You are on page 1of 11

ENAE 488C Computational Fluid Dynamics

Study of Air Flow over Two Side-By-Side


2-D Circular Cylinders
Mario Nirman Mondal
UID# 113269066
Department of Aerospace Engineering
University of Maryland, College Park, MD

Final Project, Spring 2015

Page 1

Study of Air Flow over Two Side-By-Side


2-D Circular Cylinders
Mario Nirman Mondal
UID# 113269066

Abstract
In this project, the air flow over two identical cylinders, which were placed side-by-side,
has been studied. The cylinders were represented in two-dimension by a circle, and flow domain
was created and discretized using the commercial grid generation software, Pointwise. The flow
was modeled as steady, viscous, turbulent and compressible. CFD++ was used as a CFD solver
to get the solution of the flow field. After solving the flow field, the qualitative analysis of
streamlines was done using Tecplot. The distance ratios between two cylinders, L/D (L = center
to center distance, D = Diameter of the single cylinder) were varied to study the flow behavior at
different distances. The variation of coefficient of lift (Cl) and the coefficient of drag (Cd) of the
two cylinders was examined. Lift-coefficient, Cl decreases for upper cylinder as the distance
increases. Conversely, Cl increases for the lower cylinder as the lower cylinder as the distance
increases. At some point, Cl of both the cylinders converges to same value, which means they
start to act like single cylinder and become free of the effect of the flow from the other cylinder.
The Reynolds number (Re) was also varied in each distant case. At different Reynolds numbers,
the flow pattern, the wakes behind the cylinders, vorticity contours and flow separation was
analyzed. The variation of coefficient of drag (Cd) as a function of Reynolds number was
examined, and it was plot with the analytical value of Cd to compare. As expected, Cd continues
to keep decreasing as Re increases. However, at critical Re=3 X 105, a drop in Cd is observed,
and it again increases afterwards. Unsteady flow behavior at low Reynolds number was also
simulated to get the idea of how the unsteady simulation works and the effect of unsteady
compared to steady flow was investigated. Finally, the grid resolution study was done in order to
observe the effectiveness of the baseline (1X) grid generation with the 2X and 4X grid points.

I. Introduction
Flow over cylinder is a fundamental fluid dynamics problem of practical importance.
Double circular side-by-side cylindrical configurations are widely used in engineering
applications e.g. the flow around bi-planes, twin-jet engine aircraft, wind turbine farms, electrical
poles, chimneystacks etc. The goal of the project is to investigate the flow behavior around two
side-by-side cylinders by varying the distances between two cylinders and also varying the
Reynolds number. Actually, the distance ratio L/D was varied, where L = center to center
distance, D = Diameter of the single cylinder/Characteristics length/Chord length. Because the
study was done from very low Mach (M=0.000043) region to subsonic region (M=0.43), the far

Page 2

field of the flow was placed 50 times far off the


characteristics length in each direction so that the
disturbance cannot migrate from upstream or downstream
of the flow.
The flow was modeled as steady, viscous,
turbulent and compressible flow. The flow field was
discretized and the grid was generated using grid
generation software, Pointwise. CFD++ was used as a
CFD solver to get the solution of the flow field. After
solving flow field, the qualitative and quantitative analysis
was done. The qualitative analysis was done by getting the
streamlines of flow using Tecplot, and the quantitative
analysis and plots were made using Microsoft Excel.

Figure 1: Basic Geometry

II. Theory and Equations


he relations between the coefficients and the forces are as follows:
;

Where,

Cl = Coefficient of Lift
Cd = Coeffiient of Drag
L = Lift force
D = Drag force
q = Dynamic Pressure
= Density of the fluid (in this case, Air)

S = Surface area of the body = Chord length (C) X Span (b)


Since a circle is being modeled, surface area S=
, r = radius of the circle
Also, the equation that relates Reynolds number to free stream velocity is given below:

Reynolds number,
Where,

d = characteristics length = diameter of the cylinder = 0.1m


= viscosity of air = 1.7894 x 10-5 kg/(m.s)
= Density of the fluid (in this case, Air) = 1.225 kg/m3

In order to vary the Reynolds number, the free stream velocity, V was varied.

Page 3

III. Methods
1. Grid Generation:

The geometry was created in Pointwise: A circle of diameter 0.1m (which was mirrored
later to get the symmetrical flow domain of another 2-D cylinder).
For the baseline (1X) grid, 90 grid points were imposed on the circle (4 per point).
By default, the grid point distribution on the circle was d=0.00349 and it was kept as
the spacing distribution of the points on the connecting lines.
Wall spacing of s=0.0001 was chosen for all the grids.
Farfield boundary was placed approximately 50 chord lengths/diameter away from the
cylinders in all directions.
Structured O-grid was generated around one cylinder up to 5*D distance so that the
boundary layer effects can be captured as good enough as possible.
Points were clustered around the cylinder in order to resolve boundary layer.
After creating the O-grid, the whole domain was divided into 4 rectangles and H-grid was
generated around one cylinder up to farfield.
Then, the whole domain was mirrored about X-axis to get the flow grid domain of the
other cylinder.

Figure 2: Grid (Zoomed in) for L/D=3

Three different grids were generated to vary L/D (distance ratio of the two cylinders).
1.) L/D = 1.5
: Number of cells 52,080
2.) L/D = 3
: Number of cells 55,500
3.) L/D = 5
: Number of cells 59,040
In order to do the grid resolution study, for L/D=3 two more grids were made as follows:
1.) Baseline (1X) grid points : Number of cells 55,500
2.) 2X grid points
: Number of cells 206,276
3.) 3X grid points
: Number of cells 793,512

Page 4

2. CFD++ Solver Setup:

The Wizards was used to set the equation type, initialize the domain with given
quantities, set up turbulence model with k - , set up fluid properties of air and set up time
integration quantities.
Compressible perfect gas equation was used.
The solver was run at steady-state condition.
The dimension and units were used in S.I system.
Freestream X-velocity was varied to match with the desired Reynolds number according
to the equation of the Reynolds number.
Re = 100
X-velocity = 0.0146 m/s
Re = 200
X-velocity = 0.0292 m/s
Re = 10000
X-velocity = 1.4607 m/s
Re = 100000
X-velocity = 14.6073 m/s
Re = 1e+6
X-velocity = 146.0735 m/s
Initial condition of the domain was set at sea level condition (Altitude 0 Km).
Pressure
= 101325 Pa
Temperature = 288 Kelvin
Freestream turbulence intensity was set at 3%.
Simple turbulence model (2-equation model) was used.
Turbulent/laminar viscosity ratio was used as 50.0.
In all cases, turbulence was modeled from the first iteration.
Boundary conditions were set adiabetic viscous wall function for airfoil wall, and
characteristics-based inflow/outflow for the farfield.
Spatial order of accuracy 2nd order accurate
Simulations were run at different X-velocity of each of the L/D case. Therefore, 15
steady simulations were run.
Convergence criterion was fulfilled by residual drops of 6.0 orders of magnitude.
There was one unsteady flow was run for L/D=3 and Re=200. For the unsteady flow
parameters was selected as below:
Strouhal number was assumed to be 0.2
Vortex shedding frequency was calculated using the Strouhal equation:
Strouhal no, St = (f*D)/u
f = vortex shedding frequency; D=cylinder diameter; u = X-velocity
From the frequency, f, time period, t s was calculated.
1000 time steps were wanted to resolve the time period, t s.
Therefore, time step size Dt was calculated as Dt = (t s)/1000
In order to run the unsteady simulation, up to 300 iterations were run as steady-state
so that the solution converges. Then, upto 1500 iterations were made to get the
unsteady solution.

Page 5

3. Flow visualization in Tecplot, Data collection and Plotting in Microsoft Excel:

After CFD++ ran the simulations, residual plots and the flow visualization plots were
observed and analyzed.
Streamlines of the flow were plotted in Tecplot.
Data was recorded for approximately last 50 iterations and average was calculated.
Entries for lift and drag coefficients were created in Microsoft Excel.
Plots of Cl vs. Re; Cl vs. L/D; Cd vs. L/D; Cd vs. Re, and Cl vs. Re as a function of L/D
were generated and compared.
For L/D=5 and L/D=3, Cd vs. Re was compared to the experimental value. Experimental
values were taken from Fundamentals of Aerodynamics by John D. Anderson Jr.

IV. Results and Discussion


1. Coefficient of lift (Cl) and Coefficient of drag (Cd) vs. Distance ratio (L/D):
0.04

Cl vs L/D

0.03

Coeff of Lift, Cl

0.02
0.01

0
-0.01
-0.02
-0.03

Cl of Lower Cylinder

-0.04

Cl of Upper Cylider

-0.05
0

3
L/D

Figure 4: Cl vs L/D

Cd vs. L/D

0.06

0.05
Coeff of Drag, Cd

When the cylinders were close (L/D=1.5),


flow past two side-by-side cylinders
interact with each other largely resulting
higher (+) ve lift on the upper cylinder
and higher (-)ve lift on the lower cylinder.
As the distance between the cylinders was
increased, the coefficient of lift (Cl) starts
to decrease because flow interactions
become less. As the flow interaction
decreases, the two cylinders starts to
behave as they were single cylinder.
On the other hand, coefficient of drag (Cd)
behaves differently. When the cylinder
were really close (L/D=1.5), the drag was
lower than when they were at L/D=3.
Also, when we increase the distance, as
we increase the L/D ratio, the drag starts
go down to the point as they were alone.
The reason behind that is when the
cylinders are really close because of the
flow interaction from both the cylinders
the force in the X-direction is higher, also
the wakes created past the cylinders
interact with each other causing lower
drag than at L/D=3.

0.04
0.03
0.02
Cd of Lower Cylinder
0.01

Cd of Upper Cylinder

0
0

3
L/D

Figure 3: Cd vs L/D

Page 6

2. Coefficient of lift (Cl) and Coefficient of drag (Cd) vs. Distance ratio (L/D):
At low Reynolds number (Re) somewhat higher lift-coefficient (Cl) was observed than at higher.
However, at very high Re higher Cl is observed. As from figure 5, the result is clear enough.
5.00E-02

Cl vs Reynolds No.

4.00E-02
3.00E-02
2.00E-02
1.00E-02

Cl_Up_Cyl_L/D=1.5
Cl_Up_Cyl_L/D=3
Cl_Low_Cyl_L/D=3
Cl_Low_Cyl_L/D=1.5

0.00E+00
-1.00E-02
-2.00E-02
-3.00E-02
-4.00E-02
-5.00E-02
1.00E+00

1.00E+01

1.00E+02

1.00E+03
1.00E+04
Reynolds Number

1.00E+05

1.00E+06

1.00E+07

Figure 5

Cd

At low Reynolds number, the flow is completely laminar. As the Re increases, turbulence starts
to affect and causes flow separation, which, in turn, decreases lift. However, at high Reynolds
number, the flow velocity becomes high enough to ignore the viscous effect and the pressure
gradient becomes high resulting high lift.
After examining the Cd vs. Reynolds
Cd vs. Reynolds No. for L/D=3
0.14
number plot, it can be seen that the
Cd Lower_Cyl
Reynolds number drops at Re=200 and
0.12
Cd Upper_Cyl
increases and stays close up to 100000, and
again drop after 1e+5 and starts increase. As
0.1
we expected, from 100<Re<200 flows
0.08
transition happens from laminar to
turbulent. For this reason drag drop around
0.06
at Re=200. Re=3 x 105 is called critical
Reynolds number where a significant drag
0.04
drop happens. At the flow separation point
0.02
of the cylinder transition to turbulent flow
takes place about 120 around the body.
0
This transition to turbulent flow and
1
10
100
1000
10000 100000 1000000
corresponding thinner wake reduces
Reynolds Number
Figure 6

Page 7

pressure drag on the cylinder and is responsible for the precipitous drop in Cd at Re = 3 x 105.
The experimental result is compared with the computation value in figure 7.

Cd Vs. Reylonds No. Comparison


100

10

Cd vs Re : Experimental

Cd

Cd_Up_Cyl_L/D=5
Cd_Up_Cyl_L/D=3

0.1

0.01

0.1

10

100

1000

10000

100000 1000000 10000000

Reynolds No
Figure 7

Our CFD result predicted drag less than the experimental value. There might be several reasons:
1) Experimental data was for single cylinder, but the CFD results were done for side-by-side
cylinders. Though the cylinders were far, they were not free from flow interactions that
cause underestimation of coefficient of drag (Cd).
2) Boundary layer might need to be resolved more to achieve the solution accuracy.
3) Experimental results are for 3-D cylinder whereas the CFD result is for 2-D circle.
3. Vorticity study:
Vorticity is measure of the rationality of the flow. The more the flow is rotational, the higher the
vorticity. From figure 8, it can be easily understood that the vorticity doesnt change much in
magnitude as the distance ratio (L/D) varies. However, vorticity clearly depends on Reynolds
number. At low Reynolds number, vortices form at the front face of the cylinders, and they die
away after passing the cylinder. However, at high Reynolds number, vorticity is really high at the
back face of the cylinders. Actually, at high Reynolds number flow separation occurs at 120 of
the body and the flow becomes highly turbulent. Because of turbulence in flow and the flow
separation at the body, vorticity is really higher than order of magnitude 4 at the flow separation
points. Vortices continue to move throughout the turbulent flow, but their strength decrease in
magnitude as they go far from the body. As shown in figure 8, when the cylinders are close (L/D
= 1.5) vortices are little affected by the vortices created by the other cylinder. When they are far
(L/D=3) vortices are unaffected.

Page 8

Re= 200 , L/D = 1.5

Re = 1e+6, L/D = 1.5

Re = 200, L/D = 3

Re = 1e+6, L/D = 3

Figure 8: Vorticity at different Re and L/D

4. Study of wake as L/D and Reynolds number varies:


The streamlines of the flow were plot using Tecplot, the wakes
of the flow were investigated qualitatively. When the L/D is
low at 1.5, the pattern of wakes were observed as expected.
The gap flow was biased to one side, resulting in the formation
of a narrower wake behind one cylinder and wider wake
behind the other. Steady and fluctuating fluid forces acting on
the cylinders were decomposed for the narrower wake and the
wider wake flow patterns. At higher distance, L/D=5, as
expected, two-street of vortex pattern are formed mostly
unaffected by each other (As shown in the picture next page).

Re = 200, L/D = 1.5

At lower Reynolds number, large wake patterns are formed.


Conversely, at higher Reynolds number, small wake patterns
are formed. Wakes are the phenomena of laminar flow. At high
Reynolds number, flow becomes turbulent soon enough it
leaves the body. As a result, a very small wake patterns are
seen at high Reynolds number flow.

Re = 100000, L/D = 1.5

Page 9

Re = 100000, L/D = 5

Re = 200, L/D = 5

5. Failure of unsteady simulation:


At very low Reynolds number region, the flow is expected to be highly unsteady. In order to see
the unsteady effects and the Von Karman vortex street was run. However, the unsteady
simulation result wasnt much different than the steady state simulation result. More experiment
and knowledge was needed to resolve the Strouhal number, vortex shedding frequency and the
time step size to get the visible unsteady solution.
6. Grid resolution study:
Grid resolution study was done at Re = 100000 and L/D=3 by increasing the number of points by
2X and 4X. The simulations were run and compared to the baseline data. The results are below:
Gridpoints
Baseline (1X)
2X
4X

No of cells
55,500
206,276
793,512

Percent difference in Cl =

Low_Cyl
Cl
-3.50E-03
-4.69E-03
-4.59E-03

Up_Cyl
Cl
3.50E-03
4.69E-03
4.59E-03

Low_Cyl
Cd
5.30E-02
5.55E-02
5.51E-02

Up_Cyl
Cd
5.30E-02
5.55E-02
5.51E-02

= 25.37%

Both the 2X and 4X yield the similar result. There was issue with wall spacing for the baseline
L/D=3. Most probably, grid solving wasnt good enough to capture the boundary layer effect. If
we would have worked with 2X as baseline, the result might have been better. Because the
project timeline was reached, there wasnt enough time to go back and redo the simulations.
Lack of grid refinement actually affected the results overall.

Page 10

V. Conclusion
The goal of the project was to investigate the flow behavior around two side-by-side
cylinders by varying the distances between two cylinders and also varying the Reynolds number.
The goal has been achieved because a wide range of study of lift and drag coefficients, vorticity,
streamlines, wakes, flow pattern as a variation of the distance and Reynolds number. It was
found that how the coefficient of lift and drag are affected by the Reynolds number. Different
wake patterns at different distances of the cylinders were investigated. Grid generation and flow
simulation techniques were learned and practiced. Though the grid resolution wasnt accurate,
the solution was converged so quickly (see figure). It can be said to conclude that the farther the
cylinders and the higher the Reynolds numbers, the better the results.

Residual plot at Re = 100000 and L/D = 1.5

References
Anderson, Jr., John D. Fundamentals of Aerodynamics. Fifth Edition. New York: McGraw-Hill,
2011. Print.
Alam, Md. Mahbub, M. Moriya, and H.Sakamoto. Aerodynamic characteristics of two side-byside circular cylinders and application of wavelet analysis on the switching phenomenon.
Journal of Fluids and Structures 18.3-4 (2003): 325-346. ScienceDirect. Web. 20 May. 2015.
< http://www.sciencedirect.com/science/article/pii/S0889974603001075>
Chen, Bo, and Wan-Ping Li. Near-wake flow characteristics of two side-by-side circular
cylinders close to a wall. Acta Mechanica 222.3-4 (2011): 295-307. Springer International
Publishing AG. Web. 20 May. 2015. <http://link.springer.com/article/10.1007%2Fs00707011-0538-3>
Shao, J., and C. Zhang. Numerical Studies of Flow Past Two Side-by-Side Circular Cylinders.
New Trends in Fluid Mechanics Research (2009): 73. Springer International Publishing AG.
Web. 20 May. 2015. < http://link.springer.com/chapter/10.1007%2F978-3-540-75995-9_14>

You might also like