking at optimal
geometries one could, then, imagine that the wall viscosity effect is to replace the solid
wall by a fluicl wall located in the crest plane which would induce the same friction
but would smooth turbulent perturbations; there would then be a rearrangement of
wall struct ures which would tend to lower turbulence production and, consequently,
overall friction.
Acknowledgements
These numerical studies performed, both, at Il\IST and ONERA/CERT have been
supported by the "Service Technique des Programmes Aeronautiques".
References
[1] Bacher E.V., Smith C.R. : AIAA Journal, Vol. 24, W8, 1986, pp. 1382-1385.
[2] Bechert D.W., Bartenwerfer 1\1. : J. Fluid Mech., Vol. 206, 1989, pp. 105-130.
[:3] Bushnell D.M. : AGARD Rep. 723, 1985, pp. 5.1-5.26.
[4] Choi 1(.S. : J. Fluid Mech., Vol. 208, 1989, pp. 417-458.
[5] Coustols E. : AIAA Paper 89-0963 (1989).
[6] de SainL-Victor X. : CERT Internal Technical Report (1987).
[7] Djenidi L. : Ph.D. Dissertation (1989).
157
[S] Djenidi 1., Liandrat J., Anselmet F., Fulachier L. : 2nd European Turb. Conf.
Berlin, (19SS).
[9] Djenidi 1., Liandrat J., Anselmet F., Fulachier L. : Applied Scientific Research,
Vol. 46, 19S9, pp. 263-270.
[10] Djenidi L., Liandrat J., Anselmet F., Fulachier L. : Drag Reduction in Fluid
Flows. 4th lnt. ConL on Turbulent Drag Reduction, Davos, 19S9, pp. 3.5-4l.
[11] Fulachier L., Djenidi L., Anselmet F. : LM.S.T Report (19S7).
[12] Khan M.M.S. : AlAA Paper S6-1127 (19S6).
[13] Launder B.E., Li S. : Applied Scientific Research, Vol. 46, 19S9, pp. 271-2S0.
[14] Lowson M.V., Bates J.H.T., Jewson A.R. : Drag Reduction in Fluid Flows, 4th
Int. Conf. on Turbulent Drag Reduction, Davos, 19S9, pp.77-S.3.
[15] Pulles C.J.A., Prasad K.K.,Nieuwstadt F.T.M. : Applied Scientific Research,
Vol. 46, 19S9, pp. 197-20S.
[16] Rohr J., Anderson G.W., Reidy 1.W. : Drag Reduction in Fluid Flows. 4th Int.
ConL on Turbulent Drag Reduction, Davos, 19S9, pp.263-270.
[17] Schneider G.E., Zedan }.II. : Numerical Heat Transfer, Vol. 4, 19S1.
[IS] Van DOOl'maal J.P., Raithby G.D. : Numerical Heat Transfer, Vol. 7, 19S4.
[19] Walsh l\U., Lindeman A.M. : AIAA Paper S4-0347, (19S4).
[20] Walsh "\l.J., Anders Jr. J.B. : Applied Scientific Research, Vol. 46, 19S9, pp.
255-262.
[21] Wilkinson S.P., Anders J.B., Lazos B.S., Bushnell D.M. : Int. Conf. on Turbulent
Drag l1cduction by passive means, London, 19S7.
TURBULENT BOUNDARY LAYER OVER A RIBLETED
SURFACE WITH TANDEM MANIPULATORS
USING SURF ACE DRAG BALANCES
ABSTRACT
Low speed wind tunnel tests have been carried out to assess the combined drag
reduction effects of longitudinally grooved or "ribleted" surfaces and upstream tandem
turbulence manipulators or LEBUs (Large Eddy Break-Up devices). Skin friction and
surface drag measurements are presented covering the regions upstream, over and
downstream of the ribleted plate, both with and without LEBUs. Surface drag surveys
using servo-controlled, floating element balances indicated non-linear drag reduction
effects when the two devices are used together. On their own the asymmetrically shaped
triangular riblets produced an average drag reduction of 5% over the ribleted section
based on local direct balance measurements.
NOMENCLATURE
159
160
1. INTRODUCTION
The energy crisis of the early 1970's stimulated a great deal of research and
development work on the aerodynamic drag reduction of both the wings and fuselages of
aircraft. Two particularly innovative techniques were successfully pioneered by NASA
Langley and associated groups; i.e. "riblets", consisting of suitably contoured
longitudinally grooved surfaces, and "Large Eddy Break-Up devices" or LEBUs, formed
by transverse bands or aerofoils mounted parallel to the surface inside the turbulent
boundary layer (cf. [1],[2]). Recent international meetings have high lighted the world-
wide research effort in these two areas ([3],[4],[5],[6]), both from the applied and the
more fundamental phenomenological point of view. There now seems to be some
agreement that drag reductions of up to 6-8% are possible with riblets, at least under
laboratory conditions. Controversy still exists however ([2],[7],[8]) as to whether one
can actually achieve overall drag reductions with LEBUs, taking into account the drag
penalty of the device itself and its support.
Despite extensive efforts to explain the mechanisms by which the riblets and the
LEBU s work, detailed descriptions are still subject to much debate. An excellent analysis
and documentation of the turbulent characteristics of the boundary layer downstream of
the LEBU has been given recently by Lemay in his Ph.D. dissertation [12].
In the present paper we have not attempted to address the detailed aspects of how
the turbulent structure is modified by riblets or LEBUs, but have restricted our efforts to
monitoring local changes in skin friction and surface drag arising from the replacement of
a smooth surface by a riblet section, both with and without the presence of upstream
LEBUs.
Our particular choice of both LEBU sand rib lets was guided by experimental
work, in both the U.S.A. and Europe, aimed at determining the best configurations to
161
achieve maximum overall drag reductions. Much of this work has been well summarized
in various papers presented very recently at the Fourth International Conference on Drag
Reduction held in Davos [6]. It has been suggested that tandem LEBU devices offer the
best possibility of a net overall drag reduction because of reduced parasitic drag due to the
downstream blade lying in the wake of the upstream manipulator. In low speed flows it
appears that the best position for the LEBU s is in the outer part of the boundary layer (hlo
'" 0.8), whilst in the subsonic, compressible regime (M '" 0.8) they should be placed
nearer the centre (h/o '" 0.5).
In the case of riblets, early work at NASA Langley demonstrated that the key
parameters were the height (h) and spacing (s) Reynolds numbers h+ and s+, based on
the local friction velocity U t (i.e. h+ = U'thlu and s+ = U'ts/u). The general consensus
is that small but significant reductions in smooth wall skin friction are possible with the
appropriate scaling of these two parameters (e.g. h+ < 10 and s+ < 30 : cf. figure 10).
We thus chose a riblet height of 0.2 mm, with a pitch of 0.5 mm, corresponding to
"optimized" h+ and s+ values of around 8 and and 20 for our mid-speed tunnel
conditions.
One of the principal objectives of our investigation was to provide direct local
measurements of both surface drag and skin friction, using a series of self-nulling, servo-
controlled, floating element balances developed over a number of years in the Mechanical
and Electrical Engineering Departments at Laval University. Such measurements would
seem to be particularly appropriate in the case of "rough" or "ribleted" surfaces, where
one might expect that traditional "wall law" techniques for determining surface drag
would be adversely influenced by uncertainties regarding the "effective origin" of the
"equivalent smooth surface".
2. EXPERIMENTAL ARRANGEMENTS
All the tests were conducted in the Laval 61 x 46 x 480 cm low speed wind tunnel
which has been extensively used in a number of previous drag reduction studies
([13],[14],[15]). Figure 1 illustrates this wind tunnel which possesses an IBM-PC
controlled automatic traversing mechanism, data acquisition system and speed control,
described in the previous references. The data presented in the previous paper and also at
the 1st Canadian Symposium on Aerodynamics ([19]), covers a first series of local skin
friction and surface drag measurements, involving both riblets and LEBU s. The
measurements were made in a boundary layer developing along the floor of the tunnel
without the use of any tripping device.
162
The tests covered a 2.5 metre section downstream of the trailing edge of the rear
LEBU blade (cf. figure 2), corresponding to approximately one hundred boundary layer
thicknesses (~ = x~/Di = 100: 0i = 21.5 mm: 0i being the boundary layer thickness at x~
= 0). The leading edge of the ribleted plate, 89 cm in length (~ = 41.4), was situated
some 45.5 cm (~ = 21.2) downstream of the tandem LEBU. The riblets were preceeded
and followed by smooth plates, with the riblet peaks mounted flush with the upstream
and downsteam smooth sections. Both the smooth and ribleted plates were equipped
with removeable plugs permitting the insertion of the skin friction and surface drag
balances. Initial measurements over an instrumented smooth plate at the riblet test section
provided the comparative smooth wall data. The main tests were carried out with a free
stream velocity of 16 metres/sec at the entrance of the working section.
The LEBUs used in this experiment were a pair of thin steel blades in tandem,
each with a thickness of 0.15 mm (0.0070i), a chord length of 25 mm (1.l60i) and a
spacing of l30mm (6.050i: cf. figure 3). They were placed at the same position in the
tunnel as used by Lemay et al. [14], who employed manipulators with a chord length of
19 mm and a thickness of 0.4 mm. We retained the same spacing and heights of the
manipulators in the boundary layer (h/Oi = 0.32, 0.48 and 0.75) as in Lemay's
experiments (ref. [12]).
Our original intention had been to machine a symmetrical v-shaped riblet pattern
on an aluminium plate. However, a misalignment of the cutting tool resulted in the
asymmetrical saw-tooth wave shape illustrated in figure 4, which we decided to test. The
base angles of the riblets were 60 and 30, with height (h) and spacings (s) respectively
of 0.2 mm and 0.5 mm. As mentioned earlier, the latter dimensions were chosen to
provide a range of spacing and height parameters susceptible to produce the best possible
drag reductions over the ribleted surface based on work by NASA Langley and other
investigators (i.e. s+ = D'Cs/u < 30 and h+ = D'Ch/u < 10: [9]), and compatible with our
range of tunnel speeds (8 < De < 24 rn/s). The plate possessed five inserts for mounting
the floating element drag balance. The ribleted plate, the four removeable plugs, the
floating element and the surrounding surface of the balance casing (cf. figure 6) were all
machined at the same time, to achieve an identical surface geometry and good "mountain-
valley" matching between the various elements.
The standard twin-floating head balance illustrated in figure 5 was used for the
smooth wall measurements, whilst two special single headed versions were built for the
ribleted, "rough" wall tests (cf. figure 6). In the first series of tests, described here, the
163
larger square floating head (25 x 25 mm) balance was employed. The basic principle
behind all the balances is the same. The floating elements (#11 and 19: figure 5) are
mounted on four flexures (#17), which are very stiff in the transverse direction, but free
to move in the longitudinal direction aligned with the skin friction, or surface drag forces
to be measured. Under the action of perturbing forces the floating head starts to move.
This movement is immediately detected by a linear voltage differential transformer
(LVDT:#13), which provides the error signal to drive a small linear motor (#12) forming
part of a closed loop, self-nulling, servo system, with the current in the motor coil
directly proportional to the perturbation force acting on the floating head. Due to the very
small weight of the floating elements (e.g. 1.5 gm for the standard 28.6 mm dia. smooth
head), the balances can be calibrated directly by mounting them in the vertical position,
attaching weights (10, 50, 100, 500 mg etc) and reading the output voltage across a
precision resistor in series with the motor coil. The calibration factor is adjusted to give a
1 millivolt output for 1 mg force. All the balances possess a sensitivity of better than 1
milligram, with a full scale range of over 5 gf. Fuller details can be found in reference
[13] and by referring to the assembly drawings and corresponding nomenclature
illlustrated in figure 5.
The distribution of Cf with the riblets alone is shown in figure 8, where the
surface drag and skin friction data has been normalized with respect to Cfo, the skin
friction coefficient over the smooth plate alone (cf. figure 7). A nearly constant reduction
in Cf of around 5% ( 2%) was registered over the ribleted section, with relaxation to the
smooth wall data some 25-30 boundary layer thicknesses downstream.
Results of the riblets and LEBU s combined, compared with both the rib lets and
LEBU alone are also illustrated in figure 11 respectively for the three different heights of
the LEBU. The data does not seem to indicate a simple linear additive effect. Ahead of
the ribleted plate the combined effect seems to be worse, whilst downstream the drag
reduction is improved, with an apparently much longer relaxation length. These effects
do seem rather suspicious. Behind the riblet surface, where the effects of the riblets on
their own die out fairly quickly, one might expect the "combined" drag reduction to
picture rather closely that of the LEBU alone. Further tests are underway to investigate
this apparent anomaly. Tests will be repeated over a range of tunnel speeds with the two
drag balances illustrated in figure 6 and their smooth headed counterparts.
4. CONCLUSIONS
-The standard method of determining the overall drag reduction of the plate from
momentum thickness measurements at the leading and trailing edges of the ribleted plate
produced very scattered data, with rather optimistic drag reductions compared with the
balance data. We tend to place more confidence in the direct surface drag measurements,
than the momentum thickness estimates which could be adversely affected by similar
factors to those influencing the Von Karman momentum equation technique.
-The combined effects of the LEBUs and rib lets were roughly additive over the
region of the ribleted plate, but the data indicated a much longer downstream relaxation
length than one might expect from the single manipulator data. Further experiments are
presently underway to determine whether this is actually a true picture. In particular,
besides repeating the skin friction and surface drag measurements, we would like to
monitor carefully variations, or changes in the "two-dimensionality" of the flow over the
length of the tunnel.
REFERENCES
[1] Bushnell, D.M .. "Turbulent drag reduction for external flows", AGARD-Report
No 723 on Aircraft Drag Prediction and Reduction, 1985.
[2] Walsh, M.J., Anders, I.B., "RibletlLEBU research at NASA Langley", Applied
Scientific Research, Vol. 46, No.3, July 1989.
[3] Savill, A.M., Truong, T.V., Ryhming, I.L., "Turbulent drag reduction by
passive means: a review and report on the first European drag reduction meeting",
Journal of Theoretical and Applied Mechanics, Vol. 7, No.4, 1988.
[5] Coustols, E., Savill, A.M., "Resume of important results presented at the Third
Turbulent Drag Reduction Working Party", Applied Scientific Research, Vol. 46, No.3,
July 1989.
[7] Anders, J.B., "Large-Eddy breakup devices as low Reynolds number airfoils",
SAE Paper No. 86-1769, presented at the Aerospace Technology Conference and
Exposition, Long Beach (CA), October 1986.
[8] Anders, I.B., "LEBU drag reduction in high Reynolds number boundary layers",
AIAA Paper No. 89-1011, AIAA 2nd Shear Flow Conference, Tempe (AZ), March
1989.
166
[9] Walsh, M.J., Lindemann, A.M., "Optimisation and application of riblets for
turbulent drag reduction", AIAA Paper No. 84-0347, 1984.
[10] Savill, A.M., Koury, E., Sebastian, W., Squire, L.c., "Combined effect of
manipulators and riblets on turbulent boundary layer skin friction", Fourth International
Conference on Drag Reduction, Davos, Switzerland, July 1989.
[13] Nguyen, V.D., Dickinson, J., Jean, Y., Chalifour, Y., Anderson, J., Lemay, J.,
Haeberle, D., Larose, G., "Some experimental observations of the law of the wall behind
large-eddy-breakup devices using servo-controlled skin friction balances", AIAA Paper
No. 84-0346, January 1984.
[14] Lemay, J., Provenr;al, D., Gourdeau, R., Nguyen, V.D., Dickinson, J., "More
detailed measurements behind turbulence manipulators including tandem devices using
servo-controlled balances", AIAA Paper No. 85-0521, March 1985.
[15] Pineau, F., Nguyen, V.D., Dickinson, J., Belanger, J., "Study of flow over a
rough surface with passive boundary layer manipulators and direct wall drag
measurements", AIAA Paper No. 87-0357, 1987.
[16] Beauchamp, C.H., Philips, R.B., "Riblet and polymer drag reduction on a
axisymmetric body", presented at the Symposium on Hydrodynamic Performance
Enhancement for Marine Applications, Newport Rhode Island, Oct. 1988.
[17] Reidy, L.W., Anderson, G.W., "Drag reduction for external and internal
boundary layers using riblets and polymers", AIAA Paper 88-0138, Jan. 1988.
[19] Nguyen, V.D., Dickinson, J., Jean, Y., Chalifour, Y., Smaili, A., Page, A.,
Paquet, F., "Boundary layer over a ribleted surface with tandem turbulence manipulators
using surface drag balances", Proceedings, 1st Canadian Symposium on Aerodynamics,
Ottawa, December 1989
167
Figure 1: The Laval 2'x 1 1/2' low speed wind tunnel: dimensions in mm.
2270 mm - - - - - - I...
Figure 2: Position of the LEBU s, ribleted plate and various test stations.
t
~ -----lh'-
130mm
(6.050i)
\\\\\\\\\\\\\\\\\\\\\\\\\\\~\\\\\
\_-----ill
'---------{.y
@-----
Figure 6: Self-nulling, servo controlled drag balances for riblet investigation with their
associated electronic control unit: present data obtained with the larger (25x25mm),
square floating head.
170
3.5.,.-------------------,
lJ. prest. 1
o prest.2
+ bal.
o
o
~ 3.0
)(
o
U
2.5 - t - - - . . - - - - . - - - - r - - - - , , - - - - - . - - - - 1
o 50 100 150
J : x/deltal
.t
Figure 7: Skin friction variation, Cf/Cfo vs x/Oi ' along the completely smooth test
./
section: 0i =21.5 mm, the boundary layer thickness at the end of the last LEBU blade:
L'l. Preston tube: 0 Preston tube (Lemay [4]): + Balance data.
Table: Variation of 0, Re and Ue versus x/oi :.:: J'.
J
171
I. i l
Riblets
Riblets
.
1.00
0.90
,.
I I
,
o 50 100 150
5: ..~deltai
Figure 8: Local drag coefficient Cf along test section compared with smooth wall data
Cfo: balance data.
..
10
f:. exp.1
.
/:::,. exp. 2
- 0
.. . i .. " ~! .
~ /:::,.
><
~ .. /:::,. f:.
U
<1. 10 f:.
/:::,.
f:.
/6 IS ,sT ZO
20 4S'
4 6 10
h+
Figure 9: Overall drag reduction, i1Cd ICd vs h+, on the ribleted plate based on
momentum thickness analysis (section 3.2)
~~
(> Reidy et 01 [17]
!::, Liu et 01
1.05 Coustols [18]
.2
(.)
B
1.00
0.95
t~t~ Low-speed film dolo
4% bond
0.90 0
10 20 30
s+
Figure 10: Low speed drag data for riblets: M.J. Walsh & J.B. Anders; [2] p. 256
172
1.1
A Riblets
I.,
ALP.32
+
0
I".! .-;1
1.0 .A
Riblets 06~R ALB.32
.~ ALR.32
+
,g 0.9
b 0 AI + +
( a)
u ~ + + +1
t I + I
+
I
0.8
0
I
0.7
0 50
.s :x1deltai
.t
100 150
A
1.1
Riblets
.. Riblets., ALP.48
AI
0
A6A
1.0
OA ,I I'.i~ + ALB.48
ALR.48
,g IA ~!~AI + (b)
g 0.9
+ +
u Ii I ++ +
0.8 If 1+ + + + +1
0.7
0 50 100 150
S' lfIdeltai
~
1.1
A Riblets
0 ALP.75
.. Riblets
i "1 AAA + ALB.75
1.0 .A I I ALR.75
I. Gl + +1
0.8 -I-_~=;;;:;;;;;;;;;L---r_ _.--_...--~
o
t: ~deltai
50 100 150
Figure 11: Cf/Cfo vs fBi measurements: A riblets: LEBU alone: + LEBU plus
riblets: 0 LEBU alone (Preston tube) (a) h/oi = 0.32 (b) h/oi = 0.48 (c) h/oi = 0.75
List of referees
173
Mechanics
From 1990, books on the subject of mechanics will be published under two series.
FLUID MECHANICS AND ITS APPLICATIONS
Series Editor: R. Moreau
Aims and Scope of the Series
The purpose of this series is to focus on subjects in which fluid mechanics plays a fundamental
role. As well as the more traditional applications of aeronautics, hydraulics, heat and mass transfer
etc., books will be published dealing with topics which are currently in a state of rapid develop-
ment, such as turbulence, suspensions and multiphase fluids, super and hypersonic flows and
numerical modelling techniques, It is a widely held view that it is the interdisciplinary subjects that
will receive intense scientific attention, bringing them to the forefront of technological advance-
ment. Fluids have the ability to transport matter and its properties as well as transmit force,
therefore fluid mechanics is a subject that is particularly open to cross fertilisation with other
sciences and disciplines of engineering, The subject of fluid mechanics will be highly relevant in
domains such as chemical, metallurgical, biological and ecological engineering. This series is
particularly open to such new multidisciplinary domains.
1. R.T. Haftka, Z. GUrdel and M.P. Kamal: Elements of StrIlclllral Optimi:ation. 2nd rev.ed.,
1990 ISBN 0-7923-0608-2
2. J.J. Kalker: Threedimel1siol1ai Elastic Bodies. 1990 ISBN 0-7923-0712-7
1. M.A. Krasnoselskii, P.P. Zabreiko, E.L Pustylnik and P.E. Sbolevskii: Integral
Operators in Spaces of Sllmmable Functions. 1976 ISBN 90-286-0294-1
2. V.V. Ivanov: The Theory of Approximate Methods and Their Application to the
Numerical Solution of Singular Integral Equations. 1976 ISBN 90-286-0036-1
3. A. Kufner, O. John and S. Pucfk: Function Spaces. 1977 ISBN 90-286-0015-9
4. S.G. Mikhlin: Approximation on a Rectangular Grid. With Application to Finite
Element Methods and Other Problems. 1979 ISBN 90-286-0008-6
5. D.G.B. Edelen: Isol'ector Methods for Equations of Balance. With Programs for
Computer Assistance in Operator Calculations and an Exposition of Practical Topics of
the Exterior Calculus. 1980 ISBN 90-286-0420-0
6. R.S. Anderssen, F.R. de Hoog and M.A. Lukas (eds.): The Application and Numerical
Soillfion of Integral Equations. 1980 ISBN 90-286-0450-2
7. R.Z. Has'minskil: Stochastic Stability of Differential Equations. 1980
ISBN 90-286-0100-7
8. A.I. Vol'pert and S.L Hudjaev: Analysis in Classes of Discontinuous Functions and
Eqllations of Mathematical Physics. 1985 ISBN 90-247-3109-7
9. A. Georgescu: Hydrodynamic Stability Theory. 1985 ISBN 90-247-3120-8
10. W. Noll: Finite-dimensional Spaces. Algebra, Geometry and Analysis. Volume 1. 1987
ISBN Hb 90-247-3581-5; Pb 90-247-3582-3
MECHANICS OF CONTINUA
Editors: W.O. Williams and G.fE. Oravas
1. G.C. Sih (ed.): Methods of Analysis and Solutions of Crack Problems. 1973
ISBN 90-01-79860-8
2. M.K. Kassir and G.C. Sih (eds.): Three-dimensional Crack Problems. A New Solution
of Crack Solutions in Three-dimensional Elasticity. 1975 ISBN 90-286-0414-6
3. G.c. Sih (ed.): Plates and Shells with Cracks. 1977 ISBN 90-286-0146-5
4. G.c. Sih (ed.): Elastodynamic Crack Problems. 1977 ISBN 90-286-0156-2
5. G.c. Sih (ed.): Stress Analysis of Notch Problems. Stress Solutions to a Variety of
Notch Geometries used in Engineering Design. 1978 ISBN 90-286-0166-X
6. G.c. Sih and E.P. Chen (eds.): Cracks in Composite Materials. A Compilation of Stress
Solutions for Composite System with Cracks. 1981 ISBN 90-247-2559-3
7. G.c. Sih (ed.): Experimental Evaluation of Stress Concentration and Intensity Factors.
Useful Methods and Solutions to Experimentalists in Fracture Mechanics. 1981
ISBN 90-247-2558-5