977988, 1997
Pergamon
P I I S07351933(97)000833
T U R B U L E N T PIPE F L O W O F P O W E R  L A W F L U I D S
M.R. Malin
C H A M Limited, W i m b l e d o n
London SW19 5 A U
UK
ABSTRACT
This paper reports on the numerical computation of the turbulent flow of powerlaw fluids
in smooth circular tubes. The turbulence is represented by means of a modified version of an
existing twoequation turbulence model. Numerical results are presented for the fullydeveloped friction factor and velocity profile, and compared with experimental data. The
model is shown to produce fairly good agreement with experiment over a wide range of
values for the powerlaw index and generalised Reynolds number. O 1997 ElsevierScienceLtd
Introduction
Flows of nonNewtonian fluids through pipes are relevant in many engineering applications,
especially within the manufacturing, process and wastewater industries. The turbulent flow of these
fluids is less common than laminar flow, but turbulence may be encountered in some situations, e.g.
drilling hydraulics, sewage transport, and applications involving relatively high heat transfer rates. The
present paper will deal with the pipeline flow of timeindependent viscous fluids described by the powerlaw rheological model, which relates the shear stress to the strain rate via the consistency index K and
the powerlaw index n. For values of n<i, the fluid is pseudoplastic (shear thinning), and for values of
n>l, it is dilitant (shear thickening). If n= 1 the fluid is Newtonian.
Extensive experimental and theoretical studies of turbulent nonNewtonian pipe flows were carried
out by Metzner and coworkers during the 1950s [I3]. These workers indicated how the friction factor f
varies with generalised Reynolds number Re [l] in the laminar, transitional and turbulent regimes. For
the fullydeveloped turbulent flow of powerlaw fluids, Dodge and Metzner (DM) [3] developed a semi977
978
M.R. Malin
theoretical expression for the mean velocity profile together with a correlation for f versus Re. The study
revealed that a decreasing n slightly delays transition to higher Re, and that friction factors reduce with
decreasing values of n. Subsequently, several workers [47] derived alternative expressions for the mean
velocity profile, and the relative merits and demerits of these have been discussed elsewhere [810].
More recently, Hartnett and Kostic [I 1] examined the available turbulent frictionfactor correlations, and
found that the DM correlation [3] produced the best agreement with the available measurements. The
transition regime has been considered by Reed and Pilehvari [12] who developed a procedure to calculate
transitional friction factors by combining the laminar and turbulent friction factors of DM [3].
Mohammed et al [13] developed a method for calculating the fullydeveloped friction factor and mean
velocity profile by numerical integration of the meanmomentum equation. The Reynolds stresses were
modelled by analogy with the powerlaw model, and by use of Prandtl's hypothesis [14]. Their
calculations showed fairly good agreement with the measurements of Bogue [15] over a wide range of Re
and n, although the model coefficients were evaluated by reference to this data. The planar turbulent flow
of a powerlaw fluid through a film bearing was calculated numerically by Pierre and Boudet [16], but no
comparisons were made with experimental data. These workers also closed the Reynolds stresses by
analogy with the powerlaw model, but the eddy viscosity was determined by use of the lowReynoldsnumber twoequation ke turbulence model of Lam and Bremhorst (LB) [ 17].
The present study also employs the LB lowRe ke model [17], but the more conventional practice of
determining the eddy viscosity from the linear Boussinesq stressstrain relationship is adopted. Earlier
work [18] has demonstrated that the LB model performs quite well for the calculation of Bingham plastic
fluids in pipelines. This earlier work is extended here to powerlaw fluids, so as to calculate the frictional
resistance, along with the velocity profile, by solving numerically the basic flow equations for fullydeveloped laminar and turbulent flow in smoothwalled tubes. The performance of the model is assessed
by comparing predictions with the available experimental information. It will be shown that the original
LB ke model exhibits significant quantitative deficiencies with regard to the friction factor at low values
of the powerlaw index. Therefore, a modification to the viscous damping is proposed which is inactive
for Newtonian fluids, but is shown to improve the predictions for nonNewtonian fluids.
Mathematical Model
For compactness, the Reynoldsaveraged meanflow equations for a steady turbulent incompressible
flow are written as follows:
V.(U)=O ; V.(pUU)=V.(p{v+vdVU)Vp
+V.(p{v+v,l(VU) ~)
979
(1)
where p is the fluid density, p the pressure, U the velocity vector, vt is the turbulent eddy viscosity arising
from closure of the Reynolds stresses through use of the Boussinesq stressstrain relationship, and v is
the apparent kinematic viscosity of the powerlaw fluid which is given by:
/" 1
.~(nI)/2
v=(KIp)['~(A:A)
I
A = 0.5[VU+ (VU) T]
(2)
where K is the consistency index, n is the powerlaw index, A is the deformation tensor, and the
superscript T denotes that the transpose of the dyadic VU is taken. For pseudoplastic fluids, v decreases
with increasing deformation rate, whereas for dilitant fluids it increases with increasing deformation rate.
The eddy viscosity v, is determined from the LB ke model [17], which uses transport equations for
the turbulent kinetic energy k and its rate of dissipation e. This model calculates vt from:
v,
c,,f,,k ~ / e
(3)
where Ca0.09 and f~ is a damping function, to be defined below. The turbulence parameters k and e are
calculated from the following transport equations:
V . ( p U k ) = V . (p(v+vdoi) V k ) + P(Pk  e)
(4)
(5)
(6)
The model coefficients are given by ak=l.0, ot=1.314, C~t=1.44, Cu=1.92; and the damping functions f~,
fl and t"2are determined from:
f/~ =[I exp(0.0165Ren/nl/4)]2(! +20.51Ret) ; f~ = 1 + (0.05 / f~)3 ; f2 = l+exp(Ret2) (7)
The damping function f~ differs from that proposed by Lam and Bremhorst [17] only in that it
includes the empirical parameter n TM, which will be shown to improve the accuracy of the predictions
when dealing with strongly nonNewtonian fluids. The Reynolds numbers Ren and Ret are defined by
980
M.R. Malin
Re, = .qr~y, / v and Re r = k 2 / ( E V ) , where y, is the normal distance to the wall. This distance is easy
to compute for the present geometry, but difficult for an arbitrary boundary topography in multidimensional problems, ttere, it is estimated from a general, economical method proposed by Spalding
[ 19], which involves the solution for a scalar variable, ~ , which obeys the differential equation V2~ =1
within the fluid, and which equals zero within solid materials and at noslip surfaces. The wall distance is
deduced from the solution for O from consideration of a simple geometry, namely that between two
parallel walls, and then to presume that the resulting relationship:
y~ = [ ( V O ) 2 + 2 ~ ] 1 n _Vqb
(8)
The flow is onedimensional, axisymmetric and fullydeveloped, and boundary conditions are needed
only at the flow axis and wall boundary. At the flow axis a zeroflux condition is employed for all
variables, while at the wall k=0, ~)f/3y=0, and the axial velocity w=0. The model equations are solved
numerically with the finitevolume solution procedure embodied in the generalpurpose PttOENICS
computer code [19]. The solution is obtained iteratively by means of a solver option which determines
the axial pressure gradient from overall continuity for a specified mass flow rate. Typically, the
calculations utilise 120 radial grid cells, with the grid spacing increasing in geometric progression away
from the wall. The progression ratio is between 1.02 and 1.05, and the nearwall grid node is located at y
{=pw.~2")y"/K}_0.5,where w.{=('t:ctp) I'~} is the friction velocity and xw is the wall shear stress.
Results and Discussion
Numerical computations are performed leading to the friction factor for fullydeveloped laminar and
turbulent flow. The parameters are the Fanning friction factor f=2xw/(p wb2 ), the powerlaw index n and
the generalised Reynolds number of Metzner and Reed [l]:
2n
Re =
OWb
(9)
981
produced by the modified LB model. The experimental friction factors in the laminar and turbulent
regimes are represented, respectively, by the analytical [ I ] and DodgeMetzner [3] correlations:
16 / Re
All
..~.u
.~
#.(2_,)/2.L
x~ = n~75lgl(Kes
0.4
s n'2
(10)
In the limiting case of a Newtonian fluid, n=l and the generalised Reynolds number Re reduces to the
conventional Reynolds number, and the DodgeMetzner correlation reduces to the KarmanNikuradse
correlation for Newtonian fluids.
1.8E81
1.8E82
n
1.2
1.8
8.8
Predictions
8.6
o o o
8.4
DOt
1.8E83
1.8E+82
1.8E+83
1.BE'84
Re
1.8E+85
FIG. 1
Frictional resistance: original LamBremhorst model
In purely laminar flow, Figs. 1 and 2 show that there is perfect agreement between the numerical
predictions and the analytical correlation. As noted in earlier work [18], the LB ke model tends to
overestimate the friction factor for turbulent Newtonian fluids (n=l) at Re > 2.104, whereas at lower
Reynolds numbers the model shows fairly close agreement with the measurements.
982
M.R. Malin
1 .BE81
I
1.BE82
1.2
1.8
8.8
 
co~Pelation
Turbulent corrolation
Laminar
o I"1oi
1.8EB3
8.6
Predictions
1.8E+62
ii
11
1.6E +B3
8.4
l,I
I. 8E +64
Re
I. BE "85
FIG. 2
Frictional resistance: modified LamBremhorst model
In the turbulent regime, it is evident from Fig. 1 that the unmodified turbulence model significantly
underestimates the frictional resistance for highly nonNewtonian pseudoplastics (n=0.4), whereas for
dilitant fluids (n=l.2) the friction factor is progressively overestimated with increasing Re. These
discrepancies arise because the original fo damping function ( which depends on n via the apparent
viscosity in the Reynolds numbers Re, and Ret ) produces too low vt values in the former case, and too
high values in the case of dilitant fluids. Fig. 2 reveals that the modified fg damping function significantly
reduces this oversensitivity of the original model, as fairly good agreement is achieved for all values of n
over the entire Reynoldsnumber range.
It is interesting to comment on the ability of the model to predict the critical Reynolds number Rec at
which the flow undergoes transition to turbulence. In the calculations this is defined by the point where f
achieves its minimum value, but it should be mentioned that no attempt has been made to determine the
precise values of Rec. It is seen from Fig.2 that the calculated transition occurs at essentially the same
Reynolds number for 0.6<n<l.0, whereas experiments [3] suggest that Rec increases slowly with
decreasing values of n. In the calculations, this feature is observed for strongly nonNewtonian fluids
983
(n0.4) where transition occurs at Rec=3,000, which agrees quite well with the value reported by Dodge
and Metzner [3]. However, an adequate prediction of transition cannot really be expected unless the
model is extended to account for the intermittent nature of the transitional flow.
2,S
un~
.=2,.o
Re = 5 0 9
2.0
1.5
1.0
0.5
o
e.o
O.O
I
9.Z
P r o d i c t ions
Amtlgtica I
a~ 0
%
I
0.6
I
0.4
I
6.8
r/R
1.0
FIG. 3
Laminar axial velocity profiles
The implementation of the power law model in the numerical solution procedure is further verified by
comparing the predictions obtained for the fullydeveloped velocity profile in laminar flow at Re=500
with those obtained by the exact analytical solution [9,10]:
.~w = ~ 3 n+ 1 [1 wb
n+ 1
R) n+l)ln]
(11)
where R is the pipe radius. The analytical and numerical results are shown in Fig. 3 for different values
of n at Re=500. It can be seen that the predictions are in excellent agreement with the analytical values,
and the figure illustrates the wellknown effect of n on the velocity profile; i.e. for pseudoplastics the
profile becomes progressively flatter; and for dilitant fluids the profile becomes progressively linear. Of
course, for Newtonian fluids the velocity profile is parabolic.
984
M.R. Malin
Figures 4 and 5 compare the predicted meanvelocity profiles with those measured by Bogue [15] for
fullydeveloped turbulent flow. Fig. 4 shows results obtained at Re=197,000 for a Newtonian fluid, and
at Re=107,000 for a slightly nonNewtonian fluid (n=0.825). Fig. 5 presents results for a moderately nonNewtonian fluid (n=0.615) at Re=23,240, and for a strongly nonNewtonian fluid (n=0.465) at
Re=12,900. In general there is good agreement between the calculated and measured profiles, although
the lack of any data beyond r/R=0.9 precludes validation of the model closer to the wall.
1.5
W/Idb
[]
1.8
 8.S
Predlction~
Da~a
[15]
Re = 1 . 9 7 E 5
n
= 1.8
1.5
t,l,,"lJb
1.8
P~dlctlons
Data
8.S
[1S]
Ra = 1 . 8 7 E 5
n = 6.825
8.B
8.8
I
8.Z
I
8.4
I
8.6
I
8.8
r/R
1.8
FIG .4
Turbulent mean velocity profiles for Newtonian and slightly nonNewtonian fluids
For fully turbulent flow at Re~O,000,velocity profiles in w+y + coordinates are plotted in Figure 6
for various values of n. The figure compares numerical predictions of the mean velocity with the
following semitheoretical 'twolayer' velocity distribution:
w+ = mm
( w I+ ,w,+ )
; w~ =fy+),/n
4x/2
w+ =2.46n25tln(y+)lln +A(n)+B(~,n)] 0 .nL
2
(12)
where w+=w/w., w~ is the laminar sublayer velocity [3], w+ is the velocity in the turbulent region [7],
and
985
 ~ n+
n~
~ } p ~   ~ ( 
. ) ] (13)
wherein ~=y/R and y is the distance from the wail. The velocity predictions of Fig. 6 are also plotted in
Figure 7, but normalised in terms of wb and R. The corresponding calculated profiles of turbulent kinetic
energy are presented in Figure 8.
1.5
W/IJb
1.8
Predictions
8.5
Data
[151
Be
= 1.29E4
Be
n
= 2.324E4
= I].615
8.465
1.5
W/Wb
1.8

Predictions
Data
8.5
8.8
8.8
[15)
8.2
8.4
8.8
8.8
r/D
1,8
FIG. 5.
Turbulent mean velocity profiles for moderate and strongly nonNewtonian fluids
Figure 6 shows that the thickness of the viscous sublayer reduces with decreasing n, a trend which
might have been anticipated from the laminar velocity profiles shown earlier in Fig. 3. The influence of
turbulence is to transport highermomentum fluid towards the wall, thereby leading to broader velocity
profiles, as is evident from the profiles shown in Fig. 7. The velocity predictions shown in Fig. 6 agree
fairly well with the logarithmic distributions, except in the transition zone which is not correctly
described by the twolayer approximation. It is interesting to note that the limits of the transition zone
indicated by the present model agree reasonably well with those proposed by Clapp [4]: 5n < y+ < y~,
where y~ is the boundary between the transition and turbulent zones y~ = exp[(3.8 + 3.05n) / 2.22].
986
M.R, Malin
I
i
30
Re = 4E4
n=O .4
n=0.6
//
W+
~]
n=O .B
20
o
Q
10
P v e d ict ions
1.0E01
1.0E+00
,,IHI
NonHewtonian
* *,*,,I
1.0E+01
1,0E+02
'
'l'l'l
loglaw
[71
lllll
1.8E+03 y+ 1.0E+04
FIG. 6
Effect of n on turbulent mean velocity profiles plotted in wall coordinates
0.6
0.8
0.6
0.0
0.2
0.4
FIG. 7
Effect of n on turbulent mean velocity profiles
r/R
1.8
987
Finally, Figure 8 demonstrates the effect of the powerlaw index on the turbulence energy profiles at
Re=40,000.
It is seen that in addition to the narrowing of the viscous sublayer, there is also a
pronounced attenuation of the turbulence with decreasing powerlaw index. From Figs. 6 and 7, it is
evident that the velocity profiles become progressively flatter owing to the increase in apparent viscosity,
and hence turbulence production diminishes over the entire pipe cross section.
8.815
2
R/W b
_n
Re=dE4
8.81e
1.2
"
8.G
8.805
8. B88
8.8
I
8.2
I
8.4
I
8.6
I
8.8
t/B
1.8
FIG. 8
Effect of n on turbulenceenergy profiles
Conclusions
A series of numerical computations has been performed to calculate fullydeveloped laminar and
turbulent flow of powerlaw fluids in smooth tubes. A modified version of the LamBremhorst ke
model has been tested against experimental data on the friction factor and mean velocity profile for
various generalised Reynolds numbers, with different values of the powerlaw index n. Generally, the
model produced fairly good agreement with the measured data, but the delay of transition to turbulence
was predicted only for strongly nonNewtonian fluids, i.e. n<0.6.
988
M.R. Malin
References