Vous êtes sur la page 1sur 19

Journal of Fluids and Structures 82 (2018) 473–491

Contents lists available at ScienceDirect

Journal of Fluids and Structures


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

Aerodynamic characteristics of a rounded-corner square


cylinder in shear flow at subcritical and supercritical
Reynolds numbers
Yong Cao *, Tetsuro Tamura
Department of Architecture and Building Engineering, Tokyo Institute of Technology, 4259 Nagatsuta-cho, Midori-ku, Yokohama,
Kanagawa 226-8502, Japan

graphical abstract

highlights

• Planar shear-inflow effects are investigated on a rounded-corner square cylinder.


• Re dependency is discovered from subcritical to supercritical regime.
• Shear effects are smaller and nearly symmetry is preserved at supercritical Re.
• Strong asymmetry on two sides of cylinder is found at subcritical Re.
• The mechanism for Re dependency under shear inflows is explored.

article info a b s t r a c t
Article history: Shear-inflow effects on flows past a rounded-corner square cylinder is investigated when
Received 16 March 2018 Reynolds number (Re) increases from subcritical regimes (Re=2.2×104 ) to supercritical

* Corresponding author.
E-mail address: cao.y.aa@m.titech.ac.jp (Y. Cao).

https://doi.org/10.1016/j.jfluidstructs.2018.07.012
0889-9746/© 2018 Elsevier Ltd. All rights reserved.
474 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Received in revised form 13 July 2018 regimes (Re=1.0×106 ), with a focus on the latter Re in this study. The dimensionless shear
Accepted 27 July 2018 parameter K ranges from 0 to 1.0. Results show that shear inflows have different effects on
the aerodynamic characteristics of the cylinder in two Re regimes, and generally shear ef-
fects are smaller when Re is in the supercritical region. Specifically, mean and r.m.s pressure
Keywords: distributions remain in the same shapes under all shear inflows tested at supercritical Re;
Square cylinder
however they witness strong dependency on K at subcritical Re. Moreover, the directions
Shear inflow
Reynolds number effect
of mean lifts at supercritical Re are opposite to those at subcritical Re. Correspondingly,
Rounded corner the mean flow patterns observed in the uniform flow are overall preserved in shear flows
Supercritical regime at the supercritical Re: the free stream flows along the cross section and separates from
Subcritical regime the trailing corners on both sides of the cylinder, despite of the existence of small-degree
asymmetry in the near-wake. However, the flow patterns are significantly changed at the
subcritical Re by the sufficiently high shear inflows: the shear layer in suspension tends
to reattach on the side wall of the high-velocity side and eventually separates from the
trailing edge while the other on the low-velocity side becomes farther from the side wall.
It results in the near-wake forming along an axis rotated in the clockwise direction from
the horizontal axis, being opposite to the anticlockwise direction at supercritical Re. The
mechanism for Re effects is further explored. The preservation of overall flow pattern
at supercritical Re even under the strong shear effects is suggested mainly due to the
greater favorable pressure gradient and the induced greater velocity acceleration before
flow separation on the low-velocity side.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction

The cross-section of a given bluff body significantly influences the time-averaged flow topology and the time-dependent
behavior of the vortex-shedding process. The aerodynamics of circular and square cylinders, two typical shapes, has been
attracting a great deal of research interest because of its immense scientific and practical significance. On the other hand, the
lower-degree-of-bluffness bodies are also widely applied in the engineering. The Reynolds number (Re) effect of these bodies
is a rather unexplored field. Nevertheless, some previous studies have reported Re effects when the free-stream condition is
modeled by uniform flow on the basis of trapezoidal-shape bridge girders (e.g., Schewe and Larsen, 1998), twin-box girders
(e.g., Laima et al., 2018), elongated rectangular cylinders (e.g., Schewe, 2013), and corner-modified square cylinders (e.g.,
Cao and Tamura, 2017; van Hinsberg et al., 2017, 2018).
Among other low-degree-of-bluffness bodies, corner-modified cylinders are of specific interest in this study. They have
widely been introduced to reduce the mean drag and fluctuating lift force due to the vortex shedding in the wake, including
but not limited to rounded corners, chamfered corners, and recessed corners (Kwok et al., 1988; Kawai, 1998; Tse et al., 2009;
Tamura et al., 1998). The radius effects of rounded corner itself were studied by Lamballais et al. (2010) based on a semi-
infinite 2D body presenting a single curved leading edge, and the curvature was found to deeply influence the separation
bubble dynamics. In terms of a rounded-corner square cylinder, the past studies have reported the strong Re effects (Delany
and Sorensen, 1953; Carassale et al., 2014; Cao and Tamura, 2017; van Hinsberg et al., 2017, 2018). Re is defined as U0 D/ν ,
where U0 is the free-stream velocity, D is the cylinder height in the cross section and ν is the kinematic viscosity. Among
others, Cao and Tamura (2017) clarified that strong Re effects exist in the forces and local pressures as Re increases from o
(104 ) to o (106 ), accompanied by the changeover of flow patterns. Unlike the complete separation from the frontal corners
at subcritical Re, the free stream overall flows along the cross sections of cylinder, separates from the leeward corners and
generates twin vortices behind the cylinder at supercritical Re. At a micro level, the flow experiences laminar separation and
flow reattachment near the frontal corner, resulting in a small-size separation bubble. It is followed by the development of
turbulent boundary layer (TBL) on the side face and turbulent separation near the leeward corner. This type of flow has
a much smaller recirculation region behind the cylinder and a low turbulent kinetic energy. van Hinsberg et al. (2017,
2018) experimentally investigated the Re sensitivity on a rounded-corner square cylinder under the combined effects of
surface roughness and corner radius. For a smooth cylinder, a significant dip was observed in the curves of CD -Re or St-Re
relationships in the critical-Re regime at zero angle of attack.
Strong Re influences on rounded-corner square cylinders have been discovered in the uniform inflow when Re increases
from o (104 ) to o (106 ). This study continues to investigate the Reynolds number effects of a rounded-corner square cylinder,
especially under the planar shear inflows with velocity gradient across the cross-section of a cylinder (Fig. 1) by setting Re
equal to o(104 ) and o(106 ). The relevant cases of planar shear flows in engineering applications might be a pipeline system
that is parallel to the ground (e.g., Bearman and Zdravkovich, 1978), a cylinder in the shear separation region of the upstream
cylinder (e.g., Kiya et al., 1979), or a bridge deck immersed in an atmospheric boundary layer. The measure of shear strength
is a dimensionless shear parameter K. It is defined as K = G(D/U0 ) = (dU /dy)(D/U0 ), where U0 is the mean velocity at the
center of the inflow plane, and G = dU /dy is the planar velocity gradient. G is constant in the linear shear flow. In particular,
the inflow is ideally uniform when K = 0. Cao and Tamura (2018) systematically investigated the planar shear effects on
a rounded-corner square cylinder at the subcritical Re = 2.2 × 104 , after the studies on circular, square and rectangular
cylinders under planar shear effects (e.g., Kiya et al., 1980; Sumner and Akosile, 2003; Kang, 2006; Cao et al., 2010; Cheng et
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 475

Fig. 1. Schematic of planar shear inflow approaching a rounded-corner square cylinder.

al., 2007; Lankadasu and Vengadesan, 2008, 2009; Cao et al., 2014). However, all of the previous studies are restricted to the
subcritical Re or lower, whose detailed review could be accessed in Cao and Tamura (2018).
The present numerical study aims at elucidating the influences of Reynolds numbers on aerodynamic characteristics of a
rounded-corner square cylinder under the shear inflows, in particular when Re increases from o (104 ) to o (106 ). Furthermore,
the mechanism for Reynolds number effects is expected to be clarified.

2. Numerical methods and validation

2.1. Numerical model and method

The set-up of problem under consideration is shown schematically in Fig. 2(a). A square cylinder which is geometrically
modified by rounding corners with a radius of r /D = 0.167 (where r is the projected length of the rounded corners onto
the frontal cylinder wall), following the previous investigations (Cao and Tamura, 2017, 2018). Unless otherwise stated, the
length quantities are scaled by D, and the velocity quantities are scaled by the velocity at the center of inlet U0 . The governing
equations are solved in the generalized curvilinear coordinate system. The filtered Navier–Stokes and continuity equations
for incompressible large-eddy simulation (LES) are written in the dimensionless form:
∂ (Jui ) ∂ ( ∂ ∂ξ m ∂ 1 ∂ξ m ∂ξ n ∂ ui
( ) ( )
σ m
)
+ m JUm ui = − m J p + J + (1)
∂t ∂ξ ∂ξ ∂ xi ∂ξ m Re ∂ xj ∂ xj ∂ξ n i

∂ (JU m )
=0 (2)
∂ξ m
⏐ ⏐
∂ξ m ⏐ ∂x ⏐
where Um (= ∂ x ui ) is the contravariant velocity perpendicular to the faces of the grid cells, and J (= ⏐ ∂ξ mi ⏐) is the Jacobian
i
of the transformation between the physical space and computational space. The subgrid-scale (SGS) stress is expressed as
σim = JUm ui − JUm ui , which needs to be modeled. As examined by Cao and Tamura (2017), the dynamic mixed model (DMM)
proposed by Zang et al. (1993) is more advantageous than the dynamic Smagorinsky model (Germano et al., 1991; Lilly,
1992) for predicting separation bubble at the supercritical Re. For this reason, DMM is adopted throughout this study.
The code is based on the finite difference method, which has been successfully applied to flows past bluff bodies at
subcritical and supercritical Re (Ono and Tamura, 2002; Tamura and Ono, 2003; Cao and Tamura, 2016, 2017, 2018). A
fractional step method originally proposed by (Kim and Moin, 1985) is employed to advance the solutions of velocity and
pressure in time. The time marching of the momentum equation is hybrid, viz., the Crank–Nicolson scheme is applied to
the viscous term and an explicit 3rd order Runge–Kutta method is used for the convective term. The detailed numerical
procedure is in Cao and Tamura (2016). Spatial discretization is treated as the second-order central difference generally.
However, the convective term is approximated by using the 4th order central difference scheme. To avoid numerical
instability, numerical dissipation is added through the convective term; the amount of numerical dissipation is controlled by
a parameter αND . The detailed formulas are available in Cao and Tamura (2016) or Chapter 3.5.3 in Kajishima and Taira (2017).
In this study αND = 0.2 is generally used, which means the numerical dissipation is very small in comparison with αND = 3
of the 3rd order upwind scheme proposed by Kawamura and Kuwahara (1984) and αND = 1 of the UTOPIA (uniformly 3rd
order polynomial interpolation algorithm) scheme. The exception is the larger value αND = 1.0 for the cases in the shear
flows (i.e., K ̸ = 0) when Re = 2.2 × 104 , because the asymmetric vortex shedding from two sides of cylinder brings about
undesirable numerical oscillation.
The grid system used for the computations at Re = 1.0 × 106 is shown in Fig. 2. The same grid topology applies
to Re = 2.2 × 104 . This grid topology in the x-y plane consists of an ellipse around the cylinder (with the semi-minor
476 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Fig. 2. (a) Numerical model and whole grid system at Re = 1.0 × 106 ; (b) close-up around the cylinder; and (c) close-up near the rounded corner.

axis in the horizontal direction), joined by two rectangles in the front and in the wake region. It is a modified version of
Zhang et al. (1995). The current topology attaches a rectangle in the front to permit the easy input of user-defined inlet
boundary condition (e.g., turbulent inflow in Tamura and Ono, 2003). In comparison with the traditional O-type, this grid
has the advantage of fine grid resolution in the wake due to the elliptical grid region around the cylinder and the additional
rectangular grid region in the wake. As examined by the energy spectra in Cao and Tamura (2017), the large-extent inertial
sub-ranges featured by −5/3 slope are predicted accurately in the wake, which indicates the sufficiency of the spatial
resolution.
The size of the computational domain is 30D × 24D on the (ξ , η) or (x, y) plane, resulting in a blockage ratio of 4.2% for
the whole computation domain. This study focuses on flows on cross sections influenced by planar shear inflows, indicating
that the computational size and resolution on the cross-section planes play a more important role. Considering the balance
with the computational resources, the present study used the fine resolution on the cross-section plane and fine resolution in
spanwise direction, but relatively smaller spanwise lengths (i.e., Lz /D = 4 at Re = 2.2 × 104 and Lz /D = 1 at Re = 1.0 × 106 )
compared with the authors’ previous study which focused on turbulence transition and three-dimensional flows around the
same cylinder configuration (Cao and Tamura, 2017). In addition, the aerodynamic forces and pressures are generally similar
to those obtained by longer Lz , and hence the present spanwise lengths are thought to be reasonable for the planar-shear-
effect analysis. Other computational information is summarized in Table 1. The total cell number (Ncell ) is 7.3 million and
96.7 million for the cases in the subcritical- and supercritical-Re regimes respectively. The cell size in the spanwise direction
is 0.05D (subcritical Re) and 0.00625D (supercritical Re). The height of the first cell nearest the cylinder wall is 5.59 × 10−4
at Re = 2.2 × 104 and 4.08 × 10−5 at Re = 1.0 × 106 . When K ranges between 0 and 1.0, the maximum and averaged
values of y+ over the whole cylinder (denoted by y+ max and yavg respectively) are ymax = 1.13 ∼ 1.49 and yavg = 0.32 ∼ 0.34
+ + +

at Re = 2.2 × 10 , and ymax = 2.71 ∼ 3.21 and yavg = 1.07 ∼ 1.24 at Re = 1.0 × 10 . The non-dimensional time step
4 + + 6

∆t ∗ (= ∆tU 0 /D) is 5.0 × 10−4 and 1.0 × 10−4 accordingly. After the statistically stationary flows are achieved, the statistics
are computed for a time span of over 23 shedding periods in order to assure the statistically convergence. The cylinder is a
no-slip wall without any wall functions; the spanwise end boundary conditions are periodic; the upper and lower boundaries
are free-slip. The zero-gradient condition of pressure is applied for the above boundaries. The outlet is set as the convective
condition for all velocity components and pressure. The shear inflows are imposed through the inlet boundary condition,
in addition to the uniform inflow (K = 0.0). The modeling approaches of the shear inlets will be further examined and
discussed in the next section.
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 477

Table 1
Summary of grid information for a rounded-corner square cylinder in shear flows when K =0-1.0. Lz and ∆z are spanwise lengths and span resolutions,
respectively; ∆y is the height of the first cell nearest the cylinder wall; y+
max and yavg are the cylinder-maximum and cylinder-averaged y ; ∆t is the
+ + ∗

non-dimensional time step.


Re Grid (ξ ×η×ζ ) Ncell Lz SGS ∆z ∆y y+
max y+
avg ∆t ∗
4 6 −4
2.2×10 300×300×81 7.3×10 4 DMM 0.05 5.59×10 1.13∼1.49 0.32∼0.34 5.0×10−4
1.0×106 1201×500×161 96.7×106 1 DMM 0.00625 4.08×10−5 2.71∼3.21 1.07∼1.24 1.0×10−4

Fig. 3. Velocity profiles along the vertical direction (y direction) at the subcritical and supercritical Re: (a) K = 0.1; (b) K = 0.4; (c) K = 1.0.

2.2. Classification of shear-inflow modeling

The shear strength imbedded in the inflow is not high in most of previous studies. It is a need to extend to very high
shear strength because it may happen at some situations (e.g., the extreme wind environment). After a preceding paper (Cao
and Tamura, 2018), this study continues to cast new light on the strongly and extremely high shear effects, with a focus on
the Reynolds number effects. Furthermore, a structure is often attacked by the local extreme wind events. For these local
situations, one of modeling approaches is that the linearity at inlet is restricted within a limited vertical region. That is, the
velocity profiles at the inlet are set as U(y) = Ky + 1.0 (when U0 and D are assumed to be 1.0) only in the central vertical
region with a constant velocity gradient; out of this vertical region, the velocity is truncated to be uniform. An example
can be referred to the ‘‘inflow’’ profile at K = 0.4 in Fig. 3(b). In the lower outer region, U(y) is truncated to zero in order
to ensure the absence of reverse flow out of the computational domain, which is in the same manner as Cao et al. (2010).
As a consequence, the vertical extent of linear shear flow at inlet is smaller than the computational height. Indeed it is
somehow different from the ideal linear shear inflow where the linearity spreads upwards and downwards as far as infinity.
Shear-inflow modeling approaches are firstly examined and evaluated when K varies from 0.1 to 1.0.
The mean velocity profiles along the vertical direction are shown in Fig. 3 when the measurement stations are located
at x = −1 and 0, and when K varies from 0.1 to 1.0. When K equals 0.1, the vertical influence region by the bluff body is
approximately from −2.5 to 2.5 (normalized by D) at Re = 2.2 × 104 , beyond which the profiles recover to the specified
linear shear profile. The influence region becomes narrower at the supercritical Re = 1.0 × 106 . It indicates the outer regions
are not affected by the cylinder. At K = 0.4, the profile at x = 0 has been very close to the profile x = −1 within the vertical
region from −2.5 to 2.5, the vertical range of linearity which is determined at inlet. The above observation applies to both
Reynolds numbers, which means that the flow around the body is not influenced significantly by the height limitation of
the linearity. Furthermore, it will be shown by the flow field that the centers of primary vortices in the wake are located
within y = [−1.5, 1.5] at Re = 2.2 × 104 , and within y = [−1, 1] at Re = 1.0 × 106 . These ranges are sufficiently smaller
than the vertical ranges of linear shear inflows at K = 0.1 and 0.4. Thus the effects of limited vertical range of linearity
could be neglected for K = 0.1 and 0.4 qualitatively and quantitatively, and the inflows are regarded as the appropriate
approximation of ideal linear shear inflows which are characterized only by K.
When K = 1.0, the velocity profiles around the cylinder (Fig. 3(c)) show that the inflows cannot be considered as ideal
linear shear inflows in the quantitative sense, and cannot be characterized only by K. Meanwhile, the blockage ratio (BR,
defined as the ratio of the projected area of cylinder upon the cross-section area of linear shear inflow) is introduced together
in order to consider the localization of the shear inflow. The linear shear parameter in the limited vertical region is K = 1.0,
resulting in BR = 50%. The meaning of localization of linear shear inflow was discussed in more details by Cao and Tamura
(2018). In the qualitative sense, the case at the nominal K = 1.0 could be loosely regarded as an approximation of an ideal
linear shear inflow where the equivalent/effective shear parameter is smaller than the nominal shear parameter.
Thus the shear inflows under consideration are rigorously classified into ‘‘linear shear inflows’’ when K is not greater
than 0.4, and ‘‘local and extreme linear shear inflows’’ when K = 1.0 and BR = 50%, which will be separately discussed in
Sections 3 and 4.
478 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Table 2
Comparison of bulk parameters with available experiments in the uniform inflow at both subcritical and supercritical Reynolds numbers.

regime Studies Re r/D CD −C pb CL CL St
4
Carassale et al. (2014) 2.7×10 0.067 1.77 1.28 −0.005 0.75 0.137
Carassale et al. (2014) 2.7×104 0.133 1.45 0.97 −0.05 0.52 0.138
Wang and Gu (2016) 3.5×105 0.150 1.45 1.12 – – 0.142
Subcritical-Re
van Hinsberg et al. (2017) 7.5×104 0.160 1.26 0.97 −0.02 0.53 0.137
Tamura et al. (1998) 6.0×104 0.167 1.11 – – 0.39 0.140
Delany and Sorensen (1953) subcritical 0.167 ∼1.24 – – – –
Present 2.2×104 0.167 1.37 1.04 0.04 0.72 0.144
Delany and Sorensen (1953) supercritical 0.167 ∼0.56 – – – –
Supercritical-Re van Hinsberg et al. (2017) 1.0×106 0.16 0.59 0.69 −0.016 0.11 0.27
Present 1.0×106 0.167 0.58 0.71 0.016 0.41 0.253

2.3. Numerical validation

2.3.1. Subcritical-Re regime


The numerical methods have been examined and validated in the previous publications for the flow past a square
cylinder with sharp edges at Re = 2.2 × 104 (Cao and Tamura, 2016). After checking the grid sensitivity, the obtained
numerical results agree well with the available experimental, DNS and LES studies, including global aerodynamic forces,
pressure distributions and flow fields. Thus, the discretization methods and grid resolutions follow those of Cao and Tamura
(2016). In this study, the treatment of rounding corner is adopted, which is slightly different from the previous study. The
accuracy of predicting the subcritical flow past a rounded-corner square cylinder is estimated again. As introduced before,
the experimental measurements are only available in the uniform inflow for the rounded-corner square cylinder, and the
shear inflow past the rounded-corner square cylinder was not investigated previously. Thus the quantitative validation is
conducted only on the occasion when K = 0. Nevertheless, the qualitative comparisons with the circular and square cylinder
in the subcritical-Re shear flow was performed as much as possible and provided more evidences for the validity of numerical
methods (Cao and Tamura, 2018).
Table 2 shows the comparison of bulk parameters with the previous experiments, including mean drag and lift coefficient
(C D and C L ), mean base suction (−C pb ), r.m.s lift coefficient (CL′ ) and Strouhal number (St = fv s D/U0 , where fv s is the vortex
shedding frequency). The general agreement of bulk parameters with the experiments is actually obtained. In Fig. 4(a)–(b),
the mean and r.m.s pressure distributions (denoted by C p and σp ) are compared with the available experiments (Carassale et
al., 2014; Wang and Gu, 2016) when the flow configurations and conditions are close. The mean streamwise velocity profiles
in the wake are shown in Fig. 4(c), in contrast with Hu et al. (2006). Considering the small dispersion of cylinder geometries
and Re, it is concluded that the present numerical results in space and time are consistent with the experimental results.

2.3.2. Supercritical-Re regime


At the supercritical Re, Cao and Tamura (2017) validated carefully the numerical turbulence model and numerical
discretization approaches by predicting the flow past a circular cylinder at the supercritical Re = 6.0 × 105 , which is
thought of as a challenge in the field of numerical simulations of flow past a bluff body. The process of laminar separation
and flow reattachment was successfully predicted in the vicinity of the cylinder surface. The global aerodynamic forces, St
and pressure distributions are consistent with the classical experimental data of flow past a circular cylinder. In terms of
supercritical flows past a rounded-corner square cylinder, the comparison of bulk parameters is conducted in Table 2. It can
be seen that the mean drag and lift forces and St number are very close to the experiments, while the fluctuating lift is larger
than van Hinsberg et al. (2017). The overestimated CL′ is possibly due to the shorter spanwise length of the cylinder adopted
in this study (Lz /D = 1), as analyzed about the spanwise length effects by Cao and Tamura (2015, 2017).

2.4. Effects of spanwise lengths and resolutions

2.4.1. Subcritical Re
With the increasing demand of higher accuracy and the technical support of computational power, the spanwise lengths
and resolutions have become increasingly important numerical issues. On basis of the fully-separated flow around a bluff
body (i.e., circular cylinder or square cylinder at subcritical Re), Cao and Tamura (2015, 2016) observed that the refinement
of spanwise resolutions does not result in any notable differences on the aerodynamic characteristics, provided that the
spanwise resolutions are restricted less than 0.05. By analogy, the present ∆z = 0.05 is believed to be sufficient for
subcritical-Re simulations. However, the spanwise length was found to be a much more influential factor; a longer spanwise
length induces the large phase variation in the vortex shedding along the span (e.g., vortex division and subsequent oblique
shedding). The three-dimensional wake patterns occur intermittently and infrequently compared with parallel shedding.
The details are accessible in Cao and Tamura (2015). Concerning the planar-shear-inflow effects in this study, a further
elongated spanwise length is not expected to bring about any significant changes in the main flow features.
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 479

Fig. 4. Comparison of pressure and velocity distributions with the previous experiments when the inflow is uniform at subcritical Reynolds numbers: (a)
mean pressure distributions; (b) r.m.s pressure distributions; (c) mean streamwise velocity profiles along the streamwise direction.

Fig. 5. (a) Streamlines of span- and time-averaged velocity; (b) distributions of time-averaged and fluctuating pressures when the spanwise lengths are
Lz = 1 and 4 at Re = 1.0 × 106 .

2.4.2. Supercritical Re
The effects of the spanwise length were examined in detail for the supercritical Re flow past a rounded-corner square
cylinder, as has been reported in the appendix of Cao and Tamura (2017). When Lz is increased from 1.0 to 4.0, the time-
averaged flow fields and aerodynamic characteristics (i.e., C D , −C pb , mean formation length Lf ) remain nearly unchanged
in the quantitative sense, which can be testified by the mean streamlines and the C p distributions in Fig. 5. However, the
shorter Lz = 1 tends to somewhat overestimate CD′ and CL′ , as well as σp on the side and rear cylinder faces (see Fig. 5(b)).
Nevertheless, σp shapes are generally similar when Lz equals 1 and 4. It means the major features can be generally captured
even when Lz = 1, which will be further examined hereafter.
In the instantaneous sense, the main vortical structures are examined. Fig. 6 illustrates the wake vortical structures at two
spanwise lengths Lz = 4 and 1. The 3D black iso-surfaces mean the constant instantaneous pressure. The horizontal plane is
colored by instantaneous streamwise velocity component, attached by 2D streamlines based on the vector of (U, W ). Apart
from the spanwise Karman vortices, the clear longitudinal vortices are observed in the braid region between the primary
Karman vortices, leaving the imprints on the horizontal plane (indicated by the mushroom-type structures). Karman vortices
and longitudinal vortices are qualitatively similar regardless of the spanwise lengths at Lz = 4 and 1. Moreover, the spanwise
wavelength of ∼0.7D is roughly estimated for longitudinal vortices at both spanwise lengths. Small-scale turbulent motions
are shown by Fig. 7(a) in the wake, and Fig. 7(b) in the attached turbulent boundary layer. The small-scale turbulent motions
seem to be very similar when the spanwise lengths are 4 and 1, which is supposed to be mainly determined by spanwise
resolutions instead of spanwise lengths.
The effects of spanwise resolutions was examined previously by Ono and Tamura (2008) when ∆z = 0.04, 0.0133, 0.0067
for the supercritical-Re flow past a circular cylinder and found that only ∆z = 0.0067 could accurately estimate the turbulent
reattachment. Later, Lehmkuhl et al. (2014) adopted ∆z = 0.0078. It is worth noting that Bruno et al. (2012) also observed the
480 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Fig. 6. Karman and longitudinal vortices in the wake at the supercritical Re: (a) Lz = 4; (b) Lz = 1. The 3D black iso-surfaces mean the constant instantaneous
pressure. The horizontal plane is colored by streamwise velocity component, and attached by the streamlines of (U , W ).

Fig. 7. Small-scale turbulent motions: (a) in the wake at the supercritical Re when Lz = 1; (b) in the turbulent boundary layer when Lz = 4 and 1, where
the horizontal planes are located at y = 0.501 and the color means the distributions of u = U − U. (For interpretation of the references to color in this
figure legend, the reader is referred to the web version of this article.)

sensitivity of spanwise resolutions (i.e., ∆z = 0.21 ∼ 0.05, normalized by the thickness of cylinder) on the separation bubble
of a rectangular cylinder at subcritical Re. Obviously, the requirement for predicting the separation bubble at supercritical
Re is much more rigid than at subcritical Re. On the basis of the similar numerical methods and turbulent model as Ono and
Tamura (2008), Cao and Tamura (2017) has reproduced the supercritical flow phenomenon around a circular cylinder when
∆z is smaller than the above references (∆z = 0.00625). Thus the present ∆z = 0.00625 is also believed to be sufficient for
the supercritical flows past a rounded-corner square cylinder. Nevertheless, the small-scale turbulent motions are illustrated
at the supercritical Re in Fig. 7. Cao and Tamura (2017) quantitatively examined the high-frequency reproduction by the
energy spectra and the spanwise correlation of spanwise velocity components in all typical regions around the cylinder
(including the wake, the separated shear layer and the attached TBL) when Lz = 4. Fig. 8 provides a supplementary evidence
when Lz = 1. All the examinations indicated that the small-scale turbulent motions are sufficiently reproduced in the present
spanwise resolution irrespective of spanwise lengths used in this study.
In conclusion, the spanwise resolution in this study (∆z = 0.00625) is examined to be sufficient to reproduce most
of the important vortical structures and small-scale turbulent motions, irrespective of spanwise lengths Lz = 4 or 1. The
spanwise lengths from 4 to 1 have the effects on the intermittent and infrequent occurrence of spanwise waviness and/or
the quantitative strength in the large-scale vortex shedding, and the induced quantitative differences in the fluctuation-
related quantities. But Lz = 1 is supposed not to influence significantly the consideration of the planar shear effects on the
supercritical flows past a rounded-corner square cylinder.
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 481

Fig. 8. Two-point correlation of the spanwise velocity fluctuation at the supercritical Re when Lz = 1: (a) in the wake; (b) in the recirculation region behind
the cylinder; (c) in the attached turbulent boundary layer. Note that the first and second points are separated by 10∆z. The counterparts when Lz = 4 are
accessible in Cao and Tamura (2017).

Fig. 9. Variations in bulk parameter factors (%) with K at Re = 2.2 × 104 and 1.0 × 106 .

3. Linear shear effects at subcritical and supercritical Re

3.1. Aerodynamic forces and pressures

The factors of bulk parameters (denoted by fK ) are ( adopted to) evaluate the shear effects on aerodynamic characteristics
at the shear parameter K. They are defined as fK = CF ,K − CF ,0 /CF ,0 × 100, where CF ,K and CF ,0 are the bulk parameters
when the shear parameters are K and 0 respectively. The variations in fK at Re = 2.2 × 104 and 1.0 × 106 are shown in
Fig. 9. It should be noted that the frontal stagnation point θstag is the original angle, rather than the factor. When the shear
parameter increases from 0 to 0.4, the frontal stagnation points move towards the high-velocity side in both subcritical and
supercritical Re regimes. Furthermore, the movement angles are similar in values. Irrespective of Reynolds numbers, the
time-averaged lifts C L factors possess larger values and are more sensitive to the shear inflows than the time-averaged drags
C D and base suction pressures −C pb . St tends to increase with K for both Re regimes. After comparing the factors at two Re
regimes, one could discover that the factors at Re = 1.0 × 106 are smaller than those at Re = 2.2 × 104 on the whole (except
the similar θstag and St). It indicates that the shear inflows have the weaker influences on the supercritical-Re flows past
the cylinder than on the subcritical-Re flows. In particular, the sensitive C L factors at the supercritical Re are much smaller
than those at the subcritical Re. Even at K = 0.4, the C L factor is small in number at the supercritical Re, which means
that the asymmetry of flow pattern is relatively weak between the high- and low-velocity sides. It is beyond the common
understanding that strong asymmetry occurs on two sides of cylinder in the highly shear inflow as observed in the lower Re
regimes. Moreover, the signs of C L in the subcritical shear flows are opposite to those at the supercritical Re. The former C L is
exerted on the cylinder from the low-velocity side to the high-velocity side, while the latter acts from the high-velocity side
to the low-velocity side. Interestingly, the direction of C L observed in the supercritical shear inflows past a rounded-corner
square cylinder is the same as the circular cylinders at the subcritical Re (Sumner and Akosile, 2003; Cao et al., 2007).
The mean and r.m.s pressure distributions on the cylinder are shown in Fig. 10 at the subcritical and supercritical Re
when K ranges from 0 to 0.4. As was focused on in Cao and Tamura (2018), the strong dependency on the shear parameters
is observed at Re = 2.2 × 104 . Specifically speaking, the mean pressure experiences a dramatic reduction near the frontal
corner on the high-velocity side when K increases to 0.4, and gradually recovers on the side wall of the high-velocity side
in Fig. 10(a). It is because the flow starts to reattach on the upper side wall, causing a separation bubble downstream of the
482 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Fig. 10. Variation of pressure distributions at the subcritical and supercritical Re when K ranges between 0 and 0.4: (a) mean pressures; (b) r.m.s pressures.
FC means the frontal corner and LC is the leeward corner.

frontal corner. A peak of r.m.s pressure on the upper side wall is found instead of the plateau at K = 0.4 in Fig. 10(b), which
is caused by the temporal oscillation of reattachment point. The r.m.s pressures are overall decreased on the bottom wall,
which is associated with the farther distance between the separated shear layer on the low-velocity side and the bottom
wall.
In contrast with the subcritical Re, the strong dependency of pressure distributions on the shear parameters is no longer
observed at the supercritical Re. It means that close attention should be paid when one attempts to apply the results on basis
of subcritical Re or lower into the practice at very high Re. At different levels of shear parameters, the pressure distributions
possess the similar shapes when Re = 1.0 × 106 , including mean and r.m.s pressures. Specifically, the mean pressures
show the local extreme minimum values in Fig. 10(a) and the r.m.s pressures have the local extreme maximum values in
Fig. 10(b) around the corners at various K. The similarity in pressure distributions at the supercritical Re signifies that the
flow types around the cylinder in the shear inflows are possibly similar to that in the uniform flow. Bearing it in mind,
detailed check could show slight tendencies varied with the shear parameters. For instance, the C p peaks corresponding to
the frontal stagnation points are gradually moved from the center of the frontal wall face to the high-velocity side when
K increases from 0 to 0.4. On the corners, side and rear walls, the values of C p tend to decrease on the whole by the shear
inflow in comparison with the uniform inflow. But the amount of C p decrease differentiates in these regions of cylinder
surface. In what follows, how the local pressures determine the direction of total mean lift will be analyzed. The magnitudes
of negative pressures on the leeward corners increase on both sides of the cylinder, and eventually offset each other. The
pressures on the side wall on the low-velocity side seem not to be influenced greatly by the shear inflows. Thus, the side
wall on the high-velocity side (marked by the red rectangle ‘‘p1’’ in Fig. 10(a)) and the frontal corner on the low-velocity side
(marked by ‘‘p2’’ in Fig. 10(a)) are mainly taken into account. The former in the ‘‘p1’’ region tends to cause the positive lifts
(acting from the low-velocity side to the high-velocity side), while the latter in the ‘‘p2’’ region tends to induce the negative
lifts (acting from the high-velocity side to the low-velocity side). The negative lifts (Re = 1.0e6 in Fig. 9(d)) mean that the
suction on the frontal corner of the low-velocity side (‘‘p2’’ region) dominates in determining the direction of lift.

3.2. Flow field

The flow fields will be described as a result of interaction between the cylinder and the shear inflows in order to explain
the features presented in the aerodynamic forces and pressures. For clarity, the explanations will be provided in three flow
regions: around the cylinder (−0.5 ≤ x/D < 0.5); in the near wake or base region (0.5 ≤ x/D < 2.5); in the far wake
(2.5 ≤ x/D).
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 483

Fig. 11. Variation of mean flow topologies in shear inflows around the cylinder at the subcritical and supercritical Re: (a) K = 0, Re = 2.2 × 104 ; (b)
K = 0.4, Re = 2.2 × 104 ; (b) K = 0, Re = 1.0 × 106 ; (d) K = 0.4, Re = 1.0 × 106 .

3.2.1. The region around the cylinder (−0.5 ≤ x/D < 0.5)
Fig. 11 exhibits the mean flow topologies around the cylinder influenced by the shear inflows in the subcritical and
supercritical Re regimes respectively. When the shear inflow (K = 0.4) is imposed, the flow asymmetry between the high-
and low-velocity sides is observed to different degrees at two Re. Specifically speaking, the asymmetry at the subcritical Re is
much greater than at the supercritical Re. According to Fig. 11(b), the free stream on the high-velocity side at Re = 2.2 × 104
firstly separates from the frontal corner, reattaches on the upper wall and separates again from the trailing corner. A
separation bubble forms downstream the frontal corner. But the lower shear layer detaches completely from the frontal
corner and has a farther distance from the bottom wall than that at K = 0. It results in a highly asymmetric near wake
behind the cylinder.
A question arises: how the shear inflows influence the flow patterns at the supercritical Re = 1.0 × 106 , which have
turbulent boundary layers attached on the side walls instead of the shear layers in suspension? From Fig. 11(c)–(d), the
symmetry in the mean flow topologies seem not to be significantly broken by the shear inflows even at the high shear rates.
It is further evidenced by the similarity of the distributions of dimensionless mean wall shear stresses in Fig. 12 when K is
between 0 and 0.4. The overall flows are still along the cross section and separate from the trailing corners on both sides of
cylinder. In the micro view, the separation bubbles near the frontal corners are rarely changed in size by the shear inflows.
Twin vortices form behind the cylinder despite the existence of slight asymmetry between the upper and lower vortices. The
small-degree asymmetry in the near wake will be further discussed in the next subsection. In terms of the region around
the cylinder, the preservation of mean flow type is obtained at the supercritical Re when the shear parameter varies from 0
to 0.4, which interprets the nearly unchanged shapes of pressure distributions in the shear inflows.

3.2.2. Near wake region (0.5≤x/d <2.5)


In Fig. 11, the locations of the saddle point in the near wake are labeled. At the subcritical Re, the saddle point is located at
(x, y) = (1.57, 0.02) in the uniform inflow, while it is moved to (1.91, −0.14) in the shear inflow K = 0.4. It means that the
recirculation zone (upstream of the saddle points) is elongated and shifted to the low-velocity side under the effect of the
shear inflows. At the supercritical Re, the saddle point is at (x, y) = (1.05, 0.01) in the uniform inflow and slightly moves to
(1.07, 0.08) in the shear inflow. The recirculation zone is slightly elongated and shifted to the high-velocity side by the shear
inflows. Thus by comparison, the effects of shear inflows are very weak on the recirculation zones of the supercritical-Re
cases. Another way to reflect the influence of Reynolds number on the averaged near-wake region is the local wake width
(dw ). According to the suggestions proposed by Norberg (1986), dw is estimated based on the positions of local maximum
Reynolds normal stress (uu). Their distributions are shown in Fig. 13. Meanwhile, the rotation of the near-wake centerlines
is schematically attached. Being consistent with the recirculation zone, the centerline of the near wake at Re = 2.2 × 104
is inclined towards the low-velocity side under the effects of the shear inflows, which indicates the vortex shedding along
an axis rotated in a clockwise direction. However, the near wake at Re = 1.0 × 106 shows an opposite tendency; it rotates
in the anticlockwise direction when the shear inflows are imposed. Because of that, the curvature of streamlines near the
trailing corner of the low-velocity side is raised, which is possibly one of the reasons for the increase in the pressure suction
with the increase in K in Fig. 10(a).
484 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Fig. 12. Distributions of dimensionless time-averaged wall shear stress at Re = 1.0 × 106 when K ranges from 0 to 0.4, where θstag is the stagnation angle.

Fig. 13. Distributions of local wake widths when K ranges from 0 to 0.4: (a) Re = 2.2 × 104 ; (b) Re = 1.0 × 106 . Meanwhile, the rotation of the near-wake
centerline is schematically presented inside the cylinder.

3.2.3. The far wake region (x/D ≥ 2.5)


In the uniform flow, the occurrence of periodic vortex shedding was found behind the cylinder at both subcritical and
supercritical Reynolds numbers in Cao and Tamura (2017), together with a number of small-scale eddies. Of more interest
in this study are the large-scale vortices due to their primary roles in determining the aerodynamic pressures and forces.
For studying the dynamics of large-scale vortices and their interaction, the conditional sampling technique is conducted to
instantaneous flow field so that the interference of small-scale turbulent motions can be removed. It is conceptually based on
the dominant role of organized Karman vortices and their nearly periodic nature. Following Reynolds and Hussain (1972),
the motions can be classically decomposed into the mean component, periodic fluctuation and chaotic component. The
phase-averaged component then becomes a sum of mean component and periodic fluctuation. The flows at three constant
phases during half of a shedding cycle are typically displayed, i.e., ϕ = −π/2, π/2 and 0. The reference signal is the time
series of total lift coefficient CL , and the three phases correspond to the troughs and peaks of total lift and the middle phases
between them (as sketched in Fig. 14). The phase-averaged spanwise vorticity fields at the three phases are shown in Fig. 14
at Re = 2.2 × 104 and 1.0 × 106 under different shear parameters K = 0, 0.1 and 0.4. Generally speaking, the phase-
averaged flow structures are largely topologically similar to the instantaneous ones. Prior to any deep discussion about
the vortex dynamics in the wake, a quantitative measure is preferable of the vortex swirling strength and trajectory. The
methods follow Cantwell and Coles (1983), Hu et al. (2006) and Cao and Tamura (2018). The computations are based on the
phase-averaged spanwise vorticity field. The quantitative data at Re = 2.2 × 104 are available in Cao and Tamura (2018)
and not repeated herein. For the cases at Re = 1.0 × 106 , this study shows the vortex strength measured by the circulation
⟨Γ ⟩ because the maximum vorticity and the circulation have the similar tendencies. The trajectory is tracked by means of
the vorticity-weighted locations (⟨x∗ ⟩ and ⟨y∗ ⟩) which is computed according to Eqs. (3) and (4) (where ∆A = ∆x∆y, and
∆x and ∆y are the interval distances in the x and y directions). They are shown in Fig. 15 at the supercritical Re = 1.0 × 106
when K varies from 0 to 0.4, where a series of markers (open circles, asterisks or open triangles) connected by a solid line
belong to one vortex at the two or three phases out of −π/2, 0 and π/2 in order.
∑ ( x ) ( ⟨ωz ⟩D ) ∆A
∗ ij 2
⟨x ⟩ D ij U0
ij D
= ⟨Γ ⟩
(3)
D
U0 D
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 485

Fig. 14. Phase-averaged spanwise vorticity field at Re = 2.2 × 104 and 1.0 × 106 when K ranges from 0.0 to 0.4. The figures have the range of x = [−1, 14]
and y = [−3, 3].

∑ ( y ) ( ⟨ωz ⟩D ) ∆A
∗ ij 2
⟨y ⟩ D ij U0
ij D
= ⟨Γ ⟩
(4)
D
U0 D

When K ranges between 0 and 0.4, regardless of Re regimes, the interaction between the separated shear layers on both
sides help cut the energy supply to the vortex that is forming behind the cylinder, and finally the vortex shedding ensues.
Another common property in the shear inflows at two Re regimes is that the vortex strengths on the high-velocity side are
increased while those on the low-velocity side are weakened, see Fig. 7(b) in Cao and Tamura (2018) for the subcritical Re
and Fig. 15(a) for the supercritical Re. When the very high shear inflow K = 0.4 is imposed, the anticlockwise vortices are
too weak to observe in the very far wake for both Reynolds numbers. However, the effects of shear inflows on the vortex
trajectories are opposite at the subcritical and supercritical Reynolds numbers. Whereas the vortex street at the subcritical
Re is shifted to the low-velocity side (see Fig. 8 in Cao and Tamura, 2018), the vortex street at the supercritical Re is shifted
to the high-velocity side as shown in Fig. 15(b). Particularly in terms of Re = 1.0 × 106 , the shift of vortex street in the far
wake is consistent with the anticlockwise rotation of the centerline in the process of the generation and shedding of Karman
vortex in the near wake as shown qualitatively in Fig. 11(d) and quantitatively in Fig. 13(b).

4. Local and extreme linear shear effects at subcritical and supercritical Re

In this section, the Re effects in local and extreme linear shear inflows past the cylinder is investigated when Re increases
from 2.2 × 104 to 1.0 × 106 . As mentioned before, the case at the nominal K = 1.0 could be qualitatively thought of as a loose
approximation of an ideal linear shear inflow where the equivalent/effective shear parameter is smaller than the nominal
shear parameter, although it is not the present purpose to quantify the equivalent/effective K. Thus, the presentation of the
results at K = 1.0 and BR = 50% is sometimes accompanied with the comparison to those at the lower shear parameters
which have been shown in the preceding section.
486 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Fig. 15. Vortex strength (left) and vortex trajectory (right) in the phase-averaged field at Re = 1.0 × 106 . A series of markers (open circles, asterisks or
open triangles) connected by a solid line belong to one vortex at the two or three phases out of −π/2, 0 and π/2 in order.

Fig. 16. (a) Streamlines of span- and time-averaged velocity; (b) variation of shear effects on phase-averaged vortex shedding at K = 1.0 and BR = 50% at
the subcritical and supercritical Re.

The streamlines of span- and time-averaged velocity are shown in Fig. 16(a) at K = 1.0 for two different Reynolds
numbers, together with the locations of saddle points in the near wake. At Re = 2.2 × 104 , the mean flow topology at
K = 1.0 obviously differentiates from that at K = 0. The features induced by shear inflows are enhanced compared with
K = 0.4. Specifically, the separation bubble between the flow separation and reattachment points on the high-velocity side
become smaller than at K = 0.4; the separated shear layer on the low-velocity side become much further from the bottom
wall, inducing a very long and wide recirculation region behind the cylinder. Quantitatively speaking, the saddle point in the
near wake is located at x = 3.88, which is much more downstream than x = 1.91 at K = 0.4. By contrast, at Re = 1.0 × 106 ,
the mean flow topology at K = 1.0 seems generally similar to that at K = 0, only with quantitative differences. One of the
quantitative differences is that the saddle point at (x, y) = (1.16, 0.13) moves more downstream and further towards the
high-velocity side than the location of (1.07, 0.08) at K = 0.4.
Fig. 16(b) shows the vortex dynamics in the wake quantified by the phase-averaged span-wise vorticity, which is
processed in the same manner as Fig. 14. At the subcritical Re = 2.2 × 104 , the first anticlockwise primary vortex behind
the cylinder is nearly steady in phase. It is weak in strength and very far from the clockwise vortex on the high-velocity side.
Consequently the interaction between the vortices of opposite signs behind the cylinder is actually suppressed. As a result,
the flow detached from the upper wall behaves like a turbulent plane mixing layer, which is hardly influenced by the shear
layer and the vortex on the low-velocity side. The large-scale coherent vortices are gradually rolled up in the mixing layer.
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 487

Fig. 17. (a)–(b) The behaviors of separated shear layers in the shear inflow of K = 1.0, BR = 50% at Re = 1.0 × 106 at two instants corresponding to
the peak and trough of CL , where the dotted and dashed black lines approximately represent the locations of separated shear layers. (c) uu distribution at
K = 1.0 and BR = 50% when Re = 1.0 × 106 , where the location of x = 0.6 is illustrated by the vertical dashed white line. (d) Profiles of fluctuating velocity
(uu) along the vertical direction at x = 0.6.

However at Re = 1.0 × 106 , in the presence of overall flow separation from the trailing corners, the interaction between
two separated shear layers on the high- and low-velocity sides is still clearly observed during the process of the roll-up of
primary vortices. For this reason, even under the extreme and local shear effect, the alternative vortex shedding is preserved
for at least the first two vortices behind the cylinder.
By concluding the above results with K ranging from 0.1 to 1.0, the shear effects are not significant on the overall
symmetry flow patterns at the supercritical Re compared with those at the subcritical Re. This is one of main conclusions of
this study. Additionally, one might be curious about the small-degree asymmetry at the supercritical Re. For this purpose,
closer examination of local flows is conducted, including the flapping of the shear layers separated from the trailing corners,
and TBL attached on the side walls. Their dynamic behaviors are shown in Figs. 17 and 18, taking K = 1.0 as an example. Two
typical instants are selected that correspond to the peak and trough of total lift coefficient (CL ). The locations of separated
shear layers are approximately represented by the dotted and dashed black lines in Fig. 17(a)–(b). The dotted lines are at the
instant of CL peak, and the dashed lines are at the instant of CL trough. The differences between the shear layers on the high-
and low-velocity sides are as follows. Firstly, the averaged location of the shear layer on the low-velocity side inclines in
the anticlockwise direction by a larger angle than that on the high-velocity side. Secondly, the shear-layer oscillation on the
low velocity side, over a phase lag of π , has a smaller extent than that on the high velocity side. The smaller-degree flapping
of the lower shear layer is possibly because of the overwhelming and continuous attraction from the stronger vortex on
the high-velocity side. As a result, the fluctuating velocities (uu) in the shear layer on the low-velocity side tend to become
smaller than those on the high-velocity side when the strong shear inflows are imposed, which is quantified in Fig. 17(c)–(d).
The close-ups of boundary layers attached on two side walls are shown in Fig. 18. Being developing on both side walls, the
TBL on the low-velocity side is thicker than on the high-velocity side. It is mainly due to a lower potential velocity out of the
boundary (despite of small difference) on the low-velocity side, consequently a lower TBL Reynolds number.

5. Discussion of the results and their mechanisms

This study has evidenced the great difference between the shear-inflow effects at the subcritical and supercritical Re.
The schematics of time-averaged flow patterns are shown in Fig. 19. The movement of the frontal stagnation points is not
significantly influenced by the increase in Re regimes. But the regions on two sides of cylinder and behind the cylinder
appear to be strongly dependent on Re. In the subcritical Re regime, the separated shear layers are suspended on both sides
of cylinder in the uniform inflow. When the sufficiently high shear inflow imposed, the shear layers in suspension are easily
488 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Fig. 18. The behaviors of turbulent boundary layers in the shear inflow of K = 1.0, BR = 50% at Re = 1.0 × 106 at two instants corresponding to the peak
and trough of CL : (a) at the t ∗ of CL peak; (b) at the t ∗ of CL trough.

Fig. 19. Schematics of mean flow patterns influenced by shear inflows: (a) subcritical Re; (b) supercritical Re, where the upper is in the uniform inflow and
the lower is in the strong shear inflow.

influenced. The one on the high-velocity side tends to reattach on the side wall and eventually separates from the trailing
edge, while the other on the low-velocity side becomes farther from the side wall. The center line of twin vortices behind the
cylinder is inclined in the clockwise direction. However, in the supercritical Re regime, the free stream flows along the cross
section of the cylinder, and separates from the trailing corners in the uniform flow, in spite of small-size separation bubble
around the frontal corners. The attached flow along the side walls seems not to be easily affected by the shear inflows. It
remains in the attachment status on both side walls. Note that the small-size separation bubbles are still existent. Eventually,
the separated flow on the high-velocity side tends to be more parallel to the free stream direction than on the low-velocity
side. The center line of the twin vortices in the near wake is slightly inclined in the anticlockwise direction.
The shear effects in the subcritical Re regime are investigated and analyzed by a lot of researchers before, reviewed by
Cao and Tamura (2018). But at the supercritical Re, the mechanism for the preservation of the nearly symmetric flow in the
shear inflow needs to be clarified. It is believed to be caused by the significant reduction in the velocity gap between the high-
and low-velocity sides. Firstly, the significant decrease in the velocity gap on two sides of cylinder is evidenced in Fig. 20.
In the shear inflow, the acceleration of velocity on the low-velocity side is obviously greater than that on the high-velocity
side compared with the profiles of oncoming flow. This phenomenon is called ‘‘self-adjust’’ mechanism at the subcritical Re
by Cao et al. (2010), Cao et al. (2014) and Cao and Tamura (2018). This study confirms that it is also true for supercritical
Re. Furthermore, this mechanism is considerably strengthened at the supercritical Re. That is to say, at the subcritical Re,
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 489

Fig. 20. Profiles of mean streamwise velocity along the vertical direction at several streamwise stations x = −0.5, 0, and 0.5: (a) Re = 2.2 × 104 ; (b)
Re = 1.0 × 106 .

the velocity peaks on the low-velocity side (marked by the red ellipses in Fig. 20(a)) have a noticeable disparity between
the uniform and shear inflows; by contrast, the velocity peaks on the low-velocity side (marked in Fig. 20(b)) are greatly
accelerated and become very close to those in the uniform inflow. It is also applicable to all the stations when x ranges from
−0.5 to 0.5. More importantly, the values of peaks are quite comparative to those on the high-velocity side (slightly smaller
to be exact).
The engine of greater velocity acceleration on the low-velocity side at the supercritical Re is expected to be examined.
One can notice from Fig. 20(b) that the velocity gap between the high- and low-velocity sides at x = −0.5 has become
rather small. It indicates that the laminar boundary layer on the frontal face has a great contribution, which is thought not
to be influenced by the shear-layer and wake behavior in the mean sense. Thus the mean pressure gradients along the flow
direction on the frontal faces (‘‘R1’’) and the upstream part of the frontal corners (‘‘R2’’) are emphasized, as shown in Fig. 21.
It should be noted that the circumferential regions ‘‘R1+R2’’ (i.e., the regions of α and β less than 45◦ ) are prior to the
mean flow separation points for both Reynolds numbers and both inflow conditions. At K = 0.4 and Re = 2.2 × 104 , the
favorable pressure gradients are comparable between the ‘‘R1’’ regions of the low- and high-velocity sides. The pressure
gradients in the ‘‘R2’’ region of the high-velocity side are larger than those on the low-velocity side. Being opposite to the
subcritical Re, the favorable pressure gradients at Re = 1.0 × 106 in both ‘‘R1’’ and ‘‘R2’’ regions of the low-velocity side are
obviously larger than those of the high-velocity side. As a result, the fluid flow quickens its pace on the low-velocity side and
closes the velocity gap on the high- and low-velocity sides brought by the shear inflows. Eventually the flow has reached the
supercritical-Re status before separation; the followed turbulence transition and flow reattachment are accordingly possible.
That is, the overall flow pattern even under the strong shear inflow can be possibly preserved.
Another interpretation of ‘‘self-adjust’’ phenomenon is also provided on the basis of the blocking effects and the locations
of upstream stagnation points. In the uniform inflow, the upstream stagnation point is at the center of the frontal wall. The
fluid acceleration because of the blocking effects tends to be the same in degree on two sides of the cylinder. However, in
the shear inflow, the stagnation point moves towards the high-velocity side, which forces more fluids to flow and converge
to the low-velocity side. As a result of stronger blocking effect on the low-velocity side, the velocity acceleration on the
low-velocity side is greater than on the high-velocity side. In addition, the ‘‘self-adjust’’ phenomenon in the shear inflow
past a bluff body seems an elemental example for the non-uniform upstream flows through a screen which has the effect of
reduction in the velocity variation in the downstream profiles. However, a screen is essentially, from the aerodynamic point
of view, a complicated distribution of bluff bodies (Laws and Livesey, 1978). The deeper discussions about the similarity of
aerodynamics between a screen and a bluff body are beyond the scope of the present work, but worth further investigations
in the future.

6. Conclusions

This LES study investigated Re effects on shear inflows past a rounded-corner square cylinder when Re = 2.2 × 104 and
1.0 × 106 , with a focus on the latter Re. The shear parameter K ranges from 0 to 1.0. It was shown that shear inflows have
different effects on the aerodynamic characteristics of the cylinder in two Re regimes. Generally the shear effects are smaller
in the supercritical regime. The mechanism for Re effects was also explored. The concrete conclusions are as follows:
(1) The variations of mean and r.m.s forces induced by the shear inflows are smaller at the supercritical Re than at the
subcritical Re, despite of similar movement of frontal stagnation points. In particular, the mean lifts are exerted from the
high- to the low-velocity side at the supercritical Re, which is opposite to those at the subcritical Re.
(2) The strong dependency of mean and r.m.s pressure distributions on K observed at the subcritical Re is no longer found
at the supercritical Re. This finding is associated with the preservation of mean flow patterns at various K when Re is
490 Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491

Fig. 21. Distributions of mean pressure gradients along the circumferential angle on the frontal face and the upstream part of the frontal corners before
flow separation: (a) on the high-velocity side; (b) on the low-velocity side.

supercritical. Specifically the overall flows are along the cross section and separate from the trailing corners on both sides of
the cylinder. Twin vortices form behind the cylinder despite the existence of slight asymmetry between the upper and lower
vortices. The slight asymmetry is associated to the small-degree rotation of the near-wake centerline in the anticlockwise
direction from the horizontal axis. On the other hand, at the subcritical Re and sufficiently high K, the suspended shear
layer on the high-velocity side shows the flow reattachment on the upper wall, while the shear layer on the low-velocity
side separates completely and possesses a farther distance from the bottom wall. The shear-layer behaviors result in a
highly asymmetric near wake, which has a near-wake centerline rotated in the clockwise direction from the horizontal
axis. Regardless of Re regimes, the swirling strengths of the primary vortices on the high-velocity side are increased while
those on the low-velocity side are weakened.
(3) In terms of the individual flow features at the supercritical Re, the oscillation degree of the shear layer separating from the
trailing corner on the high-velocity side is greater than on the low-velocity side, and hence the larger fluctuating velocities.
The turbulent boundary layer on the upper wall is thinner than that on the lower side wall, which is especially obvious at
K = 1.0 and BR = 50%.
(4) The mechanism for strong Re effects in the shear inflows past the cylinder was also explored. The ‘‘self-adjust’’
phenomenon is amplified at the supercritical Re: the fluid flow on the low-velocity side quickens its pace and closes the
velocity gap on the high- and low-velocity sides brought by the shear inflows, as a result of greater favorable pressure
gradients on the frontal faces and the upstream part of the frontal corners. Eventually the flow has reached the supercritical-
Re status before separation. The ‘‘self-adjust’’ phenomenon can be also understood from the perspective of the blocking
effects and the movement of upstream stagnation point. In the shear inflow, the movement of stagnation point towards the
high-velocity side forces more fluids to flow and converge to the low-velocity side. As a result of stronger blocking effect,
the velocity acceleration on the low-velocity side is greater.

Acknowledgements

The computations were conducted on the Earth Simulator at the Japan Agency of Marine-Earth Science and Technology
(JAMSTEC). The authors are very grateful to the editors and reviewers for the constructive suggestions and discussions, which
help to improve the manuscript greatly.

References

Bearman, P.W., Zdravkovich, M.M., 1978. Flow around a circular cylinder near a plane boundary. J. Fluid Mech. 89 (01), 33–47.
Bruno, L., Coste, N., Fransos, D., 2012. Simulated flow around a rectangular 5:1 cylinder: Spanwise discretisation effects and emerging flow features. J. Wind
Eng. Ind. Aerodyn. 104, 203–215.
Cantwell, B., Coles, D., 1983. An experimental study of entrainment and transport in the turbulent near wake of a circular cylinder. J. Fluid Mech. 136,
321–374.
Cao, S., Ozono, S., Hirano, K., Tamura, Y., 2007. Vortex shedding and aerodynamic forces on a circular cylinder in linear shear flow at subcritical Reynolds
number. J. Fluids Struct. 23 (5), 703–714.
Cao, S., Ozono, S., Tamura, Y., Ge, Y., Kikugawa, H., 2010. Numerical simulation of Reynolds number effects on velocity shear flow around a circular cylinder.
J. Fluids Struct. 26 (5), 685–702.
Cao, S., Zhou, Q., Zhou, Z., 2014. Velocity shear flow over rectangular cylinders with different side ratios. Comput. & Fluids 96, 35–46.
Cao, Y., Tamura, T., 2015. Numerical investigations into effects of three-dimensional wake patterns on unsteady aerodynamic characteristics of a circular
cylinder at Re = 1.3 × 105 . J. Fluids Struct. 59, 351–369.
Y. Cao, T. Tamura / Journal of Fluids and Structures 82 (2018) 473–491 491

Cao, Y., Tamura, T., 2016. Large-eddy simulations of flow past a square cylinder using structured and unstructured grids. Comput. & Fluids 137, 36–54.
Cao, Y., Tamura, T., 2017. Supercritical flows past a square cylinder with rounded corners. Phys. Fluids 29 (8), 085110.
Cao, Y., Tamura, T., 2018. Shear effects on flows past a square cylinder with rounded corners at Re = 2.2 × 104 . J. Wind Eng. Ind. Aerodyn. 174, 119–132.
Carassale, L., Freda, A., Marrè-Brunenghi, M., 2014. Experimental investigation on the aerodynamic behavior of square cylinders with rounded corners. J.
Fluids Struct. 44, 195–204.
Cheng, M., Whyte, D.S., Lou, J., 2007. Numerical simulation of flow around a square cylinder in uniform-shear flow. J. Fluids Struct. 23 (2), 207–226.
Delany, N.K., Sorensen, N.E., 1953. Low-Speed drag of cylinders of various shapes. Technical Report TN3038 NACA.
Germano, M., Piomelli, U., Moin, P., Cabot, W.H., 1991. A dynamic subgrid-scale eddy viscosity model. Phys. Fluids 3 (7), 1760–1765.
Hu, J.C., Zhou, Y., Dalton, C., 2006. Effects of the corner radius on the near wake of a square prism. Exp. Fluids 40 (1), 106.
Kajishima, T., Taira, K., 2017. Computational Fluid Dynamics, Incompressible Turbulent Flows. Springer, Cham, Switzerland.
Kang, S., 2006. Uniform-shear flow over a circular cylinder at low Reynolds numbers. J. Fluids Struct. 22 (4), 541–555.
Kawai, H., 1998. Effect of corner modifications on aeroelastic instabilities of tall buildings. J. Wind Eng. Ind. Aerodyn. 74, 719–729.
Kawamura, T., Kuwahara, K., 1984. Computation of high Reynolds number flow around a circular cylinder with surface roughness. In: AIAA 22nd Aerospace
Science Meeting.
Kim, J., Moin, P., 1985. Application of a fractional-step method to incompressible Navier–Stokes equations. J. Comput. Phys. 59 (2), 308–323.
Kiya, M., Arie, M., Tamura, H., 1979. Forces acting on circular cylinders placed in a turbulent plane mixing layer. J. Wind Eng. Ind. Aerodyn. 5 (1–2), 13–33.
Kiya, M., Tamura, H., Arie, M., 1980. Vortex shedding from a circular cylinder in moderate-Reynolds-number shear flow. J. Fluid Mech. 101 (04), 721–735.
Kwok, K.C.S., Wilhelm, P.A., Wilkie, B.G., 1988. Effect of edge configuration on wind-induced response of tall buildings. Eng. Struct. 10 (2), 135–140.
Laima, S., Jiang, C., Li, H., Chen, W., Ou, J., 2018. A numerical investigation of Reynolds number sensitivity of flow characteristics around a twin-box girder.
J. Wind Eng. Ind. Aerodyn. 172, 298–316.
Lamballais, E., Silvestrini, J., Laizet, S., 2010. Direct numerical simulation of flow separation behind a rounded leading edge: Study of curvature effects. Int.
J. Heat Fluid Flow 31 (3), 295–306.
Lankadasu, A., Vengadesan, S., 2008. Onset of vortex shedding in planar shear flow past a square cylinder. Int. J. Heat Fluid Flow 29 (4), 1054–1059.
Lankadasu, A., Vengadesan, S., 2009. Influence of inlet shear on the 3-D flow past a square cylinder at moderate Reynolds number. J. Fluids Struct. 25 (5),
889–896.
Laws, E.M., Livesey, J.L., 1978. Flow through screens. Annu. Rev. Fluid Mech. 10 (1), 247–266.
Lehmkuhl, O., Rodríguez, I., Borrell, R., Chiva, J., Oliva, A., 2014. Unsteady forces on a circular cylinder at critical Reynolds numbers. Phys. Fluids 26 (12),
125110.
Lilly, D.K., 1992. A proposed modification of the Germano subgrid-scale closure method. Phys. Fluids 4 (3), 633–635.
Norberg, C., 1986. Interaction between freestream turbulence and vortex shedding for a single tube in cross-flow. J. Wind Eng. Ind. Aerodyn. 23, 501–514.
Ono, Y., Tamura, T., 2002. Large eddy simulation using a curvilinear coordinate system for the flow around a square cylinder. Wind Struct. 5 (2–4), 369–378.
Ono, Y., Tamura, T., 2008. LES of flows around a circular cylinder in the critical Reynolds number region. In: BBAA VI International Colloquium on: Bluff
Bodies Aerodynamics and Applications (Politecnico di Milano, Milano, Italy), pp. 1–10.
Reynolds, W.C., Hussain, A.K.M.F., 1972. The mechanics of an organized wave in turbulent shear flow. Part 3. Theoretical models and comparisons with
experiments. J. Fluid Mech. 54 (02), 263–288.
Schewe, G., 2013. Reynolds-number-effects in flow around a rectangular cylinder with aspect ratio 1:5. J. Fluids Struct. 39, 15–26.
Schewe, G., Larsen, A., 1998. Reynolds number effects in the flow around a bluff bridge deck cross section. J. Wind Eng. Ind. Aerodyn. 74, 829–838.
Sumner, D., Akosile, O.O., 2003. On uniform planar shear flow around a circular cylinder at subcritical Reynolds number. J. Fluids Struct. 18 (3), 441–454.
Tamura, T., Miyagi, T., Kitagishi, T., 1998. Numerical prediction of unsteady pressures on a square cylinder with various corner shapes. J. Wind Eng. Ind.
Aerodyn. 74, 531–542.
Tamura, T., Ono, Y., 2003. LES analysis on aeroelastic instability of prisms in turbulent flow. J. Wind Eng. Ind. Aerodyn. 91 (12), 1827–1846.
Tse, K.T., Hitchcock, P.A., Kwok, K.C., Thepmongkorn, S., Chan, C.M., 2009. Economic perspectives of aerodynamic treatments of square tall buildings. J. Wind
Eng. Ind. Aerodyn. 97 (9), 455–467.
van Hinsberg, N.P., Schewe, G., Jacobs, M., 2017. Experiments on the aerodynamic behaviour of square cylinders with rounded corners at Reynolds numbers
up to 12 million. J. Fluids Struct. 74, 214–233.
van Hinsberg, N.P., Schewe, G., Jacobs, M., 2018. Experimental investigation on the combined effects of surface roughness and corner radius for square
cylinders at high Reynolds numbers up to 107 . J. Wind Eng. Ind. Aerodyn. 173, 14–27.
Wang, X., Gu, M., 2016. Experimental study on Re number effects on aerodynamic characteristics of 2D square prisms with corner modifications. Wind
Struct. 22 (5), 573–594.
Zang, Y., Street, R.L., Koseff, J.R., 1993. A dynamic mixed subgrid-scale model and its application to turbulent recirculating flows. Phys. Fluids 5 (12), 3186–
3196.
Zhang, H.Q., Fey, U., Noack, B.R., König, M., Eckelmann, H., 1995. On the transition of the cylinder wake. Phys. Fluids 7 (4), 779–794.

Vous aimerez peut-être aussi