First International Symposium on Marine Propulsors

smp’09, Trondheim, Norway, June 2009

Numerical and experimental analysis of the wake behavior of a generic

submarine propeller

Fabio Di Felice1, Mario Felli1, Mattias Liefvendahl2, Urban Svennberg2

I
Italian Ship Model Basin (INSEAN), Rome, Italy
2
Swedish Defence Research Agency (FOI), Stockholm, Sweden

ABSTRACT et al., 2004; Felli et al., 2008). Such an increased

The performance and fluid dynamics of a submarine complexity of the propeller blade geometry has enhanced
propeller, running in open water conditions, was the interest towards detailed investigation tools of the
investigated using experimental and computational propeller flow field. The present study deals with the
methods. The flow velocity was measured using Laser investigation of the wake evolution mechanisms of a
Doppler Velocimetry (LDV) and it was computed using generic isolated free-running 7-blade submarine stock
Large Eddy Simulation (LES). The wake dynamics were propeller through both experimental and numerical
investigated mainly by studying the velocity field in three approaches. The computations were carried out using
transversal planes of the wake, at three loading LES and the results were validated against the
conditions. The experimental results were also used to experimental results, both quantitatively and qualitatively.
Comparisons were made for the propeller performance, as
validate the accuracy of the simulations for the most
well as the velocity field in the near wake. The LES
important flow features.
results also provide information about the flow field
Keywords around the blades where it is difficult to measure.
Propeller, LES, LDV, wake instability, turbulence Experiments were carried out in the INSEAN large
1 INTRODUCTION Circulating Water Channel and concerned LDV
Acoustic signature reduction in a submarine is correlated measurements of the velocity distribution in three cross-
to the design of the propulsive system mostly, propeller planes of the wake at three values of the advance ratio.
emitted noise being 30-50 dB higher than the other Velocity measurements were phase-locked with the
sources typically and, unlike them, having a spectrum propeller blade. The propeller geometry is presented in
dominated by specific tonal contributions (Felli and Di §2. The experimental set-up and the computational model
Felice, 2008). Submarine propeller noise is mainly the are described in §3 and §4 respectively. Results are
consequence of tip and hub vortex vorticity fluctuations documented in §5. The results of the experiments and the
induced by both the pronounced unsteady nature of the simulations are compared and evaluated in §6. Then a
blade load, due to flow distortions of the upstream wake, brief summary is given in §7.
and the periodic impact of vortical structures against the
blades (i.e. bilge vortices, rudder horseshoe vortices,
corner vortex originating in the hull-sail intersection
region). The requirements to minimize propeller induced
noise and vibrations have oriented submarine propeller
designers towards complex configurations of the blade
system, in order to minimize the blade load gradients
along the azimuth as well as to avoid any coupling with
the perturbation induced by the incoming flow. More
specifically, a typical approach is the adoption of high
blade number and highly skewed blades with tip
unloading design. This solution allows high efficiency
while reducing the size and the speed of the propeller and Figure 1. Geometry of the INSEAN E1619 propeller.
to minimize the contribution at the blade rate frequencies,
by not having the entire leading edge hit the wake shadow 2 PROPELLER MODEL
at the same time. Moreover, blade trailing wake The propeller was the 7-bladed propeller model E1619,
undergoes a large deformation, moving the interaction designed by INSEAN (Figure 1). The model was built in
point between consecutive spirals closer to the propeller one single piece of aluminum alloy “avional” and
plane as compared to a conventional propeller (Di Felice
checked with a 6-dof CMM. The outer skin was black 4 COMPUTATIONAL METHODS
anodized to reduce laser beam reflection during the tests. The flow solver has been developed at FOI, using the
The main features of the propeller are documented in open source library OpenFOAM 1 , version 1.3. The
Table 1. OpenFOAM software library is designed to provide a
Table 1. Main propeller parameters. high-level, object-oriented framework for the
implementation of programs for the solution of coupled
Symbol INSEAN E1619 transport equations in general, and CFD in particular. The
Number of blades Z 7 discretization of the governing equations is based on the
Finite Volume Method (FVM). The FOI-solver which
Diameter (mm) D 485
have been used in the present study, implements the LES
Hub Diameter Ratio Dh /D 0.226 model by Schumann (1975). The choice of LES model is
not critical for simulation of the flow around the blades,
Pitch at r = 0.7 R P/D 1.15
and in the (very) near wake, since the flow in this region
Chord at 0.75R (mm) C0.75 6.8 does not contain large turbulent regions, but is rather
dominated by structures such as tip vortices and blade
wakes. Wall modeling based on the law-of-the wall is
3 EXPERIMENTAL SET UP AND TEST CONDITIONS applied next to boundaries where the no-slip condition is
Flow velocity was measured by means of a two- applied. The equations are solved on a rotating mesh, i.e.
component-back-scatter-LDV-system, consisting of a 5W the equations are not transformed to a rotating frame of
Argon Laser, a 2-component underwater fiber optic reference. The application of this methodology to
probe, a 40 MHz Bragg cell for the velocity versus propeller flows have been extensively tested before, and it
ambiguity removal and two TSI IFA 655 Doppler is described in detail in Bensow and Liefvendahl (2008)
processors. The probe allowed measuring the axial and and Bensow et al. (2006).
vertical components of velocity simultaneously in the test We use a cylindrical computational domain with the
section fixed frame. In view of the axi-symmetry of the following dimensions.
propeller inflow and the steady conditions of operation,
when the measurement volume is located on the vertical -2.06<x/R<5.15 r/R<3.09
radius, the axial and the vertical components correspond
to the axial and the radial ones in the propeller moving Here R is the propeller radius and x is the axial
frame respectively. Instead, locating the probe in the coordinate, which has the origin in the propeller plane
horizontal radius (y-axis), they represent the axial and the and increases downstream. Mesh generation is a crucial
tangential ones. All the measurements in a circumference practical aspect of propeller simulations. The tip-vortices
were accomplished without moving the LDV volume so and the blade wakes are very sharp flow features which
that the measurement grid in a plane is obtained by only require a fine mesh to be resolved. It is important to
radial movements (Cenedese et al.,1985). distribute the cells in a suitable manner in order to
The LDV measurement volume was traversed with an economize the number of cells, and thereby the
accuracy of about 0.01 mm. Its initial reference position computational cost. The main regions where a high mesh
was fixed by means of a special hub tip, endowed with a resolution is required are around the blades and in the
hole, providing an accuracy of 1 mm along the optical near wake.
axis and 0.5 mm in the perpendicular direction. The
tunnel water was seeded with 10μm Titanium dioxide Table 2. The number of cells in the two meshes - the coarse
particles (TiO2). Phase sampling of the velocity signals and the fine. The first row gives the total number of cells,
was performed using a rotary 3600 pulse/revolution while the second and third respectively give the number of
incremental encoder and a synchronizer; the latter tets and prims. The number N1 indicates the number of
provides the digital signal of the propeller position to the triangular surface elements used to represent a designated
TSI RMR (Rotating Machine Resolver). The propeller blade. The last row shows the number of prismatic
cell layers used on the propeller.
synchronization was given using the detection of the
Doppler signal as trigger condition: each LDV sample, Coarse Fine
tagged with the blade angular position at the measurement Total number of cells 2 568 024 4 466 630
time was arranged inside angular slots and then processed Tetrahedral cells 2 353 466 3 669 091
statistically (see Stella et al., 2000 and Felli et al., 2005 Prismatic cells 214 384 797 055
for further details). Data acquisition was accomplished by N1 8 158 16 766
using a standard PC. Velocity measurements were Prism layers 3 5
performed at the facility speed U∞=1.68 m/s changing the
propeller speed to get the values of the advance ratio Two computational grids have been constructed, with 2.6
J=0.65, J=0.74 and J=0.85. and 4.5 Mcell (1 Mcell = 106 cells) respectively. The
main part of the computational domain is meshed with
tetrahedral (tet) cells and in the boundary layer a number

1
www.opencfd.co.uk/openfoam
of cell layers of prismatic (prism) type are inserted. The manufactured propeller. When carrying out an
mesh around a blade is illustrated in figure 2. In table 2, investigation, these types of differences should be
we summarize the mesh characteristics. minimized but they will always affect any comparison of
It is a relatively common practice, at least for RANS results to some degree.
simulations, to use the rotational symmetry (7-fold in the
present case) of the propeller and only construct the mesh
for a part of the domain which corresponds to one
propeller blade. We note however that a study of the
interaction between consecutive blade wakes, and the
related wake instability, is excluded by the assumption
that the flow field has the same symmetry as the
propeller. The wake instability breaks the 7-fold
symmetry, therefore we have constructed a mesh for the
whole computational domain.

5 RESULT ANALYSIS
5.1 Performance characteristics
Open water propeller performance characteristics were Figure 3. Propeller performance characteristics.
analyzed both experimentally and numerically.
Experiments were carried out in the INSEAN towing tank
N.1 (i.e. 470m length, 13.5m width and 6.5m depth),
using the special test rig for propellers Cussons R46
having 700N maximum thrust, 40Nm maximum torque
and 2500 rpm maximum speed.

Figure 2. Mesh on and around a blade. The surface mesh of

the blade is shown as well as the volume mesh on a plane
through the domain. The boundary layer mesh (with prism
cells) can clearly be seen next to the blade.

The average blade Reynolds number, referenced to the

section at 70% of the radius, was about 230000. CFD
analysis was performed by LES using 2 different mesh
densities with 2.5M cells and 4M cells, at the advance
ratios J=0.74 and J=0.85. Results are documented in
figure 3. It is seen that the numerical simulation well
capture the propeller thrust and torque coefficients at both
the values of the advance ratio. All computed values are
within 5% of the measured values. We note that the
simulation with coarse grid (i.e. 2.5 Mcells) is observed
to be in slightly closer agreement with the experimental
results. We believe however that this discrepancy can be
explained by the unavoidable differences between the
experimental and computational set-ups, which include:
(1) Extent of flow domain. (2) Level and modeling of
inflow turbulence. (3) Experimental procedure to find
Figure 4. Streamwise evolution of the axial velocity
zero thrust. (4) Difference between computational grid
normalized with the facility speed (LDV measurements)
representation of the propeller and the actual,
5.2 Wake evolution acceleration show a dependence on the blade loading
Propeller wake evolution was dealt with considering the conditions. More specifically, the rate of acceleration
distribution of the mean and fluctuating components of Udisk/U∞ (with Udisk the averaged axial velocity over the
the velocity along three cross-sections of the wake at streamtube cross-section) in the first plane is about 60%
x=0.17R, x=0.58R and x=R. The effect of the propeller at J=0.65, 50% at J=0.74 and 40% at J=0.85. Radial and
loading condition on the wake features was investigated tangential gradients of the axial velocity become
through the analysis of the velocity field at the advance smoother more and more streamwise due to the rapid
ratios J=0.65, J=0.74 and J=0.85. turbulent diffusion of the blade wake. The cross flow
distribution is represented by the vector field in figure 5,
limitedly to the section at x=0.17R for space reason. The
cross flow velocity components show marked changes
across the blade wakes. Outside the blade wakes, the
radial flow is globally inward direct as the consequence
of the wake contraction. Indeed, the intensity of such an
inward radial component is reduces streamwise. Within
the blade wake, flow direction undergoes the typical jump
of the radial component due to the passage of the blade
trailing vorticity (Jessup, 1989). This jump, which is
modeled with no thickness according to the potential
theory, is instead widespread over a finite thickness due
to the viscosity of the fluid. An overview of the TKE
distribution is shown in figure 6.

Figure 5. Cross flow at x=0.17R for J=0.65 (top),

J=0.74(mid) and J=0.85 (bottom) (LDV measurements)

In the first plane (i.e. x=0.17R), just downstream the

blade trailing edge, a defect of axial velocity is observed
due to the blade boundary layer and the momentum
transfer from the axial to the cross components (Figure
4). The flow is mainly accelerated elsewhere with a Figure 6. Streamwise evolution of the turbulent kinetic
maximum of the axial velocity on the suction side, at energy (TKE) normalized with the average freestream
about r=0.75R, where the propeller is designed to develop kinetic energy (LDV measurements)
maximum circulation. In the blade-to-blade area, both the
intensities of the velocity defect and the rate of
This progressive bending of the blade wake is well spirals and causing the transition to instability of the
described by the contour plots of the TKE (Figure 6). propeller wake (Felli et al., 2006). The triggering position
Actually, the turbulent wake is an effective indicator for of such a mutual interaction depends strictly on the
the characterization of the trailing wake distortion. This advance ratio, once fixed the propeller geometry, and
distortion occurs with angular displacements between the tends to occur as closer to the propeller disk as the
tip vortex and the inboard blade wake which increase propeller load increases. In the present case, the iso-
streamwise progressively. The general effect of the wake contours of the Turbulent Kinetic Energy (TKE)
deformation is to stretch out the wake, moving the tip document this blade-to-blade interaction, being occurred
vortex further away from the inboard blade wake and in correspondence of the plane at x=R, independently of
possibly suppressing the roll-up. the advance ratio. In this region the trailing wake of the
actual blade shows the tendency to lose the link with its
tip vortex definitely and to be rolled up by the tip vortex
of the previous blade.
Figure 7 describes the effect of the advance ratio and the
streamwise distance on the Power Spectral Density (PSD)
of the velocity signals, in the tip vortex. The power
spectrum of the randomly acquired LDV signals was
estimated using the direct transform method (Gaster and
Roberts,1977). The energy content of the velocity signals
is concentrated at the blade frequency and its multiple-
harmonics mainly. PSD at J=0.65 exhibit the larger
number of resolved harmonics (six harmonics for J=0.65,
four harmonics for J=0.75 and J=0.85). Peak intensity at
the blade frequency reduces at the increasing J and
moving streamwise. The nature of such a phenomenon is
due to the superposition of two effects basically:
- The effect of the turbulent diffusion that smoothes the
velocity peaks in the tip vortex streamwise.
- The mechanism of instability involving a cascade
energy transfer process from the blade to the shaft
harmonic, according to Felli et al. (2006) and Felli et al.
(2008). The occurrence of a modulation at the shaft
frequency in the phase averaged velocity signals at x=R
and J=0.65 (Figure 8) as well as the appearance of some
additional harmonics in the power spectrum seems to
support this statement (Figure 7).

Figure 7. Power Spectral Density (PSD) of the velocity

signals at J=0.65 (top), J=74 (mid) and J=0.85 (bottom)

Figure 8. Angular evolution of the axial velocity in the tip

As the consequence of such a deformation, the wake of vortex at x=0.17R (top) and x=R (bottom), for J=0.65.
the actual blade starts to feel the action of the tip vortex
of the previous one at a certain point, triggering a
mechanism of mutual interaction between consecutive
6 CFD VALIDATION are also supported by the 2D plot of the axial velocity in
In this section we make a qualitative and quantitative figure 10. We see that the location and overall strength of
comparison of the experimental and computational the wake are well predicted, while the flow features are
results, with the aim of investigating and validating the more too "diffused" in the LES results. We also note that
accuracy of the LES results for the relevant flow features. the blade wakes differ slightly between the different
As noted in section 5.1, the computational prediction of blades, for the LES results. At this station this is an
the thrust and torque is in good agreement, within 5% for artificial effect, probably caused by the fact that the grid
all computed values, with the values obtained in the is not exactly symmetric.
experiments. Below, we investigate the velocity field in In general, for the velocity in the wake, the LDV
the near wake. Due to space constraints, we can only measurements give a better resolution and accuracy than
include a small number of the comparisons which have the LES in this study. With an improved mesh, with more
been made. In fact, we restrict ourselves to the first LDV cells, the LES results will naturally become more
measurement plane at x = 0.17R and the intermediate accurate. A principal advantage of the simulations, as
advance number, J = 0.74. compared to the experiments, is that the LES results
contain the velocity/pressure field in the whole
computational domain. Therefore it is possible to analyze
the flow in regions with no optical access, or where it is
difficult to measure for other reasons. Furthermore, since
the three-dimensional velocity distribution is known, it is
possible to perform other types of post-processing. This is
exemplified in figure 11, with an iso-surface of the axial
velocity which illustrates the spatial evolution of flow
structures emanating around the blade tip.

Figure 9. The axial velocity, normalized with U∞, at x =

0.17R. On the horizontal axis is the normalized radius
r/R. The radial line is chosen at an angle so that it passes
in the middle between two blade wakes.

Figure 11. (J=0.74, LES) Iso-surface of the axial velocity

Figure 10. Experimental and computational analysis of the and contour lines of the axial velocity on the plane x = 0.17R.
propeller wake at x=0.17R for J=0.74. The axial velocity
normalized by U∞, is shown. 7 DISCUSSION AND CONCLUSIONS
The problem of the wake evolution and instability for a
In figure 9, we show a comparison of the axial velocity submarine propeller, running in open water conditions,
along a radial line. It is seen that the shape and width of was investigated in the present research activity. The
the wake are well predicted. The maximum axial study concerned velocity measurements by LDV phase
acceleration is slightly closer to the center line than r/R = sampling techniques and numerical simulation of the
0.7, which is the location where the blades are designed velocity/pressure field by LES.
to give maximum circulation. The experiments indicate a On the physical side, the typical features of the propeller
higher axial velocity in the inner region of the wake. For wake in the near field (e.g. streamtube contraction, roll
other line plots, cutting through areas with large up) as well as the effect of the loading conditions on the
gradients, the simulations generally give smoother curves wake deformation and on the mechanisms of turbulent
than the experiments, and are not able to capture the diffusion and instability inception were highlighted. In
sharpness of the flow structures. The location of the flow particular, it was observed that the transition to instability
structures is however well predicted. These conclusions tends to occur closer to the propeller disk as to a
conventional propeller with a lower number of blades, the Cenedese A., Accardo L., Milone, R. (1985). Phase
spiral-to-spiral distance being smaller. Furthermore, the sampling in the analysis of a propeller wake,
occurrence of some additional harmonics in the power International Conference on Laser Anemometry
spectra of the velocity signals at the most downstream Advances and Application, Manchester, UK.
plane supports the results of literature concerning the
Di Felice F, Di Florio D, Felli M, Romano G.P. (2004).
transfers of energy in the transition wake (see Felli et al.
Experimental Investigation of the Propeller Wake at
2006, Felli et al, 2008). In this regard, additional
Different Loading Conditions by Particle Image
measurements focused on the analysis of the mechanisms
Velocimetry. Journal of Ship Research.Vol.48, N.2,
of evolution of the tip vortex in the transition wake, as in
168-190.
Felli et al. (2008), are necessary to find out the nature of
Felli M., Di Felice F. (2005). Propeller wake analysis in
such additional harmonics.
non uniform inflow by LDV phase sampling
For propeller simulations, it is possible to use software
techniques, Journal of Marine Science and
based on the boundary element method (BEM), the
Technology, Vol.10, N.4, pp.159-172.
Reynolds averaged Navier-Stokes equations (RANS) and
LES. The main motivation for the use of LES is for the Felli M., Di Felice F., Guj G., Camussi R. (2006).
investigation of large scale unsteady (in the rotating Analysis of the propeller wake by pressure and
reference frame) effects. Examples of this include the velocity phase measurements, Experiments in Fluids,
following four situations. (1) Simulation of cavitation, see N.1.
Bensow et al. (2008); (2) The propeller in crashback Felli M., Di Felice F. (2008). Characterization and
conditions; see e.g. Jang and Mahesh (2008); (3) The modulation of the acoustical signature of a marine
propeller in a non-uniform inflow; (4) The study of wake propeller operating in open water and in behind
instability, as in the present paper. As noted in section 6 condition, Exponaval 2008, Vina De Mar, Chile.
however, the mesh resolution and distribution is not Felli M., Guj G., Camussi R. (2008). Effect of the number
sufficient to investigate the wake structure, where the of blades on propeller wake evolution, Experiments in
instability sets in, with sufficient accuracy. With a Fluids, N.44.
suitable distribution of computational cells, this may be Gaster M., Roberts J.B. (1977). The spectral analysis of
accomplished with 10-20 million cells. This estimate is randomly sampled records by a direct transform,
based on the experience gained from the computations Proceedings Royal Society London, A 354, pp.27-58.
presented in this paper. In general, submarine propellers
put more severe requirements on the computational mesh Jang, H. and Mahesh, K. (2008). Large Eddy Simulation
than conventional propellers. The main reason for this is of Ducted Propulsors in Crashback. In 27th
the design which results in weaker flow structures, which Symposium on Naval Hydrodynamics, Seoul, Korea.
in turn are more easily dissipated artificially when the Jessup S. (1989). An experimental investigation of
mesh is too coarse. Submarine propellers also have more viscous aspects of propeller blade flow, The Catholic
blades, and therefore require the corresponding increase University of America, Washinngton DC.
in number of cells in the mesh, to resolve the geometric Kerwin J.E. (1986). Marine propellers, Annual review of
detail. fluid mechanics, Vol.18, pp.367-403.
Comparisons between flow field LES predictions and
LDV measurements revealed a good agreement Schumann, U. (1975). Subgrid scale model for finite
concerning both thrust and torque coefficients and difference simulation of turbulent flows in plane
location and general wake strength. The sharp flow channels and annuli. J.Comp.Phys., 18(4):376–404.
features, including blade wakes and tip vortices, are
however dissipated artificially and a finer mesh is
required to study the wake dynamics, and the associated
wake instability.

REFERENCES
Bensow, R. E. and Liefvendahl, M. (2008). Implicit and
explicit subgrid modeling in les applied to a marine
propeller.(AIAA-2008-4144).
Bensow, R. E., Huuva, T., Bark, G., and Liefvendahl, M.
(2008). Large Eddy Simulation of Cavitating
Propeller Flows. In 27th Symposium on Naval
Hydrodynamics, Seoul, Korea.
Bensow, R. E., Liefvendahl, M., and Wikström, N.
(2006). Propeller Near Wake Analysis Using LES
With a Rotating Mesh. In 26th Symposium on Naval
Hydrodynamics, Rome, Italy.