RESEARCH ARTICLE | MARCH 01 2022

An investigation of bluff body flow structures in variable

velocity flows
K. P. Sarath ; K. V. Manu 

Physics of Fluids 34, 034102 (2022)

https://doi.org/10.1063/5.0083743

CrossMark

 
View Export
Online Citation

Articles You May Be Interested In

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

A review of experiments on stationary bluff body wakes
Physics of Fluids (January 2022)

Stabilization of Reacting Fluid by Bluff Body

AIP Conference Proceedings (June 2010)

Effects of swirl number and bluff body on swirling flow dynamics

AIP Advances (February 2023)
 Physics of Fluids ARTICLE scitation.org/journal/phf

An investigation of bluff body flow structures

in variable velocity flows
Cite as: Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743
Submitted: 29 December 2021 . Accepted: 12 February 2022 .
Published Online: 1 March 2022

K. P. Sarath and K. V. Manua)

AFFILIATIONS
Department of Aerospace Engineering, Indian Institute of Space Science and Technology, Thiruvananthapuram 695547,

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

Kerala, India

a)
Author to whom correspondence should be addressed: [email protected]

ABSTRACT
The present study explores three-dimensional vortex-dynamics past a wall-attached bluff body kept in a variable velocity field with numerical
simulations. A trapezoidal pulse of mean velocity, consisting of acceleration phase from rest followed by constant velocity phase and deceler-
ation phase to rest, is imposed at the inlet of the computational domain similar to the experimental study of Das et al. [“Unsteady separation
and vortex shedding from a laminar separation bubble over a bluff body,” J. Fluids Struct. 40, 233–245 (2013)]. For a wide range of Reynolds
numbers (96  Reb  2390), acceleration Reynolds numbers (196  Rea  978), and deceleration Reynolds numbers (310  Red  1522),
different stages of flow evolution are systematically analyzed. The flow evolution starts with the formation of a primary vortex followed by a
two-dimensional circular array of spanwise vortex tubes by inflectional shear-layer instability. At a sufficiently high Reynolds number, the
shear layer vortices originated from two-dimensional fluctuations deformed by three-dimensional instabilities, giving fragmented streamwise
vorticity. In addition, long-wavelength “tongue-like structures” and short-wavelength “rib-like structures” are evident near the top wall and
the bluff body, respectively. The streamwise vorticity generation equation indicates that the spanwise vortex tubes initially tilt, resulting in
streamwise vorticity, further amplified by the vortex stretching process. The distinct flow features, including mode shape, frequency, and
growth rate associated with the shear-layer instability, are identified using the dynamic mode decomposition (DMD) algorithm. Using the
maximum growth rate criteria, the DMD technique successfully separates the coherent shear layer modes associated with two-dimensional
shear layer instability from the flow field.
Published under an exclusive license by AIP Publishing. https://doi.org/10.1063/5.0083743

I. INTRODUCTION internal flow in hydraulic devices and physiological flows (blood flow
Several computational, theoretical, and experimental studies have in the aorta), the sudden closure of valves in pipe flows, turbo-
explored flow dynamics of a bluff body maintained in a constant machinery control, vortex shedding from ship funnels, and wind flow
velocity field due to numerous practical applications. Extensive reviews over hills.
elucidating vortex dynamics of bluff body wake held in a uniform flow The wall-mounted bluff bodies kept in time-dependent boundary
field utilizing diverse standpoints have been reported in the studies by conditions can produce a fascinating gamut of flow structures. In such
Roshko,2 Williamson,3 and Thompson et al.4 Conversely, the flow cases, the thickness of the boundary layer formed over the solid surface
dynamics of the bluff body in oscillatory (zero mean) and pulsating varies in both space and time, causing a high level of complexity in the
flows (non-zero mean) have received relatively lesser attention. path of vortical structures formed from the shear/boundary layer.
The periodic coherent structures formed from a wall-mounted Numerical simulation studies are extensively used to understand the
bluff body under transient inlet conditions can influence the perfor- development of the vortical flow field and the subsequent generation
mance of many engineering systems. Various practical situations show of secondary structures.5–7 However, most numerical studies on flow
flow separation and associated instabilities originating from the bluff transition in transient inflow conditions are performed on academic
bodies. For example, the vortex shedding process from the inhibitor in geometries. Numerically predicting the transitional flows in complex
a solid rocket motor is an essential concern for design engineers. Such geometries poses a significant challenge.
vortices generated from the inhibitors can lead to thrust-oscillations in The effect of Reynolds and Strouhal numbers on the formation
launch vehicles and damage the payload. Other examples include of the primary and secondary vortices on a constricted channel with

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-1

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

time-dependent inlet flow conditions (sinusoidal variation) was stud- side of the body and the size of the separation bubble increased with
ied by Tutty and Pedley5 and Rosenfeld.6 A comparison of boundary flow time (Fig. 1). At the moderate Reynolds number (Rem  700),
layer transition and coherent structures between steady and unsteady the shear layer becomes unstable, leading to the formation of shear
flows was performed by Costamagna et al.8 They posited that the ele- layer vortices. At the high Reynolds number (Reh  2300), the shear
mentary process causing the turbulence in oscillatory flows is analo- layer breaks down quickly. A fascinating circular array of shear layer
gous to the steady flow cases. vortices was observed on the circumference of the primary large-scale
Many aspects of flow transition in oscillatory and pulsating flows vortex. Many features of the shear layer breakdown mechanism were
are delineated through the analytical solution, exponent, and transient reminiscent of the two-dimensional Bloor–Gerrard instability/second-
instability analyses. Hall and Parker9 conducted a quasi-steady linear ary vortices14,15 observed in steady inflow cases.
stability analysis of the decaying flow in a suddenly blocked channel. This study considers transitional flow past a similar complex
Detailed instability analysis of inflectional velocity profiles formed dur- obstacle, as shown in Fig. 1. The challenge in efficiently representing
ing the decelerating phase of the pulsating pipe flow was conducted by transitional flows in complex geometry is addressed by using an
Das and Arakeri.10 Good quantitative agreement with experiment and immersed boundary method in conjunction with higher-order com-
linear instability analyses was obtained for decelerating flows contain- pact schemes. The computationally economical immersed boundary
ing inflection points.10 They posited that the flow breakdown could be method will allow imposing complex geometries by using a Cartesian
due to inflectional inviscid instability. In a study with the same spirit, grid. The free-stream velocity varies temporally similar to the unsteady

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

Das et al.11 reported a detailed flow visualization in a shallow angle dif- water tunnel experimental study of Das et al.1 However, Das et al.1
fuser for trapezoidal piston motions. Ghidaoui and Kolyshkin12 rein- did not explore three-dimensional wake dynamics or the effect of
terpreted some of the experimental findings of Das and Arakeri10 acceleration and deceleration Reynolds numbers on flow instability.
based on linear and weakly nonlinear instability analysis. Nayak and Unlike sinusoidal variation of the mean flow, trapezoidal piston
Das13 recently performed a non-modal analysis of unsteady internal motion allows the analysis of flow transition in constant acceleration/
flows to investigate their stability characteristics. deceleration flow conditions. Here, the boundary layer is contingent
Das et al.1 conducted experiments in a closed-loop unsteady on adverse pressure gradient because of both the temporal and spatial
water tunnel to study the initiation of separation from a bluff body for components of the pressure gradient, which can obscure the flow
the trapezoidal-type mean flow with flow time in a channel flow. The dynamics and escalate the level of complexity.
flow configuration and salient flow features of the experiments by Das It is generally challenging to identify the origin of three-
et al.1 are depicted in Fig. 1. In the experiments of Das et al.1 at the low dimensional instability and calculate stability parameters originating
Reynolds number case (Rel  100), the flow separated in the leeward from a complex bluff body kept in a time-varying flow using classical

FIG. 1. Schematic depiction of the main flow pattern at three different Reynolds numbers as in the study of Das et al.1 Experimental figures are reproduced with permission
from Das et al., “Unsteady separation and vortex shedding from a laminar separation bubble over a bluff body,” J. Fluids Struct. 40, 233–245 (2013). Copyright 2013 Elsevier.1

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-2

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

stability theories. In this work, the dynamic mode decomposition instability is discussed in Sec. III B. Sections IV A–IV C discuss the
(DMD) algorithm and the k2 vortex identification method are used to effects of Reynolds number, acceleration Reynolds number, and decel-
determine the origin and stability parameters of the flow. To our eration Reynolds number on wake dynamics, as well as a comprehen-
knowledge, the DMD algorithm applied to accelerating or decelerating sive comparison of coherent structure dynamics from DMD with
flows is not available in the literature. In a three-dimensional flow con- spectral analysis, respectively. Finally, the results are summarized in
taining multiple lengths and time scales, it is difficult to understand Sec. V.
the instability mechanisms and educe the coherent structures mainly
II. METHODOLOGY
responsible for the energy. Such complex flows containing multiple
active degrees of freedom can be simplified by first identifying the A. Flow configuration and numerical method
dominant modes and then deriving a dynamical model for the sys- The solution of the flow problem is obtained by solving the gov-
tem’s time evolution. Dynamic mode decomposition is employed for erning equations for three-dimensional and incompressible viscous
decomposing the coherent structures and flow patterns, whose flow in the following form:
dynamics provide a more compact method for describing the flow
r  u ¼ 0; (1)
process. The DMD method has been previously applied successfully to
the investigation of many experimental and numerical simulation @u 1
¼ rp  ½rðu  uÞ þ ðu  rÞu þ r2 u þ f : (2)

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

data.16–18 In a recent review paper by Rowley and Dawson,19 different @t 2
model reduction techniques were reviewed for analyzing various fluid The computational flow solver INCOMPACT3D22,23 is used for
flow problems. As observed in previous studies, it is hard to select the solving the governing equations. This code is extensively used for
dominant modes in unsteady inflow conditions using the DMD algo- many transitional and turbulent flow studies. Extensive validation of
rithm since we are not sure whether to prioritize modes’ energy, fre- the simulation methods with values in the literature can be found else-
quency, or growth rate. Recently, DMD algorithms are implemented where.22,24 INCOMPACT3D is based on a compact sixth-order finite
successfully to identify coherent structures and stability parameters in difference scheme for spatial discretization25 with spectral treatment
many pulsating flow conditions. For instance, Jang et al.20 used ampli- for the pressure equation and a third-order Adams–Bashforth scheme
tude criteria to analyze coherent flow features and forcing frequency in for time advancement. The body is prescribed in the computational
the oscillatory flow around a cylinder. In another recent work, Liu domain by adding a body-force field in the momentum equation by
et al.21 identified the coherent flow structures in the cloud cavitating following the procedure proposed by Laizet and Lamballais22 and
flow around the hydrofoil using norm values obtained from the DMD Laizet et al.24 The Poisson equation is solved using spectral methods
algorithm. Dynamic mode decomposition can yield spatial modes by the modified wave-number formalism.22 In order to avoid spurious
based on their frequency values, making it a convenient technique for pressure oscillations due to the introduction of immersed boundary
investigating systems involving multiple frequency scales, as in the method, the pressure mesh is staggered from the velocity mesh by half
present study. a mesh.
This paper is organized as follows: Sec. II describes the numerical Figure 2 shows the schematic diagram of the three-dimensional
procedure and the theoretical formulation used for generating time- computational domain and boundary conditions used for the present
dependent inflow conditions. The results are discussed in Secs. simulation. The complex-shaped bluff body used in the study of Das
III–IV C. Validation of the numerical solver using previous experi- et al.1 (Fig. 2) is modeled using the virtual or immersed boundary tech-
mental results and grid independence analysis is presented in Sec. III. nique. The desired bluff body is formed by combining three circular
The three-dimensional instability mechanism after shear-layer arcs with a rectangular geometry. The maximum dimension of the

FIG. 2. Computational domain and boundary conditions: (a) three-dimensional view and (b) details of the bluff body.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-3

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

bluff body is 8R  6R, where R is the radius of the top circular t

up ðtÞ ¼ Up for 0  t  t0
geometry. No-slip condition on top and bottom walls is assigned t0
through the Dirichlet velocity condition. The free-slip condition is ¼ Up for t0  t  t1
applied to the right and left boundaries. At the outlet (x ¼ lx), the
ðt2  tÞ
imposed exit velocity is deduced by the following equation: ¼ Up for t1  t  t2
  ðt2  t1 Þ
@u  @u  ¼ 0 for t > t2 : (3)
þU ¼ 0:
@t x¼lx @x x¼lx
Since the velocity profiles will be fully developed at the bluff loca-
tion, providing analytical solutions helps in reducing the computa-
B. Inflow generation tional costs. Also, the most straightforward way of giving the mean
Figure 3 shows the trapezoidal temporal inflow variation used in flow at the inlet of the computational domain and calculating the
the present study. Such temporal variation in the mean flow is akin to velocity profiles was computationally demanding. The general analyti-
the piston motion in the novel experiments of Das et al.1 The single cal solution of velocity profiles for trapezoidal mean flow variation in
pulse configuration [Fig. 3(a)] has four phases, as expressed in Eq. (3). the two-dimensional channel-flow can be expressed as an infinite
In the initial phase of the trapezoidal pulse ð0 < t < t0 Þ, the mean series.26 To reduce the size of the computational domain and reduce

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

flow undergoes a constant-acceleration followed by a constant-velocity the computational cost, the time-dependent boundary condition from
phase ðt0 < t < t1 Þ and a constant-deceleration phase ðt1 < t < t2 Þ. the analytical solution given by Das and Arakeri26 is imposed at the
The mean-velocity is zero for t > t2 . Hence, the effect of acceleration inlet of the computational domain [Eq. (4)]. Usage of the analytical
and deceleration can be separately studied in phases one and three, solution has been found computationally robust.
respectively, For single pulse cases, the analytical solutions are as follows:

  
 
u 1 KA2 2K X1 v2 t
nh
for 0  t  t0 ; ¼ A1 t   e K  W;
U p t0 40 t0 nh¼1
   
u 2K X 1 v2 t
nh
v2 ðtt0 Þ
nh
for t 0  t  t1 ; ¼ A1  e K e K
 W;
Up t0 nh¼1
0     1
v2 t v2 ðtt0 Þ v2 ðtt1 Þ
  X @e K  e
1
nh nh nh
u t2  t KA2 K
e K A (4)
for t1  t  t2 ; ¼ A1 þ  2K   W;
Up t2  t1 40ðt2  t1 Þ nh¼1
t 0 t 2  t 1
0       1
v2 t v2 ðtt0 Þ v2 ðtt2 Þ v2 ðtt1 Þ
nh nh nh nh
u X1 @
e K
e K
e K
e K A
for t t2 ; ¼ 2K   W;
Up nh¼1
t0 t2  t1
" #
3  h2 cosðch vnh Þ  cosðvnh Þ
where A1 ¼ 1  c2h ; A2 ¼ 5c2h  6c2h þ 1; K¼ ; W¼ :
2  v3nh sinðvnh Þ

Here, h is the channel half height, y is the distance from the centerline profiles obtained analytically during different phases are shown in Fig.
toward the wall of the channel, ch ¼ hy and vnh , where nh ¼ 1, 2, 3, …, 1 3(b). Although the mean velocity is zero in the final phase, the analyti-
are roots of tanðvÞ ¼ v. Here, the first 50 roots of tanðvÞ ¼ v are used to cal velocity profiles are non-zero because of the residual momentum
obtain the sum of the above converging infinite series. carried from the previous stages.
The analytical solution of the fully developed laminar pulsating
flow for the variant volumetric flow rate is obtained using the Laplace C. Physical and numerical parameters
transform technique together with the Bromwich integral formula. In
The instantaneous flow time t is non-dimensionalized by t2
this technique, the governing partial differential equations with vari-
(t ¼ tt2 ). Non-dimensionalized streamwise distance is defined by
able u(x, t) are converted into ordinary differential equations with the
Laplace transforms [~ u ðx; sÞ]. The ordinary differential equation is x ¼ X0:2
2R . Wall normal distance is non-dimensionalized by the body
Y
~ ðx; sÞ, and the function is inverted to yield u(x, t) using the
solved for u height (y ¼ 6R ). The following definitions are used to define the
Bromwich integral formula. Details of the derivations of analytical sol- Reynolds number (Reb), acceleration Reynolds number (Rea), and
utions and procedures are provided in the Appendix. The velocity deceleration Reynolds number (Red),

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-4

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 3. Inflow boundary condition: (a) temporal variation of mean velocity and (b) inlet velocity profiles (u ¼ u=Up ) at four different flow times.

Up b Numerical experiments were carried out for different flow

Reb ¼ :
 parameters, as shown in Table I. The computational domain with
dimensions Lx  Ly  Lz ¼ 60R  15R  28:4R is discretized using
A standard non-dimensional number for studying flow dynamics in
accelerating flows is the acceleration Reynolds number.27,28 Finaish uniform grid points. The grid independence was assessed by compar-
et al.28 successfully analyzed the dependency between the vortex pat- ing the skin friction coefficient values at the midplane in the spanwise
terns development and the acceleration Reynolds number over an direction during the constant velocity phase for grids with
accelerating airfoil started from rest. In an accelerating flow, 513  129  241 (grid A), 1025  257  481 (grid B), and
1501  385  721 (grid C) elements. The percentage of deviation
Up ðtÞb atb between grids A and B was 10%, while it was below 1% for grids B
Rea ¼ ¼ ; and C. Figure 4(a) shows the streamwise variation of the skin friction
 
coefficient for grids A, B, and C. Results were reproduced with a negli-
where
gible difference between grids B and C, as evidenced in Fig. 4(a). The
Up Reynolds number employed in the grid independence study was the
a¼ b ¼ 2R:
and highest. Consequently, grid B is used for further simulations.
t0
qffiffi qffiffiffiffiffi Time step dependency analysis was performed by using three dif-
3
By taking a convective time scale t ¼ ba; Rea ¼ ab
2 : ferent time steps of 103 , 104 , and 105 s. In terms of accuracy and
computational economy, a time step of 104 s (CFL ¼ 0.02) was found
The same procedure is followed in a decelerating flow to obtain
to be satisfactory. Therefore, a time step of 104 s is used for all the
rffiffiffiffiffiffiffi cases. The numerical procedure is validated by comparing the vortex
db3
Red ¼ ; formation time (tv) and separation angle (b) with experimental results
2
of Das et al.1 Here, the separation angle is defined as the angle between
where the separation point and the axis of symmetry of the bluff body (taken
Up to the left of the axis of symmetry). The present simulations calculate
d¼ :
t2  t1 TABLE I. Simulation parameters.
The non-dimensional vortex formation time1 is represented by
Case Up ðm=sÞ Reb t0 ðsÞ t1 ðsÞ t2 ðsÞ Rea Red
1 Utv tv
tv ¼ : C1 0.0048 96 0.2 29.9 30.3 438 310
2 b
C2 0.0359 718 7.5 8.92 11.92 196 450
Here, tv is the time when the vortex begins to form, and Utv is the
C3 0.0717 1434 1 4.33 5.33 756 756
mean inlet velocity at tv. The following non-dimensional time in terms
of acceleration and deceleration parameters are defined using the fol- C4A 0.0956 1912 2.66 3.91 5.25 536 756
lowing relation: C4B 0.0956 1912 1.33 3.25 4.58 756 756
rffiffiffi C4C 0.0956 1912 1.33 3.25 3.84 756 1138
a
s¼t ; C4D 0.0956 1912 1.33 3.25 3.58 756 1522
b
C4E 0.0956 1912 0.8 2.98 4.32 978 756
t  t1 ðt  t1 Þd
td ¼ ¼ : C5 0.1195 2390 1.67 2.6 4.27 756 756
t2  t1 Up

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-5

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 4. Grid independence analysis and validation of numerical scheme: (a) streamwise variation of the skin-friction coefficient (Cf) for different grid sizes, (b) comparison of
vortex formation time, and (c) comparison of separation angle.

the vortex formation time by plotting the streamlines at different flow been discussed, especially in transitional and turbulent flows. These

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

instances. In contrast, Das et al.1 used high-speed dye visualization factors might account for the difference.
techniques to measure the vortex formation time. Figures 4(b) and
4(c) show the comparison of simulation results with experimental III. COHERENT STRUCTURES
results of Das et al.1 The simulation result lies within a 6% deviation A. Two-dimensional flow structures
from experimental results, as shown in Figs. 4(b) and 4(c). The simula-
tion result lies within a 6% deviation from experimental results, as First, we present numerical visualizations at different flow instan-
shown in Figs. 4(b) and 4(c). The slight difference between numerical ces in x–y planes for three Reynolds numbers (cases C1, C2, and C5).
simulation and experiment could be because of the different methods The flow parameters are precisely the same as in the water tunnel
used to evaluate the vortex formation time. Also, there could be minor experiments of Das et al.1 (cases X, I, and XI of Das et al.1). Here, the
differences in the inlet conditions between an experiment and simula- evolution of flow structures from the bluff body at three different
tion. It is not easy to maintain a perfect trapezoidal motion in an Reynolds numbers (Reb ¼ 96, 718, and 2390) is evidenced by the
experiment, which causes inlet fluctuations in the water tunnel. In instantaneous contour plot of spanwise vorticity (xz) in the center
many studies, computational noise related to numerical methods has plane of the spanwise direction (z ¼ 14:2R) as shown in Figs. 5, 6(a),

FIG. 5. Flow evolution at the low Reynolds

number (case C1). Experimental figure is
reproduced with permission from Das et al.,
“Unsteady separation and vortex shedding
from a laminar separation bubble over a bluff
body,” J. Fluids Struct. 40, 233–245 (2013).
Copyright 2013 Elsevier.1

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-6

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 6. Instantaneous spanwise vorticity depicting the flow evolution: (a) medium Reynolds number (case C2 and (b) high Reynolds number (case C5).

and 6(b). In our numerical visualization, at the low-Reynolds number “fingers”) and the interconnections between vortices in the cylinder
(Reb ¼ 96), the recirculation region grows with the flow time precisely wake experiments of Gerrard29 are also evident.
like the experimental study, as shown in Fig. 5. Here, the formed struc- Figure 7 illustrates the velocity vector field plotted over the span-
ture is laminar and two-dimensional with no sign of shear-layer insta- wise vorticity at different phases for case C5. Numerous velocity vectors
bility. For similar flow conditions, the present simulations show good are skipped in x and y directions from the obtained simulations for the
agreement with experimental observation. clarity of the image. Rolled-up laminar separation bubbles (B1 and B2)
At the moderate Reynolds number (Reb ¼ 718), the initial forma- formed over the bluff body and primary vortex are shown in Figs. 7(a)
tion of the primary vortex [marked as SL0 in Fig. 6(a)] is similar to the and 7(b). The vortices created by the shear-layer instability (SL1-SL6) do
low Reynolds number case [Fig. 6(a), t ¼ 0:28 and 0.39]. A pair of not show closed or spiraling streamline patterns, as shown in Figs. 7(b)
steady laminar separation bubbles formed over the leeward upper side of and 7(c). Coherent structures formed during the decelerating phase near
the bluff body vortices (marked as B1 and B2) is identifiable at the top and bottom wall proximity are shown in Figs. 7(d) and 7(e),
t  0:55. At this Reynolds number, the flow shows shear-layer instabil- respectively. During this deceleration period, the magnitude of other
ity. The primary vortex oscillates at flow time t  0:68. The origin of components of velocity escalates abruptly [Figs. 7(d) and 7(e)].
oscillations is near the steady separation bubbles. The shear layer roll-up
causes the formation of numerous small-scale vortices (marked as SL1 to B. Three-dimensional topology
SL4), which move around the primary vortex (t ¼ 0:92). The newly
formed shear-layer vortices eventually dominate the flow field. The Simulations in two and three dimensions are compared to
sequence of vortices formed from the shear layer is analogous to that of a better understand the nature of the shear-layer instability mechanism.
plane mixing layer (t ¼ 0:92). Flow features near the top and bottom Figure 8 shows the flow evolution of two-dimensional and three-
wall change significantly during the deceleration phase (t ¼ 0:92 and dimensional simulations for the high Reynolds number (Reb ¼ 2390).
1). In addition, deceleration retards the motion of the vortex core (SL0). The flow features are analogous for two-dimensional and three-
Deceleration creates an adverse pressure gradient and destabilizes the dimensional simulations during acceleration and constant-velocity
boundary layer. Local boundary layer separations are observed on top phases (t ¼ 0:58 and 0.94). Here, two-dimensional perturbations are
and bottom walls (WT1 and WB1, respectively). The flow deceleration inherently unstable compared to three-dimensional perturbations dur-
enhances the detachment of vortices from the near-wall vorticity layer ing the initial phase. The perturbation amplifies with time and causes
(B2 and WB1). In later flow-time, these counter-vorticity regions shed a train of vortices originating from the two-dimensional shear-layer
into the core flow from the top and bottom walls (WB1 and WT1). instability mechanism, hence can be captured with two-dimensional
A similar sequence of events is evidenced at a relatively high simulations. With increasing flow time, three-dimensional effects
Reynolds number (Reb ¼ 2390) with a higher shear layer shedding fre- appear, which could be due to the secondary instability of the sepa-
quency, as depicted in Fig. 6(b). The shear layer roll-up and shedding rated shear layer as seen in many steady flow transition conditions.30,31
of vortices (initiated from the steady separation bubbles B1 and B2) For case C5, the two-dimensional simulation results reasonably agree
are observed at t  0:44. At this Reynolds number, the shear layer with the experiment results up to the deceleration phase. A flow transi-
breaks down quickly. A fascinating circular array of shear-layer vorti- tion from a two-dimensional to a three-dimensional state can initially
ces (SL1 to SL5) is observed along the circumference of the primary be observed near the wall proximity. During the decelerating and zero
large-scale vortex (t ¼ 0:60). Multiple folding (mentioned as mean-velocity phases, three-dimensional flow characteristics, such as

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-7

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 7. Detailed view of the coherent structures observed during different phases of the flow evolution: (a) separation bubble over the bluff body, (b) early stage of vortex shed-
ding, (c) an instance of periodic vortex shedding, (d) top-wall boundary layer separation, and (e) bottom-wall boundary layer separation.

axial stretching, tilting of vortices, and subsequent instabilities, in span- We further examine the instability’s three-dimensional nature
wise directions are dominant. Positive vortices are formed from the using the lambda 2 method32 in the three-dimensional flow field. This
body, which interacts with the core flow and eventually breaks down. method is rigorously applied to accurately depict the topology of vor-
Due to this inherent limitation, the final disintegration of the vortices is tex cores in diverse classes of flows, including the wake patterns behind
not captured by two-dimensional simulations. Drastic differences in bluff bodies, transitional, and turbulent flows.4 For direct numerical
flow dynamics occur during the zero mean-velocity phases (t ¼ 1:17). simulation (DNS) data, k2 has been found to assess the size of the

FIG. 8. Comparison of three-dimensional

and two-dimensional simulations for high
Reynolds numbers (case C5): (a) two-
dimensional simulation and (b) three-
dimensional simulation.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-8

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

vortex core and the number of small scales. In this method, coherent structures. Fragmented streamwise structures and small-scale eddies
vortex structures in the flow field are recognized using the second larg- generated from the spanwise vortex rolls are evident at later flow time
est eigenvalue of the symmetric tensor created from the symmetric (t ¼ 1:23). The origin of small-scale structures from the shear rolls is
and anti-symmetric components of the velocity gradient tensor. observed in the zero mean-velocity phases.
Detailed analysis of this method is available in the work of Jeong and The topology of streamwise vorticity is depicted in Fig. 10(b).
Hussain.32 Fundamentally unstable characteristics of strained vortex sheets cause
Isometric views of the vortices identified using the k2 method at the formation of streamwise vorticity. The wavelength of the stream-
different flow instances for the medium Reynolds number (case C2) wise vortices (the spanwise distance between adjacent streamwise vor-
are shown in Fig. 9. The isosurfaces are colored according to span- tex pairs) is measured for all the spanwise rolls. Long-wavelength,
wise [Figs. 9(a) and 9(b)] and streamwise vorticity (xx) magnitude tongue-like structures are observed near the top wall. Short-
[Figs. 9(c) and 9(d)] to show the orientation of the three- wavelength, rib-like structures, characteristic of the mode B instability
dimensional structures. At t ¼ 0:88, the flow field mainly consists in the cylinder wake study, are observed near the bluff body, as shown
of two-dimensional shear-layer structures [Figs. 9(a) and 9(c)]. The in Fig. 10(c). For mode B structures in the cylinder wake, a spanwise
value of streamwise vorticity is minimal at this flow instant. The val- ffi R was proposed by Mansy et al.34 The span-
wavelength of kz  p40ffiffiffi
Re
ues are slightly higher near the side boundaries. The amplification of wise wavelength for mode B matches with the value obtained from cyl-

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

the streamwise vorticity (xx) components [Figs. 9(b) and 9(d)] indi- inder wake studies. Signatures of modes A and B are evident in the top
cate the flow transition from a two-dimensional to a three- view contours of spanwise vorticity Fig. 10(c). The process of a three-
dimensional state. Spanwise modulations of streamwise vorticity dimensional breakdown associated with streamwise vorticity is
and the tendency to generate three-dimensional structures are depicted in Fig. 10(d). The difference in the spanwise wavelength is
evident at later flow-time. The shear-layer eddies from two- evident in the top views. Streamwise vortices convected from the bot-
dimensional instabilities are deformed by three-dimensional tom wall and generated from the top wall due to the boundary layer
instabilities, as shown in Figs. 9(b) and 9(d). Alternate negative and separations dominate the flow field.
positive streamwise vorticity regions are observed in the spanwise In steady flow cases, similar modes are observed, indicative of
direction [Fig. 9(d)]. In general, since the dynamics alter in time, secondary instability. Hence, in the present case, two and three-
conditions conducive to three-dimensional instability are formed at dimensional secondary perturbations may lead to three-dimensional
the end of the decelerating phase. Here, the birth of streamwise vor- flow transition. A combination of more than one secondary instabil-
ticity is due to the progressive development of the spanwise velocity ity mechanism is present during the later flow time, as indicated by
component as the flow-time increases. Three-dimensional modula- mode A and mode B structures. The spanwise deformation of the
tions are not evident in the immense core primary vortex (SL0). separation bubble observed near the top wall is caused by the elliptic
Iso surfaces of k2 structures colored by spanwise vorticity at dif- instability of vortex cores, which results in a wavelength of the order
ferent flow instances for high Reynolds numbers (case C5) are shown of the vortex size (mode A). This mode (mode A) occurs when
in Fig. 10(a). During the deceleration phase, the flow changes from a three-dimensional disturbances are amplified in regions of two-
two-dimensional state to a three-dimensional one [Figs. 10(a)–10(d)]. dimensional elliptical streamlines. The flow structures that occur
Coherent three-dimensional modes are observed in simulations, and near the bottom wall are also characterized by three-dimensionality
the structures are similar to the other wake-transition studies.4,33 The that potentially arise due to the instability of the flow arising between
spanwise vortex rolls evolved by shear-layer instability and near-wall two consecutive vortices. Here, two-dimensional vortices become
vortices encounter axial stretching. Dominating three-dimensional distorted in the spanwise direction soon after the vortex pairing and
flow instabilities of the vortex sheets are realized for near-wall struc- eventually break up into small-scale vortices. The related structure
tures. Clear distinctions in shape and spanwise wavelength are (mode B) maintains a shorter spanwise wavelength than those
observed between the bottom-wall and top-wall formed flow amplified through mode A.

FIG. 9. Three-dimensional structures identified using k2 criteria, colored by vorticity, for medium Reynolds numbers (case c2): (a) xz contour at t ¼ 0:881, (b) xz contour at
t ¼ 1:535, (c) xx contour at t ¼ 0:881, and (d) xx contour at t ¼ 1:535.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-9

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 10. Three-dimensional structures identified using k2 criteria, colored by vorticity at different flow instances, for high Reynolds numbers (case C5): (a) colored by xz, (b)
colored by xx, (c) top-view colored by xz at t ¼ 1, and (d) top-view colored by xx at t ¼ 1.

IV. INFLUENCE OF NON-DIMENSIONAL NUMBERS The inflectional velocity profiles causing the instability are plotted
ON FLOW EVOLUTION at locations near the bluff body are plotted in Fig. 11. The velocity pro-
A. Effects of the Reynolds number on flow evolution files observed downstream of the bluff body are the superposition of
(cases C3, C4B, and C5) the confluent mixing layer (wake and shear layer) and boundary layer.
Inflectional streamwise velocity profiles are observed in all the cases
1. Instabilities of inflectional shear layers [Figs. 11(a)–11(c)]. Shear-layer instability is originated in the vicinity
The effects of the Reynolds number on flow dynamics are sys- of the inflection point on the velocity profile. This two-dimensional
tematically analyzed by keeping the acceleration and deceleration instability results in an intermittent formation of a two-dimensional
Reynolds numbers constant (C3, C4B, and C5). Important flow circular array of small vortices (SL1, SL2, …, etc.). Inflectional nature
parameters obtained from the simulation are tabulated in Table II. of flow instability is further investigated using the second derivative of
The values from the simulations generally match well with the experi- the velocity profile. Figures 11(d)–11(f) show the second derivative of
mental results. the non-dimensional velocity profiles for cases C3, C4B, and C5,

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-10

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

TABLE II. Quantitative comparison with experimental results of Das et al.1

fh havg
Case Reb tv tv; exp b bexp tsl fV fh fDMD Up
C3 1434 0.395 96 0.466 23 25.45 28.37 1.18 5.46 4.94 4.43 0.718
C4B 1912 0.396 96 0.449 25 34.33 31.14 1.50 7.5 7.32 7.33 0.659
C5 2390 0.395 17 0.447 23 29.69 30.61 1.82 10.00 10.20 9.27 0.469

ð h2  
respectively. Here, Figs. 11(d)–11(f) show inflection points for all the u u
cases, which are marked with circles in Figs. 11(a)–11(c). These h¼ 1 dy:
h1 Umax Umax
inflected points are in viscidly unstable and prone to flow instability.
As seen from the two-dimensional contour graphs of the instanta- Instantaneous momentum thickness oscillates at a frequency (fh ),
neous spanwise vorticity (Fig. 6), the flow oscillations originate at these which is nearly equal to the frequency obtained from the time-series

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

inflection points. Hence, the flow instability appears to develop at data of the vertical velocity (fv), as
 shown in Fig. 12(d). The non-
fh havg
these inflection points in velocity profiles. dimensionalized frequency Up based on average momentum
Some typical time series of vertical velocity are plotted for three
thickness is tabulated in Table II. However, here the values does not
Reynolds number cases to probe the nature of the flow unsteadiness,
remain constant.
as shown in Fig. 12(a). Here, the measurement point is near the origin
of shear layer oscillations. The shear-layer shedding process can be
seen from the time-series data of vertical velocity. The shear-layer 2. Dynamic mode decomposition of flow features
instability vortices/waves have a definite frequency, especially at a high
Reynolds number. The peak frequency of velocity fluctuation is deter- DMD analyses are conducted to understand the shear-layer
mined using spectral analysis [Fig. 12(b)]. Further time-series analysis instability mechanism further and identify the coherent structures
is made by changing the location of the measurement points. The more clearly. The method proposed by Schmid17 is used for analysis,
newly selected measurement positions are at downstream locations which is based on the snapshots for decomposing the fluid structures.
but inside the shear layer shedding regions. Similar time-series sequen- DMD uses regression of data onto locally linear dynamics
ces are obtained in the new measurement locations. xkþ1 ¼ Axk , where A is chosen to minimize jjxkþ1  Axk jj2 over
The local momentum thickness of the shear layer is calculated by k ¼ 1, 2, 3, …, N1. DMD algorithm starts with a sequential arrange-
removing the data near the top and bottom wall using the below ment of the flow field data pertaining to each instant of time. Snapshot
expression data are arranged to matrix format as

FIG. 11. Streamwise velocity profiles for different cases (x ¼ 2:3047): (a) case C3, (b) case C4B, and (c) case C5. Second derivative profiles of the velocity profile: (d) case
C3, (e) case C4B, and (f) case C5.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-11

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 12. Temporal variation of velocity, momentum thickness, and corresponding frequency spectra. (a) Wall normal velocity, (b) frequency spectra of wall normal velocity, (c)
temporal variation of momentum thickness, and (d) frequency spectra of momentum thickness.

2 3

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

u11 u21  uN1
1
Initially, the DMD algorithm is applied to three-dimensional ver-
6 1 7 tical velocity data on a sub-domain closer to the bluff body. Details of
6 u2 u22  uN1 7
6 2 7 the time step, duration, and the number of snapshots for all the cases
UN1
1 ¼ U 1 ; U 2 ; U 3 ; U 4 …U N1 ¼6 . .. .. .. 7:
6 .. . . . 7 are tabulated in Table III. The total number of snapshots selected is
4 5
large enough to ensure the DMD result converge. The obtained results
u1M u2M  uN1
M for case C5 are shown in Fig. 13. The Ritz values presented in
Singular value decomposition of U is computed as Fig. 13(a) show the stability of the modes. Here, the real and imaginary
parts of the eigenvalues are taken as horizontal and vertical axes,
U  XKV : (5) respectively. Nearly, all the values are clustered near the unit circle,
hinting that the dynamics settle on an attractor. The growth/decay
By the pseudoinverse of X, matrix A is obtained
rate and frequencies of different modes are shown in Fig. 13(b). Now,
A ¼ U0 VK1 X : (6) the mode having a frequency nearly equal to the peak frequency (fv) is
searched from the DMD analyses. Interestingly, the extracted peak fre-
A low-dimensional linear model of the dynamical system is computed quency of vertical velocity (fv) from the spectral analysis matches with
for computational economy the DMD mode with the highest growth rate, as shown in Fig. 13(b).
~ ¼X AX ¼ X U0 VK1 : The contours of DMD modes obtained from the analysis are
A (7) now arranged according to various criteria. The contours of the first
~ is a r  r projection of the full matrix into POD modes. Eigen
A three modes sorted based on amplitude, norm, and growth/decay rate
~ matrix gives
decomposition of A are shown in Figs. 13(c)–13(e), respectively. The highest mode
obtained using amplitude, norm, and growth/decay rate are indicated
~ ¼Wk:
AW (8) by the subscript A, N, and r, respectively, as shown in Fig. 13(b).
Coherent structures with alternate positive and negative regions can
DMD modes can be obtained by reconstruction of A from W,
be seen in the contours. For simplicity, in many studies, the modes are
U ¼ U0 VK1 W: (9) sorted based on the value of amplitude. However, a mode with initially
high amplitude can generally decay fast [first mode in Fig. 13(c)].
Amplitude, angular frequency, and growth rate of the modes are calcu- Also, such a method can ignore a mode with a relatively low amplitude
lated by but fast growth. In the second DMD mode, the frequency is equal to
the shedding frequency. The third mode displays a similar distribution
A ¼ U† x; (10) to the first two but with a lower frequency.
xm ¼ log Im ðkÞ=Dt; (11)
rm ¼ log Re ðkÞ=Dt; (12)
TABLE III. Dynamic mode decomposition analysis.
where Dt is the time separation between successive frames/snapshots,
A is the amplitude, xm is the angular frequency, rm is the growth
rate, / is the individual mode, and x is the corresponding left eigen- No. of Frequency
Case tstart tfinal Dt snapshots rmax (frmax )
vector of X. Frequency of each mode can be derived from angular fre-
quency using the following relation: fDMD ¼ x2pm . Approximate solution C3 1.6 2.6 0.01 100 0.0163 4.43
for future time is given by C4B 1.0 2.5 0.01 150 0.0714 7.33
X
r C5 1.87 2.60 0.01 73 0.0142 9.34
uðtÞ ¼ /k expðrm;k þ ixm;k ÞtÞA k ¼ U expðXtÞA : C5(3D) 1.87 2.60 0.01 73 0.0132 9.34
k¼1

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-12

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 13. Three-dimensional DMD analysis for case C5: (a) Ritz circle, (b) growth-rate vs frequency, (c) leading modes based on amplitude values, (d) leading modes based on
norm values, and (e) leading modes based on growth rate values.

In Fig. 13(d), the modes are sorted Frobenius norms values. structures with relatively low frequency. The third mode is the same as
Here, the mode with the shedding frequency is not among the first the first mode obtained using amplitude criteria. The first three modes
three modes derived by the norm calculation [Fig. 13(d)]. It is interest- with the highest growth rate are shown in Fig. 13(e). The first two
ing to note that using norm criteria, the first two modes are large-scale modes frequency is nearly equal to the shedding frequency obtained

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-13

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

from the spectral analysis. Here, the third mode is near the second- the primary vortex structure. The location of the low-pressure point
harmonic of the fundamental shedding frequency. In comparison with for three acceleration Reynolds numbers at various flow times are cal-
the first two modes, the spatial scale of the third mode is relatively culated (as marked in Fig. 16). The following non-dimensional param-
small. Hence, DMD analysis separated the shear layer oscillations eters are defined to track the core trajectory:
from the flow field using the growth rate criteria. Similar results are
xc  0:2
obtained by repeating the DMD analysis with spanwise vorticity data. Xc ¼ :
DMD analyses are repeated for two-dimensional data in the x-y 8R
plane (at the center of spanwise locations and full domain) obtained Figure 15 shows the variation of non-dimensionalized vortex core
pffiffi
from three-dimensional simulations for cases C3, C4B, and C5. The position (Xc) against the non-dimensionalized time scale (s ¼ t ab).
two-dimensional DMD analysis is nearly identical to the three- At early flow times (s < 3), a reasonable collapse of the location of the
dimensional DMD analysis, as shown in Table III. The growth rate of core of primary eddy is obtained with the new scaling, as shown in
different modes is obtained from the DMD analysis and plotted in Fig. Fig. 15. This regime corresponds to the time of collapse (s < 3), is
14(d). For all the cases, the fv value matches with the DMD mode with two-dimensional, and is unaffected by shear layer oscillations. With
the highest growth rate, as shown in Fig. 14. Here, all the leading the increase in flow time (s > 3), the core trajectory significantly
modes exhibit a positive growth rate. For all the cases, alternate vertical departs from each other. Furthermore, it appears that the shear-layer
velocity fluctuations are observed in the leading mode [Fig. 14(d)].

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

instabilities and three-dimensional structures cause this departure, and
Shear layer oscillations in low Reynolds numbers (case C3) are smaller they are analyzed in detail below.
and less developed than those in higher Reynolds numbers. For cases Flow evolution for three different acceleration Reynolds cases at the
C4B and C5, the leading mode tracks the secondary shear layer vortex same non-dimensional flow time (s) is shown in Fig. 16. At the early time
movement around the primary vortex. (s ¼ 2.75), the primary vortex core lies precisely at the same point for all
three cases. For the case of the lowest acceleration Reynolds number Rea
B. Effects of the acceleration Reynolds number ¼ 536, as shown in Fig. 16(a1), the flow does not generate shear-layer vor-
on flow evolution (cases C4A, C4B, and C4E) tices. At this flow time, for medium and high acceleration Reynolds num-
The effects of the acceleration Reynolds number on the flow fea- ber cases (Rea ¼ 756 and Rea ¼ 978), the shear-layer instability triggers a
ture are analyzed for three cases with varying Rea by keeping Reb and series of small-scale vortices, as shown in Figs. 16(b1) and 16(c1). The local
Red constant (cases C4A; C4B, and C4E). The development of the separation bubble on the top wall is formed further downstream for higher
vortex core point (xc) trajectory for the primary vortex is traced for all Rea cases [Figs. 16(a2), 16(b2), and 16(c2)]. Another important feature
the Rea cases. The core is identified by probing the location with worth mentioning is that the shear-layer instability features are observed
instantaneous minimum local pressure in a two-dimensional x–y earliest in the high acceleration case, while three-dimensional instability
plane. Here the low-pressure criterion effectively captures the core of features are observed in the low acceleration case. The flow structures near

FIG. 14. Two-dimensional DMD analysis for three Reynolds numbers (cases C3, C4B, and C5): (a)–(c) growth rate against frequency and (d) modes with the highest growth
rate.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-14

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

In dimensional time scale, the higher Rea case progresses in front

of the lower Rea case, resulting in an early three-dimensional structure.
The transition from 2D to 3D is initially observed at a low Reynolds
number (in the non-dimensional time frame). The change is evi-
denced by the iso-surfaces of k2 structures (colored by spanwise vortic-
ity, at s ¼ 9:48), as shown in Fig. 17(a). Coloring with spanwise
vorticity enables the identification of the shear-layer instabilities and
wall vortices with a clear distinction. At the higher Rea case, three-
dimensional flow breakdown has not been initiated at this flow
instance. The three-dimensional evolution is observed for positive vor-
tices shed from the top and bottom wall vortices for Rea ¼ 536 and
Rea ¼ 756 cases. However, at this flow time, the structures formed by
shear-layer instabilities essentially exist in a two-dimensional form, as
shown in Fig. 17(a). Spanwise waviness is clearly observable for this
case, which in later flow time breakdowns into smaller vortices.

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

The development of three-dimensionality can be analyzed by
looking at the evolution of streamwise vorticity. Figure 17(b) shows
surface contours k2 superimposed by xx. Alternate positive and nega-
tive patches of streamwise vorticity are observed over shear layer rolls
FIG. 15. Variation of the vortex center position for cases C4A, C4B, and C4E. for higher Rea cases. This indicates the initiation of 3D breakdown.
For Rea ¼ 756, the formation of wiggles in spanwise rolls coincides
the walls are relatively complex at low acceleration Reynolds number cases with regions of intense streamwise vorticity, as shown in Fig. 17(b).
[Figs. 16(a3), 16(b3), and 16(c3)]. However, small-scale eddy formation Hence, vortex stretching is a crucial factor in producing streamwise
near the bottom and top walls under the influence of the local adverse vorticity. A similar behavior is observed for the vortex structures on
pressure gradient is observed for all cases. the top wall. Production of streamwise vorticity (xx) results in vortex

FIG. 16. Flow evolution for different Rea cases: (a) C4A (b) C4B, and (c) C4E.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-15

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 17. Three-dimensional structures identified using k2 criteria colored by vorticity for different acceleration number cases (C4A, C4B, and C4E): (a) contours of xz and (b)
contours of xx.

stretching and eventual breakdown of the two-dimensional vortex frequency decreases with an increase in the acceleration Reynolds
structure. number. DMD analyses are repeated for different cases. Table IV
A quantitative comparison of vortex formation time, non- shows the maximum growth rate and the corresponding frequency
dimensionalized vortex formation time, and separation angle is made obtained from DMD analysis.
in Table IV. Here, the non-dimensional vortex formation time (tv ) is
nearly constant. By using the boundary layer parameters, a universal
C. Effects of the deceleration Reynolds number
time scale of separation independent of Reynolds number and acceler-
ation Reynolds number is obtained. The two-dimensional shedding on flow evolution (cases C4B, C4C, and C4D)
frequencies are obtained from spectra analysis of vertical velocity. As 1. Evolution of flow structures during the deceleration
discussed in Sec. IV A, the instantaneous momentum thickness oscil- phase
lates at a frequency, that is, nearly equal to the frequency obtained
from the time-series data of the vertical velocity. The values of peak To illustrate the effects of deceleration on flow features, three
frequency estimated are listed in Table IV. Here, the peak shear layer cases are analyzed with varying Red by keeping Reb and Rea as con-
stants. The effects of deceleration are analyzed by plotting instanta-
neous vorticity
 at different non-dimensionalized flow times
TABLE IV. Effects of the acceleration Reynolds number on flow features. td ¼ ttt1
2 t1
, as shown in Fig. 18. Figure 18 shows vorticity data at the

Case Rea tv ðsÞ tv b tsl ðsÞ fh fV rmax frmax starting point (td ¼ 0), quarter one (td ¼ 0:25), quarter two
(td ¼ 0:5), and at the end of deceleration phase (td ¼ 1).
C4A 536 0.66 0.391 38 22.03 2.86 8.07 8.57 0.0380 8.26 The flow dynamics of the primary vortex (SL0) and shear-layer
C4B 756 0.47 0.396 96 34.33 1.50 7.32 7.5 0.0714 7.33 vortices are nearly similar at td ¼ 0:25 for all the cases. However, the
C4E 978 0.37 0.408 99 29.69 0.93 6.25 7.04 0.1047 6.93 deceleration Reynolds number effects are seen in the dynamics of wall
vortices. In the case of smooth deceleration (at low Red), the

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-16

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 18. Evolution of spanwise vorticity during different deceleration phases: (a) C4B, (b) C4C, and (c) C4D.

detachment of vortices from the wall (B2 and WT1) is observed at an wall vorticity can be explained via a two-dimensional parent-offspring
early non-dimensionalized deceleration period (td ¼ 0.5), as shown in mechanism. Figure 19 illustrates the top and bottom wall vortex for-
Fig. 18(a3). A close-up view of the detachment of the separation bub- mation sequence. Initially, during the acceleration phase, the thickness
ble is shown in the subplot in Figs. 18(a3), 18(b3), and 18(c3). of the vorticity sheet near the top wall [positive xz ðtÞ] increases with
As the primary vortex decelerates, the separated secondary struc- flow time [Figs. 19(a1) and 19(a2)]. The advection and upward propa-
tures (WT1) in the top surface boundary layer develop rapidly, and gation of the primary vortex and shear layer structures cause lifting of
rapid boundary layer growth occurs near the secondary eddy (Wt1). the near-wall vorticity sheets, as depicted in Fig. 19(a3). Convective
Under the influence of the primary vortex structure, further small-scale vortices induce a region of adverse pressure gradient in the boundary
eddy structures soon develop near the top and bottom wall due to the layer in front of the moving vortex. Here, if the primary vortex is close
adverse pressure gradient induced by the vortex. At the end of the decel- enough to the wall and remains for a sufficient period, the lifting of
eration period [td ¼ 1, Figs. 18(a4), 18(b4), and 18(c4)], an eruption of vortex sheets and vortex-induced boundary layer separation is
the viscous flow near the top wall (near WT1) is observed for the low observed. During the deceleration phase [Fig. 19(a3)], patches of nega-
Red case [Fig. 18(a4)]. For medium and high Red cases, the deceleration tive vorticity are formed near the wall. Separation occurs in the form
causes an increase in bubble volume (WT1 and WB1), but at the end of of small size eddy, which is of the opposite rotation to the primary
the deceleration period, the vortical region would not wholly separated convective vortex. Further deceleration leads to the roll-up of vorticity
from the surface [Figs. 18(b4) and 18(c4)] as in the low Red case. sheets and detachment of vortices from the top-wall surface and ulti-
mately provokes an eruption [Fig. 19(a4)]. However, in two-
dimensional simulations, the outbreak of small-scale vortices does not
2. Vorticity generation mechanism
occur. Due to the mutual action of moving vortices and temporarily
Finally, the formation of wall vortices and the disintegration varying boundary layers, a gamut of complex and separation effects is
mechanism of the three-dimensional structures observed during the observed in the wall proximity. The vortices detach from the wall cul-
deceleration phase are discussed. The generation and roll-up of near- minating in fluid ejection from the wall region.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-17

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 19. An illustration of the flow feature near the wall: (a) top-wall and (b) bottom-wall.

A schematic of the sequence of events near the bottom-wall prox- advection stretching
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflffl}|fflffl{
imity is depicted in Fig. 19(b). The bottom-wall vortex generation
@xx @xx @xx @xx @u
mechanism is similar to the top-wall separation mechanism discussed ¼ u v w þ xx
above. However, the flow structures are complicated by the curvature @t @x @y @z @x
tilting
of the wall-attached bluff body. Here, the motion of the primary vortex zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{ diffusion
causes lifting of the near-wall vortices, as sketched in Fig. 19(b1). zfflfflffl}|fflfflffl{
@v @u @w @u
Opposite-signed vorticity rolls up near the curvature side of the bluff þ  þ r2 xx : (14)
@x @z @x @y
body immediately underneath the negative signed primary vortex
(SL0). An increase in the vertical flow component causes convection The production of xx by tilting can occur by two mechanisms. The
@v
of wall-generated positive vorticity (WB1) into the core flow. This is first mechanism (Ti1) is due to the tilting of @x by the spanwise shear
evidenced by the propagation of newly formed positive vorticity. (@u ). The second mechanism (Ti ) is the tilting of @w
@z 2 @x by the wall-
When the convected vortices are too close, merging/pairing of positive @u
normal shear (@y ). Hence, as evident from Eq. (14), the spanwise end-
vortices are observed [Fig. 19(b2)]. With further deceleration, intense,
complex local interactions between merged positive wall-generated conditions (tilting terms) significantly affect the three-dimensional
vorticity with near-wall negative vorticity are observed. The passage of vorticity mechanism. It is important to note that 3D simulations are
near-wall vortices and interactions with opposite sign vortices cause performed using periodic boundary conditions to eliminate the
the formation of new small-scale structures [Fig. 19(b4)]. For all cases boundary layer effects from the side walls. Such boundary conditions
considered, three-dimensional effects are observed during the late are intentionally imposed to match the inflow analytical boundary
stages of the decelerating phase. Further new simulations with conditions obtained by neglecting the spanwise velocity variation. As
extended steady or accelerating phases might manifest three- revealed in several experimental studies33,35 in the past, the end condi-
dimensional evolution in the early pulse phase. tions in the wind/water tunnel affect the coherent structures. Hence,
As observed in many experiments and numerical studies, the the absence of a sidewall boundary layer changes the three-
flow transition from two-dimensional to three-dimensional can be dimensional dynamics, which could be different from that obtained
related to the generation of streamwise vorticity. In order to under- with a no-slip sidewall boundary condition. Due to the strain-field cre-
stand the formation of three-dimensional structures, the incompress- ated by the evolving spanwise vortex array and the wall-effect, the vor-
ible form of the vorticity equation is considered analyzed tices initially undergo tilting resulting in the formation of streamwise
vorticity. The net streamwise vorticity production further increases by
stretching tilting
zfflffl}|fflffl{ zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{ diffusion the vortex stretching process.
zfflfflffl}|fflfflffl{ In order to analyze the contribution of each term in Eq. (14), the
Dxx @u @u @u
¼ xx þ xy þ xz þ r2 xx : (13) temporal variations of stretching and tilting terms at points near
Dt @x @y @z
the top wall (x ¼ 6:7; y ¼ 2:34) and bottom wall (x ¼ 2:26; y
In Eq. (13), the first term on the right side represents the production ¼ 0:167) are plotted in Figs. 20(a) and 20(b), respectively. These
of xx due to the stretching of vorticity-line elements. The following points are inside the three-dimensional structures formed at the end of
two terms describe the vortex line “tilting.” The last term in Eq. (13) is deceleration period. A substantial increase in tilting and stretching
the diffusion of vorticity due to viscous effects. For three-dimensional terms is observed during the end of the deceleration phase as evident
flows, Eq. (13) can be rewritten as in Figs. 20(a) and 20(b). Deceleration induces velocity variation along

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-18

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

FIG. 20. Temporal variation of streamwise vorticity generation in case C5: (a) stretching and tilting terms at a point near the bottom wall, (b) stretching and tilting terms at a
point near the top wall, (c) tilting components at a point near the bottom wall, and (d) tilting components at a point near the top wall.

the streamwise direction, affecting the stretching term. In the contour flow Reynolds number, unsteadiness has its effects on the flow-feature
plots, the strong presence of stretching terms is observed over the crip- formation. Three-dimensional vortex structures were identified using
pling parts of vortices. The tilting mechanism can be seen as a resul- k2 criteria and analyzed using spanwise and streamwise vorticity.
tant of two components [Eq. (13)]. Figures 20(c) and 20(d) show the The results obtained from three-dimensional simulation suggest
temporal evolution of the tilting term components at points near the the following picture for the development of wakes. Flow features,
top and bottom walls, respectively. The difference in temporal evolu- such as laminar separation bubble, shear layer oscillations, and associ-
tion is observed in each flow phase for each tilting component. The ated rollup, are observed during constant acceleration or constant
major contributors are the spanwise vorticity and the spanwise shear velocity phase. Subsequently, due to shear-layer instabilities, a two-
during the initial stages. Due to deceleration, the spanwise vorticity dimensional circular array of spanwise vortex tubes is developed from
magnitude reduces in the later flow phase while the wall-normal vor- the primary vortex. During the deceleration phase, the evolving span-
ticity and wall-normal shear components increase. Simultaneous to wise vortex rolls and induced near-wall circulating eddies to encounter
three-dimensional breakdown, a spike in stretching and second tilting axial stretching and flow transitioned from a two-dimensional to
term components can be identified. three-dimensional state. Vortex merging and strongly localized erup-
tions near the top and bottom walls are observed. The small-scale
V. SUMMARY AND CONCLUSIONS structures formed by the eruption penetrate the external flow field and
The work presented here attempts to understand wake dynamics subsequently mix with the large-scale spanwise vortex rolls.
in variable velocity, which is commonly observed in many engineering The effect of the Reynolds number on shear layer frequency was
situations. The flow employed to study the dynamics was a channel analyzed using momentum thickness of inflectional velocity profiles,
flow with the wall attached bluff body kept in a trapezoidal pulse of velocity spectra, and DMD analysis. A non-dimensionalized time scale
mean velocity inflow. This configuration is ideal for studying wake (s) is used to track the primary vortex core in varying acceleration
dynamics in practical situations since it contains a relatively complex cases. The low acceleration Reynolds number case showed early sign
bluff body and trapezoidal pulse, enabling a systematic study on the of 2D–3D transition in the non-dimensional time frame. Variations in
effects of acceleration and deceleration separately. deceleration effects affect the formation of wall vortices, which, in
The effects of Reynolds numbers, acceleration Reynolds numbers, turn, results in the development of three-dimensional structures.
and deceleration Reynolds numbers on flow dynamics were systemati- Qualitative and quantitative comparison of the flow evolution in dif-
cally studied. Though flow evolution has a strong dependence on the ferent deceleration cases strongly supports this observation.

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-19

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

" #
Some information concerning the shear layer instability mechanism ðeky þ eky Þ
is obtained using DMD analysis. Important mode shapes, frequencies,  ðy; sÞ ¼ /p 1  kh
u : (A6)
ðe þ ekh Þ
and growth/decay rates were identified using dynamic mode decomposi-
tion. Modes are arranged based on amplitude, norm, and growth rate To find the particular integral /p , we use the volume flow rate
values. The dominant mode, as determined by amplitude criteria, shows condition
a high decay rate, whereas the dominant mode, as determined by the
ðh " #
norm value, is a large-scale structure with relatively low frequency. With ðeky þ eky Þ
the DMD analysis, the primary and secondary harmonics associated /p 1  kh dy ¼ up ðsÞh: (A7)
0 ðe þ ekh Þ
with the shear-layer instability are obtained, and flow characteristics are
separated from the flow field using the growth rate criterion. By integrating and simplifying the above equation, we obtain /p ,
 ðy; sÞ as
and applying this in (A6), we get u
AUTHOR DECLARATIONS
Conflict of Interest ðekh þ ekh Þ  ðeky þ eky Þ
 ðy; sÞ ¼ up ðsÞ
u : (A8)
The authors have no conflicts to disclose. 1
ðekh þ ekh Þ  ðekh  ekh Þ
kh
DATA AVAILABILITY

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

The data that support the findings of this study are available Here, the trapezoidal variation of the mean velocity with time
from the corresponding author upon reasonable request. [up ðtÞ] is defined by Eq. (3). The analytical solution of the transient
velocity profile is obtained by calculating the Laplace transform of
Eq. (3) and substituting into Eq. (A8), succeeded by the inverse
APPENDIX: INFLOW GENERATION—ANALYTICAL Laplace transform calculation. The Laplace transform of the mean
SOLUTIONS velocity is given by
To obtain the inflow velocity profiles for trapezoidal mean flow Up
up ðsÞ ¼ for 0  t  t0
variation, we consider incompressible, fully developed transient fluid t0 s 2
flow with the constant thermophysical properties in an infinitely long 
Up 1 et0 s
rectangular channel with a height of 2h and zero transverse velocities. ¼  2 for t0  t  t1
t0 s2 s
Under the assumptions of uni-directionality and in the absence of body   t s 
forces, the momentum equation is given by Up 1 et0 s Up e 1
¼   for t1  t  t2
  ! t0 s 2 s 2 t2  t1 s2
@u 1 @P @2u        t2 s 
¼ þ : (A1) Up 1 et0 s Up et1 s Up e
@t q @x @y2 ¼  2  þ
t0 s2 s t2  t1 s2 t2  t1 s2
The boundary conditions are as follows: for t2  t  1:

uðh; tÞ ¼ uðh; tÞ ¼ 0 ðno slipÞ; (A2) The inverse Laplace transform of the equations is found by applying
@ ðuð0; tÞÞ Mellin’s inverse formula
uðy; 0Þ ¼ 0; ¼ 0; (A3)
@y ð
ðh 1 cþi1 ðekh þ ekh Þ  ðeky þ eky Þ
uðy; tÞ ¼ up ðsÞ  est ds:
uðy; tÞ@y ¼ up ðtÞh: (A4) 2pi ci1 kh kh
1 kh kh
0
ðe þ e Þ  ðe  e Þ
kh
Taking the Laplace transform of the momentum equation [Eq. (A9)
(A1)], the boundary conditions [Eqs. (A2)–(A4)] yield
Here, c denotes a vertical contour in the complex plane so that all
d2 u
 s  1
1 dP the singularities of the function are to the left of it.36 The Cauchy
¼
 u þ uðy; 0Þ; residue theorem is used to evaluate the complex integral. Here, the
dy2  l dx 
poles are found to be s ¼ 0 and tanðvÞ ¼ v. By adding all the residue
 ðh; sÞ ¼ u
u  ðh; sÞ ¼ 0; terms, the final solution as in Eq. (4) is reached.
u ð0; sÞ
@ (A5)
¼ 0;
@y REFERENCES
ðh 1
S. Das, U. Srinivasan, and J. Arakeri, “Unsteady separation and vortex shedding
 ðy; sÞ@y ¼ up ðsÞh:
u from a laminar separation bubble over a bluff body,” J. Fluids Struct. 40,
0
233–245 (2013).
2
The general solution of Eq. (A5) is A. Roshko, “Perspectives on bluff body aerodynamics,” J. Wind Eng. Ind.
Aerodyn. 49, 79–100 (1993).
 ðy; sÞ ¼ c1 eky þ c2 eky þ /p ;
u 3
C. H. Williamson, “Vortex dynamics in the cylinder wake,” Annu. Rev. Fluid
pffiffi Mech. 28, 477–539 (1996).
where /p is the particular integral, where k ¼ s . Applying the no- 4
M. C. Thompson, T. Leweke, and K. Hourigan, “Bluff bodies and wake-wall
slip and symmetry boundary conditions, interactions,” Annu. Rev. Fluid Mech. 53, 347–376 (2021).

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-20

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

5 22
O. Tutty and T. Pedley, “Oscillatory flow in a stepped channel,” J. Fluid Mech. S. Laizet and E. Lamballais, “High-order compact schemes for incompressible
247, 179–204 (1993). flows: A simple and efficient method with quasi-spectral accuracy,” J. Comput.
6
M. Rosenfeld, “A numerical study of pulsating flow behind a constriction,” Phys. 228, 5989–6015 (2009).
23
J. Fluid Mech. 301, 203–223 (1995). S. Laizet and N. Li, “Incompact3d: A powerful tool to tackle turbulence prob-
7
G. Sheard, T. Leweke, M. C. Thompson, and K. Hourigan, “Flow around an lems with up to o(105) computational cores,” Int. J. Numer. Methods Fluids 67,
impulsively arrested circular cylinder,” Phys. Fluids 19, 083601 (2007). 1735–1757 (2011).
8 24
P. Costamagna, G. Vittori, and P. Blondeaux, “Coherent structures in oscilla- S. Laizet, S. Lardeau, and E. Lamballais, “Direct numerical simulation of a mix-
tory boundary layers,” J. Fluid Mech. 474, 1–33 (2003). ing layer downstream a thick splitter plate,” Phys. Fluids 22, 015104 (2010).
9 25
P. Hall and K. Parker, “The stability of the decaying flow in a suddenly blocked S. K. Lele, “Compact finite difference schemes with spectral-like resolution,”
channel,” J. Fluid Mech. 75, 305–314 (1976). J. Comput. Phys. 103, 16–42 (1992).
10 26
D. Das and J. H. Arakeri, “Transition of unsteady velocity profiles with reverse D. Das and J. Arakeri, “Unsteady laminar duct flow with a given volume flow
flow,” J. Fluid Mech. 374, 251–283 (1998). rate variation,” J. Appl. Mech. 67, 274–281 (2000).
11 27
S. Das, U. Srinivasan, and J. Arakeri, “Instabilities in unsteady boundary layers M. Palmer and P. Freymuth, “Analysis of vortex development from visualiza-
with reverse flow,” Eur. J. Mech.-B/Fluids 55, 49–62 (2016). tion of accelerating flow around an airfoil, starting from rest,” AIAA Paper No.
12
M. S. Ghidaoui and A. A. Kolyshkin, “A quasi-steady approach to the instability 1984-1568, 1984, p. 1568.
28
of time-dependent flows in pipes,” J. Fluid Mech. 465, 301–330 (2002). F. Finaish, M. Palmer, and P. Freymuth, “A parametric analysis of vortex patterns
13
A. Nayak and D. Das, “Transient growth of optimal perturbation in a decaying visualized over airfoils in accelerating flow,” Exp. Fluids 5, 284–288 (1987).
29
channel flow,” Phys. Fluids 29, 064104 (2017). J. H. Gerrard, “The wakes of cylindrical bluff bodies at low Reynolds number,”

Downloaded from http://pubs.aip.org/aip/pof/article-pdf/doi/10.1063/5.0083743/16633478/034102_1_online.pdf

14
M. S. Bloor, “The transition to turbulence in the wake of a circular cylinder,” Philos. Trans. R. Soc. London, Ser. A 288, 351–382 (1978).
30
J. Fluid Mech. 19, 290–304 (1964). P. J. Schmid, D. S. Henningson, and D. Jankowski, “Stability and transition in
15
T. Wei and C. Smith, “Secondary vortices in the wake of circular cylinders,” shear flows. Applied mathematical sciences, vol. 142,” Appl. Mech. Rev. 55,
J. Fluid Mech. 169, 513–533 (1986). B57–B59 (2002).
16 31
C. W. Rowley, I. Mezić, S. Bagheri, P. Schlatter, and D. S. Henningson, Y. Zhiyin, “Secondary instability of separated shear layers,” Chin. J. Aeronaut.
“Spectral analysis of nonlinear flows,” J. Fluid Mech. 641, 115–127 (2009). 32, 37–44 (2019).
17 32
P. J. Schmid, “Dynamic mode decomposition of numerical and experimental J. Jeong and F. Hussain, “On the identification of a vortex,” J. Fluid Mech. 285,
data,” J. Fluid Mech. 656, 5–28 (2010). 69–94 (1995).
18 33
A. Seena and H. J. Sung, “Dynamic mode decomposition of turbulent cavity C. H. K. Williamson, “Oblique and parallel modes of vortex shedding in the
flows for self-sustained oscillations,” Int. J. Heat Fluid Flow 32, 1098–1110 wake of a circular cylinder at low Reynolds numbers,” J. Fluid Mech. 206,
(2011). 579–627 (1989).
19 34
C. W. Rowley and S. T. Dawson, “Model reduction for flow analysis and con- H. Mansy, P.-M. Yang, and D. R. Williams, “Quantitative measurements of
trol,” Annu. Rev. Fluid Mech. 49, 387–417 (2017). three-dimensional structures in the wake of a circular cylinder,” J. Fluid Mech.
20
H. K. Jang, C. E. Ozdemir, J. H. Liang, and M. Tyagi, “Oscillatory flow around 270, 277–296 (1994).
35
a vertical wall-mounted cylinder: Dynamic mode decomposition,” Phys. Fluids S. Behara and S. Mittal, “Flow past a circular cylinder at low Reynolds number:
33, 025113 (2021). Oblique vortex shedding,” Phys. Fluids 22, 054102 (2010).
21 36
Y. Liu, J. Long, Q. Wu, B. Huang, and G. Wang, “Data-driven modal decompo- J. W. Brown and R. V. Churchill, Complex Variables and Applications
sition of transient cavitating flow,” Phys. Fluids 33, 113316 (2021). (McGraw-Hill, 2009).

Phys. Fluids 34, 034102 (2022); doi: 10.1063/5.0083743 34, 034102-21

Published under an exclusive license by AIP Publishing