You are on page 1of 11

# International Journal of Heat and Mass Transfer 44 (2001) 18571867

www.elsevier.com/locate/ijhmt

## A computer-aided parametric analysis of mixed convection in

ducts
G.J. Hwang a, S.C. Tzeng a, C.Y. Soong b,*
a
Department of Power Mechanical Engineering, National Tsing Hua University, Hsinchu, Taiwan 30043, ROC
b
Rotating Fluids and Vortex Dynamics Laboratory, Department of Aeronautical Engineering, Chung Cheng Institute of
Technology,Taoyuan, Taiwan 33509, ROC

Abstract
This study develops a novel technique of computer-aided parametric analysis (CAPA) to formulate simple corre-
lations for thermal ow characteristics in a complex convective ow system. To demonstrate the validity of the tech-
nique, fully developed mixed convection in uniformly heated horizontal ducts of square and circular cross-sections are
employed as illustrative examples. The eects of secondary ow generated by the thermal buoyancy force are included.
The CAPA technique employs characteristic quantities, constant factors and multi-term relationships to convert the
governing equations into a set of algebraic equations. Relatively, this technique generates more extensive results than
the conventional order-of-magnitude and scaling analyses. With the aid of limited data from numerical results (or
measurements), formulas for evaluating the cross-sectional averages of axial velocity and temperature, the strength of
secondary vortices, friction factors and the heat transfer rates at various Pr and ReRa can be derived. Comparisons of
the present results and the direct numerical solutions manifest quite satisfactory performance of the CAPA technique.
Merits of the CAPA technique are: (1) less eort required in the developing procedure, (2) clear parameter-dependence
of the thermal ow characteristics, and (3) convenience in using the resultant correlations of algebraic form. 2001
Keywords: Computer-aided parametric analysis; Mixed convection; Duct ow; Secondary ow eect; Friction and heat transfer
correlations

## 1. Introduction lined the general procedure of dimensional analysis.

Although the number of non-dimensional parameters
In the 17th century, Newton originated the concept depends on the number of governing dimensional vari-
of dynamical similarity. Forty years later, Euler was the ables and their rank of dimensional matrix, there are
rst who discussed unit and dimensions in physical re- many combinations of the dimensionless parameters.
lationships. In 1822, Fourier applied the geometrical Consequently, little progress has been made after the
concept of dimension to physical quantities in heat ow early stage of development.
problems. Near the end of the 19th century, Reynolds, In recent years, great progress has been made in
Rayleigh, and many other researchers [1] successfully numerical simulation of physics and engineering, but the
applied the idea of dimensional analysis. Vaschy in 1892, direct numerical computation of full NavierStokes
Riabouchinsky in 1911 and Buckingham [2,3] reported equations and the energy equation is still not cost-
the establishment of the PI method. Buckingham out- eective. In addition, even with a large amount of
numerical results, it is still not very easy to extract the
*
Corresponding author. Tel.: +886-3-390-8102; fax: +886-3- parametric dependence and the physical senses involved.
389-1519. Therefore, simplication of the equations through
E-mail address: cysoong@ccit.edu.tw physical consideration is indispensable. Schlichting [4]
http://www.ccit.edu.tw/RFVDLab (C.Y. Soong). successfully applied an order-of-magnitude analysis

PII: S 0 0 1 7 - 9 3 1 0 ( 0 0 ) 0 0 2 4 5 - 3
1858 G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867

## Nomenclature u; v; w dimensionless velocity components in x,

y and z directions or in r, / and z
A cross-sectional area (m2 ) directions
Ai ; A0i ; A00i constant factors, i 1; 2; . . . ; 6 X; Y;Z Cartesian coordinates (m)
C parameter C1 D3e =4lv Re=w x; y; z dimensionless Cartesian coordinates
C1 axial pressure gradient dPo =dZ
N=m3 Greek symbols
C2 axial temperature gradient oT =oZ a thermal diusivity m2 =s)
(K/m) b coecient of thermal expansion (1/K)
De equivalent hydraulic diameter 4A=S h dimensionless temperature difference
(m) T Tw =C2 De PrC
f friction factor 2sw =qW
2 l viscosity N s=m2
g gravitational acceleration (m=s2 ) m kinematic viscosity (m2 =s)
h average heat transfer coecient N; n dimensional and dimensionless vorticity
(W=m2 K) (unit of N is 1/s)
Kf thermal conductivity of uid (W/(m K)) q density kg=m3
Nu Nusselt number hDe =Kf sw wall shear stress N=m2
Po axial pressure distribution which is a W; w dimensional and dimensionless stream
function of Z only (N=m2 ) function (unit of W is m2 =s)
Pr Prandtl number m=a Superscript
q constant wall heat ux (W=m2 s) cross-sectionally averaged quantity
Ra Rayleigh number gbC2 D4e =ma
Re Reynolds number W De =m Subscripts
S circumference of cross-section (m) b bulk quantity
T local temperature (K) c characteristic quantity
Tw wall temperature (K) o condition for pure forced convection
U; V ; W velocity components in X, Y and Z di- q constant wall heat ux
rections or in R, / and Z directions (m/s) w wall condition

(OMA) to the boundary layer ow over a at plate. and rational correlation for buoyancy eects in rotation-
Cheng et al. [5] used the same technique to analyze induced mixed convection ow.
mixed convection ow in the thermal entrance region of The purpose of this study is to develop a technique of
horizontal rectangular channels. One can derive dimen- parametric analysis of mixed convection. By formulating
sionless parameters from the ratios of individual terms and solving a set of simple algebraic equations and in-
in the governing equations and the associate boundary voking the aid of less numerical solutions or experimental
conditions. However, the OMA technique derives no data, a functional relationship between thermal-ow
simple algebraic relations among these parameters. characteristics and the governing parameters can be
The technique of using a two-term relation from the formulated. In the cases considered in the present work,
physical similitude is known as scaling or scale analysis only two point data obtained by numerical computations
(SA). Bejan [6] applied this SA technique to various heat (or experimental data) are needed to determine the con-
transfer problems. Krane and Incropera [7] reported a stant factors in the correlation. To keep the complete
scaling analysis of the unidirectional solidication of a physical eects, in the present computer-aided parametric
binary alloy. Due to the ow complexity, the two-term analysis (CAPA) technique, all terms in the momentum
relationship from the physical similitude may not fully and energy balance equations are retained. Simple ex-
describe the mechanism of transport phenomena. The pressions of the secondary ow eects are also introduced
conventional scaling analysis deals with a simple ow into the correlation. Finally, a set of correlations for ow
system by comparing two terms in balance equations to and heat transfer characteristics of mixed convection
get the order-of-magnitude of certain physical quantity. with the presence of secondary ow can be completed.
For a complex ow problem, only some dimensionless
parameters can be resulted by this class of techniques.
Multiple-term relation would be more appropriate. In a 2. Basic equations
recent work by Soong and Chyuan [8], for example,
three and four terms of tangential and radial momentum The thermal uid ow considered in the present
equations, respectively, were used to develop concise study is steady and laminar in a hydrodynamically and
G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867 1859

## Fig. 1. Physical models and coordinate system.

thermally fully developed region of a horizontal square Eqs. (1)(3) can be cast into the dimensionless form,
duct or a horizontal circular pipe. The axial pressure and
uon=ox von=oy r2 n RaCoh=ox; 6
temperature gradients are constant and the peripheral
wall temperature is uniform. Fig. 1(a) displays the
physical model and the coordinate system. In the present uow=ox vow=oy r2 w 4; 7
analysis, Cartesian coordinates system is applied to both
cases of square channel and circular pipe. The viscous Pruoh=ox voh=oy r2 h w; 8
dissipation and compressibility eects in the energy
and the vorticity and stream function relations, (4a) and
equation are neglected. The Boussinesq approximation
(4b), become
is assumed to be valid. After a cross-dierentiation in
cross-sectional plane, the governing equations for a n r2 w ov=ox ou=oy; 9a
thermal ow fully developed in Z direction [9] are u ow=oy; v ow=ox; 9b
written as
U oN=oX V oN=oY mr2 N bgoT =oX ; 1 where r2 o2 =ox2 o2 =oy 2 . The solution contains
two independent parameters RaC and Pr [9]. The value
of Pr ranges from 0 to 500 in the present study. Con-
U oW =oX V oW =oY 1=qoPo =oZ mr2 W ;
sequently, the boundary conditions at channel walls can
2 be written as
U oT =oX V oT =oY W oT =oZ ar2 T ; 3 ow=on w h 0; 10
where where n is a unit normal to the channel wall. The
2 boundary value of stream function at the channel walls
N r W oV =oX oU =oY ; 4a
is W 0. The symmetry condition is not assumed and
U oW=oY ; V oW=oX : 4b the computation is performed over the whole cross-
By employing the following dimensionless variables and section of the channel and it demonstrates that the ow
parameters, and temperature elds remain symmetric in the ranges of
RaC and Pr considered. In the present work, the nu-
x X =De ; y Y =De ; u UDe =m; merical solution of vorticityvelocity formulation, Eqs.
v VDe =m; w WDe =mC; (6)(8), (9a) and (9b), plays a complementary role to
h T Tw =C2 De PrC; C1 oPo =oZ; 5 support the development of CAPA. The PDE system is
discretized by utilizing the power-law scheme [10]. The
C2 oT =oZ; C C1 D3e =4ml;
boundaryvorticity method [11] is used to solve the
Ra gbC2 D4e =ma; Pr m=a; vorticity transport equation and the SIS algorithm [12] is
1860 G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867

applied for solution of the resultant dierence equations. nitudes with governing parameters. Similar to the tech-
The iteration is terminated when the variables satisfy the niques of SA and OMA, the characteristic quantities are
criterion as used for representing the corresponding ow variables,
X and the ratios of characteristic quantities for the deriv-
n1
e Fi;n1 n 5
j Fi; j =Fi; j 6 5  10 ; atives in the governing equation. As shown in Table 1,
i; j
the CAPA technique introduces a positive or negative
where F represents W; n; w, and h. The subscripts i and j sign by considering the physical meaning of each term in
show the ith and the jth grid in x and y directions, re- the governing equation. On the contrary, the SA and
spectively. The superscript n indicates the nth iteration. OMA techniques do not consider any sign. The CAPA
The numerical errors are less than 0.12%, when using the technique also uses constant factors to make up the
mesh points of 41  41. The detailed procedure of nu- dierence in the magnitudes of the ratios of character-
merical solution has been given in the literature [9] and istic quantities and the derivatives. Finally, the PI
will not be repeated here. method shows only an unknown function with non-
After computation, C converts to the Reynolds dimensional parameters, the SA technique yields only
number Re W De =m by the relation C Re=w. Fol- dimensionless parameters, and the OMA technique
lowing the conventional denitions, the expressions for provides the dimensionless partial dierential equations
the friction parameter f  Re and the Nusselt number Nu with non-dimensional parameters. The present CAPA
are technique will give a set of algebraic equations with
parameters. As seen in Table 1, the OMA and CAPA
2
f  Re 2sw =qW W De =m 2=w; 11 techniques can fully describe the physics in the transport
phenomena of the problem. The present CAPA tech-
Nu hDe =Kf w=4wh=w; 12 nique dramatically simplies the process in solving a set
of algebraic equations instead of solving PDE with the
where w and wh are the cross-sectional averages evalu- OMA technique.
ated by using Simpson's rule.
3.2. Development of parametric analysis by CAPA
technique
3. CAPA technique
Consider the case of Pr 0:73 and RaC
3.1. Comparison of CAPA and other methods 105 ReRa 1:385  104 , Fig. 2(a) shows the dimen-
sionless isotherms and constant axial velocity lines and
Before developing the CAPA for the mixed convec- Fig. 2(b) shows the stream function and secondary ow
tion ows in ducts, a comparison of CAPA and other velocity vector on cross-plane of a square duct. To il-
techniques is to be presented rst. In the CAPA analysis, lustrate the CAPA technique, Fig. 2(c) shows the sche-
the parametric correlations are directly derived from the matic diagram for dimensional temperature and axial
governing equations and it is assumed that the proles velocity on the plane x 0:5. Fig. 2(d) depicts the sec-
of velocity components and temperature distribution ondary ow velocities on the planes x 0:25, x 0:75
remain unchanged in their shapes but changed in mag- and y 0:5. The present CAPA technique assumes that

Table 1
Comparison of PI method, SA, OMA and CAPA techniques
Characterstics Analysis
PI method SAa OMA CAPA
Sign NA + PDE 
Factor NA Unit PDE Constant
TT relationb NA Oneone Oneone Multiple terms
Equation for N-D. Unknown function NAd PDEe Algebraic
parametersc
Physics in T.P.f No May be partially Fully Fully
a
SA: Scale analysis based on the book written by Bejan [6].
b
TT: Term to term.
c
N-D: Non-dimensional.
d
NA: Not applicable.
e
PDE: Partial dierential equation.
f
T.P.: Transport phenomena.
G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867 1861

Fig. 2. Field and cross-sectional average properties of mixed convection at Pr 0:73 and RaC 105 in a square duct: (a) temperature
and axial velocity; (b) stream function and vector distributions; (c) averages or characteristic quantities of temperature and axial
velocity; (d) cross-sectional ow rate.

the temperature, axial velocity, and secondary ow for dependent and independent variables, the original
patterns in the ranges of parameters investigated do not PDEs are then turned into that of algebraic form. Each
deviate much from the ones shown in this gure. The term involves a constant factor in order to make up the
cross-sectional averages of temperature and the axial approximations. In a resultant equation of n terms, ac-
velocity, respectively, are considered as the characteristic tually, only n 1 terms need this kind of factors. For
temperature and axial velocity, and the maximum value example, with scales of the velocity components and
of the stream function on the cross-plane is used to their derivatives
characterize the total secondary ow rate.
U oW=oY  Wc =De ; V oW=oX  Wc =De ;
In working procedure of the CAPA technique, by
employing the properly dened characteristic quantities oW =oX  Wc =De ;
1862 G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867

## oW =oY  Wc =De ; oW =oX 2  Wc =D2e ; In Eq. (16a), Nc Wc =D2e is used to approximate

2 2 N r2 W in Eq. (1). By substituting Tc Tw in
o W =oY  Wc =D2e ;
Eq. (15b) into (16b) and introducing dimensionless
and the factors A04 and A004 , the terms in the axial mo- stream function w Wc =m and A2  A02 A5 , one has
mentum Eq. (2) can be expressed as A02 A5
A1 w2 w ReRa 0 17a
U oW =oX V oW =oY  A04 Wc =De Wc =De ; 13a A6 Prw 1
1=qoPo =oZ C1 =q; 13b or
2
mr W  A004 mWc =D2e : 13c A1 A6 Pr 2 1 A2
w3 w w ReRa 0:
Combining Eqs. (13a)(13c) with A4  A04 =A004 and A03  A1 A6 Pr A1 A6 Pr A1 A6 Pr
1=A004 , nally, one has an algebraic relation 17b

## Wc Wc C1 Wc From [13], the solution of the above cubic equation is

A4 A03 m 2 14a
De De q De a1
wM N as Q > 0;
3 
or p 18
   / a1
C1 Wc m 2 Q cos as Q < 0;
Wc A03 A4 2 2 : 14b 3 3
q De De
where
It is seen that the inertia term of the original equation is
A1 A6 Pr 1 A2
replaced by A4 Wc =De Wc =De , in which the ratio Wc =De a1 ; a2 ; a3 ReRa;
stands for the mean secondary ow velocity U, Wc in- A1 A6 Pr A1 A6 Pr A1 A6 Pr
dicates the maximum value of stream function at vortex
center, Wc represents the mean axial velocity, and the 3a2 a21 9a1 a2 27a3 2a21
Q ; R ;
constant A4 makes up the dierence between the 9 54
magnitudes of U oW =oX V oW =oY and Wc =De q
p q
p
Wc =De . The rst term with a constant A03 on the RHS M
3
R Q3 R2 ; N
3
R Q3 R2 ;
of Eq. (14a) represents the pressure-gradient term. The
R
last one with a negative sign is for the viscous term. cos / p :
Eqs. (14a) and (14b) tells that the pressure gradient Q2
drives the ow and is balanced by the inertia force and
By employing the dimensionless mean velocity
the viscous force. This describes the physics of transport
w Wc De =mC and temperature h Tc Tw =C2 De PrC
phenomena.
and putting A3 4A03 , Eqs. (14b) and (15b) change to the
Following the procedure mentioned above, the en-
dimensionless forms,
ergy equation (3) is turned into
A3
Wc T c T w Tc Tw w ; 19
A6 A5 Wc C2 a 15a 1 A4 w
De De D2e

or A5 w
h : 20
  1 A6 Prw
Wc a
Tc Tw A5 Wc C2 A6 2 2 15b
De De Instead of solving w shown in Eq. (18), one may also
solve it reversely by nding the value of ReRa in
where (Tc Tw ) indicates the mean temperature dier- Eq. (17a) and (17b) with an increment of w from the zero
ence between the uid and the wall. Thermal convection value, i.e.,
in the axial direction, A5 C2 Wc , is balanced by the
thermal convection, A6 Wc =De Tc Tw =De , and the ReRa A1 w 1 A1 A6 w2 A6 Prw3 =A2 : 21
heat conduction, aTc Tw =D2e .
For the vorticity transport equation (1), one has With this value of w, one may obtain the dimensionless
mean axial velocity w and mean temperature dierence h
Wc 1 Wc Wc Tc Tw
A1 m 4 A02 bg 16a from Eqs. (19) and (20) for the corresponding ReRa.
De De D2e De De Following the conventional denitions of f  Re in
or Eq. (11), one can readily obtain the ratio of friction
factors as
D3e
A1 W2 W A02 bgTc Tw 0: 16b f  Re=f  Reo 1 A4 w; 22
m2
G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867 1863

## where f  Reo A3 by considering w 0 at ReRa 0. w Tc Tw

0m A02 bg : 28
For the denition of Nu in Eq. (12), the mixed mean D4e De
temperature dierence wh=w is the same as the mean
temperature dierence, h in the CAPA technique. Con- By substituting Tc Tw of Eq. (5) into (8), one has
sequently, the Nusselt number relation is 1 A2
w2 w ReRa 0: 29
A6 Pr A6 Pr
Nu=Nuo 1 A6 Prw; 23
Then one has a positive solution for w,
where Nuo is the Nusselt number for w 0. Eqs. (22) s
and (23) give simple expressions of the eects of sec- 1 1 A2
ondary ow on the friction factor and Nusselt number, w ReRa: 30
2A6 Pr 4A26 Pr2 A6 Pr
respectively. The present expressions (22) and (23) are
disclosed in the literature for the rst time. The value of stream function will be zero for nite ReRa
and Pr ! 1. The zero limiting value can be observed
3.3. Limiting case of Pr ! 0 also from Eq. (17b). The ratio of friction factors in
Eq. (22) yields
For the case of Pr ! 0, the thermal convection in the
f  Re
ow direction is balanced by the conductive heat 1 A4 w
transfer in the cross-sectional plane. The convection f  Reo
term in the cross-sectional plane is not important. By A4 p 
1 1 4A2 A6 PrReRa 1 1:0
using the characteristic quantities, the energy equation 2A6 Pr
becomes 31
Tc Tw
A5 Wc C2 a 24a for an innite Prandtl number. The ratio of Nusselt
D2e number becomes
r
or Nu 1 1
  1 A6 Prw A2 A6 PrReRa: 32
a Nuo 2 4
Tc Tw A5 Wc C2 : 24b
D2e
The constant factors can be determined by using a few
By substituting Tc Tw into the vorticity transport data from numerical solutions of the governing PDEs
equation (16b), one has (or the measurements if available). For w 0 at
ReRa 0, the numerical data of pure forced convection
A1 w2 w A2 ReRa 0: 25 determine A3 and A5 . For a given Pr and ReRa, the
mixed convection data determine A4 and A6 . However,
This equation can be obtained readily by multiplying in the present study, the values of ReRa are chosen by
Eq. (17b) by A1 A6 Pr and putting Pr 0. The solution for considering the minimum percentage RMS error at 11
positive w is points over the range of ReRa. Tables 2 and 3 list the
s factors Ai ; i 1; 2; . . . ; 6 for Pr 0500.
1 1 A2
w 2
ReRa: 26
2A1 4A1 A1
4. Results and discussion
The ratio of friction factors can be written as a function
of ReRa. In the present study, the CAPA technique yields
f  Re simple correlations for combined free and forced lami-
1 A4 w
f  Reo nar convection in ducts for Prandtl number ranging
A4 p  from 0 to 500 and ReRa 0 to around 1  105 . It covers
1 1 4A1 A2 ReRa 1 : 27 the liquid metal for small Pr, gases, water, and engine
2A1
oils for large Pr. The CAPA technique provides a
Note that the Nusselt number ratio is Nu=Nuo 1:0 method to derive simple and accurate correlation for-
for Pr 0. mulas of a complex thermal uid ow with the support
of only few data from numerical calculations or exper-
3.4. Limiting case of Pr ! 1 imental measurements.
To demonstrate the eectiveness of the CAPA tech-
In the case of large Prandtl number, Pr ! 1, the nique, ow pattern of one pair of counter-rotating
buoyancy force is balanced by the viscous force in the vortices in mixed convection duct ows is selected. The
vorticity transport equation (16a), i.e., value of the stream function is zero on the square
1864 G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867

Table 2
The corresponding factors for mixed convection in a horizontal square ducta
Factors Prandtl No. (Pr)
0 0.01 0.1 0.73 7.2 100 500
2 2 2 1
A1 1:632  10 1:846  10 2:165  10 1:128  10 1.759 40.977
A2 5:560  105 5:560  105 5:563  105 5:561  105 5:570  105 5:577  105 5:573  105
A4 1:362  102 3:431  102 4:347  102 4:761  102 2:434  102 2:541  102 2:581  102
A6 8:776  101 1:080  101 9:529  102 8:906  102 8:645  102 7:992  102
a
A3 1:404  101 and A5 4:820  102 for all Prandtl numbers. These factors are evaluated at ReRa 14:0 and 10,444.3 for
Pr 0, ReRa 14:0 and 33,094.6 for Pr 0:01, ReRa 14:0 and 61,807.5 for Pr 0:1, ReRa 14:0, and 91,591.5 for Pr 0:73,
ReRa 14:0, and 68,603.4 for Pr 7:2, ReRa 14:0, and 10,532.6 for Pr 100, and ReRa 1:4 and 140.2 for Pr 500.

Table 3
The corresponding factors for mixed convection in a horizontal circular ducta
Factors Prandtl No. (Pr)
0 0.01 0.1 0.73 7.2 100 500
2 2 2 1
A1 6:909  10 6:951  10 7:482  10 1:685  10 2.172 3.408
A2 7:743  104 7:736  104 7:736  104 7:782  104 7:791  104 7:750  104 7:750  104
A4 5:138  102 6:174  102 6:509  102 4:188  102 1:262  102 1:044  102 1:038  102
A6 1.870 3:124  101 8:940  102 5:248  102 4:177  102 3:577  102
a
A3 4:921  101 and A5 1:903  101 for all Prandtl numbers. These factors are evaluated at ReRa 49:2 and 4054.5 for Pr 0,
ReRa 49:2 and 4153.9 for Pr 0:01, ReRa 49:2 and 4166.7 for Pr 0:1, ReRa 49:2, and 92,182.8 for Pr 0:73, ReRa 49:2, and
4804.3 for Pr 7:2, ReRa 49:2, and 349.7 for Pr 100, and ReRa 49:3 and 147.8 for Pr 500.

## channel wall and along the centerline of symmetry, and

it is maximum in the sense of absolute value at the center
of counter-rotating vortices. The maximum value of the
stream function may be regarded as the secondary ow
rate of one of the vortices. It is seen that the stream
function is a signicant variable in the present para-
metric analysis. The relationships of the stream function
and ReRa to the Prandtl number in the range Pr 0500
are plotted in Fig. 3. The results show that the stream
function of CAPA ts well with the numerical solutions.
The RMS of the percentage error is about 2.79% for
Pr 0:73. The errors for the other Prandtl number are
also shown in Fig. 3.
To illustrate the accuracy of the technique on the
mean axial velocity, the numerical solution is used to
obtain A3 and A4 in Eq. (19). In Fig. 4, comparison of
the mean axial velocity with the numerical solution is
made for various Prandtl numbers. A maximum of
0.58% dierence from the numerical solution for
Pr 0:73 is observed. The dierences for the other Fig. 3. Comparison between w predicted by CAPA and nu-
Prandtl numbers are also shown in Fig. 4. merical solutions for mixed convection in a square duct.
The temperatures with ReRa for Pr 0500 are
shown in Fig. 5. The results of the simple algebraic
equation of the CAPA technique agree well with nu- with RMS dierence in percentage. The maximum dif-
merical predictions. With the CAPA technique, the ference between the CAPA results and the numerical
ratios of friction factor and Nusselt number can be ex- predictions is 0.65% for Pr 0:73. Fig. 7 shows the
pressed by Eqs. (22) and (23), respectively. Fig. 6 shows comparisons of the evaluated buoyancy eects on heat
the friction parameter f  Re=f  Reo for Pr 0500 transfer performance. It is noted that, as Nu=Nuo is
G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867 1865

## Fig. 6. Comparison between f  Re=f  Reo predicted by

Fig. 4. Comparison between w predicted by CAPA and nu-
CAPA and numerical solutions for mixed convection in a
merical solutions for mixed convection in a square duct.
square duct.

## Fig. 5. Comparison between h predicted by CAPA and nu-

Fig. 7. Comparison between Nu=Nuo predicted by CAPA and
merical solutions for mixed convection in a square duct.
numerical solutions for mixed convection in a square duct.

## plotted vs PrReRa rather than ReRa, the curves at 5. Conclusions

dierent Pr get close to each other and show weak
Pr-dependence. The RMS dierences of Nu=Nuo A technique of CAPA has been developed for de-
are slightly larger than those of f  Re=f  Reo . This is riving simple and quite accurate formulas with the aid of
due to an over-simplication of the non-linear term only few numerical data for a complex convective ow
wh=w to h. system. The CAPA technique formulates a set of alge-
The corresponding results for circular ducts are braic equations from the original conservative equations
shown in Figs. 8 and 9, respectively. These two gures with considerations of multi-term relationships, physical
demonstrate that the correlations (22) and (23) are also meaning, sign and scale of each term, and proper
available to the mixed convection in circular ducts. For constant factors for approximation. It is much easier
clarity, the value of ReRa is used in Fig. (9). to solve the resultant algebraic equations of the
1866 G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867

## Nusselt number, Nu=Nuo 1 A6 Prw, are obtained.

These two simple correlation equations have not been
reported in the literature yet. The maximum deviation
between the correlations and the numerical data is no
more than a few percent for Pr 0 to 500 and ReRa
01  105 .
For limiting cases of Pr ! 0 and Pr ! 1, the ratios
of friction factor and Nusselt number can be written
explicitly in terms of ReRa or PrReRa in Eqs. (27), (31)
and (32). These equations provide an ecient evalua-
tion of the ow and heat transfer results ReRa or
PrReRa. Finally, we have complete correlations for
ow and heat transfer characteristics of mixed con-
vection in the presence of secondary ow, which are
not possible to obtain by use of conventional scaling
analysis.
With proper considerations, the CAPA technique
could be extended to the thermal ow in rectangular
Fig. 8. Comparison between f  Re=f  Reo predicted by channels of aspect ratio other than unity. However, it
CAPA and numerical solutions for mixed convection in a cir-
should be noted that the present version of CAPA still
cular duct.
has some restrictions. It is not appropriate for the ows
turbulence at very high Re and/or very high Ra. To
implement the CAPA technique for analysis of the
above-mentioned complex ows is a challenge but
worthwhile work.

References

## [1] J. Palacios, Dimensional Analysis, Macmillan, New York,

1964.
[2] E. Buckingham, On physically similar systems: illustrations
of the use of dimensional equations, Phys. Rev. 4 (1914)
345376.
[3] E. Buckingham, Model experiments and the form of
empirical equation, Trans. ASME 37 (1915) 263296.
[4] H. Schlichting, Boundary Layer Theory, seventh ed.,
McGraw-Hill, New York, 1979.
[5] K.C. Cheng, S.W. Hong, G.J. Hwang, Buoyancy eects on
Fig. 9. Comparison between Nu=Nuo predicted by CAPA and laminar heat transfer in the thermal entrance region of
numerical solutions for mixed convection in a circular duct. horizontal rectangular channels with uniform wall heat ux
for large Prandtl number, Int. J. Heat Mass Transfer 15
(1972) 18191836.
[6] A. Bejan, Heat Transfer, Wiley, New York, 1993.
characteristic quantities than to solve the original system [7] M.J.M. Krane, F.P. Incropera, A scaling analysis of the
of partial dierential equations. unidirectional solidication of a binary alloy, Int. J. Heat
In the present study, the CAPA technique has been Mass Transfer 39 (1996) 35673579.
implemented to yield simple correlations for the laminar [8] C.Y. Soong, C.H. Chyuan, Investigation of buoyancy
eects in non-isothermal rotating ows by scaling analysis
fully developed mixed convection in horizontal square
and a novel similarity model, Chin. J. Mech. 14 (1998) 193
and circular ducts for Prandtl numbers from 0 to 500
207.
and ReRa 01  105 . The range of Prandtl numbers [9] F.C. Chou, G.J. Hwang, Combined free and forced
includes the liquid metal of small Pr, gases, water, and laminar convection in horizontal rectangular channels for
engine oils of large Pr. By the CAPA technique, the high ReRa, Can. J. Chem. Eng. 62 (1984) 830836.
ratio of cross-sectional (peripheral) averages of friction [10] S.V. Patankar, Numerical Heat Transfer and Fluid Flow,
factor, f  Re=f  Reo 1 A4 w, and the ratio of McGraw-Hill, New York, 1980.
G.J. Hwang et al. / International Journal of Heat and Mass Transfer 44 (2001) 18571867 1867

[11] G.J. Hwang, K.C. Cheng, Boundary vorticity method for [12] S.L. Lee, A strongly implicit solver for two-dimension
convective heat transfer with secondary ow application to elliptic dierential equations, Numer. Heat Transfer B 16
the combined free and forced laminar convection in (1989) 161178.
horizontal tubes, in: Proceedings of the Fourth Inter- [13] M.R. Spiegel, Mathematical Handbook of Formulas and
national Heat Transfer Conference, Versailles, Paris, vol. 4, Tables, McGraw-Hill, New York, 1968, pp. 3233.
Paper No. NC3.5, 1970.