An embedding nite element analysis of heat transfer on the surface

of circular cylinders in ow
D.C. Lo
a,
, Dong-Taur Su
b
a
Institute of Maritime Information and Technology, National Kaohsiung Marine University, Kaohsiung 805, Taiwan
b
Department of Shipping Technology, National Kaohsiung Marine University, Kaohsiung, Taiwan
a r t i c l e i n f o
Article history:
Received 30 March 2012
Received in revised form 26 June 2012
Accepted 2 July 2012
Available online 30 July 2012
Keywords:
Finite element analysis
NavierStokes equations
Convective heat transfer
a b s t r a c t
The present study establishes an embedding nite element method appropriate for solving primitive var-
iable forms of the NavierStokes equations and energy equation in a complex physical domain. The sta-
tionary solid obstacle in the ow domain is embedded in a non-uniform Cartesian grid and the governing
equations are calculated through a nite element formulation. A compact interpolating scheme near the
immersed boundaries is used to ensure the accuracy of the solution in the cut cells. We have developed a
numerical algorithm based on the operator splitting technique, balance tensor diffusivity (BTD), Runge
Kutta time-stepping method, and a bi-conjugate gradient iterative solver. Three numerical examples are
chosen to test the accuracy and exibility of the proposed scheme. Simulation of ow past a stationary
circular cylinder is conducted to validate the accuracy of the present method for solving heat transfer
problems. Flow over circular cylinders in a tandem arrangement and a staggered tube bank with convec-
tive heat transfer is computed to demonstrate the models ability to handle complex geometries.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
This paper presents an embedding nite element analysis of
heat transfer problems with uid structure interactions. In the gov-
erning equations used to illustrate incompressible viscous ows
with force-convection problems, the effect of buoyancy and com-
pressibility are neglected. Thus, the momentum and continuity
equations and energy equation are decoupled and only one-way
interaction from the governing equations is considered. This is
because the NavierStokes equations are coupled through the
velocity, but the pressure does not appear in the continuity equa-
tion. To resolve this, we adopt a decoupled numerical algorithm
based on the projection method to overcome the difculties of
coupling between the continuity and momentum equations. The
projection method has been pioneered by Chorin [1] to represent
the NavierStokes equations for the solution of multi-dimensional
ow problems. This projection method, also called the fractional
step method, has been studied over the past years by many inves-
tigators, who have examined this method by using approximations
of the advection term and the scheme used for the time discretiza-
tion [26]. The main advantage of the projection is that the incom-
pressibility conditions yield an approximate velocity eld, which
was made divergence-free from an orthogonal projection of the
apparent velocity eld. The method has been veried to simulate
complex ows and accordingly is widely used in the eld of com-
putational uid dynamics.
In this study, we have taken advantage of the operator splitting
technique, balance tensor diffusivity (BTD), and RungeKutta time-
stepping method based on the projection method for solving 2D
viscous ow problems with heat transfer. To reduce computational
effort, the system of the equations is solved simultaneously by Ja-
cobi iteration using the mass lumping technique, thus avoiding the
formulation of global matrices. Only the pressure Poisson equation
is solved by using a bi-conjugate gradient iterative scheme [7], and
this is done by storing the non-zero values during calculations. A
bi-conjugate gradient iterative scheme is adopted to solve the
simultaneous equations to obtain the solution at the global nodes.
The implementation of the present iterative solution procedure to
solve CFD problems on a personal computer is straightforward be-
cause the scheme has been efciently implemented to accommo-
date the requirement of small computer memory to store the
non-zero entries of the matrices.
Cartesian meshes have commonly been used for solving prob-
lems with irregular geometry through the body-tted or unstruc-
tured grid methods. But the computational savings of body-tted
or unstructured nite element meshes become a huge challenge
when solving problems with a moving boundary. Our method, in
contrast, uses the generation of a structural Cartesian grid rather
than that required by body-tted or unstructured nite element
meshes. On the other hand, the underlying Cartesian mesh is used
as a powerful tool for saving computational time. In the past
0017-9310/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2012.07.005

Corresponding author.
E-mail address: loderg@webmail.nkmu.edu.tw (D.C. Lo).
International Journal of Heat and Mass Transfer 55 (2012) 69166926
Contents lists available at SciVerse ScienceDirect
International Journal of Heat and Mass Transfer
j our nal homepage: www. el sevi er . com/ l ocat e/ i j hmt
several decades the immersed boundary method has become more
popular, following its introduction by Peskin [8], due to its simplic-
ity and exibility in solving moving complex boundary problems
with less computational cost and memory requirements than other
methods.
Mittal and Iaccarino [9] review important claims that the IB
method to simulate incompressible viscous ows can be classied
into two major categories, namely continuous forcing [8,1012]
and discrete forcing [1315]. An early development in the history
of the continuous forcing IB method was the note by Peskin [8] that
the NavierStokes should add a forcing term before the discretiza-
tions. A number of studies [1012,16,17] have investigated for the
modications and improvements of this method since Peskins
paper [8]. In the continuous forcing IB method, the boundaries
are modeled as pure force-generators for the simulation of the
NavierStokes equations related to xed or moving obstacles inter-
acting with a uid. It has been argued that the forcing term appear-
ing in the discrete forcing method can be either explicitly or
implicitly applied to the discretization of NavierStokes [1315].
In this paper, we present numerical solution algorithm to solve
viscous ows with heat transfer in combination with the nite ele-
ment method and an embedding method based on an interpolation
scheme to handle the immersed boundary in the complex geome-
try. The key issue in using Cartesian grid methods is the imposition
of boundary conditions at the immersed boundary. To resolve this,
we adopt a compact interpolation scheme near the immersed
boundaries that allows us to retain second-order accuracy of the
solver. In the presence of the ow past several cylinders with an
arbitrary arrangement, the values of the ow variables at the
immersed cells are uploaded using a local reconstruction
scheme involving the virtual point and the projection node at the
immersed boundary. A second order interpolation scheme is used
to solve the values of the variables at the immersed cells. Thus
the use of nite element method has enabled us to solve a mesh
generation problem of great complexity. Finite element method
is found to be an effective tool to compute the ow variables
due to its good exibility and stability. We should treat carefully
with the immersed cells and boundary condition to ensure a
conservative solution. Thus the present numerical procedure
exploits the advantages of both the interpolating scheme at the im-
mersed cells and the nite element method in the computational
domain.
Numerous experimental and numerical studies have conducted
heat-transfer over a stationary circular cylinder. Churchill and
Bernstein [18], Eckert and Soehngen [19], and Roshko [20] pro-
vided extensive discussions of the experimental results from these
studies. Using numerical methods, Momose and Kimoto [21] and
Bharti et al. [22] analyzed out the heat transfer over a stationary
cylinder at a Reynolds-number range similar to the present study.
Since the IB method is concerned, Yoon et al. [23] investigated IB
nite volume method for 2D laminar uid ow and heat transfer
past a circular cylinder near a moving wall. Recently, Kim et al.
[24] studied the natural convection induced by the temperature
difference between a cold outer square enclosure and a hot inner
circular cylinder using an IB method. Feng and Michaelides [25]
applied the IB method with a difference method to solve the ther-
mal convection in particulate ows. Wang et al. [17] presented a
direct source scheme to the simulations of natural convection be-
tween concentric cylinders, and analyzed the ow past a stationary
circular cylinder to validate the accuracy of the present method for
solving heat transfer problems.
The ow past two cylinders is a signicant topic of uidstruc-
ture interaction. The arrangement of the cylinders vs. the ow
direction of the free stream is used to simulate the hydrodynamic
interaction between two cylinders. The arrangement of the cylin-
ders with respect to the free stream ow direction can be classied
into three major categories, namely tandem the free stream ow
direction is parallel with the line of the centers of the cylinders,
transverse the free stream ow direction is perpendicular to
the line of the cylinders centers, staggered. The ow past a pair
of cylinders in tandem arrangement have been thoroughly investi-
gated numerically in [2632]. To our knowledge, there is less pub-
lication in respect of the isothermal ow past two cylinders in
tandem [3335]. Buyruk [33] used numerical method to study
the force convection heat transfer for tandem, in-line and stag-
gered cylinders congurations. Juncu [34,35] presents a computa-
tional study of the steady and viscous ow around two tandem
cylinders. He reported results for force convection heat transfer
around two tandem circular cylinders at low Reynolds numbers
using a compact nite difference method, Reynolds number vary-
ing from 1 to 30 and uid phase Prandtl number equal to 0.1, 1,
10 and 100. The present study is emphasized to the analysis of
forced convection heat transfer from two tandem cylinders in an
unsteady and viscous ow.
The contents of this paper are organized as follows. Section 2
revisits the governing equations for 2D viscous ow with forced
convection. Section 3 delineates the nite element scheme based
on projection method to discretize the governing NavierStokes
equations in primitive variables form. In Section 4, numerical re-
sults for three examples are examined: the effect of heat transfer
phenomena for ow past a circular cylinder, two cylinders in a tan-
dem arrangement, and ow past a staggered tube bank which is
composed by 77 circular cylinders with heat transfer. Concluding
remarks are presented in Section 5.
2. Governing equations
The governing equations for viscous ow can be expressed by
means of the non-dimensional form of the NavierStokes equa-
tions with forced convection in primitive variables form as:
Continuity equation
@u
@x

@v
@y
0 1
Momentum equations
@u
@t
u
@u
@x
v
@u
@y

@p
@x

1
Re
r
2
u 2
@v
@t
u
@v
@x
v
@v
@y

@p
@x

1
Re
r
2
v 3
Energy equation
@T
@t
u
@T
@x
v
@T
@y

1
Pe
r
2
T 4
for a Cartesian coordinate frame in which xy represents the hori-
zontal and the vertical direction. The following parameters are used
to dene the non-dimensional quantities in the above governing
equations and later physical explanations. u, v are the velocity vari-
ables in the x-, y-direction, respectively. p is the pressure, Re is the
Reynolds number, Pe
DU
k
RePr is the Peclet number, Pr denotes
the Prandtl number.
3. Finite element scheme
In order to discretize the time derivative of the convection
equation, consider the following equation,
@/
@t
u
@/
@x
v
@/
@y
0 5
D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926 6917
The rst order explicit time discretization is involved to discretize
the above equation as
/
n1
/
n
Dt u
n
@/
@x
n
v
n
@/
@y
n
_ _
6
where n + 1 represents the present time index level, n is the previ-
ous time index level.
According to the Taylor series expansion, the time derivative is
discretized as follows
/
n1
/
n
Dt
@/
@t
n

Dt
2
2
@/
@t
2
n
H:O:T 7
We combine Eqs. (5) and (7), and then the balance tensor diffusion
(BTD) may be included in the convection equation by explicit time
discretization as
/
n1
/
n
Dt u
n
@/
@x
n
v
n
@/
@y
n
_ _

Dt
2
2
@
@x
u
n

2
@/
@x
n
_ _ _

@
@x
u
n
v
n

@/
@y
n
_ _

@
@y
v
n
u
n

@/
@x
n
_ _

@
@y
v
n

2
@/
@y
n
_ __
8
where
BTD
Dt
2
2
@
@x
u
n

2
@/
@x
n
_ _

@
@x
u
n
v
n

@/
@y
n
_ _

@
@y
v
n
u
n

@/
@x
n
_ _ _

@
@y
v
n

2
@/
@y
n
_ __
Similarly, we employ the above formula to discretize the Navier
Stokes equations with an operator splitting scheme. Additionally,
the balancing tensor diffusivity also can be derived in the Eqs.
(2)(4). Eqs. (2)(4) can be written in a segregated form as
u
n1
u
n
Dt u
n

@u
n
@x
v
n

@u
n
@y

@p
n
@x

1
Re
r
2
u BTDu
n

_ _
9
v
n1
v
n
Dt u
n

@v
n
@x
v
n

@v
n
@y

@p
n
@y

1
Re
r
2
v BTDv
n

_ _
10
T
n1
T
n
Dt u
n

@T
n
@x
v
n

@T
n
@y

1
Re
r
2
T
n
BTDT
n

_ _
11
where
BTDu
n

Dt
2
@
@x
u
2
@T
n
@x
_ _

@
@x
uv
@T
n
@y
_ _

@
@y
vu
@T
n
@x
_ _ _

@
@y
v
2
@T
n
@y
_ __
BTDv
n

Dt
2
@
@x
u
2
@T
n
@x
_ _

@
@x
uv
@T
n
@y
_ _

@
@y
vu
@T
n
@x
_ _ _

@
@y
v
2
@T
n
@y
_ __
BTDT
n

Dt
2
@
@x
u
2
@T
n
@x
_ _

@
@x
uv
@T
n
@y
_ _

@
@y
vu
@T
n
@x
_ _ _

@
@y
v
2
@T
n
@y
_ _
Combing the continuity equation (1) and momentum equations (2)
and (3), the pressure Poisson equation can be derived to correct the
velocity equation as
r
2
p
n1

1
Dt
@u
n1
@x

@v
n1
@y
_ _
12
The u

, v

are the middle velocity variables. The pressure boundary

conditions can be written as
@p
@n

1
Re
~n r
2
~u ~n
~
f
In the present model, the pressure Poisson equation is solved using
FEM and bi-conjugate gradient iterative solver. The middle and real
velocity values are solved using the Jacobi iteration technique.
Additionally, the Jacobi iteration technique is effective for handling
time derivative equations. The spatial discretization of the equa-
tions is performed by the standard Galerkin nite element method
using four-node bilinear elements. The resulting nite element
equations can be solved as following procedure:
For step 1,
M
ij
u
n1
j
M
ij
u
n
j
Dt C
ij

1
Re
K
ij


K
ij
_ _
u
n
j
DtF
u
i
13
M
ij
v
n1
j
M
ij
v
n
j
Dt C
ij

1
Re
K
ij


K
ij
_ _
v
n
j
DtF
v
i
14
For step 2,
K
ij
P
n1
j
G
x
ij
u
n1
j
G
y
ij
v
n1
j
_ _
15
For step 3,
M
ij
u
n1
j
M
ij
u
n1
j
DtG
x
ij
P
n1
j
16
M
ij
v
n1
j
M
ij
v
n1
j
DtG
y
ij
P
n1
j
17
For step 4,
M
ij
T
n1
j
M
ij
T
n
j
Dt C
ij

1
Pe
K
ij


K
ij
_ _
T
n
j
DtF
T
i
18
where
M
ij

__
N
i
N
j
dX
C
ij

__
u
n
N
i
@N
j
@x
v
n
N
i
@N
j
@y
_ _
dX
G
x
ij

__
N
i
@N
j
@x
dX
G
y
ij

__
N
i
@N
j
@y
dX
K
ij

__
@N
i
@x
@N
j
@x

@N
i
@y
@N
j
@y
_ _
dX

K
ij

Dt
2
__
@N
i
@x
u
2
@N
j
@x

@N
i
@x
uv
@N
j
@y

@N
i
@y
vu
@N
j
@x
_

@N
i
@y
v
2
@N
j
@y
_
dX
F
u
i

_
N
i
@u
n
@x
n
x

@u
n
@y
n
y
_ _
dC
F
v
i

_
N
i
@v
n
@x
n
x

@v
n
@y
n
y
_ _
dC
F
T
i

_
N
i
@T
n
@x
n
x

@T
n
@y
n
y
_ _
dC
in which N
i
and N
j
are the shape functions.
After assembling the system and applying the initial and
boundary conditions, as described in Eqs. (13), (14) and (16)
(18), the system of the equations is solved by the Jacobi iteration.
In the Jacobi iteration, the linear system of equations is solved
iteratively using the values from the previous iteration. If a linear
6918 D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926
system of equations of the form MX = B is to be solved, the system
is modied as
MX
k1
B M MX
k
19
where M is the consistent coefcient matrix, X is the unknown vec-
tor, B is the known vector, k is the iteration number and M is a
lumped matrix. The pressure Poisson equation (15) is solved by
using the bi-conjugate gradient iterative solver. It is should be
noted that only the nonzero entries of the coefcient matrices of
the unknown variables of pressure eld are evaluated and stored
in a compressed column-vector storage form. Then a bi-conjugate
gradient iterative scheme is exploited to solve the nal equations.
Once the velocity and temperature elds are obtained, the local
and surface-averaged Nusselt numbers on the cylinder are dened
as
Nu
@T
@n
; Nu
1
L
_
L
0
Nuds 20
where n denotes the direction with respect to the walls, L denotes
the surface area of cylinder.
3.1. Inclusion of 2D Immersed boundaries
The existing interpolation functions have been widely used in
the immersed boundary and the immersed nite element methods
[1720], such as the discretized Dirac delta function [21] and the
reproducing kernel interpolation function [22,23]. This subsection
discusses the underlying approach for a Cartesian grid with im-
mersed boundaries. The immersed boundary is represented by
piecewise linear segments. Here we propose how this approach
is implemented in a situation where some of the Cartesian cells
are cut by immersed boundaries. The computational domain is di-
vided into two regions: the physical domain and the immersed cell
Fig. 1. Dark circles are the nodes of immersed cells. The black squares are projected
by associating with each dark circle as the interface point. The light circles are
extrapolation nodes by dark circles and black squares.
Fig. 2. Schematic of computational domain with immersed boundary.
0 20 40 60 80 100 120 140 160 180
0
4
8
12
Present, Isothermal
Kim & Choi(2004)
Zhang et al.(2008)
Deg.
u
Deg.
Nu
0 20 40 60 80 100 120 140 160 180
0
1
2
3
4
5
Present, Isothermal
Bharti et al.(2007)
Zhang et al.(2008)
Deg.
u
Deg.
Nu
(a)
(b)
Fig. 3. Comparison of local Nusselt number distributions on the cylinder surface
with isothermal boundary for (a) Re = 20 and (b) Re = 120.
D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926 6919
domain. The immersed cells were consisting of the uid and solid
adjacent to uid cell and solid cell in the computational domain, as
illustrated in Fig. 1. The physical domain can be easily imple-
mented because it does not consider the immersed cell. The values
of the ow variables at the immersed cells are computed using a
local reconstruction scheme involving the virtual point and the
projection node at the immersed boundary. We identify cells in
the underlying Cartesian mesh that are cut by the immersed
boundary and compute the intersection of the immersed boundary
with the sides of these cut cells. For the purpose of this discussion,
we derive one situation for the interpolation scheme, as shown in
Fig. 2(a). This gure shows that nodal point 1 external to the im-
mersed boundary is represented by the values of the ow vari-
ables, in which the desired value of node 1 is evaluated using an
interpolating scheme along the local normal to the immersed
boundary. The value of virtual nodal point v can be performed in
a compact form through four nodal points (14). Thus, the value
of /
1
can be interpolated using an immersed boundary value /
b
and virtual point value /
v
. Here, the value of /
v
is xed in the
virtual point. We should project the nodal point 1 into the im-
mersed boundary and nd an extrapolation virtual point, which
three points (b, 1, v) are the connection with a straight line and
are perpendicular to the immersed boundary. A bilinear interpola-
tion 2D shape function can be expressed as
/
v
N
1
/
1
N
2
/
2
N
3
/
3
N
4
/
4
21
In particular, the value of /
1
can be calculated by the linear approx-
imation at the two points b and v:
/
1

L
2
L
/
b

L
1
L
/
v
22
where L
1
is the length between b and 1, L
2
is the length between 1
and v, L is the length equal to L
1
+ L
2
, as illustrated in Fig. 2(a).
A compact interpolating function can be obtained by combining
the above two equations (21) and (22) as
L
L
1
N
1
_ _
/
1
N
2
/
2
N
3
/
3
N
4
/
4

L
2
L
1
/
b
23
where N
1
, N
2
, N
3
, N
4
are the bilinear shape function.
The above equation can be used as a constraint and alters the
underlying matrix operator in the rows corresponding to the
boundary cell. We may substitute Eq. (23) into the underlying ma-
trix operator by using an implicit or explicit scheme. If an implicit
scheme is used, the unknown variables are /
1
, /
2
, /
3
, /
4
. However,
we should use an iterative scheme to obtain a convergence state
during each computational stage.
4. Numerical results
In this section, heat convection with ow over one, two and 77
stationary cylinders is selected as the target cases. The simulation
results are compared with results from previous studies in the lit-
erature. The isothermal boundary condition on the circular cylin-
der is used in the present study. Since the parametric value is
concerned, the Reynolds number range is from 20 to 200. The Pra-
ndtl number is xed at 0.7 for all computational cases. Circular cyl-
inder diameter D is 1, T = 1 on the cylinder surface is specied for
isothermal cases, and the free-stream velocity U = 1.
4.1. Flow past a circular cylinder with heat transfer
Here we have an isothermal circular cylinder embedded in a do-
main. Flow in the upstreamis assumed to be uniform ow given by
u = U. The computational domain equal to 30D 15D with a circu-
lar cylinder. To vary the Reynolds number (Re = UD/t), we hold the
0 20 40 60 80 100 120 140 160 180
0
4
8
12
16
Re=20
Re=40
Re=100
Re=120
Re=200
Deg.
u
Deg.
Nu
Fig. 4. Comparison of local Nusselt number distributions on the cylinder surface
with Re = 20, 40, 100, 120 and 200, respectively.
Fig. 5. Computational domain and boundary conditions for ow over a pair of cylinders with heat transfer.
6920 D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926
velocity xed at U = 1 and vary t. Fig. 3 shows the comparisons of
the local Nusselt number distributions on the cylinder surface with
isothermal boundary for Re = 20 and Re = 120. The grid sizes were
taken as 0.02 near the cylinder, and the physical space consisted of
60,501 grids. It can be observed that the results obtained by using
the present numerical algorithm are almost in good agreement
with the results of Bharti et al. [22], Zhang et al. [16] at Re = 20
and kim and Choi [36] and Zhang et al. [16] at Re = 120 considered
in this study. Fig. 4 provides the list of local Nusselt number distri-
butions on the cylinder surface with Re = 20, 40, 100, 120 and 200,
respectively. The dened angle (h) of circular cylinder is shown in
Fig. 4. For the case of Re = 20 and 40, Fig. 4 presents the poor heat
transfer and forced convection phenomenon to the rear of the cyl-
inder on the surface. This is clearly indicated by the decreasing
trend of Nusselt number values with increase in the angle, as can
be seen from Fig. 4 for the cases of the Reynolds numbers to be
Table 1
Drag coefcient and Strouhal numbers for cylinder in tandem arrangement at L = 3D
and Re = 200.
L = 3D Meneghini
et al. [29]
Noor et al. [32] Present Dh
1
30
Present Dh
1
40
C
D1
1.0 0.992 1.04 1.02
C
D2
0.08 0.11 0.12 0.12
S
t
0.125 0.127 0.126 0.126
L=1.5D
L=3D
0 20 40 60 80 100 120 140 160 180
0
4
8
12
16
Re=200
Re=100
Re=60
Re=40
Re=20
Deg.
u
Deg.
Nu
0 20 40 60 80 100 120 140 160 180
0
4
8
12
16
Re=200
Re=100
Re=60
Re=40
Re=20
Deg.
u
Deg.
Nu
0 20 40 60 80 100 120 140 160 180
0
2
4
6
Re=200
Re=100
Re=60
Re=40
Re=20
Deg.
u
Deg.
Nu
0 20 40 60 80 100 120 140 160 180
0
2
4
6
8
10
Re=200
Re=100
Re=60
Re=40
Re=20
Deg.
u
Deg.
Nu
Fig. 6. Comparison of local Nusselt numbers for cylinder 1 (left side) and cylinder 2 (right side) on the surface for various L and Re.
D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926 6921
20 and 40. The minimum in value of the local Nusselt number
along with the cylinder on the surface at angles from 130 to 140
is obtained for the cases of Re = 100, 120 and 200. We may observe
that the local Nusselt numbers at the very front of the cylinder
(stagnation point) on the surface vary to a maximum value for
all the cases considered in the study. The above results obtained
demonstrate the efciency and robustness of the present method
to solve the NavierStokes equations and energy equation by an
embedding nite element method.
4.2. Flow past a pair of tandem cylinders with heat transfer
Apart from testing the code for the present formulation with re-
spect to the ow past a cylinder, the second example has been
developed to validate the model for the solution of ow past a pair
of cylinders in the tandem arrangement with heat transfer. Two
cylinders are assumed with isothermal on the surface with con-
stant temperature to be 1 (dimensionless). Fig. 5 shows a layout
of uniform ow over the pair cylinders with heat transfer. The
computational domain and boundary conditions for ow over a
pair cylinders with heat transfer are depicts in Fig. 5. The congu-
ration of a pair of cylinders was placed inside the ow domain in
0 6 x 6 30D and 7.5D 6 y 6 7.5D. This case was used to test the
behavior of the scheme on a non-uniformgrids generated by a sim-
ple node generator in a complicated domain. For the present cases,
we use a nonuniform grid of size 301 201 to obtain the ow and
temperature patterns. From the viewpoint of the traditional meth-
od to produce unstructured meshes or generate body-tting
meshes by coordinate transformation, our proposed method is a
simple way to handle the mesh generation. In the present study,
the two tandem cylinders were arranged with a streamwise gap
of L. For the tandem arrangement, the distance between the cen-
ters of two identical cylinders with L was varied from 1.5D, 2D
and 3D, respectively. The implementation of the initial and bound-
ary conditions are the same as those in the case for the ow around
a circular cylinder. The denition of the Reynolds number is also
based on the diameter of the cylinder. A pair of cylinders of
diameter D = 1 was placed inside the ow domain with the center
of the rst cylinder located at x = 7D and y = 0, and the location of
the second cylinder depending on various L. We simulated a series
of cases for various streamwise gaps (L) to be equal to 1.5D, 2D and
3D. For the L = 3D, the drag coefcient of upstream cylinder in-
creases and the drag of the downstream cylinder become positive
is shown in Table 1. The present simulations have good agreement
with those obtained in the literatures [29,32]. Here we consider the
two streamwise gaps of L = 1.5D and L = 3.0D, the distribution of lo-
cal Nusselt number of the cylinder 1 (left-hand side) and cylinder 2
(right-hand side) on the surface for Re = 20 and 40 was shown in
Fig. 6. As shown in cylinder 1, the maximum value of the Nusselt
number appeared at the front of the cylinder on the surface is ob-
served. Notably, cylinder 2 on the surface has the minimum value
0 20 40 60 80 100 120 140 160 180
0
4
8
12
16
Cylinder 1
Cylinder 2
Deg.
u
Deg.
Nu
Fig. 7. Comparison of local Nusselt numbers on the cylinder surface (cylinder 1 vs.
cylinder 2) for Re = 200 and L = 3D.
0 100 200 300 400
0
4
8
12
16
20
L=2D
L=3D
t
Nu
Cylinder 2
0 100 200 300 400
0
4
8
12
16
20
L=2D
L=3D
t
Nu
Cylinder 1
Fig. 8. Comparison of mean Nusselt numbers on the cylinder surface for Re = 200
and (L = 2D vs. L = 3D).
6922 D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926
of Nusselt number at the front of the cylinder at angle equal to 0.
The maximum Nusselt number with cylinder 2 on the surface is
at an angle of approximately 7080, as presented in Fig. 6. Fig. 7
presents a comparison of cylinder 1 vs. cylinder 2 for Re = 200
and L = 3D concerning the local Nusselt number on the cylinder
surface. At the front of the cylinder is an angle dened to be equal
Fig. 9. Temperature contours for various L at Re = 200 and t = 400.
D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926 6923
to 0, the cylinder 1 has a maximum value for the Nusselt number,
however cylinder 2 on the surface produces a minimum Nusselt
number. At Re = 200, the mean Nusselt number on the cylinder
surface for L = 2D vs. L = 3D is shown in Fig. 8. The state steady
solution of a mean Nusselt number is obtained with the cylinder
1 on the surface. However, we observe that the mean Nusselt num-
ber for the case of cylinder 2 and L = 3D oscillates in periodic mo-
tion with time, and the pattern of mean Nusselt number maintain a
SINE function with the time series. The present numerical study
clearly predicts the evolution of the mean Nusselt number on the
cylinder surface. Fig. 9 presents a comparison of various stream-
wise gaps varying from 1.5D, 2D and 3D at Re = 200 and t = 400
concerning the temperature distributions. A conspicuous shedding
generated at the back of the cylinder 2 is clearly observed in the
present model. However, no oscillatory heat transfer pattern was
found between the gap of the cylinder 1 and cylinder 2. Because
the gap is short enough to restrain the effect of vortex shedding.
The interference between cylinders at close proximity dramatically
changes the ow around them. It may produce unexpected forces
and pressure distributions, and the resulting forces and vortex
shedding becomes intensied or suppressed. It can be observed
that when more than one bluff body is placed in a uid ow, the
resulting heat transfer and vortex shedding phenomena are com-
pletely different from a with a single body at the same Reynolds
number.
4.3. Flow past 77 staggered cylinders with heat transfer
To verify the capability of the model to handle complex geom-
etries, ow over 77 circular cylinders was simulated to capture the
heat transfer phenomena. For this we have a computational do-
main with 77 circular cylinders embedded in it. The Reynolds
number for this example is 20, according to the average diameter
of the bodies. The computation of the dimensionless units of the
channel length is between 022 in the x-direction, 018 in the
y-direction, and the distances between two rows in streamwise
direction and transverse direction are the same which is 2. A total
number of 158,400 elements and 159,201 nodes (441 and 361 uni-
form divisions in the x, and y directions, respectively) with a time
step Dt = 0.005 are used. The uniform mesh distribution is used in
the whole computational domain. Fig. 10 displays temperature dis-
tributions for the case of ow past 77 cylinders with heat transfer
at Re = 20. Table 2 depicts the comparison of the present average
Nusselt number with the empirical corrections [37,38] and numer-
ical computation [17] for Re = 20. The present results are in close
agreement with the results of the numerical scheme [17] which
adopted an immersed boundary method. However, the average
Nusselt number on the isothermal wall have been predicted about
10% higher than the result of Grimison [37] and about 18% higher
than the result of Zhukauskas [38]. A partial explanation for this
may lie in the fact that, in the computation of the present case fully
nonlinear NavierStokes equations are considered. However,
Grimison [37] and Zhukauskas [38] derived an empirical correction
for the case of ow past a staggered tube bank with heat transfer.
Fig. 11 presents the u-contour and v-contour behavior of the steady
state solution. Fig. 12 gives the local velocity distribution of the
upstream and downstream. As we can see, the wake that formed
the cylinder surface is symmetric regarding the direction of ow
at Re = 20. The interference between two rows in streamwise
direction and transverse direction of cylinders dramatically
Fig. 10. Temperature contours for the case of ow past a staggered tube bank with
heat transfer at Re = 20.
Table 2
Comparison of the average Nusselt number.
Author Grimison [37] Zhukauskas [38] Wang et al. [17] Present
hNui 3.71 3.45 4.08 4.07
Fig. 11. The distribution of u- and v-contours at Re = 20.
6924 D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926
changes the ow around them. These results obtained indicate that
the model has been successfully applied to ow past 77 cylinders
with heat transfer, which demonstrates its capability to deal with
2D complex geometries.
5. Conclusions
In this paper, two-dimensional viscous ow problems with heat
transfer are solved by the FEM, which is based on the projection
method, which is the robust and effective scheme for application
in NavierStokes equations and energy equation simulation. The
principle is applied by embracing the operator splitting technique,
balance tensor diffusivity (BTD), RungeKutta time-stepping meth-
od, and a bi-conjugate gradient iterative solver. The numerical pro-
cedure developed in the present work is validated by comparing
the results obtained with the benchmarked problem of a ow over
one, two and 77 stationary circular cylinders. In addition, we have
taken advantage of a bi-conjugate iterative solver for the pressure
Poisson equation by storing non-zero entries of the matrices dur-
ing computations. Moreover, the mass lumped technique is further
utilized to deal with the rst order time derivative equations by Ja-
cobi iteration, eluding the linear system of equations solver. The
excellent agreement in the validation problems and the reasonable
performances of the other cases demonstrate that the present
scheme can be exploited for viscous ow problems with heat
transfer, not only with the ow past over one cylinder but also
with ow over several bodies. Therefore, it is clear that the present
study provides a basis for the FEM formulation to simulate 2D vis-
cous ow problems with heat transfer.
Acknowledgment
The support by the National Science Council, Taiwan under the
Grant No. NSC 100-2221-E-022-015 for the work is gratefully
acknowledged.
References
[1] A.J. Chorin, Numerical solution of the NavierStokes equations, Math. Comput.
22 (1968) 745762.
[2] J. Kim, P. Moin, Application of the fractional-step method to the incompressible
NavierStokes equations, J. Comput. Phys. 59 (1985) 308323.
[3] P.M. Gresho, On the theory of semi-implicit projection methods for viscous
incompressible ow and its implementation via a nite element method that
also introduces a ready consistent mass matrix. Part 1: Theory, Int. J. Numer.
Methods Fluids 11 (1990) 587620.
[4] D. Brown, R. Cortez, M. Minion, Accurate projection method for incompressible
NavierStokes equations, J. Comput. Phys. 168 (2001) 464499.
[5] W. Chang, F. Giraldo, B. Perot, Analysis of an exact fractional step method, J.
Comput. Phys. 180 (2002) 183199.
[6] D.C. Lo, D.L. Young, Y.C. Lin, Finite-element analysis of 3-D viscous ow and
mixed convection problems by the projection method, Numer. Heat Transfer A
48 (2005) 339358.
[7] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical Recipes in
Fortran 90, second ed., Cambridge University Press, New York, 1996.
[8] C.S. Peskin, Flow patterns around heart valves: a numerical method, J. Comput.
Phys. 10 (1972) 252271.
[9] R. Mittal, G. Iaccarino, Immersed boundary methods, Annu. Rev. Fluid Mech. 37
(2005) 239261.
[10] C.S. Peskin, B.F. Printz, Improved volume conservation in the computation of
ows with immersed elastic boundaries, J. Comput. Phys. 105 (1993) 3346.
[11] S. Xu, Z.J. Wang, An immersed interface method for simulating the interaction
of a uid with moving immersed boundaries, J. Comput. Phys. 216 (2006) 454
493.
[12] M.C. Lai, C.S. Peskin, An immersed boundary method with formal second-order
accuracy and reduced numerical viscosity, J. Comput. Phys. 160 (2000) 705
719.
[13] H.S. Udaykumar, R. Mittal, W. Shyy, Computation of solidliquid phase fronts
in the sharp interface limit on xed grids, J. Comput. Phys. 153 (1999) 535
574.
[14] T. Ye, R. Mittal, H.S. Udaykumar, W. Shyy, An accurate Cartesian grid method
for viscous incompressible ows with complex immersed boundaries, J.
Comput. Phys. 156 (1999) 209240.
[15] J. Kim, D. Kim, H. Choi, An immersed boundary nite-volume method for
simulations of ows in complex geometries, J. Comput. Phys. 171 (2001) 132
150.
[16] N. Zhang, Z.C. Zheng, S. Eckels, Study of heat-transfer on the surface of a
circular cylinder in ow using an immersed-boundary method, Int. J. Heat
Fluid Flow 29 (2008) 15581566.
[17] Z. Wang, J. Fan, K. Luo, K. Cen, Immersed boundary method for the simulation
of ows with heat transfer, Int. J. Heat Mass Transfer 52 (2009) 45104518.
[18] S.W. Churchill, M. Bernstein, A correlating equation for force convection from
gases and liquids to a circular cylinder in crossow, J. Heat Transfer 99 (1977)
300306.
[19] E.R.G. Eckert, E. Soehngen, Distribution of heat transfer coefcients around
circular cylinders in crossow at Reynolds numbers from 20 to 500, Trans.
ASME 75 (1952) 343347.
[20] A. Roshko, Experiments on the ow pass a circular cylinder at vary high
Reynolds number, J. Fluid Mech. 10 (1961) 345356.
[21] K. Momose, H. Kimoto, Forced convection heat transfer from a heated circular
cylinder with arbitrary surface temperature distributions, Heat Transfer
Asian Res. 28 (1999) 484499.
[22] R.P. Bharti, R.P. Chhabra, V. Eswaran, A numerical study of the steady forced
convection heat transfer from an unconned circular cylinder, Heat Mass
Transfer 43 (2007) 639648.
[23] H.S. Yoon, J.B. Lee, H.H. Chun, A numerical study on the uid ow and heat
transfer around a circular cylinder near a moving wall, Int. J. Heat Mass
Transfer 50 (2007) 35073520.
Fig. 12. Comparisons of local velocity distribution of the upstream and
downstream.
D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926 6925
[24] B.S. Kim, D.S. Lee, M.Y. Ha, H.S. Yoon, A numerical study of natural convection
in a square enclosure with a circular cylinder at different vertical locations, Int.
J. Heat Mass Transfer 51 (2008) 18881906.
[25] Z.G. Feng, E.E. Michaelides, Heat transfer in particulate ows with direct
numerical simulation, Int. J. Heat Mass Transfer 52 (2009) 777786.
[26] J. Li, A. Chambarel, M. Donneaud, R. Martin, Numerical study of laminar ow
past one and two circular cylinders, Comput. Fluids 19 (1991) 155
170.
[27] A. Slaouti, P.K. Stansby, Flow around two circular cylinders by the random
vortex method, J. Fluids Struct. 6 (1992) 641670.
[28] S. Mittal, V. Kumar, A. Raghuvanshi, Unsteady incompressible ows past two
cylinders in tandem and staggered arrangements, Int. J. Numer. Methods
Fluids 25 (1997) 13151344.
[29] J.R. Meneghini, F. Saltara, C.L.R. Siqueira, J.A. Ferrari Jr., Numerical simulation
of ow interference between two circular cylinders in tandem and side-by-
side arrangements, J. Fluids Struct. 15 (2001) 327350.
[30] S. Mittal, V. Kumar, Flow-induced oscillations of two cylinders in tandem and
staggered arrangements, J. Fluids Struct. 15 (2001) 717736.
[31] W. Jester, Y. Kallinderis, Numerical study of incompressible ow about xed
cylinder pairs, J. Fluids Struct. 17 (2003) 561577.
[32] D.Z. Noor, M.J. Chern, T.L. Horng, An immersed boundary method to solve
uidsolid interaction problems, Comput. Mech. 44 (2009) 447453.
[33] E. Buyruk, Numerical study on heat transfer characteristics on tandem
cylinders, inline and staggered tube banks in cross-ow of air, Int. Commun.
Heat Mass Transfer 29 (2002) 355366.
[34] G. Juncu, A numerical study of momentum and forced convection heat transfer
around two tandem circular cylinders at low Reynolds numbers. Part I:
Momentum transfer, Int. J. Heat Mass Transfer 50 (2007) 37883798.
[35] G. Juncu, A numerical study of momentum and forced convection heat transfer
around two tandem circular cylinders at low Reynolds numbers. Part II: Forced
convection heat transfer, Int. J. Heat Mass Transfer 50 (2007) 37993808.
[36] J. Kim, H. Choi, An immersed-boundary nite-volume method for simulation of
heat transfer in complex geometries, Korean Soc. Mech. Eng. Int. J. 18 (2004)
10261035.
[37] E.D. Grimison, Correction and utilization of new data on ow resistance and
heat transfer for cross ow of gases over tube banks, Trans. ASME 59 (1937)
583584.
[38] A. Zhukauskas, Heat transfer from tubes in cross ow, in: J.P. Hartnett, T.F.
Irvine Jr. (Eds.), Advances in Heat Transfer, vol. 8, Acedemic Press, New York,
1972, pp. 93160.
6926 D.C. Lo, D.-T. Su/ International Journal of Heat and Mass Transfer 55 (2012) 69166926