Journal of Wind Engineering

and Industrial Aerodynamics 77&78 (1998) 393—407

A numerical study of bluff body aerodynamics

in high Reynolds number flows
by viscous vortex element method
Fusen He*, Tsung-chow Su
Center for Applied Stochastics Research, Florida Atlantic University, Boca Raton, FL 33431, USA

Abstract

The viscous vortex element method was studied in the present paper, and some modifications
developed. Improvements were made to better represent the near-wall vorticity when obtaining
numerical solutions for the Navier—Stokes equations. In particular, we split the boundary
vortex sheet into two parts at each time step. One part remains a vortex sheet lying on the
boundary of the solid body, and the other enters into the flow field as a free vortex element with
a uniformly distributed vorticity. A set of kinematic relationships are used to determine the two
appropriate portions of the split, and the position of the free vortex element at the time of
release. Another improvement was the inclusion of the nonlinear acceleration terms in the flow
equations near the solid boundary when evaluating the surface pressure distribution. The
aerodynamic force coefficients were then obtained by summing up the pressure forces. By
comparing the computed surface vorticities, surface pressures, and aerodynamic force coeffi-
cients with existing numerical/experimental data in the cases of circular cylinders and aerofoils,
it was shown that the present approach is more accurate in modelling the flow features and
force coefficients without making different ad hoc assumptions for different geometries. The
algorithm is also applied to determine the aerodynamic forces for a bridge deck section. The
results are compared with experimental results obtained in the water channel laboratory at the
Center for Applied Stochastics Research at Florida Atlantic University.  1998 Elsevier
Science Ltd. All rights reserved.

1. Introduction

Vortex methods are grid-free, and therefore avoid a number of shortcomings of

Eulerian, grid-based numerical methods for solving high Reynolds number flow
problems. These include such problems as poor resolution and numerical diffusion. In
vortex methods, the continuum vorticity field is discretized into a collection of

* Corresponding author.

0167-6105/98/$ — see front matter  1998 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 6 7 - 6 1 0 5 ( 9 8 ) 0 0 1 5 9 - 7
394 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407

Lagrangian elements, known as vortex elements. These vortex elements are free to
move in the flow field which they create. The velocity field induced by these vortex
elements is a solution to the Navier—Stokes equation, and in principle the method is
suitable for high Reynolds number flows.
The study of bluff body aerodynamics has been carried out for over 100 years. Due
to its extreme complexity, our knowledge about the bluff body flow is still in the
empirical, descriptive realm [1]. Predicting the drag force correctly still constitutes
one of the basic problems in the study of the flow around bluff bodies. For a low
Reynolds number flow, the theories of Stokes and Oseen provided the theoretical
solutions for asymmetric and cylindrical bodies, respectively. Later mathematical
expansions extended their validity to a Reynolds number of up to 10. Beyond this
Reynolds number range, there is no analytical solution for a bluff body. Therefore,
numerical simulations and experimentation are the best ways to deal with the bluff
body aerodynamics problems.
Compared to the various available numerical methods, the vortex methods, by
virtue of their minimal susceptibility to numerical diffusion, without the need of
a fixed grid and far-field conditions, are well -suited for the numerical treatment of
high Reynolds number complex flows. Vortex methods rely on the discretization of
a continuous, time-dependent vorticity field into a large number of interacting vortex
“blobs”, whose position and strength determine the underlying velocity field. The
understanding of vortex dynamics is the key to the understanding of much of fluid
dynamics. And it is natural that vortex method deal directly with the interaction of
parcels of vorticity in any flow situation. Vortex methods have gained considerable
popularity in recent years and have been applied to a variety of problems.
The modern vortex method used to simulate viscous fluid flow was first introduced
by Chorin [2]. His method was based on the facts that the vorticity field in high
Reynolds number flow is compact and is confined only to the region near the solid
boundaries. Outside this region, the flow tends to resemble an inviscid flow. The
principle of this method is to use a number of Lagrangian discrete vortex elements
with finite core to simulate the evolution of a continuously distributed vorticity field.
In this method, the vorticity equation is divided into a convective and a diffusive part
and the two equations are solved sequentially. This kind of sequential rather than
simultaneous convection and diffusion of vorticity is known as the operator splitting
or fractional-step or time-splitting method. The basic algorithm is composed of three
parts: (1) the satisfaction of the no-slip boundary condition along solid boundaries, (2)
the inviscid convection by Biot—Savart law, and (3) the transport of vortex elements by
diffusion, which is usually simulated by a random walk algorithm [2].
To the best of our knowledge, the viscous vortex method often produces significant
errors on the surface vorticity distribution. In the viscous vortex method, the vorticity
creation was extended to the entire boundary by breaking the length of boundary into
intervals. Within each interval, a smoothed vortex element was created at each time
step. The strength of the vortex element was chosen to nullify the tangential velocity
and hence satisfy the no-slip boundary condition. The technique of adding a vortex
element at the boundary in response to the no-slip boundary condition and the release
of a newly created vortex element into flow field is not very accurate. According to the
 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407 395

explanation of Sethian [3], part of this inaccuracy occurs because a small motion
displacement in a newly created vortex element can cause significant variation in the
tangential velocity at the next time step. Thus, large counterrotating vortices must be
created at the next time step to enforce the no-slip condition. This can result in a large
number of vortices near the boundary with one sign which are essentially balanced by
a large number of vortices of opposite sign. The vorticity field was developed based on
the vorticity created on the body surface. Due to the significant error on the surface
vorticity distribution, it is difficult to obtain accurate surface pressure coefficients, and
therefore accurate force coefficients.
Based on a literature review and numerical experience, the major problem areas in
vortex method that need more attention are vorticity creation and diffusion from
body surface, force calculations, flow field resolution, and computer requirements and
computing time, etc. For example, Koumoutsakos and Leonard [4] used the par-
ticle-strength-exchange method and the Biot—Savart law together with fast summa-
tion algorithms to reduce the O(N) computational cost to O(N) for the direct
numerical simulations of flow past an impulsively started circular cylinder in the
range of Reynolds number from 40 to 9500. In their calculation, an efficient fast
summation algorithm was implemented that allows a large number of computational
elements, up to a million, thus producing unprecedented high-resolution simulations.
The simulation was made on CRAY YMP, and only the early stage (ºt/R)5) of the
flow fields was obtained. Therefore, this particle-strength-exchange method is a robust
and adaptive scheme for theoretical study of bluff body flows. But it is not yet
practical in engineering problems due to its huge amount of computation.
One aspect of this paper is to systematically study the viscous vortex method and
developing a vortex based algorithm that uses as few ad hoc numerical parameters as
possible. For example, in the investigation of vorticity diffusion from a solid body
surface, previous studies have been based on various ad hoc assumptions. However,
there exist some physical mechanisms to govern this phenomena. The mechanism of
viscous diffusion is a topic presently being studied. The surface pressure coefficient is
also being carefully studied and trying to obtain accurate force coefficients is one
purpose of the research. The purpose of the study is to develop an algorithm where the
computation time is not huge but which can still obtain satisfactory accuracy in
practical engineering problems. The other aspect of this paper is to demonstrate the
versatility of the developed vortex algorithm to solve practical problems in engineer-
ing applications. The present study is concentrated on the flow past bluff bodies. In
the present paper, the numerical study includes the flow fields and forces on circular
cylinders, aerofoils and bridge decks at high Reynolds number flows.

2. Numerical method

The viscous vortex element method was used in the present paper to simulate the
flow around bluff bodies in high Reynolds number flows. The step-by-step approach
of this method can be obtained from Ref. [5]. The following is a brief description
of the algorithm we used in the numerical simulation. The viscous vortex element
396 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407

method treats the continuous vorticity field by discretizing it into discrete vortex
elements. The following steps are employed in solving the Navier—Stokes equations:
(1) the generation of vorticities from the solid body surface, (2) vorticity convection,
and (3) vorticity diffusion. To obtain the distribution of surface vorticities, we impose
the no-slip and no-flux boundary conditions on the boundary on the complement of
the fluid side. Both surface vortex sheet element and source/sink element were used to
guarantee the satisfaction of zero velocity on the body surface. The detailed formula
one can refer to Ref. [5]. To calculate the inviscid vorticity convection, the Lagrangian
formulation for the particle trajectories was used. This is a typical approach in vortex
dynamics (for the detailed deduction, please refer to Ref. [3]). In the viscous vortex
method, the model of vorticity diffusion from the entire body surface is usually used.
Newly created discrete vortex sheets which satisfy the non-slip boundary conditions
will diffuse into the flow field with the elapse of time due to the viscous diffusion and
convection. And the flow field will thus be filled with vorticity which we represent by
a cloud of discrete vortices. In the present paper, an improved offset scheme [6] was
used to simulate the process of vorticity shedding from body surface. The viscous
diffusion of the vorticity in the flow field can be simulated by a random walk
scheme [2].
Three different approaches were used in the present paper to calculate the aerody-
namic forces on the bluff bodies. They are the integration approach, the Lewis
approach, and the improved Lewis approach. The integration approach [7] was
based on the information of the entire flow field. It is the most sophisticated and
accurate approach. It also needs more computation time. In the present study, the
integration approach is used as a benchmark to check the validity of simpler and more
efficient approaches.
The Lewis approach [8] is based on inviscid flow analysis. Although it can obtain
reasonable results, it does not reflect the influence of nonlinear acceleration terms. In
the present paper, an improved Lewis approach was proposed. In viscous fluid flow,
the Navier—Stokes equation is
*q 1
#(q )
)q"!
p#l
q, (1)
*t o
where q is the velocity vector, p is the pressure. At the body surface, the velocity q is
parallel to the body surface, the Navier—Stokes equation becomes
*u *u 1 *p
#u "! , (2)
*t *s o *s
where s is the tangential direction of the body surface, and n is the normal direction of
the body surface. In the vortex method, positive vorticity is defined as clockwise and
u"c. So the improved Lewis scheme becomes
1 * c *c
(p# )"! . (3)
o *s 2 *t
We follow a similar procedure to the Lewis approach [8].
 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407 397

3. Numerical results and discussion

The developed viscous vortex element algorithm is used to solve the flow past bluff
bodies in high Reynolds number flows. In the present paper, three different types of
body shape are used in the numerical simulation. They are a circular cylinder,
a NACA0012 aerofoil, and a simplified bridge deck section. The computation was
performed on a Dell 133 MHz Pentium Personal Computer at Center for Applied
Stochastics Research, Florida Atlantic University. In both cases, 50 surface vortex
elements and 50 source/sink elements were used in total. The non-dimensional time
step is 0.025, and it usually takes 7—10 days to finish one computation.

3.1. Viscous flow past a circular cylinder

The viscous flow around a circular cylinder is the first object we studied. Our study
is focused on the aerodynamic forces exerted on bluff bodies. For high Reynolds
number flows, the influence of viscosity on the forces exerted on the bluff body is
negligible compared to the forces due to the difference of surface pressure on the solid
body. The more accurate the pressure coefficient, the more accurate the aerodynamic
forces we can obtain. In the present study, we simulated the following different
Reynolds number flows, Re "10, 1.1;10, 6.7;10, and 8.4;10.
Tables 1—4 compare the force coefficients obtained by different approaches with
different Reynolds numbers. The experimental value of drag force at Re "10 is 1.0.
When the Reynolds number is in the range from 10 to 5;10, the drag force
coefficient varies in the range of 1.1—1.2. We can see that the integration method gives
better results at Re"10, but gives poor results at other Reynolds numbers. Both the
Lewis approach and the improved Lewis approach yield better results in these cases.

Table 1
Force coefficient at Re"1.0;10

Average C Average C RMS C RMS C

   
Integration 1.01 !0.12 1.02 0.71
Lewis approach 1.37 !0.16 1.39 0.98
Improved Lewis approach 1.39 !0.24 1.40 0.96

Table 2
Force coefficient at Re"1.1;10

Average C Average C RMS C RMS C

   
Integration 0.91 !0.11 0.93 0.57
Lewis approach 1.20 !0.15 1.22 0.77
Improved Lewis approach 1.19 !0.15 1.21 0.76
398 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407

Table 3
Force coefficient at Re"6.7;10

Average C Average C RMS C RMS C

   
Integration 0.85 0.04 0.90 0.61
Lewis approach 1.10 0.05 1.15 0.80
Improved Lewis approach 1.09 0.05 1.15 0.79

Table 4
Force coefficient at Re"8.4;10

Average C Average C RMS C RMS C

   
Integration 0.85 !0.004 0.87 0.54
Lewis approach 1.13 !0.007 1.34 0.70
Improved Lewis approach 1.14 !0.009 1.34 0.72

To obtain better results with integration approach, we need to add many

more discrete vortex elements in the flow field, this will rapidly increase the computa-
tion time and sometimes we even need a much more powerful supercomputer to
handle it. Although it is simple, the improved Lewis approach can obtain satisfactory
results.
The time-averaged surface pressure coefficient distribution is shown in Fig. 1,
which compares the results obtained by the integration approach, by the Lewis
approach, and by Huyer et al. [7] for the case of Re"10. The comparison shows
that we obtained the same accuracy as that of Huyer et al. Fig. 2 is the comparison of
surface pressure coefficients of flow past a circular cylinder at Re"1.1;10 by the
integration approach and the improved Lewis approach with experimental results.
For Re"1.1;10, pressure minima appear at approximately 72° and 288°. The
experimental data presented in Batchelor [9] show that minimum pressures appears
at approximately 70° and 290°. The time-averaged base pressure is !1.23, and the
pressure minimum is !1.4. From the numerical simulation, we see that the pressure
minimum at 72° is !1.5, and !1.61 at 288°. The averaged base pressure is !1.13.
Base pressure is an very important parameter for determining the drag and flow
patterns. In most of the previous studies using the vortex method the base pressure is
almost twice the value of the experiments [1]. The improved Lewis approach achieves
a better accuracy.

3.2. Viscous flow past a NACA0012 aerofoil

Another case in the present study is the viscous flow around a NACA0012 aerofoil.
The flow past the aerofoil with zero angle of attack at Re"3;10 and the result with
12° angle of attack at Re"3;10 are presented in this paper. The reason we choose
 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407 399

Fig. 1. Comparison of the time-averaged surface pressure coefficient of flow past a circular cylinder at
Re"10.

Fig. 2. Comparison of the time-averaged surface pressure coefficient of flow past a circular cylinder at
Re"1.1;10.

these parameters is because we can compare them with the result obtained by Lewis
[8]. In Lewis’ numerical result, a poor accuracy at 12° angle of attack occurred.
First, we studied the flow past a NACA0012 aerofoil at zero angle of attack at
a Reynolds number of 3;10. Fig. 3 is the comparison of the surface pressure
coefficient obtained by the integration approach, the improved Lewis approach, the
Lewis approach, and experiment. Fig. 4 is the drag coefficient obtained by the
integration approach and the improved Lewis approach. Fig. 5 is the comparison of
the lift coefficient obtained by the integration approach and the improved Lewis
approach. We can see that the integration approach has a much smaller fluctuation of
the force coefficients because it used the whole field information and has a higher
400 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407

Fig. 3. Time averaged surface pressure coefficient over a NACA0012 aerofoil at 0° angle of attack and
Re"3;10.

Fig. 4. Comparison of drag coefficient on a NACA0012 aerofoil at 0° angle of attack and Re"3;10.

degree of accuracy. As to the average value of drag and lift coefficients, both
approaches achieve the same numerical accuracy.
Lewis [8] used the offset scheme to successfully simulate the uniform flow past
a NACA0012 aerofoil at high Reynolds numbers. But the simulation of lift force is less
satisfactory when this aerofoil is at a 5° angle of attack, and the lift force drops rapidly
when the non-dimensional simulation time is in excess of 2. In the present paper, the
developed algorithm was used to simulate the flow past a NACA0012 aerofoil with
12° angle of attack at a Reynolds number of 3;10. We can still obtain a satisfactory
result. Fig. 6 shows the time-averaged surface pressure coefficient on the aerofoil by
the improved Lewis approach. The result is similar to that by Grant and Huyer [10].
 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407 401

Fig. 5. Comparison of lift coefficient on a NACA0012 aerofoil at 0° angle of attack and Re"3;10.

Fig. 6. Time averaged surface pressure coefficient on a NACA0012 aerofoil at 12° angle of attack and
Re"3;10.

The average lift coefficient is 0.96. Fig. 7 gives a picture of the vorticity structure of
flow past a NACA0012 aerofoil at 12° angle of attack.

3.3. Viscous flow past a bridge deck

The investigation of the aerodynamic forces acting on a vibrating structure is an

important subject in wind engineering. The character of unsteady forces influences
the response of the structure to the wind. The unsteady and three-dimensional flow in
the separating region and the interaction between the vibrating body and the un-
steady flow makes the flow field extremely complicated. Both theoretical study and
experimental measurement face great difficulties. Experimental investigations on the
402 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407

vibration of a bridge model have often been carried out on a two-dimensional deck
model with restricted degrees of freedom. In the present paper, the aerodynamic forces
of a two-dimensional bridge deck model in the forced torsional vibration condition
was studied by experimentation and numerical simulation. The experiment was
conducted in a water channel at Florida Atlantic University. The pitching excitation
at a frequency of 1.2 Hz and an amplitude of 1.8° is nearly sinusoidal. The test section
of the water channel of the Center for Applied Stochastics Research is 0.6 m wide,
0.4 m deep and 8 m long. A two-dimensional model was used in the experiment. The
model is 13 cm in chord length, 1.75 cm thick, with a cross-section of a bridge deck,
and is 56 cm in length spanwise.

Fig. 7. Vorticity distribution of flow past a NACA0012 aerofoil at t"10 with 12° angle of attack and
Re"3;10.

Fig. 8. Force coefficient on a bridge deck by the improved Lewis approach at 0° angle of attack and
Re"1.8;10.
 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407 403

The numerical simulation was performed on cases of flow past the bridge deck with
a Reynolds number of 1.8;10. Figs. 8 and 9 are the force coefficients obtained by the
improved Lewis approach and the integration approach. The average drag coefficient
in the improved Lewis approach is 0.061, and 0.059 in the integration approach. In the
integration approach, the information on the entire flow field is used to calculate
the force coefficient, the fluctuation of the force coefficient is much smaller than that in
the improved Lewis approach. Since only the time-averaged quantities are suitable for
analysis in the viscous vortex method based on the random walk viscous diffusion
scheme, the improved Lewis approach is more efficient than the integration approach.
Fig. 10 is the comparison of the surface pressure coefficient on the bridge deck

Fig. 9. Force coefficient on a bridge deck by the integration approach at 0° angle of attack and
Re"1.8;10.

Fig. 10. Pressure coefficient on a bridge deck by different approaches at 0° angle of attack and
Re"1.8;10.
404 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407

Fig. 11. Force coefficient on a bridge deck at 10° angle of attack and Re"1.8;10.

Fig. 12. Time averaged surface presure coefficient on a bridge deck at 10° angle of attack and
Re"1.8;10.

obtained by different approaches. It shows that both approaches achieve the same
accuracy in the surface pressure coefficient.
The viscous flow past the bridge deck at 10° angle of attack with the same Reynolds
number was also studied. Fig. 11 shows the drag and lift coefficients exerted on the
bridge deck. The average drag coefficient is 0.30 and the average lift coefficient is 1.44.
Fig. 12 is the time-averaged surface pressure coefficient on the bridge deck.
The uniform flow past an oscillating bridge deck with 1.8° amplitude oscillation was
simulated and compared with the experimental results. Fig. 13 is the spectrum
analysis of the lift coefficient on the bridge deck, we see that there exists a dominant
frequency at 6.2 Hz, which is same as the experiment. Fig. 14 shows the vortex
structure in the adjacent area of the bridge deck.
 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407 405

Fig. 13. The power spectrum of the lift force coefficient on a bridge deck with 1.8° amplitude oscillation and
Re"1.8;10.

Fig. 14. Vorticity distribution of flow past a bridge deck with 1.8° amplitude oscillation and Re"1.8;10.

Fig. 15 is the lift coefficient on the bridge deck at 10° angle of attack and 5° angular
oscillation. It shows the relation of the angle of attack with lift coefficient. Fig. 16
shows the same result as a time history. The unsteady fluid dynamic force on the
model depends upon both the instantaneous flow field and the changing rate of the
flow velocity field around the model. Generally, the generation and shedding off of the
vortices, and the changes of the separation region, have a dominant influence on the
flow field. While the angle of attack is increasing, the vortex over the upper surface of
the model is almost stationary. While the angle of attack is decreasing, the vortices
406 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407

Fig. 15. The lift coefficient on a bridge deck at 10° angle of attack and 5° amplitude oscillation versus the
angle of attack at Re"1.8;10.

Fig. 16. The time history of the lift coefficient on a bridge deck at 10° angle of attack and 5° amplitude
oscillation and Re"1.8;10.

over the upper surface are moving much faster toward the trailing edge. The change of
the position of the vortices could be readily observed in the experiments. From the
generalized Blasius theorem for the fluid dynamic forces acting on a two-dimensional
body in an unsteady flow, the changing rate of the complex potential corresponds to
the generation of fluid force on the body. Thus moving vortices would produce
additional force, which might increase the lift while the angle of attack is decreasing.
This might be the cause of the generation of negative slope of the lift—angle of attack
(c —a) curve.

 F. He, T. Su/J. Wind Eng. Ind. Aerodyn. 77&78 (1998) 393–407 407

4. Conclusion

This paper presented an improved algorithm for the viscous vortex element
method. The present approach leads to a more accurate modelling of surface vortex
sheet strength distribution and other flow characteristics. The computation time of
the approach is moderate. It has been used to solve the viscous flow past bluff bodies,
such as circular cylinder, aerofoil, and bridge deck, in high Reynolds number flows
and given satisfactory results. It can be useful in the study of the unsteady phenom-
enon in practical engineering problems.

References
[1] A. Roshko, Perspective on bluff body aerodynamics, J. Wind Eng. Ind. Aerodyn. 49 (1993) 79—100.
[2] A.J. Chorin, Numerical study of slightly viscous flow, J. Fluid Mech. 57 (1973) 785—796.
[3] J.A. Sethian, A brief overview of vortex methods, in: Vortex methods and vortex motion, K.E.
Gustafson and J.A. Sethian (Eds.), SIAM (1991) pp. 1—32.
[4] P. Koumoutsakos, A. Leonard, High-resolution simulations of the flow around an impulsively started
cylinder using vortex methods, J. Fluid Mech. 296 (1995) 1—28.
[5] F. He, Numerical study of bluff body aerodynamics by vortex methods, Ph.D. Dissertation, Florida
Atlantic University, 1998.
[6] F. He, T.C. Su, On the numerical treatment of vorticity diffusion from a boundary element in the
discrete vortex element method, ASCE 11th Engineering Mechanics Conf., Fort Lauderdale, FL,
vol. 2, 1996, pp. 844—847.
[7] S.A. Huyer, J.R. Grant, J.S. Uhlman, A vortex element representation of two-dimensional unsteady
separated flow fields, AIAA paper 94-0075, 1994.
[8] R.I. Lewis, Vortex Element Methods for Fluid Dynamic Analysis of Engineering Systems, Cambridge
University Press, Cambridge, 1991.
[9] G.K. Batchelor, An introduction to Fluid Dynamics, Cambridge University Press, Cambridge, 1967,
p. 341.
[10] J.R. Grant, S.A. Huyer, Development of lagrangian vorticity methods for calculating unsteady flows,
NUWC Division Newport Technical Digest, August, 1996, pp. 45—59.