A Numerical Study of Geometric Effects on Vortex Shedding From Elongated

Bluff Bodies
Zhigang Liu1, Gregory A. Kopp2
1

Ph.D. Candidate, Boundary Layer Wind Tunnel Laboratory, University of Western Ontario,
London, Ontario, Canada, zliu49@uwo.ca
2
Professor, Boundary Layer Wind Tunnel Laboratory, University of Western Ontario, London,
Ontario, Canada, gakopp@uwo.ca

ABSTRACT
Strouhal numbers for flat plates with various nosings are obtained from numerical simulations
with a two-dimensional, high-resolution vortex particle method at Re = 600. The well-known
step-wise variation of chord-based Strouhal numbers (Stc) with chord-to-thickness ratios (c/t) for
rectangular plates is replicated, and Stc for flat plates with a series of different triangular
nosings are also presented in a similar way. The flow fields are explored and the effects of
nosing shapes on vortex shedding from elongated bluff bodies are discussed.

INTRODUCTION
Much progress toward understanding flows around rectangular plates has been achieved in the
past three decades. Based on wind tunnel experiments on rectangular plates at Re = 14,80031,100, Parker & Welsh [2] suggest that there exist four possible vortex shedding regimes
depending on the chord-to-thickness ratio (c/t, referred to as elongation ratio hereafter), which
are summarized by Stokes & Welsh [3] as follows: (a), short plates (c/t < 3.2): the leading-edge
shear layers interact directly in the wake to form a regular vortex street; (b), longer plates (3.2 <
c/t < 7.6): the leading-edge shear layers reattach to the plate "periodically"; following
reattachment the separation bubble "grows" beyond the trailing edge letting fluid in the
recirculation region enter the wake to form vortices, and a regular vortex street is usually
observed; (c), still longer plates (7.6 < c/t < 16): the leading-edge shear layers always reattach
upstream of the trailing edge, and the separation bubble tends to "grow and divide in a random
manner" and discrete vortices are generated; these vortices arrive at the trailing edge randomly
and interfere with separation at the trailing edge and no regular shedding is observed; (d), very
long plates (c/t > 16): the vortices shed from the separation bubble are diffused upstream of the
trailing edge, where the re-developed boundary layers separate again and interact in the wake to
form a regular vortex street.
Nakamura & Nakashima [1] argue that the shear layers in regime (b) can be characterized
by the "impinging-shear-layer instability" (ISLI), a concept originally developed for cavity flows
[4,5], whereby a leading-edge shear layer becomes unstable in the presence of a sharp trailingedge corner. Since this is a single-layer instability, it exists even when a splitter plate is installed
in the wake. The frequency of this single layer instability is almost constant at Stc = fvc/U = 0.6,
which is consistent with their observations for elongated cylinders with c/t = 2-5. More wind
tunnel experiments on rectangular cylinders covering an elongation-ratio range of c/t = 3-16 are
reported by Nakamura et al. [6]. Regular vortex shedding was observed for c/t = 3-15 at Re =
1,000, but disappeared for c/t > 12 at Re > 2,000. They argue that this is because the ISLI,
although still present in the flow, is "masked" by the increased turbulence fluctuations at higher
Reynolds numbers, and can be excited by external forcing. A step-wise variation of Stc with c/t

is identified by Nakamura et al., whereby Stc is nearly constant at 0.6n for each step, in which n
is considered to be the wave number of the instability on the cylinder surface. It is hypothesized
that the instability wavelength is locked to the chord due to "some nonlinear flow processes"
until jumps occur at certain elongation ratios. They speculate that the feedback from the
"impinging edge" tends to affect the vorticity generation at the separation point and greatly
enhances the "shear-layer instability." Naudascher & Wang [7], and Naudascher & Rockwell [8]
suggest that the flow in regimes (a) and (d) are characterized by "leading-edge vortex shedding"
(LEVS) and "trailing-edge vortex shedding" (TEVS), respectively. For regimes (b) and (c), the
flow is governed by "a flow instability equivalent to that of impinging shear layers," whereby
vortices are produced in the unstable leading-edge shear layers and are convected downstream.
Weak pressure pulses are generated when they pass the trailing edge and are fed back upstream
to the leading edge to control the formation of new vortices. Based on numerical simulations at
Re 700, Hourigan et al. [9] suggest that the trailing-edge vortex shedding also plays an
important role in the self-sustained flow oscillations around long rectangular plates. They argue
that trailing-edge vortices formed by the separation of the developing boundary layer after
upstream reattachment can only form between the arrivals of leading-edge vortices. TEVS can
generate a stronger pressure pulse than does a leading-edge vortex passing the trailing edge,
which travels upstream to regulate the "preferred" shedding frequency to be that of the TEVS.
Tan et al. [10] have shown, through numerical simulations, that the change of the vorticity flux
into the leading-edge shear layer due to the effect of trailing-edge shedding is well correlated
with the variation of the total vorticity flux, whereas that due to the passage of leading-edge
vortices is poorly correlated. Mills et al. [11,12], and Tan et al. [10] speculate that, the naturally
preferred "global oscillation frequency" of the flow around a long rectangular plate is the one
that leads to the peak base suction, or the most intense trailing-edge shedding, which generates
the "dominant" pressure pulse feedback to the leading edge. It is speculated that since the
vortices that are periodically shed from the leading edge are convected downstream at a nearly
constant velocity Uc independent of c/t, it takes longer for leading edge vortices to arrive at the
trailing edge of a longer plate. However, it always takes an integer number of shedding cycles
for the leading edge vortices to arrive at the trailing edge regardless of c/t values, because their
generation is controlled by the feedback pressure pulse generated at the trailing edge, which
occurs only between the successive arrivals of leading edge vortices. Tan et al. [10] argue that
the trailing-edge shedding, which is globally unstable, can only exist over a narrow frequency
band, and hence it may be controlling the frequency selection. As c/t increases, the preferred
frequency of trailing-edge shedding may no longer be possible for certain elongation ratios
because of "interference" from leading-edge vortices, and a "nearby" frequency is selected at
which trailing-edge shedding could occur and the shedding jumps to a higher mode. As
Reynolds number is increased above 2000 for long rectangular plate, the relatively weak
feedback mechanism may be "broken down" by the enhanced inherent turbulence in the flow, but
the same locking effect may be excited by external forcing with appropriate frequency and
amplitude.
The studies on rectangular plates have shed new light on understanding VIVs of bridges.
However, due to the lack of knowledge regarding the effects of geometric details (e.g., nosing
shapes, and asymmetry of upper and lower surfaces, etc.), application of the main findings from
these studies, such as the classification of flow regimes based on elongation ratios, and the stepwise variation of Stc, to flows around bridge decks is still limited. Bridge deck nosings are
usually designed to form obtuse angles with the upper and lower surfaces, which lead to smaller
separation angles and earlier reattachment, and thus, significantly modify the flow patterns. In
the present paper, we will focus on flows around flat plates with various triangular nosings.

Numerical simulations are considered a preferable tool for such studies due to the large
parameter space and the need for flow visualizations close to the body surfaces.
Vortex particle methods have several distinct advantages over conventional grid-based
methods. As pointed out by Anderson & Greengard [13] and Cottet et al.[14], time-step
limitations are far less restrictive for vortex methods than for grid-based methods because vortex
methods are free of convection-related stability conditions, that is, large time steps may be used
and substantial savings of computational time over grid-based methods can be achieved (see also
[15]). This is in addition to considerable time savings due to the fact that the vorticity field that
needs to be solved in vortex methods usually exists only in small fractions of the whole flow
field. Far-field boundary conditions are automatically satisfied for vortex particles with a
compact support and hence there is no need for a large computational domain to minimize
blockage effects. Another advantage of vortex methods over grid-based methods is that the
interactions between computational elements mimic the physics they represent and hence no
additional criteria for "critical parts" of the flow (which are not always known a priori) are
necessary [16,17,15]. Moreover, the Lagrangian formulation means that the there is no
numerical diffusion associated with convection, and the dynamics are not constrained by a
prescribed minimum scale and thus is a robust way for high-Reynolds number flows simulations.
In the present work, numerical simulations of flows around flat plates with various
nosings at Re = 600 are performed with a 2-D high-resolution vortex particle method. The wellknown step-wise variation of step-wise variation of Stc with c/t for rectangular plates are
replicated, and the effects of nosing shapes on vortex shedding from elongated bluff bodies are
investigated.

NUMERICAL METHOD AND SIMULATIONS

A 2-D high-resolution vortex particle method is used for the present simulations; for detailed
implementation and validation of the code the reader is referred to [18]. Three-dimensional high
resolution simulations are still far too computationally expensive for the present study. 2-D
simulations, of course, will not capture any 3-D effects. It is pointed out by Mittal &
Balachandar [19] that, in 2-D simulations, all the energy extracted from the mean flow field is
used to drive the in-plane velocity fluctuations, while part of the energy is spent on sustaining
spanwise velocity fluctuations in 3-D computations, leading to a reduction in in-plane
fluctuations of velocities and Reynolds stresses. It is suggested that the effect of higher in-plane
Reynolds stresses in the near wake on surface pressures is one of the main reasons for the overprediction of drag in 2-D simulations, and the discrepancies are dependent on the geometry.
They also find that the vortex formation region is usually shorter in 2-D than in 3-D results,
which, although has been shown to have little direct effect on the mean drag, brings the intense
Reynolds stress region closer to the body, and is considered to be partly responsible for larger
amplitudes of lift fluctuations predicted in 2-D simulations. On the other hand, 2-D simulations
have also been shown to produce reasonably accurate predictions in many situations, especially
for flows around sharp-edged bluff bodies, see [18] for examples and a complete discussion.
The Reynolds number, Re = 600, in the present study was chosen as a compromise
between physical requirements and computational costs. At low Reynolds numbers 3-D effects
are less pronounced and 2-D simulations would be more accurate, and a lower Reynolds number
is also desirable to reduce computational costs. However, the wind tunnel experiments at Re =
200-1000 by Nakamura et al. [20] have shown that the step-wise variation of Stc for rectangular
plates only occurs above Re = 300. When Re < 250, the separated shear layers appear to be
steady and the trailing edge shedding dominates the wake, which leads to an almost linear

variation of Stc with c/t. This is consistent with the experimental results of Lane and Loehrke
[21] who found that the separation bubble on a long blunt plate is steady up to Re = 260. It is
also shown by Lane & Loehrke that the length of separation bubble increases with Reynolds
number up to Re = 325, and then shrinks with further increase in Reynolds number and
asymptotes to 4-5 plate thicknesses after Re = 700. Through two-dimensional numerical
simulations of separated-reattaching flows over a normal plate followed by a long splitter plate,
Najjar & Vanka [22] also find that "a clear shedding" of vortices from the separation bubble only
occurs above Re > 250. On the other hand, flow visualizations by Sasaki & Kiya [23] suggest
that the leading-edge separation bubble on a blunt flat plate becomes three-dimensional at Re >
320. Strouhal numbers obtained at relatively low Reynolds numbers, both experimentally (e.g.
water tunnel experiments at Re = 490 by Mills et al. [12]) and numerically (e.g. 2-D spectral
element method simulations at Re = 400 Tan et al. [24]), are in good agreement with the
experimental data Re = 1000 by Nakamura et al. [6]. Nakamura et al. [20] also report that an
increase in Reynolds number from 300 to 1000 only lead to higher levels of "random
modulations" in the velocity signals, and the shedding frequencies in the wake of rectangular
plates remain unchanged. It is also demonstrated by Tafti & Vanka [25] that 2-D LES
simulations of flows around rectangular plates at Re = 1000 were able to predict time-dependent
features, such as leading-edge shedding frequencies and vortex convection velocities, in good
agreement with experiments at much higher Reynolds numbers. Actually, as long as the
Reynolds number is high enough for the leading-edge shedding to occur, similar characteristics
of flows around long rectangular plates have been reported over three decades of variation in
Reynolds number; see Table 1 for a summary of such observations. In other words, the flows
around sharp edged rectangular plates are insensitive to the variations of Reynolds numbers for
Re > 300, hence a comparison between the results obtained at various Reynolds numbers are
reasonable. This also justifies the use of 2-D simulations at a moderate Reynolds number in the
present work to study this phenomenon.
Table 1. Experiments and numerical simulations where step-wise variation of Stc is observed
Source

Re

Method

Nakamura et al. [20]

Tan et al. [24, 10]
Mills et al. [12]
Nakamura et al. [6]
Mills et al. [11]
Stokes & Welsh [3]
Parker & Welsh [2]
Nakamura & Nakashima [1]

300 1,000
400
490
1,000 3,000
5,000 1,3000
8,000 44,300
14,800 31,100
50,000 300,000

Wind tunnel experiments

Numerical simulations
Water tunnel experiments
Wind tunnel experiments
Wind tunnel experiments
Wind tunnel experiments
Wind tunnel experiments
Wind tunnel experiments

In the present simulations, flat plates with various nosings and elongation ratios (c/t)
ranging from 3 to 10 were placed in a uniform flow with free stream velocity U. The Reynolds
number based on free stream velocity and the thickness of the plates was 600. To balance
accuracy and computational costs, the normalized time step was set to T = 0.05. A unity
overlap ratio was used throughout the simulations, and the particle spacing was chosen to be
h/t = 1.29110-2 so that t/h2 = 0.5, where t is the time step. The solid boundaries were
discretized into panels of lengths h, and the sharp corners were "rounded" with radius r =
for better numerical behavior. The fluid was impulsively set into motion with velocity U at the

beginning of the simulation, and remeshing was applied every 5 time steps. All the
computations were carried out on a desktop computer, and the simulations were performed long
enough for the starting effects to phase out. In most cases 15-25 stable shedding cycles were
observed, with 3-5 million particles in the flow fields toward the end of the simulations.

RESULTS AND DISCUSSIONS

The step-wise variation of the chord-based Strouhal number Stc = fvc/U, where fv is the shedding
frequency determined from the power spectra of lift history as well as instantaneous velocities at
a number of locations in the wake, with c/t is shown in Figure 1, in comparison with
experimental and numerical results available in the literature. A second symbol for the same set
of data indicates that a secondary peak was observed in spectral analysis for certain elongation
ratios.

Figure 1: Step-wise variation of chord-based Strouhal numbers Stc for rectangular plates

It can be observed from Figure 1 that the present results compare very well to the LES
simulations by Ohya et al. [26], and spectral element method simulations by Tan et al. [24]. For
c/t = 3-6, the present results match the experimental results almost exactly. For larger c/t, small
discrepancies can be observed, and they tend to be a little larger for c/t = 9-10. A possible
explanation of these disparities is that the speed-up of the flow around the plates due to blockage
in experiments may result in a higher convection speed of leading-edge vortices and hence a
higher shedding frequency. In their 2-D spectral element method simulations, Tan et al. [10]
also find that Stc = 0.55n instead of 0.6n (n = 1,2,3) commonly observed in experiments. They
speculate that higher Reynolds numbers, and hence three-dimensional effects in experiments
may lead to higher average convection velocities of leading-edge vortices, which in turn result in
higher Strouhal numbers. An additional simulation at a higher resolution with T = 0.025 and
t/h2 = 0.5 was performed for c/t = 7 to check the dependence of the results on numerical
resolution, and virtually no change in Strouhal number was observed, which indicates that the
chosen resolution (T = 0.05 and t/h2 = 0.5) is adequate to resolve the flows under
investigation. The good agreement between the present 2-D simulations and the observations in

experiments at higher Reynolds number suggests that, the present simulations with the vortex
particle code are capable of capturing the physics of interest in such flows, a point which is
explicitly examined in [18].

Figure 2: Sketch of a flat plate with triangular nosings

Figure 3: Chord-based Strouhal numbers for flat plates with various nosings at Re = 600

A sketch of a flat plate with triangular nosings is shown in Figure 2. The angle was
chosen to be 60, 75, 82.5, 90, 120 and 150 in the present simulations, and a rectangular plate
may be fit into this series with = 180. Figure 3 shows the variations of chord-based Strouhal
numbers (Stc) for a series of flat plates with triangular nosings. The dashed straight line is a
reference line with a slope of 0.2, and a symbol with a "+" at its center indicates that a secondary
peak was observed in spectral analysis. Also shown for comparison is the Stc variation for flat
plates with a streamlined leading edge (half Rankine Oval with a width to height ratio of 2) and a
rectangular trailing edge. Figure 3 shows that there are two limits for Stc variations: horizontal
lines with jumps at certain c/t ratios (lower limit, rectangular plates), and a straight line (upper
limit, plates with streamlined leading edge) where Stc varies almost linearly with c/t. It can be
inferred from the discussions in the Introduction about the mechanism leading to the step-wise
variation of Stc for rectangular plates that, when leading-edge vortices are strong enough to
interfere with the trailing-edge shedding, Stc tends to follow the horizontal line as in the case of
rectangular plates. For plates with a streamlined leading-edge, the trailing-edge shedding
dominates the wake and hence an almost linear variation of Stc is observed. By reducing the
separation angle at the leading edge (which is half of ), the transition from the lower to upper

limit appears to be continuous: Stc jumps occur earlier, and the sizes of which tend to be smaller,
until finally they disappear and Stc variation follows the straight line (e.g., Stc variation for plates
with 60 triangular nosings in Figure 3). Additional simulations for plates with a triangular
leading edge and a rectangular trailing edge have shown that, the shape of trailing-edge has only
a minor influence on the shedding frequency.

Figure 4: Streamwise RMS velocities for flat plates with various nosings and c/t = 7; upper half: just
upstream of trailing-edge separation, lower half: one thickness downstream of trailing-edge separation

Figure 4 shows the streamwise RMS (root-mean-square) velocities (urms) for plates with
various nosings and c/t = 7 just upstream (upper half) and 1 thickness downstream (lower half)
the trailing-edge separation point, in which y is measured from the center of the plate in the
cross-scream direction. It is speculated that the secondary peaks on the urms curves are the
"footprints" of large scale leading-edge vortices, hence a secondary peak on the curve just before
trailing-edge separation (TES) is considered to be an indication of strong leading-edge vortices
covecting past the trailing edge, and a leading-edge vortex is considered to have survived the
interaction with trailing-edge shedding if a secondary peak is observed on the urms curve in the
near wake. In Figure 4, secondary peaks are observed on both urms curves for all plates with
82.5. This is consistent with the observation that, for c/t = 7, Stc for plates with 82.5 lie on
the horizontal line (lower limit), and those for 75 lie close to the straight line (upper limit).
A closer look at the urms curves for = 75 reveals that there is a small secondary peak, although
hardly discernible in Figure 4 due to the scaling, on the one just before TES, but which
disappears in the near wake. That is, statistically, the leading-edge vortices do not survive the
interaction at the trailing edge and the wake is dominated by trailing-edge shedding. However,
the interaction still leads to slight deviation of the Stc-c/t relationship for = 75 from the upper
limit as observed in Figure 3.

CONCLUSIONS
There are two limits for Stc variations for flat plates with triangular nosings: horizontal lines with
jumps at certain c/t ratios for blunt plates where leading-edge shedding and its interaction with

trailing edge shedding control the wake (lower limit), and a straight line for plates with a
streamlined leading edge where trailing-edge shedding dominates the wake (upper limit). There
appears to be a continuous transition from the lower to upper limit: Stc jumps occur earlier for
more streamlined plates, and the sizes of jumps become smaller, until they disappear to reach the
upper limit and the wake is dominated by the shedding at the trailing edge.

REFERENCES
[1] Y. Nakamura, M. Nakashima, Vortex excitation of prisms with elongated rectangular, H and T crosssections, J. Fluid Mech. 163 (1986) 149-169
[2] R. Parker, M.C. Welsh, Effects of sound on flow separation from blunt flat plates, Int. J. Heat Fluid
Flow 4 (1983) 113-127
[3] A.N. Stokes, M.C. Welsh, Flow-resonant sound interaction in a duct containing a plate, II: square
leading edge, J. Sound Vib. 104 (1986) 55-73
[4] D. Rockwell, E. Naudascher, Self-sustaining oscillations of flow past cavities, J. Fluids Eng. 100
(1978) 152-165
[5] D. Rockwell, E. Naudascher, Self-sustained oscillations of impinging free shear layers, Annu. Rev.
Fluid Mech. 11 (1979) 67-94
[6] Y. Nakamura, Y. Ohya, H. Tsuruta, Experiments on vortex shedding from flat plates with square
leading and trailing edges, J. Fluid Mech.222 (1991) 437-447
[7] E. Naudascher, Y. Wang, Flow-induced vibrations of prismatic bodies and grids of prisms, J. Fluids
Struct. 7 (1993) 341-373
[8] E. Naudascher, D. Rockwell, Flow-induced vibration: An engineering guide, A.A. Balkema,
Rotterdam, 1994
[9] K. Hourigan, M.C. Thompson, B.T. Tan, Self-sustained oscillations in flows around long blunt plates,
J. Fluids Struct. 15 (2001) 387-398
[10] B.T. Tan, M.C. Thompson, K. Hourigan, Flow past rectangular cylinders: receptivity to transverse
forcing, J. Fluid Mech. 515 (2004) 33-62
[11] R. Mills, J. Sheridan, K. Hourigan, Response of base suction and vortex shedding from rectangular
prisms to transverse forcing, J. Fluid Mech. 461 (2002) 25-49
[12] R. Mills, J. Sheridan, K. Hourigan, Particle image velocimetry and visualization of natural and
forced flow around rectangular cylinders, J. Fluid Mech. 478 (2003) 299-323
[13] C. Anderson, C. Greengard, On vortex methods, SIAM J. Numer. Anal. 22 (1985) 413-440
[14] G.-H. Cottet, B. Michaux, S. Ossia, G. VanderLinden, A comparison of spectral and vortex methods
in three-dimensional incompressible flows, J. Comput. Phys. 175 (2002) 702-712
[15] P.D. Koumoutsakos, Multiscale flow simulations using particles, Annu. Rev. Fluid Mech. 37 (2005)
457-487

[16] J.T. Beale, A. Majda, Vortex methods. I: Convergence in three dimensions, Math. Comput. 39 (1982)
1-27
[17] J.T. Beale, A. Majda, Vortex methods. II: Higher order accuracy in two and three dimensions, Math.
Comput. 39 (1982) 29-52
[18] Z. Liu, G.A. Kopp, Numerical simulations of bluff-body flows with a high-resolution vortex method,
in preparation
[19] R. Mittal, S. Balachandar, Effect of three-dimensionality on the lift and drag of nominally twodimensional cylinders, Phys. Fluids 7 (1995) 1841-1865
[20] Y. Nakamura, Y. Ohya, S. Ozono, R. Nakayama, Experimental and numerical analysis of vortex
shedding from elongated rectangular cylinders at low Reynolds numbers 200-103. J. Wind Eng. Ind.
Aerodyn. 65 (1996) 301-308
[21] J.C. Lane, R.I. Loehrke, Leading edge separation from a blunt plate at low Reynolds number, J.
Fluids Eng. 102 (1980) 494-496
[22] F.M. Najjar, S.P. Vanka, Numerical Study of Separated-Reattaching Flow, Theor. Comput. Fluid
Dynam. 5 (1993) 291-308
[23] K. Sasaki & M. Kiya, Three-dimensional vortex structure in a leading-edge separation bubble at
moderate Reynolds numbers, J. Fluids Eng. 113 (1991) 405-410
[24] B.T. Tan, M.C. Thompson, K. Hourigan, Flow around long rectangular plates under cross-flow
perturbations, Int. J. Fluid Dynam. 2 (1998), Article 1 (available at
http://elecpress.monash.edu.au/ijfd/1998_vol2/paper1/paper1.html)
[25] D.K. Tafti, S.P. Vanka, A three-dimensional numerical study of flow separation and reattachment on
a blunt plate, Phys. Fluids A 3(1991) 2887-2909
[26] Y. Ohya, Y. Nakamura, S. Ozono, H. Tsuruta, R. Nakayama, A numerical study of vortex shedding
from flat plates with square leading and trailing edges, J. Fluid Mech. 236 (1992) 445-460