International Journal of Heat and Fluid Flow 38 (2012) 40–49

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Fluid Flow

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

Flow instability in baffled channel flow

Changwoo Kang, Kyung-Soo Yang ⇑
Department of Mechanical Engineering, Inha University, Incheon 402-751, Republic of Korea

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

Article history: Flow instability in baffled channel flow, where thin baffles are mounted on both channel walls periodi-
Received 28 November 2011 cally in the direction of the main flow, has been numerically investigated. The geometry considered here
Received in revised form 21 July 2012 can be regarded as a simple model for finned heat exchangers. The aim of this investigation is to under-
Accepted 13 August 2012
stand how baffle interval (L) and Reynolds number (Re) influence the flow instability. With a fixed baffle
Available online 18 October 2012
length of one quarter of channel height (H), ratios of baffle interval to channel height (RB = L/H) between 1
and 4 are considered. The critical Reynolds number of the primary instability, a Hopf bifurcation from
Keywords:
steady flow to time-periodic flow, turned out to be minimum when RB = 3.08. The friction factor (f) is
Baffled channel
Flow instability
strongly correlated with the critical Reynolds number for RB 6 2.5. For the particular cases of
Hopf bifurcation RB = 1.456 and RB = 1.0, we performed Floquet stability analysis in order to study the secondary instability
Floquet stability analysis through which time-periodic two-dimensional flow bifurcates into three-dimensional flow. The results
obtained in this investigation are in good agreement with those computed from full simulations, and
shed light on understanding and controlling flow characteristics in a finned heat exchanger, quite bene-
ficial to its design.
Ó 2012 Elsevier Inc. All rights reserved.

1. Introduction periodic solution (Roberts, 1994). In the case of baffled channel

Laminar–turbulent transition is often caused by flow instability,
and regarded as an essential process towards fully turbulent flow
(Herbert, 1983, 1988). In many engineering applications, suppress-
ing or triggering transition is widely used depending upon the de-
sired type of flow, laminar or turbulent, required by the particular
application. Therefore, understanding the process of laminar–
turbulent transition is very important in flow control, and must
be preceded by a study of the flow instability associated with the
specific flow configuration.
Flows in heat exchangers or turbulence enhancers are often char-
acterized by finned channel flows. The thin plates mounted on chan-
nel walls (‘‘baffles’’) play an important role in enhancing the heat
transfer not only by enlarging the fluid contact area but also by
destabilizing the flow field. The flow instability caused by the baffles
triggers laminar–turbulent transition. Consequently, mixing greatly
increases in the disturbed flow field, resulting in more effective heat
transfer (Howes et al., 1991; Cheng and Huang, 1991).
Flow instability in general two-dimensional (2D) flow has been
well explained by a bifurcation theory (Roberts, 1994; Yang, 2000;
Battaglia et al., 1997). Two different types of bifurcation are often
observed in 2D channel flow. One is Fold bifurcation where a stea-
dy solution bifurcates into another steady one, and the other is
Hopf bifurcation where a steady solution bifurcates into a time-
flow in which baffles are mounted on the channel walls symmetri-
cally in the vertical direction and periodically in the streamwise
direction, extensive studies were carried out both experimentally
(Roberts, 1994) and numerically (Yang, 2000), and reported the
characteristics of Hopf bifurcation associated with the primary
flow instability.
In this investigation, a parametric study has been carried out in
order to elucidate the effects of baffle interval (L) and Reynolds
number (Re) on the flow instabilities in baffled channel flow. The
baffle length (B) is fixed as H/4 as in the previous studies (Roberts,
1994; Yang, 2000). Here, H represents the channel height. Firstly,
we compute the critical Reynolds number for Hopf bifurcation
for each value of the ratio (RB = L/H) under consideration, searching
for the most unstable baffled channel flow. Secondly, for some se-
lected cases of RB, we perform Floquet stability analysis in order to
elucidate the secondary instability associated with that particular
flow configuration, leading to three-dimensional (3D) flow. The
geometric configuration of a baffled channel is depicted in Fig. 1.
The flow configuration considered here can be regarded as a simple
model for finned heat exchangers, and the results obtained in the
current parametric study provide useful information for designing
such devices.
2. Formulation and numerical methods

⇑ Corresponding author. Tel.: +82 32 860 7322; fax: +82 32 868 1716. The current investigation requires a parametric study where
E-mail address: ksyang@inha.ac.kr (K.-S. Yang). numerous numerical simulations must be performed with various

0142-727X/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatfluidflow.2012.08.002
 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49
Fig. 1. Flow configuration. L, H, and B denote baffle interval, channel height, and
baffle length, respectively.
values of RB and Re. This kind of parametric study demands consid-
erable amount of computing resources; the computing efforts can
be significantly reduced by employing an immersed boundary
method which facilitates implementing the baffles on a Cartesian
grid system (Kim et al., 2001; Gilmanov and Sotiropoulos, 2005;
Yang and Balaras, 2006; Su et al., 2007; Mittal et al., 2008). See
Fig. 2.
The governing equations for 2D incompressible base flow, mod-
ified for the immersed boundary method, are as follows:
r  u ¼ 0;
@u 1
¼ r  ðuuÞ  rp þ r2 u þ f;
@t Re
where u (or u, v), p and f represent velocity vector, pressure, and the
momentum forcing for imposing the ‘‘immersed baffles’’, respec-
tively. All the physical variables except p are nondimensionalized
by the mean bulk velocity (Um) and channel height (H). Pressure
is nondimensionalized by a reference pressure (Pref) and the
dynamic pressure. Reynolds number is defined as Re = UmH/m,
where m is the kinematic viscosity of the fluid. The governing equa-
tions were discretized by a finite-volume method on a nonuniform
staggered Cartesian grid system (Fig. 2). A second-order central dif-
ferencing was employed for spatial discretization of the derivatives.
A hybrid scheme was used for time advancement; the nonlinear
terms and the forcing term f were explicitly advanced by a third-
order Runge–Kutta scheme, and the diffusion terms were implicitly
advanced by the Crank–Nicolson method. A fractional step method
(Kim and Moin, 1985) was employed to decouple the continuity and
the momentum equations. The Poisson equation resulting from the
second stage of the fractional step method was solved by a multi-
grid method. For detailed description of the numerical method used
in the current investigation, see Yang and Ferziger (1993).
0.8
0.6
y
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2
x
Fig. 2. Grid system. The baffles shown here are for illustration only. The actual
baffles in the computation have zero thickness.
41
3. Choice of parameters and boundary conditions
The no-slip condition was employed at all solid boundaries
including the ‘‘thin’’ baffles of zero thickness; the baffles were im-
posed by the immersed-boundary method. The flow was assumed
to be periodic in the streamwise direction (x). Therefore, we actu-
ally consider an infinitely long channel with the baffles mounted
periodically in x (Fig. 1), even though the actual computational
domain contains only one period in x (Fig. 2). We studied the
effects of the domain size on the base flow and the Floquet modes
by doubling and tripling the one-period domain size in the stream-
wise direction, and no noticeable difference was seen among those
cases.
Simulation of unsteady streamwise-periodic flow in a channel
can be classified into one of the following two cases. In one case,
mass flux is fixed in time, but pressure difference between the inlet
and the outlet of the channel (Dp) fluctuates. In the other case,
mass flux fluctuates while Dp is fixed in time. We adopted the for-
mer approach by following You et al. (2000). The numerical resolu-
tion increases up to 256  256 (for RB P 3) computing cells in x
and y directions, respectively, as Re increases. Further refinement
shows less than 0.23% of difference in the growth rate of the
primary instability.
ð1Þ
4. Results and discussion
ð2Þ
4.1. Onset of the primary instability
Streamlines of the steady symmetric solutions at Re = 60 and
Re = 80 with RB = 1.456 are shown in Fig. 3. The flow does not
undergo any instability at these Reynolds numbers. At a lower Re
(Fig. 3a), two recirculation regions are identified between the
neighboring baffles; they merge into one structure at a higher Re
(Fig. 3b). Above a certain critical value of Reynolds number (Rec),
however, the flow undergoes a Hopf bifurcation (Roberts, 1994)
leading to a solution periodic in time. As a quantitative measure
of the primary instability which causes the bifurcation, we define
Vcl as follows (Roberts, 1994):
Z L
1
V cl ¼ jv ðx; H=2Þj dx; ð3Þ
L 0
which measures the deviation from zero velocity along the channel
centerline. In the case of a steady solution such as in Fig. 3, the flow
(a) Re=60
1.4
(b) Re=80
Fig. 3. Streamlines of steady flow for RB = 1.456.
42 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49

150 1994). Fig. 4 presents the time histories of Vcl and Re for
10
0
RB = 1.456. The Reynolds number seems to be almost constant at
Re = 130 as intended. In the initial stage of simulation which started
10-2 0.35 from the uniform flow combined with low-amplitude random
140
noise, instability does not occur. At approximately t = 13, a Hopf
10-4 bifurcation is triggered, and Vcl grows exponentially afterwards,
0.30
150 160 170
yielding the growth rate (r) of the most unstable mode as the slope

Re
Vcl

130 of the linear portion of the curve. Finally, nonlinearity sets in at

10-6
approximately t = 95, and Vcl periodically oscillates. The growth
rates computed for various Re values are presented in Fig. 5; nega-
10-8
120 tive growth rates are obtained from the exponentially decaying Vcl
Vcl
curves. It is seen that the critical Reynolds number (Rec) is about
10 -10 Re 103 for RB = 1.456, being in good agreement with the previous stud-
ies (Roberts, 1994; Yang, 2000). Fig. 6 shows instantaneous stream-
10 -12 110 lines at equal time interval during one period of flow oscillation at
0 50 100 150 200
t Re = 130 with RB = 1.456. Recirculation regions are generated just
behind each baffle, and travel downstream along the channel wall
Fig. 4. Vcl and Re plotted against time for RB = 1.456. to reach the next baffle. The main stream periodically deflects
towards the upper and lower walls in an alternating manner. The
recirculation regions are counter-rotating in an alternating way.
2 The critical Reynolds number strongly depends upon RB. In Fig. 7,
correlation of Rec with RB is presented. Below RB = 1.5, Rec drastically
1 increases as RB decreases. This means that flow in a baffled channel
Rec ≈103.23 tends to remain steady and symmetric in y as the baffle interval is
shortened. The Rec does not change significantly above RB = 1.5; the
0
σ

minimum value of Rec is identified at RB = 3.08 by a local polynomial

fit. One can also identify a local peak at RB = 2.31. From flow anima-
-1 Present tions that are not shown here, it is seen that there is only one recir-
Roberts(1994) culation region traveling between two neighboring baffles for
-2
RB 6 2.25, while there are two traveling recirculation regions for
50 100 150 200 2.5 6 RB 6 4.0.
Re Fig. 8 presents the variation of friction factor (f) with RB at
Re = 130,
 along
  with Rec. Here, the friction factor is defined as
Fig. 5. Growth rate of the primary instability (Hopf bifurcation) plotted against Re
for RB = 1.456. f ¼ 2 Dp=L H= qU 2m where q represents fluid density, and the
over-bar denotes time averaging. One can notice that there exists
a strong correlation between f and Rec for RB 6 2.5. In this range
is symmetric in y, and Vcl is zero. However, if an instability occurs, of RB, flow with a lower Rec demands a higher mean pressure gra-
the symmetry breaks up and Vcl grows in time. Physically, this dient, and vice versa. However, such a correlation is not obvious for
means that the absolute flow rate across the centerline increases. RB > 2.5; friction factor monotonically decreases with increasing RB,
Therefore, Vcl can serve as a measure of the instability (Roberts, indicating a diminishing influence of baffles on the flow.

Fig. 6. Streamlines of the unsteady flow during one period for RB = 1.456 and Re = 130; (a) t = 0T, (b) t = 1/5T, (c) t = 2/5T, (d) t = 3/5T, (e) t = 4/5T and (f) t = 5/5T.
 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49 43

260 0.4
240
220
0.3
200
180
Rec

⎢A ⎢
0.2
160
140
120 0.1
100
80 0
1 1.5 2 2.5 3 3.5 4 60 80 100 120
RB t

Fig. 7. Critical Reynolds number for Hopf bifurcation plotted against RB. Fig. 9. The growth and saturation of the amplitude of perturbation mode for
RB = 1.456 and Re = 130.

130 0.3 0.30

Rec
0.25
f
120 0.20
0.25
0.15
Rec

110 0.10
f

0.2 0.05

100 0.00
0.00 0.05 0.10 0.15

0.15
90 Fig. 10. The derivative of the amplitude logarithm plotted against the square of the
1.5 2 2.5 3 3.5 4 amplitude for RB = 1.456 and Re = 130.
RB

Fig. 8. Variations of friction factor at Re = 130 and the critical Reynolds number for 4.3. Onset of the secondary instability
Hopf bifurcation depending on RB.

4.3.1. Floquet stability analysis

4.2. Nonlinear characteristics of the primary bifurcation
To determine the nonlinear characteristics of the primary insta-
bility (i.e. Hopf bifurcation), the following Landau model proposed
by Landau and Lifshitz (1976) as a series expansion, was used.
dA
¼ ðr þ ixÞA  lð1 þ icÞjAj2 A þ    :
dt
Here, A(t) represents the ‘‘complex’’ amplitude of the perturba-
tion mode from the base flow. The constant r is the linear growth
rate of the perturbation, and x represents the angular oscillation
frequency during the linear growth of the perturbation. The Landau
constant is denoted by c. The sign of l determines the type of tran-
sition. A positive value of l implies a supercritical (non-hysteretic)
transition, while a negative value indicates a subcritical (hyster-
etic) one (Sheard et al., 2004). In the current investigation, the
smoothed amplitude of Vcl was taken as the amplitude of the per-
turbation mode (|A|), and its growth and saturation are shown in
Fig. 9. The sign of l is determined practically by the behavior of
dlnjAj=dt near |A|2  0. Fig. 10 presents the dlnjAj=dt plotted against
|A|2 computed for RB = 1.456 and Re = 130. The vertical axis inter-
cept gives the linear growth rate of Vcl, and the slope at |A|2 = 0
determines the nonlinear bifurcation characteristics. A negative
slope is consistent with a supercritical bifurcation, whereas a posi-
tive slope denotes a subcritical bifurcation (Sheard et al., 2004).
The current transition turned out to be a supercritical bifurcation
(Fig. 10).
The onset of the secondary instability leading to a 3D flow can
be detected by a Floquet linear stability analysis (Schatz et al.,
1995; Barkley and Henderson, 1996) in which an instantaneous
velocity field of the baffled channel flow is decomposed into a 2D
base flow with a period T (U(x, y, t) = U(x, y, t + T)) and a three-
dimensional perturbation velocity field (u0 (x, y, z, t)) as follows:
uðx; y; z; tÞ ¼ Uðx; y; tÞ þ u0 ðx; y; z; tÞ: ð5Þ
ð4Þ
Substituting Eq. (5) into the Navier–Stokes and continuity equa-
tions, and then linearizing them, one can obtain the following gov-
erning equations for the perturbation velocity field,
r  u0 ¼ 0; ð6Þ
@u0 1 0
¼ r  ðu0 U þ Uu0 Þ  rp0 þ r2 u0 þ f : ð7Þ
@t Re
Here, the additional terms for the immersed boundary method
are also included. The no-slip condition is employed at all solid
boundaries including the ‘‘thin’’ baffles of zero thickness (u0 = 0),
while the perturbation velocity and pressure fluctuation are as-
sumed to be periodic in the streamwise direction (x) with a period
of L. By defining the operator L so that L(u0 ) is the right-hand side of
the linearized equation, Eq. (7) can be written symbolically as
@u0
@t
= L(u0 ). The general solution of this equation can be expressed
as a sum of Floquet modes, u ~ ðx; y; z; tÞ expðctÞ, where c(r + ix)
is the Floquet exponent and each Floquet eigenfunction u ~ is a
time-periodic function. To extract the periodic characteristics of
the perturbation, a normalized Floquet mode (u  ) is defined as
44 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49

¼u
u ~ expðixtÞ. Instability of the base flow U is determined by the 10 -5 2
Floquet multipliers, l  exp(cT); |l| > 1 indicates exponentially
growing perturbation. The Floquet multipliers can be obtained
from the eigenvalues of L; u ~ represents the corresponding eigen- 10 -6
1.5
functions. Since velocity and pressure fluctuations are assumed
to be homogeneous in the spanwise direction, they can be ex-
10 -7
pressed by an inverse Fourier transform in z as follows:

N ( t)
  Z   1
u0 1 ^
u
0
ðx; y; z; tÞ ¼ ðx; y; b; tÞeibz db; ð8Þ 10 -8
p ^
1 p
Multiplier
where b = 2p/k represents the spanwise wave number and k is the 0.5
10 -9 Norm
corresponding spanwise wavelength of a disturbance. Since Eqs.
(6) and (7) are linear, modes with different |b| can be decoupled.
The governing equations for each disturbance wave are similar to 10 -10 0
0 10 20 30 40 50
Eqs. (6) and (7), except for the replacement of the gradient operator
t
» with »b = (o/ox, o/oy, ib). Recently, a one-dimensional (1D) power-
type method was applied by Sheard et al. (2003) to estimate the Fig. 11. Temporal variations of the L2 norm of the perturbation velocity field and
maximum magnitude of the Floquet multipliers by computing the the Floquet multiplier for Re = 120 and b = 1.0.
following ratio:
0.3
jljmax  Nðt þ TÞ=NðtÞ; ð9Þ
Re =100
Re =106
where N(t) is the L2 norm of the perturbation velocity field at an in- 0.2 Re =110
Re =120
stant of time. This method was verified by Blackburn and Lopez Re =130
(2003). In this study, we use the power-type method in conjunction
with an immersed boundary method (Yoon et al., 2010) to calculate σ 0.1
the Floquet instability of the baffled channel flow. For the sake of
convenience, the term ‘‘Floquet multiplier’’ implies the one that 0
has the maximum magnitude among the Floquet multipliers from
now on, and the subscript ‘‘max’’ is dropped.
-0.1
Eqs. (6) and (7) were temporally and spatially discretized in the
same way as for the base flow (see Section 2). The 2D time-periodic 0 0.5 1 1.5 2
base flow was first computed with 192  192 cells in the x and y
β
directions, and thirty-two snapshots were saved over exactly one
period of the fully developed periodic flow. They were fed to Eq. Fig. 12. Growth rate plotted against spanwise wave number at Reynolds numbers
(7), being Fourier interpolated at each time step. For the Floquet as shown.
stability analysis, a numerical resolution of 192  192 cells in the
number for mode II tends to slightly decrease as Re increases. In
x and y directions was also used.
the case of Re = 130, both mode I and mode II are unstable, but
In this section, we fix RB = 1.456 as in the previous studies
mode I dominates.
(Roberts, 1994; Yang, 2000). In Fig. 11, temporal variations of
Another unstable mode (hereafter called ‘mode III’) appears in
N(t) and |l| are presented for Re = 120 and b = 1.0. Random noise
the range of 6 < b < 9 with further increasing Re (see Fig. 14). The
was used to initialize the perturbation velocity fields (u0 ) at t = 0.
critical Reynolds number for mode III is found to be approximately
After an initial decay, the L2 norm of the perturbation velocity field
148 with the corresponding b = 7.71. It should be noted that mode
starts to grow at t  2, and establishes a linear growth at t  10
II is dominant for Re = 153, while mode III dominates in the case of
after which the Floquet multiplier becomes constant (|l| = 1.30).
Re = 187. The corresponding spanwise wave numbers are 3.15 and
Since |l| > 1.0, this particular mode (b = 1.0) turns out to be
7.48, respectively, which are in good agreement with the previous
unstable to 3D disturbance.
results of Yang (2000). By using full 3D simulations, he found that
Variation of the growth rate (r) with b are depicted in Fig. 12 for
the most unstable spanwise wave numbers are 3.14 for Re = 153
several Reynolds numbers near Rec = 103. The dashed line repre-
and 7.58 for Re = 187, respectively.
sents r = 0, meaning neutral stability. As Re increases, the range
To confirm the existence of modeI, an additional full 3D simula-
of unstable b becomes larger, and both the maximum value of r
tion was performed at Re = 120 with a computational domain of
and the corresponding value of b increase. For example, the most
which the spanwise size (W) was 4pH. Flow was assumed to be
unstable modes are b = 1.01 for Re = 120 and b = 1.063 for
periodic in the spanwise direction (z), and a numerical resolution
Re = 130, respectively. It should be noted that there exist some
of 192  192  512 cells (in the x, y and z directions, respectively)
unstable waves of small b (i.e. large wavelengths) even for the Rey-
was employed. An instantaneous velocity field of the 2D base flow
nolds numbers close to the critical Reynolds number for the 2D
was expanded in z, and combined with a small-amplitude random
Hopf bifurcation. This trend was also found in some other flows
noise (of order 107) to be an initial flow field for the 3D simula-
with wall-mounted obstacles (Amon and Patera, 1989).
tion. We adopted the same criterion for the secondary instability
Growth rates over a wider range of b are shown in Fig. 13 for
as in Yang (2000), that is,
Re = 120, 125, and 130. In the case of Re = 120, the growth rate is
less than 0.0 for all b greater than 1.6, confirming that only one Z H Z L
1
type of unstable mode (hereafter called ‘mode I’) can occur at a W cl ¼ jwðx; y; W=2Þjdxdy: ð10Þ
LH 0 0
low spanwise wave number at Re = 120. For a higher Re, however,
another unstable mode (hereafter called ‘mode II’) develops in the Here, Wcl represents an averaged magnitude of spanwise velocity
range of 3 < b < 5. The critical Reynolds number for mode II turned component (w), and is computed on the middle plane in z. Fig. 15
out to be approximately 125 as seen in Fig. 13. The spanwise wave presents the time evolution of Wcl. A linear growth is clearly
 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49 45

100
0.2 Re =120
Re =125 t ≈73.27
Re =130 10-1
t ≈82.21
0.1 t ≈91.13
10-2

0
σ

10-3

-0.1 10-4

10-5
-0.2
10-6
0 1 2 3 4 5 6 0 2 4 6 8 10
β β

Fig. 13. Growth rate plotted against spanwise wave number at Reynolds numbers Fig. 16. Energy spectra of spanwise Fourier modes at Re = 120, computed based on
as shown. the instantaneous spanwise profiles of w taken from the 3D simulations at
x = 0.728, y = 0.5.

0.8
Re =130
Re =145
0.6 Re =148
Re =153
Re =187
0.4
σ

0.2

-0.2

0 2 4 6 8 10 12 14
β

Fig. 14. Growth rate plotted against spanwise wave number at Reynolds numbers
as shown.

100
Fig. 17. Instantaneous vortical structures (Q contours) at a nonlinear stage from the
10-1 3D simulation at Re = 120.

10-2
4.3.2. Floquet modes
10-3
Flow structure of the mode of a given b can be visualized with
Wcl

10-4 the corresponding Floquet mode. Two-dimensional base flow is

10-5 periodic with a period T (U(x, y, t) = U(x, y, t + T)), and exhibits the
following RT symmetry (Reflection–Translation symmetry).
10-6
Uðx; y; tÞ ¼ Uðx; y; t þ T=2Þ Vðx; y; tÞ ¼ Vðx; y; t þ T=2Þ: ð11Þ
10-7
0 50 100 150 200 250 300
Here, U and V are the velocity components of U in the x and y direc-
t
tions, respectively. It should be noted that the center line is set as
Fig. 15. Wcl plotted against time from the 3D simulation. y = 0 for convenience in this section. Similarly, a Floquet mode also
exhibits an RT symmetry. Fig. 18 presents contours of the stream-
wise vorticity component (x ~ x ) of the Floquet mode corresponding
identified approximately between t = 40 and t = 100. After that, to Re = 120, b = 1.0 (mode I) during one time period of the base flow.
nonlinearity prevails, and the flow becomes fully saturated approx- Bright and dark contours represent positive and negative values,
imately after t = 200. A spanwise profile of w was taken at x = 0.728, respectively. As clearly seen in Fig. 18, x ~ x satisfies the following
y = 0.5, and fast Fourier transformed in z for three distinct instants odd RT symmetry.
(t  73.27, 82.21, 91.13) in the linear stage, yielding the dominant x
~ x ðx; y; tÞ ¼ x
~ x ðx; y; t þ T=2Þ: ð12Þ
sharp peak at b = 1.0 for all the cases as seen in Fig. 16. This result
is in good agreement with the Floquet analysis shown in Fig. 12. The odd RT symmetry of x
~ x is equivalent to the mode A of Barkley
Fig. 17 presents a top view of the instantaneous 3D vortical struc- and Henderson (1996) characterized by the following relations.
tures at a nonlinear stage (t = 333.6) by using contours of the second 8
< u
> ~ ðx; y; tÞ ¼ u
~ðx; y; t þ T=2Þ
invariant of the velocity gradient tensor (Q) (Jeong and Hussain,
1995). The dominant spanwise mode (b = 1.0) is clearly identified.
Mode A :
>
v
~ ðx; y; tÞ ¼ v~ ðx; y; t þ T=2Þ ð13Þ
: ~ ~ y; t þ T=2Þ:
wðx; y; tÞ ¼ wðx;
46 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49

Fig. 18. Contours of the streamwise vorticity component (x ~ x ) of the Floquet mode
during one period of the base flow for Re = 120 and b = 1.0; (a) t = 1/4T, (b) t = 2/4T,
(c) t = 3/4T and (d) t = 4/4T. Fluid flows from left to right. Bright and dark contours
represent positive and negative values, respectively. The contour levels are
arbitrary.
Fig. 19. Contours of the streamwise vorticity component (x ~ x ) of the Floquet mode
during one period of the base flow for Re = 125 and b = 4.0; (a) t = 1/4T, (b) t = 2/4T,
(c) t = 3/4T and (d) t = 4/4T. Fluid flows from left to right. Bright and dark contours
represent positive and negative values, respectively. The contour levels are
arbitrary.

Fig. 21 presents the instantaneous 3D structures of the domi-

In Fig. 19, contours of x
~ x are presented for the Floquet mode
nant modes by Q contours. Each flow field has been constructed
corresponding to Re = 125, b = 4.0 (mode II) during one time period by superposition of the perturbation velocity field corresponding
of the base flow, revealing an even RT symmetry as follows: to the given b that is close to the most unstable spanwise mode
at the given Re on the underlying 2D flow. Fig. 21a–c reveals the
x
~ x ðx; y; tÞ ¼ x
~ x ðx; y; t þ T=2Þ: ð14Þ vortical structures of modes I, II and III, respectively. Vortical struc-
tures generated alternatingly at the tips of the baffles propagate
The even RT symmetry of x
~ x is equivalent to the mode B of Barkley downstream, and develop the spanwise waviness as seen in Fig. 21.
and Henderson (1996) characterized by the following relations:

8 4.3.3. Secondary instability for RB = 1.0

~ ðx; y; tÞ ¼ u
< u
> ~ ðx; y; t þ T=2Þ
In this section, we study the secondary instability for another
Mode B :
>
v
~ ðx; y; tÞ ¼ v~ ðx; y; t þ T=2Þ ð15Þ case of RB = 1.0. Fig. 22 shows the time evolutions of N(t) for
:~ ~ y; t þ T=2Þ:
wðx; y; tÞ ¼ wðx; b = 2.8 and 4.0. Here, the steady flow at Re = 230 was employed
as the base flow. It should be noted that there is an unstable mode
Fig. 20 shows contours of x~ x of the Floquet mode correspond- (here referred to as mode S, b = 2.8) even though Re = 230 is below
ing to Re = 150, b = 7.7 (mode III) during one time period of the the critical Reynolds number of the primary instability (Rec 
base flow, revealing the same type of RT symmetry as in the 255.7, Hopf bifurcation) obtained for 2D flow. Fig. 23 presents
mode II. the growth rate of N(t) plotted against b for various values of Re.
 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49 47

Fig. 20. Contours of the streamwise vorticity component (x ~ x ) of the Floquet mode
during one period of the base flow for Re = 150 and b = 7.7; (a) t = 1/4T, (b) t = 2/4T,
(c) t = 3/4T and (d) t = 4/4T. Fluid flows from left to right. Bright and dark contours
represent positive and negative values, respectively. The contour levels are
arbitrary.

The critical Reynolds number for the 3D instability is identified as

Rec2 = 223.6 and the dominant spanwise wave number is consis-
tently found as b = 2.8 (corresponding to the spanwise wavelength,
k = 2p/b = 2.244). To confirm the existence of the 3D instability
mode (mode S) in the steady-state regime of 2D flow, we also con-
ducted a full 3D simulation in a way similar to that in the case of
RB = 1.456, and found that the steady flow at Re = 230 is indeed
destabilized by this 3D disturbance. In the linear stage of flow
evolution, the mode of b = 2.8 was also identified as the dominant
one in the full 3D simulation. The structure of mode S is depicted in
Fig. 24 where contours of streamwise vorticity (x ~ x ) of the Floquet
mode are presented for Re = 230, b = 2.8. The structure is steady
and completely symmetric with respect to the centerline as
Fig. 21. Instantaneous Q contours of the Floquet mode; (a) Re = 120 and b = 1.0, (b)
follows: Re = 125 and b = 4.0, (c) Re = 150 and b = 7.7. Each flow field has been constructed
x
~ x ðx; y; zÞ ¼ x
~ x ðx; y; zÞ: ð16Þ by superposition of the perturbation velocity field corresponding to the specific b on
the base flow.
48 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49

10 -7

β =2.8
β =4.0

10 -8
N (t )

10 -9

10 -10
0 10 20 30
t

Fig. 22. Temporal variations of the L2 norm of the perturbation velocity field for
Re = 230 and RB = 1.0.

0.4
Re =220 Fig. 25. Q contours of the Floquet mode for Re = 230 and b = 2.8, RB = 1.0. The flow
0.3 field was constructed by superposition of the perturbation velocity field corre-
Re =230
sponding to the specific b on the base flow.
Re =240
0.2 Re =250

5. Conclusion
0.1
σ

Flow instability in baffled channel flow, including the primary

0 and secondary instabilities, has been numerically studied. The baf-
fles, mounted periodically in the streamwise direction on both
-0.1 channel walls, were implemented on a Cartesian grid system by
an immersed boundary method. The critical Reynolds number for
-0.2 the primary instability depends upon RB, being minimum at
0 1 2 3 4 5
RB = 3.08. There exists a strong correlation between friction factor
β and the critical Reynolds number for RB 6 2.5. In this range of RB,
Fig. 23. Variations of growth rate with spanwise wave number.
flow with lower Rec demands a higher mean pressure gradient,
and vice versa. Floquet stability analysis has been performed for
RB = 1.456 in order to identify the most unstable spanwise mode
at a given Re. Three dominant modes (I–III) with distinct spanwise
wave numbers were identified. Each mode has its own RT symme-
try; odd RT symmetry for mode I, and even RT symmetry for modes
II and III. The most unstable spanwise wave number of each mode
obtained by the Floquet stability analysis is in good agreement
with the result obtained by full 3D simulation. Furthermore,
Floquet stability analysis has been performed for RB = 1.0 also. It
turned out that the 3D instability (mode S) appears when the base
flow is steady, and its Floquet mode is also steady. Our results shed
light on complete understanding of flow instability in baffled chan-
nel flow which has great applicability in engineering.

Fig. 24. Contours of the streamwise vorticity component (x ~ x ) of the Floquet mode
for Re = 230 and b = 2.8, RB = 1.0. Fluid flows from left to right. Bright and dark
contours represent positive and negative values, respectively. The contour levels are
arbitrary.
The corresponding 3D flow structure is presented in Fig. 25 where
only the lower half of the domain is shown for clarity. The Q con-
tours were obtained by superposition of the dominant Floquet
mode on the base flow. The spanwise distortion is clearly seen.
The mode S is different from modes I–III found in the case of
RB = 1.456 in the sense that it appears when the 2D base flow is
steady. Therefore, care must be taken in interpreting the stability
analysis for Hopf bifurcation, especially in the cases of small RB
where base-flow unsteadiness is suppressed by the baffles periodi-
cally placed at a short interval.
Acknowledgments
This work was supported by the National Research Foundation
of Korea (NRF) grant funded by the Korea government (MEST) (No.
2012R1A2A2A01013019).
References
Amon, C.H., Patera, A.T., 1989. Numerical calculation of stable three-dimensional
tertiary states in grooved-channel flow. Physics of Fluids A 1, 2005–2009.
Barkley, D., Henderson, R.D., 1996. Three-dimensional Floquet stability analysis of
the wake of a circular cylinder. Journal of Fluid Mechanics 322, 215–241.
Battaglia, F., Tavener, S.J., Kulkarni, A.K., Merkle, C.L., 1997. Bifurcation of low
Reynolds number flows in symmetric channels. AIAA Journal 35, 99–105.
Blackburn, H.M., Lopez, J.M., 2003. On three-dimensional quasi periodic Floquet
instabilities of two-dimensional bluff body wakes. Physics of Fluids 15, L57–
L60.
 C. Kang, K.-S. Yang / International Journal of Heat and Fluid Flow 38 (2012) 40–49
Cheng, C.H., Huang, W.H., 1991. Numerical prediction for laminar forced convection
in parallel-plate channels with transverse fin arrays. International Journal of
Heat and Mass transfer 34, 2739–2749.
Gilmanov, A., Sotiropoulos, F., 2005. A hybrid Cartesian/immersed boundary
method for simulating flows with 3D, geometrically complex, moving bodies.
Journal of Computational Physics 207, 457–492.
Herbert, Th., 1983. Subharmonic three-dimensional disturbances in unstable plane
shear flows. AIAA Paper, 83–1759.
Herbert, Th., 1988. Secondary instability of boundary layers. Annual Review of Fluid
Mechanics 20, 487–526.
Howes, T., Mackley, M.R., Roberts, E.P.L., 1991. The simulation of chaotic mixing and
dispersion for periodic flows in baffled channels. Chemical Engineering Science
46, 1669–1677.
Jeong, J., Hussain, F., 1995. On the identification of a vortex. Journal of Fluid
Mechanics 285, 69–94.
Kim, J., Moin, P., 1985. Application of a fractional-step method to incompressible
Navier–Stokes equations. Journal of Computational Physics 59, 308–323.
Kim, J., Kim, D., Choi, H., 2001. An immersed-boundary finite-volume method for
simulations of flow in complex geometries. Journal of Computational Physics
171, 132–150.
Landau, L.D., Lifshitz, E.M., 1976. Mechanics, third ed. Pergamon.
Mittal, R., Dong, H., Bozkurttas, M., Najjar, F.M., Vargas, A., von Loebbecke, A., 2008.
A versatile sharp interface immersed boundary method for incompressible
flows with complex boundaries. Journal of Computational Physics 227, 4825–
4852.
49
Roberts, E.P.L., 1994. A numerical and experimental study of transition processes in
an obstructed channel flow. Journal of Fluid Mechanics 260, 185–209.
Schatz, M.F., Barkley, D., Swinney, H.L., 1995. Instability in a spatially periodic open
flow. Physic of Fluids 7, 344–358.
Sheard, G.J., Thompson, M.C., Hourigan, K., 2003. From spheres to circular cylinders:
the stability and flow structures of bluff ring wakes. Journal of Fluid Mechanics
492, 147–180.
Sheard, G.J., Thompson, M.C., Hourigan, K., 2004. From spheres to circular cylinders:
non-axisymmetric transitions in the flow past rings. Journal of Fluid Mechanics
506, 45–78.
Su, S.W., Lai, M.C., Lin, C.A., 2007. An immersed boundary technique for
simulating complex flows with rigid boundary. Computers and Fluids 36, 313–
324.
Yang, K.S., 2000. Numerical investigation of instability and transition in an
obstructed channel flow. AIAA Journal 38, 1173–1178.
Yang, J., Balaras, E., 2006. An embedded-boundary formulation for large-eddy
simulation of turbulent flows interacting with moving boundaries. Journal of
Computational Physics 215, 12–40.
Yang, K.S., Ferziger, J.H., 1993. Large-eddy simulation of a turbulent obstacle flow
using a dynamic subgrid-scale model. AIAA Journal 31, 1406–1413.
Yoon, D.H., Yang, K.S., Choi, C.B., 2010. Flow past a square cylinder with an angle of
incidence. Physics of Fluids 22, 043603.
You, J., Choi, H., You, J.Y., 2000. A modified fractional step method of keeping a
constant mass flow rate in fully developed channel and pipe flows. Journal of
Mechanical Science and Technology 14, 547–552.