Ocean Engineering xxx (2017) 1–20

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Design of ducted propeller nozzles through a RANSE-based

optimization approach
Stefano Gaggero a, *, Diego Villa a, Giorgio Tani a, Michele Viviani a, Daniele Bertetta b
a
University of Genoa, Department of Electrical, Electronic, Telecommunications Engineering and Naval Architecture, Via Montallegro, 1, 16145 Genoa, Italy
b
Fincantieri S.p.A. Naval Vessel Business Unit, Naval Architecture Dept. (MM-ARC), Genoa, Italy

A R T I C L E I N F O A B S T R A C T

Keywords: Marine propellers design requirements are always more pressing and the application of unusual propulsive
Ducted propellers configurations, like ducted propellers with decelerating nozzles, may represent a valuable alternative to fulfill
Accelerating nozzles stringent design constraints. Accelerating duct configurations were realized mainly to increase the propeller ef-
Decelerating nozzles ficiency in the case of highly-loaded functioning. The use of decelerating nozzles sustains the postponing of the
Optimization cavitating phenomena that, in turn, reflects into a reduction of vibrations and radiated noise. The design of
RANSE
decelerating nozzle, unfortunately, is still challenging. The complex interaction between the propeller and the
OpenFOAM
nozzle, both in terms of global flow feature and local (tip located) phenomena, is not yet fully understood. No
extensive systematic series, as in the case of accelerating configurations, are available and the design still relies on
few measurements and data. On the other hand, viscous flow solvers appear as reliable and accurate tools for the
prediction of complex flow fields and their application for the calculation of ducted propeller performance and
nozzle flow was almost successful. Hence, using CFD as a part of a design procedure based on optimization, by
combining a parametric description of the geometry, a RANSE solver (OpenFOAM) and a genetic type algorithm
(the modeFrontier optimization environment), is the obvious step towards an even more reliable ducted propeller
design. An actuator disk model is adopted to include efficiently the influence of the propeller on the flow around
the duct; this allows avoiding the weighting of the computational effort that is necessary for the calculations of the
thousands of geometries needed for the indirect design by optimization. Design improvements, in model scale, are
measured by comparing, by means of dedicated fully resolved RANSE calculations, the performance of the
optimized geometries with those of conventional shapes available in literature. For both nozzle typologies,
dedicated shapes reducing the risk of cavitation and increasing the delivered thrust are obtained, showing the
opportunity of customized nozzle design out of usual systematic series. In addition, by analyzing the results of the
optimization histories, appropriate design criteria are derived for both accelerating and decelerating nozzle
shapes.

1. Introduction
Marine propellers design requirements are, nowadays, always more
pressing. Not only maximum efficiency but also comfort and environ-
mental demands and regulations have to be satisfied. The application of
unusual propulsive configurations may represent a valuable alternative
to fulfill these constraints. Contra-rotating and tip loaded blades (CLT
and Kappel like geometries) were used to improve propulsive efficiency,
together with Energy Saving Devices, like pre- and post-swirl stators,
PBCF or Mewis Ducts. Combinations of these devices mounted on PODs
(i.e. through the EU project TRIPOD, S
anchez-Caja et al., 2013) were
exploited to maximize the effects and to provide flexible propulsive
systems. Their ability to reduce side effects like induced pressure pulses
and vibrations by operating the propellers in more uniform inflow, with
minimum optimal diameters and far from the hull, has been proven over
the years.
Ducted propellers represent an additional, valid, answer to current
design requirements. In the case of vessels for which requirement of high
thrust is critical in the low-speed operation range, or when the screw is
limited in diameter, the use of accelerating nozzles is widely documented
in literature. With accelerating nozzles, the duct increases the flow rate
through the propeller, which consequently operates at a more favorable
loading. The nozzle by itself produces a certain positive thrust.
This conceptual idea dates back to Stipa (1932) and Kort (1934) but

* Corresponding author.
E-mail address: stefano.gaggero@unige.it (S. Gaggero).

https://doi.org/10.1016/j.oceaneng.2017.09.037
Received 14 February 2017; Received in revised form 21 July 2017; Accepted 24 September 2017
Available online xxxx
0029-8018/© 2017 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Gaggero, S., et al., Design of ducted propeller nozzles through a RANSE-based optimization approach, Ocean
Engineering (2017), https://doi.org/10.1016/j.oceaneng.2017.09.037
S. Gaggero et al.
probably the works by Van Manen (1954, 1957, 1962), Van Manen and
Superina (1959) and Van Manen and Oosterveld (1966) represent some
of the most accurate reviews of ducted propellers performance and
design guidelines. By combining even simple theoretical considerations
and systematic measurements, they provided a rather wide overview of
the influence of some geometrical parameters of nozzles on efficiency,
correlating non-dimensional parameters and performance in case of both
highly- and lightly-loaded propellers. They identified the nozzle profile
number 19A as the optimal compromise, having performance in towing
and pushing conditions not appreciably inferior to those of considerably
longer nozzles. This shape, fifty years later, is still the default choice in
the case of accelerating ducted propellers applied to a variety of boats
and vessels (tugboats, towboats, and supply vessels) which require
increased propulsive efficiency, especially near the bollard
pull condition.
In the same works, also the potentialities of flow decelerating type of
nozzle were pointed out. By reducing the flow rate to the impeller, a local
increase of the static pressure is achieved: this may be effective in
retardation of propeller cavitation phenomena at a cost, unfortunately, of
efficiency reduction and of an additional drag represented by the nega-
tive thrust delivered by the duct itself. An improvement of the cavitation
characteristics of the propeller can be obtained, thus, only if the gain in
static pressure exceeds the unfavorable effect of the increased screw
loading, necessary to compensate the duct drag. In addition, the risk of
duct cavitation (at leading edge or midchord, depending on the propeller
loading and the camber of the nozzle) represents another issue, which
has to be accounted for when evaluating the usefulness of the deceler-
ating configuration.
The design of decelerating nozzles consequently seems to better
comply with the constraints circa radiated noise and vibrations strictly
related to cavitation suppression. However, it is still challenging and
poorly addressed in the literature. Despite the wide employment of this
propeller configuration (i.e. fast carriers and navy vessels for which side
effects reduction is significantly more important than propulsive effi-
ciency), there are several design and analysis issues still open, starting
from the downsides previously evidenced. The complex interaction be-
tween the propeller and the nozzle, both in terms of global flow feature
and local, tip located, phenomena, is not yet fully understood. No sys-
tematic series, widely diffused in the case of accelerating configurations,
are available; thus, the design still relies on few measurements and data.
Oosterveld (1970) carried out some measurements with decelerating
ducts where the influence of some geometrical features of the duct was
investigated. The analyses were however limited to a single propeller
geometry (a finite-chord Kaplan propeller, model K5-100) operating in-
side the duct. Only qualitative trends can thus be derived for the appli-
cation of those nozzles in the cases of unloaded (zero chord at the tip)
propellers. From a theoretical/numerical point of view, the availability of
literature data is even scarcer. Apart classical momentum theories
(Oosterveld, 1970; Van Manen and Oosterveld, 1966) applied to decel-
erating duct geometries, only a few applications are available. Abdel--
Maksoud et al. (2010) developed a design and analysis (Yu et al., 2013)
method for multi-component propulsors. Çelik et al., 2011 investigated
the optimum duct geometry for passenger ships. Gaggero et al. (2012,
2013) developed a design approach based on classical lifting line/lifting
surface approaches further refined by using Boundary Element Methods
and optimization algorithms to improve the performance of a propeller
surrounded by a decelerating nozzle. Results, confirmed by dedicated
RANSE calculations and model scale measurements at the cavitation
tunnel, concerned however only the well-established problem of pro-
peller blades design, using given nozzle shapes. The inclusion of the
nozzle shape in the optimization process, indeed, would have been
possible. However, it would have excessively stressed the applicability of
BEM over its inherent limitation (thick geometries, subjected to flow
separation), leading to nozzle designs significantly conditioned by the
scarce reliability of the computational approach in dealing with complex
viscous flows.
2
Ocean Engineering xxx (2017) 1–20
Recently Bontempo et al. (2014, 2016) proposed, again in the
framework of potential flow based theories, some analyses of deceler-
ating duct shapes with momentum theory and semi-analytical actuator
disk models to account for radial distributions of forces, including
tangential ones, and derive relationships between propeller load and
nozzle performance. All these calculations, however, neglect again the
role of viscosity. This, instead, significantly influences duct and propeller
forces by interacting with the flow at the duct/propeller tip gap giving
rise to typical tip leakage vortexes, as demonstrated by Gaggero et al.
(2013), Haimov et al. (2011), S
anchez-Caja et al. (2008),
Abdek-Maksoud and Heinke (2002), Hoekstra (2006), Baltazar et al.
(2013) also in the simpler case of accelerating nozzles.
In the context of accelerating duct shapes, instead, significant de-
velopments and applications can be recognized. Since the work of Kerwin
et al. (1987) and Hughes (1993, 1997), several models have been pro-
posed to deal partially with the design and particularly to the analysis of
accelerating ducted propellers. Boundary Element Methods were pro-
posed to efficiently account for vortical wake alignment (Baltazar et al.,
2015; Kinnas et al., 2012a), influence of duct boundary layer on tip
propeller loading (Baltazar et al., 2012) or presence of the blunt trailing
edge (Kinnas et al., 2009, 2012b; Baltazar et al., 2015), typical of
accelerating configurations but critical for the application of BEM. Good
predictions of propulsive forces were achieved using RANSE, which
encouraged direct full-scale applications or the definition, through nu-
merical analyses, of scaling rules (Bhattacharyya et al., 2016a, 2016b).
Simultaneously the combined use of RANSE and simplified propeller
models (i.e. actuator disks) highlighted the usefulness of such calcula-
tions for a preliminary design of the duct. Zondervan et al. (2006),
Hoekstra (2006), Tamura et al. (2010), for instance, successfully applied
an actuator disk to investigate the influence of the propeller on the flow
around the nozzle and to derive simple correlations circa induced ve-
locities useful in the design stage. Kinnas et al. (2013) and Bosschers et al.
(2015) further extended the computational model by including the pro-
peller action through distributed momentum sources (i.e. body forces) on
Euler or RANSE solvers (as done for self-propulsion analyses by Villa
et al., 2011). This inclusion efficiently solved the interactions (separation
of the flow on the inner duct surface or thickening of the viscous
boundary layer) between propeller and duct improving, at a relatively
low computational cost, the reliability of both potential based (for the
propeller) and viscous based (for the duct) solvers. Only propellers inside
Nozzle 19A were, however, analyzed without any attempt to improve the
overall propulsive performance by modifying the duct shape.
Considering the need of a reliable and efficient design tool for both
accelerating and decelerating propeller nozzles, the scarcity of data and
design guidelines, and the satisfactory results achieved by numerical
analyses, in the present work optimization is exploited to investigate the
possible enhancement of nozzles performance provided by the applica-
tion of high-fidelity numerical tools embedded into an automatic process.
Thousands of different geometries, defined by a parametric description,
need to be tested, using viscous approaches to account for the relevant
phenomena characterizing “thick” ducts and verified against appropriate
design criteria (specifically derived for both accelerating and deceler-
ating configurations). This leads to the definition of an optimal geometry
that, due to the multi-objective nature of design, is, of course, a balance of
contrasting purposes. Only thank to the favorable ratio between accuracy
and computational efficiency of simplified representations of the influ-
ence of the propeller on the nozzle performance, such the ones repre-
sented by actuator disks, the process is possible in a reasonable
computational time compatible with the usual design routine. Actuator
disk models, already proposed with success for the analysis of acceler-
ating nozzles performance, are at first validated on the basis of the
measurements carried out at the cavitation tunnel of the University of
Genoa. Then, they are employed also for the analysis of decelerating
ducts and, through optimization, for the improvement of the perfor-
mance of reference nozzle geometries. The first purpose of this paper is to
demonstrate the reliability of the proposed design approach. Then, by
S. Gaggero et al.
using the amount of numerical data provided by the systematic analyses
inherent to any optimization process, it is possible to identify correlations
and trends among parameters and performance, providing useful
guidelines in the peculiar case of decelerating configurations.
The successful application of the design by optimization of model
scale nozzles by using the simpler actuator disk model would represent a
significant step towards a more reliable design approach. Once combined
with any potential based design approach for the propeller (a simpler
lifting line or a BEM, as proposed in Gaggero et al., 2012, 2013) the
hybrid BEM/RANSE would extend the reliability of current geometry-
optimization algorithms for the design of the entire propulsive system
directly in full scale: RANSE methods would account for flow separation
phenomena on the “thick” duct, whereas BEM will be used for prediction
of forces on the “thin” propeller blades where separation is less probable.
2. Designs of ducts through optimization
The design of propeller nozzles, despite the promising results ach-
ieved with numerical calculations, is mainly based on systematic series.
Van Manen (1962), Van Manen and Oosterveld (1966) derived simplified
correlations and identified key points for both accelerating and decel-
erating shapes considering efficiency, the risk of flow separation and
critical cavitation indexes for both the propellers and the surrounding
nozzles. In the case of accelerating duct geometries, a substantial increase
of efficiency was obtained at higher screw loading with relatively long
nozzles that instead have to be the shortest possible in lightly loaded
conditions to limit the nozzle drag which is substantial in reducing the
total propulsive efficiency. Similar analyses were proposed to investigate
the influence of maximum thickness and maximum camber over length
ratio and equivalent nose tail angle on the pressure on the outer duct
surface, controlling flow separation as well. Guidelines for decelerating
geometries were even scarcer. A favorable increase of the static pressure
at the impeller was evidenced only at low propeller thrust coefficients
with shorter nozzle length to minimize drag loss and the consequent
reduction of efficiency.
These key points turn, naturally, into the opposite objectives of a
design through numerical optimization. For an accelerating nozzle,
which is usually designed to operate close to bollard pull, the maximi-
zation of the duct thrust at a given propeller loading conflicts with the
need of reasonably high values of inner static pressure to contrast cavi-
tation inception. Most of the thrust delivered by the duct depends by the
suction on the inner surface. In bollard pull conditions, a certain risk of
cavitation could be accepted (until the propeller does not suffer from
excessive thrust breakdown). In the case of functioning at higher advance
coefficients, a geometry that allows simultaneously to maximize the duct
thrust and to obtain reasonable values of static pressure against cavita-
tion inception would be preferred. On the contrary, for a decelerating
duct propeller operating at relatively high advance coefficients, maxi-
mization of the static pressure with minimum increased duct drag turns
into the design objective. Avoiding simultaneously the risk of cavitation
on the outer surface of the nozzle in very critical working conditions has
to be considered as well. In parallel, the risk of flow separation has to be
accounted. This has been monitored experimentally and addressed in the
traditional and simplified design momentum theories with criteria on the
maximum sectional lift coefficient of the nozzle profile (Van Manen
(1962), Van Manen and Oosterveld (1966)). The analysis of the duct
forces via viscous calculations inherently includes the possible detri-
mental effects of flow separation on duct performance.
A real design would require accounting for the interactions taking
place between the duct and the propeller to exploit the maximum from
both the designs: a nozzle providing the maximum thrust and the highest
possible level of inner static pressure and a propeller optimized (effi-
ciency, thrust and cavitation, depending on the application) for that
resulting flow. Such design is, of course, possible at a considerable
computational cost by analyzing the propulsive system as a whole.
However, it could take advantage of preliminary, independent, designs of
3
Ocean Engineering xxx (2017) 1–20
nozzles and propellers by themselves: an optimal duct (depending on its
accelerating or decelerating nature) to be adopted for the iterative defi-
nition of an optimal propeller for those given conditions of flow. The use
of actuator disk models (as proposed in this work) to include the pro-
peller into the calculations of the duct performance, indeed, does not
permit any type of interactions on propeller forces being these ones
prescribed and independent by the performance (and the resulting
induced inflow) of the nozzle. Only by a blade redesign, it would be
possible to have a sort of feedback and to design the entire propulsive
system for a given total thrust. Similarly, by using a lifting line or a more
accurate BEM nested into the design loop, it would be possible to include
the changes of the propeller loading distribution due to the different
inflow of a modified nozzle shape in the optimization process of the
nozzle. This approach, of course, would still not consider the constraint
on the total delivered thrust. In both cases, being duct and propeller
performances dependent on each other (loading distribution changes
inflow which, in turn, changes loading distribution), additional inner
iterations would be required for any given nozzle shape. By assuming an
unchanged propeller thrust (i.e. unchanged actuator disk intensity),
instead, the analysis of the optimal geometries has to be restricted to
those nozzle shapes that delivering the thrust of the reference geometry
(i.e. not changing the working condition), improve performance in terms
of inner static pressure. This inherently implies that also nozzle induced
velocities to the propeller are assumed not to be reasonably affected by a
modified geometry that, having to provide the same thrust of the refer-
ence shape, should also be significantly similar in terms of induction. Any
other case (i.e. an increase of the duct thrust at unchanged inner static
pressure or heavily changed velocity field to the propeller) has to be
regarded as the first step towards the redesign of propeller blades. In this
case, the propeller design will take advantage of the unloading which is
allowed by the higher thrust delivered by the newly designed duct.
A list of possible design objectives that could be exploited consists,
consequently, in:

For accelerating duct shapes, conceived for highly loaded
functioning:
○ Maximization of the nozzle thrust close to bollard pull condition
○ Minimization of the nozzle drag at higher advance coefficients,
when the ducted propeller provides the propulsive thrust for other
operating conditions of the ship (e.g. cruising speed for tugs; other
possible “mixed conditions” like those typical of a take home
propulsor, which needs higher thrust density than a conventional
one, with the necessity to operate at a certain speed)
○ Maximization of the inner static pressure in any operating
conditions
○ Avoidance of nozzle cavitation

For decelerating duct shapes:
○ Minimization of the nozzle drag at the design condition
○ Maximization of the inner static pressure at the design condition
○ Avoidance of nozzle cavitation
A design by optimization, definitely, can be explained as an automatic
“trial and error” procedure that, by analyzing hundreds (thousands) of
different geometrical configurations, iteratively changes their shape to
achieve a Pareto convergence of the designs towards the satisfaction of
the outlined objectives. For this purpose, an efficient tool for the char-
acterization of flow features and nozzle performance by including the
propeller action in the simplified form of an actuator disk (Sections 3.1
and 3.2) is necessary. A parametric description of the geometry to
manage automatically shape variations in accordance with given con-
straints (Section 3.3) is required as well. The need to handle opposite
objectives requires the use of multi-objective optimization algorithms
that, in present calculations, are of genetic type. By exploiting the anal-
ogy of natural selection, subsequent generations are created by
combining the best characteristics of each pre-computed combinations
via crossover and mutation. These algorithms, moreover, allow to
S. Gaggero et al.
avoiding local minima, with respect to which computationally efficient
gradient-based methods are more prone, by the inclusion of a certain
randomness in the selection of the characters of any subsequent gener-
ations allowing more unrestrained analyses. This feature was considered
important for the analysis of trends and of the influence of combinations
of parameters. The entire process has been coded into the modeFrontier
(Esteco, 2016) optimization environment.
3. Numerical methods
3.1. Flow solver and computational domain
Detailed analyses of duct performance cannot be achieved by simply
using inviscid flow models. Flow separation, except by using heavily
approximated formulations, is beyond potential flow calculations. Blunt
trailing edges, in the particular case of accelerating ducts, pose significant
difficulties for the application of the Kutta condition. Local modification
of the duct shape (Bosschers et al., 2015) or inclusion of transpiration
sources have been proposed but the necessary calibration of such models
is far from the accuracy needed even in comparative analyses as those
represented by design by optimization. Viscous models are definitely
needed and RANSE approximations with suitable turbulence models can
adequately answer all the specific issues related to design procedures.
In this study, OpenFOAM (Weller et al., 1998; the OpenFOAM
Foundation, 2017) has been employed. OpenFOAM is a collection of li-
braries suited for the solution of partial differential equations using a
Finite Volume approach and cell-centered collocated variables with
face-based implementation to allow for arbitrary cell shapes.
Pre-processing, solvers, and post-processing tools are fully scriptable,
which is of fundamental importance when a completely automatic pro-
cess has to be setup.
For this particular application, where steady flow with respect to a
blade fixed reference frame is expected, the simpleFOAM solver with the
SST k-ω turbulence model (Menter et al., 2003) has been used. Continuity
and RANSE equations are solved in a segregated form, with a SIMPLE
based pressure-velocity correction and under-relaxation to achieve
steady state solutions. Second order accurate schemes in space have been
preferred for momentum and turbulence equations.
The use of actuator disks instead of actual propeller blades results in a
fully axisymmetric flow solution. Therefore, the computational domain
can be reduced to a blade sector smaller than that of a blade passage.
In OpenFOAM, there are two possible formulations to exploit the
rotational symmetry, which however condition the type of flow that can
be modeled. The first consists in a 3D angular sector of sufficient
amplitude (Fig. 1a), arranged with periodic (cyclic) interfaces allowing
the inclusion of the propeller slipstream rotation by means of momentum
sources representative of the forces acting on the propeller blade in the
tangential direction. Periodicity requires the use of a certain number of
cells in the tangential direction (as in Section 3.2) to take properly into
account the tangential component of the momentum. The second
Fig. 1. Computational domains for the analysis of ducted propeller performance. (a) Cyclic type boundary conditions. (b) Wedge type boundary conditions.
4
Ocean Engineering xxx (2017) 1–20
formulation exploits “wedge type” boundary conditions for purely
axisymmetric calculations. For 2-dimensional axisymmetric cases, the
geometry is specified as a wedge of small angle (<5 ) with a thickness of
1 cell, running along the plane of symmetry, straddling one of the co-
ordinate planes, as shown in Fig. 1b. With this arrangement, only mo-
mentum sources in the axial directions are allowed and all the flow
features different from those on the meridian plane (i.e. slipstream
rotation) are neglected. The choice of the most appropriate computa-
tional domain depends on the influence of the slipstream rotation on duct
forces. The analyses by Hoekstra (2006) showed a certain importance of
slipstream which was calculated using a formulation able to account for
the tangential effect also by using only an equivalent two-dimensional
meridian plane. In the present case, the need of computationally effi-
cient calculations would encourage, on the contrary, the use of the effi-
cient setup consisting in the 2-dimensional wedge formulation, thus
neglecting the 3-dimensional nature of the flow. The preliminary analysis
proposed for the reference geometries in Section 4 would definitely
clarify the influence of slipstream in the case of the calculations carried
with OpenFOAM, allowing for the selection of the most appropriate
setup. As proposed by Hoekstra (2006), also the propeller hub has been
included in the calculations. Its presence, in fact, has a certain positive
influence on the thrust delivered by the duct: momentum sources are
distributed over a small area, inducing higher velocities that change the
angle of attack of the nozzle and the resulting pressure distribution.
Details of the hub shape are also important. The use of semi-infinite (aft
or forward) rather than infinitely long shafts has a different influence on
flow streamlines affecting the performance of the nozzle. In present
calculations, the focus is on the definition of a procedure for nozzles
design rather than on the exact characterization of the flow. Conse-
quently, the simplest adoption of an infinitely long shaft was deemed
sufficient to account at least for its principal influence. Similarly, a slip
boundary condition, without the need of appropriate prism layers, was
considered sufficient to catch its influence on the flow to the nozzle. For
both cyclic and wedge calculations, anyhow, the computational domain
consists of an angular sector (25 when periodic boundary conditions are
employed, 5 in the case of the “wedge” formulation) with the inlet
placed 2 propeller diameters in front of the nozzle and the outlet 5
propeller diameters aft the nozzle. The external boundary lies on a cy-
lindrical surface 4 propeller diameters from the propeller shaft. A sum-
mary of the boundary conditions for both the configurations is reported
in Table 1.
3.2. Propeller model and grid arrangement
The influence of the propeller on the flow around nozzles has been
modeled by using an actuator disk. The actuator disk can be seen, defi-
nitely, as an infinite bladed propeller. The load of the propeller is
distributed over the entire area of the propeller, neglecting the real
material nature of a finite number of blades. Even if more detailed rep-
resentations of blade forces are available (Bosschers et al., 2015), by
S. Gaggero et al.
Table 1
Summary of boundary conditions.
Patch Velocity Pressure
∂p
Inlet Fixed
∂n ¼0
(depending on J)
∂U p¼0
∂n ¼ 0
Outlet
∂U p¼0
External ∂n¼0
∂p
Nozzle No-slip
∂n ¼0
Shaft/hub Slip Slip
Interfaces Wedge/Cyclic Wedge/Cyclic
using local momentum sources, their application for the characterization
of flow in the nozzle design process seems redundant. Only the radial
distribution of load is, in fact, really influential. The tip loading has a
positive influence on duct forces; this is largely confirmed by the pref-
erence, in the case of accelerating nozzle shape, of Kaplan-like propeller
geometries, with a finite chord at tip fostering the development of higher
forces at outer radii. Unloaded tips introduce a substantial risk of flow
separation on the inner surface of the duct (which can be further enlarged
by the simplified model represented by the actuator disk). This effect can
be mitigated by dedicated nozzle shapes, as those that for instance may
be derived from an integrated design approach as the one proposed in
this paper.
Hoekstra (2006) proposed a simple three-parameter model to
describe the propeller loading and to allow for variations of the radial
position of the maximum thrust. Similar approaches are currently
available in many commercial CFD software and often consider radial
variations of the loading in accordance with Goldstein optimum criteria
(Goldstein, 1929).
In present calculations, instead, the radial distribution of forces (axial
and tangential, depending on the type of simulation and on the compu-
tational domain, as in Section 4) have been derived from Boundary
Element Method calculations. They are subsequently loaded into RANSE
by using interpolating tables and polynomials through a dedicated li-
brary specifically developed to extend the computational functionalities
of OpenFOAM. For both the accelerating and the decelerating cases, the
optimization has been carried out starting from reference geometries
(Gaggero et al., 2012) designed and tested at the cavitation tunnel.
Exactly these geometries were used to obtain a numerical estimation of
the loading distributions and to define the design points in correspon-
dence to which the design of the new nozzles has been carried out. The
same set of data served also for the preliminary validation of the
computational model to be employed in the optimization process.
The proposed actuator disk is a finite thickness disk set equal to the
average longitudinal extension “cut out” by the propeller (Fig. 2). No
investigations circa the influence of the disk thickness have been carried
out and the simplest possible model has been used in the light of
computational efficiency. As discussed by Hoekstra (2006), also the
chord-wise load distribution (i.e. the distribution of forces in the axial
Fig. 2. Actuator disk representations into the computational domain (cyclic or wedge). Colors represent the strength of the momentum sources per unit volume. (a) Cyclic type boundary
conditions. (b) Wedge type boundary conditions. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
5
Ocean Engineering xxx (2017) 1–20
ω k
Fixed (depending on U), Fixed (depending on U),
1% turbulence intensity 1% turbulence intensity
∂ω ∂k
∂n ¼ 0 ∂n ¼ 0
∂ω ∂k
¼0
∂n ¼0
∂n
Wall Function Wall Function
Slip Slip
Wedge/Cyclic Wedge/Cyclic
direction) could have a certain influence on duct performance. Blade
forces are largely due to suction on the back side, while the pressure jump
due to a momentum source is evenly distributed between over- and
under-pressure. The unsymmetrical distribution of pressure exerted by
the propeller blades could have been taken into considerations by
employing a sequence of actuator disks of different magnitude aiming at
reproducing a sort of chord-wise distribution of load in order to partially
address the uneven pressure distribution between sides of the blade. This
arrangement, definitely needed when the interactions between pressure
fields are of primary importance (i.e. for the evaluation of the thrust
deduction during self-propulsion calculations) has, however, a certain
computational cost that seems unnecessary in this phase.
Grids were arranged accordingly to the computational domain
selected for the analyses. A structured multi-block layout has been
preferred to describe in details the most important features of the noz-
zles. In the case of accelerating duct configuration, essentially an O-grid
mesh has been arranged to have better resolution at the blunt trailing
edge. C-type grids were adopted, instead, in the case of decelerating
geometries to manage the sharp trailing edge of the nozzle. An example
of the meshes is shown in Fig. 3. Both the arrangements for cyclic and
wedge type boundary conditions are presented. The meridian sections
show clearly the multi-block nature of the meshes, which were arranged
with the OpenFOAM blockMesh utility. Meshes, in the case of the wedge
arrangement, consist of a one cell thick layer of structured hexahedral
elements with appropriate grading to comply with wall treatment
modeling through wall functions. In the case of 3-dimensional calcula-
tions with periodic boundary conditions, exactly the same cell arrange-
ment is used and the angular amplitude of the sector is filled with 25
layers of cells.
The influence of the grid resolution in the case of the wedge type
boundary condition (that, as highlighted in Section 4, will be used for the
optimization activity) was investigated. In order to carry out this inves-
tigation, a reference computational mesh (i.e. Medium) and two other
meshes, one with reduced (Coarse) and one with increased resolution
(Fine) arranged using a refinement factor of 2 have been considered.
Richardson extrapolation was applied to derive, in principle, extrapo-
lated values at zero grid spacing from a series of lower-order discrete
values. Values of the nozzle thrust coefficient
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Fig. 3. Multi-block Structured mesh arrangements around typical accelerating and decelerating nozzles. (a) Accelerating type nozzle, cyclic b.c. (b) Decelerating type nozzle, cyclic b.c. (c)
Accelerating type nozzle, meridian plane, wedge b.c. (d) Decelerating type nozzle, meridian plane, wedge b.c.

(CT Nozzle ¼ TNozzle =0:5ρπR2prop: V 2 ) were analyzed at equivalent propeller
advance coefficients of J ¼ 0.8 (accelerating duct) and of J ¼ 103.4% of
Jref. (decelerating duct). For both geometries, the convergence trend is
particularly good. Especially for the accelerating duct configuration, the
medium mesh arrangement allows for negligible errors with respect to
the extrapolated value (Table 2). Also for the decelerating configuration,
the difference is well below 1%. Particularly low values of the Grid
Convergence Index (Roache, 1994), in addition, justify the use of the
Medium mesh arrangement in the light of computationally efficient and
reliable calculations throughout the design by optimization. Finally, as
per Fig. 4, also the iterative convergence of both analyses with this grid
setup is good and consequently 800 iterations were selected for all the
calculations of the optimization process.
3.3. Nozzles parametric description
The parametric description of nozzles is rather simple. Nozzles are
definitely hydrofoils that can be handled, analogously to what already
done in the case of propellers (Gaggero et al., 2016a, 2016b) and wings
(Olivucci and Gaggero, 2016), by using B-Spline curves for both camber
Table 2
Discretization errors for the nozzles. Wedge type boundary condition.
Cells Accelerating Nozzle Decelerating Nozzle
CT Nozzle CT Nozzle/CT ref.
and thickness distributions. In present calculations, however, a slightly
different approach has been adopted. In order to avoid unfeasible shapes
(i.e. maintain the inner duct surface in correspondence to the propeller
on a cylindrical surface of a given radius) it was preferred to describe
with B-Spline curves directly the inner and the outer contour of the
nozzle to facilitate the application of the geometrical constraints (Ver-
nengo et al., 2016). The control points defining the B-Spline curves, as
shown in Fig. 5, are free to move with some “compatibility” constraints.
At leading edge (and at trailing edge, in the case of the sharp closure of
decelerating nozzles) the control points of back and face sides have to
move as one, and only in the radial direction. The control points that
handle the central part of the inner surface have to be altered, instead,
only axially to keep it rectilinear at propeller location.
With these assumptions, the description of the accelerating duct
consists of 11 free parameters while the decelerating nozzle has 7 degrees
of freedom. In both cases, a scaling factor in the axial direction is added to
change anisotropically the length/diameter ratio. The ranges of variation
of the parameters are summarized in Table 3. Choices, also based on
previous experiences, are motivated by the need of a trade-off between
allowed modifications (the widest possible) and fairness of the geometry,
which, in turn, is facilitated in any case by the use of B-Spline curves.
However, for design purposes this description is not intuitive and usual
parameters, like the angle of attack of the nozzle, the maximum camber
and thickness (plus their chord-wise position) or entrance and exit radius
can be easily derived and monitored throughout the design process.

Coarse 4400 0.1887 0.01948 4. Validation of the simplified propeller model

Medium 17,600 0.2381 0.01858
Fine 70,400 0.2379 0.01846
The validation of the computational tools (actuator disk model,
Extrapolated 0.2379 0.01844 wedge or cyclic type boundary conditions) has been carried out consid-
Approximated 0.10% 0.66% ering two ducted propellers for which measurements of forces and flow
relative error (Fine are available (Gaggero et al., 2012, 2013). The first one is an accelerating
– Medium meshes) duct propeller designed for a take home propulsor. It is equipped with a
Extrapolated relative 0.10% 0.76%
error for Medium
conventional Nozzle 19A. Rather than exclusively favor bollard pull ef-
Mesh ficiency, the design was carried out to increase the propulsive thrust of a
GCI 0.00% 0.13% relatively small propeller operating at a certain speed. The second one is

6
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Fig. 4. Iterative convergence (residuals and forces) of the solver. (a) Accelerating type nozzle, L1 residuals convergence. (b) Decelerating type nozzle, L1 residuals convergence. (c)
Accelerating type nozzle, thrust coefficient convergence. (d) Decelerating type nozzle, thrust coefficient convergence.

Fig. 5. Parametric description of nozzles shape. Accelerating duct case.

a decelerating duct propeller developed to postpone cavitation for ap- geometries, propellers are unloaded at the tip and have a small chord,
plications in which reduction of noise (or better, retardation of cavita- which made the selection of the optimal nozzle shape problematic for the
tion) is important. In both cases, differently from usual Kaplan like reasons discussed in the introduction. By using a Boundary Element

7
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Table 3
B-Spline control points on the local nozzle reference system, the range of variability and the compatibility conditions.

Accelerating Nozzles Decelerating Nozzles

x/L y/L x/L y/L

P1 inner 0 (fixed) 0.15–0.20 P1 inner 0 (fixed) 0.05–0.1

P2 inner 0 (fixed) 0.06–0.14 P2 inner 0 (fixed) 0–0.02
P3 inner 0.1–0.25 0 (fixed) P3 inner 0.3–0.55 0 (fixed)
P4 inner 0.329 (fixed) 0 (fixed) P4 inner 0.7–0.85 0 (fixed)
P5 inner 0.65–0.85 0 (fixed) P5 inner 1 (fixed) 0.08–0
P6 inner 1 (fixed) 0–0.04

P1 outer 0 (fixed) y/L P1 inner P1 outer 0 (fixed) y/L P1 inner

P2 outer 0 (fixed) 0.21–0.25 P2 outer 0 (fixed) 0.14–0.24
P3 outer 0.07–0.3 0.16–0.23 P3 outer 0.4–0.8 0.16–0.24
P4 outer 0.5–0.8 0.08–0.14 P4 outer 1 (fixed) y/L P5 inner
P5 outer 1 (fixed) 0.05–0.08

Length scaling 0.75–1.5 Length scaling 0.75–1.5

TNozzle
Method, the radial distributions (Fig. 6) of axial and tangential forces on CT Nozzle ¼1 (2)
2
ρπRprop: V
2 2
the propeller blades were derived and momentum sources were assigned
over the actuator disks coherently. Predicted forces exerted by the noz-
CT prop:
zles in the case of both cyclic type (for which also tangential forces have τ¼ (3)
been included in the analysis) and wedge type boundary conditions were CT prop: þ CT Nozzle

compared to those measured during experiments. Results, as shown in

Figs. 7 and 8, are summarized in two ways. As usual, thrust coefficient TNozzle:
KT ¼ (4)
values of propeller and nozzle are compared in terms of thrust ratio τ in ρN 2 D4
order to verify that the relative contribution of the duct to the total thrust
Results, for both the nozzle geometries, are in good accordance with
increases with increasing propeller loading. Coefficients, respectively CT
experimental data. The agreement between calculations and measure-
prop and CT Nozzle, are non-dimensional with respect to the inflow velocity
ments is good, especially from medium to heavy loading conditions. The
V and the area of the propeller disk. Also thrust coefficient KT, assuming
inclusion of the tangential forces as an additional momentum source
the revolution rate (N) of the propeller during experiments and its
distribution, when cyclic boundary conditions are adopted, has a favor-
diameter as reference to derive non-dimensional quantities, is used for
able effect on the numerical calculations. As pointed out in Hoekstra
comparison purposes.
(2006), the additional pressure drop in the swirling slipstream due to the
Tprop: modeling of tangential velocities increases the pressure difference be-
CT prop: ¼ 1 (1) tween the inner and the outer duct surface, raising the duct forces to
2
ρπR2prop: V 2
values closer to measurements. This effect is particularly evident at

Fig. 6. Axial and tangential force distributions. Reference stands for the propeller loading as computed by the Boundary Element Method for the given blade geometry. Loaded and
Unloaded stand for tip modified distributions used to assess the sensitivity of the design process to local changes of load. (a) Accelerating ducted propeller. (b) Decelerating duc-
ted propeller.

8
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Fig. 7. Comparison of measured and computed nozzle thrust. Accelerating duct. (a) Comparison of the nozzle over propeller thrust ratio (τ) as a function of the propeller thrust coefficient
CT prop. (b) Comparison of thrust coefficients (KT) as a function of the propeller advance coefficient J.

Fig. 8. Comparison of measured and computed nozzle thrust. Decelerating duct. For confidentiality reasons, values are given non-dimensional with respect to a reference operative
condition. (a) Comparison of the nozzle over propeller thrust ratio (τ) as a function of the propeller thrust coefficient CT prop. (b) Comparison of thrust coefficients (KT) as a function of the
propeller advance coefficient J.

higher propeller loadings. In lightly loaded conditions, differences be-
tween computations and model tests are larger. The inclusion of
tangential velocities in the numerical simulation does not contribute
significantly to improve the correlation with experiments by modifying
the pressures. Especially in the case of the accelerating duct geometry,
the thrust of the nozzle is significantly overestimated. This difference
may be explained by considering that the flow predicted by the actuator
disk models is axisymmetric and neglects the tip leakage vortex; this can
alter the streamlines path in the circumferential direction and produce a
negative interaction with the duct blunt trailing edge (Bosschers
et al., 2015).
Accelerating duct geometries, however, are primarily designed to
operate at high propeller loading. The discrepancy at very high advances,
not contemplated in present design activity, can be consequently
accepted, as well as the approximations introduced by the wedge type
boundary conditions setup. The comparative nature of a design by
optimization, indeed, relies more on the ability to identify relative dif-
ferences among designs, rather than on the absolute accuracy of the
calculations. The computational efficiency that can be exploited with the
wedge type boundary conditions largely compensates for the slightly less
accurate results.
Streamline patterns and inner pressure fields
(CP ¼ 2ðp  pref : Þ=ðρV 2 Þ) for the two nozzle geometries are shown in
Fig. 9. The position of the leading edge stagnation point is evident, as
well as how streamlines change with propeller loading. Streamlines, as
already evidenced in Hoekstra (2006), approach Nozzle 19A almost
along its camber line for a propeller thrust coefficient CT prop. of about 4.
On the contrary, in lightly loaded conditions, the risk of a separation
bubble on the outer side is clear, completely in agreement with the design
of the nozzle, devoted to highly loaded functioning. For the decelerating
nozzle shape, it is possible to appreciate the increase of the inner pressure
close to the design point (CT prop./CT prop. ref. ¼ 0.10, J ¼ 103.4% of Jref.),
together with streamlines approaching the nozzle at its ideal angle
of attack.
It is worth noting that calculations have been carried out with a sort of
“GAP” model. The flow passing through the gap between the propeller tip
and the inner duct surface is, indeed, strongly influenced by viscosity.
Simplified representations have been proposed to include these effects in
Boundary Element Methods calculations. It is the case of Hughes (1993),
who proposed to “fill” the gap between the propeller tip and the inner
surface of the duct with an additional strip of chord-wise panels. The
strength of dipoles and sources distributed over these panels was pre-
scribed through a transpiration velocity in accordance with empirical
coefficients to account for the influence of viscosity on the flow rate
(reduced with respect to pure potential approximations) through the gap.
Calculations with momentum sources by Hoekstra (2006) partially
showed the presence of a recirculation zone in correspondence to the
propeller plane for relatively highly load conditions, with non-reattached
streamlines as the tip clearance increases or the loading at the tip is
reduced. The extension of this zone was, however, overpredicted with
respect to the few experimental measurements available. Bosschers et al.
(2015) further analyzed this phenomenon and concluded that the

9
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Fig. 9. Streamlines and inner pressure distribution at different loading conditions. Accelerating (Nozzle 19A, left) and decelerating (right) ducts. For confidentiality reasons, the
reference decelerating nozzle shape cannot be disclosed. (a) CT prop. ¼ 12.0 (J ¼ 0.3) (b) CT prop. ¼ 4.0 (J ¼ 0.5) (c) CT prop. ¼ 0.35 (J ¼ 1.1) (d) CT prop./CT ref. ¼ 0.41 (J ¼ 60% of Jref.)
(e) CT prop./CT ref. ¼ 0.11 (J ¼ 100% of Jref.) (f) CT prop./CT ref. ¼ 0.02 (J ¼ 140% of Jref.).

presence of a recirculation zone is due to the loading of the propeller that
generates an upstream velocity in the gap where no momentum sources
are present. Momentum sources, even in the case of viscous calculations,
cannot account for the influence of a “physical” surface as that of the
propeller blades. Body forces are almost “transparent” for the flow that is
solely accelerated and not deviated by the “obstacle” represented by the
blades. The circumferential averaged distribution of forces realized by
the actuator disk further prevents any asymmetry of the flow, avoiding
the generation of the tip leakage vortex that could mitigate the risk of
separation. Hence, as proposed by Bosschers et al. (2015), a gap model is
required also in the case of viscous calculations with momentum sources.
Such model could be easily realized, perfectly in analogy with the one
proposed for BEM calculations, by placing momentum sources of reduced
intensity in the clearance between the propeller tip and the inner surface
of the nozzle. Exactly as the propeller forces modeled through the radi-
ally varying actuator disk, also the “GAP” flow model is applied uni-
formly in the circumferential direction, averaging its influence all over
the ring across the propeller tip, consistently with the modeling of the
propeller with an actuator disk. In addition, this approach is the only
possible with the wedge type computational domain. More complex

Fig. 10. Influence of the “GAP” model on separation on the inner surface of Nozzle 19A. CT prop. ¼ 4.0 (J ¼ 0.5). (a) Without the “GAP” model, CT Nozzle ¼ 1.07 (CT Nozzle Exp. ¼ 1.587). (b)
With the “GAP” model, CT Nozzle ¼ 1.58 (CT Nozzle Exp. ¼ 1.587).

10
S. Gaggero et al.
Fig. 11. Pareto diagram of the accelerating duct design by optimization. Correlation be-
tween nozzle thrust and average inner pressure at the higher advance coefficient
(J ¼ 0.80, CT prop ¼ 1.23).
propeller representations, like the local distributions of momentum
sources resembling each single blade adopted in Villa et al. (2011),
would have allowed for the use of more sophisticated “GAP” approxi-
mations, which could be based, for instance, on the exact location of
blade tips. As shown by Bosschers et al. (2015), however, even the in-
clusion of such a simplified “GAP” flow model dramatically helps in
avoiding recirculation. The “GAP” model provides a flow in the trailing
part of the duct that is closer to the average flow obtained from fully
resolved RANSE calculations. The use of the Boundary Element Method
(that includes the simplified model by Hughes) to define the radial dis-
tribution of forces, finally, provides naturally the strength of these
additional momentum sources.
An example of application is shown in Fig. 10. Without the “GAP”
model a serious recirculation zone, resulting in a significant change
(reduction) of the thrust delivered by the nozzle, can be evidenced in
correspondence to the propeller disc. Streamlines instead are aligned
with the duct inner surface and force is better in agreement with mea-
surements when momentum sources fill the clearance between the pro-
peller and the nozzle.
5. Optimization of nozzle shapes
The design of nozzles through optimization was achieved starting
from two reference geometries available at the cavitation tunnel of the
University of Genoa. In both cases, the design consists in the analysis of
an initial population of 150 members. This population was selected by
using the quasi-random Sobol sequencing, resulting in a sufficiently
uniform distribution in the design space of the possible geometrical
Fig. 12. Pareto diagrams of the accelerating duct design by optimization. Correlation between nozzle thrusts (left) and average inner pressures (right) computed in bollard pull and at the
higher advance coefficient conditions. (a) Correlation of thrusts. (b) Correlation of pressures.
11
Ocean Engineering xxx (2017) 1–20
Table 4
Functioning conditions for the accelerating duct design.
Condition 1 Condition 2
(i.e. bollard pull) (higher advance)
CT prop. 12 1.23
J (equivalent propeller 0.3 0.8
advance coefficient)
combinations. The optimization workflow lets the initial population
evolve for 15 generations; this, in the light of preliminary calculations
and the relatively low number of free parameters, was considered suffi-
cient to achieve convergence. Results of the analyses are summarized in
Sections 5.1 and 5.2.
5.1. Accelerating ducts
The outcomes of the design by optimization of the accelerating duct
are summarized in the Pareto diagrams of Figs. 11 and 12. Among the
design activities taken into consideration, the one involving the accel-
erating duct is the more complex. It considers, indeed, multiple working
conditions, as outlined in Table 4. The nozzle is designed to provide the
maximum thrust close to bollard pull condition that in numerical cal-
culations was approximated with a relatively low propeller advance co-
efficient to avoid the complications related to zero inflow. At the same
time, minimization of the nozzle drag at higher advance coefficients
(J ¼ 0.8 in the present case) was required, together with maximization of
inner static pressure at both functioning. The static pressure was moni-
tored through the average pressure on a propeller disc in front of the
actuator disc at x/D ¼ 0.15 from the propeller plane, as in Fig. 5. The
propeller loading is that of Fig. 6a. Compared to the parametric load
distributions of Hoekstra (2006) that resembled tip loaded Kaplan like
propeller force distributions due to the finite chord at the tip, the present
distribution of force is particularly unloaded. This particular feature, that
was preferred for the nature of the application, further reinforces the
need for the “GAP” model (even though simplified) to limit the risk of
flow separation on the inner of the nozzle evidenced by Bosschers
et al. (2015).
Results of the optimization process show certain margins of im-
provements, which can be verified by considering some relevant geom-
etries among those belonging to the Pareto Frontier. As outlined in the
discussion, without redesigning the propeller, the recommended optimal
nozzles are those operating at constant thrust coefficient in order not to
change the design point of the propulsion system. The almost constant
nozzle thrust allows also minimizing the influence of the nozzle on the
propeller inflow and, in turn, on its performance. Depending on the
importance of bollard pull with respect to routinely operative conditions
at higher advance coefficient, there are multiple possible choices. In the
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Table 5
Geometrical characteristics and performance of the selected designs for the accelerating duct case (reference points and camber line definition slightly different from those proposed for the
reference Nozzle 19A by Van Manen and Oosterveld, 1966).

Nozzle 19A ID 2172 ID 1987 ID 1736 ID 1835 ID 1056

L/D 0.5000 0.4075 0.3875 0.4825 0.4625 0.6525

Rinlet/D 0.5913 0.5860 0.5885 0.5905 0.5765 0.5850
Routlet/D 0.5218 0.5238 0.5245 0.5248 0.5235 0.5298
cambermax/L 0.05421 0.05007 0.06600 0.06872 0.05164 0.04264
thicknessmax/L 0.16452 0.21127 0.20288 0.14951 0.15582 0.11475
x max camber/L 0.2910 0.3437 0.3527 0.3617 0.4576 0.3288
x max thickness/L 0.2768 0.3171 0.3288 0.2437 0.2656 0.2910
Incidence angle [ ] 7.900 8.661 9.384 7.785 6.569 4.862

CT Nozzle (J ¼ 0.3, CT prop. ¼ 12) 6.731 6.725 5.829 7.370 7.080 7.510
CT Nozzle (J ¼ 0.8, CT prop. ¼ 1.23) 0.238 0.236 0.237 0.237 0.267 0.282
CP (J ¼ 0.3, x/D ¼ 0.15) 9.363 8.368 7.268 9.866 9.239 11.105
CP (J ¼ 0.8, x/D ¼ 0.15) 1.019 0.861 0.815 1.096 1.010 1.278

present design, where the reference propeller/nozzle system was
designed for take home purposes (i.e. provide a reasonably high thrust
with a small propeller at a relatively high advance coefficient) it was
considered mandatory, at first, to improve the performance at the higher
advance coefficient. In order to limit cavitation phenomena, moreover,
the increase of the inner nozzle pressure in this condition is another
important objective of the design that turns into the second most
important aspect for the selection of an optimal geometry. Among the
Pareto geometries, ID 2172 is the optimal balance between the opposite
design requirements. Providing a constant thrust at the higher advance
coefficient, ID 2172 increases, at that condition, the inner static pressure
at its maximum without worsening the performance at bollard pull
(constant thrust) in correspondence to which a certain improvement in
terms inner pressure can be appreciated too. Other geometries presup-
pose trade-offs among objectives. At constant nozzle delivered thrust for
J ¼ 0.80, ID 1987 provides the maximum increase of static pressure
(appreciable also at bollard pull) at the cost of a sensible reduction of
bollard pull thrust (Fig. 12). On the opposite, ID 1736 maximizes bollard
pull performance at constant thrust for J ¼ 0.80 with a detrimental in-
fluence on cavitation avoidance. By accepting the redesign of the pro-
peller, ID 1835 exploits the maximum thrust from the propeller without
any significant influence on pressure fields while ID 1056 is devoted only
to the maximization of nozzle delivered thrust regardless the inner
pressure field. Detailed descriptions of the geometrical modifications
necessary for these achievements are summarized in Table 5 while in
Figs. 13 and 14 the computed pressure and velocity fields are shown for
the selected geometries at both the advance coefficients (i.e. propel-
ler loadings).
The presentation of the results in the Pareto diagrams shows some
correlations among objectives. Obviously, there is a relation between
nozzle thrusts at different advance coefficients (propeller loadings): a
geometry that maximizes bollard pull performance improbably will
provide poor performance in terms of thrust at different operative con-
ditions. At the higher advance coefficient, results are a bit more scattered
and there is a certain degree of freedom, exploited for the selections of
the candidates for the detailed analyses of Figs. 13 and 14. The correla-
tion between inner pressures is even stronger and almost linear. By
analyzing the performance of all the geometries tested during the opti-
mization process, some important correlations among performance and
geometrical parameters can be furthermore discussed. Data are collected
in the charts of Figs. 15 and 16 where the Student t-test was used to es-
timate the influence (and the significance, i.e. the trustworthiness of the
devised relationship) of the input variables (the design parameters) on
the outputs (Esteco, 2016). By combining variances, mean values and
number of variables in the lower and in the upper part of the design
space, essentially the t-test provides a measure of the strength of the
relationship between the outputs and the inputs, that is summarized in
the pie chart of Fig. 15. Another output of the analysis is the kind of

Fig. 13. Streamlines and inner pressure distribution for the selected optimal geometries. Bollard pull condition (J ¼ 0.3, CT prop. ¼ 12). (a) Reference Nozzle 19A. (b) ID 1056. (c) ID 1835.
(d) ID 1736. (e) ID 1987. (f) ID 2172.

12
S. Gaggero et al.
Fig. 14. Streamlines and inner pressure distribution for the selected optimal geometries. Higher advance coefficient condition (J ¼ 0.8, CT prop. ¼ 1.23). (a) Reference Nozzle 19A. (b) ID
1056. (c) ID 1835. (d) ID 1736. (e) ID 1987. (f) ID 2172.
Fig. 15. Overall Student t-test for design objectives of the accelerating duct.
relationship (direct or inverse) among inputs and outputs, highlighted by
the bar chart of Fig. 16. Even if the number of elements over which the
statistical analysis is carried out is rather limited, some observations circa
the observed influences can be drawn, also in the light of the very low
values of significance obtained by the t-test (the lower the significance,
the higher the trustworthiness of the effects, direct or inverse, between
inputs and outputs).
The length of the nozzle, as well as the maximum nozzle thickness,
have a significant importance on design objectives. Higher values of
thrust, especially close to bollard pull, are achieved with longer nozzles.
The direct effect of nozzle length in correspondence to lightly loaded
propeller functioning is lower, exactly as the trends evidenced by Van
Manen and Oosterveld (1966). Increased thrust is directly influenced by
the decrease in thickness and of incidence angle. The exit radius (defined
in Fig. 5) has a non-negligible influence on delivered thrust while the role
of the camber (value and position) is marginal, as already evidenced by
the analyses carried out in the case of conventional Nozzles 18, 19 and
20. When looking at the maximization of the inner pressure, these vari-
ables have exactly an opposite influence. The improvement of cavitating
performance favors shorter but thicker nozzles, higher angles of attack
and lower exit radii, encouraging configurations and trends that some-
how tend towards those typical of the decelerating nozzles of Section 5.2.
This demonstrates the complexity of the design and the need of a
trade-off between contrasting objectives also in the light of constraints
(regarding geometry or structural strength) that, presently, have not been
considered. This analysis and the similar one proposed in next Section for
the decelerating nozzle case has not to be considered, however, fully
exhaustive. The number of configurations accounted for the t-test,
13
Ocean Engineering xxx (2017) 1–20
indeed, is probably not large enough. In particular, it is not equally
distributed over the design space to account evenly for the dependencies
and the relationships existing between free parameters and objectives.
The use of a sampling of the design space based on results of an opti-
mization activity, in addition, directs the candidates for the t-test towards
the geometries and the choice of the parameters that favor the objectives
of the design. The candidates in the upper and in the lower part of the
design space, in turn, could be not equally divided, affecting the accuracy
of the test, whose reliability has to be considered in relation to this choice
of design space and range of variations of the parameters. In this sense,
some contrasting results of the analysis could be explained. For instance,
the opposite influence of the angle of attack with respect to the inlet
radius (that in some way is dependent on a combination of the angle of
attack itself and the nozzle length) on the average pressure, as well as the
role of the maximum thickness, could be ascribed to an insufficient and
not perfectly appropriate choice of the elements for the t-test.
In the absence of dedicated experimental analyses, an indirect vali-
dation of the outcomes of the design by optimization has been proposed
by using 3D RANSE calculations of the selected geometries. OpenFOAM
calculations were carried out based on the numerical setup for propeller
performance evaluation devised in Gaggero and Villa (2017), using a
polyhedral mesh between 1.4 and 1.6 Million cells per blade depending
on the geometry under investigation. The reliability of the proposed
calculations has been preliminary verified in the case of the reference
Nozzle 19A arrangement by comparing, in Table 6, the computed results
with the available experimental measurements. The overall agreement at
both the advance coefficients under investigation is good. The propeller
thrust is slightly underestimated and torque is reasonably predicted. The
S. Gaggero et al.
Fig. 16. Effect size of each geometrical parameter on the design objectives. Accelerating duct. (a) Effect on CP (J ¼ 0.3, CT prop. ¼ 12). (b) Effect on CP (J ¼ 0.8, CT prop. ¼ 1.23). (c) Effect on
CT Nozzle (J ¼ 0.3, CT prop. ¼ 12). (d) Effect on CT Nozzle (J ¼ 0.8, CT prop. ¼ 1.23).
Table 6
Comparison between measured and predicted performance. Accelerating ducted propeller inside Nozzle 19A.
Exp. KT Prop. Δ%
RANSE
J ¼ 0.3 0.4269 0.4168 2.36%
J ¼ 0.8 0.3091 0.3046 1.46%
thrust delivered by the nozzle is in good agreement with measurements
close to the bollard pull condition (J ¼ 0.30). At higher J, the differences
are amplified by the low reference value with respect to which small
absolute discrepancies turn into high percentage errors; these, however,
appear practically negligible when looking at the total (propeller plus
nozzle) thrust. Close to bollard pull, the difference is less than 1% and
calculations show an underpredicted total thrust of about 0.5%. At
J ¼ 0.80 the overestimation is about 2.8%, perfectly in line with the
assumption of calculations.
The five geometries identified by means of the optimization were
analyzed (Fig. 17) with the same numerical setup of the reference Nozzle
19A configuration. The percentage differences with respect to the nu-
merical results of the reference geometry, both in terms of delivered
thrust and average pressure coefficient in front of the propeller are
collected in Table 7. The complexity of the three-dimensional flow
around a ducted propeller, already evidenced in the case of the issues
related to the gap and the risk of flow separation, is sensibly simplified
with the adoption of actuator disk models. Hence the comparison is
significant only for what regards trends rather than in terms of absolute
value of improvements of one geometry with respect to the others. In this
sense, the results collected with fully resolved RANSE calculations are
satisfactory. Close to bollard pull condition, fully resolved calculations
underestimate the increase (decrease) of nozzle thrust with respect to the
calculations with the simplified actuator disk. A different propeller
loading at bollard pull (not accounted during optimization) and the
Ocean Engineering xxx (2017) 1–20
Exp. 10KQ Δ% Exp. KT Nozzle Δ%
RANSE RANSE
0.9867 0.9862 0.05% 0.2561 0.2629 þ2.64%
0.7712 0.7826 þ1.48% 0.0414 0.0557 þ34.44%
three-dimensional interaction of the flow may justify these differences. It
is remarkable, however, that the highlighted differences do not invali-
date the results of the design. With the exception of ID 1736 and ID 1835
(which do not have the same ranking at J ¼ 0.30), the trend and the
ranking evidenced at the lower advance coefficient by the simplified
optimization process are confirmed. The same ID 1736, ID 1835 plus ID
1056 nozzle geometries are those that provide the highest increase of the
delivered thrust at bollard pull which, however, turns into a worsening
(reduction) of the propeller performance (thrust), with obvious conse-
quences discussed in the following on the inner pressure distributions. At
J ¼ 0.80, nozzles ID 2172, ID 1736 and ID 1987 deliver almost the same
thrust of the reference configuration, as expected by the optimization
(Fig. 11). Only ID 1056 does not confirm the results achieved with the
actuator disk. This geometry should have been the one delivering the
higher thrust at both the advance coefficients with the poorer perfor-
mance in terms of average pressure distribution to the propeller. While
the substantial worsening of the performance in terms of inner pressure
distribution is confirmed, this geometry is not the one providing the
highest increase of nozzle thrust at J ¼ 0.80 with respect to the reference
Nozzle 19A. In this respect, ID 1835 performs better. Also the increase of
inner pressure, foreseen by the optimization, is generally confirmed by
the fully resolved RANSE calculations. The improvements of ID 1987 and
ID 2172 at both advances are clear and in line with the results of the
optimization. ID 1736, instead, has an opposite trend with respect to the
optimization process. With the actuator disk, this geometry should have
14
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Fig. 17. Pressure coefficient distribution on propeller and duct at J ¼ 0.8. Fully resolved RANSE calculations. Accelerating duct design. Pressure coefficient based on propeller rate or
revolution (CpN ¼ 2ðp  pref : Þ=ðρN 2 D2 Þ). (a) Reference Nozzle 19A. (b) ID 2172. (c) ID 1987. (d) ID 1736. (e) ID 1835. (f) ID 1056.

inflow as a consequence of sensibly modified nozzle shaped) predicted by

Table 7 the fully resolved RANSE calculations and not accounted by the fixed
Optimized accelerating nozzle performance. Differences with respect to the Nozzle 19A
configuration.
body forces distribution adopted throughout the optimization process.
For those two configurations (but also for ID 1056, as discussed previ-
ID 2172 ID 1987 ID 1736 ID 1835 ID 1056
ously) the predicted propeller thrust is lower than that of the propeller
ΔT Nozzle, 0.08% 13.39% 9.51% 5.19% 11.58% with the Nozzle 19A (about 4% for both the geometries at J ¼ 0.8).
Optimization,
Lower propeller loading turns immediately into a lower decrease of
J ¼ 0.30
ΔT Nozzle, RANSE, 0.92% 2.71% 2.07% 2.32% 2.50%
pressure on the suction side and the consequences on the average pres-
J ¼ 0.30 sure in front of the propeller are obvious. On the other hand, geometries
(ID 2172 and 1987) selected to provide the same nozzle thrust at J ¼ 0.80
ΔT Nozzle, 0.94% 0.40% 0.58% 12.20% 18.35%
Optimization, without improvements at bollard pull (both the analyses shows a
J ¼ 0.80 reduction of the delivered thrust in this condition) are those that avoid
ΔT Nozzle, RANSE, 0.52% 1.01% 2.54% 16.12% 5.91% any significant modification of the propeller functioning. At both oper-
J ¼ 0.80
ating points, fully resolved RANSE calculations predict propeller per-
ΔCP, Optimization, 10.62% 22.37% 5.37% 1.33% 18.61% formance with a maximum of 1% variation with respect to the reference
J ¼ 0.30 configuration with Nozzle 19A, proving the assumption of the design at
ΔCP, RANSE, 17.72% 24.07% 2.56% 8.01% 11.29%
constant nozzle thrust.
J ¼ 0.30

ΔCP, Optimization, 15.44% 20.04% 7.55% 0.84% 25.41%

J ¼ 0.80 5.2. Decelerating ducts
ΔCP, RANSE, 27.78% 35.65% 2.84% 10.60% 23.15%
J ¼ 0.80
The optimization of the decelerating duct has been carried out by
considering the propeller loadings of Fig. 6b. As partially discussed in the
provided a reduction of the inner pressure (7.55% with respect to the introduction, the loading distribution has a certain influence on the
average pressure inside Nozzle 19A at J ¼ 0.80) while resolved RANSE nozzle performance that could be exploited during the design. The
calculations suggest an increase (þ2.84%) at the same advance coeffi- reference propeller was designed, as in the case of the accelerating duct,
cient. Similarly, ID 1835 provides increases substantially higher than to postpone as much as possible cavitation inception (Gaggero et al.,
those evidenced in the optimization process. These differences may be 2012), which of course affects the tip at first. Through optimization,
ascribed to the different functioning of the propeller (due to a modified therefore, also the influence of propeller tip loading on nozzle shape was
investigated as a first step in the direction of a combined propeller/nozzle
Table 8 design. Any redesign of the propeller, in fact, turns into a different load
Functioning conditions for the decelerating duct design and influence of tip loading on the distribution and it is important to evaluate which could be the influence
computed performance of the reference nozzle shape. of such modifications on the nozzle performance. An unloaded tip blade
Load BEM Tip Loaded Tip Unloaded obviously improves the inception speed of the propeller. The possible
CT prop./CT prop. ref. 0.10 0.10 0.10 reduction of the nozzle thrust or of the inner static pressure, however,
J (equivalent propeller 103.4% of J ref. 103.4% of J ref. 103.4% of J ref. could nullify the positive influence of tip unloading. On the contrary, the
advance coefficient) negative effects related to tip loading (i.e. anticipated cavitation incep-
CT Nozzle/CT prop. ref. 0.0182 0.0186 0.0170 tion) could be mitigated by an increase of the inner pressure or by the
CP at x/D ¼ 0.15 0.035 0.041 0.018
possibility to operate the propeller at a most favorable point thanks, for

15
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

reduction of the pressure. This behavior, within the simplification of

circumferentially averaged body forces, can be explained by the change
of the flow around the duct, induced by the different loading distribution.
With the tip loaded, the angle of attack of the duct increases, and the
stagnation point moves towards the inner surface of the nozzle. This local
increase of pressure overbalances its reduction due to the higher load (i.e.
higher suction) at the tip. The shape of the reference nozzle, with the
usual converging geometry at the trailing edge, inherently reduces the
risk of flow separation observed in the case of accelerating duct in lightly
loading conditions. Also the need of the “GAP” model should be dis-
cussed in the case of decelerating duct geometries in the light of the
different behavior of the flow at the inner nozzle surface. The satisfactory
results collected during the validation analyses, which were computed in
any case (accelerating and decelerating) with momentum sources also in
the gap region, authorize the application of the same model also for the
optimization activities.
Optimizations have been performed at the propeller design advance
Fig. 18. Pareto diagram of the decelerating duct design by optimization. Reference pro-
peller loading distribution.
coefficient, in correspondence to the same propeller loading condition
used for the preliminary assessment of the influence of tip. The design
was carried out by requiring the maximization of the inner pressure (also
in this case the average pressure extracted on a plane 0.15D in front of the
propeller disc) and the minimization of the nozzle drag. The results of the
calculations with different loading distributions have been collected
through Pareto diagrams among which the one for the reference distri-
bution is shown in Fig. 18. Trends are very similar regardless the loading
distribution and the Pareto frontiers show an almost linear correlation
between thrust and inner pressure.
The comparison of the Pareto frontiers (shown in Fig. 19) for the three
cases under investigation is more significant. At constant nozzle thrust
(drag, in this case), that is the condition that minimizes the need of a
propeller redesign (needed only if induced velocities to the propeller are
significantly changed), the very low overpressure with respect to undis-
turbed flow in front of the disk of the reference geometry is more than
doubled. At constant inner pressure, the drag of the nozzle can be
reduced by about 20%, which may represent a certain margin for the
Fig. 19. Comparison of the Pareto frontier for different propeller loading distributions.
redesign of a propeller in a more favorable, for cavitation inception,
unloaded working condition.
Results of the optimization activities are somehow in contrast with
instance, to a reduction of the nozzle drag. The tip loaded distribution has
the preliminary verification of the influence of tip loading. From the
been realized by simply increasing the value of the delivered propeller
analysis of the reference geometry in Table 8, the unloading at tip proved
forces only on the outer actuator disk sections. The unloaded tip distri-
to be effective in reducing the nozzle drag with a slight worsening (i.e.
bution, instead, couples the decreasing of the forces at outer radii with
reduction) of the inner pressure. Nozzle designs obtained through opti-
the shift of the position of the maximum of the distribution towards the
mization in the case of the unloaded tip show, instead, a simultaneous
center of the disk.
improvement of both the objectives (reduction of drag and increase of
A preliminary analysis of the influence of the propeller loading dis-
inner pressure) with respect to designs obtained with the reference and
tributions (at constant total propeller loading) on the reference nozzle
the tip loaded propeller load distributions. Designs obtained with the
performance is reported in Table 8. Calculations have been carried out
unloaded tip are overall better both at any nozzle thrust (providing the
for the propeller design condition and are summarized in the same table:
highest inner pressure) and at any inner pressure level (providing the
loading the tip slightly increases the average static pressure in front of the
lower drag). The selected optimal shapes have the main geometrical
actuator disk at the cost of a negligibly higher drag. Unloading, on the
characteristics collected in Table 9 while non-dimensional pressure fields
contrary, reduces the drag of the nozzle at the cost of a more significant

Table 9
Geometrical characteristics and performance of the selected designs for the decelerating duct case (CT prop./CT prop. ref. ¼ 0.10, J ¼ 103.4% of Jref.).

Reference Load Loaded Tip Unloaded Tip

ID 2201 ID 2192 ID 1930 ID 1914 ID 2231 ID 1948

L/D 0.4400 0.3850 0.4250 0.4375 0.3950 0.3850

Rinlet/D 0.5370 0.5493 0.5500 0.5500 0.5383 0.5455
Routlet/D 0.4738 0.4800 0.4758 0.4798 0.4743 0.4785
cambermax/L 0.03930 0.03775 0.02997 0.02594 0.04315 0.03957
thicknessmax/L 0.12513 0.17058 0.15210 0.14653 0.13693 0.16360
x max camber/L 0.5830 0.6292 0.6231 0.6322 0.6047 0.5892
x max thickness/L 0.2437 0.2740 0.2198 0.2276 0.2518 0.2601
Incidence angle [ ] 8.180 10.197 9.910 9.122 9.203 9.872

CT Nozzle/CT prop. ref. 0.0182 0.0150 0.0182 0.0154 0.0180 0.0148

CP 0.086 0.034 0.091 0.041 0.093 0.037

16
S. Gaggero et al.
Fig. 20. Streamlines and inner pressure distribution for the selected optimal geometries with the reference propeller loading distribution (J ¼ 103.4% of Jref., CT prop./CT prop. ref. ¼ 0.10).
(a) Reference Geometry. (b) ID 2201. (c) ID 2192.
Fig. 21. Streamlines and inner pressure distribution for the selected optimal geometries with the propeller tip loaded distribution (J ¼ 103.4% of Jref., CT
Reference Geometry. (b) ID 1930. (c) ID 1914.
Fig. 22. Streamlines and inner pressure distribution for the selected optimal geometries with the propeller tip unloaded distribution (J ¼ 103.4% of Jref., CT
Reference Geometry. (b) ID 2231. (c) ID 1948.
and streamlines on the meridian plane are presented in Figs. 20–22. For
geometries improving the increase of pressure at the propeller plane, it is
possible to observe stagnation points slightly moved to the inner part of
the nozzle. The suction peak on the outer duct surface is not sensibly
affected by the geometrical modifications and the risk of sheet cavitation
on the nozzle, even though not directly monitored throughout the opti-
mization process, is not appreciably enlarged for the selected geometries.
The substantial identity of performance turns into very close shapes
regardless the distribution of forces of the actuator disk: geometries with
higher inner pressure are generally thinner and with a smaller exit sec-
tion. Also the length is decreased to avoid high losses due to friction but,
with respect to the analyses of the parameters influence in the case of
accelerating nozzles, its importance is scaled down. All the optimal ge-
ometries have a nozzle length substantially lower than the reference
geometry, regardless the objective that is addressed.
A thorough estimation of the relations existing between geometrical
modifications and objectives is outlined in Figs. 23 and 24. Also for the
decelerating nozzle design, the Student t-test, with the same limitations
highlighted for the accelerating duct case, has been used to verify the
relationships and their nature, direct of inverse, between modifications of
main parameters outlined in Fig. 5 and effects on objectives. Due to the
very similar design outcomes with different choices of the propeller
loading distribution, only the reference case is considered. As in the case
17
Ocean Engineering xxx (2017) 1–20
prop./CT prop. ref. ¼ 0.10). (a)
prop./CT prop. ref. ¼ 0.10). (a)
of accelerating duct geometries, the objectives are influenced by the
parameters in opposite ways. The exit radius, the position of the
maximum thickness, the maximum camber (and its position) and the
incidence angle turn into the most influencing geometrical characteris-
tics. The nozzle length, as for the discussion circa the selected optimal
shapes, marginally influences performance improvement.
Exactly as done in the case of the accelerating nozzle case, a valida-
tion of the results of the design by optimization was carried out by means
of fully resolved RANSE analyses. Only the reference propeller loading
distribution (i.e. geometries ID 2192 and ID 2201) was analyzed, being
the tip loaded and unloaded distributions only case studies without a real
propeller geometry available. In the case of the reference geometry
(Table 10), numerical calculations with OpenFOAM reasonably predict
the propeller performance, with results in line with those already pro-
posed by Gaggero et al. (2012, 2013) using a different RANSE solver
(StarCCMþ). The overprediction of the nozzle drag, when looking at the
percentage differences, is again amplified by the very low reference
value. Its influence on the total propulsive thrust (numerically under-
estimated by less than 0.7%) is, however, negligible.
The results of the optimization based on the simplified model of the
actuator disk are, also in the case of the decelerating nozzle configura-
tion, confirmed by the fully resolved RANSE analyses (Fig. 25) with few
discrepancies worth of discussion (Table 11). ID 2201 provides a nozzle
S. Gaggero et al. Ocean Engineering xxx (2017) 1–20

Fig. 23. Overall Student t-test for design objectives of the decelerating duct. Reference propeller loading distribution.

Fig. 24. Effect size of each geometrical parameter on the design objectives. Reference propeller loading distribution. (a) Effect on CP. (b) Effect on CT Nozzle.

actuator disk) on the flow field, as discussed in Section 3.2. A propeller

Table 10
Comparison between measured and predicted performance. Decelerating ducted propeller
produces an asymmetrical pressure jump: the suction side is commonly
inside the reference geometry. characterized by a reduction of the pressure that is two or three times
higher (in magnitude) with respect to the pressure increase on the
ΔKT Prop. Δ10KQ ΔKT Nozzle
pressure side of the blade (Gaggero et al., 2017). On the contrary, a
J ¼ 103.4% of Jref. 3.19% 6.14% 24.34%
momentum source always generates a symmetrical pressure jump equally
distributed between decrease of pressure in its front and increase in its
thrust slightly higher than the reference geometry but reasonably in line aft. This means that with the actuator disk the reduction of pressure in
with the outcomes of the optimization, which identified this geometry as front of the propeller is underestimated, resulting in the higher (with
the one providing the maximum increase of inner pressure at constant respect to fully resolved propellers with RANSE) inner pressure seen
nozzle thrust (Fig. 19). The increase of the inner pressure, indeed, is throughout the optimization process.
substantial (about þ30%) but far from the predictions with the actuator The relative (with respect to the reference geometry performance)
disk which foreseen a more than doubled inner pressure. Reasons for this increase of nozzle thrust of ID 2192 is similarly predicted by both the
discrepancy can be found in the influence of momentum sources (i.e. the approaches. Fully resolved RANSE calculations, however, identify a

Fig. 25. Pressure coefficient distribution on propeller and duct. Fully resolved RANSE calculations. Decelerating duct design. Pressure coefficient based on propeller rate or revolution
(CpN ¼ 2ðp  pref : Þ=ðρN 2 D2 Þ). (a) Reference Geometry. (b) ID 2201. (c) ID 2192.

18
S. Gaggero et al.
Table 11
Optimized decelerating nozzle performance. Differences with respect to the reference
geometry.
ID 2201 ID 2192
ΔT Nozzle, Optimization, J ¼ 103.4% of Jref. 0.17% þ17.57%
ΔT Nozzle, RANSE, J ¼ 103.4% of Jref. þ3.90% þ16.59%
ΔCP, Optimization, J ¼ 103.4% of Jref. 146% 3%
ΔCP, RANSE, J ¼ 103.4% of Jref. 27.55% 8.29%
certain increase of the inner pressure, not evidenced in the case of the
actuator disk (ID 2192 was selected as the geometry providing the
maximum increase of nozzle thrust at almost constant pressure). This
discrepancy, again, can be explained by the different working condition
of the propeller. Nozzle ID 2192 is the shortest and has the highest
incidence angle. The inflow to the propeller is such that the propeller
delivered thrust results lower with respect to reference geometry by
about 4.2%. Exactly as discussed in the case of the accelerating nozzle
geometries, unloading the propeller reduces the suction on the back side
of the blade with an obvious influence on the average inner pressure in
front of the propeller.
These results, together with the analyses carried out for the acceler-
ating nozzle configurations, mainly confirm the robustness of the opti-
mization approach. In parallel, however, they suggest once more the
need of a feedback to the propeller loading/actuator disk momentum
distribution inside the optimization loop. In addition, a selection of an
optimal candidate based on a posteriori fully resolved RANSE analyses on
a reasonable set of geometries selected through the (simplified) optimi-
zation procedure, as already proposed in the case of conventional highly
loaded propellers of Gaggero et al. (2017b) could furthermore refine
the choice.
6. Conclusions
In this paper, an optimization approach for the design of the accel-
erating and decelerating nozzles for ducted propeller applications has
been proposed. The design workflow consists of a parametric description
of nozzle shape, in a fully scriptable mesh generation tool based on the
OpenFOAM blockMesh library, in the efficient RANSE solver simpleFOAM
and in an automatic post-processing of the data together combined
through the modeFrontier optimization environment. A simplified rep-
resentation of the propeller action on the flow field (an actuator disk with
radially varying momentum sources) was used to achieve the necessary
computational efficiency required by a design through optimization
involving RANSE calculations. A preliminary validation confirmed the
reliability of the computational model with all its simplifications (wedge
type calculations, GAP model).
Traditionally, accelerating nozzles are designed mainly for highly
(tip) loaded propellers while for decelerating type shapes the availability
of design guidelines is even more scarce. The use of tip unloaded pro-
pellers, motivated by increasingly important side effects, furthermore
stresses the design requirements of nozzles which may suffer from flow
separation and worsening of performance. The proposed design by
optimization meets these requirements: to provide a reliable tool in order
to customize the nozzle shape based on different objectives and con-
straints, allowing for the full exploitation of the potentialities of ducted
propellers also in the case of non-usual propeller functioning conditions
and loadings.
In the specific case of the accelerating duct, it was possible to identify
a set of trade-offs configurations for the multiple operative conditions
simultaneously addressed in the optimization. Improving the cavity
inception speed appeared as a possible objective, without any detri-
mental effects on thrust at both bollard pull and at the higher advance
coefficient selected to resemble the routinely functioning of the propel-
ler. Similarly, the optimization procedure demonstrated the possibility to
improve the propulsive efficiency at unvaried cavitation inception speed
19
Ocean Engineering xxx (2017) 1–20
by maximizing the delivered thrust of the nozzle and, in the end, by
redesigning the propeller in the more favorable (unloaded) functioning
condition allowed by the increase of the nozzle thrust.
A first step towards the combined design of the whole propulsive
system was proposed in the case of the decelerating nozzle. The adoption
of the actuator disk modeling avoids any possible interactions from the
propeller point of view, which instead, would be necessary to account for
the modification of the inflow to the propeller due, at first, to the mod-
ifications of the nozzle shape and vice versa. For the decelerating duct,
consequently, three simultaneous designs were carried out with different
propeller loading distributions at constant delivered thrust. A change in
the inflow turns immediately, indeed, into a modification of sectional
angles of attack. A preliminary assessment of the influence of a different
propeller on duct performance (or the investigation on how the duct has
to be modified to take advantage of a particular loading distribution) was
possible by simply modifying the momentum source distribution. Also in
this case, the outcomes of the design by optimization were encouraging.
It was demonstrated that a dedicated nozzle design, as in the previous
case, supports simultaneously the increase of the inner pressure and the
reduction of drag. Moreover, the opportunity to exploit even small
modification of the radial load distribution of the propeller to further
increase the propulsive performance was verified, showing the benefits
of a custom design based on current requisites. In addition, in both the
accelerating and the decelerating case, useful correlations between usual
design parameters and nozzle performance were identified to support
design choices.
Dedicated fully resolved RANSE calculations mainly confirmed the
outcomes of the optimization for both the configurations. Trends, rather
than absolute values, and, in general, the ranking of the analyzed optimal
geometries, were verified, demonstrating the feasibility of the proposed
design by optimization for both accelerating and decelerating nozzles.
These analyses, however, confirmed some of the limitations discussed
throughout the paper, of the adopted simplified models and the necessity
of dedicated investigations on few critical points was evidenced. The role
of the tip loading in case of accelerating duct geometries, especially for
what regards the flow features on the nozzle inner surface, would require
dedicated analyses even if the use of highly tip unloaded propellers
would nullify the scope of these propulsors, devised for heavy loading
functioning. Within the simplifications of the actuator disk model, opti-
mization would provide also in this case useful design guidelines and the
opportunity to customize the nozzle shape. RANSE calculations, more-
over, clearly shown a certain degradation of the propeller performance
with specific nozzle shapes (those delivering the highest thrust) in both
accelerating and decelerating configurations, which obviously could not
be evidenced by a fixed momentum source distribution.
A redesign of the propeller geometry in the actual inflow for any
analyzed nozzle geometry (turning into an updated body forces distri-
bution) or a modification of the body forces radial distribution based on
the performance of the given propeller subjected to a modified inflows
appear to be the first unavoidable steps for even more reliable and per-
forming designs. The propeller could take advantage of a more favorable
inflow, of a higher static pressure and of a global unloading of the blade:
these achievements are due to the different design point allowed by the
increased thrust (or reduced drag) provided by the optimal nozzle. A
feedback mechanism in the optimization loop, indeed, would be defi-
nitely necessary in order to account for the mutual influence between the
propeller and the newly designed ducts to carry out a combined design
for constant delivered thrust. The simpler procedure could consist in a
computationally efficient lifting line design tool embedded into an inner
calculation loop. For any nozzle shape, a certain number of iterations
allows defining the optimal propeller loading distribution dependent by
the nozzle inflow to the propeller, in turn, influenced by the performance
of the propeller itself. Dedicated fully resolved RANSE calculations,
extended to a wider set of candidates could be employed, finally, as a post
processing of the optimization results for the definitive choice of the
optimal configuration. More wisely, RANSE calculations could be
S. Gaggero et al.
adopted in order to continuously correct results throughout the design
process, increasing the reliability of the proposed procedure, leading to
the simultaneous design of the nozzle and of the propeller by employing
the well-established procedure for propeller optimization including the
systematic variation of the duct driven by high-fidelity viscous
calculations.
References
Abdel-Maksoud, M., Heinke, H.J., 2002, July. Scale effects on ducted propellers. In:
Proceedings of the 24th Symposium on Naval Hydrodynamics, Fukuoka, Japan,
pp. 744–759.
Abdel-Maksoud, M., Steden, M., Hundemer, J., 2010, September. Design of a multi-
component propulsor. In: Proceedings of 28th Symposium on Naval Hydrodynamics,
Pasadena, pp. 12–17.
Baltazar, J., Falc~
ao de Campos, J.A.C., Bosschers, J., 2012. Open-water thrust and torque
predictions of a ducted propeller system with a panel method. Int. J. Rotating Mach.
2012 https://doi.org/10.1155/2012/474785. Article ID 474785, 11 pages.
Baltazar, J., Rijpkema, D., Falcao de Campos, J.A.C., Bosschers, J., 2013, May.
A comparison of panel method and RANS calculations for a ducted propeller system
in open-water. In: Proceedings of the Third International Symposium on Marine
Propulsors (SMP2013), Launceston, Tasmania, Australia.
Baltazar, J., Falcao de Campos, J.A.C., Bosschers, J., 2015. Potential flow modelling of
ducted propellers with a panel method. In: Proceedings of the Fourth International
Symposium on Marine Propulsors (SMP2015). Austin, Texas, USA.
Bhattacharyya, A., Krasilnikov, V., Steen, S., 2016a. Scale effects on open water
characteristics of a controllable pitch propeller working within different duct designs.
Ocean. Eng. 112, 226–242.
Bhattacharyya, A., Krasilnikov, V., Steen, S., 2016b. A CFD-based scaling approach for
ducted propellers. Ocean. Eng. 123, 116–130.
Bontempo, R., Cardone, M., Manna, M., Vorraro, G., 2014. Ducted propeller flow analysis
by means of a generalized actuator disk model. Energy Procedia 45, 1107–1115.
Bontempo, R., Cardone, M., Manna, M., 2016. Performance analysis of ducted marine
propellers. Part I–Decelerating duct. Appl. Ocean Res. 58, 322–330.
Bosschers, J., Willemsen, C., Peddle, A., Rijpkema, D., 2015. Analysis of ducted propellers
by combining potential flow and RANS methods. In: Proceedings of the Fourth
International Symposium on Marine Propulsors (SMP2015), Austin, Texas, USA.
Çelik, F., Dogrul, A., Arõkan, Y., 2011. Investigation of the optimum duct Geometry for A
Passenger ferry. In: Proceedings of the 9th Symposium on High Speed Marine
Vehicles.
Esteco, S.R.L., 2016. ModeFRONTIER 2016 User's Manual.
Gaggero, S., Villa, D., 2017. Steady cavitating propeller performance by using
OpenFOAM, StarCCMþ and a boundary element method. Proc. Inst. Mech. Eng. Part
M J. Eng. Marit. Environ. 231 (2), 411–440. https://doi.org/10.1177/
1475090216644280.
Gaggero, S., Rizzo, C.M., Tani, G., Viviani, M., 2012. EFD and CFD design and analysis of
a propeller in decelerating duct. Int. J. Rotating Mach. 2012 https://doi.org/
10.1155/2012/823831. Article ID 823831, 15 pages.
Gaggero, S., Rizzo, C.M., Tani, G., Viviani, M., 2013, May. Design, analysis and
experimental characterization of a propeller in decelerating duct. In: Proceedings of
the Third International Symposium on Marine Propulsors (SMP2013), Launceston,
Tasmania, Australia.
Gaggero, S., Gonzalez-Adalid, J., Sobrino, M.P., 2016a. Design of contracted and tip
loaded propellers by using boundary element methods and optimization algorithms.
Appl. Ocean Res. 55, 102–129.
Gaggero, S., Gonzalez-Adalid, J., Sobrino, M.P., 2016b. Design and analysis of a new
generation of CLT propellers. Appl. Ocean Res. 59, 424–450.
Gaggero, S., Villa, D., Viviani, M., 2017. An extensive analysis of numerical ship self-
propulsion prediction via a coupled BEM/RANS approach. Appl. Ocean Res. 66,
55–78.
Gaggero, S., Tani, G., Villa, D., Viviani, M., Ausonio, P., Travi, P., Bizzarri, G., Serra, F.,
2017b. Efficient and multi-objective cavitating propeller optimization: an application
to a high-speed craft. Appl. Ocean Res. 64, 31–57.
Goldstein, S., 1929. On the vortex theory of screw propellers. In: Proceedings of the Royal
Society of London, Series a, vol. 123, pp. 440–465, 1929.
Haimov, H., Vicario, J., Del Corral, J., 2011, June. Ranse code application for ducted and
endplate propellers in open water. In: Proceedings of the Second International
Symposium on Marine Propulsors.
Hoekstra, M., 2006. A RANS-based analysis tool for ducted propeller systems in open
water condition. Int. Shipbuild. Prog. 53 (3), 205–227.
20
Ocean Engineering xxx (2017) 1–20
Hughes, M.J., 1993. Analysis of Multi-component Ducted Propulsors in Unsteady Flow.
PhD. Thesis. Massachusetts Institute of Technology, USA.
Hughes, M.J., 1997. Implementation of a special procedure for modeling the tip clearance
flow in a panel method for ducted propellers. In: Proceedings of the Propellers/
Shafting ’97 Symposium, Virginia Beach, Va, USA.
Kerwin, J.E., Kinnas, S.A., Lee, J.-T., Shih, W.-Z., 1987. A surface panel method for the
hydrodynamic analysis of ducted propellers. Trans. Soc. Nav. Archit. Mar. Eng. 95.
Kinnas, S.A., Chang, S.-H., Yu, Y.-H., He, L., 2009. A hybrid viscous/potential flow
method for the prediction of the performance of podded and ducted propellers. In:
Proceedings of Propeller/Shafting ’09 Symposium. Virginia Beach, VA, USA.
Kinnas, S.A., Yu, X., Tian, Y., 2012a. Prediction of propeller performance under high
loading conditions with viscous/inviscid interaction and a full wake alignment
model. In: Proceedings of 29th Symposium on Naval Hydrodynamics, Sweden.
Kinnas, S.A., Jeon, C.-H., Tian, Y., 2012b. A hybrid viscous/potential flow method for the
prediction of the wetted and cavitating performance of ducted propellers. In: SNAME
Propellers/Shafting Symposium 2012, VA, USA.
Kinnas, S.A., Jeon, C.H., Purohit, J., Tian, Y., 2013. Prediction of the unsteady cavitating
performance of ducted propellers subject to an inclined inflow. In: Proceedings of the
Third International Symposium on Marine Propulsors (SMP2013), Launceston,
Tasmania, Australia.
Kort, L., 1934. Der Neue Dusenschrauben-antrieb. Werft, Reederei und Hafen.
Menter, F.R., Kuntz, M., Langtry, R., 2003. Ten years of industrial experience with the SST
turbulence model. Turbul. heat mass Transf. 4 (1), 625–632.
Olivucci, P., Gaggero, S., 2016. A framework for the design by optimization of hydrofoils
under cavitating conditions. In: Proceedings of the 7th European Congress on
Computational Methods in Applied Sciences and Engineering, ECCOMAS Congress
2016. Crete, Greece.
Oosterveld, M.W.C., 1970. Wake Adapted Ducted Propellers. Tech. rep. 345. Netherlands
Ship Model Basin, Wageningen, Netherlands.
OpenFOAM Foundation, 2017. OpenFOAM 4.1.0. The OpenFOAM Foundation. www.
openfoam.org.
Roache, P.J., 1994. Perspective: a method for uniform reporting of grid refinement
studies. ASME J. Fluids Eng. 116.
S
anchez-Caja, A., Pylkk€anen, J.V., Sipil€a, T.P., 2008, October. Simulation of the
incompressible viscous flow around ducted propellers with rudders using a RANSE
solver. In: In 27th Symposium on Naval Hydrodynamics.
S
anchez-Caja, A., P
erez-Sobrino, M., Quereda, R., Nijland, M., Veikonheimo, T., Gonz
alez-
Adalid, J., Saisto, I., Uriarte, A., 2013. Combination of pod, CLT and CRP propulsion
for improving ship efficiency: the TRIPOD project. In: Proceedings of Third
International Symposium on Marine Propulsors (SMP2013), Launceston, Tasmania,
Australia.
Stipa, L., 1932. Experiments with intubed propellers. L’Aerotecnica 1932, 923–953.
Translated by D. M. Miner, NACA TM 655.
Tamura, Y., Nanke, Y., Matsuura, M., Taketani, T., Kimura, K., Ishii, N., 2010.
Development of a high performance ducted propeller. In: Proc. 21st Int. Tug Salvage
Convent. Exhibit. (ITS), pp. 1–8.
Van Manen, J.D., 1954. Open Water test series with propellers in nozzles. Int. Shipbuild.
Prog. 1 (3).
Van Manen, J.D., 1957. Recent research on propellers in nozzles. Int. Shipbuild. Prog. 4
(36).
Van Manen, J.D., 1962, May. Effect of Radial Load Distribution on the Performance of
Shrouded Propellers. Paper 7, Transactions of the Royal Institution of Naval
Architects.
Van Manen, J.D., Oosterveld, M.W.C., 1966. Analysis of ducted propeller design. Trans.
SNAME 74, 522–561.
Van Manen, J.D., Superina, A., 1959. The design of screw propellers in nozzles. Int.
Shipbuild. Prog. 6 (55), 95–113.
Vernengo, G., Bonfiglio, L., Gaggero, S., Brizzolara, S., 2016. Physics-based design by
optimization of unconventional supercavitating hydrofoils. J. Ship Res. 60 (4),
187–202.
Villa, D., Gaggero, S., Brizzolara, S., 2011, March. Simulation of ship in self propulsion
with different CFD methods: from actuator disk to potential flow/RANS coupled
solvers. In: Proceedings of International Conference-developments in Marine CFD
RINACFD2011, London England, pp. 1–12.
Weller, H.G., Tabor, G., Jasak, H., Fureby, C., 1998. A tensorial approach to
computational continuum mechanics using object-oriented techniques. Comput.
Phys. 12 (6), 620–631.
Yu, L., Greve, M., Druckenbrod, M., Abdel-Maksoud, M., 2013. Numerical analysis of
ducted propeller performance under open water test condition. J. Mar. Sci. Technol.
18 (3), 381–394.
Zondervan, G.J., Hoekstra, M., Holtrop, J., 2006. Flow analysis, design and testing of
ducted propellers. In: Proc., Propellers/Shafting 2006 Symposium, Virginia, USA.