23rd International Symposium on Transport Phenomena

Auckland, New Zealand

1922 November 2012

CFD Study of Turbulent Cross-flow in an In-line Tube Bundle

Youngmin Bae1, Han-Ok Kang1 and Young-In Kim1

1
Korea Atomic Energy Research Institute
Daeduk-daero 989-111, Yuseong-gu, Daejeon, 305-353, South Korea

Abstract great deal of experimental and numerical studies have been

conducted for several decades, to understand the detailed features
This study numerically investigates the turbulent cross-flow in an of shell-side cross-flow over tube bundles [2,9] as well as to
in-line tube bundle consisting of 10 rows of rods arranged with a develop correlations for pressure drop and heat transfer
constant pitch-to-diameter ratio of 1.5. The three-dimensional coefficients [4,12]. With regard to the velocity and turbulence
unsteady incompressible flow is computed using a large eddy distributions, extensive measurements have also been made using
simulation approach at a Reynolds number of 27000 based on the laser Doppler anemometry (LDA) and particle image velocimetry
inlet velocity and tube diameter. The validity of the numerical (PIV) [1,6]. Clarifying the detailed structures of a cross-flow in
results is assessed by comparing the time-averaged streamwise tube bundles, however, still remains difficult and expensive in
velocity distributions behind the tubes and separation points with experiments.
the experimental data. The flow instability originated from the Understanding the flow characteristics in tube bundles is a
streamwise vortices formed in the near-wake is discussed in the critical prerequisite for an effective and reliable design of a heat
paper, with the emphasis on the spatio-temporal characteristics of exchanger. The flow in tube bundles is characterized by the
wall pressure fluctuation as well as the wavelength of spanwise three-dimensional, unsteady motion of separated shear layers,
non-uniformity. anisotropic vortices over a wide range of length scales and their
interactions, and a high level of turbulence intensity [5,6], which
Nomenclature are strongly dependent on the tube configuration, arrangement,
Cs Smagorinsky constant and flow condition. These complexities often make a numerical
d tube diameter simulation of the tube bundle flow a challenging task. For
p pressure instance, a Reynolds averaged NavierStokes (RANS) approach
p' pressure fluctuation has been widely used for the prediction of a turbulent flow in
Q second invariant of velocity gradient many practical applications, but it has been recognized that
Re Reynolds number RANS models severely underestimate the turbulence level and
S* pitch-to-diameter ratio may not be suitable for the calculation of a tube bundle flow [9].
Sij strain rate tensor On the other hand, a large eddy simulation (LES) technique has
St Strouhal number the potential to provide accurate predictions of turbulent statistics,
u x-direction velocity component and thus it is now considered as a promising tool for studying the
U mean streamwise velocity fundamental features of a turbulent flow in tube bundles [2,3,5].
U0 inlet velocity In the present study, we numerically investigate the turbulent
v y-direction velocity component cross-flow in an in-line tube bundle, which is composed of 10
w z-direction velocity component rows of rods arranged with a pitch-to-diameter ratio of 1.5. By
x streamwise coordinate means of a commercial computational fluid dynamics (CFD)
xL x-direction distance from tube center code, Fluent 12.0, a three-dimensional LES is performed at a
y transverse coordinate Reynolds number of 27000 based on the inlet velocity and tube
y+ non-dimensional wall distance diameter. The obtained results such as the instantaneous vortex
y first grid spacing at the wall structures, mean velocity profiles, and separation angles are
z spanwise coordinate compared with the experimental data [6] to assess the validity of
the present simulation. The three-dimensional instability
Greeks originated from the streamwise vortices formed in the near-wake
filter width is also discussed in this paper, particularly focusing on the time-
ij Kronecker delta spatial characteristics of wall pressure fluctuation as well as the
wavelength wavelength of spanwise non-uniformity.
kinematic viscosity of fluid
circumferential location
Computational set-up
s separation angle
fluid density Governing equations and numerical methods
ij subgrid-scale stress tensor
z spanwise vorticity In the present study, we consider a three-dimensional, unsteady,
incompressible turbulent flow. Under the assumption of a single-
phase constant-property Newtonian fluid, it is governed by the
Introduction filtered NavierStokes equations as
A turbulent flow in tube bundles has received much attention in a
variety of heat transfer applications. More specifically, a shell- ui
side cross-flow continues to be one of the major concerns in the 0 (1)
design of heat exchangers, steam generators, evaporators, etc. A xi
Figure 1. Schematic of turbulent flow in an in-line tube bundle: Figure 3. Iso-surfaces of normalized second invariant of velocity
x-z (top) and x-y (bottom) planes gradient Q=372 coloured with spanwise vorticity z (21 levels
between -50 and 50)

ui ui u j 1 p 2 ui ij diameter ratio of S*=1.5. In the transverse direction, there are one

(2)
t x j xi xi xi xi full rod and two half-rods of diameter d and span width 3.3d. The
computational domain is extended from 10d upstream to 33.5d
where the overbar denotes the spatial filtering operation. For the downstream in the streamwise direction (or x-direction), and 3d
subgrid-scale (SGS) stresses ij in equation (2), we employ herein in the transverse direction (or y-direction). Note that this is
the classical Smagorinsky model, based on the fact that the almost the same setup as in the experiment of Iwaki et al. [6]
influence of subgrid-scale model is insignificant in case of the except for the number of transverse tubes.
tube bundle flow [10]. The deviatory part of the SGS stress is
As for the boundary conditions, uniform velocity U0 is prescribed
therefore given by
at the inlet, while the outflow boundary condition is imposed at
the outlet. The Reynolds number based on the inlet velocity and
1
ij kk ij 2C s 2 S S ij (3) tube diameter is 27000. The flow periodicity is assumed in both
3 the y- and z-directions. The tube surfaces are treated as stationary
no-slip smooth walls.
where ij is the Kronecker delta, Cs is the Smagorinsky constant
(Cs=0.065), and is the filter width. In equation (3), Sij is the Mesh configuration
strain rate tensor for the resolved scale: A continuous body-fitted mesh is generated by making use of the
geometry and mesh building intelligent toolkit (GAMBIT). As
1 u u j depicted in Figure 2, the computational grids consisting of
S ij i (4)
2 x j xi 4814400 hexahedral volume cells are clustered into tube surfaces,
where 160 and 40 grid points are distributed uniformly in the
circumferential and spanwise directions (or z-direction),
S ij 2S ij S ij
1/ 2
(5) respectively. It should also be noted that the first grid spacing at
the wall is set to be y=0.0008d, and this corresponds to a non-
The governing equations are solved using the commercial CFD dimensional wall distance of y+=1.5.
code, Fluent 12.0, which is based on the finite volume method
(FVM). In the present LES, a segregated and double precision Results and discussion
solver is utilized with semi-implicit methods for pressure linked
Flow characteristics
equations (SIMPLE) algorithm for pressure-velocity coupling [8],
a second-order central differencing method for discretization, and To understand the basic features of a turbulent flow over the in-
a second-order implicit method for time advancement. line tube bundle, instantaneous flow structures are shown in
Figures 3 and 4, using the iso-surface of the second invariant of
Computational domain and boundary conditions
the velocity gradient and streamwise velocity distribution,
Figure 1 illustrates the computational domain and corresponding respectively. It is shown that complex flow phenomena such as
boundary conditions used in our simulation. The in-line tube the interaction of separated shear layer with a downstream row,
bundle is composed of 10 rows of tubes arranged with a pitch-to- three-dimensional vortical structures over a wide range of length

Figure 2. Mesh configuration for LES of turbulent flow through Figure 4. Instantaneous streamwise velocity distributions u/U0
an in-line tube bundle (31 levels between -2 and 4)
 (a) (b) (c)
Figure 5. Comparison of streamwise velocity development U/U0 behind the tubes at (a) the first, (b) second, and (c) third rows: present
solution, measurement by Iwaki et al. [6]

scales, and a high velocity jet behind the last row changing its LES and the measured data is shown to be reasonably good,
direction intermittently are effectively resolved in the LES. Two particularly in the downstream of the second row where the
distinctive flow regions are clearly discernable in the streamwise separation angle does not change significantly. In the experiment
velocity distribution, i.e., the high velocity region in the narrow by Iwaki et al. [6], it was reported that the wake structure behind
passage between adjacent tubes and the recirculation region the first row is much different from the others, leading to
behind the tubes. It is also interesting to note that the increased width of the recirculation region and an upward
recirculation flows differ from row to row. As was shown in the movement of the separation point at the first row. In our
experiment of Iwaki et al. [6], an asymmetric vortex pattern of a simulation, the separation point of the first row is about 90,
large vortex accompanied by a small vortex is most frequently while the separation points at the other rows are in the range of
observed behind the tubes, whereas a symmetric pair of vortices 100~120. This result is consistent with the previous observation
or a single large vortex seldom forms. Furthermore, the vortex
pattern in the wake region is found to be nearly 180 out of phase
with those in the neighboring rows except for the first row. The
wake structures behind the other rows also show similar results.
Comparison with experiment
To further assess the validity of the present LES, Figure 5
compares the time-averaged streamwise velocity (normalized by
U0) distributions behind the first three rows, at several locations
downstream from each tube center. Note that the time-averaged
velocity components in the LES are computed as an average of
the instantaneous values over 40 non-dimensional time (t*=tU0/d),
after the flow reaches a quasi-periodic stage. It can be seen that
the mean velocity profiles agree fairly well with the measurement
[6] in both the inter-tube region and the recirculation region. The
development of a streamwise velocity with distance downstream
(e.g. increase of non-uniformity) is also found to be well
predicted in the present LES.
Figure 6 shows a comparison of the separation points on the
Figure 6. Comparison of separation angle with the experimental
tubes, which are defined by the angle from the front stagnation
data: present solution, measurement of Iwaki et al. [6]
point of each tube. The overall agreement between the present

(a)
(b)
Figure 7. Time histories of spanwise wall pressure fluctuation di-
stribution p'/U02 at (a) =90 and (b) =180 (25 levels between [5] Hassan, Y.A. and Barsamian, H.R., Tube bundle flows with
-1.2 and 1.2)
[6], indicating that the present LES provides reliable predictions
of a turbulent flow across tube bundles.
Spanwise non-uniformity
Figure 7 presents spatio-temporal characteristics of the wall
pressure fluctuation (normalized by U02) at two circumferential
locations up to the fourth row, where is defined by the angle
from the front stagnation point of each tube. At the tube top
(=90) of the first row, the pressure fluctuation exhibits a quasi-
periodic spanwise structure. This non-uniformity has a
wavelength of around one tube diameter /d~1, which is similar
to the mode-B instability of the flow over a single cylinder [7,11].
Further downstream, the spanwise non-uniformity becomes more
chaotic, and the pressure fluctuation increases until a fully
developed state is reached. A similar growth in the pressure
fluctuation is also found at the rear end of the tube (=180).
Moreover, it is interesting to note here that any pressure marks
corresponding to the typical vortex shedding frequency of
St=fd/U0=0.1~0.2 is not discernible at the first four rows. Instead,
the pressure fluctuation associated with the shear layer instability
whose non-dimensional period is approximately 0.65 (or St=1.5)
is noticeable in the present study.
Conclusions
In this study, we numerically investigate a turbulent cross-flow in
an in-line tube bundle consisting of 10 rows of rods arranged
with a pitch-to-diameter ratio of 1.5. With the aid of the
commercial CFD code Fluent 12.0, a large eddy simulation is
performed at a Reynolds number of 27000 based on the inlet
velocity and tube diameter. By comparing the time-averaged
streamwise velocity distributions behind the tubes and the
separation points with the experimental data, it has been shown
that LES provides reliable predictions of a turbulent flow across
the in-line tube bundle. The spatio-temporal characteristics of the
wall pressure fluctuation are also investigated, with a brief
discussion on the spanwise non-uniformity originated from the
streamwise vortices formed in the near-wake. More in-depth
investigations will be pursued in a future study.
Acknowledgments
This study has been performed under a contract with the Korean
Ministry of Educational Science and Technology. The authors
would like to acknowledge the support from KISTI super-
computing center through the strategic support program for the
supercomputing application research [No. KSC-2012-C1-03].
References
[1] Balabani, S., Bergeles, G., Burry, D. and Yanneskis, M.,
Velocity characteristics of the crossflow over tube bundles,
7th International Symposium on Applications of Laser
Anemometry to Fluid Mechanics, 2, 1994, 39.3.139.3.8
[2] Barsamian, H.R. and Hassan, Y.A., Large eddy simulation
of turbulent crossflow in tube bundles, Nuclear Engineering
and Design, 172, 1997, 103122.
[3] Benhamadouche, S. and Laurence, D., LES, coarse LES, and
transient RANS comparisons on the flow across a tube
bundle, International Journal of Heat and Fluid Flow, 24,
2003, 470479.
[4] Gaddis, E.S. and Gnielinski, V., Pressure drop in cross flow
across tube bundles, International Chemical Engineering, 25,
1985, 115.
the large Eddy simulation technique in curvilinear
coordinates, International Journal of Heat and Mass
Transfer, 47, 2004, 30573071.
[6] Iwaki, C., Cheong, K. H., Monji, H. and Matsui, G., PIV
measurement of the vertical cross-flow structure over tube
bundles, Experiments in Fluids, 37, 2004, 350363.
[7] Nakamura, H. and Igarashi, T., Unsteady heat transfer from
a circular cylinder for Reynolds numbers from 3000 to
15000, International Journal of Heat and Fluid Flow, 25,
2004, 741748.
[8] Patankar, S.V., Numerical heat transfer and fluid flow,
Hemisphere Publishing Co., New York, 1981.
[9] Paul, S.S., Ormiston, S.J. and Tachie, M.F., Experimental
and numerical investigation of turbulent cross-flow in a
staggered tube bundle, International Journal of Heat and
Fluid Flow, 29, 2008, 387414.
[10] Rollet-Miet, P., Laurence, D. and Ferziger, J., LES and
RANS of turbulent flow in tube bundles, International
Journal of Heat and Fluid Flow, 20, 1999, 241254.
[11] Williamson, C.H.K., Three-dimensional wake transition,
Journal of Fluid Mechanics, 328, 1996, 345407.
[12] Zukauskas, A. and Ulinskas, R., Banks of plain and finned
tubes, in: Heat Exchanger Design Handbook, Hemisphere
Publishing Co., New York, 1987.