Sei sulla pagina 1di 16

PHYSICAL REVIEW E 98, 053106 (2018)

Three-dimensional transition of zero-pressure-gradient boundary layer by impulsively and


nonimpulsively started harmonic wall excitation

Pushpender Sharma,1,* Tapan K. Sengupta,1,† and Swagata Bhaumik2,‡


1
Department of Aerospace Engineering, IIT Kanpur, Kanpur-208016, India
2
Department of Mechanical Engineering, IIT Jammu, Jammu-181221, India

(Received 29 June 2018; revised manuscript received 23 October 2018; published 26 November 2018)

A three-dimensional (3D) transition route for zero-pressure-gradient boundary layer over a flat plate is com-
putationally investigated, for impulsive and nonimpulsive startup of harmonic wall excitation. A monochromatic
frequency of excitation is chosen to perturb the boundary layer. The exciter placed near the leading edge
of the flat plate is stapled in the spanwise direction, as in the classical experiment by Klebanoff, Tidstrom,
and Sargent [J. Fluid Mech. 12, 1 (1962)]. The computational domain includes the leading edge and the 3D
Navier-Stokes equation is solved first for the equilibrium flow without any excitation. This is followed by
computing the disturbed flow with the exciter started in either an impulsive or in a nonimpulsive manner. Detailed
description of the transition process is provided for the startup cases, which shows that the transition is essentially
caused by the spatiotemporal wave front for the moderate frequency of excitation. The present investigation
follows the formulation and methodology given by Bhaumik and Sengupta [Phys. Rev. E 89, 043018 (2014)],
which specifically reported only the impulsive startup of harmonic wall Gaussian circular patch excitation for a
moderate frequency that led to the aligned pattern of  vortices.

DOI: 10.1103/PhysRevE.98.053106

I. INTRODUCTION reported successfully for two-dimensional (2D) turbulence in


[6] and 3D routes of transition in [7] for the ZPG boundary
Zero-pressure-gradient (ZPG) flow past a flat plate is the
layer.
canonical problem to study transition to turbulence. This
The main feature of the simulations is to show that the
is done by studying the instability of an equilibrium flow.
spatiotemporal wave front (STWF) is primarily responsible
Early researchers have used Blasius boundary layer as the
for causing transition to turbulence. These [6,7] reproduced
equilibrium flow, whose instability indicates the onset of the
the experimental observations [9] and [10], where the flow
transition process. Although the Blasius boundary layer is
was excited time harmonically, which was started impulsively.
used for instability and receptivity studies [1–4], the results
For moderate to high frequency excitations, two aspects have
are restrictive. The Blasius profile is not valid near the leading
been noted from the simulations: (i) the STWF is created at the
edge of the plate [5], and exclusion of leading edge alters the
exciter location and grows in space and time, leaving behind
instability route qualitatively. This is avoided in [6,7], where
the TS wave, which appears to have been frozen in space and
the leading edge of the plate is included in the computational
time, due to its spatial properties and (ii) once the STWF
domain for solving the Navier-Stokes equation (NSE). This
is created and it convects downstream, it has the feature of
is adopted here also, to study the receptivity to 3D harmonic
spawning subsequent STWF, upstream of its predecessor.
wall excitation cases, where the harmonic excitation is started
STWF has been shown in [12,13] from the solution of
in impulsive and nonimpulsive manners.
the Orr-Sommerfeld equation obtained using the Bromwich
Instability of the ZPG boundary layer has been studied
contour integral method (BCIM) developed in [14], to explore
in [8] for spatial growth to show the presence of Tollmien-
the existence of spatiotemporal disturbance growth. Thus the
Schlichting (TS) waves. Existence of the TS wave has been
origin of STWF is by a linear receptivity mechanism, and
shown experimentally in [9] requiring a quiet wind tunnel
it is theoretically different from convective instability (de-
with a ribbon vibrated at a single frequency inside the bound-
termined from the branch-point singularities in the complex
ary layer. Corresponding 3D disturbances were shown later
frequency and wave-number plane for inviscid disturbance
in [10,11], where a vibrating ribbon was restrained from
equations) [15]. Relevance of BCIM is in circumventing the
vibrating at locations along the spanwise direction periodi-
restrictive assumption used in the signal problem, which
cally, thereby enforcing spanwise periodicity. Experimental
enforces a fluid dynamical system to display response at
excitation to create turbulence deterministically encourages
the frequency of excitation, an idea proposed in [16,17].
one to follow the same route computationally. This has been
However, the authors in [14] reasoned that, for an unstable
system, excited at a constant frequency, the startup is by the
Heaviside function, which excites all frequencies during the
*
pksharma@iitk.ac.in onset. The transfer function of the dynamical system will
† select the dominant natural frequency. This is the basis of
tksen@iitk.ac.in
‡ developing BCIM for OSE, where the Bromwich contours are
swagata.bhaumik@iitjammu.ac.in

2470-0045/2018/98(5)/053106(16) 053106-1 ©2018 American Physical Society


SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

simultaneously chosen in wave number and circular frequency the literature: (a) the K-type route (after Klebanoff et al.
planes. [10]) and (b) the N-type or H -type route [11]. The K-type
There have been efforts to explain the existence of STWF route is followed by a fundamental wave with an amplitude
for the wave propagation problem in electromagnetics in [18]. modulation in spanwise direction leading to the formation of
One also notes that [2] reported that the STWF is due to the “peaks” and “valleys” in the perturbed region and the aligned
fact that, for the simulations, the disturbances are introduced formation of  vortices [22]. Appearance of spikes at late
for t > 0 into a flow that was, prior to that, steady and undis- stages of transition due to ring vortices being snatched away
turbed; the propagating, time-periodic Tollmien-Schlichting from the tip of the  vortices is the most characteristic feature
waves are preceded by a wave front that is essentially a wave of the K-type route of flow transition [11,23–25]. Theoretical
packet. Of course, the nonlinear behavior of the leading wave explanation of a K-type route is given following the resonant
packet would be very different from that of pure Tollmien- triad interaction mechanism [26,27], when one 2D wave and a
Schlichting waves. The more complicated developments in pair of 3D oblique waves of identical fundamental frequency
this leading wave packet are not considered here, but will be interact with each other [11]. The H -type route of transition
the subject of another detailed investigation. However, since is initiated by the interaction of a 2D fundamental wave and a
then such an investigation has not been reported. Chomaz pair of 3D oblique subharmonic waves [11,28], such that the
[19] has advanced the concept of STWF, as a consequence phase angle between these waves is favorable for a parametric
of finite initial perturbations applied over a finite streamwise resonance [26]. This leads to a rapid amplification of the
extent, which separates from the undisturbed state. The au- subharmonic amplitude and to the formation of staggered
thor studied nonlocalized finite amplitude excitation, different arrangement of  vortices. The breakdown of the  vortices
from the STWF shown by the linear and nonlinear receptivity in the H -type route is characterized by a gradual broadband
analysis. filling of the disturbance spectrum without the appearance of
As the STWF convects downstream it experiences sec- spikes as noted for the K-type route [29]. These above obser-
ondary and nonlinear growth leading to 2D turbulence as vations are valid when waves rather than wave packets (as in
shown in [6], and to 3D turbulence as shown in [7], estab- STWF) are considered. Consequently, the dynamic scenario
lishing the centrality of STWF as the precursor of turbulence. of flow transition by STWF is more rich and interesting in
The existence and growth of STWF have been verified from content.
the DNS of 2D NSE in [6,20] and 3D NSE in [7], matching Another mechanism of flow transition is described in
with the essential features of experiments in [9] and [10] [30,31], which is called the “lift-up effect” due to algebraic
caused by time-harmonic wall excitation started impulsively. growth and streak instability. Recently, this mechanism is
All these computations used impulsive startup and that raises summarized in [32]. This describes the instability of the slow-
a question: what will happen if the startup of the harmonic speed meandering velocity streaks following the theoretical
wall excitation is nonimpulsive? For 2D transition, this is works of Ellingsen and Palm [33]. This follows the invis-
addressed in [21] to show that STWF still plays the same role, cid instability analysis in the turbulent zone for streamwise
even if the startup process is smooth. This is the motivation of independent disturbances and also requires the wall-normal
the present research, where we start the harmonic wall excita- velocity in the transitional zone to be time invariant. These
tion with a finite rate for the 3D transition. Can we reproduce two assumptions are shown here not to hold true when
the experimental features of [10], as it was shown in [7] transition is caused by the amplifications of STWF. The
for impulsive startup of the harmonic excitation? The current wall-normal velocity changes significantly due to the down-
investigation is undertaken to check the effects of the startup stream propagation and amplification of STWF invalidating
process of time-periodic monochromatic wall excitation on the assumptions made in the theoretical formulation of these
the appearance, growth of STWF, and eventual transition to mechanisms.
turbulence for a zero-pressure-gradient boundary layer. The We also note that the STWF discussed here is distinct
exciter is in the form of a spanwise modulated wall-normal from those studied in experimental, theoretical, and numerical
excitation source, as it was done in [9,10]. approaches in [4,34–36], where wave packets are created
It has been reported in [21] that the actual process of by pulse excitation and not by continuous time-harmonic
transition follows essentially two routes. (i) For moderate- excitation, as in [9,10]. The DNS approaches in [6,20] to
to high-frequency excitations, STWF causes transition, while simulate experiments of [9–11] have used a simultaneous
the TS wave packet remains distinctly rooted near the exciter blowing-suction (SBS) strip to show the local solution, TS
and, due to the noninteracting nature of these two elements wave packet, and the STWF.
during transition, it is termed as a class-N transition. ii) For In the next section, we briefly state the formulation of
low-frequency excitation, the STWF and TS wave packet the governing NSE used here and the associated auxiliary
continually interact to cause transition and hence it is termed conditions for its solution. Also in this section, we provide
as a class-I transition route. Here, we investigate the moderate the computational details of the performed DNS. In Sec. III,
frequency excitation case for 3D disturbance flow, with the corresponding results for 3D transition routes are discussed,
monochromatic wall excitation started impulsively and non- for different startup of time-harmonic wall excitation. Effects
impulsively, with different acceleration rates. One notes that, of startup on STWF are described in Sec. IV, followed by
in the present reported results, the transition is not caused by description of global flow field in Sec. V. In Secs. VI and
the spatially modulated TS wave packets. VII spectrum and time accurate data of the 3D disturbance
Conventionally, two main routes of the boundary layer field are provided. The manuscript ends with the summary and
transition following TS wave breakdown are identified in conclusion in Sec. VIII.

053106-2
THREE-DIMENSIONAL TRANSITION OF ZERO- … PHYSICAL REVIEW E 98, 053106 (2018)

II. GOVERNING EQUATION, FORMULATION, AND taking the curl of NSE in the primitive variable formulation.
NUMERICAL METHODOLOGY The velocity field is determined by vector Poisson equations.
This formulation and its numerical treatment is given in [7,41]
DNS of fluid flows have been performed and reported in
in detail, and a short description of the governing equa-
literature with different formulations of NSE. For 2D simula-
tion is given in Appendix of this paper. References [6,7,20]
tions, NSE has been used in [6,20] in stream function-vorticity
have successfully created STWF showing its regeneration
formulation, which is unlike the variants of NSE used in
mechanism, leading to fully developed turbulent flows with
[37,38], where the authors used the same formulation, but
predicted dependence of 2D and 3D spectra for turbulence.
added hypo- and hyperviscosity terms, in addition to temporal
The numerical results for skin friction and shape factor also
damping and assorted forcing terms composed of Gaussian
matched excellently with results in [43,44].
and white noises. We also note that in many primitive variable
The advantage of using velocity-vorticity formulation is
formulations either periodicity is imposed [39] or an error-
that the direct evaluation of pressure is avoided as the govern-
prone implicit-explicit (IMEX) time integration method has
ing vorticity transport equations and velocity Poisson equa-
been used [3]. A critique on methods used to solve transi-
tions do not contain pressure. Accuracy in DNS demands
tional and turbulent flows has been given in [40]. For the
that the dependent variables are divergence free, i.e., velocity
computations, we solve the NSE using the vorticity-velocity
and vorticity are solenoidal. Analytically the vorticity field is
formulation given in [41], in a long domain, with deterministic
divergence free, as [∇ · (∇ × V ) ≡ 0]. However numerically
excitation to simulate the results of the experiments in [10].
it isn’t so, specially on collocated grids [41]. Thus a staggered
For computation of 3D transitional flows over a flat plate,
compact difference scheme [45] is used to obtain the spatial
NSE have been solved in the primitive variable formulation
derivatives. In the adopted staggered grid method, the velocity
in [3], and in velocity-vorticity formulation in [2,42]. None
components are evaluated at the face centers and the vorticity
of these references reported finding STWF and, in some of
components are computed at the edge centers. This technique
these, random noise has been injected through the inflow to
has been found to minimize the error in the requirement
trigger transition. The reason that these works using derived
of the divergence-free vorticity condition. Apart from the
or primitive variables could not show the STWF can be
optimized version of the spatial discretization scheme [named
traced to the domain size used in the simulations of [1–3].
as optimized staggered compact scheme (OSCS)], an opti-
In Fig. 1, we indicate domains used by different researchers
mized dispersion relation preserving three-stage Runge-Kutta
and this is shown with respect to the neutral curve obtained
method (ORK 3 ) is used for time integration [46]. The exten-
by linear stability theory. In [1,3], the domain only covers
sive validation of the formulation using the 3D parallelized
a very small portion of the spatially unstable zone right of
code is shown in [41], where the problems of cubic lid driven
the tip of the neutral curve as shown in Fig. 1. In contrast,
cavity and transition on a flat-plate boundary layer have been
successful computation efforts in [7,41] used a domain that is
solved.
significantly larger and is able to capture STWF from onset,
for very low amplitude of excitation.
In the present study, governing NSE is solved using A. Computational domain
the velocity-vorticity formulation in a longer domain with
leading-edge effects included. All the quantities have been The computational domain, as shown in Fig. 2, is
nondimensionalized with the help of free stream speed (U∞ ) 50.05 units long in the streamwise x direction; one unit in the
as the velocity scale and L as the length scale. This length wall-normal, y direction and the domain spans from −0.25
scale is so chosen that one defines the Reynolds number to 0.25 in the spanwise, z direction. The number of points
by ReL = U∞ L/ν = 105 , which appears with the viscous used in the computations are 2501 × 351 × 49, in x, y, and
diffusion term. The corresponding time scale is given by z directions, respectively. To account for the leading-edge
L/U∞ , which is also used to nondimensionalize the vorticity. effects, a small section ahead of the leading edge of the flat
The vorticity governed by transport equations are obtained by

[1];

FIG. 1. ω vs Reδ∗ plot. Various domain lengths used by different


authors are compared in the figure, including the present work. FIG. 2. Schematic of the 3D computational domain.

053106-3
SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

plate is used, with the input plane located at xin = −0.05. The 0.2
exit plane is located at xout = 50.
The grid is generated such that the points are clustered 0.18
Blasius Profile, x = 1.5
near the leading edge of the plate and uniform after x = 5. Blasius Profile, x = 7.5
0.16 Blasius Profile, x = 15
The grid points are also clustered near the wall of the plate to Blasius Profile, x = 40
accurately resolve the boundary layer. Clustering in both the Mean Flow, x = 1.5
Mean Flow, x = 7.5
0.14
streamwise and wall-normal directions is performed using a Mean Flow, x = 15
Mean Flow, x = 40
tangent hyperbolic function, as suggested by [47]. It has been 0.12
shown in [48] that such a grid produces minimum aliasing y
error during computation. Uniformly spaced grid points are 0.1 x = 1.5
used in the spanwise direction. NSE is solved in the trans- 1

formed (ξ , η, ζ ) plane, such that x = x(ξ ), y = y(η), and 0.08


0.8
0.6
z = z(ζ ). Grid transformation in the x direction is given for 0.4
0  ξ  ξ1 (which corresponds to xin  x  xs ) as 0.06 0.2

  0

tanh[βx (1 − ξ )] 0.04
0.99 0.995 1 1.005
x(ξ ) = xin + (xs − xin ) 1 − (1)
tanh(βx )
0.02
While for ξ1  ξ  1 [which corresponds to xs  x(ξ ) 
xout where uniform grid spacing is used],
   0.2 0.4 umean 0.6 0.8 1
βx ξ − ξ1
x(ξ ) = xs + (xs − xin ) , (2)
tanh(βx ) ξ1 FIG. 3. Equilibrium flow profiles obtained by solving the Navier-
Stokes equation compared with the Blasius boundary layer solution.
−xs
where ξ = 1+A 1
1
and A1 = ( xxout
s −xin
)( tanh(β
βx
x)
). Here, xs = 5
is used in the reported simulations. The grid-transformation
function along the wall-normal direction is given as
  C. Computing equilibrium flow from NSE for 3D computations
tanh[βy (1 − η)]
y(η) = ymax 1 − , (3) The solution of NSE for the 3D domain is initialized with
tanh(βy ) the result obtained for a 2D flat plate of the same dimensions
where 0  η  1. Here, βx and βy are stretching parameters in x and y directions, with the leading-edge included. In
used in the streamwise and wall-normal directions, respec- Fig. 3, the 2D velocity profile obtained by solving NSE is
tively. In the study for all computed cases, we have used compared with the Blasius boundary layer profile at various
βx = βy = 2. streamwise locations. The profiles are plotted for y between
zero and 0.1, to focus on the boundary layer, near the wall.
Observable difference between the two profiles is seen near
B. Boundary conditions
the leading edge of the boundary layer, shown in the zoomed
The inlet (ABCD) as shown in Fig. 2 is supplied with inset. Such differences disappear as one moves downstream.
a uniform flow with free stream nondimensional velocity of Despite the small difference of the flow field excluding the
u = 1. At the outflow plane (EFGH), convective Sommerfeld leading edge portion of the plate, the presence of the leading
boundary conditions on velocity and vorticity are applied, as edge inside the computational domain brings in qualitative
explained in [41] with details and validation. This type of changes in the evolution of the disturbance field, as has been
boundary condition is also used in the DNS of flow over a flat shown earlier in [7,50,53]. This 2D solution is mapped onto
plate exposed to adverse pressure gradient in [49,50]. As the the 3D domain by stacking the solution in the (x, y) plane, for
plate has the sharp leading edge PQ, for zero angle of attack, all the z stations. With this initialization, the 3D solver is run
ABPQ represents the stagnation flow plane and symmetry for sometime to reach the 3D equilibrium state, till the time
conditions are imposed on the boundary segment ABPQ, derivatives of vorticity components fall below a prescribed
i.e., ∂u/∂y = ∂w/∂y = ∂ωy /∂y = 0 and v = ωz = ωx = 0. threshold.
In DNS literature, researchers also use a fringe or buffer
region at the outflow to prevent reflection and/or enforce
periodicity in the streamwise direction, as in [39,51,52]. Use
of compact scheme in the present research also helps attenuate III. RECEPTIVITY TO SPANWISE-MODULATED WALL
the variables near the outflow, an inherent property of com- EXCITATION
pact scheme near the outflow, as explained in [48]. Periodic The spanwise-modulated time-harmonic wall excitation is
boundary conditions are considered along the z direction. imposed to trigger transition for the flow past a flat plate.
At the top of the computational domain (CGHD), boundary The amplitude of excitation that has been used in [7,41] with
conditions used for velocity components are: u = 1, w = 0 impulsive start is further modulated for nonimpulsive start by
and the v component (wall-normal component) of the velocity multiplying the amplitude with an error function as described
is evaluated using Eq. (A12). For the vorticity components, below.
the following conditions are used: ωx = ωz = 0, along with The wall-normal excitation (A) is obtained by multiply-
∂ωy
∂y
= 0, as explained in [7,41]. ing the amplitude in the streamwise direction (Ax ) with the

053106-4
THREE-DIMENSIONAL TRANSITION OF ZERO- … PHYSICAL REVIEW E 98, 053106 (2018)

amplitude in the spanwise direction (Az ), and these are given


as follows:

Ax = Af × 0.5 × [1.0 + cos(2π xex )], (4)


 

Az = sin z , (5)
λz
A = Ax × Az , (6)
where the basic amplitude is given by Af = 0.01, which is
also the value chosen in [7] for impulsive start. The exciter
is centered around xexciter = 1.5 and has a streamwise extent
from x1 = 1.455 up to x2 = 1.545, as marked in Fig. 2. There-
fore, xex in Eq. (4) is defined as xex = (x − xexciter )/(x2 − x1 )
and x1  x  x2 . The signal is constructed in such a way
that it has two periods of a sinusoid in the spanwise direction
within z = ±0.25. This implies that the spanwise wavelength
(λz ) is 0.25. Startup of the time-harmonic wall excitation (with
frequency Ff ) is constructed in alternate form for wall-normal
excitation as follows. FIG. 4. Error function (αerr ) variation, as given in Eq. (9), is
(i) For impulsive start with the Heaviside function, H (t ), plotted for different startup cases.
by
vex = A sin(Ff ReL t ) H (t ). (7)
excitation causes additional time scales and frequencies to be-
(ii) For nonimpulsive smooth start created by the error come effective via convolution. These are shown in the insets
function (αerr ): of Fig. 4, with the Fourier transform of the associated error
vex = A sin(Ff ReL t ) αerr , (8) function for the different startup cases. While the amplitude
for the lowest frequency (nondimensional value of 0.0099)
where Ff = 2
2πf ν/U∞ = ω/ReL . Here, the physical fre- is same (=0.0799) in the FFTs for all the cases, the higher
quency, f (in Hz), is nondimensionalized by the viscous time frequency distributions are different for different cases. It is
2
scale (ν/U∞ ), whereas the alternative nondimensional circu- to be noted that a smooth startup, as compared to an abrupt
lar frequency is obtained as ω = 2πf L/U∞ , using convective jump startup of the impulsive case is preferred due to the
time scale (L/U∞ ). All the simulations have been performed band-limited nature of the excitation as shown in the insets of
for Ff = 1 × 10−4 , which was also used in [7]. In Eqs. (7) Fig. 4. At the same time, the aim of the present investigation
and (8), t is the nondimensional time starting from zero and is to ascertain whether the STWF is an attribute of wideband
H (t ) decides that the impulsive start of vex comes into effect excitation of impulsive start or not. It is for this reason that
at t  0. For the nonimpulsive start given in Eq. (8), vex is two nonimpulsive startup cases are investigated with different
obtained by modulating the time-harmonic excitation by a bandwidths and amplitude distributions of excitation. Simi-
smooth functional variation (αerr ), as shown in Fig. 4 and larly stability property of boundary layer over a plate with
given below. This function (αerr ) is constructed using an error front facing step with error function distribution using DNS
function by has been studied in [54]. Due to the smooth variation of area
   decrease causes the flow to accelerate and stabilize. Thus
t − t0
αerr = 0.5 1 + erf √ , (9) space and time variations of flow field excitations are seen
2 π αE
to be useful tools to study flow instability by DNS.
where t0 is the time about which αerr is centered, which For the choice of computational parameters here, Ff =
is taken here as t0 = 4. Note that αE → 0 corresponds to 1 × 10−4 and ReL = 105 , we get Ff ReL = 10 = 2πf L/U∞ .
the impulsive startup case and nonzero values of αE are for Thus we have two equations for three unknowns (L, U∞ , and
nonimpulsive cases, with higher values implying smoother f ), if we consider the working fluid as air with kinematic
startup. Three different cases are considered here: one with viscosity, ν = 1.5 × 10−5 m2 /s and an auxiliary input is the
impulsive start (αE → 0) and the other two nonimpulsive spanwise wavelength of the exciter given by λz = 0.25. If
startup cases, with αE = 0.005 (near-impulsive case) and ReL is treated as the unit Reynolds number, then the above
αE = 0.1 for a smoother startup. With respect to results in
[6,7], this frequency is viewed as moderate to high frequency.
The cases considered here are listed in Table I. TABLE I. Startup cases considered.
In case 1, excitation comes into effect immediately once
the receptivity computations are started from the equilibrium Case no. Type of start αE t0 Ff
state. In cases 2 and case 3, the wall excitation comes into 1 Impulsive 0 0 1.0 × 10−4
effect gradually, with the error function variation centered 2 Nonimpulsive 0.005 4 1.0 × 10−4
around t = 4, as shown in Fig. 4. A similar study was reported 3 Nonimpulsive 0.1 4 1.0 × 10−4
for the 2D case studies in [21]. Such startup at the onset of wall

053106-5
SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

FIG. 5. 3D isometric views of disturbance streamwise velocity (ud ) in (x, z) plane given by y = 0.00215, for different indicated times.
First, second, and third columns in the figure represent case 1, case 2, and case 3, respectively.

conditions yield U∞ = 1.5 m/s, which along with Ff = 10−4 the spanwise modulated excitation is used, as in [53]. Apart
gives f = 2.25 Hz. As we have considered periodicity in from this, we see a STWF leading the disturbance structure,
the spanwise direction, the dynamical system is excited by for both the frequencies of excitation of Ff = 0.5 × 10−4
the superharmonic of this fundamental frequency. Thus the and 1 × 10−4 in [53]. It has been clearly shown that STWF
response for the lowest frequency is related to the Klebanoff creates the turbulent spots at various isolated spanwise lo-
mode, as explained in [55]. cations. These spots gradually enlarge and elongate along
streamwise and spanwise directions and finally merge into
fully developed turbulent flow. Hence STWF is the precursor
IV. EFFECTS OF STARTUP ON EVOLUTION OF
of transition, as shown in [6,7,53] for 2D and 3D transition
SPATIOTEMPORAL WAVE FRONT
following impulsive excitation. These simulations match with
The role of STWFs in creating turbulence has been ex- the experimental observations in [9,10].
plored before in [7,53] for a 3D transition, when the time- The present research is undertaken to investigate nonim-
harmonic excitation is started impulsively. In the first refer- pulsive startup with respect to impulsive startup for 3D transi-
ence, the boundary layer is excited by a Gaussian circular tion by time harmonic excitation. Here, simulation results are
patch (GCP) exciter, in a spanwise periodic domain: −0.5  obtained for three cases for a frequency of Ff = 1 × 10−4 .
z  0.5. It was seen that GCP excitation creates oblique 3D It has been noted in [6,21] that such moderate frequency
TS wave packets, whereas planar TS waves are created when excitation causes N-type of transition, i.e., the STWF and the

053106-6
THREE-DIMENSIONAL TRANSITION OF ZERO- … PHYSICAL REVIEW E 98, 053106 (2018)

FIG. 6. 3D isometric views of disturbance streamwise velocity (ud ) in (x, z) plane (y = 0.00215) at indicated times displaying secondary
and nonlinear stages of disturbance growth. First, second, and third columns in the figure represent case 1, case 2, and case 3, respectively.

TS wave packet do not interact with each other. Therefore, perceptible small waves, marked as A for case 3, ahead of the
the motivation for the present 3D simulation is also to check exciter. In the following row of Fig. 5, three strips of STWFs
whether STWF is at all created and whether transition to are seen for cases 1 and 2, even though the times are different.
turbulence is different for such excitation with nonimpulsive For case 3, four strips of STWFs can be seen at t = 18, which
startup. Also the present study with spanwise modulated can happen due to persistence of acceleration over a longer
excitation will enable us to compare with experimental results period. In the third row of Fig. 5, case 1 shows the highest
in [10,11]. growth at t = 15, whereas growth is slightly higher for case
To look at the evolution for all the three cases of Table I, 3, as compared to case 2 at a relatively similar time.
the flow is explored, close to the wall, in Figs. 5 and 6, with Figure 6 shows the secondary and nonlinear stages of
3D isometric views of disturbance streamwise velocity (ud ) at growth of STWFs, and eventual transition to turbulence via
indicated times for the plane given by y = 0.00215. Figure 5 the formation and amalgamation of turbulent spots. As seen
shows the near-field solution with views of ud at earlier in frames (d1), (d2), and (d3) of Fig. 6, instabilities in the
times. During this stage, the STWF is noted in the frames form of kinks appear across the whole span over the SWTFs,
of the second row. The frames (a1), (a2), and (a3) plotted as noted at around t = 19, t = 26, and t = 27 for the three
at early times show prominent local solution for all the three cases, respectively. The noted spots in the spanwise modulated
cases near the exciter. Cases 1 and 2 appear to have similar excitation case are different from those noted in the Gaussian
evolution, as the startup is almost similar. One can see the circular patch excitation case reported in [7]. In [7], the initial

053106-7
SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

(a) The spanwise locations are chosen such that the growth is
A = Anti-node
A
0.01
A N = Node tracked along nodes and antinodes along the spanwise exciter
Y
Z location; see Fig. 7(a). The primary growth for impulsive
N N
0
N N N startup occurs around t = 10, while for the nonimpulsive
-0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25
cases, this occurs at around t = 15. The saturation of udm ,
A A after nonlinear growth, can be seen to happen at around t = 21
-0.01
for case 1 at all the spanwise stations shown, whereas similar
(b) saturation occurs slightly later: at t = 27 for case 2 and at
t = 30 for case 3. For the nonimpulsive start cases, saturation
0 (i) 0 (ii)
10 10

-2 -2 occurs later as the maximum excitation [given in Eq. (8)]


10 10
udm udm occurs later, due to the fact that the startup is decided by
10
-4
10-4 the error function, which is centered around t = 4—causing a
-6
time lag of various events and phenomena. Despite the startup
-6
10 10 for case 3 being slower than case 2, both the nonimpulsive
10
-8
10
-8 cases grow almost at a similar rate and their udm saturates
t t almost at the same time. Thus, unlike the GCP excitation
0 10 20 30 40 50 0 10 20 30 40 50 case in [7], here for the spanwise modulated excitation cases,
10
0 (iii) 10
0 (iv) variation of properties of STWF is not noted at all the four
spanwise stations for all the three cases, as seen in Fig. 7(b).
-2 -2
10 10
udm udm
-4
10-4 10
V. GLOBAL FLOW FIELD: FROM RECEPTIVITY TO
-6 -6
10 10 TURBULENT SPOT STAGE
10
-8
10
-8
The growth of disturbances in the flow field for space-time
t t evolution of disturbances is traced next. Streamwise variation
0 10 20 30 40 50 0 10 20 30 40 50
of ud is plotted at indicated z locations in Fig. 8 for case 2.
FIG. 7. (a) Nodal and antinodal locations marked for the span- These spanwise locations are chosen such that these corre-
wise variation of imposed disturbance (vex ). (b) Maximum dis- spond to nodal and antinodal locations of the exciter, with
turbance streamwise velocity (udm ) for STWFs compared at y = the stations z = 0 and −0.125 at the nodal locations, whereas
0.00215 for all three cases at four z locations. Here (solid red line: z = −0.0625 and −0.1875 are at the antinodal locations, as
case 1), (: case 2), and (: case 3). Panels (i) and (iii) are the shown in Fig. 7(a).
antinodal locations at z = −0.1875 and z = −0.0625, respectively, Figure 8 shows the evolution of ud from t = 20 to 70,
whereas (ii) and (iv) are the nodal location at z = −0.125 and z = 0, with a time gap of 10, at the nodal and antinodal spanwise
respectively. locations. This figure shows the evolution of ud in the whole
domain, using the same abscissa and ordinate in all frames.
One of the distinctive features noted in these frames of Fig. 8
turbulent spots noted were a zone of high velocity fluctuations is the qualitatively different evolution pattern for the distur-
surrounded by regions of significantly smoother variations in bance along z = 0. This is due to the fact that the symmetry
flow variables. Emmons experimentally noted the appearance of the computed flow about this midplane causes the flow
of similar sporadic spots when the flow is subjected to tur- to behave more in a 2D manner, and that shows events at
bulent disturbances from the free-stream [56,57], which were lower wave numbers for the disturbance along this plane of
also reported in [58–60] by performing DNS. For the span- symmetry. In comparison, other spanwise stations show 3D
wise modulated excited cases, reported here, the initial spots evolution of disturbances, which involves vortex stretching,
cover almost the entire spanwise extent of the computational creating higher wave number fluctuations.
domain and were not noted in the references cited above. The The STWF can be spotted at t = 20, near x = 10 in
appearance of instability of STWFs is a common feature for frame (a) of Fig. 8. The STWF is more apparent due to its
all three cases, which also shows that the mechanism of the larger amplitude at the antinodal locations at t = 20 than at
instability is the same for all the cases. In frames (e1), (e2), nodal locations. The STWF, which is barely perceptible at
and (e3), the instability is very rapid for all the three cases, t = 20, suffers rapid growth due to primary and secondary
while STWFs convect downstream with time. The region instabilities, and by t = 30 one notices disturbances spreading
affected by STWFs keeps elongating with time and, in frames over large wave numbers, except for the spanwise station,
(f1), (f2), and (f3), one sees the appearance of extensive z = 0. A zoomed view of this rapid growth is shown with
turbulence spots in the flow, filling up the whole domain respect to displacement thickness Reynolds number (Reδ∗ ) for
subsequently. See the video in the Supplemental Material [61] case 2 in the Supplemental Material [61]. The leading part of
for complete space-time evolution of disturbances. STWFs convects faster, as compared to the trailing part, and
In Fig. 7(b), the growth of maximum disturbance stream- eventually convects out of the domain as seen in frame (f) at
wise velocity (udm ) is associated with the STWFs, and is t = 70. The trailing part of STWFs eventually settles down to
compared for the three cases at four spanwise z locations. a fixed location between x = 13 and x = 15 at later times of
This is a semilog plot with the ordinate in log scale for udm . t  40. The maximum amplitude of disturbances is also seen

053106-8
THREE-DIMENSIONAL TRANSITION OF ZERO- … PHYSICAL REVIEW E 98, 053106 (2018)

0.4 (a) t = 20 z = -0.1875


z = -0.125
ud 0.2 z = -0.0625
z = 0.0
0
5 10 15 20 x 25 30 35 40 45 50

0.4 (b) t = 30

ud 0.2
0
5 10 15 20 x 25 30 35 40 45 50

0.4 (c) t = 40

ud 0.2
0
5 10 15 20 x 25 30 35 40 45 50

0.4 (d) t = 50

ud 0.2
0
5 10 15 20 x 25 30 35 40 45 50

0.4 (e) t = 60

ud 0.2
0
5 10 15 20 x 25 30 35 40 45 50

0.4 (f) t = 70

ud 0.2
0
5 10 15 20 x 25 30 35 40 45 50

FIG. 8. Disturbance streamwise velocity (ud ) plotted as a function of x for case 2 at y = 0.00125. Nodal locations: z = 0 (dotted line) and
−0.125 (dashed line). Antinodal locations: z = −0.0625 (dash dotted line) and −0.1875 (solid line).

to be saturated and bound within a maximum value of 0.5, length. Thus there are twice as many waves as there are for the
which is about half the free stream speed. exciter. Also, at x = 2.34 and x = 3.65 (approximately), one
notices the signal to have maximum or minimum amplitude
for ud .
Near-field solution evolution and receptivity One also notices another feature of the local solution,
One very interesting feature for all the three cases of impul- that is associated with very large wavelength for streamwise
sive and nonimpulsive starts is that downstream of the exciter variation of ud , as seen in Fig. 9. This can be understood
between x = 1.5 and 3 eventual superposition of waves in by looking at the spectrum of startup of the time-harmonic
time happens in such a way that crests of the response are excitations shown in Fig. 4. In the inset of this figure, it is
aligned with the nodes of the exciter. Similarly, the troughs noted that for the lowest frequency (2.25 Hz, that can be
of the response are always along the antinodes (maxima or associated with the Klebanoff mode as reported in [62]), all
minima locations) of the exciter, as shown in Fig. 9 for case 2. the cases have identical amplitude and such low frequency
The other two cases (case 1 and case 3) also display identical excitation creates the Klebanoff mode, as explained in [55].
features and hence are not shown. In this mode of motion (also known as the breathing mode),
Furthermore, it can also be noted from Fig. 9 that the a large streamwise length of the boundary layer is seen to
spanwise wavelength of the variation of ud occurring between execute heaving motion, as seen in Fig. 9. This is seen to be
x = 1.5 and x = 6 is half of the exciter’s spanwise wave- identical for all the three cases studied here.

053106-9
SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

and spanwise wave numbers. Multiple horizontal cellular


structures are noted in frames (d1) and (d3) of Fig. 10, for
the fundamental and superharmonics of β. Additionally, a
figure showing the comparison of the 2D spectrum of ud at
y = 0.00215 for nonimpulsive startup case 2 and case 3 is
also shown in the Supplemental Material [61].

VII. 3D ANALYSIS OF THE TIME-ACCURATE DATA


We have already noted that during 3D transition distur-
bances grow more rapidly, as compared to 2D transition—
although in both the cases transition is caused by STWF. This
is due to the presence of vortex stretching present only for 3D
flows. It is noted previously that the disturbances start to grow
FIG. 9. Zoomed view of ud at y = 0.00215 for nonimpulsive rapidly after t = 26 for case 2 and case 3 as compared to case
start case 2, showing local solution, near-field solution, and TS wave 1 for which the onset of instability is after t = 20. The growth
regions at t = 10. of disturbance is studied using ud , in (x, z) plane located close
to the wall. We note that the cellular structures are formed in
the spectral plane in Fig. 10, which defines the flow behavior
VI. SPECTRUM OF SPATIAL DATA
near the wall. To explain flow features during 3D transition,
here we trace the disturbances with the help of stream traces
The 2D spectrum for ud in the (x, z) plane for y = 0.00215 and λ2 criterion of [63] for extracting the coherent structures
is plotted in Fig. 10 to note the dominant streamwise and in the flow field. These aspects are discussed next.
spanwise wave numbers (α, β) near the wall, with results
plotted in the (α, β) plane. The results for case 1 are shown in
Fig. 10. The first column of the figure shows the FFT of the A. Investigating 3D flow field using stream-trace patterns
full domain data for the indicated times. The second column The stream traces are followed, as shown in Fig. 11, for
shows the spectrum of the local or near-field solution (i.e., case 2 at the indicated time in the domain. Here case 2 is
for 0  x  6). The third column shows the spectrum of the considered, as it shares the features of case 1 and case 3.
STWF alone, i.e., for 6  x  20. This way of looking at Rake of particles are injected, along the spanwise direction,
part of the domain and performing FFT helps in identifying near the leading edge of the plate at y = 0.00363 in frame
contributions coming from each part of the overall solution, (a) and in the free stream away from the wall in frame (b).
while the present calculations are performed in a nonperiodic The stream traces are colored with the contour values of v
domain. component of velocity. We notice rapid vertical eruptions of
As seen in Fig. 10 for t = 10, the spectrum obtained from the stream traces in the STWF region between t = 25 and t =
the full domain and the local solution is almost similar as 30 (not shown here), and later the stream traces start to curl
seen in frames (a1) and (a2), respectively, while the STWF up, manifesting the presence of streamwise and wall-normal
has very small contribution noted at β = 25 in frame (a3). components of vorticity. Frame (a) of Fig. 11 at t = 65 shows
The wave number β  25 corresponds to the input excitation, the trajectory of the coherent pair of vortices which starts off
i.e., β = 2π/λz = 2π/0.25 = 25.13, where λz is the span- from the wall and travels downstream with increasing height.
wise wavelength of the exciter defined before. Additionally, This is illustrated with clarity in the figure, by showing only
the streamwise variation of ud contributes to multiple wave the half of the spanwise domain in the range −0.25  z  0.
numbers seen at t = 10, for the imposed β = 25. The zoomed In this frame, one can notice that the tracers follow the plate
view of ud variation has been shown in the vicinity of the parallelly up to x  15 and then the stream traces converge,
exciter in Fig. 9. As seen from this figure, the spanwise wave- before lifting off in the wall-normal direction, while con-
length of the signal immediately downstream of the exciter vecting downstream. The traces describe a pair of ascending
is half the exciter’s spanwise wavelength. Therefore, a peak helical structures, implying strong presence of streamwise and
is also noted for higher harmonics of β = 25. The near-field wall-normal components of vorticity. These are reminiscent of
solution is indicated by almost a fixed (α, β) combination peak and valley structures of transitional and turbulent flows
as noted in the middle column. One notices the presence of in [64]. If these two streamwise roller structures correspond to
the Klebanoff mode to remain near the origin (β = 0) of the the subsequent peaks, then the intervening space corresponds
spectral plane. to a valley. This is the state of affairs for tracers released from
For case 1 shown in Fig. 10 at early times, the contribution the near-wall region.
from STWF is prominent at β = 25, with its superharmon- A complimentary picture is obtained for tracer particles
ics also visible at t = 15. Thereafter the global spectrum is released from the free stream, as depicted in frame (b) of
dominated by the contribution from STWF, while the local Fig. 11. These tracer particles move parallel to the wall, up to
solution remains invariant. Higher spanwise wave numbers x  15, and thereafter the particles rapidly descend towards
are rapidly excited after t  15. The STWF suffers rapid the wall. As described above, such motions are due to filling
growth at around t = 21 due to nonlinearity, as noted in up of the valleys by the energetic fluids from the free stream.
Fig. 6, enriching the spectrum by filling up of the streamwise These structures also describe a streamwise descending

053106-10
THREE-DIMENSIONAL TRANSITION OF ZERO- … PHYSICAL REVIEW E 98, 053106 (2018)

t = 10 10 t = 10 10 t = 10 10
100 100 100
9 9 9
8 8 8
80 7
80 7
80 7
β ~ 25
(Local Solution) 6 6 6
60 5 60 5 60 5
β

β
4 4 4
40 3 40 3 40 3
2 2 2
20 1 20 1 20 1
0 0 0
0 0 0
(a1) 0 20 α 40 60 80 (a2) 0 20 α 40 60 80 (a3) 0 20 α 40 60 80
t = 15 t = 15 t = 15
100 100 100
80 80 80
60 60 60
β

β
40 40 40
20 20 20
0 0 0
(b1) 0 20 α 40 60 80 (b2) 0 20 α 40 60 80 (b3) 0 20 α 40 60 80
t = 19 t = 19 t = 19
100 100 100
80 80 80
60 60 60
β

β
40 40 40
20 20 20
0 0 0
(c1) 0 20 α 40 60 80 (c2) 0 20 α 40 60 80 (c3) 0 20 α 40 60 80
Max = 859.65
100
t = 24 Min = 3.18999 × 10−6 100 t = 24
100
t = 24

80 80 80
60 60 60
β

40 40 40
20 20 20
0 0 0
(d1) 0 20 α 40 60 80 (d2) 0 20 α 40 60 80 (d3) 0 20 α 40 60 80

FIG. 10. Spectrum of disturbance streamwise velocity (ud ) at y = 0.00215 for impulsive start case 1. First column: spectrum of full-domain
data. Second column: spectrum of local and near-field solution alone (for x < 6). Third column: spectrum of wave-front part alone (for
6 < x < 20).

helical structure. Thus the associated vortical structures have equation). For such cases, the linearized disturbance equation
positive streamwise and negative wall-normal components of is given as
vorticity. The spherical dots in Fig. 11 are extracted using a
Du
Dωx

postprocessing tool’s feature for tracing vortex cores using = 0, = 0. (10)


the velocity gradient eigenmode method. As time progresses, Dt Dt
the liftoff or the ejection is observed near the wall, while the u
and ωx
indicate disturbance u velocity and streamwise
flow is seen to sweep the void created in the intervening space component of vorticity. The first equation of Eq. (10) gives
between lifted-off parts of the flow.
∂u
∂U
Flow transition by “lift-up mechanism” is summarized in + v
= 0, (11)
[32], to describe algebraic instability of low-speed veloc- ∂t ∂y
ity streak [30,31,65]. This follows the theoretical works of where v
is the disturbance v velocity and U denotes the
Ellingsen and Palm [33], who showed that a finite disturbance streamwise component of the mean flow. Also, the linearized
independent of a streamwise coordinate may lead to instabil- streamwise vorticity equation is given as
ity of linear flow, even if the basic velocity is independent  
of any inflection point. The idea stems from the tempo- ∂ ∂v
∂w

ral instability of inviscid stability equations (i.e., Rayleigh’s − = 0. (12)


∂t ∂z ∂y

053106-11
SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

FIG. 11. Stream-trace pattern and vortex core evolution at t = 65 for nonimpulsive start case 2 showing ejection and entrainment.

Following Eq. (12), Ellingsen and Palm [33] noted that v


is ringlike structures are more prominent for an impulsive start
time invariant and u
grows algebraically as u
(t ) = u
(0) − (case 1), as compared to nonimpulsive cases.
v
∂U
∂y
t. Equation (12) is the basis of the algebraic inviscid
instability, which states that the lift-up mechanism requires VIII. SUMMARY AND CONCLUSION
that v
is independent of time. This has been described as one
of the mechanisms for the instability of streamwise streaks, 3D routes of transition are investigated for harmonic wall
which leads to secondary and higher-order instabilities excitation for a zero-pressure-gradient boundary layer, with
[66–68]. Present simulations show that these assumptions are excitation started impulsively and in a smooth nonimpulsive
not appropriate. It can be shown that the wall-normal compo- manner. The present investigation follows the formulation
nent of velocity (v) in the region of STWF growth changes and methodology given in [7,53], which investigated only
with time, which also makes the disturbance v velocity (v
) the impulsive startup cases for two different frequencies to
change with time. delineate between K and H routes of transition. The sim-
ulations performed for deterministic excitation successfully
reproduced the corresponding experimental investigations re-
B. Extracting coherent structures using λ2 criterion ported in [9,10] with a vibrating ribbon excited near the
Visualization of structures in 3D transitional and turbulent flat plate. For moderate to high frequency of excitations, it
flows is challenging compared to that for similar 2D flows, due has been found that the TS wave packet stays rooted near
to the presence of vortex stretching in 3D flows. The vortical the exciter, while the STWF evolves into fully developed
structures dominate in these flows and are of great importance turbulence via the formation of turbulent spots. The turbulent
to educe the coherent structures correctly in the 3D flow. state is marked by the appearance of coherent structures
For capturing the vortical structures, the λ2 isosurfaces have known as the  vortices. For moderate frequency excitation,
been extracted, as proposed by [63]. The λ2 is the second these vortices are aligned, while for the lower frequency
eigenvalue of the symmetric matrix Sik Skj + ik kj , which of excitations, it has been noted that the  vortices are
represents the rate of strain tensor, with Sij and ij as the staggered.
symmetric and antisymmetric part of the velocity gradient In a recent 2D transition study, Ref. [21] investigated cases
tensor, respectively. where the time-harmonic excitation, as employed in [9,10],
Figure 12 shows the λ2 isosurfaces with a value of −1.0 has been started in both impulsive and nonimpulsive manners
at the indicated times for all the three cases. The isosurfaces to elucidate the differences for the ideal impulsive startup and
are colored with the contours of u component of velocity. The its variations with finite smooth acceleration. Once again the
ringlike  structures can be seen clearly near x = 23 and 25. transition routes have been seen with the TS wave packet not
These structures go through a breakdown near x = 27 into interfering with the STWF for moderate to high frequencies
smaller structures, as seen downstream. The isocontours of λ2 of excitation, while for the lower frequencies, the TS wave
are plotted at t = 45 for case 1, whereas case 2 and case 3 packet is seen to continuously interact with the STWF. In the
are plotted for t = 50 in the figure. It can be seen that the present study, the similarities and differences between impul-
streamwise extent of the iso surfaces is seen to be similar sive and nonimpulsive startups have been investigated for a
for all the cases. The structures between x = 15 and x = 22 3D route of transition for a moderate frequency corresponding
look similar except for case 3, where these are slightly lifted to a K-type transition. The exciter is placed near the leading
up. The presence of streamwise vortical structures, alluded edge of the flat plate and it is punctuated in the spanwise
to in Fig. 11 for case 2, are clearly visible in this figure for direction, as has been reported in [10,11]. The computational
all the three cases. It is interesting to note that the longer domain includes the leading edge and the 3D Navier-Stokes
sustained acceleration during the startup stage for case 3 equation is solved first for the equilibrium flow, without any
causes these streamwise structures to lift up more. Also the excitation, using the velocity-vorticity formulation used in

053106-12
THREE-DIMENSIONAL TRANSITION OF ZERO- … PHYSICAL REVIEW E 98, 053106 (2018)

FIG. 12. Comparison of λ2 = −1 isosurfaces, colored by u velocity, for all the three cases. (a) Case 1 at t = 45, (b) case 2 at t = 50, and
(c) case 3 at t = 50.

[7,41]. This is followed by computing the disturbed flow with and coherent structures by the λ2 criterion proposed in [63].
the exciter started in an impulsive and nonimpulsive manner. The composite view one obtains shows the existence of peak-
Detailed description is provided for all the startup cases to valley splitting of the spanwise variation of vortical structures.
show that the transition is essentially driven by the STWF for We also explain the ejection and sweep events during the
this case of moderate frequency of excitation. turbulent stage of the flow, with the help of stream-trace plots.
In the present study, we have obtained results to show Thus the present study clearly shows the flow field from
that, irrespective of the type of startup, all the three cases laminar to turbulent state, with detailed information of the
considered show the transition to be dominated by STWF, transition process. In the near future, we would like to extend
with sudden rapid growth of STWF after some time following the present studies for other frequencies to further characterize
the startup. This rapid growth stage takes the STWF to affect the various routes of transition.
a progressively larger streamwise stretch of the flow to the *
turbulent state. The near-field flow is described, explaining
how different scales are created in the streamwise and span-
APPENDIX: GOVERNING EQUATIONS AND
wise direction. The spectrum clearly shows cellular structures,
FORMULATIONS
including the Klebanoff mode, due to the lowest frequency
contents arising due to the startup process, as reported experi- Here, a brief description of the governing equations and
mentally by [62] and explained in [55]. A combination of the formulation is provided for the benefit of the readers. The
attributes of spanwise periodicity and exciter wavelength, one governing vorticity transport equations in nondimensional
notices the above mentioned cellular structure in streamwise form, as obtained from incompressible NSE, are
and spanwise wave number plane.
One of the main features of the present work is also to 
∂ 1 2
 × V ) =
+ ∇ × ( ∇ , (A1)
show the 3D flow features with the help of stream traces ∂t ReL

053106-13
SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

where vorticity vector and velocity vector are given by   =  


 1 1 ∂ωy 1 ∂ωx
(ωx , ωy , ωz ) and V = (u, v, w), respectively, in the trans- fζ = (vωx − uωy ) + − . (A8)
formed plane. Although by definition the vorticity vector is ReL h1 ∂ξ h2 ∂η
 = 0, to preserve the solenoidality of the
solenoidal, i.e., ∇ ·  The velocity Poisson equations ∇ 2 V = −∇ ×   are
numerically computed vorticity field, here we use the rota- solved to obtain the corresponding velocity fields, which are
tional variant of the (V , )
 formulation of NSE as discussed
given in transformed plane as
in [69,70]. The vorticity transport equation in the rotational
 
formulation is as given in [7], ∂ωy ∂ωz
∇ξ ηζ u = h1 h2
2
− h3 h1 , (A9)
∂ ∂ζ ∂η
+ ∇ × H = 0, (A2)  
∂t ∂ωz ∂ωx
∇ξ2ηζ v = h2 h3 − h1 h2 , (A10)
∂ξ ∂ζ
where H = (  × V + 1 ∇ × ).
ReL
 The simulations are per-  
formed in the transformed (ξ, η, ζ ) plane, where x = x(ξ ), ∂ωx ∂ωy
∇ξ2ηζ w = h3 h1 − h2 h3 , (A11)
y = y(η), and z = z(ζ ) have finer resolution near the wall and ∂η ∂ξ
the leading edge of the plate. The individual equations for the with the diffusion operator ∇ξ2ηζ given as
vorticity components in the transformed plane are given as      
  ∂ h2 h3 ∂ ∂ h3 h1 ∂ ∂ h1 h2 ∂
∂ωx 1 ∂fζ 1 ∂fη ∇ξ2ηζ = + + .
+ − = 0, (A3) ∂ξ h1 ∂ξ ∂η h2 ∂η ∂ζ h3 ∂ζ
∂t h ∂η h3 ∂ζ
 2  The Poisson equations are solved for u and w components of
∂ωy 1 ∂fξ 1 ∂fζ
+ − = 0, (A4) velocity given by Eqs. (A9) and (A11), whereas the v compo-
∂t h ∂ζ h1 ∂ξ
 3  nent (wall-normal component) of the velocity is calculated by
∂ωz 1 ∂fη 1 ∂fξ integrating the continuity equation from the wall as
+ − = 0, (A5)
∂t h1 ∂ξ h2 ∂η  η 
h2 ∂u h2 ∂w
where h1 , h2 , and h3 are the scale factors of the transforma- v(ξ, η, ζ ) = v(ξ, 0, ζ ) − + dη. (A12)
0 h1 ∂ξ h3 ∂ζ
tion given by h1 = ∂x/∂ξ , h2 = ∂y/∂η, and h3 = ∂z/∂ζ . In
Eqs. (A3) to (A5), the terms fξ , fη , and fζ are given by This ensures the satisfaction of the velocity solenoidality
  condition at the discrete nodes of the domain numerically.
1 1 ∂ωz 1 ∂ωy Moreover, it also imposes the boundary condition on the v
fξ = (wωy − vωz ) + − , (A6) component of the velocity at the far-field boundary in Fig. 2.
ReL h2 ∂η h3 ∂ζ
  This methodology also saves the additional time otherwise
1 1 ∂ωx 1 ∂ωz
fη = (uωz − wωx ) + − , (A7) spent on solving the Poisson equation for the v component.
ReL h3 ∂ζ h1 ∂ξ

[1] H. Fasel and U. Konzelmann, Non-parallel stability of a flat- [9] G. B. Schubauer and H. K. Skramstad, Laminar-boundary-layer
plate boundary layer using the complete Navier-Stokes equa- oscillations and transition on a flat plate, J. Aeron. Sci. 14, 69
tions, J. Fluid Mech. 221, 311 (1990). (1947).
[2] H. F. Fasel, U. Rist, and U. Konzelmann, Numerical investi- [10] P. S. Klebanoff, K. D. Tidstrom, and L. M. Sargent, The three-
gation of the three-dimensional development in boundary-layer dimensional nature of boundary-layer instability, J. Fluid Mech.
transition, AIAA J. 28, 29 (1990). 12, 1 (1962).
[3] T. Sayadi, C. W. Hamman, and P. Moin, Direct numerical [11] Y. S. Kachanov, Physical mechanisms of laminar-boundary-
simulation of complete H -type and K-type transitions with layer transition, Annu. Rev. Fluid Mech. 26, 411 (1994).
implications for the dynamics of turbulent boundary layers, [12] T. K. Sengupta, A. K. Rao, and K. Venkatasubbaiah, Spatio-
J. Fluid Mech. 724, 480 (2013). temporal growth of disturbances in a boundary layer and energy
[4] K. S. Yeo, X. Zhao, Z. Y. Wang, and K. C. Ng, DNS of based receptivity analysis, Phys. Fluids 18, 094101 (2006).
wavepacket evolution in a Blasius boundary layer, J. Fluid [13] T. K. Sengupta, A. K. Rao, and K. Venkatasubbaiah, Spa-
Mech. 652, 333 (2010). tiotemporal Growing Wave Fronts in Spatially Stable Boundary
[5] T. K. Sengupta, Instabilities of Flows and Transition to Turbu- Layers, Phys. Rev. Lett. 96, 224504 (2006).
lence (CRC Press, Boca Raton, FL, 2012). [14] T. K. Sengupta, M. Ballav, and S. Nijhawan, Generation
[6] T. K. Sengupta, and S. Bhaumik, Onset of Turbulence from the of Tollmien–Schlichting waves by harmonic excitation, Phys.
Receptivity Stage of Fluid Flows, Phys. Rev. Lett. 107, 154501 Fluids 6, 1213 (1994).
(2011). [15] P. Huerre and P. A. Monkewitz, Absolute and convective insta-
[7] S. Bhaumik and T. K. Sengupta, Precursor of transition to bilities in free shear layers, J. Fluid Mech. 159, 151 (1985).
turbulence: Spatiotemporal wave front, Phys. Rev. E 89, 043018 [16] M. Gaster, On the generation of spatially growing waves in a
(2014). boundary layer, J. Fluid Mech. 22, 433 (1965).
[8] P. G. Drazin, and W. H. Reid, Hydrodynamic Stability (Cam- [17] D. E. Ashpis and E. Reshotko, The vibrating ribbon problem
bridge University Press, Cambridge, UK, 2004). revisited, J. Fluid Mech. 213, 531 (1990).

053106-14
THREE-DIMENSIONAL TRANSITION OF ZERO- … PHYSICAL REVIEW E 98, 053106 (2018)

[18] L. Brillouin, Wave Propagation and Group Velocity (Academic tsunami, and transition to turbulence, Phys. Fluids 29, 124103
Press, New York, 1960). (2017).
[19] J.-M. Chomaz, Global instabilities in spatially developing [37] A. Bracco and J. C. McWilliams, Reynolds-number depen-
flows: non-normality and nonlinearity, Annu. Rev. Fluid Mech. dency in homogeneous, stationary two-dimensional turbulence,
37, 357 (2005). J. Fluid Mech. 646, 517 (2010).
[20] T. K. Sengupta, S. Bhaumik, and Y. G. Bhumkar, Direct [38] L. M. Smith and V. Yakhot, Finite-size effects in forced
numerical simulation of two-dimensional wall-bounded turbu- two-dimensional turbulence, J. Fluid Mech. 274, 115
lent flows from receptivity stage, Phys. Rev. E 85, 026308 (1994).
(2012). [39] M. Skote and D. S. Henningson, Direct numerical simulation of
[21] S. Bhaumik, T. K. Sengupta, and Z. A. Shabab, Receptivity to a separated turbulent boundary layer, J. Fluid Mech. 471, 107
harmonic excitation following nonimpulsive start for boundary- (2002).
layer flows, AIAA J. 55, 3233 (2017). [40] T. K. Sengupta, A critical assessment of simulations for tran-
[22] W. S. Saric and A. S. W. Thomas, Experiments on the subhar- sitional and turbulent flows, in Advances in Computation,
monic route to turbulence in boundary layers, in Turbulence and Modeling and Control of Transitional and Turbulent Flows,
Chaotic Phenomena in Fluids, Proceedings of the International edited by T. K. Sengupta, S. K. Lele, K.R. Sreenivasan,
Symposium, Kyoto, Japan, edited by T. Tatsumi (North-Holland, and P. A. Davidson (World Sci. Pub. Co., Singapore, 2016),
Amsterdam, 1984), pp. 117–122. pp. 491–532.
[23] F. R. Hama and J. Nutant, in detailed flow-field observations in [41] S. Bhaumik and T. K. Sengupta, A new velocity–vorticity
the transition process in a thick boundary layer, in, Proceedings formulation for direct numerical simulation of 3D transitional
of Heat Transfer and Fluid Mechanics Institute, edited by A. and turbulent flows, J. Comput. Phys. 284, 230 (2015).
Roshko, B. Sturtevant, and D. R. Bartz (Stanford Univ. Press, [42] W. Würz, D. Sartorius, M. Kloker, V. I. Borodulin, Y. S.
Palo Alto, 1963), pp. 77–93. Kachanov, and B. V. Smorodsky, Detuned resonances of
[24] V. I. Borodulin and Y. S. Kachanov, Experimental study of Tollmien-Schlichting waves in an airfoil boundary layer: Ex-
nonlinear stages of a boundary layer breakdown, in Nonlinear periment, theory, and direct numerical simulation, Phys. Fluids
Instability of Nonparallel Flows, IUTAM, edited by S. P. Lin, 24, 094103 (2012).
W. R. C. Phillips, and D. T. Valentine (Springer, Berlin, 1994), [43] H. Schlichting, Boundary-Layer Theory (McGraw-Hill, New
pp. 69–80. York, 1968).
[25] U. Rist and Y. S. Kachanov, Numerical and experimental in- [44] F. M. White, and I. Corfield, Viscous Fluid Flow (McGraw-Hill
vestigation of the K-regime of boundary-layer transition, in Higher Education, Boston, 2006).
Laminar-Turbulent Transition, IUTAM, edited by R. Kobayashi [45] M. K. Rajpoot, S. Bhaumik, and T. K. Sengupta, Solution of lin-
(Springer, Berlin, 1995), pp. 405–412. earized rotating shallow water equations by compact schemes
[26] A. D. D. Craik, Non-linear resonant instability in boundary with different grid-staggering strategies, J. Comput. Phys. 231,
layers, J. Fluid Mech. 50, 393 (1971). 2300 (2012).
[27] L. Brandt, C. Cossu, J.-M. Chomaz, P. Huerre, and D. S. [46] T. K. Sengupta, M. K. Rajpoot, and Y. G. Bhumkar,
Henningson, On the convectively unstable nature of optimal Space-time discretizing optimal DRP schemes for flow
streaks in boundary layers, J. Fluid Mech. 485, 221 (2003). and wave propagation problems, Comp. Fluids 47, 144
[28] Y. S. Kachanov and V. Y. Levchenko, The resonant interaction (2011).
of disturbances at laminar-turbulent transition in a boundary [47] P. R. Eiseman, Grid generation for fluid mechanics computa-
layer, J. Fluid Mech. 138, 209 (1984). tions, Annu. Rev. Fluid Mech. 17, 487 (1985).
[29] T. C. Corke and R. A. Mangano, Resonant growth of three- [48] T. K. Sengupta, High Accuracy Computing Methods: Fluid
dimensional modes in transitioning Blasius boundary layers, Flows and Wave Phenomena (Cambridge University Press, New
J. Fluid Mech. 209, 93 (1989). York, 2013).
[30] M. T. Landahl, A note on an algebraic instability of inviscid [49] W. Wu and U. Piomelli, Effects of surface roughness on a
parallel shear flows, J. Fluid Mech. 98, 243 (1980). separating turbulent boundary layer, J. Fluid Mech. 841, 552
[31] P. Luchini, Reynolds-number-independent instability of the (2018).
boundary layer over a flat surface: optimal perturbations, [50] T. K. Sengupta, S. De, and S. Sarkar, Vortex-induced instability
J. Fluid Mech. 404, 289 (2000). of an incompressible wall-bounded shear layer, J. Fluid Mech.
[32] L. Brandt, The lift-up effect: the linear mechanism behind 493, 277 (2003).
transition and turbulence in shear flows, Eur. J. Mech.-B/Fluids [51] M. A. Bucci, D. K. Plckert, C. Andriano, J.-C. Loiseau, S.
47, 80 (2014). Cherubini, J.-C. Robinet, and U. Rist, Roughness-induced tran-
[33] T. Ellingsen and E. Palm, Optimal disturbances and bypass sition by quasi-resonance of a varicose global mode, J. Fluid
transition in boundary layers, Phys. Fluids 18, 487 (1975). Mech. 836, 167 (2018).
[34] M. Gaster and I. Grant, An experimental investigation of the [52] G. Bonfigli and M. Kloker, Secondary instability of crossflow
formation and development of a wave packet in a laminar vortices: validation of the stability theory by direct numerical
boundary layer, Proc. R. Soc. London A 347, 253 (1975). simulation, J. Fluid Mech. 583, 229 (2007).
[35] M. A. F. Medeiros and M. Gaster, An experimental investiga- [53] S. Bhaumik, T. K. Sengupta, and V. Mudkavi, Different routes
tion of the formation and development of a wave packet in a of transition by spatio-temporal wavefront, in Advances in Com-
laminar boundary layer, J. Fluid Mech. 399, 301 (1999). putation, Modeling and Control of Transitional and Turbulent
[36] S. Bhaumik and T. K. Sengupta, Impulse response and spatio- Flows, edited by T. K. Sengupta, S. K. Lele, K. R. Sreenivasan,
temporal wave-packets: The common feature of rogue waves,

053106-15
SHARMA, SENGUPTA, AND BHAUMIK PHYSICAL REVIEW E 98, 053106 (2018)

and P. A. Davidson (World Sci. Pub. Co., Singapore, 2016), pp. Applied Mechanics, edited by J. P. Den Hartog and H. Peters
68–83. (Wiley, New York, 1939), pp. 294–310.
[54] H. Xu, J.-E. W. Lombard, and S. J. Sherwin, Influence of [63] J. Jeong and F. Hussain, On the identification of a vortex,
localised smooth steps on the instability of a boundary layer, J. Fluid Mech. 285, 69 (1995).
J. Fluid Mech. 817, 138 (2017). [64] S. K. Robinson, Coherent motions in the turbulent boundary
[55] T. K. Sengupta, M. T. Nair, and V. Rana, Boundary layers ex- layer, J. Fluid Mech. 23, 601 (1991).
cited by low frequency disturbances – Klebanoff mode, J. Fluids [65] M. T. Landahl, Wave breakdown and turbulence, SIAM J. Appl.
Struct. 11, 845 (1997). Math. 28, 735 (1975).
[56] H. W. Emmons, The laminar-turbulent transition in a boundary [66] D. S. Henningson, A. Lundbladh, and A. V. Johansson,
layer-Part I, J. Aerosol. Sci. 18, 490 (1951). A mechanism for bypass transition from localized distur-
[57] P. S. Klebanoff, Characteristics of turbulence in boundary bances in wall-bounded shear flows, J. Fluid Mech. 250, 169
layer with zero pressure gradient, NACA Technical Report No. (1993).
NACA-TR-1247, 1955 (unpublished). [67] S. Berlin, A. Lundbladh, and D. Henningson, Spatial simula-
[58] L. Brandt, P. Schlatter, and D. S. Henningson, Transition in tions of oblique transition in a boundary layer, Phys. Fluids 6,
boundary layers subject to free-stream turbulence, J. Fluid 1949 (1994).
Mech. 517, 167 (2004). [68] P. Andersson, M. Berggren, and D. S. Henningson, Optimal
[59] B. Cantwell, D. Coles, and P. Dimotakis, Structure and entrain- disturbances and bypass transition in boundary layers, Phys.
ment in the plane of symmetry of a turbulent spot, J. Fluid Fluids 11, 134 (1999).
Mech. 87, 641 (1978). [69] S. Bhaumik, Ph.D. thesis, Indian Institute of Technology, Kan-
[60] R. G. Jacobs and P. A. Durbin, Simulations of bypass transition, pur, 2013.
J. Fluid Mech. 428, 185 (2001). [70] S. Bhaumik and T. K. Sengupta, On the divergence-free condi-
[61] See Supplemental Material at http://link.aps.org/supplemental/ tion of velocity and vorticity in velocity-vorticity formulation of
10.1103/PhysRevE.98.053106 for additional video and figures. incompressible Navier-Stokes equation, in Proceedings of the
[62] G. I. Taylor, Some recent developments in the study of tur- 20th AIAA CFD Conference, Honolulu, Hawaii, USA (AIAA,
bulence, in Proceedings of the 5th International Congress for Reston, VA, 2011), paper AIAA-3238.

053106-16

Potrebbero piacerti anche