Journal of Hydraulic Research Vol. **, No. * (****), pp. 493 501
**** ************* *********** ** ********* Engineering and Research
Modi ed log-wake law for turbulent ow in smooth pipes
Loi log-train e modi e pour coulement turbulent en conduite paroi lisse
JUNKE GUO, Assistant Professor, Department of Civil Engineering, National University of Singapore, 10 Kent Ridge Crescent,
Singapore 119260. Email: abqp36@r.postjobfree.com
PIERRE Y. JULIEN, Professor, Engineering Research Center, Department of Civil Engineering, Colorado State University,
Fort Collins, CO 80523, USA. Email: abqp36@r.postjobfree.com
ABSTRACT
A modi ed log-wake law for turbulent ow in smooth pipes is developed and tested with laboratory data. The law consists of three terms: a log term,
a sine-square term and a cubic term. The log term re ects the restriction of the wall, the sine-square term expresses the contribution of the pressure
gradient, and the cubic term makes the standard log-wake law satisfy the axial symmetrical condition. The last two terms de ne the modi ed wake
law. The proposed velocity pro le model not only improves the standard log-wake law near the pipe axis but also provides a better eddy viscosity
model for turbulent mixing studies. An explicit friction factor is also presented for practical applications. The velocity pro le model and the friction
factor equation agree very well with Nikuradse and other recent data. The eddy viscosity model is consistent with Laufer s, Nunner s and Reichardt s
experimental data. Finally, an equivalent polynomial version of the modi ed log-wake law is presented.
R SUM
Une loi log-tra n e modi e est d velopp e pour d crire les coulements turbulents dans les conduites parois lisses. La loi modi e consiste en trois
termes: un terme logarithmique, un terme de tra n e, et un terme cubique de condition limite axiale. Le terme logarithmique repr sente les conditions
de la paroi, le terme de tra n e d nit le gradient de pression de l coulement, et le dernier terme satisfait la condition limite au centre de la conduite.
Le terme de tra n e d crit physiquement les conditions de m lange turbulent caus es par le gradient de pression. Le coef cient de frottement et le
coef cient de viscosit tourbillonnaire sont d crits de fa on analytique. Les quations de pro ls de vitesse et de coef cient de frottement sont en
accord avec les donn es exp rimentales de Nikuradse et autres donn es r centes. Le mod le de viscosit tourbillonnaire est consistent avec les donn es
exp rimentales de Laufer, Nunner et Reichardt. Finalement, une version polynomiale de la loi log-tra n e modi e est galement pr sent e.
Keywords: Pipe ow; turbulence; logarithmic matching; log law; log-wake law; velocity pro le; velocity distribution; eddy viscosity;
friction factor.
1 Introduction This study emphasizes the velocity pro le away from the pipe
wall where yu / > 30 given = uid kinematic viscosity. The
Fully developed turbulent ow in circular pipes has been inves- law of the wall or the logarithmic law proposed by von Karman
tigated extensively not only because of its practical importance, and Prandtl (Schlichting, 1979, p. 603) is widely used in this
but also for the extension of the results to open-channel ows and region. For smooth pipe ow, it is written as
boundary layer ows. The rst systematic study of the velocity
u 1 yu
= ln +A (2a)
pro le in turbulent pipe ows may be credited to Darcy in 1855
u
(Schlichting, 1979, p. 608) who deduced a 3/2-nd-power velocity
where = the von Karman constant. Nikuradse (1932) has
defect law from his careful laboratory measurements, i.e.
veri ed (2a) with his classical experiments. When (2a) was
umax u y 3/2
= 5.08 1 (1) rst proposed, the value of was suggested to be 0.4. Most
u R
researchers subsequently preferred the value of 0.41 (Nezu and
in which umax = the maximum velocity at the pipe axis, u =
Nakagawa, 1993, p. 51). Recently, Barenblatt (1996, p. 278)
the time-averaged velocity at a distance y from the pipe wall,
showed that for very large Reynolds number, the value of von
u = the shear velocity, and R = the pipe radius. Equation (1)
Karman constant tends to
is seldom used because it is invalid near the pipe wall with =
2
y/R
3e
(White, 1991, p. 415) or O Connor s law (1995) may be applied.
Revision received February 17, 2003. Open for discussion till February 29, 2004.
493
494 Junke Guo and Pierre Y. Julien
deviation from the log law, de ne a law of the wake. It is noted
that although (5a) or (5b) satis es the axial symmetrical condi-
tion, the constant derivative of the linear correction term brings
an additional shear stress in the near wall region, which slightly
perturbs the law of the wall. It is therefore concluded that the
linear function is not the best axial boundary correction.
The purpose of this paper is to develop a physically-based
velocity pro le model for turbulent pipe ows called the modi ed
log-wake law. It is proposed to improve upon Eq. (5b) in the light
of a better physical interpretation of the wake law. The proposed
modi ed log-wake law is also tested with experimental velocity
Figure 1 The law of the wall or the logarithmic law.
measurements in pipes. Analytical relationships for the turbulent
eddy viscosity and for the friction factor are then compared with
The value of A varies in literature; it is about 5.29 0.47 accord-
experimental data.
ing to Nezu and Nakagawa (1993, p. 51). Equation (2b) will be
examined later in this paper.
Equation (2a) is applicable away from the wall where yu / > 2 Development of the modi ed log-wake law
30, as shown in Fig. 1. It is not only valid for steady ow, but
is also frequently used as a reference condition in unsteady ow This section formulates the modi ed log-wake law in smooth
simulations (Ferziger and Peric, 1997, p. 277). This is because pipes. First a theoretical analysis is considered. Section 2.2
even in unsteady ows, the wall shear stress predominates in the proposes the basic structure of the modi ed log-wake law.
near-wall ow, and the in uence of inertial forces and pressure Section 2.3 de nes the axial boundary correction function. The
gradient are vanishingly small. nal formulation of the modi ed log-wake law is written in
Laufer (1954) found that the logarithmic law (2a) actually velocity-defect form in Section 2.4.
deviates from experimental data when = y/R > 0.1 0.2.
Coles (1956) further con rmed this nding and claimed that
the deviation has a wake-like shape when viewed from the 2.1 Theoretical analysis
freestream. Thus, he called the deviation the law of the wake.
Consider fully developed turbulent ow with homogeneous den-
Based on Coles digital data, Hinze (1975, p. 98) proposed the
sity through a pipe of radius R . For convenience, cylindrical
following expression for the wake function, i.e.
coordinates are used with the x -axis coinciding with the axis of
2
W = sin2 the pipe, as shown in Fig. 2(a). One can show that the continuity
(3)
2 equation is automatically satis ed and none of the ow variables
in which is Coles wake strength. Finally, one can modify the depend on . The momentum equation in the r -direction gives
logarithmic law by adding the wake function, i.e. that the pressure p is a function of x alone, i.e.
u 1 yu 2 p
= +A + sin2
ln (4) =0 (6)
u 2 r
This is called the log-wake law. When applied to pipe ows, in which p = dynamic pressure. The only nonzero component
the reader can easily show that the log-wake law (4) does not of velocity is the axial velocity u(r) or u(y) where y = distance
satisfy the axial symmetrical condition, i.e., the velocity gradient from the wall, i.e. y = R r . With reference to Fig. 2(b),
is nonzero at the pipe axis. Besides, the physical interpretation for steady and incompressible ow, the force balance in the x -
of the wake function is not clear in pipe ows. direction gives
Similar to ows in narrow open-channels (Guo and Julien,
2 (R y)dx + ( + d ) 2 (R y dy)dx
2001), to correct the velocity gradient at the pipe axis, Guo (1998)
dp [(R y)2 (R y dy)2 ] = 0
proposed the following velocity pro le model for pipe ows,
u 1 yu 2
in which = local shear stress. Neglecting the 2nd order terms,
= ln +A+
sin2 (5a)
u 2
the above equation reduces to
The law of the wall Law of the wake
2 dxdy + d 2 (R y)dx dp 2 (R y)dy = 0
or
umax u 1 2
= (ln + 1 ) + cos2 Dividing by 2 dxdy, the above becomes
(5b)
u 2
d dp
+ (R y) (R y) = 0
in which the Coles wake strength is due to the pressure-
dy dx
gradient. The log term in (5a) expresses the restriction of the
This can be rearranged as
wall, the sine-square term indicates the effect of the pressure-
d [(R y) ]
gradient, and the last term re ects the axial boundary correction. dp
(R y) + =0 (7)
According to Coles, the last two terms in (5a), which are the dx dy
Modi ed log-wake law for turbulent ow in smooth pipes 495
Figure 2 Scheme of a developed turbulent pipe ow.
This is the momentum equation in the x -direction. Integrating the second term, one can apply (9) to (13) and eliminate the
the above with respect to y and applying the wall shear stress w pressure-gradient. Integrating (13) gives
at y = 0 gives ( 2)d
u d
= + (14)
(1 )f (1 )f u
R dp (2R y)y
R w
= + (8)
R y 2 dx (R y)R Clearly, the solution of the above equation requires the knowledge
of f that is complicated and unknown. Therefore, this paper
Although the relation tries to construct an approximate velocity pro le model, based
on a physical and mathematical reasoning.
R dp
w = (9)
2 dx
2.2 Approximation of the velocity pro le model
which can be found from ( = 1) = 0, can further reduce (8)
to a linear function, the current form in (8) has a clear physical The effect of the wall shear stress is often expressed by the law
interpretation. Its rst term expresses the effect of the wall shear of the wall (2a). This implies that the rst integral of (14) must
stress, and the second term re ects the effect of the pressure- reduce to the logarithmic law (2a) near the wall. It is then assumed
gradient. Near the wall, the effect of the pressure gradient can be that the rst integral can be approximated by
neglected, and the uid shear stress is balanced by the wall shear
d 1 yu
= + A + F1 stress. ln (15)
(1 )f Applying the eddy viscosity model,
in which F1 is a correction function of the effect of the wall on
du the core ow region. Obviously, F1 must satisfy
t = t (10)
dy
F1 ( = 0) = 0 (16a)
in which t = turbulent shear stress, = uid density, and
and
t = eddy viscosity, to (8) and neglecting the viscous shear
F1 ( = 0) = 0
stress gives (16b)
R dp (2R y)y Condition (16a) keeps F1 negligible near the wall, and condition
du R w
= +
t (11)
R y 2 dx (R y)R (16b) guarantees F1 does not bring a shear stress to the near wall
dy
region.
According to previous experience (Hinze, 1975, p. 730), one can For simplicity, one can de ne the second integral of (14) as
assume an eddy viscosity as
( 2)d
F2 = (17)
(1 )f w y
t = R f (12)
R Like F1, F2 must be negligible and not bring a shear stress near
the wall, i.e.
in which f is an unknown function. Applying the de nition of
F2 ( = 0) = 0
the shear velocity u = w / and the normalized distance (18a)
= y/R into (11) and (12) leads to
and
(2 )
1 du 1 R dp 1 F2 ( = 0) = 0 (18b)
= + (13)
(1 )f 2 dx u2 (1 )f u d
This is equivalent to removing the wall restriction in the ow
This shows that the pipe velocity pro le is a result of the effects direction. Therefore, the effect of the pressure gradient can be
of the wall shear stress (the rst term) and the pressure-gradient considered a wall-free shear, like a jet, except that the pressure
(the second term). After making clear the physical meaning of gradient is the driving force in pipe ows while the inertia is
496 Junke Guo and Pierre Y. Julien
the axis = 1. In terms of Taylor series, all odd derivatives at
the driving force in a developed jet. Furthermore, the pipe ow
= 1 must be zero, i.e.
can be considered a superposition of a wall-bounded shear and a
wall-free shear. Because of the symmetrical condition, F2 must
1 d 2u 1 d 4u
u = umax + (1 )2 + (1 )4 +
reach its maximum value at the pipe axis and satisfy
2! d 2 4! d 4
=1 =1
F2 ( = 1) = 0 (18c)
To correct the modi ed log-wake law to the third order term,
According to (18b) and (18c), one can approximate the derivative letting
of F2 as
d 3u
=0
F2 sin (19) d 3 =1
The integration of (19) with the boundary condition (18a) gives in (21) where (25) gets applied, one can show that
2
n=3
F2 = sin2 (26)
(20a)
2
or
2.4 The modi ed log-wake law and its defect form
( 2)d 2
= sin2 (20b)
(1 )f 2 Combining (21), (22), (25) and (26) leads to the following
in which the constant 2/ gets buried in in the integration modi ed log-wake law:
and is introduced as per Coles wake function. Clearly, the
3e 3
u 3e yu 2
= + A + 2 sin
sine-square function in pipe ows is due to the effect of pressure- ln (27)
u 2 2 23
gradient. The value of might vary with a Reynolds number
the modi ed law of the wake
the law of the wall
slightly, but a universal constant might be good enough for large
Since the last two terms in the above equation express the devia-
Reynolds number ows.
tion from the law of the wall, following Coles (1956), they de ne
Substituting (15) and (20b) into (14) gives
the law of the wake in this paper, i.e.
u 1 yu 2
= +A + + F1 sin2
ln (21)
3e 3
2
u 2
W = 2 sin (28)
2 23
The last two terms disappear near the wall, thus the values of
To distinguish it from the standard sine-square wake law, this
and A should be the same as those in the law of the wall. It is
then suggested = 2/( 3e) 0.42. The value of A will be paper calls (28) the modi ed wake law. Furthermore, (27) is
called the modi ed log-wake law that is graphically represented
replaced with the maximum velocity umax by using the velocity
by Fig. 3. Equation (27) has at least two advantages over the
defect formulation in Section 2.4. Besides, it is shown later that
= ts the modi ed log-wake law well with standard log-wake law. It shows that the law of the wake in pipes
the value of
results from the pressure-gradient and the axial symmetrical con-
experimental data. Therefore, this paper assumes
dition. It meets the symmetrical condition at the pipe axis where
2
= = 0.42 (22) the standard log-wake law fails.
3e
To eliminate A from the modi ed log-wake law (27), one can
introduce the maximum velocity umax at the axis = 1 to the
2.3 The axial boundary correction modi ed log-wake law. From (27), one obtains
Because of the axial symmetry, the velocity gradient must be zero umax 3e Ru 1
= +A+2
ln (29)
at = 1. From (21), one has u 2 3
1 du 1
= + F1 (1) = 0 Furthermore, eliminating A from (27) and (29) gives the velocity
u d
=1 defect form of the modi ed log-wake law
which gives
umax u 1 3
3e
= ln + + 2 cos2 (30)
1 2 3 2
u
F1 (1) = (23)
This is the most important result of this study.
From (16b) and (23), one can assume
n 1
F1 = (24)
3 Determination of the maximum velocity and comparison
with recent experiments
in which n > 1. Integrating the above equation and applying
(16a) gives
3.1 Determination of the maximum velocity
n
F1 = (25)
n The maximum velocity in (30) plays the role of wall friction that
Since pipe ows are completely symmetrical about the axis, will be discussed in Section 5. This section directly correlates the
mathematically, the pipe velocity pro le is an even function about maximum velocity umax with Reynolds number Ru / . A plot of
Modi ed log-wake law for turbulent ow in smooth pipes 497
Near wall region
R
Core region
=1
Pipe
axis
y
Near wall region
(a) The effect of (b) The effect of the (c) The boundary (d) The composite
the wall pressure-gradient correction velocity profile
Figure 3 Components of the modi ed log-wake law.
3.2 Comparison with recent experiments
This section rst examines the applicability of (30) to describe
individual velocity pro les. The universality of the parameters
is then tested by plotting all data points according to the defect
form (30). Zagarola (1996) at Princeton University measured 26
mean velocity pro les with different Reynolds numbers between
3.1 104 and 3.5 107 . The test pipe was smooth and had a
nominal diameter of 129 mm. The complete descriptions of the
experimental apparatus and experimental data can be found on
the web site http://www.princeton.edu/ gasdyn/ or in Zagarola
(1996).
To illustrate the procedures of analysis, take Run 16 for an
example where R = 6.47 cm, u = 0.7 m/s and = 1.07
10 6 m2 /s. One can calculate that
Ru
= 45290
Figure 4 The relation of umax /u versus Ru / in loglog coordinates.
and from (31c) one gets
Nikurasde s (1932) classical data and the recent Princeton Uni-
versity experiments (Zagarola, 1996) is shown in Fig. 4. One can umax
= 31.2
see that for Ru / 2 104, the data ts the following power law
well, Using the above procedures, all 26 pro les are obtained and plot-
ted in Fig. 5. One can conclude that: (i) the basic structure of the
1/16
modi ed log-wake law is correct; (ii) Eq. (30) can replicate the
umax R u
= 16.55 (31b)
experimental data very well; (iii) the empirical Eq. (31c) for the
u
maximum velocity works very well; (iv) the modi ed log-wake
According to Guo s (2002) logarithmic matching, an accurate law tends to a straight line in the semilog plot near the wall and
curve- tting equation is then obtained for any Reynolds number, then coincides with the log law there; and (v) the zero velocity
gradient at the pipe axis can be clearly seen from all the experi-
1/16
1/8
mental pro les which show that the axial boundary correction is
umax R u 1 Ru
= 9.9 1+ (31c)
necessary.
u 3720
Besides, according to the defect form, all 26 pro les including
where the shape transition parameter = 1 in Guo s (2002) 1040 data points are also plotted in Fig. 6 where all data points
method provides excellent agreement with experimental data, as fall in a narrow band. This shows that the model parameters,
shown in Fig. 4. and n are universal constants.
498 Junke Guo and Pierre Y. Julien
1
100
1 2 3 4 5 17 6 7
0.8
10 1 0.6
RUN Re
4
3.16 10
1
4.17 104
2
0.4
4
5.67 10
3
4
7.43 10
4
4
9.88 10
10 2
5
1.46 105
6 0.2
1.85 105
7
6
3.10 10
17
12 3 4 5
5 17 6 7
0
0 5 10 15 20 25
0 5 10 15 20 25
u (m/s)
u (m/s)
100 1
18 19 8 9 10 11
0.8
10 1 0.6
RUN Re
0.4
5
2.30 10
8
5
3.10 10
9
10 2 4.09 105
10
0.2
5
5.39 10
11
4.42 106
18
18 19 8 9 10 11 6
6.07 10
19
0
0 5 **-**-**-**-*-*-** 15 20 25
u (m/s) u (m/s)
100 1
12 13 14 15 16
0.8
10 1 0.6
RUN Re
0.4
5
7.52 10
12
6
1.02 10
13
10 2 1.34 106
14
0.2
1.79 106
15
12 13 14 15 16 6
2.35 10
16
0
0 5 **-**-**-**-*-*-** 15 20 25
u (m/s) u (m/s)
100 1
0.8
10 1 0.6
RUN Re
0.4
7.71 106
20
7
1.02 10
21
1.36 107
22
10 2 1.82 107
23
0.2
7
2.40 10
24
2.99 107
25
7
3.53 10
26
0
0 *-**-**-**-**-** 35 0 *-**-**-**-**-** 35
u (m/s)
u (m/s)
Figure 5 Comparison of the modi ed log-wake law (30) with Zagarola s (1996) experimental data through individual pro les.
= w (1 )
4 Implication for eddy viscosity (33)
Neglecting the viscous shear stress, from (10) and (33), one can
Eddy viscosity is important when studying turbulent mixing.
show that the eddy viscosity can be expressed by
With the modi ed log-wake law, the eddy viscosity can now
be determined. First, applying (9) to (8) gives the shear stress 1
t
= (34)
distribution: Ru (1/u )(du/d )
Modi ed log-wake law for turbulent ow in smooth pipes 499
(b) 14
(a) 14
12
12
10
10
(umax u)/u*
u)/u
8
8
max
6
6
(u
4
4
Modified log wake law (30)
Data of Zagarola (1996) 2
2
0
03 2 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
10 10 10 10
Figure 6 Veri cation of the model universal constants with the recent Princeton University experiments.
0.1
which is consistent with the mixing length model. Near the pipe
axis, 1, one has
0.09
sin (1 )
sin
0.08
=
1 1
0.07
Substituting it into (35) yields
0.06
/(Ru )
t 1
0.05
= 0.059 (36b)
5
u = 1.05 m/s, Re = 4.1 10, Laufer
(3 3e/2) + 2
t
Ru
4
u = 0.13 m/s, Re = 4.1 10, Laufer
0.04
4
u* = 0.49 m/s, Re = 3 10, Nunner
The constant eddy viscosity near the axis corresponds to an
0.03 u* = 0.15 m/s, Re = N/A, Reichardt
asymptote of a parabolic law (Hinze, 1975, p. 732).
u = 0.20 m/s, Re = N/A, Reichardt
0.02
u = 0.45 m/s, Re = N/A, Reichardt Equation (35) may be the best result so far for the eddy vis-
The eddy viscosity model (35)
0.01
cosity model in pipe ows. It may also be used to study some
Asymptotes (36a) and (36b)
complicated turbulent ows such as a wave turbulent boundary
0
0 0.2 0.4 0.6 0.8 1
= y/R layer ow. All previous velocity pro le models, including the
log law, the log-wake law and the power law, cannot produce
Figure 7 Comparison of the eddy viscosity model with experimental
the maximum eddy viscosity at 0.3 and the constant eddy
data (Data source: Ohmi and Usui, 1976).
viscosity near the axis.
The velocity gradient from the modi ed log-wake law (30) is
5 Friction factor
1 du 3e 1
= 2 + sin
The friction factor is an essential parameter in pipe designs and
u d 2
numerical simulations. It is de ned as
Substitution into (34) gives