CFD Modeling of Heat Transfer in Turbulent

Pipe Flows
S. S. Thakre and J. B. Joshi
Dept. of Chemical Technology, University of Mumbai, Matunga, Mumbai}400019, India

Twel®e ®ersions of low Reynolds number k- e and two low Reynolds number Reynolds
stress turbulence models for heat transfer were analyzed comparati®ely. Predictions of
the mean axial temperature, the radial and axial turbulent heat fluxes, and the effect of
Prandtl number on Nusselt number were compared with the experimental data. The
model by Lai and So from the k- e group and Lai and So from the Reynolds stress
group had the best o®erall predicti®e ability for heat transfer in turbulent pipe flow. The
Lai and So model was attributed to its success in the predictions of flow parameters
such as mean axial ®elocity, turbulent kinetic energy, eddy diffusi®ity, and the o®erall
energy dissipation rate. The k- e models performed relati®ely better than the Reynolds
stress models for predicting the mean axial temperature and the Nusselt number. This
qualitati®e and quantitati®e study found the need for more sophisticated near-wall exper-
imental measurements and the accuracy of the dissipation (of turbulent energy) and the
pressure-scrambling models.

Introduction
In the recent years, low Reynolds number modeling has in an eddy-diffusivity model, in which the eddy diffusivity for
been widely used for the flow predictions in the turbulent momentum is evaluated by either a mixing length model or
shear flows. The low Reynolds number modeling approach the k-e model, and the eddy diffusivity for heat is evaluated
incorporates either a wall-damping effect or a direct effect of using Prt . The experimental evidence suggests that Prt is a
molecular viscosity, or both, on the empirical constants and number very near unity, and its departure from unity is not
functions in the turbulence transport equations. A fairly com- great except for very low Prandtl number fluids. However,
plete review of the low Reynolds number k-e modeling of the experimental studies show that the value of Prt increases
turbulent shear flows has been given by Patel et al. Ž1985. steeply and far exceeds unity as the wall is approached.
and Hrenya et al. Ž1995.. The models reviewed by Patel et al. There is apparently a possibility of analyzing the behavior
Ž1985. and Hrenya et al. Ž1995. were focused on the flow of this parameter by using the near-wall turbulent heat-flux
patterns. Since the heat-transfer process depends entirely on models. In one such attempt, Lai and So Ž1990b. have shown
flow pattern, the model that gives good flow predictions can that Prt is approximately constant for most of the pipe; how-
be expected to give good heat transfer predictions as well. ever, it increases steeply and far exceeds unity as the wall is
However, this statement is in the form of a question. There approached. Therefore, any heat-transfer model that as-
are few possible reasons for this limitation: Ž1. there have sumes a constant Prt needs very careful analysis, even for
been rare attempts to extend the flow knowledge from low simple flows like turbulent pipe flow. However, it must be
Reynolds number models to heat transfer; Ž2. the limitation pointed out that so far no suitable near-wall second moment
as regards the accuracy of modeling the energy-dissipation closure for scalar flux transport has been developed. This is
equation Žwhich directly affects the flow quantities and due, in part, to a lack of detailed near-wall scalar flux mea-
thereby heat transfer .; and Ž3. the difficulty of the turbulent surements and partially to the unavailability of an asymptoti-
Prandtl number Ž Prt . concept. The Prt concept is widely used cally correct near-wall Reynolds stress model. Hence, com-
prehensive work is still needed for the estimation of Prt , us-
Correspondence concerning this article should be addressed to J. B. Joshi. ing turbulent heat-flux models.

1798 September 2000 Vol. 46, No. 9 AIChE Journal

 The shortcomings just cited motivated us to critically ana- Ž1981., Chein Ž1982., Nagano and Tagawa Ž1988., Myong and
lyze the validity of both the low Reynolds number k-e and Kasagi Ž1990., Lai and So Ž1990a., Shih and Mansour Ž1990.,
the Reynolds stress models for the process of heat transfer. So et al. Ž1991., Fan et al. Ž1993., Yang and Shih Ž1993., and
The study was restricted to the turbulent pipe flow condition, Cho and Goldstein Ž1994.. In the present work, more atten-
a relatively simple state, yet of the utmost importance to the tion is paid to the heat-transfer aspects. For brevity in the
chemical engineering community. As a first step, it was subsequent discussion, these models are referred to as JL,
thought desirable to critically analyze the existing low LS, LB, CH, NT, MK, LSO, SM, ST, FLB, YS, and CG model,
Reynolds-number two-equation k-e models and Reynolds respectively.
stress models for the flow patterns. Numerical codes were
developed for more recent versions of twelve low Reynolds The low Reynolds number model
number k-e and two low Reynolds number Reynolds stress
models for the turbulent pipe flow condition. The knowledge For steady, isothermal, incompressible, fully developed tur-
of flow pattern was then used for the numerical prediction of bulent pipe flow, the set of governing equations for flow and
the heat-transfer process. These numerical heat-transfer pre- heat are given as follows.
dictions were compared with the experimental data of Gowen Flow equations
and Smith Ž1967., Kader Ž1981., Bremhorst and Bullock Ž1970,
1973., Hishida et al. Ž1986., and Mizushina et al. Ž1971.. The 1 d du 1 dp
present article focuses more on the heat-transfer aspects. 0s r Žn qnT .
ž / ž / y , Ž1.
r dr dr r dz c

Heat Transfer Analysis from the k - e Models where

The low Reynolds number k-e model of turbulence has
been widely used in numerical simulations due to its simplic- k2
n T sCm fm . Ž2.
ity and, to some extent, capability for predicting wall-bounded e
turbulent flows. Although these models work well for the flow
predictions, they offer some limitations when applied to heat- The general form of the transport equations that determine k
and mass-transfer calculations at solid boundaries. In these and e in the low Reynolds number models, simplified for the
situations they can yield unrealistic predictions for the flux of flows considered here Žfully turbulent pipe flow., is given as
energy and matter, as well as for the profiles of the corre-
sponding scalar quantities. A literature review shows that the
existing near-wall formulations yield reasonable predictions 1 d nT dk du 2
nq qnT ye y D
for friction factors and thus for heat-transfer rates when the
Prandtl number is of the order of unity, but fail at large
0s
r dr
r
žž sk / /
dr ž / dr
Ž3.

Prandtl numbers. At higher Prandtl numbers or at low ther- 1 d nT de Ce 1 f 1 n T e du 2

nq
mal diffusivity, n T values very near the wall have substantial
influence on heat-transfer rates.
0s
r dr
r
žž sk / / dr
q
k ž / dr
In view of the shortcomings just cited, it was thought desir-
Ce 2 f 2 e 2
able to critically analyze the low Reynolds k-e models for the y qE Ž4.
near-wall heat-transfer predictions. The general computing k
procedure for heat transfer is to first obtain the flow predic-
tions and then extend this knowledge for the heat-transfer where Cm , Ce 1 , Ce 2 , s k , and se are the same empirical con-
calculations. For this purpose, twelve different low Reynolds stants found in the standard high Reynolds number k-e mod-
number k-e models have been considered. The models se- els. The functions fm , f 1 , and f 2 , and, in some cases, the D
lected for the comparison are those of Jones and Launder and E terms are included in the low Reynolds number mod-
Ž1972., Launder and Sharma Ž1974., Lam and Bremhorst els in order to render the models valid in the near-wall re-

Table 1. Numerical Values for the Constants Cm , Ce 1 , Ce 2 , s k , and se for the Low Reynolds Number k - e Models
S. No Researchers Model Cm Ce 1 Ce 2 sk se
1. Jones and Launder Ž1972. JL 0.09 1.45 2.0 1.0 1.3
2. Launder and Sharma Ž1974. LS 0.09 1.44 1.92 1.0 1.3
3. Lam and Bremhorst Ž1981. LB 0.09 1.44 1.92 1.0 1.3
4. Chein Ž1982. CH 0.09 1.35 1.8 1.0 1.3
5. Nagano and Tagawa Ž1988. NT 0.09 1.45 1.9 1.4 1.3
6. Shih and Mansour Ž1990. SM 0.09 1.45 2.0 1.3 1.3
7. Lai and So Ž1990a,b. LSO 0.09 1.35 1.8 1.0 1.3
8. Myong and Kasagi Ž1990. MK 0.09 1.4 1.8 1.4 1.3
9. So et al. Ž1991. SZS 0.096 1.5 1.83 0.75 1.45
10. Yang and Shih Ž1993. YS 0.09 1.44 1.92 1.0 1.3
11. Fan et al. Ž1993. FLB 0.09 1.4 1.8 1.0 1.3
12. Cho and Goldstein Ž1994. CG 0.09 1.44 1.92 1.0 1.3

AIChE Journal September 2000 Vol. 46, No. 9 1799

 Table 2. Summary of the Damping Functions fm , f 1 , and f 2 for the Low Reynolds Number k - e Models
Model fm f1 f2
JL exp wy2.5rŽ1q R Tr50.x 1 1y0.3 exp Žy R T2 .
LS exp wy3.4rŽ1q R Tr50. 2 x 1 1y0.3 exp Žy R T2 .
LB w1yexp Žy0.0165R y .x2 1qŽ0.05rfm . 3 1yexp Žy R T2 .
Ž1q20.5rR T .
CH 1yexp Žy0.0115 yq . 1 1yŽ2r9. exp ŽyŽ R Tr6. 2 .
NT w1yexp Žy yqr26.5. 2 x 1 1y0.3 exp wy
Ž1q4.1rR T3r4 . Ž R Tr6.5. 2 x4w1yexp
Žy yqr6.x2
SM 1yexp Žy6=10y3 yq y4= 1 1y0.22 exp wy
10y4 q2
y q2.5=10y6 yq3 y4.0=10y9 yq4 . Ž R Tr6. 2 x4ere
LSO 1yexp Žy0.0115 yq . 1q1y0.6 exp wy 1yŽ2r9. exp wyŽ R Tr6. 2 x
Rer10 4 x4 exp wy
Ž R Tr64. 2 x
MK w1yexp Žy yqr70.x 1 1yŽ2r9. exp wyŽ R Tr6. 2 x4
Ž1q3.45r'R T . wŽ1yexp Žyyqr5.x2
SZS Ž1q3.45r'R T . tanh Ž yqr115. 1 1
YS Ž1q1r'R T .w1yexp Ž1.5=10y4 R y 'R TrŽ1q'R T . 'R TrŽ1q'R T .
y5.0=10y7R 3y y1.0=10y10 R 5y .x1r2
FLB 0.4 f wr'R T qŽ1y0.4 f wr'R T . 1 1yŽ2r9 exp wy
w1yexp Žy R yr42.63.x3 Ž R Tr6. 2 x4 f w2
where
f w s1yexp y'R yr2.3qŽ'R yr2.3y
R yr8.89.w1yexp Žy R yr20.x 3 4
CG 1y0.95 exp Žy5=10y5 R T2 . 1 1y0.222 exp wyŽ R T2 r36.x

gion. The models vary within the general framework in terms ­T 1 ­ n nT ­T

of the assigned constants, damping functions, and the terms
D and E as seen from Tables 1, 2, and 3, respectively.
u
­x
s
r ­r žž
r
Pr
q
Prt / /
­r
. Ž5.

Heat-Balance Equations. The following thermal-energy We consider heat transfer in a pipe with constant heat flux
equation defines the heat-transfer problem: through the wall. The convective term on the lefthand side of

Table 3. Summary of the D and E Terms, and Wall Boundary Conditions for k and e for the Low Reynolds Number k - e
Models
Wall Boundary
Model D E Conditions
JL 2 n Ž d'k rdr . 2 2 nn T Ž d 2 urdr 2 . 2 kse s0
LS 2 n Ž d'k rdr . 2 2 nn T Ž d 2 urdr 2 . 2 kse s0
LB 0 0 k s 0; e s n d 2 krdr 2
CH 2 n kry 2 Žy2 nery 2 . exp Žy0.5 yq . kse s0
­k ­ k2
NT 0 0 s0 e s y 2
­y ­y
SM 2 n Ž d'k rdr . 2 nn T Ž d 2 urdr 2 . 2 k s 0; e s n d 2 krdr 2
LSO 0 2 n C2 f 2 Ž erk .Ž d'k rdr . 2 k s 0; e s 2 n Ž d'k rdr . 2
T
qexp wyŽ R Tr64. 2 x
ž9 2
C y2
/
= erk Ž e y2 n Ž d'k rdr ..
yŽ1r2 k .Ž e y2 n kry 2 . 2 .
MK 0 0 k s 0; e s n d 2 krdr 22
d'k
2 n C2 f 2 Ž erk .Ž d'krdr . 2 k s 0; e s n
SZS 0

qexp wyŽ R Tr64. 2 xwyŽ2 erk .

ž /
dr

Ž e y2 n Ž d'krdr . 2 .
qŽ3r2 k .Ž e y2 n kry 2 . 2 x
YS 0 nn TrŽ d 2 urdr 2 . 2 k s 0; e s n d 2 krdr 2
­e
FLB 0 0 k s 0; s0
­r
­e
CG 0 Se k s 0; s0
­r
Se s1.44Ž1y fm .wŽ2 nn T Ž d 2 urdr 2 . 2 q2 n Ž d'k rdr . 2 Ž erk ..x
qmaxw0.83Ž e 2rk .Ž lrC1 y y1.Ž lrC1 y . 2 ,0.0x

1800 September 2000 Vol. 46, No. 9 AIChE Journal

Eq. 5 was normalized according to Yakhot et al. Ž1987.. The near-wall scalar measurements and, partially, to the unavail-
heat equation ŽEq. 5. was normalized as: rqs rrR; uqs urut ; ability of an asymptotically correct near-wall Reynolds-stress
TqsTrT U , where T U s qrCP r ut , and nqs nrut R, nq Ts model.
n Trut R. In view of this shortcoming, the model proposed by Lai
and So Ž1990a. looks promising. Lai and So Ž1990a. ŽLSO.,
Boundary conditions developed a near-wall Reynolds-stress turbulence model that
can correctly predict the anisotropy of the turbulent normal
Since the flow is axisymmetric, only the boundary condi- stresses. The success of that model motivated Lai and So
tions at the wall and the symmetry line are required to the Ž1990b. to extend the momentum approach to model turbu-
specified: lent heat transport near a wall. The modeling approach was
similar to that outlined in Lai and So Ž1990a. and was based
du dk dT
r s 0; s s s0 on the limiting wall behavior of the heat-flux transport equa-
dr dr dr tions. This way, the modeled equation was valid all the way
dT to the wall and the assumption of a temperature wall func-
r s R; us k s 0; qsy k . Ž6. tion, and a constant turbulent Prandtl number was not re-
dr
quired. Prud’homme and Elghobashi Ž1986. ŽPEL. proposed
As summarized by Patel et al. Ž1985., three boundary condi- a second-order closure for momentum and the algebraic stress
tions have been employed in the past for dissipation variable. model for the heat transfer. In the present work, some modi-
These are e w s 0.; Ž ­er­ y . w s 0; e w s n Ž ­ 2 kr­ y 2 . w or the fications have been incorporated in the heat-transfer part.
equivalent e w s n Ž2 kry 2 . w . The boundary conditions used by The algebraic stress model for heat was replaced with the
various authors are summarized in Table 3. Equation 5, along second moment heat closure of LSO. This modification was
with the boundary conditions ŽEq. 6., was solved by the nu- mainly due to our interest in the low Reynolds number sec-
merical technique outlined in the fourth section. Since the ond moment closures for heat, and to present a comparative
heat transfer and the momentum equations present a set of study of the predictive capability of these two models for heat
coupled equations, the eddy viscosity and the axial velocity transfer.
profiles obtained from flow are used for the solution of the
thermal energy equation ŽEq. 5.. Their iterative procedure Model of Lai and So
was implemented for the simulations of the momentum equa- Flow Equations. Lai and So Ž1990a. solved the full
tions to obtain the complete flow details, and these results Reynolds stress closure along with the mean axial velocity
were used for the solution of the heat equation. Thus, only and the dissipation rate equation:
one iteration was enough for getting the temperature profile.
The only input used for solving the heat problem was the du rut2
Prandtl number. The turbulent Prandtl number was assigned n yuX ®Xq s0 Ž7.
dr R
a constant value of Prt s1.
1 d k duX 2 1 d k duX ®X
r n qC s ® 2
ž q/ rC s 2uX ®X
Heat-Transfer Analysis from the Turbulent r dr e dr r dr e dr
Heat-Flux Models
2 1 2 du
Although much progress has been achieved in recent years y2 1y a q b q a U f w, 1 uX ®X
ž /
3 3 3 dr
in the modeling of the Reynolds stress transport equations,
the modeling of the scalar field, on the other hand, is still e 2 2
rather primitive. This is because, the model development of yC1Ž 1y f w , 1 . uX 2y k y e Ž 1y f w , 1 .
ž /
k 3 3
the heat flux depends largely on the availability and correct-
ness of turbulent stresses predicted by the Reynolds-stress e uX 2
model. Furthermore, heat-flux transport is influenced by more y f w, 1 s0 Ž8.
k 3®X 2
than one time scale. Consequently, it is more difficult to
achieve a closure of the heat-flux transport equations than
the Reynolds-stress equations. In addition, a shortage of reli-
ž 1q
2k /
able and relatively accurate near-wall heat-flux measure- 1 d k d®X 2 2 k dwX 2
ments also contributes to the slow development of a near-wall r n q3C s ®X 2
ž y C s ®X 2
/
r dr e dr r e dr
turbulence model for heat fluxes. Realizing the limitation of
the calculation methods based on Prt , researchers try to im- k 4 2n
prove the modeling by turning to two equation k-e and the C s wX 2 ®X 2ywX 2 q 2 wX 2y®X 2
y Ž . Ž .
r e 2
r
algebraic flux models for heat transport. Despite some suc-
2 du e X2 2
cess, it is still believed that the most reliable prediction meth- y Ž a y2 b y a U f w, 1 . uX ®X yC1 Ž 1y f w , 1 . ® y k ž /
ods are those based on a second moment closure. The reason 3 dr k 3
is that the turbulent interactions that generate the Reynolds
2 e e ®X 2
stresses and heat fluxes can be treated with less empiricism. y e Ž 1y f w, 1 . y2 f w , 1 ®X 2y4 f w , 1 s0 Ž9.
A detailed literature review shows that, so far, no suitable 3 k k 3®X 2
near-wall second moment closure for scalar flux transport has
been developed. This is due, in part, to a lack of detailed
ž 1q
2k /
AIChE Journal September 2000 Vol. 46, No. 9 1801
 Table 4. Numerical Values of Constants for the Low Reynolds Number Reynolds Stress Models
Researchers Model C1 C2 C3 Ce 1 Ce 2 Cs Ce aU a b g
Prud’homme PEL 1.17 0.3 0.123 1.45 1.9 0.25 0.15 Ž1r11. Ž1r11. Ž1r55.
and Ž8qC2 . Ž8C2 y2. Ž30C2 y2.
Elghobashi
Ž1986.
Lai and So LSO 1.5 0.4 } 1.35 1.8 0.11 0.15 0.45 Ž1r11. Ž1r11. Ž1r55.
Ž1990. Ž8qC2 . Ž8C2 y2. Ž30 C2 y2.

1 d k dwX 2 2 d k The model constants used in the preceding closures are given
r n qC s ®X 2
ž q C s wX 2 ®X 2ywX 2
/ Ž . in Table 4. The expressions for the damping functions em-
r dr e dr r dr e
ployed in these models are given in Table 5.
Heat-Balance Equations. Lai and So Ž1990b. solved the full
4 k 2 k dwX 2 turbulent heat-flux closures along with the mean axial tem-
q C wX 2 ®X 2ywX 2
2 s Ž . q C s ®X 2
r e r e dr perature equation:
2 du
y Ž a q b y a U f w, 1 . uX ®X 1 ­ ­T 1 d ­T
3 dr a
r ­r ž / r
­r
y
r dr
Ž r®X T X . s u ­ x . Ž 13.
e 2 2n 2
yC1Ž 1y f w , 1 . ž wX 2y k q 2 ®X 2ywX 2 y e Ž 1y f w , 1 .
/ Ž .
k 3 r 3 The heat-flux terms ®X T and uX T X were expressed by the fol-
lowing equations:
e wX 2
y f w, 1 s0 Ž 10 .
k 3®X 2 1 d a yn k d®X T X
ž 1q
2k / r dr žžr nq
3
q2C su
e
X2
® / dr /
1 d k duX ®X 1 d k d®X 2 2 k X2
®X T X a y n ®X T X ­T
r n q2C s ® X2 X X
C su w y nqž / y®X 2
r dr ž e / dr
q
r dr
rC s u ®
e dr
y
r e r 3 r2 ­r

­T e 'k
1 k dwX 2 wX 2 du yuX ®X y Ž 1y f w, u . C1 u ®X T XyC1 u , w ®X T X
y Cs u ® q2 X X
y Ž 1y a q a f w, 1 . ® U X2 ­x k Ry r
r e dr r dr
e 1 e
y f w, u ®X T Xy f w , u 1q ž / ®X T Xs 0 Ž 14 .
X2
du e X X k Pr k
y bu y g k
Ž . dr yC Ž1y f 1 w,1. u®
k
1 d a yn k duX T X
r nq qC su ®X 2
e
y f w, 1 uX ®Xy2 f w , 1
k
e
k
X X
u®
X2
s0 Ž 11 . r dr žž 2 e / dr /
3®
d®X T X ­T ­T
ž 1q
2k / q
1 d
r dr
k
rC su uX ®X
ež dr
yuX ®X
­x
yuX 2
­x /
1 d k de du
r n qCe ®X 2 yCe 1 Ž 1q s f w , 2 . uX ®X du e du
r dr že dr / dr y®X T X
dr
y Ž 1y f w, u . C1 u uX T XqC2 u , w ®X T X
k dr
ee 7 ee 1 eU2 f w, u 1 e
yCe 2 fe q fw , 2 ž Ce 2 y2 / y s 0. Ž 12 . y ž 1q / uX T Xs 0. Ž 15 .
k 9 k 2 k 2 Pr k

Table 5. Summary of the Damping Functions and Wall Boundary Conditions for Turbulent Kinetic Energy Dissipation Rate
for the Low Reynolds Number Reynolds Stress Models
Wall Boundary
Model fm f s1 f s2 f Ž s. f w1 fw 2 fe Conditions
y1
RT ­e
} } } }
PEL exp wŽy3.4.r
Ž1q R Tr50 .2 x
ž 1q
10 / 1yŽ1y
Ž1.4rCe 2 .. exp wy
­r
s0

Ž R Tr36. 2 x
LSO 1yexp Ž0.0115 y 4 . } } } exp wyŽ R Tr exp wyŽ R Tr 1yŽ2r9. exp wy e s 2 n Ž d'k rdr . 2
150. 2 x 64. 2 x Ž R Tr6. 2 x

1802 September 2000 Vol. 46, No. 9 AIChE Journal

Model of Prud’homme and Elghobashi specified. These are
Flow Equations. Prud’homme and Elghobashi Ž1986.
solved the full Reynolds-stress closure along with the mean usuX 2s®X 2swX 2suX ®Xs 0
axial velocity and the dissipation rate equation:
­'k
e s 2n
n
du
dr
yu ® q X X
rut2
R
s0 Ž 16.
ž / ­y
at rsR

dT
X2
qsy k , ®X T XsuX T Xs 0 at rsR
1 d k du 2 dr
X2
r n qC s fm
ž ® / y2 Ž 1y a . q
ž Ža qb .
r dr e dr 3
de duX 2 d®X 2 dwX 2
s s s s 0, uX ®Xs 0 at r s0
2 du e 2 dr dr dr dr
X X X2
q /
fs u ® yC1 u q e Ž C1 y1 . s 0 Ž 17.
3 dr k 3 ­T duX T X
s s0 ®X T Xs 0 at r s 0, Ž 22.
X2
­r dr
1 d k d® du
r n qC s fm
ž ®X 2 / q2uX ®X b
r dr e dr dr The energy-balance analysis over a control volume between x
and x q dx and the fully developed condition gives
2 du e du
y Ž a qg . uX ®X yC1 ®X 2q2C3uX ®X f Ž s.
­T 2q
3 dr k dr
s . Ž 23.
X2 X2
­x r Cp ub R
k ® yw
y2 ž n qC s fm wX 2
e /Ž r
. Ž 18 .
To obtain the solution for the governing equations, first the
dependent and independent variables are normalized in the
1 d k dwX 2 2 du following manner: u by ut , uXi uXj by ut2 , e by ut3rR, ŽTw yT .
r n qC s fm
ž wX 2 / y Ž a q b . uX ®X by T U , uXiT X by ut T U , and r by R. The expression for normal-
r dr e dr 3 dr
ized heat-transfer coefficient ŽNusselt number. was obtained
in the following form:
e 2
yC1 wX 2q e Ž C1 y1 .
k 3
2 ReU Pr
Nuqs . Ž 24 .
X2
k X2 Ž® ywX 2 . Tq
y2 ž n qC s fm w / s0 Ž 19 .
e r
This equation was used for the numerical predictions of Nus-
X X
selt numbers from both the k-e and Reynolds-stress models.
1 d k du ® du e
r n qC s fm
ž ®X 2 / y b1 q Ž C1 q f Ž s . . uX ®X
r dr e dr dr k Method of Solution
The solution procedure consisted of two steps: the first step
k uX ®X du
q ž n qC s fm wX 2 yC3/ Žu X2
y®X 2 . dr f Ž s. s0 Ž 20 . was to solve the momentum equations to obtain the mean
e r2 axial velocity, the normal stresses, the Reynolds stresses, the
turbulent kinetic energy Ž k ., and the turbulent kinetic energy
1 d k de e du e2 dissipation rate Ž e .. Since the momentum and heat-flux
r n qCe fm
ž ®2 / yCe 1 uX ®X yCe 2 fe equations present a set of coupled equations, the flow infor-
r dr e dr k dr k
mation obtained from step one was used in step two: to ob-
2 2 tain the mean temperature and turbulent heat-flux distribu-
e d'k k X2
d2u
qCe 2 fe n q2 n
k ž / dr e
fm ®
ž / dr 2
s 0. Ž 21 . tions. Thus, the solution for the coupled momentum]heat
model necessitates the solution of four ordinary differential
equations for k-e models and nine for the Reynolds-stress
Heat-Balance Equations. Prud’homme and Elghobashi models. The governing equations for both, the k-e models as
Ž1986. proposed an algebraic-stress model for the heat trans- well as Reynolds-stress closures, are ordinary differential
port. In the present work, however, the algebraic-stress model equations, and therefore can be solved by any iteration
was replaced with the turbulent heat-flux model of Lai and scheme for split boundary-value problems. The dependent
So Ž1990b. given by Eqs. 13]15. variables were normalized by ut and ut rR Ž u by ut , k by ut2 ,
and e by ut3rR ., while the r coordinate was normalized by R
so that the integration was carried out from symmetry line to
Boundary conditions the wall. Thus formulated, Res ut Rrn s Ž ut r2 u 0 . R D be-
Since the flow is axisymmetric, only the boundary condi- came the only input parameter to the fully developed flow
tions at the wall and the symmetry line are required to be calculations, where R D s u 0 2 Rrn , and u 0 is the mean axial

AIChE Journal September 2000 Vol. 46, No. 9 1803

velocity at the pipe center. Similarly, for heat transfer, Pr s64.0; and Kader Ž1981. for Pr s95 and 170. In the present
was the only input parameter needed to be defined. investigation, the low Prandtl number data Ž Pr s 5.7. of
A finite-volume technique proposed by Patankar Ž1980. was Gowen and Smith Ž1967. and high Prandtl number data Ž Pr
used for the solution of the four coupled nonlinear ordinary
differential equations for k-e models and nine equations for
Reynolds-stress models. The integration was carried from
center to the wall. In all the cases, a nonuniform grid was
used to carry out the calculations. Grid generation is one of
the important aspects of the numerical simulation. The ro-
bustness of any numerical code depends on the effectiveness
and stability of the grid-generation scheme employed for in-
vestigation. In the present work, the following two grid-gen-
eration techniques were employed:
1. By locating more than half of the points in the range
rrRG 0.9. This method gave good predictions for both k-e
and Reynolds-stress closures. However, it was observed that
for high Reynolds number, it shows inefficiency to depict the
peak in the turbulence quantities. Moreover, for high grid
densities Žabove 120 grid points., as should be used for higher
Reynolds number and Prandtl number flows, this method
leads to the numerical error.
2. By obtaining an increasingly fine mesh as the wall is
approached. This grid system was found to be most effective
for both the k-e and Reynolds-stress closures. Moreover, this
method worked fine for high grid densities Žup to 180 grid
points., and hence was useful at higher Reynolds number and
Prandtl number flows.
The first technique was found to be useful only for the LB
model. For the remaining models, the second technique was
preferred due to the reasons just cited. The predictions ob-
tained from both techniques Ž1. and Ž2. were found to be
insensitive to the grids, since doubling the number of grid
points changed the solution profiles by less than 1%. Al-
though the influence of grid density on flow is expected, the
degree of sensitivity is more for heat transfer. This is espe-
cially true for the case of higher Prandtl number, the reason
being that, at higher Prandtl numbers, the thickness of the
thermal boundary becomes smaller than the viscous sublayer.
Hence, the heat transfer gets confined to a very small region
near the wall. In order to effectively capture the steep tem-
perature gradients in this region, it is essential to define fine
grid size in the near wall region. In the present work, the grid
densities used ranged from 70 to 120 points. The number and
concentration of grid points used depended on the Reynolds
and Prandtl numbers under consideration.

Results and Discussions

Comparison of numerical predictions with the experimental
data for k- e models
In order to validate the low Reynolds number k-e models
for heat transfer, the fully developed pipe flow temperature
measurements with constant wall heat-flux boundary condi-
tions were chosen. Unlike the measurements of flow quanti-
ties, such as the mean velocity and Reynolds stresses, the ac-
curate experimental data on the temperature profiles are
scarce. Moreover, there are very few cases where measure-
ments are reported for higher Prandtl numbers. The few im- Figure 1. Comparison of the mean axial temperature
portant experimental investigations in this regard are those predictions of k - e models with the experi-
of Johnk and Hanratty Ž1962. for Pr s 0.72; Gowen and Smith mental data of Gowen and Smith (1967) for
Ž1967. for Pr s 5.7 and 14.3; Slarciauskas et al. Ž1973. for Pr Prandtl number s 5.7.

1804 September 2000 Vol. 46, No. 9 AIChE Journal

s95 and 170. of Kader Ž1981. have been used to test the
validity of the models.
The predictions of mean axial temperature profiles for Pr
s 5.7 are compared with the experimental data of Gowen
and Smith Ž1967. in Figure 1. It can be noted that the tem-
peratures were normalized with respect to the heat flow rate,
heat capacity, and friction velocity, so that the Reynolds
number independence is attained. It can be observed from
Figure 1 that only the LB, LSO, and CG models agree fairly
well with the experimental data. Most of the models were
found to underpredict the experimental data. The compari-
son was relatively better for the LSO model over the whole
range of the experimental data.
The predictions for Pr s95 and 170 have been validated
with the experimental data of Kader Ž1981.. The results are
shown in Figure 2. Figure 2a shows the comparison for JL,
LS, and LB models. It can be observed that none of these
models compare well with the experimental data. The JL
model underpredicts, while the LS and LB models overpre-
dict, the experimental data. Of the three, the LS model com-
pares fairly well with the experimental data. Figure 5 indi-
cates that the underpredictions of the JL model and the over-
predictions of the LS model may be due to the higher k pro-
files of the former and the lower of the latter. However, both
the models, in accordance with the experimental data, were
able to depict the location where the profiles change slope.
The profile of the LB model was quite unrealistic in the sense
that it neither gave the proper location of the change of slope
Žthe location was observed to be close to the wall., nor did it
give the proper wall value Žthe wall value was 300 for LB, as
compared to 80 for JL and LS.. This may be due to the im-
proper damping functions used, leading to the inaccurate
predictions of n T andror the turbulent kinetic energy dissi-
pation rate in the near-wall region. The predictions of the
dissipation variable Ž e . in the near wall region reflects on the
eddy viscosity profiles. Hence, any model that fails to cor-
rectly predict e in the near-wall region, will lead to inaccu-
rate n T , and consequently the temperature predictions. The
temperature predictions obtained from CH, NT, and SM are
given in Figure 2b. The CH model underpredicts the experi-
mental data, while NT and SM compare well with the experi-
mental data. The SM model slightly underpredicts the exper-
imental data. The comparison of the LSO, MK, and SZS
models is given in Figure 2c. The LSO model slightly over-
predicts the experimental data toward the central region;
otherwise, the comparison is good. Both the MK and SZS
models give values that are too low compared to the experi-
mental data. The comparison of the YS, FLB, and CG mod-
els is given in Figure 2d. It can be observed that CG is in
good agreement with the experimental data, while FLB and
YS underpredict the experimental profile. The reasons for
the observed discrepancies in various models may be the same
as explained for Figure 2a. Figure 2. Comparison of the mean axial temperature
The comparison for Pr s170 is given in Figure 3. In this predictions of k - e models with the experi-
case, the experimental data by Kader Ž1981. are available in mental data of Kader (1981) for Prandtl num-
the near-wall zone. The comparison for Pr s170 sets a cru- ber s 95.
cial test for the validation of the models for temperature, be-
cause of the very small thermal boundary-layer thickness. The
temperature profiles from various models have been com- model underpredicted and LS overpredicted the experimen-
pared in Figures 3a]3d. The trend observed for JL, LS, and tal data, while the LB model predictions were quite unrealis-
LB in Figure 3a was similar to Figure 2a. Here, too, the JL tic. In Figure 3b the SM and NT models were in better agree-

AIChE Journal September 2000 Vol. 46, No. 9 1805

 were found to be in good agreement with the experimental
data. In spite of these discrepancies, most of the models were
found to be in good agreement with the experimental data
for yqF 2.
From the comparison study of Figures 1, 2, and 3, it can be
observed that the LSO model is in overall good agreement
with the experimental data for either of the three Prandtl
numbers. The remaining models faired poorly for either of
the three Prandtl numbers. Since the LSO model compares
well, it was thought desirable to analyze the relative perfor-
mance of these models for the predictions of flow parameters
such as axial velocity, turbulence kinetic energy, and eddy
diffusivity. For the first two parameters, the experimental data
of Durst et al. Ž1995. was used. For eddy diffusivity, the ex-
perimental data of Schildknecht et al. Ž1979. was used. Since
the near-wall region is of importance for the prediction of
temperature profiles, attention has been focused on this re-
gion, particularly for comparison of the k and n T profiles.
The results are given in Figures 4, 5, and 6, respectively. From
Figure 4 it can be seen that the JL, CH, NT, SM, LSO, MK,
SZS, and YS models show very good agreement with the ex-
perimental data of mean axial velocity Ž0 - yq- 200.. Practi-
cally all the models Žexcept LS. are able to predict the k
profiles ŽFigure 5. in the yq range of 0 to 100. For eddy
diffusivity ŽFigure 6., fairly good predictions are made by the
JL, LS, LSO, and MK models.
It was also thought that it was desirable to establish overall
energy balance. The total Žviscous and turbulent . energy dis-
sipation rate can be calculated by the following equation:

D P =Vol.flow rate
eexpt s . Ž 25 .
Volume of the pipe

Normalizing the above equation Ž e by ut3rR and u by ut ., we

get

eq
expts fu
q3
. Ž 26.

The predicted energy dissipation rate consists of viscous and

turbulent energy dissipation rates:

eq
predicteds e turbulentq e mean,
q q
Ž 27.

where

2
­ uq
eq
means n
q
ž / Ž 28.
­ rq

­ uXi ­ uXi
eturbulent s n . Ž 29.
­ xj ­ xj
Figure 3. Comparison of the mean axial temperature
predictions of k - e models with the experi-
mental data of Kader (1981) for Prandtl num- The comparison between the experimental and the pre-
ber s170. dicted dissipation rates is presented in Table 6. It can be
seen that the LSO, CG, and FLB models predict the energy
dissipation rate within 4%. The difference is large for the
ment with the experimental data, while the CH model under- other models.
predicted. Once again, the results obtained were similar to From the foregoing discussion it is clear that the LSO
Figure 2b. From Figures 3c and 3d, the LSO and CG models model is the best if we consider the predictive ability for all

1806 September 2000 Vol. 46, No. 9 AIChE Journal

 Figure 5. Comparison of the turbulent kinetic energy
Figure 4. Comparison of the mean axial velocity predic- predictions of the k - e model with the experi-
tions of the k - e model with the experimental mental data of Durst et al. (1995) for Reynolds
data of Durst et al. (1995) for Reynolds num- number s7442.
ber s7442.
four parameters u, k, n T , and the energy balance. This abil-
ity of the LSO model probably explains its success in the pre-
diction of temperature profiles, as shown in Figures 1 to 3.

AIChE Journal September 2000 Vol. 46, No. 9 1807

 Table 6. Numerical Simulation and the Energy Balance
Calculations for Res 22,000
No. of No. of
Grid Points Iterations Energy
Required Convergence Required Input Total eq
Model ŽRange. Criteria ŽRange. ŽEq. 26. ŽEq. 27.
JL 75]85 10y4 ]10y5 100]200 35.61 29.90
LS 75]85 10y4 ]10y5 100]200 35.61 32.47
LB 75]85 10y4 ]10y5 100]200 35.61 33.14
CH 75]85 10y4 ]10y5 100]200 35.61 31.42
NT 75]85 10y4 ]10y5 150]250 35.61 33.26
SM 75]85 10y4 ]10y5 200]300 35.61 31.63
LSO 75]85 10y4 ]10y5 150]250 35.61 34.39
MK 75]85 10y4 ]10y5 150]250 35.61 33.25
SZS 75]85 10y4 ]10y5 100]200 35.61 30.84
YS 75]85 10y4 ]10y5 150]250 35.61 32.26
FLB 75]85 10y4 ]10y5 100]200 35.61 34.94
CG 75]85 10y4 ]10y5 200]250 35.61 34.94

Although the LSO model has been shown to be the best,

additional work is still needed for closure predictions. It is
known that, in addition to eddy diffusivity, there are other
factors that can contribute to the poor predictions. These are
the assumption of the constant turbulent Prandtl number and
the near-wall measurement limitations. In the present inves-
tigation, the turbulent Prandtl number was assigned a con-
stant value of 1.0. If a constant turbulent Prandtl number is
assumed, the entire thermal field depends only on eddy dif-
fusivity. In fact, some previous experimental evidence sug-
gests that the turbulent Prandtl number does not remain con-
stant; rather, it takes a value higher than unity very near the
wall. Second, the discrepancy observed between the numeri-
cla predictions and experimental data may be due to the un-
certainity of the data caused by the experimental difficulties
involved in measurements close to the wall, which are needed
at high Prandtl numbers.

Comparison of numerical predictions with the experimental

data for Reynolds-stress models
In order to validate the near-wall turbulent heat-flux mod-
els of Prud’homme and Elghobashi Ž1986. and Lai and So
Ž1990b., the data of Gowen and Smith Ž1967. and Kader
Ž1981., Bremhorst and Bullock Ž1970, 1973., and Hishida et
al. Ž1986. were used for checking the predictive capability of
models as regards to the mean axial temperature, the normal
and axial heat fluxes, and for the calculation of the Nusselt
number. The comparison for the normal and axial turbulent
heat fluxes is hindered due to the scarcity and the reliability
of the experimental data. In this respect, the preceding ex-
perimental measurements were found to be reasonably accu-
rate for the comparison. The calculations were carried out to
assess the correctness and validity of the near-wall heat-flux
closure as regards its ability to predict the mean temperature
profiles, the normal, and axial turbulent heat fluxes.
LSO compared their mean temperature profiles with the
experimental data of Johnk and Hanratty Ž1962. for very low
Figure 6. Comparison of the eddy viscosity predictions Prandtl number of Pr s 0.7. The predictions of mean tem-
of the k - e model with the experimental data perature profiles were obtained for two cases: in the pres-
of Schildknecht (1979) for Reynolds number ence as well as in the absence of the wall reflection term.
s 22,000. This was done to quantify the effect of this term on the tem-

1808 September 2000 Vol. 46, No. 9 AIChE Journal

 pressure-scrambling and diffusion terms. In Figure 9, the
predictions from LSO model were once again fairly good
compared to the PEL model. This analysis indicates that the
LSO model is in the best overall agreement with the experi-
mental data for the Prandtl numbers under consideration.
The comparisons of the calculated and measured heat
fluxes are shown in Figures 10 and 11. Comparison of the
predicted turbulent axial heat flux with the experimental data
of Bremhorst and Bullock Ž1970, 1973. is given in Figure 10,
while comparison of predicted turbulent radial heat flux with
the experimental data of Hishida et al. Ž1986. is given in Fig-
ure 11. For streamwise turbulent heat flux ŽuX T X., it was ob-
served that both models overpredict the experimental data.
The discrepancy was large near the wall, since the difference
in the peak values of predicted and experimental data was
observed to be higher at that point. According to LSO Ž1990b.,
the discrepancy between the model predictions and experi-
ments could be attributed to measurement errors generated
Figure 7. Comparison of the mean axial temperature by flow distortion due to buoyancy, and also partly to the
predictions of Reynolds stress models with simple one-time scale model of pressure scrambling. Shih and
the experimental data of Gowen and Smith Lumley Ž1986. argued that the pressure-scrambling process is
(1967) for Prandtl number s 5.7. influenced by more than one time scale and proceeded to
propose modifications to the pressure-scrambling model. The
preceding reasons justify the discrepancy observed between
perature profiles. In the present case, however, the predic- the predictions and the experimental data for the estimation
tions were obtained with the consideration of the wall reflec- of streamwise turbulent heat flux, and implies the need for
tion terms. Figure 7 shows the comparison for Pr s 5.7, while more sophisticated models so that the present approach can
Figures 8 and 9 give the comparison for Pr s95 and Pr s170, again be used to derive an asymptotically correct near-wall
respectively. It can be seen from these figures, that none of turbulent heat-flux model.
the models compare over the whole range of experimental The radial turbulent heat-flux comparison is given in Fig-
data. For Pr s 5.7 and 95, it was observed that, the predic- ure 11. The accurate estimation of this variable is crucial since
tions from the LSO model agree well with the experimental it is the controlling factor for the mean temperature and
data up to yq$8; however, a steep increase up to yqs15 streamwise turbulent heat-flux calculations. It can be seen
was observed. The predictions from the PEL model were ap- that the LSO model compares well with the experimental data
preciably lower than the experimental data in the near-wall in the near wall region up to yq$12. However, for yq%15,
region Ž yq$10.. It can be seen from Figure 8, that for yq% predictions start deviating, and there is a sudden flatness in
100, the predictions from both models gave an irregular pat- the curve in the region 12 $ yq$ center of the column. In-
tern. The irregularity in LSO arose at a lower yq as com- deed, the flatness is observed in the experimental curve, too,
pared to the PEL model. It is believed that the discrepancy is but it occurs at larger values of the normalized wall distance,
due to the incorrect high Reynolds-number modeling of the yq% 25. The predictions from the PEL model disagreed with

Figure 8. Comparison of the mean axial temperature Figure 9. Comparison of the mean axial temperature
predictions of Reynolds stress models with predictions of Reynolds stress models with
the experimental data of Kader (1981) for the experimental data of Kader (1981) for
Prandtl number s 95. Prandtl number s170.

AIChE Journal September 2000 Vol. 46, No. 9 1809

Figure 10. Comparison of the turbulent axial heat-flux
predictions with the experimental data of
Bremhorst and Bullock (1970, 1973). Figure 13. Comparison of the turbulent kinetic energy
predictions of the Reynolds stress models
with the experimental data of Durst et al.
(1995) for Reynolds number s7442.

the experimental data both near the wall and toward the cen-
ter. Near the wall, the predictions were higher, while a sub-
stantial decrease was observed toward the center. The accu-
rate prediction of ®X T X near the wall region from LSO is re-
flected in the mean temperature predictions as seen from
Figures 7, 8, and 9. It can be seen that the good prediction of
this quantity Žobtained by LSO. compares well with the mean
axial temperature.
Similar to the k-e model, the success of the LSO model
under the RSM category can also be explained on the basis
of its ability to predict the flow parameters such as mean
axial velocity Ž u., turbulent kinetic energy Ž k ., and the radial
Figure 11. Comparison of the turbulent radial heat flux rms fluctuating velocity. The results are shown in Figures 12,
predictions with the experimental data of 13, and 14, respectively. It can be seen that the predictions of
Hishida et al. (1986). the LSO model are very close to the experimental data.

Figure 12. Comparison of the mean axial velocity pre- Figure 14. Comparison of the rms radial velocity com-
dictions of the Reynolds stress models with ponent predicted by Reynolds stress models
the experimental data of Durst et al. (1995) with the experimental data of Durst et al.
for Reynolds number s7442. (1995) for Reynolds number s7442.

1810 September 2000 Vol. 46, No. 9 AIChE Journal

 Conclusions
The heat-transfer analysis presented in this study shows
that the LSO model of the k-e and Reynolds-stress groups
were in overall good agreement with experimental data for
the three Prandtl numbers. The disagreement for the axial
heat flux ŽuX T X. predictions was due to one time scale model
of pressure scrambling. The comparative analysis between the
k-e and RSM models for the mean temperature and the Nus-
selt number predictions favor the applicability of the k y e
model for heat transfer. Even though the RSM model over-
comes the assumptions of isotropy and the constancy of tur-
bulent Prandtl number, its applicability to heat transfer is
suspected. This is mainly due to the speculation that the
heat-flux transport is more complicated than momentum
transport, and is more likely to be influenced by two or more
Figure 15. Comparison of the mean axial temperature time scales than by one. The second reason may be due to
predictions of the k - e and the Reynolds the inappropriate near-wall modeling of the dissipation and
stress models with the experimental data of the pressure-scrambling terms. The predictive ability of the
Kader (1981) for Prandtl number s170. k-e models is expected to improve when the Prt variation
near the wall is included. Further, the overall discrepancy
observed in both the k-e and the RSM models for heat trans-
fer can be attributed to the incorrect near-wall modeling of
Comparison of the k- e and Reynolds-stress models for heat the dissipation term and the lack of detailed near-wall tem-
transfer perature and scalar flux measurements at higher Prandtl
The mean temperature profiles obtained from the k-e and numbers.
the Reynolds-stress models are compared in Figure 15. For
this study, the NT model from the k y e group and the LSO
model from the RSM group were selected. It can be seen Notation
that the overall comparison is better for the k-e model. The
CP sspecific heat at constant pressure, kJ ? kgy1 ? 8Cy1
second comparison was sought for the predictive capability of C1 sconstant in the k-e and Reynolds-stress models
the k-e and RSM models for the heat-transfer coefficients. C1 u , C2 u sconstants used in turbulent heat-flux model
The predicted Nu vs. Pr profile was compared with the ex- C1 u , w , C su sconstants used in turbulent heat-flux model
perimental data of Mizushina et al. Ž1971.. The comparison is Dsterm contained in the k equation
given in Figure 16. It can be seen from Figure 16 that the NT Ž dprdz .c sconstant axial pressure gradient
Esterm contained in the e equation
model compares well with the experimental data up to Pr F
f sfriction factor
100. At higher Prandt numbers, however, the model slightly fm , f 1 , f 2 sdamping functions used in low Reynolds number k-e
underpredicted the experimental data. On the other hand, and Reynolds-stress models
comparison of the LSO ŽReynolds-stress model. model was fe , f w, 1 sdamping functions used in low Reynolds number
poor for the whole range of Prandtl numbers. Reynolds-stress models
f w, u sdamping function used in low Reynolds number turbu-
lent heat-flux model
k sturbulent kinetic energy, m2 ? sy2
kq snormalized turbulent kinetic energy Ž krut2 .
P spressure, N ? my2
X
p sfluctuating pressure, N ? my2
qsconstant wall heat flux, W ? my2
qeff seffective radial heat flux, w s Ž k q r C p e H . ­ Tr­ y x, W ?
my2
Rsradius of pipe, m
R D snormalized radius of pipe, Ž2 Ru 0rn .
R T sturbulent Reynolds number Ž k 2rne .
R y sturbulent Reynolds number based on y Ž k 1r2 yrn .
ResReynolds number based on mean velocity Ž2 Rurn .
ReU sReynolds number based on friction velocity Ž Rutrn .
ScsSchmidt number
St sStanton number
T smean axial temperature, 8C
Tw smean axial temperature at the wall, 8C
T U stemperature based on friction velocity Ž s qrC p r ut ., 8C
Tq snormalized mean axial temperature ŽTrT U .
usmean axial fluid velocity, m ? sy1
Figure 16. Comparison of the Nusselt number predic- u b sbulk mean axial velocity of fluid, m ? sy1
tions of the k - e and the Reynolds stress ut sfriction velocity w s y R Ž dprdz . crŽ 2 r . x
'
models with the experimental data of X
u sfluctuating axial fluid velocity, m ? sy1
Mizushina et al. (1971). uq snormalized mean axial fluid velocity Ž urut .

AIChE Journal September 2000 Vol. 46, No. 9 1811

 X X X X
u T snormalized turbulent axial heat flux Ž su T rut T U .
q
Johnk, R. E., and T. J. Hanratty, ‘‘Temperature Profiles for Turbu-
X Xq X X U
® T sturbulent radial heat flux Ž s®rT rut T . lent Flow of Air in a Pipe,’’ Chem. Eng. Sci., 17, 881 Ž1962..
®smean radial fluid velocity, m ? sy1 Jones, W. P., and B. E. Launder, ‘‘The Prediction of Laminarization
X
® sfluctuating radial velocity component, m ? sy1 with a Two Equation Model of Turbulence,’’ Int. J. Heat Mass
w smean tangential fluid velocity, m ? sy1 Transfer, 15, 301 Ž1972..
X
w sfluctuating tangential velocity component, m ? sy1 Kader, B. A., ‘‘Temperature and Concentration Profiles in Fully
y snormal distance from wall Ž Ry r . Turbulent Boundary Layers,’’ Int. J. Heat Mass Transfer, 24, 154
yq sdimensionless wall distance Ž yutrn . Ž1981..
z saxial coordinate Lai, Y. G., and R. M. C. So, ‘‘On Near Wall Turbulent Flow Model-
ing,’’ J. Fluid Mech., 221, 641 Ž1990a..
Greek letters Lai, Y. G., and R. M. C. So, ‘‘Near Wall Modeling of Turbulent
Heat Fluxes,’’ Int. J. Heat Mass Transfer, 33, 1429 Ž1990b..
a , a U , b , g sconstants in low Reynolds number Reynolds-stress Lam, C. K. G., and K. Bremhorst, ‘‘A Modified Form of the k y e
model Model for Predicting Wall Turbulence,’’ J. Fluids Eng., 103, 456
k sthermal conductivity, W ? my1 ? 8Cy1 Ž1981..
m smolecular viscosity of fluid, kg ? my1 ? sy1 Launder, B. E., and B. I. Sharma, ‘‘Application of Energy Dissipa-
m t sturbulent viscosity of fluid, kg ? my1 ? sy1 tion Model of Turbulence to the Calculation of Flow Near a Spin-
r sdensity of fluid, kg ? my3 ning Disc,’’ Lett. Heat Mass Transfer, 1, 131 Ž1974..
n smolecular kinematic viscosity of liquid, m2 ? sy1 Miszushina, T., F. Ogino, Y. Oka, and H. Fukuda, ‘‘Turbulence Heat
n T sturbulent kinematic viscosity or eddy diffusivity, m2 ? sy1 and Mass Transfer Between Wall and Fluid Streams of Large
s ssurface tension, N ? my1 Prandtl and Schmidt Numbers,’’ Int. J. Heat Mass Transfer, 14, 1705
s k sturbulent Prandtl number for k Ž1971..
se sturbulent Prandtl number for e Myong, H. K., and N. Kasagi, ‘‘A New Approach to the Improve-
e sturbulent kientic energy dissipation rate, m2 ? sy3 ment of k y e Turbulence Model for Wall Bounded Shear Flows,’’
eq snormalized turbulent kinetic energy dissipation rate JSME Int. J., 33, 63 Ž1990..
Ž e Rrut3 . Nagano, Y., and M. Tagava, ‘‘A Improved k y e Model for Bound-
e˜ spseudodissipation variable Ž e y D . ary Layer Flows,’’ J. Fluids Eng., 112, 33 Ž1988..
e H seddy diffusivity for heat, m2 ? sy1 Patankar, S. V., Numerical Heat Transfer and Fluid Flow, McGraw-
D P spressure drop, N ? my2 Hill, New York Ž1980..
Patel, V. C., W. Rodi, and G. Scheuerer, ‘‘Turbulence Models for
Near Wall and Low Reynolds Number Flows: A Review,’’ AIAA J.,
Literature Cited 23, 1308 Ž1985..
Bremhorst, K., and K. J. Bullock, ‘‘Spectral Measurements of Tem- Prud’homme, M., and S. Elghobashi, ‘‘Turbulent Heat Transfer Near
perature and Longitudinal Velocity Fluctuations in Fully Devel- the Reattachment of Flow Downstream of a Sudden Pipe Expan-
oped Pipe Flow,’’ Int. J. Heat Mass Transfer, 13, 1313 Ž1970.. sion,’’ Numer. Heat Transfer, 10, 349 Ž1986..
Bremhorst, K., and K. J. Bullock, ‘‘Spectral Measurements of Turbu- Schildknecht, M., J. A. Miller, and G. E. A. Meier, ‘‘The Influence
lent Heat Transfer in Fully Developed Pipe Flow,’’ Int. J. Heat of Suction on the Structure of Turbulence in Fully Turbulent Pipe
Mass Transfer, 16, 2141 Ž1973.. Flow,’’ J. Fluid Mech., 90, 67 Ž1979..
Chein, K. V., ‘‘Prediction of Channel and Boundary Layer Flows with Shih, T. H., and J. L. Lumley, ‘‘Influence of Time Scale Ratio on
a Low Reynolds Number Turbulence Model,’’ AIAA J., 20, 33 Scalar Heat Flux Relaxation; Modelling Sirivat and Warhaft’s Ho-
Ž1982.. mogeneous Passive Scalar Fluctuations,’’ J. Fluid Mech., 162, 211,
Cho, H. H., and R. J. Goldstein, ‘‘An Improved Low Reynolds Num- Ž1986..
ber k ] e Turbulence Model for Recirculating Flows,’’ Int. J. Heat Shih, T. H., and N. N. Mansour, ‘‘Modelling of Near Wall Turbu-
Mass Transfer, 37, 1495 Ž1994.. lence,’’ Engineering Turbulence Modeling and Experiments, Elsevier,
Durst, F., J. Jovonavic, and J. Sender, ‘‘LDA Measurements in the Amsterdam Ž1990..
Near Wall Region of Turbulent Pipe Flow,’’ J. Fluid Mech., 295, Slarciauskas, A. A., A. A. Pedisius, and A. A. Zukauskas, ‘‘Universal
305 Ž1995.. Profiles of the Temperature and Turbulent Prandtl Number in the
Fan, S., B. Lakshminarayan, and M. Barnett, ‘‘Low Reynolds Num- Boundary Layer on a Plate in a Stream of Fluid,’’ Tr. Akad. Nauk
ber k y e Model for Unsteady Turbulent Boundary Flows,’’ AIAA Lit. SSR, Ser. B, 2 Ž1971. ŽEng. Transl.: Heat Transfer So®. Res., 5,
J., 31, 1777 Ž1993.. 623 Ž1973...
Gowen, R. A., and J. W. Smith, ‘‘The Effect of the Prandtl Number So, R. M. C., H. S. Zhang, and C. G. Speziale, ‘‘Near Wall Mod-
on Temperature Profiles for Heat Transfer in Turbulent Pipe elling of the Dissipation Rate Equation,’’ AIAA J., 29, 2069 Ž1991..
Flow,’’ Chem. Eng. Sci., 22, 1701 Ž1967.. Yakhot, V., S. A. Orszag, and A. Yakhot, ‘‘Heat Transfer in Turbu-
Hishida, M., Y. Nagano, and M. Tagawa, ‘‘Transport Processes of lent Fluids]1. Pipe Flow,’’ Int. J. Heat Mass Transfer, 30, 15 Ž1987..
Heat and Momentum in the Wall Region of Turbulent Pipe Flow,’’ Yang, Z., and T. H. Shih, ‘‘New Time Scale Based k y e Model for
Proc. Int. Heat Transfer Conf., Vol. 3, C. L. Tien et al., eds., Hemi- Near Wall Turbulence,’’ AIAA J., 31, 1191 Ž1993..
sphere, Washington, DC, p. 925 Ž1986..
Hrenya, C. M., E. J. Boilo, D. Chakrabarti, and J. L. Sinclair, ‘‘Com-
parison of Low Reynolds Number k y e Turbulence Model in Pre-
dicting Fully Developed Pipe Flow,’’ Chem. Eng. Sci., 12, 1923
Ž1995.. Manuscript recei®ed May 24, 1999, and re®ision recei®ed Mar. 14, 2000.

1812 September 2000 Vol. 46, No. 9 AIChE Journal