Вы находитесь на странице: 1из 9

Ind. Eng. Chem. Res.

2002, 41, 6393-6401 6393

A Reinterpretation of the Turbulent Prandtl Number†


Stuart W. Churchill‡
Department of Chemical Engineering, University of Pennsylvania, 311A Towne Building, 220 South 33rd
Street, Philadelphia, Pennsylvania 19104

The eddy viscosity at any location in fully developed turbulent flow in a round pipe is shown by
an exact analysis to be simply the ratio of the shear stress due to the time-averaged turbulent
fluctuations in the velocity to that due to the molecular motion and thus independent of its
heuristic diffusional origin. The eddy conductivity in fully developed turbulent convection is
similarly shown to be simply the ratio of the corresponding contributions to the radial heat flux
density. However, while providing a quantitative physical rationale for the eddy viscosity and
the eddy conductivity, the analysis that leads to this insight at the same time eliminates their
conceptual value except in a historical sense. On the other hand, the turbulent Prandtl number,
as redefined directly in terms of the time-averaged fluctuations, remains an essential parametric
variable. The local fraction of the radial heat flux density due to turbulence (the replacement
for the eddy conductivity as a variable) appears, on the basis of both experimental evidence and
asymptotic analyses, to be independent of geometry and the thermal boundary condition(s) and
thereby a universal function only of the molecular Prandtl number and the local fraction of the
shear stress due to the turbulence. It follows that the turbulent Prandtl number shares this
restricted functionality. Although the functional and numerical behavior of the turbulent Prandtl
number remains to this day incompletely defined either experimentally or theoretically, the
rate of heat transfer, as represented by the Nusselt number, is fortuitously insensitive to these
uncertainties.

Introduction have a lesser variance and could be more readily


The time-averaged once-integrated differential rep- predicted or more simply correlated than -Fu′v′ itself.
resentation for the conservation of momentum in the The corresponding expressions for the transport of
radial direction for fully developed turbulent flow of a energy in the negative radial direction, with the ad-
fluid with invariant physical properties in a round tube ditional postulate of negligible viscous dissipation, are
may be expressed as
∂T
j ) -k + FcT′v′ (3)
du ∂y
τ)µ - Fu′v′ (1)
dy
and

( )
The new terms such as -Fu′v′ that arise from time
averaging are sometimes called the Reynolds stresses ∂T kt ∂T
j ) -(k + kt) or -k 1 + or
because Reynolds1 was the first to space average the ∂y k ∂y
partial differential equations of conservation in order ∂T
to obtain a more tractable representation. In one- -kT (4)
∂y
dimensional flows, the contribution of the turbulent
fluctuations to the local shear stress (the rate of Equation 4 has generally been reexpressed as
momentum transfer in the negative radial direction),
namely, Fu′v′, has generally been modeled in terms of
an eddy viscosity µt, or a total viscosity µT, thereby [ ( )( ) ]
j ) -k 1 +
cµ kt µt ∂T
k cµt µ ∂y
or

[ ( )( ) ]
converting eq 1 to
Pr µt ∂T

( )
-k 1 + (5)
du µt du du Prt µ ∂y
τ ) (µ + µt) or µ 1+ or µT
dy µ dy dy
(2) or as

This model was proposed more than a century ago by


Boussinesq2 with the hope and expectation that µt would
j ) -k ( )( )( )
kT cµ µT ∂T
cµT k µ ∂y
or -k ( )( )
Pr µT ∂T
PrT µ ∂y
(6)


This paper is dedicated to William R. Schowalter not only with the hope and expectation that the eddy conductivity
in recognition of his technical and educational contributions kt or the total conductivity kT will be better behaved than
but also in appreciation of his leadership in the National
Academy of Engineering and the American Institute of Chemi- FcT′v′ and, in turn, that the turbulent Prandtl number,
cal Engineers. His presence in those councils was reassuring cµt/kt, and/or the total Prandtl number, cµT/kT, will be
at critical junctures. even better behaved in terms of variance, prediction,

Tel.: 215-898-5579. Fax: 215-573-2093. E-mail: Churchil@ and correlation. The expectations and hopes relative to
seas.upenn.edu. the behavior of µt, µT, kt, kT, Prt, and PrT are generally
10.1021/ie011021k CCC: $22.00 © 2002 American Chemical Society
Published on Web 07/30/2002
6394 Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002

accepted as being fulfilled, as attested by their continued bulent flow. Equation 1 then takes the form
usage to this day, although some purists have cast scorn
on this entire heuristic diffusional concept because of
its lack of a mechanistic rationale.
Until recently, the modeling of fully developed tur-
(1-
y+
a )
+
du+
[1 - (u′v′)++] ) +
dy
(8)

bulent flow has generally been based on eq 2 with an


empirical correlating equation for µt. The experimental If τ were a dependent variable, the expression (u′v′)++
values of µt upon which such a correlating equation is ) -Fu′v′/τ might appear to be the equivalent of a
based are subject to considerable uncertainty because similarity transformation. However, for fully developed
of their determination from (-dp/dx)(r/2)/(du/dy) - µ and flow in a round tube, τ bears a known linear relationship
in particular the determination of du/dy from measured to the independent variable τw. The term -pu′v′/τ is
values of u{y}. The modeling of fully developed turbu- simply a convenient representation for -Fu′v′/τw(1 -
lent convection has correspondingly been based on eq 5 y/a).
with an additional correlating equation or constant Equation 8 may be integrated formally in terms of R
value for Prt and often the additional postulate of a ) 1 - y+/a+ to obtain the following expression for the
constant value or a linear variation of the heat flux velocity distribution:
density with radius. The experimental values of kt upon
which a correlating equation for Prt is based are subject
a+
to even greater uncertainty than those of µt because of
2
u+ )
2
∫R1 [1 - (u′v′)++] dR2
2 (9)
their determination from j/(-∂T/∂y) - k ) [(Fc/r)∫r0 u
(∂T/∂x) dr ]/(-∂T/∂y) - k and in particular the determi-
2
This integral expression may, in turn, be integrated over
nation of ∂T/∂x from T{x,y}. The turbulent Prandtl the cross section and then reduced by means of integra-
number, as determined from the ratio of experimental tion by parts to obtain the following single integral for
values of µt and kt, is obviously subject to even greater the mixed-mean velocity and the Fanning friction factor,
uncertainty, particularly for large values of Pr, for which namely, f ≡ 2τw/Fum2:
the temperature profile develops very near the wall, and
for small values of Pr, for which kt is very small relative +
to k.
Theoretical results, experimental results, and cor-
2 1/2
()
f
) um+ ≡ ∫01u+ dR2 ) a4 ∫01[1 - (u′v′)++] dR4
relating equations for Prt were recently reviewed ob- (10)
jectively and in considerable detail by Kays,3 and to
avoid duplication, that review will be accepted with only The “unity” term of the integrand of eqs 9 and 10 can,
one minor reservation as a point of reference with of course, be integrated analytically, but that does not
respect to the state of the art. Furthermore, the primary constitute a significant simplification.
objective of the analysis presented here is to reinterpret These integral expressions are only marginally sim-
the results reviewed by Kays rather than to reevaluate pler than their counterparts in terms of µt, but the
them. reduction to a single integral for um+ does not follow so
A New Representation for Turbulent Flow and directly or obviously in terms of µt. More importantly,
Convection. Prompted by the appearance of essentially eliminating the derivative between eqs 1 and 2 reveals
exact computed values of the turbulent shear stress by that
direct numerical simulation (DNS), Churchill and Chan4
readdressed the question of the merits in terms of µt -Fu′v′ (u′v′)++
modeling, prediction, and correlation of the introduction ) ) (11)
µ τ + Fu′v′ 1 - (u′v′)++
of the eddy viscosity in eq 1 and were surprised to
discover, on the favorable side, that this quantity could This is a very significant result because of the revelation
be expressed directly in terms of the time average of that µt/µ is simply the ratio of the contributions of the
the fluctuating velocities without recourse to a diffusive turbulent fluctuations and the molecular motions to the
mechanism involving a gradient of the time-averaged transport of momentum and thereby completely inde-
velocity, while, on the unfavorable side, that this pendent of its heuristic diffusional origin involving the
quantity served no unique useful function for flow in a time-mean velocity gradient. As a consequence of eqs
round tube or between parallel plates and was singular 9-11, there does not appear to be any future role for
at one radial position and negative over an adjacent the eddy viscosity except in a historical sense. It should
finite range of the radius in circular concentric annuli. not be inferred from eq 11 that µt is proportional to µ
Accordingly, they suggested abandoning the eddy vis-
cosity concept and simply expressing eq 1 in “wall” because (u′v′)++ will, in general, be an implicit function
units, namely, as of µ. The analogue of (u′v′)++ for the transport of
energy is (T′v′)++ ) FcT′v′/j, namely, the local fraction
y+ du+ of the radial heat flux density due to the turbulent
1- ) + (u′v′)+ (7) fluctuations. Introduction of this dimensionless quantity
a+ dy+ as well as the less familiar counterpart of u+, namely,
T+ ≡ k(τwF)1/2(Tw - T)/µjw, into eq 3 results in
where (u′v′)+ ) -Fu′v′/τw. They further devised a gen-
eralized correlating equation for (u′v′)+ as a function of j dT+
[1 - (T′v′)++] ) (12)
y+ and a+. Churchill5 subsequently concluded that jw dy
(u′v′)++ ≡ -Fu′v′/τ, namely, the local fraction of the
shear stress due to fluctuations in velocity, was an even Then from eqs 4 and 12 and the definition of T+, it
better dimensionless variable for the modeling of tur- follows that
Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002 6395

kt FcT′v′ (T′v′)++ (u′v′)++ and (T′v′)++. It is remarkable that the critics


) ) (13) who were scornful of the eddy viscosity, eddy conductiv-
k j - FcT′v′ 1 - (T′v′)++
ity, and turbulent Prandtl number concepts because of
their lack of a mechanistic rational did not appear to
and, in turn, from eqs 5 and 6 that recognize this simple justification. On the other hand,

[ ] ( )
the many analysts who applied these concepts without
Prt (u′v′)++ 1 - (T′v′)++ -u′v′ j - FcT′v′ apology may be given the benefit of the doubt.
) ) (14) Having said the above, the eddy viscosity and eddy
Pr (T′v′)++ 1 - (u′v′)++ cT′v′ τ + Fu′v′ conductivity, although shown retroactively to be valid
concepts, are inferior to the fractions of the shear stress
and
and the radial heat flux density due to turbulence as

[ ]
operative variables in three respects. First, the momen-
PrT 1 - (T′v′)++ τ j - FcT′v′ tum and energy balances in terms of the fluctuating
) ) (15)
Pr 1 - (u′v′)++ j τ + Fu′v′ quantities are more transparent and understandable
because of the absence of heuristic quantities. Second,
the eddy viscosity is singular and negative in all
Eliminating (T′v′)++ between eqs 14 and 15 indicates
channels with unequal shear stresses on two bounding
that
surfaces, such as in a concentric circular annulus or in
++
a parallel-plate channel with surfaces of unequal rough-
1 (u′v′) 1 - (u′v′)++
) + (16) ness. The quantity (u′v′)++ is also inapplicable in this
PrT Prt Pr case, but (u′v′)+ is well behaved. Third, the formal
integral solutions for the velocity distribution and the
and provides for interchangeability between Prt and PrT. friction factor, as illustrated by eqs 9 and 10, and for
Equations 13-15 are very significant results in the the temperature distribution and Nusselt number, as
same sense as eq 11 in that kt/k, Prt/Pr, and PrT/Pr are illustrated subsequently, are much simpler in terms of
revealed to be simple functions of (u′v′)++ and (T′v′)++ the fractions of the shear stress and the radial heat flux
and thereby independent of their heuristic diffusive density than in terms of the eddy viscosity and eddy
origin. Thus, the eddy conductivity, as well as the eddy conductivity.
viscosity, no longer appears to have any essential role Implementation of the New Model for Convec-
except possibly in a historical sense. On the other hand, tion. Substituting for (T′v′)++ in eq 12 from eq 14
the turbulent Prandtl number and the total Prandtl results in

[ ( )]
number, as redefined by eqs 14 and 15, respectively,
have continued essential utility even if their diffusion- ++
j Pr (u′v′) dT+
ally based names are misleading. The total Prandtl ) 1+ (17)
number divided by the Prandtl number is seen from eq jw Prt 1 - (u′v′)++ dy+
15 to be equal to the fractional transport of energy by
molecular motion divided by its counterpart for the Although eq 17 incorporates two parametric variables,
transport of momentum, while the turbulent Prandtl namely, Prt/Pr and (u′v′)++ as compared to one in eq
number divided by the Prandtl number is seen from eq 12, namely, (T′v′)++, it is more tractable with respect
14 to be equal to the ratio of the radial transport of to numerical and functional evaluations because of the
momentum due to the turbulent fluctuations to that due
well-documented functionality of (u′v′)++ and the very
to the molecular motion divided by the corresponding
constrained variance of Prt/Pr except in regimes and
ratio for the radial transport of energy. Again, it should
regions in which it has a very limited influence.
not be inferred from eq 13 that kt is proportional to k or
For reasons that are not related to the turbulent
from eqs 14 and 15 that Prt and PrT are proportional to
Prandtl number, it is convenient, following Reichardt,6
Pr because the right-hand side of these expressions will
to introduce into eq 17 the perturbation-like quantity γ
generally be a function of Pr as well (u′v′)+.
defined by
An Alternative but Slightly Inferior Interpreta-

() ( )
tion. Rather than avoiding the eddy viscosity on the
j τw j y+
basis of eq 8 or eliminating it from existing models for 1+γ) ) / 1- + (18)
flow on the basis of eq 11, the latter expression could jw τ jw a
equally well be interpreted as a confirmation of the
intuition of Boussinesq in recognizing or postulating thereby obtaining

( )[ ( )]
that the transport of momentum by the turbulent ++
fluctuations is a diffusive process in terms of the radial y+ Pr (u′v′) dT+
(1 + γ) 1 - ) 1 + (19)
gradient of the time-averaged velocity even though he a+ Prt 1 - (u′v′)++ dy+
had no idea that the effective diffusivity might be
expressed exactly and completely in terms of a well- Equation 19 may be integrated formally in terms of R
defined physical property, namely, -Fu′v′/τ. The same to obtain the following analogue of eq 9:
argument applies to the eddy conductivity by virtue of
eqs 12 and 13. It then follows that the turbulent Prandtl a+ (1 + γ) dR2
∫R
( )
1
number is indeed analogous to the true Prandtl number T+ ) 2 (20)
2 ++
with the diffusional coefficients for the turbulent fluc- Pr (u′v′)
1+
tuations substituted for the molecular ones and thereby Prt 1 - (u′v′)++
properly named. Furthermore, eqs 9 and 10 as well as
all of the ensuing analysis herein could be expressed and in turn T+, weighted by u+/um+, can be integrated
with equal validity in terms of µt and kt rather than of over the cross section to obtain
6396 Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002

2a+
( )
+
a+ the wall where (u′v′)++/[1 - (u′v′)++] can be approxi-
) Tm+ ≡ ∫01T+ uu + dR2 ) ∫01 mated by 0.7 (y+/10)3 and γ can be set equal to zero.

[ ]] (
Nu 2
Then, insofar as Prt approaches a finite value Prt*as y+

)
m
(1 + γ) dR2 u+
∫R f 0, eq 19 can be integrated in closed form to obtain an

[
1
dR2 (21) algebraic expression for T+, from which it follows that
u m+
2
++
Pr (u′v′)
1+
Prt 1 - (u′v′)++

Equation 20 is more complex than its analogue, eq 9,


Nu f 0.076343 ( ) ()
Pr
Prt*
1/3
Re
f
2
1/2
(25)

by virtue of 1 + γ and Prt/Pr, and eq 21 relative to eq The existence of these three special solutions for Nu is
10 by virtue of u+/um+ as well, a mathematical reflection very fortuitous in view of the considerable uncertainty
of the greater physical complexity of convection as concerning Prt, both functionally and numerically in a
compared to flow. It may be noted that the temperature general sense, as documented by Kays.3 Surprisingly,
distribution, as given by eq 20, is not directly analogous a significant reduction in that uncertainty has not
to the velocity distribution as given by eq 9 for Pr ) 1, occurred in the intervening 7 years since his review,
as has often been postulated, nor even for Pr ) Prt, except possibly in the followup by Weigand et al.8 It
owing to the nonlinear heat flux density distribution as should be noted, as implied above, that these several
represented by γ. conclusions regarding the dependence of Nu on Pr and
Implementation of eq 21 to obtain numerical values Prt could also have been drawn if µt/µ were substituted
of Nu, as described by Bo et al.7 and others, is beyond
for (u′v′)++/[1 - (u′v′)++] in eqs 17, 19-21, and 23.
the scope and objective of this analysis, but the exami-
Equations 19-21 and 23 may be reexpressed in terms
nation of eq 21 itself is within that objective because it
of PrT rather than Prt as follows:
helps to identity the role of the function labeled Prt in

( ) ( )
the prediction of Nu. For the special case of uniform
y+ PrT dT+
heating or cooling at the wall of a round tube, (1 + γ) 1 - +
[1 - (u′v′)++] ) + (26)
Pr

( )
a dy
u+
1
∫0R
2

( )
1+γ) dR2 (22)
um+ a+ PrT
R2 T+ )
2
∫R1 (1 + γ)[1 - (u′v′)++]
2
Pr
dR2 (27)

[ ]
and eq 21 may be reduced by integration by parts to
obtain the following single integral: 2a+
Nu
) Tm+ )
a+
2
∫01 ∫R1 (1 + γ)[1 - (u′v′)]++
2 ( )
PrT
Pr
dR2

( )
8
Nu ) (23) +
(1 + γ)2 dR4 u
∫01 dR2 (28)

[ ]
+
++ um
Pr (u′v′)
1+
Prt 1 - (u′v′)++ and
The dependence of Nu on both Pr and Prt then stems
only from the term Pr/Prt in the denominator of the
integral. On the other hand, for the other common
Nu ) 8/ ∫01(1 + γ)[1 - (u′v′)++] ( ) PrT
Pr
dR4 (29)

thermal boundary conditions, namely, that of a uniform


wall temperature, Equations 26-29 appear to have a simpler structure
than eqs 19-21 and 23, and they do have some utility

( )
+ + for that reason in a qualitative sense. However, this
1+γ)
1
R2
∫0R uu + TT +
2
dR2 (24) impression of simplicity is somewhat misleading. For
example, Nu does not approach zero as Pr f ∞, as might
m m
be inferred from eq 29. Also, Prt is a more sensitive
Because of the presence of T+/Tm+ in the integrand of variable than PrT, and, perhaps for that reason, all
eq 24, γ now depends on Prt, Nu depends on Prt by correlating equations appear to have been constructed
virtue of γ as well as by its explicit presence, reduction for Prt rather than PrT.
of eq 21 to a single integral by integration by parts is Alternative Models and Variables for the Pre-
no longer possible, and an iterative scheme is required diction of Turbulent Flow and Convection. The
for the solution of eqs 20, 21, and 24 or their differential mixing length l of Prandtl,9 as defined by
counterparts.
(du
2

dy )
The dependence of Nu on Prt may be seen to vanish Fl 2 ) -Fu′v′ (30)
from eqs 21 and 23 for Pr ) 0, for which condition Nu
is thereby a function only of the rate of flow and the
thermal boundary condition. Prt also vanishes from has been widely used in the past as an alternative to
these two expressions for Pr ≡ Prt insofar as the this the eddy viscosity, and a thermal mixing length has
value of Prt is invariant with radius, a condition which, sometimes been used as an alternative to the eddy
remarkably enough, appears to occur to at least a good conductivity. Although the use of the mixing length has
degree of approximation. However, despite the conse- been fading in recent years, it was frequently cited in
quent great simplification of eqs 21 and 23 for this the past as being conceptually superior to the eddy
particular value of Pr, Nu remains a function of Pr in viscosity because of its mechanistic origin as an ana-
the region encompassing that value. For Pr f ∞, the logue, however strained, of the mean free path in the
temperature field develops almost completely very near kinetic theory of gases. In a round tube the mixing
Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002 6397

length is readily shown to be related to (u′v′)++ as measurements for all rates of flow, but the disappear-
follows: ance for b+ < 300 of the semilogarithmic regime of the
time-mean velocity distribution, which characterizes
the flow in the “turbulent core near the wall” (30 < y+
(u′v′)++
(l +)2 ) < 0.1b+), renders questionable the extrapolation of the

( )
(31)
y+ results of the DNS calculations for the turbulent core
1 - + [1 - (u′v′)++]2 to higher values of b+. The effective restriction of DNS
a
calculations to parallel-plate channels is less serious in
that the analogy of MacLeod19 implies that the values
Because (u′v′)++ is finite at the center line, l +, and
thereby l, may be seen from eq 31 to be unbounded at of (u′v′)++ and u+ are the same function of y+ and a+ in
that location, thereby negating their very basis. This a round tube as of y+ and b+ in a parallel-plate channel.
serious failing was apparently overlooked for over half On the other hand, because of their potential precision,
a century because of (1) insufficiently precise data for the execution of otherwise identical DNS calculations
for these two geometries for a wide range of values of
u′v′ and u in that region, (2) uncritical examination of b+ and a+ would provide a more severe test of the
such data, (3) a fortuitous insensitivity of calculations accuracy and limitations of that analogy than do the
for f and Nu to values of l in that region, and (4) the best current experimental measurements. Calculations
valid prediction, based on the experimentally observed using DNS for heat transfer have been carried out for
linearity of l to y in “the turbulent core near the a wide range of values of Pr, but the various calculated
wall,” of a semilogarithmic dependence of u on y in that
region. However, the semilogarithmic relationship has values of T′v′, as contrasted with those of u′v′, are not
since been derived by speculative and asymptotic di- in complete agreement with one another or with experi-
mensional analysis without recourse to a mechanistic mental results. In particular, the values of Sct (the
model such as the mixing length. Furthermore, the analogue for mass transfer of Prt for heat transfer), as
corresponding relationship for the velocity distribution predicted in the pioneering work of Papavassiliou and
in “the turbulent core near the center line”, which was Hanratty20 using a Lagrangian form of DNS for very
derived on the basis of a finite value of the mixing length large values of Sc (analogous to Pr), contradict the
at the center line has since been shown to be in serious postulate of a limiting value of Sct and thereby of Prt
error. (See, for example, work by Churchill.5) The mixing for y+ f 0 that leads to eq 23. Because the most reliable
length is also singular at the central plane between experimental data for heat transfer generally agree with
parallel plates and shares the failure of the eddy eq 24, including the 1/3 power dependence on Pr for large
viscosity in channels with differing shear stresses on values of Pr, this discrepancy may be due to (1) the
the walls. upper limit of Pr ) 100 for all fluids, whereas values of
The κ- model of Launder and Spalding10 was de- Sc are virtually unlimited in magnitude, (2) a funda-
signed to predict µt and thereby turbulent flow, but after mental difference between the turbulent transport of
20 years its implementation remains quite arbitrary and energy and species very near the wall, or (3) the
approximate overall and essentially correlative rather limitation of the pertinent DNS calculations to b+ ) 150.
Renormalization group theory (RGT) has been used
than predictive near the wall. The κ--u′v′ model was to predict PrT directly rather than by detailed solution
utilized successfully by Hanjalić and Launder11 to of the equations of conservation of momentum and
predict the behavior in an annulus, which an eddy
energy for u′v′ and T′v′ as with DNS. Results of great
viscosity model cannot do. The κ--u′v′-T′v′ model purported generality are achieved at the price of ide-
utilized by Suzuki et al.12 for an annulus might, in alizations of unknown consequence. Yahkot et al.21 and
principle, predict Prt. However, both of these latter others have devised the following structure for this
models are quite empirical overall and essentially purpose:

( ) ( )
correlative near the walls.
The large eddy simulation (LES) model pioneered by R/(1+2R)
1 (1+R)/(1+2R) 1
Schumann13 is predictive, in principle, for u′v′ and T′v′ -R +1+R
PrT PrT µ
and thereby Prt in the turbulent core but incorporates )
the eddy viscosity in the subgrid formulation for the 1 1 µ + µt
-R +1+R
region very near the wall. Kawamura et al.14 used the Pr Pr
LES model for convection as well as for flow, but they or 1 - (u′v′)++ (32)
postulated a value of Prt ) 0.85 for the subgrid region
rather than predicted it. where
One methodology for the prediction of u′v′ and T ′v′
and thereby the prediction or avoidance of Prt for the 2 1/2
entire cross section of a channel without any empiricism
is that of DNS, which was apparently first conceived
([
R) 1+81+ ( d )] )
- 1 /2 (33)

by Orszag and Patterson15 in the context of homoge- An iterative method of solution is obviously required for
neous and isotropic turbulence. However, the great early PrT. Prt can, in turn, be calculated by means of eq 16.
promise of this concept has not yet been completely Yahkot et al. proposed a value of 21/3 ) 7 for d in eq
fulfilled in that numerical results for confined shear 33, but in a “Note Added in Proof” stated that 9/3 ) 3
flows are limited primarily to parallel-plate channels might be better. Elperin et al.22 and Itazu and Nagano23
and to rates of flow barely above the minimum value of both proposed 14/3. On the other hand, an arbitrary
145 for b+ for fully developed turbulence. The predicted value of d ) 24/3 ) 8 produces better agreement with
values of several investigators, including Kim et al.,16 experimental data and the subsequently mentioned
Lyons et al.,17 and Rutledge and Sleicher,18 for (u′v′)++ empirical correlating equations for Prt. Yahkot et al.
near the wall are in good agreement with experimental argue that eqs 32 and 33 are free of any “experimentally
6398 Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002

Table 1. Comparison of Predictions of Prt for a+ ) 5000 and y+ ) 500


Pr 0.0001 0.001 0.01 0.1 0.7 0.8633 1.0 10 100
Pr(u′v′)++/ 0.01857 0.1857 1.857 18.57 130.0 160.3 185.7 1857 18570)
[1 - (u′v′)++]a
eq 35 (N&S) 1201 63.80 3.366 1.003 0.9457 0.9458 0.9463 0.9525 0.9553
eq 34b (J&R) 115.9 12.35 2.000 0.9650 0.8664 0.8633 0.8615 0.8512 0.8501
eq 36 (Kays) 38.54 4.619 1.227 0.8877 0.8553 0.8544 0.8538 0.8504 0.8500
eqs 32 and 33 (Yahkot et al.)c 42.98 4.855 1.216 0.8957 0.8643 0.8633 0.8627 0.8593 0.8590
a (u′v′)++ is based on eq 5 of Bo et al.7 b With a coefficient of 0.0115. c Based on d ) 8.

adjusted parameters”, the quantity d apparently being provides a reasonable representation for all values of
exempted as a “physical dimension”. In any event, a Pr.
number of the idealizations made in the derivations of Hill27 (also see Notter and Sleicher28) concluded from
eqs 32 and 33, and in particular that of homogeneous theoretical considerations involving isotropic turbulence
turbulence, would appear to confine their applicability and a linear temperature gradient that Prt approaches
to the turbulent core (y+ > 30) and perhaps to “the inverse proportionality to the equivalent of Pr (u′v′)++/
turbulent core near the wall” (30 < y+ < ∼0.1a+).
[1 - (u′v′)++] as Pr f 0. Notter and Sleicher developed
the following correlating equation based on that result
Functional Dependence and Correlating and the apparently false premise that Prt f 1 as
Equations for Prt (u′v′)++/[1 - (u′v′)++] f ∞:
Abbrecht and Churchill24 determined experimentally
the eddy conductivity as well as the eddy viscosity in Prt ) {1 + φ}/{0.025Pr(u′v′)++/[1 - (u′v′)++] + φ}

{ }
the developing temperature field generated by a step 10
in wall temperature in the fully developed flow of air 1+ (35)
in a round tube. Their derived values of kt are somewhat 35 + (u′v′) /[1 - (u′v′)++]
++

less certain than those of µt because of the necessity of


evaluating the longitudinal as well as the radial tem- where φ ) 90Pr3/2[(u′v′)++/[1 - (u′v′)++]]1/4.
perature gradients, but the invariance on the mean of Kays found that most experimental and computed
kt and Prt with axial distance is unambiguous evidence values of Prt in the turbulent core (y+ > 30) could be
of their independence from the temperature field and represented satisfactorily with an empirical expression
thereby of the thermal boundary condition. The close that may be rewritten in terms of (u′v′)++ rather than
correspondence of their values of Prt with those mea- ut/µ as follows:

[ ]
sured by Page et al.25 in fully developed turbulent flow
++
between parallel plates at different uniform tempera- 0.7 1 - (u′v′)
tures provides not only a confirmation of the indepen- Prt ) 0.85 + (36)
Pr (u′v′)++
dence of Prt from the thermal boundary condition but
also an indication of independence from geometry for
However, he suggested that 2.0 was a better coefficient
a+ ) b+. Abbrecht and Churchill thereby concluded that
than 0.7 for liquid metals (Pr , 1).
Prt is a function only of µt/µ and Pr. In retrospect, this
It is informative to compare the predictions of eqs 32
conclusion may or may not be valid for the regime very
and 34-36 functionally and numerically. Equations 32
near the wall because of the increased uncertainty of
the values they determined for µt and kt as y+ f 0. This and 35 predict a separate dependence of Prt on (u′v′)++
conclusion, that is, the dependence of Prt on µt/µ [or and Pr, whereas eq 36 predicts a dependence only on
(u′v′)++] and Pr only, has since been dismissed as both the combined variable Pr(u′v′)++/[1 - (u′v′)++]. This
obvious and disputed and apparently has never been result is perhaps explicable in that Hill27 derived this
proven or disproven conclusively, although it is implied combined dependence as an asymptote for Pr f 0,
by eq 32 and most empirical correlating equations. whereas eq 36 stretches the applicability of that as-
Kays,3 with some justification, questioned the ap- ymptote for all values of Pr. The neglect of a dependence
plicability of this restricted dependence to the region on (u′v′)++/[1 - (u′v′)++] by Jischa and Rieke can
near the wall and near the center line on the grounds perhaps be interpreted as an approximation of this term
of insufficient evidence and, having overlooked the work by a mean value that results in only a slight error
of Abbrecht and Churchill, speculated on the basis of a insofar as the term 0.85 is dominant.
comparison of some separate experimental results for Equation 32 implies the existence of a value of Pr )
uniform heating and uniform wall temperature that Prt Prt ) PrT that is independent of (u′v′)++, thus justify-
may depend on the thermal boundary condition. The ing the reduction of eqs 19-21, 23, and 26-29 to
extreme difference in the thermal boundary conditions expressions free of Prt and Pr for this condition. On the
utilized in the work of Abbrecht and Churchill and in other hand, both eqs 35 and 36 predict a slightly varying
that utilized by Page et al. would appear to be a more dependence of Prt and PrT on (u′v′)++ for all values of
critical and sufficient test. Pr.
Turning to correlating equations for Prt , Jischa and Representative predictions of Prt by eqs 32 and 34-
Rieke26 concluded that the dependence of Prt on y+ and 36 are listed in Table 1. Values of d ) 8 in eq 33 and a
Re, while real, is of second-order and that the simple coefficient of 0.0115 (rather than 0.015) in eq 34 are
expression postulated for this comparison in the interest of con-
sistency. The value of a+ ) 5000 was chosen to represent
Prt ) 0.85 +
0.015
(34) a moderate rate of flow and the value of y+ ) 500 to
Pr represent the outer range of the semilogarthmic regime
Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002 6399

Table 2. Comparison of Predictions of PrT for a+ ) 5000 and y+ ) 500


Pr 0.0001 0.001 0.01 0.1 0.7 0.8633 1.0 10 100
eq 34a (J&R) 0.01867 0.1839 0.9681 0.9223 0.8653 0.8633 0.8621 0.8553 0.8547
eqs 32 and 33 (Yahkot et al.)b 0.01866 0.1798 0.7388 0.8591 0.8633 0.8633 0.8634 0.8635 0.8635
a With a coefficient at 0.0115. b Based on d ) 8.

Table 3. Comparative Dependence of Nu for a+ ) 5000 on Extreme Expressions for Prt and PrT
Pr 0.0001 0.001 0.01 0.10 0.70 1.0 10 100
Nu1/Nu2 1.000 1.010 1.120 0.9913 1.0076 1.006 0.9844 0.9743
(Prt)1/(Prt)2 7.962 4.025 1.432 1.003 1.085 1.094 1.119 1.124
(PrT)2/(PrT)1 1.0001 1.009 1.154 1.003 1.085 1.094 1.119 1.124
a Prt based on eq 35 (N&S). b Prt based on eq 34 (J&R).

of the velocity distribution. The agreement for large and The actual uncertainty in the value of Nu resulting from
moderate values of Pr is very good except for the the use of eq 32 or 36 for y+ > 30 and eq 34 for y+ < 30
predictions of Notter and Sleicher, which are signifi- would be expected to be far less, probably less than 1
cantly higher. The predictions of Kays and Yahkot et or 2% for all values of Pr and a+. It may be concluded
al. are in fair agreement for small values of Pr, but those that improved numerical values, predictive equations,
of Jischa and Rieke and of Notter and Sleicher are much and/or correlative equations for Prt, while eagerly
higher. The appearance of an almost exact agreement awaited in an intrinsic sense, will not greatly impact
between the predictions of Yahkot et al. and Kays in existing calculated values of Nu such as those of Bo et
Figure 2 of the latter is to some extent an artifact of al.7 for round tubes, Danov et al.29 for parallel-plate
the compressed logarithmic coordinates. channels, and Bo et al.30 for concentric circular annuli.
The predictions of PrT by eq 34 of Jischa and Rieke
are compared in Table 2 with those of Yahkot et al. Summary and Conclusions
using d ) 8 in eq 33, for the same conditions as those
for Table 1. It is noteworthy that the agreement is very The eddy viscosity ratio µt/µ has been shown to be
close for Pr g 0.7 and for Pr e 0.001 despite the great equal to the ratio of the local fraction of the shear stress
structural difference of eqs 32 and 34. This is a rate of momentum transfer due to the turbulent fluc-
somewhat fortuitous consequence, on the one hand, of tuations, namely, (u′v′)++ ) -Fu′v′/τ, to that due to the
the dominance of the limiting value of PrT ) 0.85 for
molecular motions, namely, 1 - (u′v′)++, and thereby
even moderate values of Pr and, on the other hand, the
independent of its heuristic diffusive origin. The eddy
approach of PrT to Pr/[1 - (u′v′)++] for Pr f 0 per eq 16 conductivity ratio kt/k has similarly been shown to be
and thereby its independence from the expression used equal to the ratio of the local fraction of the radial heat
to predict Prt. These same two generalizations apply to flux density due to the turbulent fluctuations, namely,
the predictions (not shown) of PrT by eqs 35 and 36.
(T′v′)++ ) FcT′v′/j, to that due to the molecular mo-
Dependence of Nu on Prt tions, namely, 1 - (T′v′)++, and thereby also indepen-
dent of the heuristic diffusive origin of kt. Despite their
The principal applicability of an expression for the justification as physically meaningful quantities in
prediction or correlation of values of Prt is in the round tubes and parallel-plate channels, the eddy
prediction of values of Nu by means of eq 21 or 23 or viscosity and eddy conductivity are inferior in behavior
their equivalent. It may be inferred from the reduced and in implementation to (u′v′)++ or (u′v′)+ and T′v′)++
forms of eqs 21 and 23 for Pr ) 0 and Pr ) Prt and the even in those two geometries.
limiting form for Pr f ∞ that the power dependence of Turbulent flow and convection can be described com-
Nu on Pr/Prt increases from zero to a maximum value
approaching unity at a value of Pr/Prt ) 1 and then pletely and with clarity in terms of (u′v′)++ and (T′v′)++
decreases to a value of 1/3 as Pr f ∞. The less than linear while wholly avoiding the eddy viscosity, the eddy
power dependence of Nu on Pr/Prt for all values of Pr conductivity, and/or the corresponding mixing lengths.
thus fortuitously reduces the uncertainty in Nu arising This approach might be feasible with undergraduate
from the uncertainty in Prt. This reduction is illustrated students, who have nothing to unlearn, were it not for
in Table 3, in which the ratio of the values of Nu the pervasive presence of the latter quantities in the
calculated by Bo et al.7 for a+ ) 5000 and uniform literature of the past and present.
heating using eqs 34 and 35 for Prt is compared with On the other hand, the turbulent Prandtl number
the corresponding ratios of values of Prt and PrT. The ratio, Prt/Pr, which was originally defined as the ratio
difference in the predicted values of Nu is very small of µt/µ to kt/k but which now may be reexpressed in
except at Pr ) 0.01, for which it is 12% as compared to terms of (u′v′)++ and (T′v′)++, as in eq 16, and inter-
corresponding differences of 43% in Prt and 15% in PrT. preted as the ratio of the fractions of the local rate of
The 2.6% difference in Nu for Pr ) 100 is a consequence momentum transport due to turbulent fluctuations and
of the 12.5% overprediction of the limiting value of Prt molecular motions divided by the corresponding ratio
by eq 35, thereby a proof rather than an exception with for energy transport, has a continued essential role in
respect to the insensitivity of Nu to Prt. Almost identical the prediction of convection, although its symbol and
results are obtained for uniform wall temperature and name may be vestiges of the past.
for other values of a+. It should be emphasized that the Abbrecht and Churchill24 found experimentally that
comparisons in Table 3 were deliberately made using a kt/k is a function only of µt/µ and Pr and thereby
simplistic expression (eq 34) and a more complex but independent of the thermal boundary condition. The
questionable one (eq 35) for Prt in order to exaggerate same dependence and independence obviously extends
the dependence of Nu on the expression used for Prt. to the turbulent Prandtl number ratio Prt/Pr. The close
6400 Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002

agreement of their results for kt/k and Prt/Pr with those γ ) jτw/jwτ - 1
of Page et al.25 for convection between parallel plates at  ) rate of dissipation of turbulent energy
different uniform temperatures confirms this finding κ ) kinetic energy of turbulence
and implies that Prt/Pr is independent of geometry as µ ) dynamic viscosity
well when expressed in terms of µt/µ ) (u′v′)++/[1 - µt ) eddy viscosity
(u′v′)++] and Pr. µT ) µ + µt ) total viscosity
The only significant advance in the measurement and F ) specific density
prediction of Prt since the recent review of Kays3 τ ) shear stress
appears to be the Lagrangian DNS calculations of φ ) 90Pr3/2{(u′v′)++/[1 - (u′v′)++]}1/4
Papavassiliou and Hanratty20 for very large Schmidt
numbers. The latter results contradict the postulate of Subscripts
a finite value of Prt at the wall that is the basis for the m ) mixed-mean value
experimentally confirmed 1/3 power dependence of Nu w ) at wall
on Pr. The explanation of this discrepancy is not yet
resolved, but the practical consequences for heat trans- Acknowledgments
fer are unimportant because all normal fluids have
values of Pr < 100. The very discerning comments and constructive sug-
Predictions of Nu, which are the principal application gestions of the anonymous reviewers resulted in sig-
of predictions for Prt, are fortuitously very insensitive nificant improvement of the manuscript and are grate-
to Prt. While improved measurements, predictions, and fully acknowledged. The investigation that led to the
correlating equations for Prt are eagerly awaited, their unexpected findings reported herein might not have
impact on the currently predicted results for Nu can be been undertaken without the prompting of Professor
expected to be only of second-order. Tetsu Fujii following a seminar by the author at Kyushu
University.
Nomenclature
Literature Cited
a ) radius of the round tube
a+ ) a(τwF)1/2/µ (1) Reynolds, O. On the Dynamical Theory of Incompressible
b ) half-spacing of parallel plates Viscous Fluids and the Determination of the Criterion. Proc. R.
b+ ) b(τwF)1/2/µ Soc. London 1985, 186, 123.
c ) heat capacity at constant pressure (2) Boussinesq, J. Essai sur la theorie des eaux courantes. (A
d ) “arbitrary” coefficient in eq 33 Study of the Theory of Water Flows). Mem. Acad. Sci. Fr. 1877,
f ) 2τw/Fum2 ) Fanning friction factor 23, 1.
(3) Kays, W. M. Turbulent Prandtl NumbersWhere are We?
j ) heat flux density in the y direction
J. Heat Transfer, Trans. ASME 1994, 116, 284.
k ) thermal conductivity
(4) Churchill, S. W.; Chan, C. Turbulent Flow in Channels in
kt ) eddy conductivity Terms of the Turbulent Shear and Normal Stresses. AIChE J.
kT ) k + kt ) total conductivity 1995, 41, 2513.
l ) mixing length (5) Churchill, S. W. New Simplified Models and Formulations
l + ) l (τwF)1/2/µ for Turbulent Flow and Convection. AIChE J. 1997, 43, 1125.
Nu ) 2ajw/k(Tw - Tm) ) Nusselt number (6) Reichardt, H. Die Grundlagen des Turbulent Wärmeüber-
p ) pressure traganges. Arch. Gesamte Wärmetech. (The Principles of Turbulent
Pr ) cµ/k ) Prandtl number Heat Transfer). 1951, 2, 129.
Prt ) cµt/kt ) Pr(u′v′)++[1 - (T′v′)++]/(T′v′)++[1 - (u′v′)++] (7) Bo, Y.; Ozoe, H.; Churchill, S. W. The Characteristics of
Fully Developed Turbulent Convection in a Round Tube. Chem.
) turbulent Prandtl number Eng. Sci. 2001, 56, 1781.
PrT ) cµT/kT ) Pr[1 - (T′v′)++]/[1 - (u′v′)++] ) total (8) Weigand, B.; Ferguson, J. R.; Crawford, M. E. An Extended
Prandtl number Kays and Crawford Turbulent Prandtl Number Model. Int. J. Heat
R ) 1 - y+/a+ Mass Transfer 1997, 40, 4192.
Re ) 2a+um+/µ ) Reynolds number (9) Prandtl, L. Eine Beziehung zwischen Wärmeaustausch and
Sc ) Schmidt number Strömungswiderstand der Flüssigkeit. (An Analogy between Heat
Sct ) turbulent Schmidt number Transfer and Friction in Fluid Flow). Phys. Z. 1910, 11, 1072.
T ) time-averaged temperature (10) Launder, B. E.; Spalding, D. B. Mathematical Models of
T+ ) k(τwF)1/2/(Tw - T)/µjw Turbulence; Academic Press: York, London, 1972.
(11) Hanjalic, K.; Launder, B. E. A Reynolds Stress Model of
T′ ) fluctuating component of temperature
Turbulence and Its Application to Thin Shear Flows. J. Fluid
T′v′ ) time-averaged value of T′v′ Mech. 1972, 52, 609.
(T′v′)++ ) FcT′v′/j (12) Suzuki, K.; Tohkaku, A.; Sato, T. Liquid Metal Turbulent
u ) time-averaged x component of velocity Heat Transfer in Concentric Annuli. Proceedings of the 8th
u+ ) u(F/τw)1/2 International Heat Transfer Conference; Hemisphere Publ. Corp.:
Washington, DC, 1980; Vol. III, p 969.
u′ ) fluctuating x component of velocity
(13) Schumann, U. Subgrid Scale Model for Finite Difference
u′v′ ) time-averaged value of u′v′ Simulation of Turbulent Channel Flows in Plane Channels and
(u′v′)+ ) -Fu′v′/τw Annuli. J. Comput Phys. 1975, 18, 376.
(u′v′)++ ) -Fu′v′/τ (14) Kawamura, H.; Nakamura, S.; Satake, S.; Kunagi, T. Large
Eddy Simulation of Turbulent Heat Transfer in a Concentric
v′ ) fluctuating y component of velocity
Annulus. Thermal Science & Eng. 1994, 2, No. 2, 16.
x ) axial coordinate (15) Orszag, S. A.; Patterson, G. S. Numerical Simulation of
y ) distance from the wall Three-Dimensional Homogenous Isotropic Turbulence. Phys. Rev.
y+ ) y(τwF)1/2/µ Lett. 1972, 28, 75.
Greek Symbols (16) Kim, J.; Moin, P.; Moser, R. Turbulence Statistics in Fully
Developed Turbulent Channel Flow at Low Reynolds Number. J.
R ) coefficient defined by eq 33 Fluid Mech. 1987, 177, 133.
Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002 6401

(17) Lyons, S. L.; Hanratty, T. J.; McLaughlin, J. B. Large- (25) Page, F., Jr.; Schlinger, W. G.; Beaux, D. K.; Sage, B. W.
Scale Computer Simulation of Fully Developed Turbulent Channel Point Values of Eddy Conductivity and Viscosity in Uniform Flow
Flow with Heat Transfer. Int. J. Numer. Methods Fluids 1991, Between Parallel Plates. Ind. Eng. Chem. 1982, 44, 424.
13, 999. (26) Jischa, M.; Rieke, H. B. About the Prediction of Turbulent
(18) Rutledge, J.; Sleicher, C. A. Direct Simulation of Turbulent Prandtl and Schmidt Numbers from Modeled Transport Equation.
Flow and Heat Transfer in a Channel. Part I. Smooth Walls. Int. Int. J. Heat Mass Transfer 1979, 22, 1547.
J. Numer. Methods Fluids 1993, 15, 1051. (27) Hill, J. W. Estimation of Eddy Diffusivities in Isotropic
(19) MacLeod, A. C. Liquid Turbulence in a Gas-Liquid Turbulence. Preprints of the 6th Annual AIChE Meeting, San
Absorption System. Ph.D. Thesis, Carnegie Institute of Technol- Francisco, 1972.
ogy, Pittsburgh, PA, 1951. (28) Notter, R. D.; Sleicher, C. A. A Solution to the Turbulent
(20) Papavassiliou, D. V.; Hanratty, T. J. Transport of a Passive Graetz ProblemsIII. Fully Developed and Entry Region Heat
Scalar in a Turbulent Channel Flow. Int. J. Heat Mass Transfer Transfer Rates. Int. J. Heat Mass Transfer 1972, 27, 2073.
1977, 40, 1303. (29) Danov, S. N.; Arai, N.; Churchill, S. W. Exact Formulations
(21) Yahkot, V.; Orszag, S. A.; Yahkot, A. Heat Transfer in and Nearly Exact Numerical Solutions for Convection in Turbulent
Turbulent Fields. 1. Pipe Flow. Int. J. Heat Mass Transfer 1987, Flow Between Parallel Plates. Int. J. Heat Mass Transfer 2000,
30, 15. 43, 2767.
(22) Elperin, T.; Kleeorin, N.; Rogachevskii, I. Isotropic and (30) Bo, Y.; Ozoe, H.; Churchill, S. W. Nearly Exact Numerical
Anisotropic Spectra of Passive Scalar Fluctuations in Turbulent Solutions for Turbulent Convection in an Internally Heated
Fluid Flow. Phys. Rev. E 1996, 53, 3431. Circular Concentric Annulus. In review.
(23) Itazu, Y.; Nagano, Y. Modeling of Thermal Eddy Diffusivity Received for review December 17, 2001
Based on Renormalization Theory (In Japanese). Trans. Jpn. Soc. Revised manuscript received May 29, 2002
Mech. Eng. 1997, B36, 3072. Accepted June 6, 2002
(24) Abbrecht, P. H.; Churchill, S. W. The Thermal Entrance
Region in Fully Developed Turbulent Flow. AIChE J. 1960, 6, 268. IE011021K

Вам также может понравиться