Vous êtes sur la page 1sur 7

International Journal of Heat and Fluid Flow 41 (2013) 27–33

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Fluid Flow


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

Numerical investigation of flow through a triangular duct:


The coexistence of laminar and turbulent flow
Gertraud Daschiel a,⇑, Bettina Frohnapfel a, Jovan Jovanović b,c
a
Karsruhe Institute of Technology, Institute of Fluid Mechanics, Kaiserstr. 10, 76131 Karlsruhe, Germany
b
Technical University of Darmstadt, Center of Smart Interfaces, Petersenstr. 32, 64287 Darmstadt, Germany
c
University of Erlangen-Nürnberg, Institute of Fluid Mechanics, Cauerstr. 4, 91058 Erlangen, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Experimental studies of turbulent flow through a triangular duct with small apex angle of 11.5° per-
Received 10 October 2012 formed by Eckert and Irvine (1956) show flow laminarisation in the corner region of the duct. This effect
Received in revised form 13 March 2013 is re-investigated using direct numerical simulation (DNS). In order to analyze the impact of duct corners
Accepted 26 March 2013
on the flow behavior, results for the friction factor and the mean velocity profiles arising from different
Available online 28 April 2013
non-circular duct geometries, namely a square duct, an equilateral triangular duct and isosceles triangu-
lar ducts with an apex angle of 11.5° and 4° are compared with the results for circular pipe flow. Within
Keywords:
the cross-sections of the ducts with the small apex angles, regions of essentially laminar and turbulent
Drag reduction
Laminarization
properties are found to exist simultaneously. From analysis of the turbulent quantities performed exem-
DNS plary for the 11.5°-triangle we gain further insights into the mechanisms responsible for flow laminari-
Non-circular duct flow zation which are supported by analytical considerations for statistically axisymmetric turbulence.
Ó 2013 Elsevier Inc. All rights reserved.

1. Introduction insights into the mechanisms responsible for flow laminarisation


and skin-friction drag reduction can be gained from careful analy-
Flow through isosceles triangular ducts with small apex angles sis of this specific flow.
were first studied by Eckert and Irvine (1956). Their experimental For detailed investigation of the flow field and its statistical
investigations were mainly motivated by a critical observation in properties, a direct numerical simulation (DNS) of Eckert’s duct
practical applications: the walls of certain types of heat exchangers geometry is carried out. The simulations are run at Reh = 4500 as
were found to partly become too hot. The regions of reduced heat results from Eckert and Irvine (1956) show a significant laminar
transfer were detected in corner regions of non-circular passages. as well as turbulent region under this condition. In addition, the
Thus, extensive studies were carried out in order to examine the flow in a square, an equilateral triangular and a triangular duct
influence of corner regions on the resulting heat transfer. consisting of an even smaller apex angle than Eckert’s triangle,
During these experimental studies of flow through a triangular namely 4° is simulated at a similar hydraulic Reynolds number.
duct with an apex angle of 11.5°, the authors hit on a remarkable The results for the friction factor and the mean velocity profiles
fact: using flow visualization techniques, they observed a coexis- arising from the different duct shapes are compared in order to
tence of laminar and turbulent flow over a certain range of hydrau- gain a general impression of the impact of corner regions on turbu-
lic Reynolds numbers where the flow is expected to be fully lent flow. Experimental results for the friction factor for non-circu-
turbulent. Actually, it was found that the laminar flow behavior lar ducts which are available in literature, e.g. Hartnett et al.
close to the small apex angle is responsible for the locally reduced (1962), Nikuradse (1930), Eckert and Irvine (1960), and Carlson
heat transfer. In following investigations Eckert and Irvine (1960) and Irvine (1961), complement the presented analysis. Turbulent
measured the pressure drop in a fully developed triangular duct flow through a square duct was also investigated intensively in
flow up to Reh = 30,000, with Reh being the Reynolds number based DNS studies in the past, e.g. by Gavrilakis (1992) and Pinelli et al.
on the hydraulic diameter Dh. They found that the friction factor (2010). The corresponding data for the flow field serve for valida-
was reduced considerably compared to the well established corre- tion of the present code for calculation of non-circular duct flows.
lation of Blasius (Schlichting, 1979) for the entire turbulent regime. The flow field in triangular shaped ducts at the present Reynolds
In the context of today’s interest in flow control, these observa- number was rarely studied in experiments or DNS. Rokni and
tions provide a promising starting point: it is expected that further Gatski (2001) report predictions of flow through an equilateral
triangular duct using an explicit algebraic stress model. The impact
⇑ Corresponding author. of corners with relatively small angles on turbulent duct flow was
E-mail address: gertraud.daschiel@kit.edu (G. Daschiel). investigated in DNS studies by Raiesi et al. (2011) and Fukushima

0142-727X/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatfluidflow.2013.03.016
28 G. Daschiel et al. / International Journal of Heat and Fluid Flow 41 (2013) 27–33

and Kasagi (2002). For flow through skewed ducts with the apex viscous sublayer in the turbulent regime of the duct flow is re-
angle of 30° first indications for flow laminarisation are observed solved at least with three grid points.
in both of the above mentioned studies. However, for these ducts The cells are expanded towards the  core  region
 þof
 the ducts
coexisting laminar and turbulent regions were not detected. Due where the spacing is restricted to Dxþ 2 core  Dx3 core 6 5  5
to the differences in the duct shapes the results for the triangular based on us calculated from the Balsius correlation.
ducts presented in the following can be only compared qualita- The grid resolution expressed in terms of the ratio of the grid
tively but not quantitatively to these data. spacing to Kolmogorov length is another important issue for
assessment the grid resolution for turbulence simulations. The va-
lue of the Kolmogorov length scale estimated according to Lam-
2. Procedure mers et al. (2012), gþ  =U Þ1=4 , yields gþ
u
K ¼ ð0:25Re s sþ  b  þ K  1:5
resulting
 þin the grid
 þ resolution D x2 wall  Dx3 wall 6 1:0gþK 
The spacial discretisation of the square duct is performed using þ þ þ þ þ
2:8gK ; Dx2 core  Dx3 core 6 3gK  3gK and x1 ¼ 5:7gK based on
an orthogonal, structured mesh. In contrast, the cross-sectional  s resulting from DNS for the geometry shown in Fig. 1c.
u
plane of the equilateral triangle as well as of the triangular ducts The grids are extruded in streamwise (x1) direction. The stream-
with very small apex angles of 11.5° and 4° is spatially discretized wise mesh spacing is set to Dxþ 1 ¼ 9:5 according to findings of
with an unstructured grid consisting of prism layers resolving the Gavrilakis (1992) for DNS of square duct flow. The total streamwise
near wall regions and a hexahedral core mesh. The cross-section extensions are set to Lx1 ¼ 5Dh for all cases, which also corresponds
planes of the geometries with details of the mesh at critical posi- to the set-up in Eggels et al. (1994). Chin et al. (2010) investigated
tions
 þ  are shown  in Fig. 1. The mesh is initially set up to ensure the influence of the pipe length on turbulence statistics calculated
Dx2 wall  Dxþ3 wall 6 1:6  4:6 where the viscous length is based from DNS for circular pipe flow. According to their results the pres-
on the fluid viscosity and a friction velocity us estimated a priori ent duct length is sufficient for studying the mean flow and sec-
based on the Blasius correlation (Schlichting, 1979). The simulation ond-order turbulent statistics.
results
 show  a smaller
 skin friction than the Blasius law, such that A second-order finite volume method is applied for the spacial
Dxþ2 wall  Dxþ3 wall 6 1:4  4:2 is fulfilled for the triangular duct approximation of the continuity and Navier–Stokes equations on
shown in Fig. 1c if the viscous length scale is calculated on the ba- a collocated-mesh arrangement. The equations are coupled for
sis of the average wall shear velocity, u  s . Since the values of us vary the calculation of the pressure using the PISO-algorithm. Variables
along the side walls of non-circular ducts, the normalization based on the cell faces are approximated by second order linear interpo-
on the local wall shear velocity gives the most realistic impression lation. In general, the use of a collocated mesh arrangement does
of the grid quality.
 Based
 on this definition, the grid resolution en- not ensure the conservation of the kinetic energy (Ferziger and
sures Dxþ þ
2 wall  Dx3 wall 6 1:83  5:27 where the upper bounds Perić, 2008). However, it is commonly used for turbulence simula-
of the resolution are reached at about x3/h = 0.8. Results at this crit- tions in complex geometries due to its simpler form in curvilinear
ical position are discussed in Section 4. Overall, it is found that the

(a) (b)

(c) (d)
Fig. 1. Cross section of the duct geometries with mesh details. (a) Square duct; (b) equilateral duct; (c) triangular duct (a = 11.5°); and (d) triangular duct (a = 4°).
G. Daschiel et al. / International Journal of Heat and Fluid Flow 41 (2013) 27–33 29

coordinates. A similar numerical method is introduced by Felten


and Lund (2006) and tested in terms of conservation errors in a
large eddy simulation of a plane channel flow. They observe good
agreement with reference data provided that the simulation is
run at a sufficiently high mesh resolution. As the resolution within
the present investigation is similar to the one used for DNS studies
of non-circular duct flows in literature (Gavrilakis, 1992) we as-
sume numerical dissipation effects to be of minor importance.
The simulations are carried out under a constant flow rate con-
dition. Time discretisation is achieved using the second order im-
plicit Euler-backward scheme and the time step is chosen to
ensure a Courant number of Comax < 0.2.
The initial velocity fields are set using parabolic velocity profiles
with superposition of random three dimensional fluctuation of 10%
of the bulk velocity Ub. The simulations were run for approximately
five turnover times t⁄ with t⁄ = Dh/us without averaging. After an
initial drop of the pressure gradient a fully turbulent flow field is
observed to develop. After this development period simulations
were extended in order to perform time averaging for at least 20 Fig. 2. Friction factor f plotted against the hydraulic Reynolds number Reh. Analytic
turnover times. Due to the restricted possibility to use symmetries solution for circular laminar pipe flow: f = 64/Reh, Blasius correlation for turbulent
flow: fh ¼ 0:316=Re1=4 (Schlichting, 1979). (a) Square duct: analytic solution for
within the duct cross-sections, long averaging times were found to h
laminar flow: f = 57/Reh, open symbols: measurements from Hartnett et al. (1962),
be necessary to obtain reliable flow statistics. For the triangular solid symbol: DNS; (b) equilateral triangular duct: analytic solution for laminar
duct with apex angle of 11.5° the averaging time was extended flow: f = 53/Reh, open symbols: measurements from Nikuradse (1930), solid
to 34 turnover times or 9500u  2s =m in order to obtain second order symbol: DNS; (c) triangular duct (a = 11.5°): laminar solution for a circular sector:
fh = 50.3/Reh (Eckert and Irvine, 1956), open symbols: measurements from Eckert
statistics for this geometry. A similar averaging time was used by
and Irvine (1960), solid symbol: DNS; and (d) triangular duct (a = 4°): laminar
Fukushima and Kasagi (2002) for DNS of ducts with rhombic cross analytic solution for a circular sector: fh = 48.85/Reh (Carlson and Irvine, 1961), open
sections. In the following sections, h  i marks quantities which are symbols: measurements from Carlson and Irvine (1961), solid symbol: DNS.
averaged along the homogeneous streamwise direction, x1, and in
time.
gradient is significantly decreased close to the small apex angles
while on the opposite side the velocity gradient is increased com-
3. Comparison of flow properties in different duct geometries
pared to the circular pipe. However, this loss does not annihilate
the benefits resulting from the sharp corner of the triangles. Thus
3.1. Friction factor f
it can be concluded that by increasing the duct region in which
the flow tends to the laminar state, the friction factor is decreased.
The results for the friction factor f from present simulations (so-
lid symbols) are plotted in Fig. 2, together with corresponding
experimental data (open symbols). A good agreement for all duct 4. Coexistence of laminar and turbulent flow in a 11.5° isosceles
geometries investigated can be observed. triangular duct
In the laminar regime (Reh < 2300) analytical solutions for the
friction factor are available. For this flow regime the values emerg- Within this section we further analyze the results correspond-
ing from the non-circular duct geometries are lower compared to ing to the 11.5°-triangular duct. For the Reynolds number of the
the solution for the circular pipe. In contrast to the behavior in present investigation, Eckert and Irvine (1956) found laminar flow
the laminar regime, in turbulent flows the correlation of Blasius to persist in the narrow corner region whereas in the remaining
typically provides a good estimate for the friction factor in ducts domain the flow was fully turbulent. From detailed analysis of this
of different cross-sectional shapes (Schlichting, 1979). The present specific flow situation we may expect to gain fundamental insights
results, however, show a surprising tendency: while for the square into the process of reverse transition.
duct the friction factor is only slightly lower, this effect becomes
more pronounced for the triangular duct shapes. The tendency to 4.1. Mean flow field hUii in the duct
lie below the empirical correlation of Blasius becomes stronger if
the angle of the duct is reduced (Fig. 2a–d). The triangular shape In order to provide a general impression of the flow field, con-
with a = 11.5° (case (c)), for example, shows a constant reduction tour plots of the mean velocity hU1i, the kinetic energy of turbu-
of f for about 20% over the entire turbulent regime investigated lence k = 0.5 hu u i and the magnitude of the secondary flow
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i i
(Eckert and Irvine, 1956). hU 2 i2 þ hU 3 i2 are shown in Figs. 4–6. As in experiments of Eckert
and Irvine (1956) a stable laminar region was detected close to the
3.2. Mean velocity profile in laminar and turbulent flow acute angle up to x3/h  0.25. Moving towards the core region of
the duct, the velocity profile starts to deviate from the laminar par-
The profiles of the streamwise component of the mean flow ex- abolic shape and finally transforms into the fully turbulent profile.
tracted from DNS are plotted in Fig. 3 along the x3-axis of the ducts. This behavior is supported by the contour plot of the turbulent ki-
For case (c), measured data from Eckert and Irvine (1956) are in netic energy in Fig. 5.
good agreement with results from DNS. A characteristic phenomenon of turbulent flow through non-
A comparison of the turbulent velocity profiles with the laminar circular ducts is the appearance of secondary flows of Prandtl’s sec-
velocity profiles for the same geometry and the fully turbulent ond kind (Pöschl, 1926). As expected, a secondary flow pattern ap-
velocity profile for a circular pipe (Fukagata and Kasagi, 2002) at pears in the turbulent part of the triangular duct in form of
a similar Reynolds number show the laminarizing effect in the cor- counter-rotating vortices directed towards the corners as illus-
ner regions of the ducts. Especially for cases (c) and (d) the velocity trated in Fig. 6. It is observed that the magnitude of this secondary
30 G. Daschiel et al. / International Journal of Heat and Fluid Flow 41 (2013) 27–33

square, Reh=1000
3.0 DNS square, Reh=4410
DNS pipe, Reh=5310
2.5

2.0
<U1>/Ub

1.5

1.0 Fig. 4. Contour plot of the mean streamwise velocity field hU1i normalized
with Ub.
0.5
(a)
0.0
0.0 0.2 0.4 0.6 0.8 1.0
x3/h

triangle, Reh=1000
3.0 DNS triangle, Reh=4500
DNS pipe, Reh=5310
2.5

Fig. 5. Contour plot of the kinetic energy k = 0.5 huiuii normalized with the mean
2.0
<U1>/Ub

 2s resulting from DNS.


wall shear velocity u

1.5

1.0

0.5
(b)
0.0
0.0 0.2 0.4 0.6 0.8 1.0
x3/h

triangle, Reh=1000
3.0 DNS triangle, Reh=4500
Eckert, Reh=4590
2.5 DNS pipe, Reh=5310

2.0
<U1>/Ub

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Fig. 6. Contour plot of the magnitude of the secondary flow field hU 2 i2 þ hU 3 i2
1.5 normalized with Ub together with the secondary flow developing at corners. Note
that the vectors are scaled by the factor two close to the acute angle in order to
1.0 highlight the direction of the secondary flow in this part of the duct.

0.5
motion, namely up to 2.5% of the bulk velocity, is slightly larger
(c) than that in square duct flow where Gavrilakis (1992) found
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.9%. By contrast, the secondary motion directed towards the acute
x3/h angle is very weak and vanishes when approaching the laminar
part of the flow. In previous simulations of turbulent flow in ducts
triangle, Reh=1000 with an apex angle of 30° this strong damping of the secondary
3.0 DNS triangle, Reh=4500 motion was not observed (Raiesi et al., 2011; Fukushima and
DNS pipe, Reh=5310
Kasagi, 2002), which is presumably linked to the fact that lamina-
2.5
rization was not observed.
2.0 More details about the modifications of the mean velocity field
<U1>/Ub

can be gained in the comparison of streamwise profiles plotted


1.5 along the x2-direction at several spanwise positions as shown
Fig. 7. The mean streamwise velocity is normalized with the corre-
1.0 sponding local wall shear velocity us (instead of us or u
 s ). In the fol-
lowing the subscript ‘+’ will refer to normalization with us . For
0.5
square duct flows such scaling was shown to be appropriate for
(d) the investigation of the near wall region (Gavrilakis, 1992). The
0.0
0.0 0.2 0.4 0.6 0.8 1.0 characteristic profiles corresponding to the viscous sublayer given
x3/h by hU 1 iþ ¼ xþ2, and the logarithmic region described by
hU 1 iþ ¼ ð1=jÞxþ
2 þ B with j = 0.41 and B = 5.2 (Pope, 2011) are also
Fig. 3. Comparison of the laminar and turbulent velocity profiles along the x3-axis of the included in the plots. Fig. 7a shows a velocity profile which is rep-
ducts with the velocity profile of turbulent circular pipe flow at Reh = 5310 (Fukagata
resentative for the fully turbulent part of the duct. In Section 2 it
and Kasagi, 2002). (a) Square duct: results for Reh = 1000 and Reh = 4410; (b) equilateral
triangular duct: results for Reh = 1000 and Reh = 4500; (c) triangular duct (a = 11.5°): was discussed, that the wall shear stress in streamwise direction
results for Reh = 1000 and Reh = 4500, experimental data from Eckert and Irvine (1956) at reaches its maximum value at x3/h = 0.8 and consequently the
Reh = 4590; and (d) triangular duct (a = 4°): results for Reh = 1000 and Reh = 4500. mesh spacing in viscous scales is coarsest at this position. The re-
G. Daschiel et al. / International Journal of Heat and Fluid Flow 41 (2013) 27–33 31

20 x3-direction some affinity to the profiles in plane channel flow


DNS trinagle
<U1>+=x2+ are expected. These profiles for similar friction Reynolds numbers
log law Res ¼ us d=m are also included, where 2d is estimated by the width
15
of the duct at a certain x3-position. The exact values for Res are: (a)
Res = 160, (b) Res = 85 and (c) Res = 40 while channel data for
<U1>+

10 Res = 150 (a) and Res = 100 (b) and (c) are used for comparison.
The Reynolds stresses in Fig. 8a representing the core region of
5 the duct in general are in good agreement with the corresponding
channel flow data. The slight underestimation of the channel data
(a) x3/h=0.8 in the near wall region is also observed by Gavrilakis (1992) for
0
0.1 1 10 100 square duct flow and is related to the scaling with the local friction
x2+ velocity us . At a position closer to the acute angle shown in Fig. 8b
the magnitude of all stress components is considerably reduced
20 compared to the turbulent channel data for the lowest Reynolds

15 8
(a) x3/h=0.8 <u1u1>
7 <u2u2>
<U1>+

<u3u3>
10 6 <u1u2>
<u1u3>
5
Kuroda, 1990
5

+
4

<uiuj>
(b) x3/h=0.5 3
0
0.1 1 10 2
+
x2 1
0
20
-1
0 20 40 60 80 100 120 140
15 x2+
<U1>+

8
10
7
(b) x3/h=0.5
5 6
5
(c) x3/h=0.3
+

0 4
<uiuj>

0.1 1 10
3
x2+
2
Fig. 7. Mean streamwise velocity hU1i normalized with the local wall shear velocity 1
us plotted along the x2-direction at different spanwise positions: x3/h = 0.8 (a),
x3/h = 0.5 (b) and x3/h = 0.3 (c) together with the characteristic profile for the 0
viscous sublayer hU 1 iþ ¼ xþ þ þ
2 and the log-region hU 1 i ¼ ð1=jÞx2 þ B with j = 0.41
-1
and B = 5.2 (Pope, 2011). 0 10 20 30 40 50 60 70 80
x2+
sults show, that the expected linear velocity profile in the near wall 8
region is reproduced satisfactorily with the present grid resolution. (c) x3/h=0.3
In the center of the duct hU1i+ approaches a logarithmic distribu- 7
tion which is a common property of turbulent wall bounded flows. 6
The present data are laying slightly above the relation given by 5
Pope (2011) for plane channel flow. This observation is made in
+

4
<uiuj>

general for turbulent duct flows (Eggels et al., 1994; Gavrilakis,


1992). 3
When moving to positions closer to the corner shown in Fig. 7b 2
and c typical deviation from the turbulent logarithmic distribution
1
can be observed which illustrate the development towards a para-
bolic flow profile indicating the laminarization of the flow. 0
-1
0 5 10 15 20 25 30 35 40
4.2. Statistical flow properties
x2+
The suppression of turbulent flow properties towards the acute
Fig. 8. Reynolds stresses huiuji normalized with the local wall shear velocity us
angle that accompanies the observed flow laminarization is inves-
plotted along x2-direction at different spanwise positions: x3/h = 0.8 (a), x3/h = 0.5
tigated in the following. In Fig. 8 the dominant components of the (b) and x3/h = 0.3 (c) together with the profiles for plane channel flow at similar
Reynolds stress tensor are plotted along the x2-direction at several values of Res. (a) Res = 150 (Kuroda and Kasagi, 1989; Kuroda, 1990), (b) and (c)
spanwise positions. Due to the very elongated duct shape in Res = 100 (Kuroda and Kasagi, 1989; Kuroda et al., 1989; Kuroda, 1990)
32 G. Daschiel et al. / International Journal of Heat and Fluid Flow 41 (2013) 27–33

number available. This tendency increases when moving even fur- 0.20
u1,rms
ther in negative x3-direction (Fig. 8c) where all components of the u2,rms
Reynolds stress tensor except of hu1u1i are almost supressed sug- 0.15 u3,rms

ui,rms/Ub
gesting a highly anisotropic state of turbulence.
The observed trend in the dynamics of turbulence is further 0.10
analyzed in the invariant space formed by two scalar invariants
(II = aijaji; III = aijajkaki) of the anisotropy tensor ðaij ¼ ui uj =q2  0.05
1=3dij Þ. From this representation of turbulent quantities it can be
concluded in general that the turbulent dissipation rate at the wall 0.00
0.0 0.2 0.4 0.6 0.8 1.0
can be entirely supressed if the velocity fluctuations in the near- x3/h
wall region satisfy local axisymmetry with invariance to rotation
about the axis aligned with the mean flow direction (here x1) Fig. 9. Development of the normalized rms-values of the velocity fluctuations ui,rms
(Jovanović and Hillerbrand, 2005). For such turbulence the stream- along the x3-axis.
wise intensity is much larger than the intensities in the other two
directions which are identical. Due to this property, the wall loca-
tion corresponds to the one-component state of turbulence. In this
special state, turbulent kinetic energy cannot be amplified near the
wall and therefore turbulence must decay, leading to flow lamina-
rization (Jovanović et al., 2006). These conclusions are in close
agreement with results of direct numerical simulations which dis-
play high drag reduction when turbulence in the viscous sublayer
is manipulated to tend towards the one-component state in an
axisymmetric fashion (Frohnapfel et al., 2007). In the following
we show that these fundamental deductions about flow laminari-
zation from an initially turbulent state hold also in duct flows of
particular cross-section configurations where coexistence of lami-
nar and turbulent flow regimes are observed. The starting point is
the equation for the mean flow:

@U i @U i @ui uk 1 @P @2 Ui
þ Uk þ ¼ þm : ð1Þ
@t @xk @xk q @xi @xk @xk
Fig. 10. Trajectory through the anisotropy invariant map for turbulent pipe flow
The third term on the left side corresponds to the turbulent trans- (Fukagata and Kasagi, 2002) for Reh = 5310 and along the centerline of the triangular
duct: 0.25 6 x3/h 6 0.8. For x3/h < 0.25 the mean velocity profile was found to have a
port. Physically, it represents the interaction of turbulent fluctua-
parabolic shape and the flow thus is expected to be laminar.
tions with the mean flow field.
For a statistical axisymmetric state of turbulence ui uk can be ex-
pressed as
This kind of trajectory in the anisotropy-invariant map was pre-
ui uk ¼ Adik þ Bki kk ; ki ð1; 0; 0Þ; ð2Þ viously found to be characteristic for the laminar to turbulent tran-
sition process of natural disturbances in external flows (Jovicić
with A and B being scalar functions so that x1 is the axis of invari-
et al., 2006) and pipe flows (Nishi, 2009). The tendency of increased
ance (Jovanović, 2004). Eq. (1) can be rewritten for such turbulence
near-wall anisotropy is also characteristic for a group of turbulent
as follows:
drag reduction mechanisms like for example the one achieved with
 
@U i @U i 1 @ 1 2 @2 Ui polymer additives (Frohnapfel et al., 2007).
þ Uk ¼ P þ qq2 þ qB þm ; ð3Þ Present and future studies will focus on different duct geome-
@t @xk q @xi 3 3 @xk @xk
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} tries where the coexistence of laminar and turbulent flow is found.
P
Based on further analysis of these flows we aim at extracting
2 ⁄
where q is the trace of ui uk . P corresponds to a generalized pres- knowledge for the design of smart duct geometries for flow control
sure such that this equation leads to solutions which coincide with purposes.
solutions for fully developed laminar flow. This suggests that a ten-
dency towards axisymmetry in ui uk leads to flow laminarization
and therefore to a large viscous drag reduction effect (for the class
of flows considered in this study). Similar deductions are also re- 5. Conclusion and outlook
ported by Lammers et al. (2012).
If the statistical properties of the flow are analyzed for the trian- Flow laminarization is observed in corner regions for a selection
gular duct along the x3-axis, moving away from the small apex an- of non-circular ducts. This flow behavior leads to friction factors
gle, i.e. from the laminar to the turbulent part of the flow, one finds below the Blasius solution typically used in engineering predic-
that velocity fluctuations in the x1-direction grow faster than in tions for the pressure drop in non-circular ducts. The reduction
x2- and x3-direction (see Fig. 9). In Fig. 10 this development is plot- of the friction factor is considerably larger for triangular ducts with
ted in the anisotropy-invariant map (Lumley and Newman, 1977) a small apex angle in which flow regions with essential laminar
in comparison to the trajectory of a turbulent circular pipe flow. and turbulent properties are found to coexist. Further investiga-
It can be seen that the trajectory across the map develops closer tions of this particular flow situation in an isosceles triangular duct
to its right boundary for the triangular shaped duct, which repre- with apex angle of 11.5° show that turbulent quantities in the core
sents a statistically axisymmetric state of the velocity fluctuations. region of the duct behave similar as in turbulent channel flow.
Moreover, the anisotropy in the corner region of the duct is When moving towards the acute angle of the duct, the turbulent
increased. These observations are in agreement with the conclu- properties of the flow are found to be progressively suppressed
sions drawn from the above given analysis of Eq. (3). resulting in a parabolic laminar velocity distribution.
G. Daschiel et al. / International Journal of Heat and Fluid Flow 41 (2013) 27–33 33

It is suggested by analytical considerations that statistically axi- Fukushima, N., Kasagi, N., 2002. Turbulent momentum and heat transfer in ducts of
rhombic cross section. In: Proceedings of Heat transfer 2, Grenoble, France, pp.
symmetric turbulence represents a flow state with reduced inter-
207–212.
action of the turbulent quantities with the mean flow field Gavrilakis, S., 1992. Numerical simulation of low-Reynolds-number turbulent flow
leading to flow laminarization. Analysis of the flow field in the tri- through a straight square duct. J. Fluid Mech. 244, 101–129.
angular duct in the framework of invariant theory support this Hartnett, J., Koh, J., McComas, S., 1962. A comparison of predicted and measured
friction factors for turbulent flow through rectangular ducts. J. Heat Transfer 84,
conclusion. The coexisting laminar and turbulent flow has charac- 82–88.
teristic properties that are also observed in the laminar to turbu- Jovanović, J., 2004. The Statistical Dynamics of Turbulence. Springer-Verlag, Berlin.
lent transition process in external and circular pipe flows and in Jovanović, J., Frohnapfel, B., Škaljić, E., Jovanović, M., 2006. Persistence of the
laminar regime in a flat plate boundary layer at very high Reynolds number.
drag reduced turbulent flows. This observation forms the starting Therm. Sci. 10, 63–96.
point for our ongoing work which aims at the development of no- Jovanović, J., Hillerbrand, R., 2005. On peculiar property of the velocity fluctuations
vel duct geometries for flow control purposes. in wall-bounded flows. Therm. Sci. 9, 3–12.
Jovicić, N., Breuer, M., Jovanović, J., 2006. Anisotropy-invariant mapping of
turbulence in a flow past an unswept airfoil at high angle of attack. J. Fluid
Acknowledgments Eng. 128, 559–567.
Kuroda, A., 1990. Direct Numerical Simulation of Couette-Poiseuille Flows. Dr. Eng.
Thesis, The University of Tokyo, Tokyo.
The authors acknowledge support of Project FR2823/2-1 of the Kuroda, A., Kasagi, N., 1989. Fully Developed 2-d Channel Flow, Thtlab DNS
German Research Foundation (DFG) and the ‘Center of Smart Inter- Database.
faces’ at TU Darmstadt. Kuroda, A., Kasagi, N., Hirata, M., 1989. A direct numerical simulation of the fully
developed turbulent channel flow. In: International Symposium on
Computational Fluid Dynamics, Nagoya, pp. 1174–1179.
References Lammers, P., Jovanović, J., Frohnapfel, B., Delgado, A., 2012. Erlangen pipe flow: the
concept and DNS results for microflow control of near-wall turbulence.
Carlson, L., Irvine, T., 1961. Fully developed pressure drop in triangular shaped Microfluidics Nanofluidics. http://dx.doi.org/10.1007/s10404-012-0972-0.
ducts. J. Heat Transfer 83, 441–444. Lumley, J., Newman, G., 1977. The return of isotropy of homogeneous turbulence. J.
Chin, C., Ooi, A., Marusic, I., Blackburn, H., 2010. The influence of pipe length on Fluid Mech. 82, 161–178.
turbulence statistics computed from direct numerical simulation data. Phys. Nikuradse, J., 1930. Turbulente Strömungen in nicht-kreisförmigen Rohren. Ing.-
Fluids 22, 115107. Arch. 1, 306–332.
Eckert, E., Irvine, T., 1956. Flow in corners of passages with noncircular cross Nishi, M., 2009. Laminar to Turbulent Transition in Pipe Flow Through Puffs and
sections. Trans. ASME 78, 709–718. Slugs. Suedwestdeutscher Verlag, Germany.
Eckert, E., Irvine, T., 1960. Pressure drop and heat transfer in a duct with triangular Pinelli, A., Uhlmann, M., Sekimoto, A., Kawahara, G., 2010. Reynolds number
cross section. J. Heat Transfer, 709–718. dependence of mean flow structure in square duct turbulence. J. Fluid Mech.
Eggels, J., Unger, F., Weiss, M., Westerweel, J., Adrian, R., Friedrich, R., Nieuwstadt, F., 644, 107–122.
1994. Fully developed turbulent pipe flow: a comparison between direct Pope, S., 2011. Turbulent Flows. Cambridge University Press, Cambridge.
numerical simulation and experiment. J. Fluid Mech. 268, 175–209. Pöschl, T., 1926. Zweiter Internationaler Kongreß für Technische Mechanik in
Felten, F., Lund, T., 2006. Kinetic energy conservation issues associated with the Zürich. (12. bis 17. September 1926). Naturwissenschaften 14, 1029–1032.
collocated mesh scheme for incompressible flow. J. Comput. Phys. 215, 465–484. Raiesi, H., Pollard, A., Piomelli, U., 2011. Direct numerical simulations of turbulence
Ferziger, J., Perić, M., 2008. Numerische Strömungsmechanik. Springer-Verlag, induced secondary motion in square and skewed ducts. In: Proceedings of TSFP-
Berlin. 7, Ottawa, Canada.
Frohnapfel, B., Lammers, P., Jovanović, J., Durst, F., 2007. Interpretation of the Rokni, M., Gatski, T., 2001. Predicting turbulent convective heat transfer in fully
mechanism associated with turbulent drag reduction in terms of anisotropy developed duct flows. Int. J. Heat Fluid Flow 22, 381–392.
invariants. J. Fluid Mech. 577, 457–466. Schlichting, H., 1979. Boundary-Layer Theory, seventh ed. McGraw-Hill Book
Fukagata, K., Kasagi, N., 2002. Highly energy-conservative finite difference method Company, NewYork.
for the cylindrical coordinate system. J. Comput. Phys. 181, 478–498.

Vous aimerez peut-être aussi