Computational Fluid Dynamics – A.A.

2024-2025

Study of ground effect on NACA4412 in incompressible

flow
Silvia Bon, 10680658
Massimo Campanella, 10671690 Group number 14

Giovanni Cernotto, 10706345 Lecturers:

Dr. Barbara Re
Alessandro Colombo, 10671826 Dr. Andrea Rausa
Lorenzo Pronzello, 10705855

1. Problem statement
The purpose of this report is to
study the ground effect of a 2D
NACA 4412 profile under steady
subsonic conditions. The idea is to
evaluate the main aerodynamic co-
efficients for different heights above
the ground and different angles to
characterize the behavior of the air-
foil. To do so, a numerical anal-
ysis is carried out using the open
source software SU2. The results
obtained from the study are com-
pared with experimental and numer-
ical data from papers [2] [4]. Figure 1: Flow speed along the airfoil.
The ground effect is a well-known
physical phenomenon that can be
exploited to increase the efficiency of a wing in proximity to the ground. When an airfoil is
in this condition, the flow under the profile is considerably modified as it is constraint to become
parallel to the surface below. The resultant effect is, for a positive angle of attack, an increase
in the pressure distribution on the lower surface due to the air cushion that is generated [2] [4].
For a real 3D wing, another significant effect of the ground proximity is the reduction of drag.
This happens because the ground often pushes the wing tip vortices outward along the span,
leading to a decrease in downwash angle and induced drag. However, this is not true considering
only a 2D case and, for this reason, the focus of this report is on value the CL and Cp of the
airfoil. Flying in this condition is something that every plane experiences during landing and
take-off, but the gain of efficiency was also exploited in some specific airplane such as the russian
Ekranoplan.

1
2. Mesh
2.1. Mesh construction
During this project, two different simulations were performed: one to measure the influence of
the ground effect and one without the ground, for comparison. For both cases, the same profile
geometry was used: a NACA 4412 profile (Figure 2) with a chord length of 1 m, consisting of 201
points with a higher concentration of points at the leading and trailing edges to improve shape
accuracy. To generate a smoother mesh, a blunt trailing edge was used, as visible in Figure 3.

Figure 2: NACA 4412 profile. Figure 3: Detail of blunt trailing edge.

For the case regarding the ground effect, represented in Figure 6, a rectangular farfield was
created, measuring 30m length and 12m height. The profile was then placed at various heights
above the ground (specifically 0.05, 0.2, 0.6 and 1 normalized by the chord length), at different
angles of attack: 0 °, 2 °, 4 °, 6 °, 8 °, and 10 °. The presence of the wall on the lower boundary
of the farfield made rotating the profile a more practical solution compared to altering the inlet
velocity. In the case without the ground effect, an O-grid (Figure 5) mesh was designed around
the same profile, without a wall under the wing, to not consider any ground effect.

Figure 4: Mesh example of the case with wall. Figure 5: O-grid mesh example of the case
without the wall.

An unstructured mesh was generated for both cases, because it allows more flexibility when
changing angle of attack and ground distance. To model the boundary layer, a structured field
was incorporated. The height of the first cell was calculated by imposing y + = 1, allowing for a
good direct computation inside the boundary layer. The freestream velocity was set equal to 100
m/s, the density at 1.225 kg/m3 , and the viscosity at µ∞ = 1.716 × 10−5 N·s/m2 . Consequently,
the first cell thickness near the wall was set to y = 0.0000076 m. The same height was used for
the profiles of the two cases and for the lower side of the rectangular field, which represents a
solid wall (Figure 7).
From this thickness, a gradient of 1.1 was applied to expand the rectangular cells, ensuring a
smoother transition to the unstructured grid. Additionally, a finer grid spacing was used near
the leading and trailing edges due to higher pressure and velocity gradients, which require higher

2
Figure 6: Mesh example of the box behind the Figure 7: Example of the structured mesh of
airfoil. the boundary layer.

accuracy.
As visible in the results of the first simulations related to the wall effect, the wake region exhibits
significant velocity and pressure gradients. Consequently, this region was not neglected, and a
dedicated box was incorporated into the mesh to cover it. A detail of the box shape can be seen
in Figure 6. To ensure an effective design, a box length of 15m and a height of 0.5m were chosen.
Additionally, we linked the refinement level of the box to the same parameter used for the rest
of the grid to maintain consistency. The mesh design strategy was based on defining the cell
dimensions (and consequently the total number of nodes and cells) using a single parameter. In
this way, increasing that parameter scaled the cell sizes while maintaining the proportion on the
mesh accuracy between the different regions.

2.2. Grid convergence

To ensure the reliability and accuracy of the CFD analysis results, a mesh convergence study
was conducted. This process was essential to verify that the aerodynamic coefficients of lift
(CL ) and drag (CD ) became independent of the mesh resolution. A total of five mesh versions
were generated and analyzed, each with progressively refined resolution, increasing the node
density, and focusing on critical areas of the domain, such as the leading and trailing edges of
the airfoil, as well as regions with high pressure and velocity gradients. The analysis showed that
starting from the fourth mesh, variations in CL and CD values were negligible, indicating that
the results had become stationary and were no longer influenced by further refinement of the
mesh. As a result, the fourth mesh was identified as optimal, providing the best balance between
result accuracy and computational cost. Although the fifth mesh demonstrated only minimal
differences compared to the fourth, it significantly increased computational time without any
tangible improvement in the quality of the result.
Table 1, Table 2, Figure 8 and Figure 9 show the trends of the aerodynamic coefficients as a
function of the number of nodes. For these graphs, the configuration with alpha 0 and 8 degrees
and a minimum ground clearance of h/c = 0.05 was chosen. As we can see, the variation of the
coefficients decreases until it approaches a value from which it no longer deviates as the number
of nodes increases.

3
 Mesh Nodes Cells CL CD Mesh Nodes Cells CL CD
1 24496 33245 0.13237 0.01613 1 24724 33701 1.11815 0.06011
2 67226 103202 0.19450 0.01357 2 68688 106162 1.18368 0.05153
3 211540 360768 0.22707 0.01250 3 214250 366188 1.22201 0.04809
4 312240 546637 0.23201 0.01239 4 318178 558513 1.22761 0.04774
5 526332 948920 0.23522 0.01230 5 525081 946426 1.23428 0.04750

Table 1: Mesh convergence at α = 0◦ Table 2: Mesh convergence at α = 8◦ .

Figure 8: Mesh convergence at α = 0◦ .

Figure 9: Mesh convergence at α = 8◦ .

The same convergence study was conducted for the O-grid configuration for the airfoil without
the wall. The analysis is carried out with an angle of attack of 4 °, as shown in Table 3 and
Figure 10.

4
 Mesh Nodes Cells CL CD
1 34969 63263 0.77339 0.01088
2 111193 209368 0.79810 0.01069
3 180022 343834 0.78533 0.01007
4 360099 698675 0.77900 0.00999

Table 3: O-grid mesh convergence at α = 4◦ .

Figure 10: O-grid mesh convergence at α = 4◦ .

3. Numerical simulation methodology

3.1. Governing equations
To ensure the proper selection of the governing equations, the following assumptions were con-
sidered, in order to reduce the computational cost:
– subsonic flow, with Mach<0.3 (0.294), therefore it can be approximated as incompressibile
– steady flow
– two dimensional case
Taking into account these hypotheses, the selected numerical model is INC_RANS. This choice
is justified by the need to capture the effect of turbulence without directly solving the full Navier-
Stokes equations. Indeed, this model is an adequate trade-off between a reliable approximation
and a reasonable computational cost.
An appropriate turbulence model was essential to accurately capture the phenomena under
investigation. The Shear-Stress Transport (SST) model, combining features of k − ω and k − ϵ
models, was selected for its ability to provide accurate predictions of the behavior of the boundary
layer and the flow characteristics near the surfaces, while maintaining robustness in the outer
flow. These qualities made it the ideal choice for RANS simulations, offering a balanced trade-off
between computational efficiency and predictive accuracy.

3.1.1 Physical data

The following data were used in this study to define the numerical model:
– Freestream density ρ∞ = 1.225 kg/m3
– Freestream velocity on x axis u∞ = 100 m/s
– Freestream viscosity µ∞ = 1.716 × 10−5 N·s/m2 , imposed as constant
– Reynolds number Re = 7.14×106 considering the chord l = 1m as the characteristic lenght

5
3.1.2 Boundary conditions
For the RANS solver, the correct boundary conditions were set for every section of the domain,
as defined in Subsec. 2.
A farfield configuration was applied to the corresponding parts of the domain. For the wall
boundaries, which include both the airfoil surface and the wall itself, a heat flux marker was
required and set to zero. Since in the case study the airfoil is considered moving over the surface,
it was necessary to insert a moving wall condition. To simulate the movement of the airfoil, the
velocity of the wall was set equal to the freestream velocity u∞ .

3.1.3 Boundary conditions without wall

When examining the case without the wall, the same conditions were established, except for the
wall, which was removed.

3.2. Numerical methods

In this section, an analysis of the numerical methods employed to solve the problem is provided.
First, the reference quantity selected to determine the convergence of the method is the pressure
root mean square residual. Its value was set to 10−9 : at this point the target values are stable
as shown in Figure 11 and Figure 12, which are examples corresponding to the case with α =
4°, ground distance = 0.6, and the chosen refinement.

Figure 11: Example of the variation of CL Figure 12: Example of the variation of the
with respect to the number of iterations. pressure root mean square residual with re-
spect to the number of iterations.

First, an implicit time discretization scheme was applied to both the flow and turbulence nu-
merical methods, specifically adopting the implicit Euler scheme. The choice is justified by its
unconditional numerical stability [3]. Consequently, it was possible to set the CFL (Courant-
Friedrichs-Lewy) number higher than one (2.5) and it was imposed as adaptive to allow for a
dynamic optimization of the time steps, improving the efficiency of the simulation.
The weighted least squares method was used for spatial gradients due to its robustness even in
the presence of unstructured meshes, such as the one under consideration.
Considering the simplicity of the geometry in the case study, the steady state, the flow velocity,
and the absence of discontinuities, the numerical method selected to solve the PDE is the Flux
Difference Splitting (FDS). To achieve a second-order upwind scheme, therefore a more accurate
solution, the MUSCL scheme was used. For the same reason of absence of discontinuities, no

6
slope limiter was used. An alternative would have been a central scheme method such as JST
or Lax-Fredrich but second order upwind schemes are in general more accurate [1]. Moreover, a
first order scalar upwind method was used to evaluate the turbulence part, since it guarantees
a lower computational cost.
To solve the linear system the FGMRES solver was chosen with the ILU (Incomplete LU) pre-
conditioner to accelerate and enhance convergence by improving the conditioning of the matrix
of the system.

4. Comparison with previous studies

This section reports and analyses the results obtained with previous papers on the ground effect
for the NACA 4412 profile. The available reference data originate from a numerical analysis [4]
and an experimental study [2] carried out under conditions closely aligned with those utilized
in this report. Based on the simulation results and the findings reported in the literature, it
becomes evident that the observed trends align closely with the results obtained in this study;
however, the obtained values show discrepancies. To address this problem, few actions were
implemented.
The first step involved modifying the mesh construction. The farfield was simplified with larger
cells and the wake length was reduced to four times the chord length behind the airfoil. This
allowed for a reduction in cell size near the airfoil and the wall without significantly increasing
the computational effort.
The second step focused on adjusting the airfoil geometry. Specifically, the original profile was
truncated at 99.05% of the chord length of the complete profile, while the modified version
features a cut at 99.85% (Figure 13). This adjustment ensures that the aerodynamic behavior
of the cut airfoil is closer to that of the sharp-edged one.

Figure 13: Detail of the modified blunt trailing edge.

By performing the preliminary simulation, it is observed that the first step induces minimal
changes in the results, whereas the second step significantly enhances the coefficients, bringing
them closer to both experimental and numerical reference values. A convergence study was then
performed for the new mesh with the modified airfoil, focusing on the most challenging scenario:
an angle of attack of 10 degrees and a ground height of 0.05 m.
It is observed that, compared to the previous case, the number of nodes and cells appears to be
reduced, as shown in Table 4. This reduction isn’t caused by a globally lower refinement level, but
it is due to a reduction on the refinement level of the farfield, while the grid in the most sensitive
areas maintains a high level of detail. After conducting this study, mesh 4 is selected and used
for simulations at the other heights and angles presented in the paper, allowing comparison. [2]
[4]

7
 Mesh Nodes Cells CL CD
1 62487 80346 1.4841 0.0522
2 74362 97680 1.4542 0.0531
3 91630 123772 1.4697 0.0542
4 159979 230844 1.4903 0.0579
5 244137 369440 1.4877 0.0564

Table 4: Mesh convergence at α = 10◦ ,

ground distance = 0.05m.

Figure 14: Mesh convergence at α = 10◦ , ground

distance = 0.05m..

In Figure 15 and Figure 16 a comparison of the CL values obtained at ground distance 0.05 m
and 1.0 m is presented. The graphs show the values obtained from the two papers, the data
with the new configuration, and the previous one. As can be observed, the improvements made,
in both geometry and mesh construction, bring the obtained values closer to those from the
literature, almost overlapping the experimental data [2].

Figure 15: CL - α diagram at gd = 0.05 m. Figure 16: CL - α diagram at gd = 1 m.

As can be observed, the new configuration is very close to the one obtained experimentally.
In particular, it is possible to calculate the Root Mean Square Error (RMSE) to measure the
difference between the reference values and the obtained ones. Considering the case with ground
distance of 1 m, it can be observed that in the previous case, RM SEpast = 0.1544, while in the
new configuration, RM SEnew = 0.0497. This highlights a reduction in the error of 67.82%.
The same approach was followed to improve the O-grid simulation for the airfoil without wall
and the overall results are presented in the following section.

8
5. Results
5.1. CL − α
The variation of the lift coefficient as a func-
tion of the distance from the ground clearly
demonstrates how the latter influences the in-
tensity of lift. As the distance to the ground
decreases, for positive angles of attack, the lift
coefficient tends to increase significantly, em-
phasizing the importance of the ground effect
in enhancing aerodynamic performance.
The figure 17 presents a graph illustrating the
variation of CL with changes in the angle of
attack for different ground clearances. It is
evident that at lower angles, the lift is smaller
at reduced distances from the ground, whereas
at higher angles, the lift increases for closer
ground proximities. Notably, around 4◦ the
curves intersect, demonstrating the described
effect.
From the graph, it can also be observed that
the slope of the curves becomes steeper as
the ground clearance decreases, indicating a
stronger sensitivity of lift to angle of attack
in close proximity to the ground. This trend Figure 17: CL - α diagram for different ground
highlights the critical role of the ground effect distances.
in modifying aerodynamic performance based
on the angle of attack and clearance height.

The data presented in Figures 18, 19, 20 e 21 for angles of attack of 0◦ , 4◦ , 8◦ , and 10◦ show
that the trend of the lift coefficient as the height from the ground increases remains relatively
consistent. The graphs display the variation of the lift coefficient CL as a function of the height-
to-chord ratio h/c for different angles of attack, comparing the results from various studies: the
"Past results", the "Improved results" after geometry modifications, the numerical data [4], and
the experimental results [2].
It is observed that CL generally decreases as h/c increases, highlighting the ground effect, which
is more pronounced at lower heights (h/c < 0.4). The graphs reveal that the profile geometry
modifications described in Section 4 have led to a significant performance improvement. This
brings the results much closer to the data obtained from [2] and [4], compared to those derived
from the earlier configuration with the original geometry. This confirms the effectiveness of the
optimizations implemented.
It can thus be concluded that the ground effect leads to an increase in the aerodynamic perfor-
mance of the profile, confirming its crucial role in characterizing the aerodynamic behavior of
lifting surfaces near the ground.

9
 Figure 18: CL - gd diagram at α = 0◦ . Figure 19: CL - gd diagram at α = 4◦ .

Figure 20: CL - gd diagram at α = 8◦ . Figure 21: CL - gd diagram at α = 10◦ .

5.2. CP
As mentioned in Section 4, the airfoil used in the simulations has a blunt trailing edge, which
results in few anomalously high CP values near the trailing edge. These values are non-physical,
as they do not align with the expected aerodynamic behavior. To ensure accurate results, these
points were excluded from the analysis.

According to Figure 17 the ground effect strongly depends on the angle of attack of the profile.
For small ones, CL decreases approaching the ground, while for big angles the effect is the op-
posite.
To explain the difference between the two phenomena, the CP diagrams for α = 0◦ and α = 8◦
are considered.

10
 Figure 22: CP diagram at α = 0◦ . Figure 23: NACA 4412 airfoil and the ground
at α = 0◦ .

As visible in Figure 22, for an angle of attack of 0◦ , ground effect exhibits a unique behavior
compared to configurations with high positive angles of attack. Under these conditions, the
distribution of the pressure coefficient (CP ) reveals a region of negative pressure on the lower
surface of the airfoil. In the Figure 23 the NACA4412 airfoil is represented with its lowest
point at 13% of the chord length, where the minimum pressure is expected. Downstream of this
point, a divergent region forms between the airfoil and the ground, creating an adverse pressure
gradient due to flow divergence. The adverse pressure gradient is particularly high at h/c =
0.05 because it results from flow separation. The flow cannot reattach to the surface. Although
separation without reattachment is uncommon at these Reynolds numbers, the decelerated flow
below the airfoil cannot overcome the increasing pressure and detaches. This leads to a minimum
CP value on the lower surface and a corresponding pressure reduction. Meanwhile, the pressure
on the upper surface remains largely unchanged.
The net result is a reduction in total lift, as the negative contribution from the lower surface
suction outweighs any positive effects. This behavior highlights a critical limitation of ground
effect at α = 0◦ : instead of enhancing lift, the configuration results in a net downward force,
reducing the efficiency of the airfoil in proximity to the ground [2] [4].
An important factor contributing to this phenomenon is the reduction in effective angle of attack.
As the airfoil approaches the ground, the flow over the upper surface is less diverted upward due
to the proximity of the ground.
This effect reduces the effective angle of attack, meaning that the flow over the upper surface
does not accelerate as much as in free-stream conditions. As a result, the depression at the
leading edge of the airfoil is less pronounced, further contributing to the reduction of lift at low
ground clearances.

11
 Figure 24: CP diagram at α = 8◦ . Figure 25: NACA 4412 airfoil and the ground
at α = 8◦ .

As shown in Figure 17, for angles of attack greater than approximately 3.5◦ , there is a reversal
in the behavior of the airfoil as the relative ground clearance (h/c) decreases. For the case of
a high angle of attack, α = 8◦ , significant changes in the pressure coefficient (CP ) distribution
over the airfoil surfaces are introduced by ground effect, as illustrated in the Figure 24. These
changes influence both lift generation and flow behavior. At this angle of attack, the lower
surface forms a convergent region with the ground, Figure 25, leading to increased pressures
across the entire lower surface as h/c decreases. This increase in pressure is primarily due to
favorable pressure gradients in the converging passage between the airfoil and the ground. The
lift increase observed is mainly attributed to the more pronounced rise in pressure on the lower
surface, which outweighs the reduced suction on the upper surface. The pressure distributions
reveal a significant rise in CP values near the leading edge of the lower surface, followed by a
gradual recovery downstream [2] [4].
For these larger angles of attack, additional effects, such as flow compression and flow blockage,
further influence the pressure distribution.

12
6. Conclusions
In this study, the ground effect on a NACA 4412 profile under steady subsonic conditions was
analyzed. The results highlight a clear increase in lift coefficient as the ground clearance decreases
for high positive angles of attack, confirming the importance of ground effect in enhancing
aerodynamic performance. This phenomenon is particularly pronounced at higher angles of
attack, and small distance from the ground, as it is demonstrated studying the case for h/c =
0.05, emphasizing its relevance for landing and take-off configurations.
The simulations demonstrated good agreement with experimental and numerical data from the
literature, especially after modifications to the airfoil geometry, in particular getting the blunted
trailing edge as far downstream as possible. The 67.82% reduction in RMSE showcases the
effectiveness of the improvements implemented to enhance result accuracy.
However, this study has some limitations. The two-dimensional approach does not account
for three-dimensional effects, such as the reduction of induced drag, which significantly impact
real-world aerodynamics. Additionally, the use of a blunt trailing edge, necessary for numerical
stability, introduced deviations in pressure values near the trailing edge, and then it could affect
the aerodynamics coefficients.
Future work could include three-dimensional simulations and more accurate modeling of trailing
edges. Despite these limitations, this study provides a robust and detailed analysis of the
aerodynamic effects of ground proximity on a classic airfoil, offering a solid foundation for more
advanced research.

References
[1] SU2 Multiphysics Simulation and Design Software - Convective Schemes. https://su2code.
github.io/docs_v7/Convective-Schemes/.

[2] Takasaki T. Ahmed M. R. and Kohama Y. Aerodynamics of a NACA4412 airfoil in ground

effect. Technical report, AIAA Journal, 1 2007.

[3] John D. Anderson. Computational Fluid Dynamics. McGraw-Hill Education, pages 152–155,
1995.

[4] Liu P. Qu Q., Wang W. and Agarwal R. K. Airfoil aerodynamics in ground effect for wide
range of angles of attack. Technical report, AIAA Journal, 4 2015.

13