Available online at www.sciencedirect.

com

International Journal of Heat and Fluid Flow 29 (2008) 8393

www.elsevier.com/locate/ijh

Parallel and non-parallel laminar mixed convection ow

in an inclined tube: The eect of the boundary conditions
A. Barletta *

Universita` di Bologna, Dipartimento di Ingegneria Energetica, Nucleare e del Controllo Ambientale (DIENCA), Laboratorio di Montecuccolino,
Via dei Colli 16, Bologna I-40136, Italy

Received 29 October 2006; received in revised form 29 April 2007; accepted 19 July 2007
Available online 27 September 2007

Abstract

The necessary condition for the onset of parallel ow in the fully developed region of an inclined duct is applied to the case of a cir-
cular tube. Parallel ow in inclined ducts is an uncommon regime, since in most cases buoyancy tends to produce the onset of secondary
ow. The present study shows how proper thermal boundary conditions may preserve parallel ow regime. Mixed convection ow is
studied for a special non-axisymmetric thermal boundary condition that, with a proper choice of a switch parameter, may be compatible
with parallel ow. More precisely, a circumferentially variable heat ux distribution is prescribed on the tube wall, expressed as a sinu-
soidal function of the azimuthal coordinate # with period 2p. A p/2 rotation in the position of the maximum heat ux, achieved by set-
ting the switch parameter, may allow or not the existence of parallel ow. Two cases are considered corresponding to parallel and non-
parallel ow. In the rst case, the governing balance equations allow a simple analytical solution. On the contrary, in the second case, the
local balance equations are solved numerically by employing a nite element method.
 2007 Elsevier Inc. All rights reserved.

Keywords: Mixed convection; Non-axisymmetric heat ux; Laminar ow; Parallel ow; Inclined duct

1. Introduction equations, thus giving a chance to nd exact solutions at

The subject of laminar mixed convection in vertical or
inclined ducts deserves a special interest for its applications
in the design of cooling systems for electronic devices or for
solar collectors. Several authors have presented theoretical
or experimental investigations most of which have been
reviewed, for instance, in Aung (1987). The literature of
the last decades includes many papers presenting theoreti-
cal investigations of buoyancy-induced ows in vertical
or inclined ducts. Some of these papers (Lavine, 1988;
Barletta and Zanchini, 1999, 2001; Chamkha et al., 2002;
Buhler, 2003; Weidman, 2006; Magyari, 2007) describe
analytical solutions with reference to fully developed paral-
lel ows. In fact, parallel ow represents the condition that
allows a drastic simplication of the governing balance
*
Tel.: +39 051 6441703; fax: +39 051 6441747.
E-mail address: antonio.barletta@mail.ing.unibo.it
least for the simpler geometries: parallel plane channel, cir-
cular or annular duct, rectangular duct. However, if one
deals with inclined ducts, the condition of parallel ow
can be considered as an exception rather than a rule. This
conclusion is a direct consequence of the observation that,
in an inclined duct, the buoyancy force vector has a non-
vanishing projection on the plane of the duct cross-section.
In the fully developed region, the transversal components
of the buoyancy force are normally responsible for the
onset of a secondary ow that makes velocity a non-paral-
lel helicoidal vector eld. It must be pointed out that, in
this reasoning, geometry matters. In fact, for a parallel
plane channel, parallel ows are still possible when the
channel is inclined, for the more commonly employed
thermal boundary conditions (Lavine, 1988; Barletta and
Zanchini, 2001). For dierent geometries of the duct
cross-section, as circular, annular or rectangular, this is
not true.

0142-727X/$ - see front matter  2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.ijheatuidow.2007.07.008
84 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393
Nomenclature
a dimensionless function, Eq. (10)
A function of X and Y, Eq. (9)
b dimensionless constant, Eq. (10)
B constant, Eq. (9)
g gravitational acceleration
g modulus of the gravitational acceleration
Gr Grashof number, Eq. (10)
k thermal conductivity
Nu Nusselt number, Eq. (38)
P dierence between the pressure and the hydro-
static pressure
Pr Prandtl number, Eq. (10)
qw(#) incoming wall heat ux, Eq. (2)
q0 constant wall heat ux, Eq. (2)
r dimensionless radial coordinate, Eq. (10)
R radial cylindrical coordinate
R0 duct radius
Re Reynolds number, Eq. (10)
T temperature
T0 average temperature in a duct cross-section, Eq.
(11)
u dimensionless axial velocity component, Eq.
(10)
u1(r), u2(r) dimensionless functions of r, Eq. (23)
Uz axial velocity component
U0 average velocity in a duct cross-section, Eq. (12)
U velocity
In a recent paper (Barletta, 2005), a criterion to establish
whether parallel ow in an inclined duct is possible or not
has been discussed. This criterion is in fact a necessary con-
dition for parallel ow and, as such, precludes the possibil-
ity of a parallel velocity eld when it is not fullled. In
Barletta (2005), it is shown that parallel ow is possible
only if the temperature gradient, the unit vector in the axial
direction and the gravitational acceleration are everywhere
coplanar vectors. The necessary condition is tested in the
case of a parallel plane channel, showing that the widely
studied boundary condition of isothermal walls with
unequal temperatures is compatible with parallel ow.
However, the compatibility holds only if the channel is
tilted in the direction orthogonal to the boundary planes.
If, on the other hand, the channel is tilted in the direction
parallel to the boundary planes, no parallel ow is possible
(Barletta, 2005).
The aim of the present paper is to extend the analysis
presented in Barletta (2005), in order to show that proper
thermal boundary conditions may allow the onset of paral-
lel ow also for duct geometries dierent from the parallel
plane channel. Reference is made to an inclined circular
tube. The prescribed boundary condition is a simple non-
axisymmetric thermal boundary condition, namely a wall
heat ux sinusoidally varying in the angular direction. In
u0 dimensionless 2D velocity vector, Eq. (10)
x, y dimensionless Cartesian coordinates, Eq. (10)
X, Y, Z Cartesian coordinates
Y unit vector in the Y-direction
Greek symbols
a thermal diusivity
b coecient of thermal expansion
ek dimensionless kinetic energy of secondary ow,
Eq. (42)
# angular cylindrical coordinate
H dimensionless temperature, Eq. (10)
K dimensionless parameter, Eq. (33)
l dynamic viscosity
m kinematic viscosity
n dimensionless switch parameter, Eq. (2)
. mass density
.0 mass density at temperature T0
u tilt angle of the duct axis
X dimensionless parameter, Eq. (33)
Superscripts, Subscripts
0
2D vector obtained by projection in the (X, Y)-
plane
max, min maximum, minimum value in a duct cross-
section
linear cases, this boundary condition yields what can be
considered a fundamental solution for the analysis of non-
axisymmetric ows by means of Fourier series (Barletta
et al., 2003). It is shown that the same thermal boundary
condition may yield parallel or non-parallel ow depending
on the value of a switch parameter that yields a p/2 rota-
tion in the wall heat ux distribution. In the case of parallel
ow, the governing balance equations admit a straightfor-
ward analytical solution. In the case of non-parallel ow, a
simple analytical solution of the balance equations is not
possible and the study is performed numerically by
employing a nite element solution procedure. A further
objective of the present paper is to extend the criterion
for parallel ow discussed in Barletta (2005) and recalled
in the next section, in order to include duct ows with an
axial temperature change as well as duct ows in a uid-
saturated porous medium. The latter task is accomplished
in two short appendices.
2. Necessary condition for parallel ow
Let us analyze mixed convection ow in an inclined duct
with an arbitrary cross-sectional shape. Let us choose
Cartesian coordinates (X, Y, Z) such that the duct cross-
section belongs to the (X, Y)-plane and the Z-axis is parallel
 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393 85

to the duct axis. In other words, the duct cross-section cor- metric wall heat ux. A sketch of the duct and of the
responds to a domain D in the (X, Y)-plane. origin of the angular coordinate with respect to the gravi-
As it has been shown in Barletta (2005), by invoking the tational eld is given in Fig. 1. As it is shown by this gure,
validity of the Boussinesq approximation and by assuming the tilt angle between the duct axis and the direction of the
that the boundary conditions imply an axially invariant gravitational acceleration is u and the vector g 0 is given by
temperature eld T, i.e. oT/oZ = 0, the following statement
g0 g sin uY; 1
holds
A parallel ow solution of the local balance equations, i.e. where Y is the unit vector in the Y-direction. The incoming
a solution such that U 0 = 0, exists in the fully developed wall heat ux is expressed as
 
region only if the temperature eld is such that oT  p
g0  $0 T 0, at every position in the domain D. qw # k  q0 sin # n ; 2
oR RR0 2
In this statement, symbols U 0 and g 0 denote the two-
dimensional projections on the (X, Y)-plane of the uid where (R, #) are two-dimensional polar coordinates, while
velocity U and of the gravitational acceleration g, while n is a dimensionless switch variable which can be equal
$0 is the two-dimensional gradient (o/oX, o/oY). either to 0 or 1. On account of Eq. (2), one obtains in this
In Appendix A, it is shown that the above statement case that the circumferentially averaged wall heat ux van-
holds also for cases such that the temperature eld under- ishes. Moreover, let us assume that the eect of viscous dis-
goes an axial change, oT/oZ 5 0. Moreover, in Appendix sipation is negligible and that the ow is fully developed.
B, it is shown that the same statement holds not only for As a consequence, both the velocity eld U and the temper-
a clear uid, but also for the case of ows in a uid-satu- ature eld T are invariant in the axial direction, namely
rated porous medium according to the DarcyForchheimer oU/oZ = 0 and oT/oZ = 0. Thus, the governing equations
model. In the present paper, the term clear uid will be can be written as
used, following the common practice of treatises on con-
vective ow in porous media, in order to denote the classi- $0  U0 0; 3
0 0 0 0 0 02 0
cal NavierStokes ow when compared to seepage ow in a .0 U  $ U .0 bT  T 0 g  $ P lr U ; 4
porous medium. oP
The condition for the existence of parallel ow implies .0 U0  $0 U z .0 bT  T 0 gz  lr02 U z ; 5
oZ
that, for a non-vertical duct (g 0 5 0), this special ow solu- .0 cp U0  $0 T kr02 T : 6
tion for the fully developed regime can be found only if the
thermal boundary conditions are such that either the vector
eld $0 T is a parallel eld with the same direction as the
vector g 0 or the uid changes its temperature only in the
axial direction, i.e. $0 T 0. The latter case can hardly be
arranged in practice. On the other hand, in the former case,
two-dimensional heat transfer occurs in the direction paral-
lel to g 0 and in the axial direction. One can easily conclude
that, for a vertical duct (g 0 = 0), a parallel ow solution
always exists. Roughly speaking, the necessary condition
for parallel ow in a non-vertical duct is that the isotherms
on a plane transversal to the ow direction must be parallel
straight lines orthogonal to the direction of g 0 . Such a con-
dition can be hardly fullled whenever the duct is not a
parallel plane channel. In fact, for a plane channel, the
geometry of the boundaries allows one to get parallel
straight isotherms in the uid, for instance, by prescribing
uniform temperatures on both the boundary walls (Bar-
letta, 2005). For a dierent geometry of the duct, the nec-
essary condition for parallel ow can be fullled only by
prescribing more complicated thermal boundary condi-
tions, as it is shown in the next section with reference to
a circular duct.

3. Fully developed ow in an inclined tube

Let us consider a clear uid owing in an inclined circu-

lar duct having radius R0 and subjected to a non-axisym- Fig. 1. Sketch of the inclined tube.
86 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393

Since U and T do not depend on Z, by dierentiating Eqs. 3.1. The case n = 0

(4) and (5) with respect to Z, one obtains
If n = 0, a solution of Eqs. (2) and (13)(17) can be
oP
$0 0; 7 found such that H is a function only of the Cartesian coor-
oZ dinate y. Thus, as a consequence of Eq. (1), the condition
o2 P for parallel ow, g0  $0 T 0, is fullled. One can easily
0: 8
oZ 2 nd that

Eqs. (7) and (8) allow one to infer that eld P can be un- H y r sin #: 19
iquely represented as It must be pointed out that the boundary condition equa-
tion (2) with n = 0 is the only one that is compatible with
P X ; Y ; Z AX ; Y BZ; 9
parallel ow for u 5 0. In fact, the parallel ow condition
where B is a constant. implies that H depends only on y and, thus, Eq. (17) can be
Note that the elds U 0 and T can be determined by solv- satised in this case only if H is a linear function of y. Fi-
ing Eqs. (3), (4) and (6). In other words, these elds are not nally, the constraint equation (18) leads to the conclusion
inuenced by the axial component Uz. The latter compo- that H must be proportional to y and the proportionality
nent can be obtained as a last step of the solution proce- constant can be set to 1 by a proper redenition of q0.
dure, i.e. by solving Eq. (5). To summarize, parallel ow implies the validity of Eq.
Let us introduce the dimensionless quantities, (19) and yields the thermal boundary condition equation
(2) with n = 0.
2R0 U0 Uz T  T0 Since the ow is parallel, u 0 = 0 and the axial velocity
u0 ; u ; Hk ; component u fulls the equation
m U0 q 0 R0
X Y R 4R2 . A 2R20 B o2 u o2 u Gr cos u b
x ; y ; r ; a 0 20 ; b ; H 0; 20
R0 R0 R0 l lU 0 ox2 oy 2 4Re 2
2R0 U 0 8gbq0 R40 m together with the no-slip condition for r = 1. In order to
Re ; Gr 2
; Pr :
m km a determine u, it is convenient to express Eq. (20) in cylindri-
10 cal polar coordinates

The reference temperature T0 and the reference velocity U0 o2 u 1 ou 1 o2 u Gr cos u b

r sin # 0: 21
are given by or2 r or r2 o#2 4Re 2
Z R0 Z 2p A solution of Eq. (21) that fulls the no-slip boundary
1
T0 2 dR d#RT ; 11 condition
pR0 0 0
Z R0 Z 2p u1; # 0; 22
1
U0 2 dR d#RU z : 12 can be sought in the form
pR0 0 0
ur; # u1 r u2 r sin #: 23
Thus, Eqs. (3)(6) can be rewritten as
By substituting Eq. (23) in Eq. (21), one obtains a pair of
ou0x ou0y ordinary dierential equations
0; 13
ox oy d2 u1 1 du1 b
  0; 24
ou0 ou0 oa o2 u0x o2 u0x dr2 r dr 2
u0x x u0y x  2 ; 14
ox oy ox ox2 oy 2 d2 u2 1 du2 u2 Gr cos u
!  2 r 0; 25
ou0y ou0y Gr sin u oa o2 u0y o2 u0y dr2 r dr r 4Re
u0x u0y H 2 2 ; 15
ox oy 2 oy ox2 oy subjected to the boundary conditions
 2 2  u1 1 0 u2 1: 26
ou ou Gr cos u ou ou
u0x u0y Hb2 ; 16
ox oy 2Re ox2 oy 2 One obtains
 
Pr 0 oH 0 oH o 2 H o2 H b
ux uy 2 2: 17 u1 r 1  r2 ; 27
2 ox oy ox oy 8
Gr cos u
On account of Eq. (11), the dimensionless temperature H u2 r r1  r2 : 28
32Re
must full the constraint
The value assumed by the constant b can be obtained by
Z 1 Z 2p imposing the constraint (12), thus yielding
dr d#rH 0: 18
0 0 b 16: 29
 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393 87

In fact, one gets the same value of b that would be obtained Gr

in the case of isothermal Poiseuille ow. This result could
have been expected since, as shown by Eq. (23), the buoy-
ancy-induced term u2(r) sin# in the expression of u(r, #) has
a vanishing average value in a duct cross-section. Finally,
the transverse distribution of the dimensionless dierence
between the pressure and the hydrostatic pressure, repre-
sented by function a(r, #), is obtained from Eqs. (14), (15)
and (19). These equations yield
Gr sin u 2 2
ar; # r sin #: 30
4
Function a(r, #) can be determined only up to an arbitrary
additive constant. In Eq. (30), this constant has been xed
so that a = 0 for r = 1 and # = 0. Eqs. (19), (23) and (27)
(30) show that the dimensionless solution is governed by
parameters Gr and Re as well as by the tilt angle u and does
not depend explicitly on Pr.
3.2. The case n = 1
If n = 1, on account of Eq. (2), the incoming wall heat
ux qw(#) is proportional to cos #, i.e. to x. Then, the gra-
dient of T cannot be everywhere parallel to g 0 that is direc-
X Gr sin u; K cos u 33
Re
and Pr. By this optimized parametrization, the limiting
case of a vertical duct (u ! 0) can be dealt with by taking
the limit X ! 0. On the other hand, the limiting case of a
horizontal duct (u ! p/2) can be treated by taking the limit
K ! 0. Note that the optimized parametrization in terms
of X and K is similar to that introduced by Lavine (1988).
Eq. (32) reveals that, for a vertical duct, the value of b is
in any case 16. Therefore, one can base the numerical solu-
tion procedure on the guess that b = 16 for every other pos-
sible value of the tilt angle u or, stated dierently, for every
possible values of X and K. The physical meaning of this
guess is that the dimensionless axial pressure drop is not
inuenced by the buoyancy eect. Stated dierently, buoy-
ancy does not aect the relation between the mass ow rate
and the axial pressure gradient. Thus, this relation is the
same that holds in the case of isothermal ow (Poiseuille
ow). In order to test the reliability of this guess, one pre-
scribes the additional constraint induced by Eq. (12),
namely
Z Z 2p
1 1
1 dr d#ru 0: 34
p 0
0

ted along the y-axis. One thus concludes, in this case, that it The method to get the numerical solution involves two
is impossible to have a parallel ow solution, unless u = 0. steps. First, one solves Eqs. (13)(15) and (17), that is noth-
For u = 0, the duct is vertical and g 0 = 0, so that the con- ing but a 2D natural convection problem in a circular cav-
dition g0  $0 T 0 is satised at every point inside the ity. Then, one uses the obtained numerical values of u 0 and
duct. In this case, a simple analytical solution is allowed. H to solve Eq. (16) and thus obtain u. While the elds u 0
Eq. (17) reduces merely to the Laplace equation for H. and H depend only on Pr and X, the eld u depends also
Hence, consistently with the thermal boundary condition, on K.
one has One can easily check that, for xed values of Pr, Eqs.
(13)(17) as well as the boundary conditions prescribed at
H x r cos #: 31
r=1
Following a procedure similar to that described in Section oH
3.1, one gets u 0; cos # 35
or
 
2 Gr undergo two fundamental symmetries
ur; # 21  r 1 r cos # ; 0 0 1
64Re ux ! u0x
0 1
b 16; ar; # 0: 32 x ! x B 0 C
B y!y C B uy ! u0y C
B C B C
This analytical solution for the vertical duct can be em- B C)B B u!u C C; 36
@ X ! X A B C
ployed as a benchmark to test the numerical solution @ H ! H A
procedure. K ! K
a!a
For u 5 0, the non-parallel laminar solution can be 0 0 1
obtained numerically by employing a Galerkin nite ele- 0 1 ux ! u0x
x!x B 0 C
ment method, implemented through the software package B y ! y C B uy ! u0y C
B C B C
Comsol Multiphysics ( Comsol, AB). Although the ow B C)B B u!u C C: 37
has a 3D nature due to the secondary ow in the xy-plane, @ X!X A B C
@ H ! H A
the numerical solution, based on Eqs. (13)(17), is obtained K ! K
by a purely 2D procedure. Eqs. (13)(17) reveal that the a!a
solution depends on four governing parameters: Re, Gr, On account of these symmetries, one can restrict the anal-
Pr and the tilt angle u. However, one can manage these ysis to positive values of both X and K.
parameters in order to hide the dependence on u. In fact, The Nusselt number can be dened as
the four governing parameters can be reduced to three by
2R0 q0 2
noticing that Eqs. (13)(17) depend only on Nu : 38
kT max  T min Hmax  Hmin
88 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393

It must be pointed out that the maximum and minimum
values of H always occur on the duct wall r = 1. In the
cases of parallel ow, dened either by Eqs. (19) and
(23)(30) or by Eq. (32), one has Nu = 1. Indeed, the con-
vective heat transfer between the hotter and the cooler
parts of the duct wall is entirely due to the secondary ow.
4. Discussion of the results
4.1. The parallel ow case (n = 0)
Poiseuille prole. On the other hand, stronger dierences
appear in the other three plots, where the ow reversal phe-
nomenon arises. The onset of ow reversal takes place next
to (r = 1, # = 3p/2), or (x = 0, y =  1). In fact, when
K > 0, this position represents the coolest one for upward
ow and the hottest one for downward ow. The value
of K corresponding to the onset of ow reversal can be eas-
ily found out by checking the sign of ou/or evaluated at
(r = 1, # = 3p/2). On account of Eqs. (23), (27), (28), one
infers that ow reversal takes place when

The parallel ow solution in the case n = 0 is repre- K > 64: 39

sented in Fig. 2. This gure displays plots of u(r, #) for dif-
ferent values of the governing parameter K. It is easily
veried that, for K = 50, the z-velocity prole presents a What happens in the case K < 0 is easily established from
maximum next to 2 and displays a not too marked asym- Eqs. (23), (27), (28), by noticing that the transformation,
metry with respect to rotations around the z-axis. In this
sense, it does not dier much from the isothermal ow K ! K; # ! # p; 40

Fig. 2. Case n = 0. Plots of u(r, #) for dierent values of K.

 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393 89

leaves u invariant. In particular, this symmetry implies Table 2

that, for K < 0, the onset of ow reversal takes place next Numerical solution (n = 1): comparison with the analytical benchmark
solution (in italic) for X ! 0 and Pr = 7
to (r = 1, # = p/2) for
K umax umin
K < 64: 41
50 2.24683 0.00000
2.24682 0.

4.2. The non-parallel ow case (n = 1) 100 2.72997 0.184710

2.73003 0.184723
The non-parallel ow is studied for a liquid with Pr = 7.
The computational domain is the dimensionless duct cross- 200 3.84788 1.21167
3.84800 1.21168
section, i.e. the unit circle. Seven structured meshes with an
increasing number of quadrangular elements from 1600 to 500 7.39422 4.73249
4900 are dened in this domain in order to test the grid 7.39431 4.73249
independence. Four quantities are monitored in order to
check the gridindependence: the maximum value umax 1000 13.3852 10.7200
13.3855 10.7200
and the minimum value umin of u, the value of the Nusselt
number Nu and the value of the quantity 2000 25.4017 22.7348
Z Z 2p 25.4018 22.7354
1 1 ru02
ek dr d# : 42
p 0 0 2 5000 61.4776 58.8119
61.4789 58.8123
The latter quantity represents the dimensionless kinetic en-
The numerical solution is obtained with X = 107.
ergy associated with the secondary ow and thus is indepen-
dent of K. The grid independence has been tested with
reference to a rather critical ow condition, namely quantity has a unique value for a vertical duct. Since this
X = 107 and K = 104. The results are compared in Table 1. special ow is a parallel one, the expected value of ek should
This table shows that the renements succeed in driving be zero. In fact, the numerically predicted value of ek is
the numerical simulations to convergence. The parameter rather small: 1019.
most aected by the grid renement is ek. However, the rel- In the following, the cases examined refer to 102 6
ative change of ek between the grid with 3600 elements and X 6 107 and 107 6 K 6 104. The choice K = 107 is a
that with 4900 elements (36% increase) is 0.03%, while the
other quantities umax, umin and Nu undergo changes around
0.005%. Moreover, the left hand side of Eq. (34) has values Table 3
lower than 2 1015 in all the cases examined, thus conrm- Numerical solution (n = 1): values of umax, umin, Nu and ek for Pr = 7
ing the validity of the guess b = 16. As a result of the grid X K = 107 K = 102 K = 103 K = 104
independence test, in the following, the mesh with 4096 ele- 10 2
umax 2.000 2.613 12.05 108.2
ments will be used. umin 0.000 0.1030 9.385 105.5
Another check on the reliability of the numerical code is Nu 1.106 1.106 1.106 1.106
ek 0.06339 0.06339 0.06339 0.06339
the comparison with the benchmark analytical solution
(Section 3.2) in the case of a vertical duct, X ! 0. Table 2 103 umax 2.000 2.258 7.605 63.72
umin 0.000 0.000 4.971 61.08
refers to dierent positive values of K and displays the cal-
Nu 1.705 1.705 1.705 1.705
culated values of umax and umin. As it is shown by this table, ek 0.6491 0.6491 0.6491 0.6491
the comparison between the numerical solution, obtained
104 umax 1.999 2.100 5.059 37.97
for X = 107, and the benchmark analytical solution reveals umin 0.000 0.000 2.440 35.34
a very good agreement, which, in most cases, holds for the Nu 2.668 2.668 2.668 2.668
rst 5 signicant digits. Being ek independent of K, this ek 2.608 2.608 2.608 2.608
105 umax 2.000 2.048 3.930 26.39
Table 1 umin 0.000 0.000 1.327 23.76
Numerical solution (n = 1): grid independence test with structured meshes Nu 3.854 3.854 3.854 3.854
having increasing renements, for X = 107, K = 104 and Pr = 7 ek 7.498 7.498 7.498 7.498

Mesh umax umin Nu ek 106 umax 2.001 2.027 3.278 19.49

umin 0.000 0.000 0.6907 16.85
1600 elements 14.9298 12.2820 7.45612 56.2996
Nu 5.417 5.417 5.417 5.417
2025 elements 14.9330 12.2856 7.45412 56.2186
ek 20.54 20.54 20.54 20.54
2500 elements 14.9349 12.2867 7.45318 56.1743
3025 elements 14.9353 12.2875 7.45256 56.1484 107 umax 2.001 2.016 2.863 14.94
3600 elements 14.9353 12.2876 7.45229 56.1323 umin 0.000 0.000 0.3073 12.29
4096 elements 14.9357 12.2881 7.45212 56.1237 Nu 7.452 7.452 7.452 7.452
4900 elements 14.9359 12.2882 7.45195 56.1149 ek 56.12 56.12 56.12 56.12
90 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393

good approximation of the limit K ! 0, i.e. the limit of a
horizontal duct. Table 3 displays the values of umax, umin,
Nu and ek. It must be pointed out that both Nu and ek,
being constructed from the elds u 0 and H, do not depend
on the governing parameter K. Table 3 reveals two impor-
tant features: the Nusselt number and the quantity ek are
increasing functions of X; the onset of ow reversal (nega-
tive values of umin) takes place with higher threshold values
of K as X increases. Physically, an increasing value of X
means a stronger secondary ow. This implies obviously

Fig. 3. Case n = 1. Plots of H (isotherms) and u 0 (proportional arrow plots) for dierent values of X.
 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393 91

the increasing values of Nu and ek, but also the delayed
onset of the ow reversal phenomenon, induced by the
smaller temperature dierences in the duct cross-section.
In fact, the ow reversal phenomenon is due to a su-
ciently intense buoyancy force that may cause locally a
uid ow in the direction opposite to the mean ow. This
suciently intense buoyancy force arises when consider-
able temperature dierences are present within the duct
cross-section.
Table 3 shows that a horizontal duct (K = 107) displays
values of umin and umax compatible with the Poiseuille
velocity prole, whatever is the value of X. This means that
the coupling between the secondary ow and the axial ow
induced by the convective derivative term in Eq. (16) has a

Fig. 4. Case n = 1. Contour plots of u for dierent values of X and K.

92 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393

negligible eect in this case. For all the values of X and K The dimensionless pressure drop in the axial direction,
considered in Table 3, the left hand side of Eq. (34) has val- b, is not inuenced by the buoyancy eect, i.e. it is inde-
ues lower than 4 1015. pendent of X and of K. The Nusselt number and the
The inuence of the parameters K and X on the velocity dimensionless kinetic energy ek associated to the second-
and temperature eld is also pointed out in Figs. 3 and 4. ary ow are both increasing functions of X.
Fig. 3 shows that, for lower values of X, the shape of the The coupling eect between the secondary ow and the
isotherms implies a mainly conductive heat transfer in the axial ow in the special case of a horizontal tube
x-direction, while a dominant thermal stratication in the (K ! 0) is induced only by the convective derivative
direction of g 0 (y-direction) occurs for higher values of X. term in the axial momentum balance and is rather small.
Fig. 3 shows also that the secondary ow is almost every- In fact, in all the cases examined, the axial velocity pro-
where directed in the #-direction, when X is small. On le for the horizontal tube has negligible dierences
the other hand, as X increases, stronger secondary ow from the isothermal Poiseuille prole.
occurs in the neighborhood of # = 0 and # = p, where
the maximum incoming and outgoing heat uxes are pre- Appendix A. Parallel ow for a clear uid
scribed. Fig. 4 reveals that the change in the shape of the
isotherms as X increases implies a displacement in the posi- In the following, it will be shown that the restrictive
tions of the maximum and minimum values of u. For assumption of an axially invariant temperature eld can
X = 102, the position of the maximum axial velocity corre- be released without altering the validity of the necessary
sponds to a value of # intermediate between 0 and p/2. For condition for parallel ow. Let us consider mixed convec-
X = 104, this position is denitely on the plane # = p/2. tion ow of a clear uid in an inclined duct, such that
The latter feature is connected to the thermal stratication the cross-section has an arbitrary shape. Let us choose a
in the direction of g 0 shown in Fig. 3. Cartesian coordinate frame (X, Y, Z) as in Section 2.
In the fully developed region, where
5. Conclusions oU o~.T ; T 0
0; 0; A:1
oZ oZ
Fully developed and laminar mixed convection in an
inclined circular tube has been analyzed. The thermal the mass, momentum and energy balance equations can be
boundary condition prescribed at the duct wall is a non- expressed according to the Boussinesq approximation as
axisymmetric heat ux varying sinusoidally in the angular $0  U0 0; A:2
direction. It has been shown that this thermal boundary 0 0 0 0 0 02 0
condition has a rather special feature: it can be made .0 U  $ U ~.T ; T 0 g  $ P lr U ; A:3
compatible with parallel ow by a proper setting of a oP
.0 U0  $0 U z ~.T ; T 0 gz  lr02 U z ; A:4
switch parameter n. In fact, parallel ow in an inclined oZ
   
duct dierent from a parallel plane channel is an excep- oT o2 T
tion rather than a rule, and this thermal boundary condi- .0 cp $0  U0 T U z k r02 T 2 lU: A:5
oZ oZ
tion provides precisely the exception. A dierent tuning of
the switch parameter or, stated dierently, a (p/2)-rota- In Eqs. (A.1)(A.5), U represents the viscous dissipation
tion of the wall heat ux distribution restores the rule, function and ~.T ; T 0 .T  .0 , where .(T) is the tem-
i.e. non-parallel ow. While, in the parallel ow case, perature-dependent mass density evaluated through the
the governing balance equations admit a simple analytical equation of state.
solution, the solution found in the non-parallel ow case If one assumes U 0 = 0, Eq. (A.3) yields
has been obtained following a numerical nite element ~.T ; T 0 g0  $0 P 0: A:6
procedure. This numerical solution has been found in 0
Since g is a constant vector, by evaluating the two-dimen-
the case Pr = 7, that means, approximately, water at
sional curl of both sides of Eq. (A.6), one obtains
room temperature. The main features of this solution
are the following: 0 $0  ~.T ; T 0 g0  $0 P  g0  $0 ~.T ; T 0
g0  $0 .T b.T g0  $0 T : A:7
The secondary ow velocity and the temperature eld
depend on a unique parameter, X, whose value is deter- Eq. (A.7) ensures the validity of the parallel ow condition
mined by the uid properties and by the duct radius, by reported at the beginning of Section 2.
the amplitude of the wall heat ux distribution, q0, and
by the tilt angle, u. As a consequence, also the Nusselt Appendix B. Parallel ow in a DarcyForchheimer porous
number depends only on X. medium
The axial ow velocity component depends on X and on
another dimensionless parameter, K, that is determined The necessary condition for the occurrence of parallel
by the same quantities involved in the denition of X ow can be stated also in the case of stationary ow in a
as well as by the mass ow rate. uid-saturated porous medium. Let us consider mixed con-
 A. Barletta / Int. J. Heat and Fluid Flow 29 (2008) 8393 93

vection ow in an inclined duct lled with a porous med- local validity of the constraint g0  $0 T 0. The proof
ium, such that the cross-section has an arbitrary shape. coincides with that given in the preceding appendix. In-
Let us assume the validity of DarcyForchheimer law as deed, if the ow is parallel (U 0 = 0), Eq. (B.4) coincides
well as of the Boussinesq approximation, so that with Eq. (A.6).
l F.
U p0 jUjU $P ~
.T ; T 0 g; B:1
K K References
where K is the permeability of the medium and F is Forch-
Aung, W., 1987. Mixed convection in internal ow. In: Kakac, S., Shah,
heimer coecient. The velocity U represents the local aver- R.K., Aung, W. (Eds.), Handbook of Single-Phase Convective Heat
age velocity of the uid often called Darcy seepage velocity. Transfer. Wiley, New York (Chapter 15).
The local mass balance still implies that $  U 0 and the Barletta, A., 2005. On the existence of parallel ow for mixed convection
local energy balance equation is given by in an inclined duct. International Journal of Heat and Mass Transfer
48, 20422049.
l
.0 cp U  $T ^kr2 T U2 ; B:2 Barletta, A., Zanchini, E., 1999. On the choice of the reference temperature
K for fully-developed mixed convection in a vertical channel. Interna-
tional Journal of Heat and Mass Transfer 42, 31693181.
where ^k is the average thermal conductivity of the uid-sat-
Barletta, A., Zanchini, E., 2001. Mixed convection with viscous dissipa-
urated porous medium. As is well known, the term lU2/K tion in an inclined channel with prescribed wall temperatures.
on the right hand side of Eq. (B.2) represents the viscous International Journal of Heat and Mass Transfer 44, 42674275.
dissipation contribution. Barletta, A., Lazzari, S., Zanchini, E., 2003. Non-axisymmetric forced and
In the fully developed region dened through Eq. (A.1), free ow in a vertical circular duct. International Journal of Heat and
Mass Transfer 46, 44994512.
the governing equations can be expressed as Buhler, K., 2003. Special solutions of Boussinesq-equations for free
$0  U0 0; B:3 convection ows in a vertical gap. Heat and Mass Transfer 39, 631
q 638.
l 0 F .0 0
U p U U02 U 2z $0 P ~ .T ;T 0 g0 ; B:4 Chamkha, A.J., Grosan, T., Pop, I., 2002. Fully developed free convection
K K of a micropolar uid in a vertical channel. International Communi-
q cations in Heat and Mass Transfer 29, 11191127.
l F. oP
U z p0 U z U02 U 2z  ~ .T ;T 0 gz ; B:5 Lavine, A.S., 1988. Analysis of fully developed opposing mixed convection
K K oz between inclined parallel plates. Warme- und Stoubertragung 23,
   
oT o2 T l
 249257.
.0 cp $0  U0 T U z ^k r02 T 2 U02 U 2z : Magyari, E., 2007. Normal mode analysis of the fully developed free
oZ oZ K convection ow in a vertical slot with open to capped ends. Heat and
B:6 Mass Transfer 43, 827832.
Weidman, P.D., 2006. Convection regime ow in a vertical slot:
It is easily shown that, as in the case of a free uid, the nec- continuum of solutions from capped to open ends. Heat and Mass
essary condition for the occurrence of parallel ow is the Transfer 43, 103109.