Introduction

In this study, a computational fluid dynamics (CFD) analysis was done in StarCCM to evaluate the perfor-
mance of the MA409 airfoil at Reynolds numbers of 41,000 through 10,000 and angles of attack spanning -8
degrees to 16 degrees. The results of the CFD analysis at a Reynolds number of 41,000 were validated using
empirical data, ensuring that the simulation of the MA409 airfoil was accurate. Then a mesh refinement
study was done to ensure that the data was mesh size independent. After this, the drag and lift coefficients
were calculated and plotted as polar plots and angles of attack sweeps, providing concrete performance met-
rics. The low-speed simulation provides insights into the airfoils’ behavior and potential applications.

Michael Achterberg developed the MA409 airfoil in the 1990s for use in F1C class free flight model aircraft
[1], [2] and it has demonstrated good climbing performance and glide endurance in numerous flyoffs. Its high
lift and low drag at low Reynolds numbers makes it an excellent candidate for low-speed gliders and wind
turbines. In addition its low drag at zero-lift improves its climb performance in the F1C class free flight
[1]. The results of the CFD analysis of the MA409 airfoil provide valuable information for designing and
optimizing systems that use this airfoil.

Figure 1: Plot of the MA409 airfoil from [3]

Simulation Setup
The simulation was set up by using StarCCM’s internal CAD tools to create a domain and the wing as two
separate pieces. The domain was setup to be 6m tall and 13m long to minimize boundary effects on the
wing. The wing was normalized to have a chord of 1m and then the tip of it was centered 2m into the
domain. This left 10m of wake room behind the wing. The wing was set up to rotate about the tip as
angle of attack changed. Then the subtract tool was used to generate the final object (See Figure 2). The
computational models selected were: Constant Density, Gas (Air), Two Dimensional, Laminar, and PISO
Unsteady. These were selected due to the nature of the problem and the instructions.

On this object three regions were created. The front, top, and bottom were selected as inlets with velocity
vectors pointing to the right (X-direction). The velocity was chosen to achieve the Reynolds values tested.
The back was selected as a pressure outlet. Finally the wing was selected as a no-slip boundary. See Figure
2 for an annotated image. All cells were initialized with the same velocity as the inlet so that the flow would
develop faster.

The simulation was set up with the base element size set to the boundary layer thickness at 1/10th of
the chord length (1). Table 1 summarizes the velocities and boundary layer thicknesses for each Reynolds
number.
r
νc
δ= (1)
10U∞

1
Figure 2: The three different regions of the simulation. The wing is positioned at 0 degrees angle of attack.

Reynolds number Input Velocity Boundary Layer Thickness Time Step

41, 000 0.6175m/s 0.00156m 0.00015s
20, 000 0.3012m/s 0.00224m 0.0003s
10, 000 0.1506m/s 0.00316m 0.001s

Table 1: Reynolds numbers and the corresponding velocities, boundary layer thicknesses, and time steps.

The mesh was then further refined by decreasing the surface growth rate to 1.1, minimum surface size to
10% of base size, and two surface controls were added. The surface growth rate and surface size were set as
such because, at high angles of attack, there was full separation which developed faster, and what appeared
to be more accurately, if the mesh was denser. The first surface control was to set the target and minimum
surface size to 4000% of base size for the inlet and outlets. This ensured that those areas were not meshed
to the same density as the wing which was unnecessary. Finally, the wing had a wake refinement added to
it, extending till the end of the region with a slight conical angle of 0.01rad and with the size set to 500% of
base size. This was so that the entire wake could be captured at decently high resolution without sacrificing
much simulation performance. All of these refinements improved the speed at which the flow converged while
also improving how well the simulations matched empirical data.

The time step was adjusted until the maximum local CFL number was approximately 0.3, and all simulations
at the same Reynolds number were run with the same time step. A list of the chosen time steps are in table
1. Each simulation was run until the simulation time reached 150% of the boundary length divided by the
velocity (2). This ensured that all initial conditions had been pushed out. Then the Cd and Cl vs iterations
plots were inspected and, if they appeared to not have reach a steadily oscillating value, were run longer.

13m
Tmin = 1.5 × (2)
U∞
Cl and Cd were calculated as follows:
Fy Fx
Cl = 1 2
, Cd = 1 2
(3)
2 ρU∞c 2 ρU∞c

Where Fy and Fx were the forces on the wing region in the domain in the y and x direction respectively. It
is important to note that, because I rotated the wing, these forces were always pointing the right direction.

2
ρ was the reference density that StarCCM was using for air, U∞ was the free air speed, and c = 1m was the
chord length of the airfoil.
As Cd and Cl oscillated due to the unsteady nature of the problem, the average over the previous 40, 000
iterations was taken. If the values were incredibly unsteady, 100, 000 iterations were used. It is important
to note that these iterations were all after the initial conditions had passed out of the domain.

Verification at Re = 41, 000 using empirical data

First, the simulation was run at Re = 41, 000 as [1] had empirical data for that case. The results are as
follows:

Figure 3: Plots of Cl and Cd against angle of attack from both CFD and Empirical data at Re = 41, 000.

As we can see from figure 3, the CFD results match the angle of attack pretty well for Cl . However, it should
start dropping at higher angles of attack but does not. A point of particular note is the 16 degrees angle
of attack, it should have less lift than the 12 degree case. This means that there is an issue in the CFD at
high angles of attack. Some possibilities include not high enough resolution in the separation region and too
small of a wake. See figure 4 and 5 that showcase the extent of the separation. Unfortunately, due to the
increased computational complexity associated with increasing the cell density in those areas this was not
explored further. It should be noted that the Cl is likely invalid at high angles of attack but does appear to
be valid from −10 degrees to 10 degrees.

The Cd matches up worse with consistently higher drag across the angle of attack sweep. Once again, it is
possible that a finer resolution mesh that could capture the separation better would improve the Cd . As the
Cd is higher in general, a finer mesh should be placed at the tip and tail of the wing to possibly improve it.
Once again however, the increased computational complexity prevented me from doing so.

Plots of the velocity and pressure provide some insight into the odd behavior at high angles of attack:

3
 Figure 4: Velocity magnitude map for Re = 41, 000 and 16 deg angle of attack

Figure 5: Pressure magnitude map for Re = 41, 000 and 16 deg angle of attack

The 16 deg angle of attack case was run for 397,937 iterations resulting in 59.69s of simulation. This should
be plenty to achieve a steady state and, as can be see in figure 13 in the appendix. It had settling into steady
oscillations so it is unlikely the error is due to the time integration an much more likely that it is due to the
mesh size and simulation setup or model.

4
Mesh Refinement Study
After validating the Re = 41, 000 case against empirical data, a mesh refinement study was done at 0 degrees
angle of attack to ensure that the solution was not mesh dependent. To do this, the base size that was set as
the boundary layer thickness was increased by ×1, ×2, ×4, ×8, ×16, ×32, ×64 and no other values were
changed. As we see in figure 6 the Cd converged to a decently steady value very quickly but the Cl took
much longer to reach a value that was steady. Really a ×0.5 simulation should have been run to validate
that the Cl would not change as it does not appear to have settled into a steady value. Once again this was
prevented by the increase in computational complexity and the deadline for this report. Due to the changing
value of Cl , I would not use a grid resolution any larger than the smallest one I did.

Figure 6: Plots of Cl and Cd against the mesh size of the CFD.

In addition, the residuals decreased as mesh size increased except for the ×64 case (See figure 7). For
some reason that case had very low residuals. This was likely due to it not capturing any of the oscillatory
behavior.

Figure 7: Plot of the residuals against the mesh size of the CFD

A characteristic of the 0 degree angle of attack case is an oscillating wake as seen in figures 8 and 9. This
provided a good reference to compare mesh sizes as larger and larger meshes would stop resolving this oscil-
lating wake.

5
 Figure 8: Velocity magnitude map for Re = 41, 000 and 0 deg angle of attack

Figure 9: Pressure magnitude map for Re = 41, 000 and 0 deg angle of attack

Results at Re = 20, 000

Based on the previous simulation validation and mesh refinement study, the CFD was assumed to be accurate
and the lift and drag for Re = 20, 000 were simulated. No changes were made except for keeping the base size
equal to the boundary layer thickness at 1/10th chord and adjusting time step to keep the CFL approximately
0.3. The results for average Cd and Cl vs angle of attack are as follows in figure 10. Individual Cd and Cl
vs iteration graphs can be found in figure 16 in the Appendix.

Figure 10: Plots of Cl and Cd against angle of attack from CFD at Re = 20, 000.

6
Results at Re = 10, 000
Re = 10, 000 was run with the same modifications as Re = 20, 000. The results for average Cd and Cl vs
angle of attack are as follows in figure 11. Individual Cd and Cl vs iteration graphs can be found in figure
16 in the Appendix as well.

Figure 11: Plots of Cl and Cd against angle of attack from CFD at Re = 10, 000.

Overall Cl vs Cd results (polar plots)

Putting both the Cd and Cl together on a polar plot (figure 12) showcases just how much more drag the
CFD had. This figure can also give insight into the overall performance of the airfoil. We definitely, see the
low drag at zero-lift and how, across a whole span of lifts drag stays decently low. As expected, the different
Reynolds numbers have similar Cl vs Cd graphs.

Figure 12: Plots of Cl vs Cd from both CFD and Empirical data.

Conclusion
In conclusion, the MA409 airfoil provides decent performance at low Reynolds number flow. We see that
there is some oddities in the simulation with respect to Cd and, at high angles of attack, Cl . A higher

7
REFERENCES REFERENCES

resolution mesh should be run to investigate wether the error is simply due to having too large or a mesh or
another issue.

Appendix

References
[1] Summary of Low-Speed Airfoil Data. Vol. 1. Virginia Beach, Virginia: SoarTech Publications, 1995.
isbn: 0-9646747-1-8.
[2] Summary of Low-Speed Airfoil Data. Vol. 5. 2012.
[3] Airfoil Tools. Dec. 18, 2022. url: http://www.airfoiltools.com/index.

Cd and Cl coefficients over time

Figure 13: Plots of Cl and Cd over time for Re = 41, 000.

8
REFERENCES REFERENCES

Figure 14: Plots of Cl and Cd over time for Re = 41, 000 as mesh size increases.

Figure 15: Plots of Cl and Cd over time for Re = 20, 000.

9
REFERENCES REFERENCES

Figure 16: Plots of Cl and Cd over time for Re = 10, 000.

10