Proceedings of the ASME 2013 Power Conference

Power2013
July 29-August 1, 2013, Boston, Massachusetts, U.S.A.

Power2013-98014

NATURAL CONVECTION HEAT TRANSFER IN HORIZONTAL CYLINDRICAL

CAVITIES: A COMPUTATIONAL FLUID DYNAMICS (CFD) INVESTIGATION.

Daniele Ludovisi Ivo A. Garza

Sargent & Lundy LLC Sargent & Lundy LLC
Nuclear Plant Analytical Group Nuclear Plant Analytical Group
55 E. Monroe St., Chicago, Illinois 60603, USA 55 E. Monroe St., Chicago, Illinois 60603, USA
Email: [email protected] Email: [email protected]

ABSTRACT g Gravitational acceleration;

Many processes in power plants involve the storage and h Convection heat transfer coefficient;
transfer of fluids including water in outdoor pipelines. Under J0 0th Bessel function of the first kind;
extreme cold weather conditions, water can freeze if allowed to J1 1st Bessel function of the first kind;
cool down to the freezing temperature. Installing insulation and k Thermal conductivity of the fluid;
maintaining adequate flow rate can sometimes prevent prob- L Pipe length;
lems. However, during extended non-processing times, there m Integer counter: 1, 2, 3, ...;
are circumstances where cool down cannot be avoided and heat Nu Nusselt number;
tracing along the piping becomes a necessity. Pr Prandtl number;
In many instances, the need for the installation of heat trac- q0 Heat transfer per unit length of the pipe;
ing is simply determined based on pipe size. However, by per- r Pipe radial coordinate;
forming accurate calculations, it is possible to determine if the R Pipe inside radius;
need for heat tracing is real or not, thus saving on installation Ra Rayleigh number;
and maintenance costs. Correlations for the estimation of the t Time;
heat transfer coefficient in horizontal cavities are not sufficiently Taverage Fluid average temperature;
documented in literature. In the present work, two-dimensional Ts Pipe surface temperature;
CFD models are used to investigate the natural convection in α Thermal diffusivity of the fluid;
water-filled horizontal pipes of different diameters. The analy- β Thermal expansion coefficient of the fluid;
sis has been carried out based on the assumption of a uniform γm mth positive roots of the equation J0 (γR) = 0;
pipe surface temperature. The Nusselt number is estimated as µ Dynamic viscosity of the fluid;
a function of the Rayleigh number and shown not to be strongly ν Kinematic viscosity of the fluid;
dependent on the Prandtl number. ρ Density of the fluid.
The analysis and the results of the numerical investigation
are presented and compared to experimental data and other cor-
relations available in literature. The documented correlation
INTRODUCTION
has an expanded range of applicability to high and low Rayleigh
Numerous correlations have been proposed for the over-
numbers, is supported by numerical and experimental results
all heat transfer by natural convection on the outer surface of
and is expressed in a simple form.
horizontal cylinders and in the space between concentric cylin-
ders. However, despite their application in a variety of engineer-
ing applications, natural convection within horizontal cylindri-
NOMENCLATURE cal cavities is not sufficiently documented. Many industrial pro-
cp Specific heat of the fluid; cesses involve the storage and transfer of fluids including water
D Pipe inside diameter; in outdoor pipelines. During extended non-processing times in

1 Copyright
c 2013 by ASME
extreme cold weather conditions, still fluid in pipes can cool (L/D > 2.5). Five different pipe diameters are analyzed: 1 inch,
down and freeze. 2 inches, 4 inches, 8 inches and 16 inches. With the Rayleigh
This study combines the results of conduction and convec- number being a function of the pipe diameter to the third power,
tion heat transfer analyses and experimental data to develop a this set of pipe diameters allows the evaluation of the system
correlation for the prediction of heat transfer from a fluid inside thermo-fluid dynamics for Ra ranging from 103 to 1010 . In all
a horizontal cylinder under quasi-steady conditions. One of the simulations, the Boussinesq approximation is used. For Ra >
applications of the results of this study is in the economic eval- 106 , the flow is turbulent in which case the transient SST k −
uation of heat tracing of horizontal pipes. ω turbulent model is employed. The commercial code STAR-
CCM+ Version 6.04 [2] is used for the simulations. For Ra <
3 · 104 , the Nusselt number shows an asymptotic behavior due to
ANALYSIS the reduced velocity in the cavity. In the limit Ra → 0, the heat
Governing non-dimensional parameters of natural convec- transfer is analyzed by using equations for the heat conduction
tion in closed cavities are the Prandtl (Pr) and Rayleigh (Ra) in a solid media.
numbers [1]. Pr is the fundamental characteristic for the ther-
mal boundary layer for natural and forced convection. Ra mea-
sures the relative effect of the buoyancy force, which drives the RESULTS AND DISCUSSION
natural convection, and the viscous force, which tends to slow The numerical simulations documented in this study con-
down the flow. It incorporates the influence of the fluid proper- sider water as process fluid with the properties as shown in Table
ties, cavity size and the driving force, buoyancy. The greater Ra, 1, which are held constant throughout each simulation. How-
the greater the buoyancy effect and the more turbulent the flow. ever, the results are provided in dimensionless form and thus
For small Ra, the flow is laminar and, for Ra approaching zero, they can be generalized for other Newtonian fluids, tempera-
the fluid becomes still thus behaving similarly to a solid. The tures and pipe sizes.
dimensionless form of the heat transfer coefficient is the Nus-
selt number Nu, which is defined as the ratio of convection heat
transfer to fluid conduction heat transfer under the same condi-
Pr ρ cp k µ β
tions. In the present work, the Prandtl, Rayleigh and Nusselt
numbers are defined as follows: [−] [ mkg3 ] [ kgJK ] [ mWK ] [ Nm2s ] [ K1 ]

ν 1 859.4 4479.9 0.623 1.38E − 4 3.91E − 04

Pr = (1)
α 6 997.6 4186.4 0.620 8.89E − 4 3.91E − 04
15 999.9 4224.5 0.561 1.99E − 3 3.91E − 04
TABLE 1. Fluid properties used in the CFD simulations.
gβ (Taverage − Ts )D3
Ra = (2)
να
The system is initialized with a uniform high fluid temper-
ature and zero angular and radial velocity. A uniform low pipe
q0 hD surface temperature is applied as boundary condition (cooling
Nu = = (3)
πk(Taverage − Ts ) k process). At each time step of the transient analysis, the overall
heat loss from the fluid and the average fluid temperature are
computed. The Rayleigh and Nusselt numbers are computed
with
based on Equations 1 to 3. Three different Prandtl numbers are
evaluated which are typical for liquid water: 1, 6 and 15 (see
q0 = πDh(Taverage − Ts ) (4) Table 1).
A review of the numerical results shows the presence of
where Taverage is the average temperature of the fluid in the pipe four periods in the simulated transient cooling process: an initial
(cross-sectional area), Ts is the temperature of the pipe wall (uni- regime of heat conduction (Period 1), a developing flow regime
form), and q0 is the heat transfer per unit length of the pipe. (Period 2), a quasi steady-state regime (Period 3) and, finally, an
This study investigates the Nusselt number as a function of asymptotic quasi heat conduction regime (Period 4) (see Figure
the Rayleigh and Prandtl numbers: 1). The developed quasi steady-state natural convection pro-
cess is characterized by the presence of a boundary layer near
the pipe wall and a central core with almost isothermal horizon-
Nu = f (Pr, Ra) (5) tal sections and a stratified temperature profile in the vertical
direction (see Figures 2 and 3). As seen in Figure 1, the cal-
CFD modeling is used to simulate the fluid flow and heat culated Nusselt number as function of Rayleigh number rapidly
transfer within two-dimensional horizontal cylindrical cavities decreases (Period 1), stabilizes by slightly increasing (Period

2 Copyright
c 2013 by ASME
 FIGURE 4. Nusselt number as a function Rayleigh number for vari-
FIGURE 1. Typical Nu vs. Ra trend in the transient simulations ous Prandtl numbers.

2), then slowly and steadily decreases (Period 3), and finally ap- Figure 4 shows the results of the numerical CFD simula-
proaches an asymptotic value (Period 4). The duration of each tions (Period 3). As seen, the Nusselt number is a weak function
period is different and depends on the mass of water in the pipe. of the Prandtl number for 1.0 ≤ Pr ≤ 15.0. Therefore, the effect
The quasi steady-state regime (Period 3) is the focus of the nu- of the Prandtl number is ignored. The results of the numerical
merical analysis presented in this study. The asymptotic values analyses are curve-fit and, for 3 · 104 ≤ Ra ≤ 1 · 1010 , the Nusselt
for Period 4 are theoretically investigated based on conduction and Rayleigh numbers are correlated by the following equation:
heat transfer principles.

Nu = 1.15 · Ra0.22 (6)

Figure 4 also shows experimental results for various mix-

tures of water and glycerol by Deaver and Eckert [3]. As seen,
there is a good agreement between the numerical and experi-
mental results. The error band of the proposed correlation is
±10% on the numerical results and ±20% on the experimental
data [3]. The experimental results also confirm that the Prandtl
number does not have a significant impact on the Nusselt num-
ber. Kuehn and Goldstein [4] also proposed a Nusselt number
correlation as a function of the Rayleigh number. This correla-
FIGURE 2. Typical isotherms at low Rayleigh numbers tion was based on engineering judgment by combining various
expressions available in literature for configurations somewhat
comparable to a horizontal cavity (natural convection within
a sphere, concentric cylinders, etc.). The documented equa-
tion was supported by the experimental data only in a narrow
range of Rayleigh numbers (Deaver and Eckert [3]). The error
band of the correlation proposed in the present study is ±20%
on the correlation derived by Kuehn and Goldstein [4] in the
range of Ra supported by experimental evidence. Martynenko
and Khramtsov [5] reported in their work a correlation which
is based on the expression originally published by Kuehn and
Goldstein [4] and added a correction factor which is function of
the Prandtl number. However, the Nusselt numbers calculated
by employing the correlation by Martynenko and Khramtsov [5]
do not agree with the experimental results of Deaver and Eck-
ert [3], the results by Kuehn and Goldstein [4] and the results
FIGURE 3. Typical velocity distribution at low Rayleigh numbers obtained in the present paper.
As the Rayleigh number approaches small values, the Nus-
selt number approaches an asymptotic value (see Figures 1 and

3 Copyright
c 2013 by ASME
4). With the water velocity being small, conduction heat transfer value for Period 4 is found. Numerical calculations show that
equations are used to estimate this asymptotic limit. The theo- this asymptotic value is not dependent on Pr or pipe size and is
retical temperature distribution in a solid cylinder as function of equal to:
time and position is well known in a closed form [6]:

Z R
2 ∞ −αγm2 t J0 (γm r) lim Nu(t) = 5.78 (13)
T (r,t) = Ts + ∑e · 2 · r0 J0 (γm r0 )F(r0 )dr0 t→∞
R2 m=1 J1 (γm R) r0 =0
(7)
where J0 and J1 are Bessel functions of the first kind, γm are CONCLUSIONS
positive roots of the transcendental equation J0 (γR) = 0 and A correlation for the prediction of natural convection heat
F(r) = T0 (r) − Ts is the initial condition. In this study, the initial transfer from a fluid inside a horizontal cylinder with a uniform
temperature distribution T0 (r) is set to be equal to a constant T0 wall temperature is reported in this study. The Nusselt number
for simplification of the calculations and it does not affect the is found to be a function of the Rayleigh number according to
results of the analysis. the following equation:
The Nusselt number is calculated using Equation 3 as Nu =
q0 /πk(Taverage − Ts ). Taverage and q0 are calculated by manipula-
tion of Equation 7, and the expression for Taverage is derived as
follows: Nu = 1.15 · Ra0.22 (14)

which is valid for 3 · 104 ≤ Ra ≤ 1010 and 1 ≤ Pr ≤ 15.

RR 0 [T (r 0 ,t) − T ]dr
r0 =0 2πr s
Taverage (t) = Ts + (8) Additionally, the Nusselt number is found to have an asymptotic
πR2
value of 5.78 for Ra → 0.
or:

∞ REFERENCES
4(T0 − Ts ) −αγm2 t
Taverage (t) = ∑ e (9) [1] Bejan, A., 2004. Convection Heat Transfer. John Wiley &
m=1 R2 γm2 Sons, Inc., Hoboken, NJ.
[2] STAR-CCM+ Version 6.04 User Guide. CD-adapco, Inc.,
The expression for q0 is derived as follows: New York, NY.
[3] Deaver, F., and Eckert, E., 1970. “An interferometric in-
 vestigation of convective heat transfer in a horizontal fluid
0 ∂ T (r,t)  cylinder with wall temperature increasing at a uniform
q (t) = −2πRk (10)
∂t r=R rate”. Proceedings of the 4th International Heat Transfer
Conference, A.I.Ch.E, New York, NY. Paper NC1.1.
or: [4] Kuehn, T., and Goldstein, R., 1976. “Correlating equations
for natural convection heat transfer between horizontal cir-
∞
2 cular cylinders”. International Journal of Heat and Mass
q0 (t) = ∑ 4πk(T0 − Ts )e−αγmt (11) Transfer, 19(10), October, pp. 1127–1134.
m=1
[5] Martynenko, O., and Khramtsov, P., 2005. Free-Convective
Heat Transfer. Springer, Heidelberg, Germany.
With the above definitions, the following expression for the [6] Ozisik, M. N., 1989. Boundary Value Problems of Heat
Nusselt is obtained: Conduction. Dover Publications, Inc., New York, NY.

∞
2
R2 ∑ e−αγm t
m=1
Nu(t) = 2t (12)
∞ −αγm
e
∑ γm2
m=1

As seen in the above equation, the Nusselt number is a func-

tion of time. As time tends to infinity, the temperature distribu-
tion in the solid media is expected to be similar to the distri-
bution in the fluid at the end of Period 4, with a warmer core
region and a cold outer boundary. Taking the limit of the Nus-
selt number function in Equation 12 for t → ∞, the asymptotic

4 Copyright
c 2013 by ASME