Mechanics and Mechanical Engineering

Vol. 20, No. 4 (2016) 579–593

⃝
c Lodz University of Technology

Development of a CFD model for propeller simulation

Mateusz Stajuda
Maciej Karczewski
Damian Obidowski
Krzysztof Jóźwik
Lodz University of Technology
Institute of Turbomachinery
mateusz.stajuda@p.lodz.pl

Received (29 September 2016)

Revised (6 November 2016)
Accepted (11 December 2016)

The article presents a development of numerical model for a single propeller simulation
and comparison of obtained results with experimental data available from a test campaign
in scale 1:1. Described simulation is a steady state computation taking advantage of
Multiple Reference Frame model implemented in Ansys CFX. The paper includes an
analysis of rotating domain thickness influence on numerical values of thrust and power.
The results indicate that this type of simulation may be sensitive to the sizing of rotating
domain especially when disc solidity is low, or when the number of blades is 2, a frequent
situation in all electric flight vehicles. The analysis shows that performing simulations,
using one domain sizing, for a number of flight scenarios requiring analysis of a few
rotational speeds can produce unintuitive results. Therefore, it is suggested to calibrate
the model, preferably by experimental results.
Keywords: propeller, CFD, electric airship, UAV, manned flight, experiment.

1. Introduction
The emergence of new all–electric aircraft concepts has resulted in an increased
interest in propellers design and optimization. New vehicle designs are presented
by technological startups, but also by major aviation companies, believing in a
development of this market sector. Promising concepts were presented for example
by Agusta Westland and E–volo. The Project Zero and Volocopter are presented in
Fig. 1a and 1b. Many other electric aircrafts are to be presented in the oncoming
years. The growth in the market of electric aircrafts is encouraged by European
Union Clean Sky 2 program [1], which goal is to reduce the fuel consumption, carbon
dioxide and nitric oxides emissions and to decrease the noise footprint of the air fleet.
Created all electric or hybrid aircrafts are characterized by a much simpler design,
in terms of mechanical components, than contemporary helicopters and often higher
580 Stajuda, M., Karczewski, M. Obidowski, D. and Jóźwik, K.

number of propulsion elements. The use of electric motor instead of a fuel powered
engines offers the possibility of relatively easy changes in rotational speed of the
propellers, which allows the designers to trim the aircraft via the rotational speed
variation. This, in turn, changes the requirements which have to be met by the
propeller design, as it should be efficient for a wide range of rotational speeds.

a) b)

Figure 1 Examples of all-electric aricarafts: a) Volocopter by E-volo [3], b) Project Zero by

Agusta Westland [2]

The rise of momentum theory, widely used for the general design of propeller blades,
dates to the beginning of 20th century with the works of Betz [4] and Glauert [5].
Betz initiated a general theory of flow through rotating disc, while Glauert enriched
it with terms accounting for momentum loss, including the tip loss factor. The tests
on different blade geometries and blade configurations were performed throughout
1950s, 60s, 70s and 80s, for example in NACA research centers by Gessow [6],
Hartman and Biermann [7] or Harrington [8]. A blade shape, including the chord
distribution, planform, twist and pitch, was the source of interest. The studies also
included ducted and coaxial configurations. The experimental research provided
necessary data for theoretical development in the field of propeller design, leading
to a creation of analytical methods such as Blade Element Theory (BET), Blade El-
ement Momentum Theory (BEMT) or Free Vortex Method (FVM). These methods
are fast and may be used for the design and performance validation of the propellers
[9]. Despite their reliability and quickness, they always require input parameters
from experiment to produce accurate results, which make them difficult to use in
order to precisely assess new designs without a complementary experimental study.
Generally, a computational fluid dynamics (CFD), offering a wide range of mod-
elling capabilities is the most precise design validation method. The generality of
CFD codes and necessity of introducing turbulence closure in Navier–Stokes equa-
tions make it difficult to create a proven computational model able to produce
coherent and reliable results, especially when highly unsteady flows, as in propeller
simulations, are expected. The CFD simulation reliability depends on many as-
pects, including computational domain sizing, mesh size and quality as well chosen
solution type (steady state or transient) and physical formulation (type of equation
solved, additional wall formulations and corrections). A validated numerical model
offers possibility of a much cheaper and more thorough analysis than experiments.
The most reliable information on propeller performance can be delivered by full–
scale tests, which are very demanding and expensive, therefore they nowadays are
performed by mainstream aircraft manufacturers having sufficient funds for such
 Development of a CFD model for propeller simulation 581

campaigns. The majority of papers concerning the topic of propellers, such as

those of Juhasz et al. [10] or Syal and Leishman [11] relies on the experimental tests
described, among others, in sources [6, 7, 8]. The articles concerning validation of
new propeller design based on full scale tests for a manned VTOL flight were not
found.
The aim of this paper is to describe the development of numerical model used
for simulations of a single propeller. The procedure of domain sizing, meshing
approach and numerical setup is well described. The article presents a study on
the influence of rotating domain size on the calculated thrust and power magnitude
for a discreet model prepared in ANSYS CFX. A short study of an interface type
is also performed. The obtained results are compared with full scale experiment
results to draw conclusions about an appropriate domain sizing.

2. Method of CFD Modelling

The modelling approach presented in this paper is a result of authors’ experience
gained during the work in the field of numerical simulations and propeller designs.
The goal was to assess the importance of model parameters on obtained results.
This can lead to a reliable and versatile simulation setup which may be used for
tests of different propeller designs and optimization of the blade shape without the
need for expensive, profound experimental studies.

2.1. Simulation approach

A steady state simulation taking advantage of Multiple Reference Frame (MRF)
approach is the model used by the authors [12]. This model allows one to connect a
rotating domain containing the propeller blade and its proximity with a complemen-
tary stationary domain representing the volume of air further from the propeller.
The two domains are connected using domain interfaces of an appropriate type,
which transfer data between stationary and rotating frames This method of sim-
ulation introduces certain approximation, as the simulation is stationary and the
blade itself does not rotate with respect to the surrounding domain. The method
is, however, industrially accepted for simulating propellers and provides reliable
results.

2.2. Computational domain

The domain (Fig. 2a) consisted of rotating propeller disc (marked in blue), mo-
tor (marked in red) and stationary surrounding domain. Only half of the physical
domain was computed as a two–bladed propeller was analysed thus allowing for ro-
tational periodicity. The simulation did not include the test bench elements, present
during experimental campaign. The propeller blade was placed in the centre of the
computational domain (Fig. 2 b). The surrounding domain size was selected on
the basis of numerical experiments performed in previous computational campaigns
[13]. The domain radius was five times bigger than the radius of the propeller and
domain height was 10 times larger than the diameter of the propeller. Such sizing
ensured that the boundary conditions were sufficiently far from the propeller disc.
582 Stajuda, M., Karczewski, M. Obidowski, D. and Jóźwik, K.

2.3. Mesh
The mesh for the simulation was generated in two different mesh generators. For
the rotating domain, where precise solution was crucial, a structural hexahedral
mesh in ICEM CFD was created. The size of the mesh elements was chosen to
provide acceptable mesh quality and to allow one to fully solve the boundary layer
at the blade, which in case of ANSYS CFX, requires y+ value below 2 [12]. The
propeller mesh consisted of around 200 elements chord–wise for capturing the pres-
sure changes along the blade chord, which have important influence on the propeller
thrust and around 300 elements span–wise, which offer much lower resolution, how-
ever sufficient to account for pressure gradients expected in this direction. A mesh
clip along the blade chord is presented in Fig. 3. The mesh was also refined in
the region of leading edge and behind the trailing edge, where the blade wake was
expected. The final mesh of the rotating domain comprised of 8 million nodes.

a)

b)

Figure 2 a) General view of the computational domain, b) view of the rotating domain

The mesh in the surrounding domain was chosen to be the hybrid one. Hexahedral
elements were used in the propeller disc proximity for better shape and size fit
of elements on both sides of the interface (Fig. 4). Further from the propeller,
a tetrahedral mesh was created. It allowed for a standard growth rate of 1.2, high
quality and reasonable aspect ratio of the elements. A use of tetrahedral mesh also
gives possibility of easy changes in a surrounding domain, for example introducing
test bench elements, without a need for changes in the meshing strategy. The mesh
in a surrounding domain was created using ANSYS Meshing, which requires less
labour and allows one to generate the mesh of more complicated geometrical models
than structural mesh created with the use of ICEM CFD.
The sizing in the regions of blade induced wake path, near the propeller blade tip
and in the areas where the prop induced eddies dissipate by exchanging energy
with the quiescent surroundings was the most important in the outer mesh. For
this purpose a growth rate from the hexahedral mesh surfaces was decreased from
 Development of a CFD model for propeller simulation 583

default 1.2 to 1.05 and sizing using body of influence was added (see Fig. 5). The
outer mesh was defined to include as little nodes as possible, to decrease the need for
computational power and minimize the computational time needed to converge the
solution. The total number of nodes in outer domain, while maintaining expected
sizing and good elements quality, was below 3 million.

2.4. Mesh independence

A mesh independence and rotating disc thickness study were performed for this
computational model. Both tests included three variants, as they allowed one to
capture the trend and draw conclusions. Mesh independence study was only per-
formed for the rotating propeller domain. The number of nodes varied from 5 to
12 million, as the number of elements along the blade span and blade chord was
changed. The sizing of the first layer, therefore y+ value, remained relatively con-
stant. Results of the study showed that the difference between observed parameters
for the least and the most dense meshes were below 1%. Consequently, the mesh
with the lowest number of nodes was chosen. Attempts in further decrease of the
mesh size resulted in low quality of the mesh elements, therefore these meshes were
not used.

Figure 3 Mesh clip inside the rotating domain

584 Stajuda, M., Karczewski, M. Obidowski, D. and Jóźwik, K.

Figure 4 Mesh structure in the proximity of propeller domain; white region is where the propeller
mesh is inserted

Figure 5 Mesh clip of the stationary surrounding domain

2.5. Numerical model setup

The Reynolds–Averaged Navier-Stokes simulation prepared for propeller analysis
was treated as fully turbulent. In the simulation, the Shear Stress Transport tur-
bulence model was used. The SST turbulence model formulation in ANSYS CFX
provides robustness and ease of application as it allows for automatic wall treat-
ment, i.e. changing from full solution to boundary layer modelling depending on
grid resolution in the model [12].
For proper thrust and power computation, the exact pressure distribution and pos-
sible separations on the blade were very important, therefore full boundary layer
resolution was desirable with y+ value below 2 on the whole propeller. Due to the
aspect ratio limits and quality of resulting elements, the zone where y+ was below
2 was limited to the outer part of the blade, where majority of the thrust is gen-
erated and majority of the power is consumed. The simulation was performed for
absolute pressure of 101325 placeStatePa. The temperature was set to 20 o C. The
fluid entering the domain had low turbulence intensity (<1%). The total energy
model was used in simulations, which ensured that the compressibility effects were
taken into account.
 Development of a CFD model for propeller simulation 585

The multiple frame of reference (MFR) model used for this simulation introduces
certain simplifications, due to which its application is not straightforward. In steady
state simulation, the domain does not physically move, but only the velocity re-
sulting from domain rotation is added in the appropriate nodes. This leads to a
situation, where the propeller disc domain is numerically rotating, while the blade
remains stationary with respect to the surrounding domain. For this reason, the
modelling of wake development downstream the blade differs along with the rotat-
ing disc thickness. Thus each MFR model requires thorough study of disc thickness
influence on the resultant values of thrust and power.

2.6. Disc thickness study

The disc thickness study appears to be very important for the MRF model in
the two bladed propeller simulation due to limited mixing of the trailing vortices.
The vorticity field shed by a propeller with such a low number of blades is thus
under the strong influence of the quiescent surroundings. A rapid transition from a
rotating frame of reference to a stationary one may result in an unphysical numerical
diffusion of the conservative quantities. In this sense, the MRF introduces a certain
simplification, as in steady state simulation, the domain motion is simulated by
adding appropriate velocity components as source terms in N-S equations, while the
domain and, therefore, mesh remain in the constant position. To transfer the fluxes
of conservative parameters between rotational and stationary domains, the possible
interface options are as follows: frozen rotor, which directly transfers the data
between adjacent cells through the interface and stage, which performs averaging
of conservative quantities over the control surfaces building the interface and then
transfers the resulting values over. In case when the blade chord is only a small part
of the domain, a frozen rotor interface is recommended option [12]. Even though
the interfaces provide correct data transfer for basic flow quantities, they may cause
a loss of information and changes in certain complex flow structures induced by the
propeller as described above. Therefore, the influence of rotating domain sizing
and resulting interface placement were assessed, as disc thickness changes the flow
characteristic around the propeller and alters the interface placement.
The disc thickness study described in this paper consisted of 4 thickness values,
ranging from 2 blade chords (2 BC) up to 4 blade chords (4 BC). The disc thick-
ness is increased symmetrically i.e. the same height is added above and below the
blade, using the rotating domain of 2 BC height as a starting point. The results
obtained with changing disc thickness showed significant variability, confirming the
importance of this parameter for obtaining reliable results.

3. Results
The described numerical model was used to simulate the propeller at three differ-
ent rotational speeds – 0.84 of nominal, nominal, and 1.16 of nominal speed. For
each RPM, four rotating domain sizes were tested. The obtained results provided
numerical parameters, such as thrust and power, but also additional information
which were used to assess the simulation quality.
An assessment of resultant y+ value on the blade was the first step for solution
validation. Fig. 6a shows the y+ parameter value in a range up to 10, indicating
586 Stajuda, M., Karczewski, M. Obidowski, D. and Jóźwik, K.

that the whole blade is included in this range. The y+ contour up to two (Fig. 6 b),
shows in which regions the boundary layer was fully solved. Presented values were
obtained for the highest 1.16 nominal rotational velocity. It is visible that the y+
parameter is below 2 for the majority of the blade area. The value is higher for the
blade inboard part, due to a meshing compromise described earlier. The separate
study aimed at obtaining the y+ value smaller than 2 for the entire blade confirmed
that, despite increased computational resources, no further improvement in thrust
nor power was achieved. The area average value of y+ parameter over the whole
blade is 1.27.

a) b)

Figure 6 Y+ parameter contour on the propeller blade: a) with range up to 10; b) with range
up to 2

a)

b)

Figure 7 a) Pressure field near the interfaces (black lines); b) velocity field near the interfaces
 Development of a CFD model for propeller simulation 587

Velocity and pressure continuity at the interfaces were another important parame-
ters which had to be assessed. The contours showing pressure and velocity transfers
through interfaces are shown in Fig. 7a and 7b. The interfaces proved to be coher-
ent and ensure physical data transfer. Even though the same level of accuracy was
observed for smaller disc thickness (2 BC), the results differed significantly. This
indicated that not only the integrity of fluxes across interfaces is to be assessed
when using the MRF model.
The converged simulations were next used to calculate thrust and power values for
four rotating domain sizes at three rotational speeds. The results showed high de-
pendence of thrust and power on the rotating domain sizing. The experimental data
was used for a comparison with obtained numerical results and recommendation of
an appropriate disc thickness. The results for different rotating domain sizing are
shown in terms of thrust and power differences from the experimental values.

Figure 8 Thrust difference from experiment for different rotating domain sizing

Figure 9 Power difference from experiment for different rotating domain sizing
588 Stajuda, M., Karczewski, M. Obidowski, D. and Jóźwik, K.

The graph in Fig. 8 presents the difference between numerical and experimental
values of thrust for three rotational velocities. Fig. 9 shows a similar dependence
for differences between experimental and numerically obtained values in terms of
power. The figures indicate that an increase of disc thickness results in an increase
of power and thrust generated by the blade. Considering the thrust, a very close
match for experimental values can be found in the range of the applied domain
sizing. The corresponding values of power are lower than the experimental ones,
with the difference increasing along the rotational speed increase.
Due to the thrust and power variation for different rotating domain sizing, an
analysis of thrust and power distribution along the blade for different disc thickness
was conducted. In Fig. 10a–10c thrust distributions along the blade for different
disc thickness are presented. A difference in the thrust value is visible along the
whole blade span, with the biggest variation for the outboard blade region especially
between the cases with lower domain span (2 and 2.67 BC) and the two with
larger rotational domain sizing (3.33 and 4 BC). The highest difference in thrust
is observed for 1.16 of nominal RPM, where the air velocity and, therefore, forces
generated on the blade surfaces are the highest. In general, the higher the disc
thickness, the bigger the thrust generated by a blade portion is. This confirms
the assumption that this mesh parameter has an impact on pressure distribution
forming around the blade sides. In the absence of the intensified flow mixing,
normally present when numerous blade aerodynamic trails are encountered, the
rate of dissipation of energy carried by vortex sheets is hindered. The location
of rotating–stationary interface seems to have an upstream influence on the base
pressures and thus has to be moved further away to allow sufficient room for loss
mechanisms to unfold.
Fig. 11 from a) to c) presents power distributions along the blade in a similar
convention as in Fig. 10. The power distribution analysis also indicates that the
amount of power produced by the blade is bigger for each blade section with in-
creasing disc thickness. Interestingly enough, in case of power, however, higher
differences are observed in the inboard blade part. Higher differences are also vis-
ible for the region close to the blade tip where especially the 2 BC case seems to
underperform.

4. Discussion
The analysis presented in this paper shows that CFD is highly sensitive to numer-
ical model design, especially when applying simplifications, such as steady state
model for a propeller. In case of MFR model, the disc thickness and, therefore,
interfaces placement plays a crucial role. The mistakes done in this aspect may
lead to significant differences in results, despite the overall good mesh quality and
attained solution convergence level.
The curves presented in Figures 8 and 9 have a very similar character for different
disc thickness. This proves the consistency of the method used in this paper. The
difference in obtained values of power and thrust implies that the numerical model of
propeller simulation should be calibrated for a given rotational speed and probably
blade geometry. The thrust and power variation from experimental results for the
same rotational speed, reached up to 15%. Such a difference may lead to important
 Development of a CFD model for propeller simulation 589

mistakes when the blade characteristics is used for a design of an aircraft.

The results show, that the smallest disc height, which, intuitively, could be assumed
as the most correct, as the smallest volume of air around the propeller is rotated,
produces much lower power and thrust values.
It may be noticed that for reproduction of the experimental results, the disc thick-
ness should grow along with the rotational speed of the propeller. This may result
from the fact that the pressure distribution on a blade depends on the vortex sheet
created by the propeller. If the disc is too thin and an interface is placed too close
to the propeller, the vortex may not be correctly simulated, which influences the
pressure distribution.
Figures 8 and 9 show that differences from experimental values either in terms of
thrust or power are present for the given disc thickness. According to the results,
the choice of disc thickness for which the value of power and thrust will be identical
with the experiment is not possible. While the error made in case of 0.84 of nominal
RPM and nominal RPM is not high, it becomes more significant for the highest
rotational speed.
It should be, however, noted that the experimental results were obtained with a test
bench, which was not included in the simulation. The influence of this structure
could lead to disturbances, which would, in turn, result in higher power consumption
for the same value of thrust. Due to this fact, the simulation results may show that
for a propeller working in a flow field not altered by the test bench elements, the
same thrust as in experiment would be obtained for lower power, or for the same
power, higher thrust value could be attained.
The thrust distribution along the blade shows that the differences are visible along
the entire blade. There are, however, regions, where the difference is more notice-
able: mainly on the outboard portion of the blade and near the tip.
An analysis of power distribution indicates that very high amount of power is gen-
erated near the tip of the blade, which results from significant energy consumption
by the forming tip vortex. The complementary effect of thrust decrease due to tip
leakage is also visible on thrust distributions. In case of power, higher difference
for changing domain sizing is visible in an inboard part of the propeller, while the
outboard part is almost identical for the three highest domain sizes. The difference
between the three highest velocities is also visible at the tip of the propeller, where
for higher disc thickness, higher power is generated.
The study considered only equidistant rotating domain size changes. The impor-
tance of domain growth on the inflow side or outflow side could also be established
for further analysis. This could give an answer whether for the MRF approach, the
distance of interface from the blade is more important for the inflow or outflow side.

5. Conclusions
This article presents the development of numerical model for a propeller simulation,
gives detailed description of the performed simulation and presents its validation
using the experimental data. The created CFD model using multiple reference
frame formulation allows one to precisely simulate the behaviour of the propeller
using steady state approach. It is more efficient than transient simulation, which is
more time consuming and requires far greater computational resources.
590 Stajuda, M., Karczewski, M. Obidowski, D. and Jóźwik, K.

a)

b)

c)

Figure 10 Thrust distribution for different rotating domain sizing: a) 0.84 of nominal RPM; b)
nominal RPM; c) 1.16 of nominal RPM
 Development of a CFD model for propeller simulation 591

a)

b)

c)

Figure 11 Power distribution for different rotating domain sizing: a) 0.84 of nominal RPM; b)
nominal RPM; c) 1.16 of nominal RPM
592 Stajuda, M., Karczewski, M. Obidowski, D. and Jóźwik, K.

The study shows that the model is sensitive to the rotating domain sizing, there-
fore it should be properly calibrated for the given blade and rotational velocity.
The sensitivity for two–bladed propeller may come from important propeller wake,
concentrated on a relatively small interface area, compared to the overall interface
extent.
The difference in results for the given disc thickness changes with the rotational
velocity. When attempting to reproduce the experimental results, for higher rota-
tional velocities, higher disc thickness is required. The need for the size increase
is a result of propeller wake growth with the rotational velocity. A part of the
differences may also come from altering the inflow conditions, which happen when
disc thickness is changed. The study of a non-symmetric disc changes could give an
answer to the observed behaviour.
The comparison of numerical and experimental results shows that different disc
thickness should be used for different rotational velocities. There is no general
formula presenting the optimal thickness, therefore further research in this area is
required to provide dependence from rotational velocity, blade geometry or other
parameters. The choice of optimal disc thickness should be possible for a certain
case, producing power and thrust values matching with the experiment. The diver-
gence of results presented in this study is attributed to an influence of test bench,
which was not included in the numerical simulation.
Changes in rotating domain thickness produce systematic changes in power and
thrust values, while not changing significantly the distribution of those quantities
along the blade. In the inboard part, the power and thrust are simultaneously
increased, while in the outboard part, the power remains almost constant while the
thrust increases significantly. This may lead to a conclusion that the changes in
the velocity triangle are not linear and more profound studies are required to fully
understand the influence of rotating domain sizing on flow structure in steady state
MRF simulations.

References
[1] Clean Sky, Overview of the Proposed Programme, 2013.
Online: www.cleansky.eu/sites/default/files/documents/20131009 cs2programme
overviewfinal.pdf.
[2] [http://newatlas.com/project-zero-electric-tilt-rotor-aircraft/26561/#gallery]
[3] e–volo GmbH, Online: http://www.e-volo.com/.
[4] Betz, A., with appendix by Prandtl, L. Screw Propellers with Minimum Energy
Loss, Gottingen Reports, 193–213, 1919.
[5] Glauert, H.: Airplane Propellers, :Aerodynamic Theory, edited by W. Du-
rand, Div. L, Vol.5, Peter Smith, Gloucester, MA, 169–269, 1976.
[6] Gessow A.: Effect of rotor blade twist and planform tapper on helicopter perfor-
mance, NACA Technical note, no. 1542, 1948.
[7] Hartman, E. P. and Birdman, D.: The aerodynamics of full scale propellers
having 2, 3 and 4, blades of Clark Y and R. A. F. 6 airfoil sections, NACA Report
no. 640, 1938.
[8] Harrington, R. D.: Full–Scale Tunnel Investigation of the Static Thrust Perfor-
mance of a Coaxial Helicopter Rotor, NACA Technical Note, 2318, 1951.
 Development of a CFD model for propeller simulation 593

[9] Leishman, G. J.: Principles of helicopter aerodynamics, Cambridge University

Press, 2006.
[10] Juhasz, O., et al.: Comparison of Three Coaxial Aerodynamic Prediction Methods
Including Validation with Model Test Data, J AM HELICOPTER SOC, 59, 3, 1–14,
2014.
[11] Syal, M. and Leishman, J. G.: Aerodynamic optimization study of a coaxial rotor
in hovering flight, J AM HELICOPTER SOC, 57, 4, 1–15, 2012.
[12] ANSYS, C. F. X. Solver modelling guide, Release 17, 2016.
[13] Karczewski, M., and Eglin P.: Application of the modified blade element method
for propeller blade design on high–speed helicopter X3, CMP-Turbomachinery, LUT,
145, 63–64, 2014.