International Journal of Heat and Mass Transfer 96 (2016) 362–370

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

journal homepage: www.elsevier.com/locate/ijhmt

Enhancement of heat transfer by a self-oscillating inverted flag

in a Poiseuille channel flow
Sung Goon Park, Boyoung Kim, Cheong Bong Chang, Jaeha Ryu, Hyung Jin Sung ⇑
Department of Mechanical Engineering, KAIST, 291 Daehak-ro, Yuseong-gu, Daejeon 34141, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: A flexible inverted flag immersed in a Poiseuille flow with heated walls was numerically modeled to
Received 5 October 2015 investigate the dynamics of the flag and its effect on convective heat transfer. An immersed boundary
Received in revised form 22 December 2015 method was used to analyze the interaction between the fluid and the inverted flag. This inverted flag
Accepted 16 January 2016
readily becomes self-oscillating because of its configuration, in which the leading edge is free to move
Available online 4 February 2016
and the trailing edge is clamped. The inverted flag has three dynamic modes according to the character-
istics of the surrounding fluid and the flag flexibility: deflected, flapping, and straight. In the flapping
Keywords:
mode, nearly 6 pairs of vortical structures are generated in the wake of the inverted flag, which include
Convective heat transfer
Inverted flag
counter vortical structures formed near the walls as well as structures generated by the interaction
Flow-structure interaction between the flag and the surrounding fluid. These vortical structures affect the thermal boundary layer
Self-oscillating flapping near the walls and the temperature field in a manner that enhances the heat transfer performance of
Vortex dynamics the channel flow.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction rectangular winglets were found to produce counter-rotating lon-

gitudinal vortices on a flat plate that increase the heat transfer
Heat exchangers are widely used in many industrial systems [3]. Protrusions such as slender delta wing or winglet-type vortex
such as refrigeration facilities, boiling facilities, and electrical generators also enhance the heat transfer at the expense of rela-
devices. High-performance heat sinks are of great importance to tively less energy loss penalty [4,5]. Mahalingam and Glezer [6]
the operation of electrical devices, which generate significant heat tested a heat sink incorporating a synthetic jet system that entrains
internally and at the same time are susceptible to the generated cool air upstream and ejects the entrained fluid through the heat
heat. The simplest heat sink models are air-cooled systems in sink channel, which was found to provide a rapid cooling process.
which the flow passes through a channel and the generated heat Although this heat sink system improves the heat transfer perfor-
is exchanged. Such heat sinks have the advantage of simplicity, mance by 140%, auxiliary devices that entrain and discharge fluid
but the flows in heat sink systems are mostly laminar, which have high manufacturing costs and make the heat sink system
reduces system efficiency. To enhance heat transfer, auxiliary more complicated. To minimize the additional manufacturing costs
devices for controlling the flow have been adopted. The introduc- and simplify heat sink systems while enhancing their heat transfer,
tion of such secondary devices requires the simultaneous consider- many researchers have installed flexible reeds with self-oscillating
ation of factors such as the manufacturing costs, the operation characteristics that enhance heat transfer under specific surround-
costs, the increased energy losses, and the improvement in the ing flow conditions [7,8]. Shoele and Mittal [8] numerically scruti-
heat transfer. nized the dynamics of a flexible reed in the presence of wall effects
Kays and London [1] proposed a heat sink system in which the and found that this heat sink system improves the heat transfer
flow and temperature fields are structured by introducing rough- efficiency by approximately 20%; their results were expressed in
ness elements or secondary fins onto its surfaces. Various types terms of the thermal enhancement factor, which is defined as the
of protrusions embedded in channel flows have been used as vor- ratio of the mean heat transfer in the channel to that in a channel
tex generators. Fiebig et al. [2] demonstrated that a heat sink sys- without any secondary devices. The self-oscillation of the flexible
tem with wings in the channel flow exhibits greater heat transfer reed is due to energy transfer from the kinetic energy of the sur-
than a heat transfer system without a secondary device. Pairs of rounding fluid to the strain energy of the reed. However, as many
previous studies have shown [9–12], the critical velocity for the
⇑ Corresponding author. Tel.: +82 42 350 3027; fax: +82 42 350 5027. flapping motion of the flexible reed is high, which undermines
E-mail address: [email protected] (H.J. Sung). the stability of the system.

http://dx.doi.org/10.1016/j.ijheatmasstransfer.2016.01.043
0017-9310/Ó 2016 Elsevier Ltd. All rights reserved.
 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370 363

Nomenclature

A flapping amplitude Uib velocity at immersed boundary

C1, C2 momentum force constant U1 mean fluid velocity at the inlet
cp heat capacity vj data sequence
D spatial structure matrix X flag position
El net energy loss Xib immersed boundary position

l
E mean net energy loss ytip y position of flag tip
F Lagrangian momentum force
f Eulerian momentum force Greek symbols
ff flapping frequency b thermal expansion coefficient
H channel height d delta function
h grid size Uj dynamic modes
k thermal conductivity coefficient c bending rigidity
K, K0 data sequence matrix g heat transfer efficiency
L inverted flag length ki dynamic mode eigenvalues
M linear dynamical system matrix l dynamic fluid viscosity
p pressure q density difference between the fluid and the flag
Pr ¼ cp l=k Prandtl number q0 fluid density
Q heat transfer q1 flag density


Q mean heat transfer x dynamic mode frequency
Re ¼ q0 U 1 L=l Reynolds number
s Lagrangian coordinate Subscripts
St = ff A/U Strouhal number
C cold
T temperature H hot
t time i inlet
U flag velocity W wall
u fluid velocity

Kim et al. [12] proposed the use of an inverted flag, which in The amplitude A is defined by the distance from the lowest peak to
contrast to typical flags has a freely moving leading edge and a the highest peak. The flapping frequency is denoted by ff. The fluid
clamped trailing edge. For an inverted flag immersed in a flow, motion is defined in the Eulerian coordinate with the origin located
the critical flow velocity of its flapping motion is relatively low at the trailing edge of the inverted flag, and the flag motion is
so the flag easily becomes unstable, which is a constraint on the described on the curvilinear coordinate s. The interaction between
energy harvesting system. An inverted flag has three dynamical the flag and the surrounding fluid was analyzed within the frame-
modes according to the conditions of flow and the material elastic work of the immersed boundary method. The fluid motion with
characteristics. In the flapping mode, the inverted flag has a high thermal convection is governed by the incompressible Navier–
flapping amplitude that is almost twice its length. The dynamics Stokes equations and the continuity equation
of inverted flags has previously been analyzed numerically and
@u 1
the vortical structures generated in their wake have been visual- þ u  ru ¼ rp þ r2 u þ f ; ð1Þ
ized [13]. Although these studies have investigated the dynamics @t Re
of inverted flags and identified many features beneficial to energy
harvesting systems, their behaviors in heat transfer processes have
r  u ¼ 0; ð2Þ
not yet been elucidated [12,13]. where u is the velocity vector, p is the pressure, f is the momentum
The main purpose of the present study is to explore the dynam- forcing term that enforces the no-slip conditions on the flag, the
ics of an inverted flag interacting with a wall. We investigated the Reynolds number is defined by Re ¼ q0 U 1 L=l, and l is the dynamic
dynamics of inverted flags in Poiseuille channel flows for various viscosity. By choosing the fluid density q0 as the characteristic den-
flag bending rigidities and Reynolds numbers. The generated vorti-
cal structures were analyzed in detail and their effects on the ther-
mal boundary layers were examined. We also compared the
dynamic modes of the vorticity to those obtained from the normal-
ized temperature fields. Further, we investigated the effects on the
heat transfer rate and the efficiency of the heat sink system with a
flexible inverted flag. The immersed boundary method was used to
analyze the interaction between each inverted flag and the sur-
rounding fluid. Section 2 describes this approach and our numeri-
cal methods. The numerical results and discussion are presented in
Section 3, and our conclusions are presented in Section 4.

2. Problem formulation

A schematic diagram of an inverted flag in a Poiseuille channel Fig. 1. Schematic diagram of an elastic inverted flag in a Poiseuille flow. The leading
flow and the coordinate system is shown in Fig. 1. The leading edge edge is free and the trailing edge is clamped. A is the flapping amplitude of the
of the inverted flag is free to move and the trailing edge is clamped. inverted flag in the y-direction. s is the curvilinear coordinate of the inverted flag.
364 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370

sity, the flag length L as the characteristic length, the bulk mean flag expressed by U = dX/dt. The Lagrangian momentum force F is
fluid velocity at the inlet U 1 as the characteristic velocity, the fluid distributed over the nearby fluid domain by the following
governing equations were non-dimensionalized and the following expression,
non-dimensional scales were introduced: L/U1 for time, q0 U 21 for
Z
pressure, q0 U 21 =L for the Eulerian momentum forcing f. The temper-
f ðx; tÞ ¼ q Fðs; tÞdðx  Xðs; tÞÞds; ð9Þ
ature defined in the Eulerian coordinates is governed by the thermal C
energy equation,
@T 1 where q is the density ratio q = q1/(q0L) and was set to 1 in the pre-
þ u  rT ¼ r2 T: ð3Þ
@t RePr sent simulation. More details of the immersed boundary method
can be found in Huang et al. [17]. Fig. 2 shows the computational
Here, Pr is the Prandtl number Pr ¼ cp l=k, where cp is the heat
domain and the boundary conditions. The height and length of
capacity and k is the thermal conductivity coefficient. The dimen-
the channel are 2L and 32L respectively. A Poiseuille flow with a
sionless temperature T is defined as T ¼ ðT  T o Þ=ðT w  T o Þ, where
mean velocity U was applied at the inlet boundary (x = 4L) and
Tw is the wall temperature and T0 is the fluid temperature at the
the Neumann boundary condition was set at the outer boundary
inlet. The fractional step method [14] was used to solve the fluid
(x = 28L). The no-slip boundary condition was applied at the lower
governing equations on a staggered Eulerian grid system. More
and upper channel walls. The temperature of the inflow T0 was 0
details of the numerical procedures of the fractional step method
and the wall temperature was Tw. Pr was set at 1 and the buoyancy
can be found in Kim et al. [14].
driven force was not taken into account in the present study. We
The motion of the flexible inverted flag is governed by
! performed the convergence test with respect to the grid size h


@2X @ @X @2 @2X and the results are shown in Table 1. Numerical results such as
¼ r  2 c 2  F; ð4Þ the flapping amplitude and the Strouhal number defined by
@t2 @s @s @s @s
ffA/U were found to converge as the grid size became dense. In
where Xðs; tÞ is the position of the inverted flag, s is the curvilinear the present study, we adopted a grid size h of 1/32 (see Case 3 in
coordinate on the flag, r is the tension force along the curvilinear Table 1).
coordinate, c is the bending rigidity, and F is the Lagrangian momen-
tum force representing the fluid–flag interaction. The flag governing 3. Results and discussion
equation was non-dimensionalized by the following characteristic
scales: the flag density r1 defined the unit of density, the flag length 3.1. Dynamics of the inverted flag
L defined the unit of the length, L/U1 defined the unit of time, q1 U 21
defined the tension force r q1 U 21 L2 defined the bending rigidity c The inverted flag was found to exhibit three different dynamical
and q1 U 21 =L defined the Lagrangian momentum force F. modes according to its flexibility: deflected, flapping, and straight.
In the immersed boundary method, the moving Lagrangian grid In the deflected mode, the inverted flag is initially deflected in one
passes across the fixed Eulerian grid as the flag is fluttering in the direction and is not restored to its original position while oscillat-
surrounding fluid, and the fluid-flag interactions can be formulated ing with a small amplitude in a deflected state. The deflected mode
in terms of the momentum forces f and F, which are added to the occurs when the elastic energy stored during the deflecting process
fluid motion Eq. (1) and the flexible flag motion Eq. (4), respec- is insufficient to overcome the kinetic energy of the surrounding
tively. In this method, a set of Lagrangian points that consists of fluid. In the flapping mode, the flexible flag is also initially
the position of the immersed boundary point Xib, is used, which deflected in one direction, as in the deflected mode. However,
moves with the local fluid. The velocity on the immersed boundary when the elastic energy stored in the flag is sufficient to overcome
Uib can be obtained by interpolating the local fluid velocity at the the kinetic energy of the fluid, the flag tends to return to its original
flag position X position. When it reaches the original position, it starts to deflect in
Z the opposite direction due to the inertial effect. This equilibrium
U ib ðs; tÞ ¼ uðx; tÞdðXðs; tÞ  xÞdx; ð5Þ between the elastic energy and the kinetic energy results in the
X flexible flag flapping through the center axis (y = 0) in a steady
where d() is the Dirac delta function, state, which is called the flapping mode. When the stiffness of
 x  y the flexible flag is too high for flapping to be initiated by the sur-
1 rounding fluid, the straight mode occurs. Fig. 3(a) shows the max-
dðxÞ ¼ 2
/ / : ð6Þ
h h h imum and minimum y positions of the inverted flag in a steady
Here, h denotes the Eulerian mesh size, h = Dx = Dy. The four-point state. The initial y position of the leading edge was 0.15 in the
delta function introduced by Peskin [15] is adopted in the present study, present simulation. The deflected mode arises in the low bending
8  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rigidity region (c < 0.4). As the bending rigidity c increases, the
>
> 1
3  2jrj þ 1 þ 4jrj  4r 2 ; 0 6 jrj < 1; elastic energy becomes sufficiently large to initiate flapping behav-
>
< 
8
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi  ior. Once flapping is initiated, the flag continuously self-oscillates
/ðrÞ ¼ 1 5  2jrj  7 þ 12jrj  4r 2 ; 1 6 jrj < 2; ð7Þ
>
> 8 by passing through the y = 0 axis. The flapping amplitudes of an
>
:
0; 2 6 jrj: inverted flag in the flapping mode are larger than those of a con-
ventional flapping flag. The amplitude of the flapping mode
The fluid motion and the flag motion are coupled through the decreases as the bending rigidity increases. As the bending rigidity
immersed boundary of the flag and the interaction force can be cal- increases, the straight mode arises in which the flag is stationary
culated by the feedback law, without any oscillations. Fig. 3(b) shows the maximum and mini-
Z t mum y positions of the leading edges. In the deflected mode, the
F ¼ C1 ðU ib  UÞdt þ C 2 ðU ib  UÞ; ð8Þ flag oscillates on one side with respect to y = 0 with only a small
0
amplitude. As the bending rigidity increases, the flexibility of the
where C1 and C2 are large negative free constants set to be 8  106 flag overcomes the kinetic energy of the surrounding fluid, which
and 250 defined by Shin et al. [16]. U is the velocity of the inverted initiates flapping behavior. When the bending rigidity is nearly
 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370 365

Fig. 2. The computational domain and the applied boundary conditions.

Table 1
Flapping amplitude and the Strouhal number St of the inverted flag in the Poiseuille as the bending rigidity increases. Fig. 5 (c = 1.1, 1.3, 1.5) also shows
flow. the stationary positions for the straight modes of the flags.
h A/L St
Case 1 1/20 0.02 0
3.2. Vortical structures and temperature fields
Case 2 1/25 1.69 0.153
Case 3 1/32 1.71 0.181
Case 4 1/40 1.71 0.187 Fig. 6 shows the vortical structures generated in the wake of a
flapping mode (c = 0.5) and schematic diagrams of these struc-
tures. During one flapping period, the vortical structures were
visualized at the instants shown in Fig. 4(b). In Fig. 6(a), the vorti-
1.1, the flag lapses into a straight mode with the leading edge at a cal structures denoted by 1 (V1) and 2 (V2) were generated by the
negative value of y. However, when the bending rigidity is between free leading edge and the clamped trailing edge respectively. When
1.15 and 1.3, flapping behavior arises. the inverted flag deflects as depicted in Fig. 6(a), the oncoming
Fig. 4 shows the time histories of the y positions of the leading fluid velocity at the leading edge is lower than that at the trailing
edges for various bending rigidities. Fig. 4(a) shows that the edge since the no-slip boundary condition is applied to the channel
inverted flag initially deflects downward and oscillates with a wall. V2 is shed further downstream than V1 due to the difference
small amplitude in a steady state. The initial y position of the flag between the velocities of the fluids passing the leading edge and
is negative, which leads to a downward deflection of the flag. As the trailing edge. When the inverted flag is immersed in a uniform
the bending rigidity increases, the flag is also initially deflected flow and far enough from the channel walls that they do not influ-
downward and the leading edge of the flag moves to the opposite ence the flag, the surrounding fluid velocities are the same at both
direction because of the elastic energy stored during the deflection, edges. When the fluid passes the leading edge of the inverted flag
as can be seen in Fig. 4(b). Fig. 4(c) shows a flapping behavior with at the position shown in Fig. 6, the fluid is accelerated by the stroke
an amplitude smaller than that shown in Fig. 4(b), for which the of the flag, which means that the fluid velocity at the leading edge
bending rigidity is lower. When the bending rigidity is 1.2, the flag is faster than that at the trailing edge, and this is also true for the
is initially deflected upward, in contrast to the flag with a lower generated vortical structures. Further downstream, V2 is a result of
bending rigidity in Fig. 4(c). In Fig. 4(d), the flag is initially the wall effect. V2 induces a fluid motion towards the boundary
deflected upward, and then becomes stationary in a steady state. layer of the lower wall, which generates the counter vortical struc-
Fig. 5 shows superimposed inverted flags with various bending ture marked 20 (V20 ). V2 moves closer to the upper wall while gen-
rigidities. The red lines indicate the flags’ initial positions and the erating V20 and the fluid at the boundary layer of the upper wall is
green lines indicate the final positions in the steady state. The induced to move into the center region by V1 and V2, as shown in
two green lines indicate the positions from peak to peak when Fig. 6(b). In Fig. 6(b), the shear layers generated by the leading edge
the flag is oscillating in a steady state. When the bending rigidity become unstable and roll up at the trailing edge, which leads to the
c is 0.2, the flag is deflected downward and oscillates with a small formation of V2. When the flag is deflected downward through the
amplitude. As the bending rigidity increases (c = 0.4 in Fig. 5), the center axis (y = 0), V1 is shed downstream, which separates V2
flag oscillates in a less deflected state. As shown in Figs. 3–5 from the shear layers on the flag. This process leads to the forma-
(c = 0.5, 1.0, 1.2), the amplitude of the flapping mode decreases tion of another vortical structure denoted by 3 (V3), which is gen-

(a) (b)
2 1

1.5 0.5
A/L

Flapping
yr/L

Deflected Straight
1 0

0.5 -0.5
Straight

0 -1
0 0.5 1 1.5 0 0.5 1 1.5

Fig. 3. (a) Flapping amplitudes for various bending rigidities and (b) the maximum and minimum y positions of the leading edges.
366 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370

(a) 1.2 (b) 1.2

A G
0.8 0.8

0.4 0.4
F

ytip/L

ytip/L
0 0
B
-0.4 -0.4
A C E
-0.8 -0.8 D

-1.2 -1.2
0 20 40 60 80 0 20 40 60 80
(c) 1.2 (d) 1.2
0.8 0.8

0.4 0.4
A
ytip/L

ytip/L
0 0

-0.4 -0.4

-0.8 -0.8

-1.2 -1.2
0 20 40 60 80 0 20 40 60 80
t t
Fig. 4. Time histories of the y values of the leading edges for various bending rigidities: (a) c = 0.4 (deflected mode), (b) c = 0.5 (flapping mode), (c) c = 1.2 (flapping mode),
and (d) c = 1.3 (straight mode).

Fig. 5. Superimposed inverted flags for various bending rigidities. The red lines indicate the initial positions of the inverted flags. The green lines indicate the positions of the
inverted flags in a steady state. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

erated at the trailing edge. In Fig. 6(d), the flag is positioned in the fluid motion also distorts the thermal boundary layers near the
opposite direction with respect to the center axis to that found in upper wall above region A in Fig. 7(a). The vortical structure V2
Fig. 6(a). At the leading and the trailing edges, V4 and V5 are gen- in Fig. 6(b) also affects the thermal boundary layers near the lower
erated in the same manner as in Fig. 6(a). Downstream, V20 is wall and induces the heated fluids at the lower wall to move into
moved closer to the upper wall by the fluid motion induced by the center axis above region B, as shown in Fig. 7(b). In Fig. 7(c),
V2, which generates another counter vortical structure V200 near as the flag deflects downward, V1 and V2 are also shed down-
the upper wall. In Fig. 6(e), the counter vortical structure (V50 ) is stream, which mixes the heated fluid near both walls with the sur-
generated near the upper wall as V5 moves closer to the lower rounding fluid because of the rotating momentums of those
wall. The fluid near the lower wall is induced to move into the cen- vortical structures. Fig. 7(d) shows that the heated fluids near the
ter region by V4 and V5, as shown in Fig. 6(e). As arises for V1 in lower wall around region C are entrained into the center region
Fig. 6(c), V4 separates V5 from the shear layers at the flag and V6 by the vortical structures V4 and V5, as in Fig. 6(d).
is generated when the flag deflected upward through the center Fig. 8 shows the vortical structures and temperature fields for
axis, as shown in Fig. 6(f). During one flapping period, nearly 6 some deflected and straight modes. The inverted flag is deflected
pairs of vortical structures are generated, including the counter downward and the vortical structures are generated by both the
vortical structures generated near the wall as well as those gener- leading edge and the trailing edge in Fig. 8(a); the oncoming fluids
ated by the flapping behavior of the inverted flag. passing between the leading edge and the lower wall roll up into
Fig. 7 shows the temperature contours in the wake of the the region behind the deflected flag, which produces a low velocity
inverted flag for c = 0.5 (flapping mode) at the same instants as region at B. Downstream, the vortical structures generated by the
in Fig. 6(a–d). The fluid near the upper wall is induced to move into trailing edge and the counter vortical structures generated near
the center axis region by V1 and V2, as in Fig. 6(a). The induced the upper wall mix the near wall fluids with center region fluids.
 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370 367

Fig. 6. Vorticity contours in the wake of the inverted flag for c = 0.5 (flapping mode) and schematic diagrams of these structures. These vorticity contours were visualized at
the instants marked in Fig. 4(b) (A–F). (a) – A, (b) – B, (c) – C, (d) – D, (e) – E, (f) – F, and (g) – G. (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)

(a)1 (b)
A A
0.5
0
-0.5 B
-1
(c)1 (d) T/ΔTw
T

0.5 A 0.97
0.83
0.69
0 B 0.55
B A 0.41
-0.5 C 0.28
0.14
-1-2 -1 0 1 2 3 4 5 6 -2 -1 0 1 2 3 4 5 6
0.00

Fig. 7. Temperature contours in the wake of the inverted flag for c = 0.5 (flapping mode) at the same instants as in Fig. 6(a)–(d).

Fig. 8. Vorticity and temperature contours in the wake of the inverted flag for (a) c = 0.2 (deflected mode), (b) c = 0.4 (deflected mode), and (c) c = 1.3 (straight mode).
368 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370

In the temperature field in Fig. 8(a), a relatively thick thermal dynamic system. The logarithmic mapping of these eigenvalues
boundary layer is evident from the lower wall at region B, which is provides the frequency xj of the dynamic system expressed as
due to the low velocity of the fluid. Fig. 8(b) shows the vortical struc- xj ¼ Imðkf Þ=Dt and the corresponding dynamic modes Uj = DTj.
tures and temperature field of a deflected inverted flag with a larger Fig. 9(a) shows the variation with frequency of the normalized
bending rigidity. The main difference between the vortical struc- energy magnitude |U|, which provides a way to measure the contri-
tures of these two cases is that the structure generated by the trailing bution of each mode to the overall energy; Fig. 9(b–d) show repre-
edge approaches the lower wall in Fig. 8(b), which means that there sentative dynamic modes using the vorticity fields generated in
is less fluid mixing and a lower amount of heat transfer than in Fig. 8 the wake of the flapping inverted flag (c = 0.5). The obtained eigen-
(a). In region B of the temperature field in Fig. 8(b), there is a thick values are complex conjugate pairs, which results in the symmetric
thermal boundary layer due to the low fluid velocity. In Fig. 8(c), spectrum with respect to x = 0 shown in Fig. 9(a). Mode 1 appears at
the shear layers near the flag in the straight mode are shown; the x = 0 with the highest energy and is the time-invariant mean vortic-
thermal boundary layers seem unaffected by the flag. ity, which has positive and negative values on the upper and lower
walls respectively that arise due to the shear layers on the walls.
The 2nd highest energy mode appears at x = 0.663 and is visualized
3.3. Dynamic mode decomposition for vorticity and temperature in Fig. 9(c). The structures have periodic behaviors in time and have
elongated shapes parallel to the walls, as shown in Fig. 9(c). The red
Dynamic mode decomposition (DMD) has been used to extract and blue contours indicate positive and negative fluctuations of the
dynamic modes from turbulent flows [18,19]. We applied DMD to vorticities respectively. The flapping frequency of the inverted flag in
temperature fields as well as flows in the wake of the flapping the flapping mode (c = 0.5) is approximately 0.104, as shown in
inverted flag so that the relationships between the generated vor- Fig. 4(b), which is approximately six times smaller than the fre-
tical structures and the distorted temperature fields could be fur- quency of the 2nd highest energy mode. As in Fig. 6, the vortical
ther analyzed. DMD extracts dynamic information from the data structures are generated by the flapping behaviors of the inverted
sequence of the flow and temperature fields and relies on a low- flag; these structures induce fluid motions near the walls that gener-
dimensional subspace after applying a high-degree polynomial to ate counter vortical structures, which results in the formation of
the data sequence. The basic premise of DMD is that each data almost 6 pairs of generated structures during one flapping period.
sequence in time is assumed to be related by a linear dynamical This process affects the frequency at which the 2nd highest energy
system, which completely determines the behavior of the dynam- mode appears. In Fig. 9(d), mode 3 contains smaller structures with
ical system. Detailed derivations of DMD have been given by Sch- a higher frequency (x = 1.97) when compared to mode 2.
mid [18], and are briefly described here. Fig. 10 shows the results of the DMD analysis for the tempera-
A data sequence of flow or temperature fields denoted by vj that ture field. Mode 1 is the time-invariant mean temperature field,
is equally spaced in time can be represented in the forms K and K0 which is similar to that found in a channel flow without any sec-
as follows: ondary apparatus, as in Fig. 10(b). In Fig. 10(c), the positive and
negative temperature fluctuations are added with a frequency of
K ¼ fv 1 ; v 2 ; . . . ; v N1 g; ð10Þ x = 0.669, which is almost the same as the frequency at which
the vorticity dynamical mode 2 appears. The structures found in
Fig. 10(c) are generally inclined in the streamwise direction, which
K 0 ¼ fv 2 ; v 3 ; . . . ; v N g: ð11Þ
implies that the vortical structures have significant effects on the
Each data sequence is assumed to be a linear dynamical system, mixing of the heated fluids near the wall with the fluids of the cen-
satisfying vj+1 = Mvj. The dynamic characteristics of the system are ter regions and generating temperature fluctuations. Mode 3 also
extracted from the eigenvalues and eigenvectors of matrix M. K appears at the same frequency (x = 1.98) as the vorticity dynamic
can be calculated as K ¼ DRW T where D contains information mode 3, and contains inclined structures, as shown in Fig. 10(d).
P
about the spatial structures, is a diagonal square matrix with
the energy rankings in the diagonal elements, and W represents 3.4. Effects of varying the bending rigidity and Re number on thermal
eigenvectors including temporal structures. A square matrix S is performance
newly defined as S ¼ DT K 0 WR1 ; the eigenvalues ki of S approxi-
mate some of the eigenvalues of the matrix M. The eigenvalues ki Fig. 11 shows the mean heat flux Q, the mechanical energy loss
and eigenvectors Tj of S include frequency information for the El, and the heat transfer efficiency g, which were calculated at a

(a) 1 (b)
γ = 0.5
|Φ|

0.5
(c) (d)

0 -5 -4 -3 -2 -1 0 1 2 3 4 5
ω
(b) 1
0
y

-1
(c) 1
0
y

-1
(d) 1
0
y

-1 0 5 10 x 15 20 25

Fig. 9. DMD analysis for c = 0.5 (flapping mode). (a) Energy magnitude and (b–d) representative dynamic modes corresponding to dominant eigenvalues in the spectrum,
visualized as the vorticity.
 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370 369

Fig. 10. DMD analysis for c = 0.5 (flapping mode). (a) Energy magnitude and (b–d) representative dynamic modes corresponding to the dominant eigenvalues in the
spectrum, visualized as the temperature.

(a) 1.1 (b) 3.5 (c) 1.2

Re = 200
1 3
1.1
0.9 2.5
El

| |
Q

deflected mode 1
flapping mode 2
0.8
straight mode
1.5 0.9
0.7 channel
channel
1
0.6 0.8
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

Fig. 11. Variations with the bending rigidity c in (a) the mean heat flux, (b) the mechanical energy loss, and (c) the heat transfer efficiency. The dashed lines in (a) and (b)
indicate the time averaged mean heat flux Q
0 and the mechanical energy loss E
lo of the channel flow without an inverted flag respectively.

cross section of the output plane that is 26L downstream from the the mean heat transfer increases as the flapping amplitude
clamped trailing edge. The net change in the fluid thermal energy increases. When compared to the case of the channel without a
is flag, the mean heat transfer performance for the channel with
Z L
the inverted flag (c = 0.6) increases by up to 160%. As the flapping
Q¼ q0 cp uðT  T 0 Þdy: ð12Þ amplitude increases, the energy loss increases, as shown in Fig. 11
L (b). In straight modes, the mean heat transfer increases little, and
The net energy loss due to the pressure drop between the inflow the energy loss increases more than the mean heat transfer
and the output plane is given by increases, which results in an efficiency that is less than 1, as
shown in Fig. 11(a–c). The maximum improvement in the effi-
Z H Z H
ciency arises when the bending rigidity of the flag is 1.0.
El ¼ pudyx¼0  pudyx¼26L : ð13Þ
0 0
Fig. 12 shows a phase diagram of the dynamic modes with
respect to the bending rigidity and the Re number. The dynamic
The time averaged mean heat transfer and mechanical energy

and E
loss are denoted Q
l respectively. The heat transfer efficiency
g was adopted to assess the enhancement of the heat transfer per-
formance by the inverted flag relative to that obtained in a channel
without any secondary devices [8,20,21], and is defined b

=E
1=3
Q
g ¼
l 1=3 ; ð14Þ
Q 0 =El0

where Q
0 and E

l0 are the time averaged values of the mean heat
transfer and the mechanical energy loss respectively, which were
measured in a channel without a flag. A system for which the heat
transfer efficiency g is above 1 is expected to improve the heat
transfer performance over that of the channel without a flag.
Fig. 11(a) shows the variation of the time averaged mean heat
flux with the bending rigidity. For deflected flags (c = 0.2 and
0.4), both the mean heat flux and the mechanical energy loss are
larger when the flag is deflected more (c = 0.2), as shown in Fig. 12. Phase diagram of the dynamic modes of the inverted flag. The contour
Fig. 11(a) and (b). In flapping modes, the time averaged value of shows the normalized flapping amplitude.
370 S.G. Park et al. / International Journal of Heat and Mass Transfer 96 (2016) 362–370

(a) 12 (b) 4 (c) 1.3

10 1.2

8 1.1
| |

El
Q 6
2
1

4 channel 0.9
channel
2 0 0.8
200 400 600 800 200 400 600 800 200 400 600 800
Re Re Re
Fig. 13. Variations with Re in (a) the mean heat flux, (b) the mechanical energy loss, and (c) the heat transfer efficiency.

mode is more affected by the bending rigidity than by the Re num- iary devices generate vortical structures and improve thermal
ber. In a region of low bending rigidity, the inverted flag generally performance.
enters the deflected mode, and exhibits the straight mode in a
region of high bending rigidity, as shown in Fig. 12. The flapping Acknowledgment
mode arises for bending rigidities between 0.7 and 1.0 regardless
of the Re number. Fig. 13 shows the variations with the Re number This study was supported by the Creative Research Initiatives
in the mean heat transfer, mechanical energy loss, and the heat (No. 2015-001828) program of the National Research Foundation
transfer efficiency. The time averaged mean heat transfer decreases of Korea (MSIP).
as the Re number increases, as evident in Fig. 13(a). The absolute
value of the heat transfer decreases as Re increases. However, the References
heat transfer performance when compared to that of the channel
without the flag is increased by up to 250% when the Re number [1] W.M. Kays, A.L. London, Compact Heat Exchangers, second ed., McGraw-Hill,
New York, 1984.
is 800. The mechanical energy loss decreases as the Re number [2] M. Fiebig, U. Brockmeier, N.K. Mitra, T. Guntermann, Structure of velocity and
increases, as can be seen in Fig. 13(b). The efficiency g in Fig. 13 temperature fields in laminar channel flows with longitudinal vortex
(c) indicates that the heat transfer system with the inverted flag generators, Num. Heat Transfer, Part A 15 (1989) 281–302.
[3] A.Y. Turk, G.H. Junkhan, Heat transfer enhancement downstream of vortex
is not efficient at a low Re number (Re = 100), but it operates effi- generators of a flat plate, Eight Int. Heat Transfer Conf., 6, San Francisco, (1986)
ciently at high Re numbers (Re = 200–800), for which the efficiency 2903–2908.
g is enhanced by a factor of 1.2 (Re = 800). [4] G. Biswas, K. Torii, D. Fujii, K. Nishino, Numerical and experimental
determination of flow structure and heat transfer effects of longitudinal
vortices in a channel flow, Int. J. Heat Mass Transfer 39 (1996) 3441–3451.
4. Conclusions [5] S. Tiggelbeck, N. Mitra, M. Fiebig, Flow structure and heat transfer in a channel
with multiple longitudinal vortex generators, Exp. Therm. Fluid Sci. 5 (1992)
425–436.
In the present study, we investigated the dynamic modes of the [6] R. Mahalingam, A. Glezer, Design and thermal characteristics of a synthetic jet
inverted flag in a Poiseuille flow and the effects of the vortical ejector heat sink, J. Electron. Packag. 127 (2005) 172–177.
structures generated by the flag on convection heat transfer. An [7] P. Hidalgo, F. Herrault, A. Glezer, M. Allen, S. Kaslusky, B.S. Rock, Heat transfer
enhancement in high-power heat sinks using active reed technology, in:
inverted flag embedded in a surrounding flow becomes unstable Proceedings of the 16th International Workshop On Thermal Investigations of
readily because of the kinematic energy of the flow. The inverted ICs and Systems THERMINIC), Barcelona, Spain, (2010) 1–6.
flag in a Poiseuille flow exhibits three dynamic modes according [8] K. Shoele, R. Mittal, Computational study of flow-induced vibration of a reed in
a channel and effect on convective heat transfer, Phys. Fluids 26 (2014)
to the bending rigidity: deflected, flapping, and straight. The vorti- 127103.
cal structures and their effects on the convection heat transfer vary [9] J.J. Allen, A.J. Smits, Energy harvesting eel, J. Fluids. Struct. 15 (2001) 629–640.
with the dynamic mode. Two vortical structures are generated by [10] B.S.H. Connell, D.K.P. Yue, Flapping dynamics of a flag in a uniform stream, J.
Fluid Mech. 581 (2007) 33–68.
the interactions between the flag and the surrounding fluid, one [11] S. Alben, M.J. Shelley, Flapping states of a flag in an inviscid fluid: bistability
from the clamped trailing edge and the other from the free leading and the transition to chaos, Phys. Rev. Lett. 100 (2008) 074301.
edge. These vortical structures induce fluid motions near the walls [12] D. Kim, J. Cosse, C.H. Cerdeira, M. Gharib, Flapping dynamics of an inverted
flag, J. Fluid Mech. 736 (2013) R1.
that lead to counter vortical structures, and nearly 6 pairs of such
[13] J. Ryu, S.G. Park, B. Kim, H.J. Sung, Flapping dynamics of an inverted flag in a
structures are generated during one flapping period. The tempera- uniform flow, J. Fluid Struct. 57 (2015) 159–169.
ture fields are distorted by the generated vortical structures, so [14] K. Kim, S.J. Baek, H.J. Sung, An implicit velocity decoupling procedure for
inclined fluctuations were found in our DMD analysis. Both the incompressible Navier-Stokes equations, Int. J. Numer. Method Fluids 38
(2002) 125–138.
mean heat transfer and the mechanical energy loss decrease as [15] C.S. Peskin, The immersed boundary method, Acta Numer. 11 (2002) 479–517.
the Re number increases. When the system with the inverted flag [16] S.J. Shin, W.-X. Huang, H.J. Sung, Assessment of regularized delta functions and
was used, the heat transfer performance was found to be enhanced feedback forcing schemes for an immersed boundary method, Int. J. Numer.
Method Fluids. 58 (2008) 263–286.
by a factor of 2.5 at Re = 800 in comparison to the channel without [17] W.-X. Huang, S.J. Shin, H.J. Sung, Simulation of flexible filaments in a uniform
a flag. When both the heat transfer enhancement and the mechan- flow by the immersed boundary method, J. Comput. Phys. 226 (2007) 2206–
ical energy loss were considered, the efficiency was found to be 2228.
[18] P.J. Schmid, Dynamic mode decomposition of numerical and experimental
increased by the inverted flag by a factor of 1.2. Our study has char- data, J. Fluid Mech. 656 (2010) 5–28.
acterized the dynamics of flexible inverted flags with a channel [19] A. Seena, H.J. Sung, Spatiotemporal representation of the dynamic modes in
effect and the fundamental mechanisms by which the self- turbulent cavity flows, Int. J. Heat Fluid Flow 44 (2013) 1–13.
[20] D.L. Gee, R. Webb, Forced convection heat transfer in helically rib-roughened
oscillating behaviors of the flag distort the temperature fields tubes, Int. J. Heat Mass Transfer 23 (1980) 1127–1136.
through the generation of vortical structures and thus improve [21] J. Tisa, J. Hwang, Measurements of heat transfer and fluid flow in a rectangular
heat transfer performance. This study has determined the key duct with alternate attached-detached rib-arrays, Int. J. Heat Mass Transfer 42
(1998) 2071–2083.
mechanisms by which heat transfer systems with simplified auxil-