Documente Academic
Documente Profesional
Documente Cultură
a r t i c l e
i n f o
Article history:
Received 12 October 2012
Received in revised form 18 February 2013
Accepted 13 March 2013
Available online 11 April 2013
Keywords:
Wall turbulence
Pipe ow
Curvature effects
Reynolds-stress budgets
Coiled tube
a b s t r a c t
Fully developed, statistically steady turbulent ow in straight and curved pipes at moderate Reynolds
numbers is studied in detail using direct numerical simulations (DNS) based on a spectral element discretisation. After the validation of data and setup against existing DNS results, a comparative study of turbulent characteristics at different bulk Reynolds numbers Reb = 5300 and 11,700, and various curvature
parameters j = 0, 0.01, 0.1 is presented. In particular, complete Reynolds-stress budgets are reported
for the rst time. Instantaneous visualisations reveal partial relaminarisation along the inner surface of
the curved pipe at the highest curvature, whereas developed turbulence is always maintained at the
outer side. The mean ow shows asymmetry in the axial velocity prole and distinct Dean vortices as secondary motions. For strong curvature a distinct bulge appears close to the pipe centre, which has previously been observed in laminar and transitional curved pipes at lower Reb only. On the other hand, mild
curvature allows the interesting observation of a friction factor which is lower than in a straight pipe for
the same ow rate.
All statistical data, including mean prole, uctuations and the Reynolds-stress budgets, is available for
development and validation of turbulence models in curved geometries.
2013 Elsevier Inc. All rights reserved.
1. Introduction
Turbulent ow in curved pipes is frequently occurring in a variety of industrial applications. Typical prominent examples are heat
and mass transfer systems where straight, bent and helically coiled
pipes are encountered in heat exchangers, chemical reactors, pipeline systems as well as components of internal combustion engines
(e.g. exhaust manifolds). Similarly, biological systems such as the
blood ow in arteries or the air ow in the respiratory system,
are also occurring mostly in bent geometries. A better understanding of the physical mechanisms in action and improved ability to
accurately model the specic uid phenomena would help to improve the performance of such devices; e.g. improved heat and
mass transfer coefcients and enhanced cross-sectional mixing,
and reduced axial dispersion, as for instance discussed by Vashisth
et al. (2008).
Compared with other (canonical) internal ows such as ow in
straight pipes and ducts, the curved pipe conguration has been
studied in detail mainly for laminar ow. In general, the curvature
causes the appearance of centrifugal forces which deects the axial
maximum velocity away from the centre towards the outer side of
the curved pipe. At the same time, the curvature gives rise to a secondary motion in the cross-section of the bent pipe, essentially due
Corresponding author.
E-mail address: azad@mech.kth.se (A. Noorani).
0142-727X/$ - see front matter 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatuidow.2013.03.005
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
17
18
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
2. Computational methodology
2.1. Flow conguration and governing equations
Fig. 1 (top) shows a schematic view of a curved pipe together
with its Cartesian (X, Y, Z) and toroidal (R, s, f) coordinate systems.
A relevant quantity in this conguration is the curvature parameter
j dened as Ra/Rc; Ra is the radius of the pipe cross-section and Rc
is the radius of curvature at the pipe centreline. This dimensionless
parameter distinguishes between mildly (j 0.01) and strongly
curved pipes (j 0.1).
The governing incompressible NavierStokes equations,
r u 0;
@u
1 2
u ru rp
r u
@t
Reb
are integrated in time for the velocity vector u and the pressure
pReb is the bulk Reynolds number based on the bulk velocity ub
and the pipe diameter D = 2Ra, i.e. Reb = 2Raub/m with the kinematic
viscosity m. The dimensionless Dean number is given by
p
Deb Reb j and is widely used in the literature to characterise
the secondary motion, generally in the laminar regime. Throughout
the present study, the horizontal and vertical cuts are referred to as
azimuthal in-plane positions of h = 0 and p/2, respectively; see the
sketch in Fig. 1 (bottom).
2.2. Numerical approach and parameters
The simulations are performed using the massively parallel
spectral-element method (SEM) solver nek5000. This code has
been developed by Fischer et al. (2008), and it solves the incom-
Fig. 1. (Top) embedded in-plane polar (r, h) and Cartesian (x, y) coordinates along
with side view of ow conguration and associated coordinate systems, (X, Y, Z) as
reference orthogonal coordinate triad and (R, s, f) as attached toroidal cordinate
system. (Bottom, left) Cut-away section of a curved pipe, with locations of the
horizontal and equatorial cross-sections indicated, h= 0 and h= p/2, respectively.
(Bottom, right) Front view with torodial coordinate system.
Fig. 2. (Left) Mesh conguration in the equatorial mid-plane of the toroidal pipe for
the higher Reynolds number cases. Element boundaries and their GaussLobatto
Legendre (GLL) points are also shown in the darker region. (Right) A view of a
quarter of the cross-sectional plane pertaining to the mesh for the same case.
19
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
15000
Turbulent
Chaotic
Reb
10000
QuasiPeriodic
5000
Laminar
0.05
0.1
0.15
0.2
Table 2
Various simulation parameters for the present study. Case names are coded based on
the bulk Reynolds number and domain congurations (S, as straight, M, as mildly
curved, and H, for strongly curved pipes). The time t is normalised by Ra/ub, starting
from the point that the ow has reached a statistically stationary state.
Case
Reb,D
Res
Sampling period
R5300S
R5300M
R6900S
R6900M
R11700S
R11700M
R11700H
5300
5300
6926
6926
11,700
11,700
11,700
0.00
0.01
0.00
0.01
0.00
0.01
0.10
180
176
228
230
360
368
400
400
2800
400
2800
400
1000
1000
0.25
Fig. 3. Tentative ow regime map following Di Piazza and Ciofalo (2011), - - - transition from stationary to quasi-periodic regime, -- transition to chaotic state,
transition to fully developed turbulent ow. j Parameters for present DNS. h DNS
performed yielding quasi-periodic state but not further discussed here.
pipe congurations with xed Reb are considered such that the
ow rate is above the critical value given by Ito (1959) for laminar-turbulent transition. Fig. 3 shows the aforementioned tentative
Rebj map by Di Piazza and Ciofalo (2011). For our DNS, we chose
three Reynolds numbers, Reb = 5300, 6900 and 11,700; for each Reb
the curvature parameters are chosen such that the fully turbulent
regime is obtained, as indicated in Fig. 3. We have also performed a
simulation with Reb = 5300 and high curvature j = 0.1, which
reached, as expected, the quasi-periodic state. We believe that this
state is quite regular rather than genuinely turbulent and therefore
we do not further study this parameter range.
At a given Reynolds number, the grid resolution of the curved
pipe congurations is chosen to be the same as that for the straight
pipe. Apart from the validation case (Reb = 6900, see Section 3.2)
the length of the domain for all cases is set to be 25Ra along the
pipe centreline. The grid spacing, measured in wall units where
the viscous length scale is based on the straight pipe friction Reynolds number at each Reb, is set such that Dr
max 6 5 with four grid
points placed below Dr+ = 1, measured from the wall, and
DRa hmax 6 5 and Dsmax 6 10. Viscous scaling is indicated by the
superscript +. The details of the computational meshes at the various Reynolds numbers are presented in Table 1.
The ow is homogeneous in the axial direction (s) and statistically steady in time (t). Therefore, all statistical data are averaged
in these two directions, and the nal average is denoted by hi. In
the straight pipe, azimuthal averaging is also used. Sampling of statistics is started after the ow has settled in the developed turbulent regime approximately after 200Ra/ub in each case. Different
statistical averaging time periods are used for the various cases
depending on the observed convergence of the data. This and other
relevant simulation parameters are listed in Table 2. A brief outline
of the procedure on how to obtain statistical averages via tensor
transformations under rotation from Cartesian coordinates to
toroidal coordinates is given in Appendix A.
In the
p
present study the mean friction velocity us is dened as
sw;s =q where sw,s indicates the streamwise component of the
mean wall shear stress and an overbar denotes averaging along
the circumference of the pipe section. The azimuthal component
of the wall shear stress sw,h is not used as a normalisation parameter since its magnitude is small compared to the streamwise component. Based on that, the friction Reynolds number is dened as
Res = usRa/m. For the purpose of this paper, the friction factor refers
to the Fanning coefcient and is computed based on the following
denition:
2
sw;s
Res
:
8
2
Reb
qub
2
f 1
is Dr
min ; Dr max 0:26; 6:74, DRa hmin ; DRa hmax 1:45; 7:31
and Ds
;
D
s
3:45;
11:27.
This
resolution
is very similar
max
min
to what will be used for the new cases, as discussed in Section 2.2.
The mean axial velocity and turbulence kinetic energy proles
(plotted in Fig. 5) show very good agreement with the literature
data of Httl and Friedrich (2001). As we xed the bulk velocity
Table 1
Resolution details for the present pipe ow meshes used for the two Reynolds numbers under consideration.
Reb
# Of elements
# Grid points
Dr +
DRah+
Ds+
5300
11,700
36,480
237,120
18.67 106
121.4 106
(0.14, 4.44)
(0.16, 4.70)
(1.51, 4.93)
(1.49, 4.93)
(3.03, 9.91)
(3.03, 9.91)
20
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
+
()
w,s
1.5
0.5
/2
3/2
Fig. 6. Circumferential distribution of the local axial wall shear stress normalised
by its averaged value. Straight pipe, - - - - R11700M, -- R5300M,
R11700H. Note that the equatorial symmetry is not being used.
0.2<us> ,k
+ +
4
3
2
1
0
1
0.5
0.5
r/R
Fig. 5. Mean axial velocity component husi+ normalised by us and turbulent kinetic
energy k+ at the equatorial mid-plane of the toroidal pipe, with r/Ra = 1 denoting
the inner side and r/Ra = +1 indicating the outer side. DNS data by Httl and
Friedrich (2001), - scaled mean streamwise velocity, - - - - turbulent kinetic
energy case R6900M.
Fig. 7. (Left) vector plot of mean secondary ow at R11700M. (Right) (left section)
contour plot of mean in-plane velocity at R11700M, (right section) Contour plot of
mean axial velocity at R11700M.
21
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
25
20
15
10
5
25
Fig. 8. (Left) Vector plot of mean secondary ow at R11700H. (Right) (left section)
Contour plot of mean in-plane velocity at R11700H, (right section) Contour plot of
mean axial velocity at R11700H.
15
<u >+
20
10
5
0
1
0.5
0.5
r/R
core region found near the inner side of the pipe from a high-speed
region close to the outer bend. This sudden change is in qualitative
agreement with the data by Boersma and Nieuwstadt (1996), Httl
and Friedrich (2000), and Httl and Friedrich (2001) at similar curvature. The experimental results by Webster and Humphrey
(1993) also show a similar quantitative behaviour in this region.
However, it should be noted that the mentioned literature data
are all at lower Reynolds numbers, or even in the transitional regime. Furthermore, similar structures are commonly observed in
the pipe centre in laminar curved pipe ows at higher Dean numbers. The existence of such a feature at higher Re in turbulence is
thus remarkable, and is clearly linked to the bulge in the streamlines discussed previously. Meanwhile, the mean radial velocity
component is displayed in Fig. 11. For low curvature parameters,
the radial velocity increases monotonically from the walls towards
a local maximum that is located in the lower section of the pipe
whereas the main effect of high curvature is the presence of a local
0.7
0.6
<ur>+
0.5
0.4
0.3
0.2
0.1
0
1
0.5
0.5
r/R
Fig. 11. Normalised mean radial velocity component at the vertical cut. For legend
see Fig. 10.
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
0.5
1
1
k
2
1
0
1
0.5
0.5
r/R
Fig. 12. Normalised turbulent kinetic energy along the vertical cut of the pipe
section. For legend see Fig. 10.
0.5
Fig. 13. Normalised Reynolds shear stress component hu0s u0r i prole along the
vertical mid-plane. For legend see Fig. 10.
7
6
5
4
3
2
1
0
1
0.5
0.5
r/R
Fig. 14. Normalised rms uctuations of the pressure along the vertical mid-plane.
For legend see Fig. 10.
@ 0 0
hu u i C ij Pij eij Psij Pdij Dij T ij
@t i j
C ij huk i
0.5
r/R
5
4
<u u >
0.5
rms
22
D
E
@ u0i u0j
@xk
0 0 @huj i D 0 0 E @hui i
Pij ui uk
uj uk
@xk
@xk
0
0
@ui @uj
eij 2m
@xk @xk
@u0j
@u0
1
Psij
p0 i p0
q
@xj
@xi
1 @
@ 0 0
Pdij
hp0 u0j i
p ui
@xj
q @xi
2 D
E
@
Dij m 2 u0i u0j
@xk
@ D 0 0 0E
uuu :
T ij
@xk i j k
5a
5b
5c
5d
5e
5f
5g
23
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
The budget terms for the various simulations in this study are
calculated according to the scheme presented in Appendix A. It is
based on tensor transformation and thus the (cumbersome)
expression of these terms in orthogonal helical coordinate system
is avoided. To our knowledge, the complete Reynolds-stress budgets have not been presented before for periodic bent pipes. These,
however, are of great importance to address three-dimensional effects in turbulence models, e.g. how empirical constants in turbulence models might be inuenced by mean streamwise vorticity
(Bradshaw, 1987). Note that the budget terms are inherently
two-dimensional, and all Reynolds-stress components are nonzero. Our results are based on long averaging times, such that
the residuals in the various components are less than 104 in viscous scaling.
8
6
4
2
0
2
4
2
8
6
4
2
0
2
4
2
1
0
2
2
2
2
2
4
2
4
0
1
2
4
2
4
0.5
0.5
r/Ra
rr
2
2
2
2
ss
0
1
8
6
4
2
0
2
4
2
Bk
8
6
4
2
0
2
4
2
rr
ss
Approaching
the pipe centre from the inner bend, the production of u0s u0s is compensated by pressurestrain and dissipation.
The process for the other normal components is similar except that
the balance is solely among pressurestrain and dissipation.
Although the level of turbulent uctuations are stronger at the outer bend, it is interesting to note that the mentioned terms are
decreasing inversely proportional to the distance from the wall
mimicking an ideal logarithmic layer at high Reynolds number
(see Hoyas and Jimnez, 2008).
The contribution of the various budget terms at the outer side of
the strongly curved pipe (Fig. 16 right section) is very similar to that
at mild curvature, Fig. 15, except that now the cross-ow budget
terms are about a factor of two lower, while the terms in the axial
budget are similar in magnitude. The pressurestrain in the axial
direction withdraws energy from streamwise production to supply
Psrr ; Pshh . These terms are the main source of energy for the lateral
and azimuthal components since their own production is insignificantly small. This process occurs in straight pipes and likewise in
mildly curved ones. However, for the strongly curved pipe, the production in the Brr budget has the same magnitude as the pressure
strain. This may indicate that the streamwise production is predominantly converted into azimuthal uctuations.
On the other hand, the budget in the inner side of the highly
curved pipe signicantly differs from that in the straight or mild
curvature congurations. Streamwise velocity uctuations energise the ow between r/Ra = 0.3 and r/Ra = 0.5. Close to the wall,
the ow is intermittent and even partially laminarised, the peak of
production is occurring at about r/Ra = 0.35. This is essentially the
position where the mean radial and streamwise velocity exhibits
high gradients and u0s u0r declines. The location of this process
coincides with the bulge in the middle of the secondary motion
map. Being located far away from the wall, the process is more
homogeneous similar to turbulence in free shear ows, rather than
comparable to near-wall turbulence.
2
1
2
0.5
0.5
r/R
Fig. 15. Pre-multiplied budgets for the normal Reynolds-stress components and the
turbulent kinetic energy k at the equatorial mid-plane of the case R11700M, with r/
Ra = 1 indicating the inner side and r/R = +1 the outer side. + Cij, Pij, - - - eij ; Psij ; Pdij , -- Dij, Tij.
Fig. 16. Pre-multiplied budgets for the normal Reynolds-stress components and the
turbulent kinetic energy k at the equatorial mid-plane of the case R11700H, with r/
Ra = 1 indicating the inner side and r/R = +1 the outer side. For legend see Fig. 15.
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
gain
24
0.4
0.4
0.2
0.2
loss
0.2
0.4
0.2
10
20
30
40
50 50
40
30
20
10
0.4
(1r/Ra)
(1r/Ra)
Fig. 17. Turbulent kinetic energy budget (major components) along vertical cut. Black represents R11700S, blue as R11700M, and red indicates R11700H case. (Left) is the
inner side of the curved pipe, and (right) is the outer side. For line caption see Fig. 15. (For interpretation of the references to colour in this gure legend, the reader is referred
to the web version of this article.)
D
E
2
u0i u0j kdij 2mT hsij i
3
mT C l k2 =e
6a
6b
with dij being the Kroneker delta, hsiji the mean strain rate tensor
dened as 1/2(@huii/@xj + @huji/@xi), and mT the turbulent eddy viscosity.D The turbulent
dissipation rate is dened as
E
e 2m s0ij du0i =dxj , and Cl denotes the turbulent viscosity constant
in the standard ke model. Equating production and dissipation
yields an expression which can be solved for the eddy viscosity,
The present DNS data can now be used to evaluate these terms to
obtain Cl. This scalar value can be essentially considered as nondimensionalised mt and serves as an indicator for examining the
validity of a traditional eddy-viscosity model such as ke. Fig. 18
Fig. 18. Computed Cl from the DNS data for turbulent pipe ow at straight and
curved congurations. (Top) from left to right R11700H, R11700S, R5300S. (Bottom)
from left to right R11700M, R6900M, R5300M. Cl in the straight pipe is circumferentially averaged.
25
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626
2.05
log10(f)
2.1
2.15
3.6. Sub-straight drag
As nal aspect in the present paper, we study the effect of
increasing Reynolds number at a xed curvature, j = 0.01 on the
Fanning friction factor f. Considering the friction factor for straight
(cases R5300S, R6900S and R11700S) and curved pipes with mild
curvature (cases R5300M, R6900M and R11700M) shows the surprising behaviour that f in the bent pipe is not always higher than
in the straight pipe despite the existence of secondary motion. As
illustrated in Fig. 19 (and its blow-up in Fig. 20), with increasing
Reynolds number, all three cases both in straight or curved conguration show a decrease in the friction factor. In cases R6900M and
R11700M the friction coefcient is larger in the curved pipe compared to their straight peer. However, for cases R5300M and
R5300S, the reversed situation can be seen. This is the same observation made by Cioncolini and Santini (2006) based on their pressure-drop experiments. Fig. 20 further shows an excellent
agreement between the DNS data and these experimental results.
To our knowledge this is the rst time that an accurate numerical
computation is conrming this effect. This opens the possibility to
further study this peculiar behaviour, as a sound physical explanation is still not given.
4. Summary and conclusions
Direct numerical simulations have been performed in order to
study the effect of curvature and Reynolds number in fully developed turbulent pipe ows. The newly obtained data has been carefully validated against previous work by Httl and Friedrich
(2001), yielding excellent agreement. The present results extend
the previous studies in both Reynolds number and curvature
parameter, reaching up to a nominal Res = 400 maintaining turbulence at a curvature of up to j = 0.1. We chose a pipe with length
L = 25Ra, which can be considered sufcient to obtain accurate statistics for the various cases.
1.7
log10(f)
1.8
1.9
2
2.1
2.2
3
3.5
4.5
log (Re )
10
Fig. 19. Fanning friction factor (f), straight pipe, coiled pipe at various Reynolds
numbers and j = 0.00964 (data obtained from Cioncolini and Santini, 2006), j DNS
data at different Reynolds numbers for straight pipe (R5300S, R6900S, R11700S), h
DNS data at different Reynolds numbers for curved pipe with j = 0.01 (R5300M,
R6900M, R11700M), - - - - and laminar and turbulent correlations by Ito (1987).
3.7
3.8
3.9
4.1
log10(Reb)
Fig. 20. Close up view of the Fig. 19.
Due to the curvature of the pipe, centrifugal forces are generated in the ow, which lead to a shift of the mean axial ow towards the outer side of the bent pipe. In consequence, the curved
streamlines may develop secondary motion perpendicular to the
primary ow. One pair of counter-rotating vortices are observed
and commonly denoted as Dean vortices. Depending on the values
of Reynolds number, here in terms of the bulk Reynolds number
Reb, and the curvature parameter j, i.e. the ratio of pipe and bend
radii, different regimes can be identied, ranging from intermittent
relaminarisation at low Reb and high j, to weakly modied turbulent pipe ow at high Reb and low j. In this study we concentrate
on those parameters which yield developed turbulent ow,
whereas transitional stages such as the quasi-periodic regime (Di
Piazza and Ciofalo, 2011) are not discussed. Nevertheless we have
observed for our highest curvature partial relaminarisation at the
inner side of the bend, characterised by high atness values of
the velocity signal, but the outer side of the pipe has always been
fully turbulent. Further studies of these relaminarisation processes
and in particular the laminar-turbulent interface might be interesting extensions of the present work.
The effect of the curvature on the secondary motion allows us to
distinguish three rather distinct regimes; zero curvature with vanishing secondary ow, mild curvature with two large counterrotating Dean cells, and strong curvature which causes the core
of these stable vortices to move towards the side walls and a distinct bulge region appears in the pipe centre. Straight and mildly
curved pipes share many similarities in most ow statistics. For
these cases, the mean axial velocity in a vertical cut is close to linear, similar to rotating channel ows.
For the strongly curved pipes, the aforementioned bulge visible
in the secondary ow has also an effect on the mean axial velocity,
which shows a distinct change of slope just below the centre of the
pipe. Together with the relevant shear stress component u0s u0r an
increased production in this region is observed, leading to
enhancement of turbulent kinetic energy in the core of the bent
pipe. This feature has already been reported in the experiments
by Webster and Humphrey (1997) and Httl and Friedrich (2000)
in the transitional (quasi-periodic) state. Here however we observe
both developed turbulence at the outer wall together with the turbulent bulge. Further studies are necessary with regards to what
extent near-wall turbulence is modied by the presence of the
core-turbulence.
The present data base also includes full Reynolds-stress budgets
for the considered cases, not available previously. These statistics
can be helpful for tuning and improving turbulence models. In particular models based on the eddy-viscosity assumption are known
to have problems in the presence of secondary motions and
stream-line curvature. It turns out that for straight and mildly
26
A. Noorani et al. / International Journal of Heat and Fluid Flow 41 (2013) 1626