Aerodynamics of Sailing Yachts

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 96

34564476 4 14 44150144615

4 813 46430
!"#

$%&'()*+,-./0'102+-3-*40
5+.67/
89:;<=:<>>?:<@AB>@CD>E@=<F<@>?:<F<AG>
@>H:CBI<H<GJ<<A<K=<:CF<AG>@ABA?F<:C;@D
>CF?D@GC9A>
LM$NOP2OQ0RS$OQTQNP

UVW0MQ5$X0SN2VSVYVP0QL0VPOWNQXQR5
2OWQQX0QL0PNRSNPPMSNR02OSPNOP2
Abstract
The present work reports the first systematic results obtained in wind tunnel and at full scale with the
experimental apparatus developed within the joint project among Politecnico di Milano, North Sails and
CSEM. The steady state upwind aerodynamics of sailing yachts are investigated through the contemporary
measurement of global forces, distributed pressures and sail flying shapes, and with numerical simulations
based on Potential Theory and Navier-Stokes equations.

The wind tunnel of Politecnico di Milano and the Sailing Yacht Lab (SYL), used for full scale investigations,
are described together with the measurement systems adopted for recording forces, pressures and sail
shapes. The aerodynamic loads are obtained through dedicated arrangements of load cells; the pressure
distributions on sail sections are evaluated with integrated systems of customized local measurement
solutions and MEMS sensors; the sail flying shapes are detected thanks to two laser scanners based of the
Time of Flight technology. The experimental procedures adopted during tests are presented and discussed
in relation with the aim of the work.

Numerical simulations of selected wind tunnel cases are performed in order to assess the capabilities of the
empirical techniques to validate Computational Fluid Dynamics (CFD) codes and to propose solid numerical
set-ups for investigating sailing yacht aerodynamics. A Vortex Lattice Method (VLM) code, written in
Matlab, is used for quick preliminary analyses of the global forces developed by the sail plan, whereas the
open source environment OpenFOAM is adopted to perform 3D simulations for investigating in details the
local flow patterns.

The experimental apparatus, both at model and full scale, proved to be extremely well suited for the
purposes of the study, giving remarkable results regarding, in particular, pressures and sail shapes. The
expected distributions are obtained during wind tunnel tests and interesting considerations arise from the
comparison with the full scale outcomes. Numerically, the results of simulations meet the preliminary
intuitions with the VLM code capable of accurately predicting global forces up to certain wind angles and
the RANS-based computations providing notable agreement with the measured local pressures.
Contents
1 Introduction ................................................................................................................................... 1
1.1 State of the Art .................................................................................................................................. 2
1.2 Utility and Aim of the Research ......................................................................................................... 5
2 Background Theory ......................................................................................................................... 6
2.1 The Yacht Motion .............................................................................................................................. 6
2.2 Sail Aerodynamics.............................................................................................................................. 8
3 Experimental tests ........................................................................................................................ 17
3.1 Wind Tunnel & Sailing Yacht Lab ..................................................................................................... 17
3.2 Testing Procedure............................................................................................................................ 19
4 Measuring systems ....................................................................................................................... 21
4.1 Force Measurements....................................................................................................................... 21
4.2 Pressure Measurements.................................................................................................................. 23
4.3 Sail Shape Detection ........................................................................................................................ 29
5 Data Management ........................................................................................................................ 31
5.1 Pre-process ...................................................................................................................................... 31
5.2 Acquisition ....................................................................................................................................... 32
5.3 Post-process .................................................................................................................................... 34
6 Numerical Methods ...................................................................................................................... 40
6.1 VLM Method .................................................................................................................................... 40
6.2 RANS Equations ............................................................................................................................... 44
7 Results ......................................................................................................................................... 55
7.1 Wind Tunnel Measurements ........................................................................................................... 56
7.2 Comparison of Wind Tunnel Measurements and Numerical Simulations ...................................... 67
7.3 Numerical Results ............................................................................................................................ 75
7.4 Full-Scale Measurements ................................................................................................................ 80
7.5 Comparison of Full Scale and Wind Tunnel Measurements............................................................ 85
8 Conclusions .................................................................................................................................. 87
9 References ................................................................................................................................... 89
1 Introduction
Despite their thousand-year old history, sailboats are very complex machines and an accurate study of their
motion still represents an open challenge for the scientific community. The main difficulty in the analysis of
these systems lies in the fact that their components interact simultaneously with two fluids, air and water.
This interaction is closely interconnected, therefore the overall performance of the boat cannot be
determined by the performance of any of its sub-system considered in isolation ([1]). The most common
strategy adopted to tackle this difficulty is to analyse separately the forces originated from wind, from
water and the ones due to weight and buoyancy. Once these are measured or estimated, comprehensive
equilibrium equations can be written in order to predict the motion of the overall system.

The present work proposes a novel method for a complete and extensive investigation of the sail plan
aerodynamics, whose study has historically been tougher than the hydrodynamics of the hull. In fact,
research about bodies moving through water is a fairly old practice; already in 1500, Leonardo Da Vinci was
interested in comparing different ship models and, during the 19th century, many towing tanks have been
built and used to test scale models of any type of hulls. To the contrary, at that time it was not possible to
reach the same level of research regarding the interaction between sails and wind due to the greater
complexity of the problem. The first experiments on sail plans using scale models date back to no more
than 50 years ago, when C.A. Marchaj and his research team at Southampton University performed tests in
a wind tunnel for the first time ([2]). From those first pioneering attempts, the study of sails aerodynamics
has been mainly motivated by the intent of optimizing a small number of high performing racing yachts;
only in the recent years, with the development of computers and numerical methods, the subject has been
broaden to a more commercial scale involving sail designers and smaller research teams.

In spite of the arrival of new techniques for studying sail aerodynamics, still nowadays sails are difficult
objects to study, commonly classified only in terms of global forces that they can produce in different wind
conditions and trims. Even though overall loads are important pieces of information, especially for naval
architects who run VPPs, they are far from being a complete description of the flow around sails. In fact,
global forces are only the sum of what happens locally, but the same overall effects might originate from
very different phenomena at smaller scales. Information of this kind are fundamental, in particular for sail
designers and sail makers who work for optimising the sail shapes: it is crucial for their applications to
deeply understand how the flow develops around the sails, how these interact with each other and the
effects the mast and rigging have on the sheets. Since these information are extremely difficult to obtain, in
the past, sails have been mainly designed and manufactured based on the experience of craftsmen and
sailors, and their work has always been considered a sort of art. This approach has certainly led to beautiful
shapes and great performances, but there is no doubt that large improvements might arise from the study
of sails’ aerodynamics from a more scientific point of view. Following this idea, in the recent years, all the
largest sail maker groups around the world have undertaken the challenge of engineering their design and
production procedures.

Within this context, Politecnico di Milano, which has been playing an active role in the sailing research
sector since the wind tunnel was launched in 2001, has recently established a collaboration with the world
leading sail maker North Sails and the Swiss technology organisation CSEM, specialised in manufacturing
small electronics components. The aim of the partnership is to set up an investigation apparatus that can
be used both in wind tunnel and at full-scale for studying sail yacht aerodynamics in a novel and more
accurate way, opening the path to a new understanding of sails.

This collaboration is part of the project of Prof. Fossati and his research team (now led by the researcher
Ing. Ph.D. Sara Muggiasca), who decided to support wind tunnel sailing activities at Politecnico with full-
scale experimental data. For the purpose, a dedicated nautical research and training centre was launched in
1
2015 at the university’s campus in Lecco ([3]): following similar successful experiences around the world
([4], [5], [6]) a dynamometric boat, named Sailing Yacht Lab, has been designed and built within the PhD
thesis project of the Ing. Ph.D. Ilmas Bayati ([7]). In order to support the activities, in parallel a novel sail
shape detection system has been developed by Ing. Ph.D. Ambra Vandone ([8]). With the contributions
from North Sails and CSEM, the joint efforts resulted in a new and comprehensive experimental apparatus
for the contemporary measurement of forces, pressure distributions and sail flying shapes ([9]).

The present work describes the overall apparatus and reports the outcomes from the experimental
campaigns performed in wind tunnel and at full scale in 2016. The great amount of experimental data is
then compared with results from numerical simulations, carried out with dedicated computational fluid
dynamic tools.

1.1 State of the Art


It is very difficult to study contemporary all the aspects of the yacht’s motion due to its simultaneous
interaction with water and air. Therefore, the research activity in the sailing field is widely based on the
procedure of obtaining hydrostatic, hydrodynamic and aerodynamic forces separately, and only
successively composing them together to obtain the overall dynamic equilibrium that describes the yacht’s
motion. In this context, the present research project focuses on the solely aerodynamics of the sail plan in
steady state conditions, i.e. under the assumption of absence of time dependent phenomena.

In the last 25 years, the increasing computational power and development of numerical methods have
shifted the attention of the research community from empirical studies to virtual aerodynamic simulations,
bringing with them both positive and negative considerations. Certainly, thanks to CFD (Computational
Fluid Dynamic) investigations, it has been possible to easily visualize flow fields around objects and to
perform much faster and less costly analyses but, on the other hand, the numerical results need to be
critically examined and not considered as unquestionable truths. Every numerical code is in fact inherently
based on assumptions and simplifications, and therefore its validity shall be carefully assessed against
empirical evidences.

For this reason, it is common practice to set-up experimental tests identifying specific parameters that can
be compared with numerical results in order to validate codes. This procedure usually involves the
measurement of global aerodynamic forces and, rarely, distributed pressures that are coupled with the sail
shapes responsible of their development in order to coherently compare experimental and numerical
results. This operation of coupling loads and shapes is far from being trivial when dealing with soft sails
since they inherently change geometry according to wind speed, wind direction and trim.

In the last years, many attempts to obtain reliable information about sail aerodynamics have been reported
in literature. Most of them were performed in wind tunnel facilities taking advantage of the controlled
environment, whereas others tried to set up experiments in real sailing conditions. All the techniques
developed throughout these works helps to constantly push forward the sailing research and represents
fruitful examples on which the entire community can rely and take advantage of for further developments.

In the following, some of the major experimental investigations reported in literature will be presented and
discussed in light of the scope of the present work. The studies are thus classified in terms of force and
pressure distribution measurement and sail shape detection systems since these are the three key aspects
around which the proposed experimental set-up is built. Moreover, some hints about the state of the art of
numerical simulations applied to sail plan study will introduce the reader to subject of CFD and the models
adopted in the present project.

2
1.1.1 Force Measurement
Until recently, the main objective of wind tunnel testing was to
measure global forces and moments developed by the sail plan on
boat models. From the early 1990s, the research group led by Prof.
Flay at the University of Auckland has been involved in several
America’s Cup campaigns developing methods and technologies for
these type of model-scale experimental studies ([10], [11]). At the
beginning of 2000s also the Politecnico di Milano took the challenge
to support some teams with its newly launched wind tunnel, further
sharpening the testing techniques for extrapolating aerodynamic
loads in different wind conditions ([1]). Fig. 1 - Yacht model in the test section of
the wind tunnel in Milan
In 1992 Milgram at MIT opened the path of full-scale
measurements with the first sail dynamometer boat, Amphetrete ([4]). This new measurement system was
developed for supporting the design of the several International America’s Cup Class (IACC) racing yachts,
being able to measure all forces and moments acting on the sails thanks
to an internal frame connected to the hull by means of six load cells. This
first pioneering attempt was followed some years later by Hochkirch and
Brandt in Berlin ([6]) with the dynamometer boat DYNA and by
Masuyama and Fukasawa ([5]) who built Fujin. All these instrumented
boats have been used in these years to gather data about both
hydrodynamic and aerodynamic loads investigating several sailing aspect
both under steady and unsteady assumptions. Given the extent of the
topic, in this section only the major works regarding the upwind steady
sailing conditions will be presented and discussed.

Milgram’s work ([4]) represented the first attempt to measure full-scale


forces and sail shapes and compute them with semi-empirical numerical
models. The detailed descriptions of the measurement system used on

Fig. 2 – Amphetetre. Schematic board, in the following years inspired other research teams around the
representation of frame and rig world. The German boat DYNA was used for investigating both the
showing the six measured forces
hydrodynamics of the hull and the aerodynamics of the sails. In
particular, Hansen and his group compared the aerodynamic coefficients obtained at full-scale and in wind
tunnel, focusing also on the interaction between sails and hull and on the importance of the depowering
procedure in wind tunnels in order to correlate model and full-scale results ([12], [13]). Furthermore, Clauss
and Heisen ran CFD simulations of DYNA in order to obtain numerically the full-scale forces; in that
occasion their sail shape detection system was based on photogrammetric methods that would have
inspired other colleagues thereafter ([14]). Masuyama and Fukasawa launched Fujin with mainly two
objectives in mind. On the one hand, they aimed at testing experimentally the mathematical model that
they had developed for simulating the aerodynamic force variation during manoeuvres ([15]) and, on the
other, they wanted to investigate the relationship between sail shape and performance for upwind
conditions. About this second scope, particularly noteworthy for the present study, they presented
remarkable plots providing comparisons of full-scale experimental forces and results obtained with two
different numerical methods, Vortex Lattice Method (VLM) and RANS-based simulations ([16], [17]).

1.1.2 Pressure Measurement


While many research groups were interested in evaluating performance of wind tunnel models and full-
scale boats in terms of global forces, others found interesting to concentrate on pressure distribution
measurements in order to deeply understand how the flow develops around the sail plan. This curiosity has
3
been legitimized by the steep growth of computational power that has allowed more and more accurate
numerical simulations that needed appropriate experimental results for comparison and validation. Some
pioneering qualitative studies of flow patterns around sails were carried out by Marchaj ([2]) using some
visualisation techniques and later by Milgram ([18]) who did tests in a pressurised recirculation channel;
however, the first experimentally relevant studies focusing on the flow along the chord of a sail and the
effects of the separation bubble produced by the mast, were performed by Wilkinson ([19]). In his
investigations, Wilkinson used a 2D rigid foil simulating the mainsail placed behind the mast and measured
the pressure distribution chord-wise. Despite the simplifications in the experiment, he was able to identify
the major flow characteristics along the chord, becoming the reference study for later investigations.
Within this area, one of the most active research groups in the last years has certainly been the pair formed
by Viola and Flay. Supported by the world leading infrastructures of the Yacht Research Unit of the
University of Auckland, they collaborated in a two-year project with the aim of investigating pressure
distributions around sails with all the three main methods used for sail aerodynamic studies: full-scale and
wind tunnel testing and numerical simulations ([20]). Focusing on their contribution on upwind sailing, at
full-scale they presented results from pressure distributions on jib and mainsail of a 24-foot sailing yacht,
relating the outcome to different wind directions and qualitative variations of sail trims ([21]). Concerning
wind tunnel tests, they have presented results of pressure measurements and compared them with
numerical simulations performed with a VLM code ([22]); afterwards, they proposed also more accurate
numerical results, obtained with RANS-based simulations ([23]). In order to provide the same inputs for the
two methods and to overcome the complications related to the installation of pressure taps on soft sails,
they decided to manufacture and instrument rigid sails resembling flying shapes of regular sails under the
effect of wind.
In the same years, another group of researchers in New Zealand measured pressure on sails, with the aim
of deducing form these data, the global aerodynamic forces acting on the yacht ([24]). In alternative to the
mentioned dynamometer boats, they claimed that this method would have allowed to measure forces
acting on any vessel, supporting its wider practicality and applicability. Besides its originality, the study is
worthy of attention from our perspective as it reports simultaneous pressure measurements and sail shape
detections at full scale.

1.1.3 Sail Shape Detection


As it has been briefly discussed previously, if the scope of an investigation is to evaluate forces or pressures
acting on sails, it is also crucial to know the actual shapes responsible of these loads. Unfortunately, the
design model used by the sail maker to manufacture the
cloth cannot be fully taken as reference, since it significantly
differs from the actual shape in navigating conditions. This
actual shape, named flying shape, is extremely difficult to
obtain and verify because depends on many contingencies
such as wind strength, boat attitude and sheets’ regulations
and “disappears” as soon as the wind stops to blow. In wind
tunnel testing, in order to overcome the problem of
detecting the flying shape in real time, a possible solution is
to adopt rigid sails whose shape is fixed and know ([22]).
This arrangement removes the issue of the shape
measurement but, besides the limited applicability to wind
tunnel experiments (cannot be used in full scale), it
introduces a higher level of approximation from the actual Fig. 3- Photograph of the rigid sail plan tested at YRU
sails’ behaviour and narrows the range of configurations in Auckland by Viola and Flay

4
that can be tested with the same model. For these reasons, alternative strategies for detecting sail shapes,
have been developed and applied by other research teams. Photogrammetry based techniques, for
example, have been employed in wind tunnel tests showing good results in controlled environments ([25],
[26]). The method consists in attaching several markers to the sails and obtain their coordinates via several
cameras that evaluate their position in space. In some cases however, the number and location of the
cameras might become quite large and inconvenient for full scale applications. Therefore, different
techniques, based for example on regular camera images processing, such as VSPARS and ASA, have been
developed. In particular, the base line principle of most of these methods consist in attaching fluorescent
stripes to the sails at known vertical locations and obtain their coordinates by post-processing the images
taken with the cameras ([27], [28], [29]).

1.1.4 Numerical Simulations


In the last decades, the main goal of experimental tests has shifted to representing validation bases for the
increasingly numerous mathematical models developed to predict the flow around sails. Focusing solely on
upwind steady state conditions, nowadays the two most applied numerical tools are the Vortex Lattice
Method (VLM), based on potential flow theory, and RANS simulations that solve on discrete volumes the
Raynolds Averaged Navier-Stokes (RANS) equations ([30], [31]).
The former was introduced by Milgram and Gentry at the end of the sixties, whereas the latter has grown
together with the increase of computational capabilities
in the last decades and many examples can be found in
literature ([23], [32], [33]). Potential flow codes have
proved to be very powerful in situations where the flow
remains attached to the sail providing good estimations
of the global aerodynamic forces in very short
computational times. On the other hand, RANS
simulations allow more detailed descriptions of the flow
phenomena around the sail plan and become
indispensable to accurately calculate the fluid motion
field in those areas where viscosity plays a significant
Fig. 4 – Streamlines and pressure distributions as results of role such as in separated flow regions and sails tips.
simulations performed by Roux

1.2 Utility and Aim of the Research


To the author’s knowledge, this is the first work that includes simultaneous acquisition of aerodynamic
forces, pressure distributions and sail shapes on soft sails both in wind tunnel and at full scale. In particular,
the research activity carried on in cooperation with North Sails and CSEM, not only aimed at developing a
complete measurement system to support sail designer activities, but also candidates itself to become a
reference experimental study for validating advanced numerical tools for steady state upwind sailing
conditions. This has been possible thanks to the fine research equipment available at Politecnico, the wind
tunnel and Sailing Yacht Lab, together with the systems developed for measuring forces and pressures, and
determining sail flying shapes. The experimental work has been extensively integrated with the simulations
performed with two numerical schemes based on potential flow and RANS equations that had been
recently developed within two other thesis projects at KTH and Politecnico, respectively ([34], [35]).

It should be noted that the simplest case of steady state upwind conditions, analysed in the present work,
represents only the first step of a more general study that, in the future, might involve similar techniques
also for downwind and dynamic investigations.

5
2 Background Theory
In order to fully appreciate the present work, it is fundamental to understand the main fluid mechanics
principles that govern the flow around the sail plan in upwind conditions. Moreover, a brief summary of the
principal concepts of sail boat motion will help the reader to think about the aerodynamics of sails not as a
stand-alone problem, but as a part of the larger physical system, like a yacht is. This last aspect will be the
starting point of our discussion.

2.1 The Yacht Motion


A sail boat is a physical system that is able to move forward thanks to the simultaneous interaction
between its components and the two fluids it is immersed in. In particular, the hull and various appendages
are submerged in water, while sails, mast and rigging are invested by air. The relative movement between
yacht’s elements and the two fluids generates different kind of forces: the aerodynamic loads are the ones
produced by the interaction with air; the hydrodynamic ones are due to water. In addition, even when not
in motion, the boat is subjected to the gravitational field, responsible of its weight, and immersed in water,
which makes it receiving a thrust directed upwards equal to the weight of the fluid displaced (Law of
Archimedes), called buoyancy force.

In case of steady state conditions, i.e. in situations without variations of boat and/or wind speed either in
direction or modulus such as sailing in light wind and calm water, these forces summed together must
satisfy a dynamic equilibrium. On the other hand, if accelerations or any kind of other time-dependent
phenomena take place, also inertial effects must be considered for equilibrium. In this work we will
consider steady state conditions only.

Since a boat might be seen as a rigid body moving in space, its motion can be defined with 6 degrees of
freedom: 3 translations and 3 rotations along the axes of the conventional clockwise reference system
𝑥𝑦𝑧 with the origin in the centre of gravity of the boat, the x-axis lying along the boat’s centreline and
pointing forward and the z-axis pointing upwards (Fig. 5).

Fig. 5 - Degrees of Freedom of the boat

6
Therefore, the equilibrium of the yacht might be expressed in terms of 6 scalar equations containing all the
mentioned forces, along the three axes 𝑥, 𝑦 and 𝑧:

1. Surge: ∑ 𝐹𝑥 = 0;
2. Sway: ∑ 𝐹𝑦 = 0;
3. Heave: ∑ 𝐹𝑧 = 0;
4. Roll: ∑ 𝑀𝑥 = 0;
5. Pitch: ∑ 𝑀𝑦 = 0;
6. Yaw: ∑ 𝑀𝑧 = 0;

The weight and buoyancy forces, directed downwards and upwards, and applied at the centre of gravity
(CG) and centre of buoyancy (CB) respectively, follow the laws of basic hydrostatics and essentially cancel
out each other in the Heave equation. On the other hand, aerodynamic and hydrodynamic loads generate
due to the relative motion between boat, air and water are expressed by fluid dynamic concepts as it will
be explained in the following: by projecting these two dynamic forces on 𝑥 and 𝑦 directions, equations 1
and 2 are set. The rotational equilibrium, expressed in the last three identities, depends on the intensity of
all force and on their points of application.

Aero and Hydrodynamic loads have many aspects in common even though they are generated by two
different fluids. The aerodynamic force 𝐹𝐴 is caused by the action of the apparent wind, the air “felt” by the
boat, and might be decomposed into two components: one large perpendicular to the flow direction,
Aerodynamic Lift and one smaller parallel to the wind, Aerodynamic Resistance (Fig. 6).

Fig. 6 - Interaction of wind and sails

Moreover, the force might be thought as applied in a single point on the sail plan called centre of
aerodynamic effort (𝐶𝐸 ). On equal terms, the appendages behave like wings under water generating the
hydrodynamic force 𝐹𝐼 applied in the centre of lateral resistance (𝐶𝐿𝑅 ). Again, it is common to identify two
components, the Hydrodynamic Lift, perpendicular to the flow, and the Hydrodynamic Resistance, parallel
to it. In this case, it is interesting to note that since the appendages below the water level are symmetric, in
order to generate a lifting force, these surfaces need an angle of attack with respect to the fluid which is
represented by the leeway angle 𝜆. This angle, which might be seen as the dependent variable in Sway and
Yaw equations, is the relative angle between the longitudinal axis of the boat and its course direction.

Even though aero and hydrodynamic lift and drag are sufficient to characterize the dynamic loads acting on
the boat, it is convenient to project them along the 𝑥 and 𝑦 directions: in this manner, it is clear which
component of the forces contributes to the forward movement and thus to the yacht velocity, 𝐹𝑥 and which
one is responsible of the heeling of the boat, 𝐹𝑦 (Fig. 7).

7
Fig. 7 - Panorama of the Aero and Hydrodynamic forces applied to the yacht under way

2.2 Sail Aerodynamics


Now, that the general picture of the main forces acting on a navigating yacht is clear, the discussion will
focus on the aerodynamics of the sail plan only. In particular, the present work deals with the sail behaviour
in upwind conditions, disregarding what happens at large wind angles.

We can say that sailing boats take advantage of the wind in different ways depending on the angle between
the yacht’s course and the wind. In case of small angles (< 90°), the boat is said to sail in upwind conditions,
while for large angles (> 90°), we talk about downwind conditions. It is important to differentiate these two
situations, because the aim of the sails in producing aerodynamic forces changes, and so does the
composition of the force that allows the boat to move forward. In case of upwind navigation, sails act as
the wings of airplanes, i.e. like airfoils: the incoming airflow interacts with them remaining attached for the
major part of the sail area, thus creating a large lift force perpendicular to the flow itself and a small
component of drag force, oriented parallel to it (Fig. 8). On the other hand, when sailing downwind, the
objective is to produce as much aerodynamic force as possible and, in order to achieve it, a large angle of
attack between the airflow and the sails is desired. In such conditions, major portions of the sheets are
characterised by flow separation, thus behaving like “bluff bodies”. The drag force produced is of the same
order of magnitude as the lift component and their contribution to drive the boat is similar.

Fig. 8 – Similar behaviour of sails in upwind conditions and wings (airfoils)

8
Even though, the joint project between Politecnico, North Sails and CSEM aims at investigating all the
possible sailing headings, it was decided to initially focus on upwind conditions due to the less complexity of
the flow phenomena involved. Indeed, the “bluff body” behaviour is much more difficult to study since flow
separation areas are characterised by a high level of turbulence and thus, in large part, by randomness,
aspect very challenging to handle in experiments and to model virtually. Moreover, the airfoil-like
behaviour of sails when moving towards the wind, allows to take advantage of the numerous experimental
and numerical methods developed throughout the years for aeronautical applications. Therefore, in the
following we will concentrate on upwind sailing condition only, covering the main aerodynamic concepts
that govern it.

First of all, it is useful to clarify which sails are used in upwind conditions. Usually, the regular set of sheets
hoisted encompasses two components: the mainsail, a triangular or trapezoidal shaped sail fastened at the
mast on the front side (luff) and at the boom on the bottom, and the headsail, triangular and reeved to the
forestay, a cable connecting the bow and the top part of the mast (Fig. 9). This is the most common sailing
configuration and the one analysed in the present study.

Fig. 9 -Yacht sailing upwind, equipped with mainsail and headsail

In the following, upwind sailing aerodynamics will be discussed in details, subdividing the subject in two
different levels: firstly, we will go through the general fluid mechanics concepts at the base of lift and drag
production and then we will focus on the description of the local patterns that characterise the airflow
around the sail plan. In this manner, the reader will be able to appreciate the importance of studying sailing
aerodynamics not only from a global point of view in terms of integral forces, but also by investigating the
complex pressure distribution around jib, mast and mainsail.

9
2.2.1 Interaction between fluids and solids
At first, let us consider a sail just like a wing, an airfoil-shaped body with a certain thickness. Imagine to
immerge this object in a fluid stream that, for simplicity, is steady, i.e. time independent. As we have
already discussed, this means, among the rest, that the flow is not turbulent. In these conditions, the fluid
motion can be represented by its streamlines, which are lines whose tangent in any point represents the
fluid velocity at that point in space. In order to better visualise the situation, let us assume also a laminar
flow, condition for which each fluid particle has the same velocity as all the others, and thus the
streamlines are parallel.

When a solid object, like a wing, is introduced in the


flow, since the fluid cannot penetrate the object,
the streamlines are deviated to either side of it (Fig.
10). The effect of this deviation is “felt” by the
object in terms of pressure on its surface. In
particular, at each point of the object’s contour, a
certain pressure, called dynamic pressure (𝑝), acts
perpendicular to the surface itself. This action is the
result of the relative motion between the fluid and
the object and should not be confused with the
static pressure (𝑃𝑆𝑇 ), which acts even in absence of
movement. In order to quantify the dynamic
pressure and to better understand its relation with
the relative velocity between the fluid and the
object, a balance of the energy contained in the Fig. 10 – Laminar flow - Streamlines representation and
system must be considered. distributed loads on the foils surface

The total mechanical energy possessed by the fluid vein, under the assumption of incompressible fluid, is
the sum of potential, kinetic and pressure energy. The potential energy is the energy stored in the system
and is often related to its position; the kinetic energy is associated with the mass and velocity of the fluid;
finally, the pressure energy is linked with the static pressure of the fluid.

In incompressible fluid mechanics, the conservation of energy is expressed with the Bernoulli’s Law:
1
𝑃𝑆𝑇 + 𝜌𝑉 2 + 𝜌𝑔ℎ = 𝑐𝑜𝑛𝑠𝑡.
2
In many applications, including the present, the changes in height are very small; therefore, the potential
term is negligible and a simplified form of Bernoulli’s Law can be used:
1
𝑃𝑇𝑂𝑇 = 𝑃𝑆𝑇 + 𝑝 = 𝑃𝑆𝑇 + 𝜌𝑉 2 = 𝑐𝑜𝑛𝑠𝑡.
2
According to this equation, on each streamline the sum of pressure and kinetic energy remains the same at
each time. In other words, if in a certain point in space the velocity increases with respect to the
undisturbed flow, the static pressure will be lower and vice versa.

By calculating the static pressure at every point on the considered object’s surface and by integrating it
over the entire surface, the non-viscous force exerted on the object by the fluid can be evaluated: it is
called form drag, 𝑝, and is associated with the shape of the object. Usually, it is expressed at each point of
the object’s surface with a normal pressure drag coefficient 𝐶𝑅𝑃 (𝑠) different for each point considered:
1
𝑝 = 𝐶𝑅𝑃 (𝑠)𝜌𝑉 2
2
10
The other contribution to the total force is called frictional resistance, 𝜏, and is related to the fluid viscosity.
Again, this can be expressed in terms of a local coefficient, skin friction drag coefficient 𝐶𝑅𝐴 (𝑠):
1
𝜏 = 𝐶𝑅𝐴 (𝑠)𝜌𝑉 2
2
By integrating these two terms along the object and summing them together, it is possible to evaluate the
total force generated by the interaction between the fluid and the object.

In order to better understand the frictional resistance, we need to give a closer look at the fluid-solid
interface on the object surface. If we consider a fixed object, the fluid particles in contact with the object
must be fixed as well. As we move away from the object’s surface, the fluid particles’ velocity increases
until the effect of the presence of the object is not “felt” anymore. At that point the flow velocity will be
the same as the undisturbed flow. Therefore, we can identify a region close to the object surface in which
the flow velocity passes from zero to the value of the undisturbed flow with a certain gradient in the
direction perpendicular to the surface. This region, called Boundary Layer, is the only region interested by
shear stresses and thus the only one in which the flow viscosity has an effect (Fig. 11).

Fig. 11 - Boundary Layer development along a non-slip wall. (a) represents the incoming undisturbed flow. (b) shows how the flow
develops in the B.L. in conditions of positive pressure gradients. Moving downstream, the gradient becomes null due to viscous
effects (c). Further downstream, the negative pressure gradients induce the flow to separate (red area), therefore increasing
vortices are shed from the surface (d), (e).

Without going into details in the description of the Boundary Layer, we can say that it evolves in
characteristics and dimensions along the chord length of the object’s surface interface. Close to the leading
edge, the boundary layer is slim and laminar, i.e. the flow streamlines are parallel and behave like thin
sheets in relative motion; then, it starts to become more and more unstable and finally turbulent due to
the larger and larger mixing effects of the momentum exchange. In turbulent regions, there are large
vortices that generate a lot of frictional resistance due to the high velocity gradients.

The transition points of the boundary layer have been studied and theorised firstly by Reynolds, who
associated this transition to the ratio between inertial and viscous forces. The non-dimensional parameter
that governs this kind of problems is the well-known Reynolds number:
𝐿𝑉
𝑅𝑒 = 𝜈

For better understanding the topics treated in this study, it is also useful to add that, in case of long
surfaces or if the rear part of the body closes abruptly, at severe angles to the undisturbed flow, the
boundary layer detaches prematurely from the object surface increasing the production of turbulence. This
phenomenon is called flow separation, and the point at which it happens, separation point. In upwind
conditions, flow separation happens only is minor areas of the sails such as at the trailing edges or behind

11
the mast, whereas downwind configurations are largely interested by this phenomenon. This is one of the
main reasons why, it is a lot easier to study upwind sailing conditions rather than downwind ones.

2.2.2 Global forces


At this point of the discussion, it is clear how bodies like wings immersed in a fluid can generate forces. In
particular, we have highlighted the two terms that form the total force: dynamic pressure 𝑝 and frictional
drag 𝜏. As we have already pointed out, in aeronautical applications, these two contributions are
considered together in a single overall aerodynamic force
1
𝐹 = 𝜌𝐶𝐹 𝐴𝑉 2
2
and decomposed along the directions perpendicular and parallel to the flow, identifying lift and drag
respectively:
1
𝐿 = 𝜌𝐶𝐿 𝐴𝑉 2
2
1
𝐷 = 𝜌𝐶𝐷 𝐴𝑉 2
2
In sailing applications, it is also interesting to evaluate which portion of the total aerodynamic force
contributes to the forward movement of the boat (driving force) and which one is responsible of the
heeling of it (side force). Therefore, it is common practice to breakdown the total force 𝐹 along the
direction of the hull centreline (x-dir) and along the direction perpendicular to it (y-dir).
1
𝐹𝑥 = 𝜌𝐶𝑥 𝐴𝑉 2
2
1
𝐹𝑦 = 𝜌𝐶𝑦 𝐴𝑉 2
2
In the above formula, 𝐶𝐹 , 𝐶𝐿 , 𝐶𝐷 , 𝐶𝑥 , 𝐶𝑦 are all non-dimensional coefficients that allow to compare results
from different boats and testing conditions. We will discuss their role more in details when the
experimental and numerical outcomes will be presented.

2.2.3 From wings to sails


The discussed aerodynamic forces are the overall result of the interaction between sails and apparent wind.
Even though these are important pieces of information, they are not a complete and exhaustive picture of
what happen around the sail plan. For this reason, it is important to have a closer look to the local
phenomena that characterise the flow.

So far, sails have been considered like wings meaning that, at small angles of attack, they deflect the
airflow, creating regions of high pressure windward and low pressure leeward, which result in production
of lift and drag. In order to understand more in details these pressure distributions around sails though,
some important features that differentiate them from wings must be pointed out. First of all, sails thickness
in negligible, contrary to wings which might be also very thick depending on their application. Secondly,
sails are soft, which means not only that their shape depends on how they are trimmed, but also that their
geometry is influenced by many external factors, such as wind strength. Thirdly, sails surfaces might be
rather irregular due to their manufacturing process and the presence of trimming equipment like holes and

12
ropes. Last but not least, the mast has a great influence on the inflow of the mainsail and therefore must be
considered in the discussion.

All these aspects contribute to give rise to local phenomena and pressure distributions that are specific for
sails in upwind conditions and that differ for jib and mainsail, mainly due to their relative positions.

2.2.4 Pressure around the sail plan


Regarding the headsail, we will refer to Crompton and Barrets’s work ([36]) who deeply investigated the
phenomenon of separation bubble behind the sharp leading edge of a flat plate at incidence. On the other
hand, Wilkinson’s investigation ([19]) will be used as guideline for presenting the airflow around the main.
Both these studies were reviewed by Viola and Flay ([37]) who explained their experimental results in terms
of conventional thin airfoil theory and aerodynamics of separation bubbles.

When dealing with pressure distributions around the sail plan, the pressure is expressed in term of static
pressure coefficient 𝐶𝑝 , which in each point defines the ratio between the static pressure in that point
minus the static pressure in the undisturbed flow, and the dynamic pressure in the undisturbed flow:
𝑝 − 𝑝0
𝐶𝑝 =
1 2
2 𝜌𝑉∞
Recalling Bernoulli’s Law, the pressure coefficient can be expressed also in terms of velocities at the
location considered and the flow velocity at the far field such that:

𝑉 2
𝐶𝑝 = 1 − ( )
𝑉∞
From these expressions, it is immediate to see that if the flow accelerates in respect of the undisturbed
velocity 𝑉∞ , the pressure coefficient becomes negative, being the local static pressure lower than the
pressure in the undisturbed flow. On the other hand, when the flow slows down 𝐶𝑝 is positive and the
static pressure higher than in the far field.

Fig. 12 ([20]) summarises the general behaviour of the


pressure coefficient around jib and main. It presents
similar trends with minor differences due to the distinct
sail geometry and relative position: both sails present a
suction region, on the leeward side, where the flow is
accelerated and the pressure drops, and a region of
positive pressure on the windward side where the flow is
slowed down. At the leading edge of both sheets, the flow
detaches from the surface and reattaches further
downstream creating the so-called separation bubble,
which is a region of recirculating flow. At the trailing edge,
depending on the sail trim and the angle of attack with the
incoming air, the flow might separate again leading to
turbulent phenomena and vortex shedding.

Pressure distribution on Main


Wilkinson, in his work, focused on the flow behaviour
Fig. 12 - Schematic representation of typical Cp
distributions along headsail and mainsail sections
13
around the mast and mainsail. He identified different regions in which attached flow, separation bubbles,
reattaching phenomena and detached flow alternate each other. With reference to his diagram (Fig. 13),
nine zones can be identified, five leeward and four windward. Note that the vertical axis in the plot is
flipped, having negative values along the positive axis direction.

Fig. 13 - Schematic 2D representation of the pressure coefficient around


mast and a mainsail section identified by Wilkinson ([19])

 In Zone 1, the flow around the mast is deflected and initially accelerated such that the pressure
drops reaching a negative peak. At the beginning of this first section, a laminar boundary layer
develops along the solid surface. Afterwards, due to the blunt geometry of the mast, the flow
separates, causing a partial recovery of the pressure.
 Zone 2 is characterised by the separation bubble; the average fluid velocity in this region remains
almost constant, therefore no major pressure gradients are observed. Close to the downstream
end of the bubble, laminar to turbulent transition occurs.
 In Zone 3, the flow tends to reattach to the surface. This interaction slows the fluid down, therefore
the pressure partially recovers.
 Zone 4 presents a flow pattern in which the turbulent boundary layer grows until separation at the
trailing edge. In this region, the gentle curvature of the sail creates a suction region that accelerates
the flow until the change of slope. After that, the sail shape is responsible of slowing down the fluid
and consequently the pressure coefficient becomes less negative.
 In Zone 5, the flow might detach from surface; this phenomenon is visible in terms of pressure, as
the gradients are almost null.

14
 Zone 6 includes the windward face of the mast. The pressure behaviour is nearly the opposite as on
the leeward side. Here the flow is decelerated, reaching an “ideal” point in which the value of
velocity is zero; this is called “stagnation point” and presents by definition 𝐶𝑝 = 1.
 Zone 7 shows a region of constant pressure, consequence of the separation bubble that develops
also on this side of the mast for the same reasons described above.
 In Zone 8, the flow reattaches to the surface and, on this side, the sail geometry causes the
deceleration of the fluid; therefore the pressure coefficient progressively grows.
 Along Zone 9, the slowed flow remains attached to the surface presenting a pressure higher than in
the undisturbed flow. At the trailing edge windward and leeward pressures must be equal as
theorised by Kutta.

Pressure distribution on Jib


The qualitative description inspired by Wilkinson’s experiments on a 2D rigid foil placed behind a cylinder,
is partially valid also when outlining the flow patterns around the jib. Even though the headsail does not
have any large bluff body in front of it, the leading edge often presents a separation bubble and the high
curvature of the sail might lead to flow separation at the trailing edge.

Fig. 14 – Expected Cp distributions along the headsail sections at varying angles of attack ([37])

With reference to Fig. 14, the separation bubble on the leeward side occurs when the angle of attack is
higher than the ideal one, that is when the sail leading edge is not exactly aligned with the streamlines of
the incoming flow. In case of soft sails, a slightly over-tightened trim of this kind is desirable in order to
prevent the sheet to flap; therefore, separation bubbles are often present.

The sharp edge of the jib causes a longer bubble in respect to the one on the mainsail, characterised by a
large recirculating region with high backflow velocity and by a laminar to turbulent transition that occurs

15
further upstream in the bubble. This higher level of turbulence causes a more energetic reattachment of
the flow which results into a steeper positive pressure gradient.

If the jib is excessively over-trimmed, i.e. the Angle of Attack is too high, the LE separation bubble becomes
very large and the positive pressure gradient related to the reattachment decreases. Moreover, at the
trailing edge, the separation point moves further upstream creating a region of constant pressure, thus
reducing the suction peak related to the sail curvature. In extreme cases, the combination of these two
phenomena leads to a monotonic pressure recovery from the suction peak to the trailing edge.

On the windward side, the separation bubble is much less pronounced than on mainsail due to the absence
of mast, leading to a more constant positive pressure coefficient that decreases only towards the trailing
edge where windward and leeward pressures must join together. This point is characterised by a more or
less negative value of pressure depending on the level of flow separation, which in turn results from the sail
trim.

Sails interaction
Even though headsail and mainsail flow phenomena have been presented separately, considering the
interactions between the two is crucial for understanding the relative amount of aerodynamic loads
attributed to the sails. In particular, referring to the common upwind situation depicted in Fig. 12, it can be
seen that the leeward side of the jib is characterised by higher 𝐶𝑝 (more negative) compared to the same
side of the main. This “enhanced” efficiency of the headsail is to be entirely attributed to sails interaction:
in fact, the mainsail, producing lift, generates a double positive effect on the jib: on the one hand, the jib’s
stagnation point is shifted towards its windward side, virtually increasing its angle of attack (Up-Wash
Effect) and, on the other, the depressed leeward side of the main accelerates the flow at the trailing edge
of the jib and, consequently, on its entire leeward side. To the contrary, the headsail makes the main
stagnation point shifting towards the leading edge, with the effect of reducing peak suction velocities and
consequently pressure gradients (Down-Wash Effect).

These are, in synthesis, the major phenomena that develop chord-wise around a common sail plan in
upwind conditions; to present them, sails have been considered only in terms of horizontal sections. This
assumption is commonly accepted as a first order simplification when describing the aerodynamics of a
slender body with high aspect ratio; however, since sails have a finite span, some 3D flow features occur
close to their lower and upper tips. In these regions, the pressure difference between the windward and
leeward sides induces the flow to move from the high pressure zone to the lower pressure causing circular
patterns around the tips. These phenomena, known as tip vortices, dissipate energy and are responsible of
the so-called induced drag.

16
3 Experimental tests
Following the example of the presented studies, the aim of this project is to investigate sailing yacht
aerodynamics in a systematic and extensive way, trying to observe experimentally what has been
presented in theory. To do that, Politecnico di Milano, North Sails and CSEM has joint their efforts to set-up
empirical tests both at model and full scale at the two facilities of the Italian technical institution: the Wind
Tunnel in Milan and the Sailing Yacht Lab on Como Lake. The investigations performed on both sites aimed
at gathering accurate data regarding forces and pressures developed by the sail plan in upwind conditions
and contemporarily measuring the flying shapes responsible of such loads.

Wind tunnel and full scale campaigns have been performed and presented within the same project since
large parts of the instrumentation used in the two types of tests are very similar, even sharing, in some
cases, the same hardware and software.

The next paragraphs will firstly introduce the Wind Tunnel and SYL in terms of facilities and testing
procedures; afterwards, the measurement systems used for determining the three quantities of interest
will be described and thoroughly discussed.

3.1 Wind Tunnel & Sailing Yacht Lab


The Wind Tunnel at Politecnico di Milano
was launched in 2001 and, since then, has
been operated for any kind of Wind
Engineering related activity, including sail
boats ([42]). The facility was designed as a
closed circuit, arranged vertically, with two
different test sections: a high speed, low
turbulence section placed at the lower level
and a low speed boundary layer section,
above it; for sailing applications, only this
latter space is used (Fig. 15). It has a working
area that spans 40m in length, 14m in width Fig. 15 – Politecnico di Milano Wind Tunnel. Architecture of the test sections
and 4m in height. This large room allows to
test up to 1:10 scaled yacht models with low
blockage effects at a maximum speed of
15m/s (Fig. 16). Moreover, the long distance
between the 14 fans and the tested model
guarantees a very stable incoming flow. This
stability is achieved also in terms of
temperature thanks to the presence of a
heat exchanger that keeps air conditions
constant during tests.

Fig. 16 – 1:10 scaled yacht model in position for test

17
The Sailing Yacht Lab (SYL) is a 10m length sailing yacht fitted with instruments for acquiring data and
variables while sailing ([3]). It was designed and built within the PhD Thesis Project of Ing. Bayati, PhD ([7])
as part of the wider project of Lecco Innovation Hub which aimed at extending the nautical research
activities of Politecnico di Milano to full-scale applications (Fig. 17)(Fig. 18). The SYL’s primary objective was
to measure aerodynamic loads in steady conditions; therefore, it was decided to sail in the northern part of
Como Lake, off shore Colico (LC), where the environmental conditions of strong wind and very limited
waves resulted optimal for such activity.

Fig. 17 - Rendering of the Sailing Yacht Lab

Fig. 18 – The yacht handled with a lifting crane for maintenance operations

18
3.2 Testing Procedure
The aim of the tests in both environments was to generate a benchmark of cases reporting forces and
pressure distributions developed around the sail plan at different wind angles and trims, and relate them
with the correspondent flying shapes.

To do that, the procedures followed in wind tunnel and at full-scale were analogous, compatibly with the
inherent environmental operating differences. The acquisition time for example, well defined in the former
case, on the lake strongly depended on wind steadiness and helmsman skills, thus resulting to be different
from one trail to the other. Moreover, at model-scale all the acquisitions were formally identical in the
number of configurations and conditions tested, whereas at full-scale the researchers had to constantly
deal with variable wind and external disturbances, which resulted in more variable test cases. Besides these
minor differences, during acquisitions the same steps were followed.

In wind tunnel, the boat model was tested at different wind angles ranging from 20° to 35° both in upright
and heeled configurations, as reported in details in Tab. 1. For each angle, an envelope curve from
maximum to minimum thrust was obtained by gradually regulating the sails. In particular, since the work
aims at statically characterise sail boat aerodynamics, once the desired trim was achieved, the acquisition
started and lasted for 60s, in order to be able of properly averaging values in the post-process phase.
Before and after each AWA change, a zero-force measurement was performed and used in post-processing
to purify measurements from possible thermal measure deviations.

Tab. 1 – Summary of the configurations tested in wind tunnel

Apparent Wind Angle


Hull Configuration 20° 22.5° 25° 27.5° 30° 35°
Upright

Heeled

At full-scale instead, it was not possible to determine the wind angle in such a schematic way; therefore,
“steady” acquisition segments of 15s were extrapolated from longer acquisitions and individually analysed
in the post-process phase, verifying a-posteriori the correspondent AWA.

An important part of the testing procedure is represented by the method used for switching configurations
and hence, how sails were trimmed. In both cases, tests have been performed following the conventional
method of drawing the so-called “depower” curves (Fig. 19). In closed hauled sailing conditions, for every
AWA tested, the procedure consists firstly in finding the optimal trim (Max power), i.e. the one that gives
the maximum driving force (𝐹𝑥 ) the sail plan can develop, and then in changing trim to evaluate the boat
behaviour in different conditions. The pieces of information acquired in this way are extremely valuable for
practical applications, since in real-sailing situations often happens that environmental conditions, such as
strong wind or heavy sea, do not allow to best trim sails; therefore, for a complete characterisation of the
boat, is important to gather data regarding non-optimum operating conditions as well.

19
Fig. 19 - Example of Depower Curve. Usually, results are expressed in terms of coefficients Cx and Cy, rather than forces Fx and Fy.
On the graph, the different configurations tested are marked: “Max power” corresponds to the trim that generates the maximum
driving force; “Over-trimmed” configuration is the result of a too tight sail regulation; “Depower” points represent the easing
procedure of the main sail.

Regarding sail regulations, in wind tunnel, the trimmer remotely controlled electrical model scale winches
with a multiple-turn knob console from the control room with direct visibility on the model, constantly
monitoring the driving and heeling forces, 𝐹𝑥 and 𝐹𝑦 ; in this way, it was possible to know when the
optimum trim was achieved. At that point, in order to depower the sail plan, the two regulations that have
been decided to apply concerned the main sail only and consisted in easing the traveller and adjusting the
sheet tension. The former action influences rigidly the entire sail, varying its angle of attack with the
incoming flow, whereas the latter, operates on the sail twist, thus on the relative openness of different sail
sections to the wind. In general, from the optimum trim, by pulling in the traveller, the effect will be
negative because the driving force decreases and the heeling increases due to flow separation effects
caused by angles of attack too severe, especially on the upper parts of the main surface. This Over-trimmed
situation is shown on the right-hand side of the plot in Fig. 19. On the contrary, by gradually dropping the
main traveller and easing the sheet, both 𝐹𝑥 and 𝐹𝑦 reduce because of progressively lower angles of attack,
obtaining the depower curve. During this procedure, the jib was not touched in order to limit the number
of variables that influence the problem.

At full scale, the process was the same except for the chronological sequence of actions. In this case, the
forces were not available in real time, hence it was more difficult to determine when maximum driving
force was achieved. Therefore, once a reasonably good trim was accomplished according the sailing
experience of the crew, the traveller was further pulled in, with the aim of intentionally over-trimming the
main sail. From this configuration, the traveller was then progressively eased by predetermined steps
indicated on the rail in order to obtain the depower curve.

Both in wind tunnel and at full scale, during each acquisition once the target configuration was achieved,
the instruments for measuring forces, pressures and flying shapes were contemporarily activated by an
operator who needed a single button to trigger the simultaneous recordings.

In the following chapter, the three measuring systems developed and adopted for the acquisition of the
variables of interests will be presented both for wind tunnel and SYL, underlying similarities and
differences.

20
4 Measuring systems
4.1 Force Measurements
4.1.1 Wind Tunnel
The measurement of forces is a well-established practice within the Wind Tunnel experimental activities of
any kind. In sail research field, the aim is to obtain global forces and moments developed by the sail plan in
certain conditions of wind angle and sail trim. In case of steady regimes, wind speed is a fixed parameter,
chosen in relation to the structural capabilities of the tested components.

The usual setup sees the model of the boat mounted on a


dynamometric balance fixed to a turning table that allows to
easily and quickly change AWA without changing the
instrumentation positioning (Fig. 20). The balance used has seven
strain-gauged channels capable of measuring, thanks to a
suitable calibration matrix, the three forces (vertical, longitudinal
and lateral) and moments around the three principal axes ([1])
(Fig. 22). An important feature of the acquisition system,
especially for sailing applications, is the possibility of visualizing in
real time the measured forces. This capability, combined with the
opportunity of remotely controlling the trimming devices, allows
to adjust sails in order to seek for maximum driving force or any
other desired configuration (Fig. 21). The possible rig and sail Fig. 22 – Dynamometric balance on which the
regulations on the model used in the wind tunnel are entirely boat model is placed
similar to what sailors can do on regular boats: rig adjustments such as fore and backstay tensioning must
be done manually before the test start, while sails are remotely trimmed during the wind action, thanks to
electrical miniaturized winches that control sheet tension and traveller position of both main and jib. In
order to simulate even more accurately the real sailing regulation mechanism, dedicated cameras are
placed in different locations around the boat so that sails are visible from “realistic” points of view.

Fig. 20 – Boat model fixed on the balance Fig. 21 – The trimmer remotely regulates the sails with direct visibility on the
beneath the hull model, supported by cameras and real time force output

21
4.1.2 Full Scale
Following the successful examples of the mentioned dynamometric boats in the U.S, Japan and Germany,
the Italian research team developed a similar concept for measuring global aerodynamic loads in real
sailing conditions.

The idea at the basis of the system is to connect the whole rigging of the sail boat to a rigid aluminium
frame entirely integrated in the hull and attached to it by means of six mono-axial load cells only (Fig. 24
and Fig. 23). Each rigging element that transmits aerodynamic forces (mast, sheets, fore and backstays,
winches, travellers, etc), is bound to such framework and any contact with the deck or internal part of the
hull is avoided ([7]). For this reason, the deck has been opened in certain points and dedicated connecting
plates were designed and manufactured in order to connect the rigs to the internal frame (Fig. 25).

Fig. 24 – Scheme of the internal aluminium frame with the Fig. 23 – Aluminium frame inside the hull of the SYL
position of the cell loads

The number and positioning of the cells have been chosen in order to obtain an isostatic structure (to avoid
any pre-load due to hyper-staticity) and to capture all the relevant loads coming from the rigs with the least
contaminations from undesired internal actions due to misalignments between forces and load cells. For
the same reasons, spherical joints have been judged optimal for not introducing any kind of non-axial
constraints within the cells (Fig. 26).

Similarly to the dynamometric balance of the wind tunnel, the electrical output of the load cells is
converted to force measures thanks to a calibration matrix that was calculated at the beginning of the
research activity but needs to be checked every time any change, even if minor, is made to the structural
arrangement.

In order to isolate the values of aerodynamic forces, the output from the load cells needs to be purified
from the contributions related to the boat self-weight and to the inertial loads associated with specific
trims and load dynamics at every time instant. This is done by measuring boat’s trim and accelerations
during navigation thanks to an inertial platform and an inclinometer.

All the information acquired during test sessions are collected and managed by dedicated built-in software,
whose functioning will be thoroughly explained in the next chapters.

22
Fig. 25 – Starboard headsail traveller. Particular of one Fig. 26 – Load Cell FZ1. The picture shows the detailed
connecting point between rigging and internal frame. connection between the hull and the aluminium frame.

4.2 Pressure Measurements


We have discussed the importance for sail makers to obtain reliable information about the pressure field
around the sail plan. Previous experiences from other research teams however, highlighted a number of
difficulties in measuring such quantity, starting from the geometry and positioning of the acquisition
system components. In fact, being the pressure taps applied on sails, on the one hand they affect the flow
field with their presence, thus introducing errors in what they measure and, on the other, they might
influence the geometry of the soft sails on which they are applied.

In order to overcome these difficulties, Politecnico di Milano, the pressure technology manufacturer CSEM
and the sail maker North Sails, joining their efforts, came up with a novel measurement system based on
MEMS sensors and dedicated tapping layouts for both model and full-scale experiments ([9]).

4.2.1 MEMS Sensors


The MEMS (Micro Electro-Mechanical Systems) sensors, integrated in the CSEM Pressure Scanners C16 (Fig.
27 and Tab. 2), represent the core devices of the pressure measurement system. A single scanner has 16
channels, each of which equipped with a dedicated sensor capable of measuring differential pressures in
the range of ±1000 Pa thanks to a new generation of piezo-resistive differential low-pressure dies. Besides
their measuring accuracy, even in ranges much lower than their full-scale (resolution of 0.01% FS), these
types of scanners were chosen also for their slim, lightweight and waterproof packages. In particular, the
height of only 6 mm and the limited weight of 50 g allow to place the scanners directly in custom built
sleeves close to the actual measurement sections without affecting the flow field in which they operate

23
The differential measurements of sensors might be
performed between the pressure of interest at a
certain location and a reference pressure taken in the
undisturbed flow or directly measuring the differential
pressures between windward and leeward sides of the
sail. In order to produce these different outputs, the
same scanners can be equipped with two distinctive
pressure flange systems: one refers the 16 pressures
measured from the as many channels to a single
reference pressure while the other provides 32 tubes,
two per pressure sensor, one facing to the windward
and the other to the leeward side of the sail (Fig. 28).
The configuration with a common reference pressure
is well-suited for wind tunnel applications since is Fig. 27 – Pressure scanner C16 with auxiliary parts
fairly simple to measure the upstream pressure with a Tab. 2 – Specification of Pressure scanner
simple Pitot-tube, whereas the differential
measurement is extremely practical in full-scale
experiments where, on the contrary, is rather difficult
to locate measuring devices on board in areas without
flow disturbance from the sail plan.

On the opposite side of the scanner box, it is located


the plug-in for the CAN cable that supplies the scanner
with electrical power, allows remote access to the
scanner configuration commands and transmits the
measurement data to the receiving computer (Fig. 27).
This CAN interface (CAN 2.0A), suitable for a network
of scanners much larger than the ones actually used
(no more than 5 scanners used in our tests), has been
preferred over a wireless solution due to its robust
data transmission capability, the guaranteed data rate
of 1Mbit/s and the possibility of direct power supply,
avoiding embedded batteries that would have
drastically increased the overall scanner weight ([9]).

4.2.2 Tubing and Tapping layout


The described scanners are the devices that actually
perform the pressure measurements, but they are
only part of the acquisition system; in order to reach Fig. 28 – Available pressure flanges for scanners. The one to
the left is used for differential measurements; the one to the
the location at which the pressure is detected, a right refers the local measurements to a single reference value
sophisticated system of tubing is indispensable. As
already highlighted, the two driving concepts to bear in mind when designing such layouts shall be the
minimum disturbance of the flow in which they are introduced and minimum alteration of sail geometry on
which they are applied. With this regard, the Politecnico, CSEM and North Sails adopted a strip technology-
based system for wind tunnel tests and developed a pad technology for full-scale applications. As it will be
described more in details in the following sections, the two different concepts were necessary because of
their characteristics not compatible with both environments.

24
4.2.3 Pressure Strips Technology
The strips used in wind tunnel experiments consists of soft PVC thin and flexible foils with embedded micro-
tubes that end at the exact measurement locations (Fig. 31). The very limited thickness (less than 1.5mm)
and the material of which they are made of, guarantee the fulfilment of the requirements of least flow
disturbance.

Fig. 31 - Strips with embedded micro-tubes. The test sections are Fig. 30 – Connection between tubes and scanners. Each
the horizontal parts on which the tubes end with small holes for sail has 40 taps per side (16 + 16 + 8). Therefore, five
detecting the flow scanners per sail are needed

Fig. 29 - Strips applied on both sails of the wind tunnel yacht model

Each strip is applied on one single side of the sail and the 40 channels are divided among three horizontal
test sections at 25%, 50% and 75% of the sail height with 16 taps at the two lower sections and 8 taps on
the higher one (Fig. 29). The taps, which are effectively the extremities of the micro-channels, are displaced
25
at variable chord lengths in order to best capture all the flow phenomena of interest. For this reason, the
tap concentration is rather fine at the leading edge, to detect the pressure peak and separation bubble, and
thins out moving towards the trailing edge where pressure gradients are less severe. A modest refinement
is present close to the TE in order to observe the effects of the Kutta condition.

The lower part of the strip is equipped with a connector that allows the link with the scanners’ flanges by
means of tiny silicon tubes, commonly used in wind tunnel experiments (Fig. 30).

All in all, the pressure strip technology is greatly suited for model scale applications because the small size
of the sails reflects in the likewise tiny dimensions of the strips. This aspect limits the addition of stiffness to
the sail cloth, preserving almost entirely the sail natural shape. This would not be the case at full-scale
where a lot more material should be used to produce a robust system. Moreover, the strip technology is
definitely little intrusive since it does not damage the fabric of the sail and could thus be applied and
removed as many times as needed. As it will be illustrated in the next paragraph, the same cannot be said
for the pad technology.

4.2.4 Pad Technology


Having clear in mind the critical aspects of the strips in case of large sails, a different technology has been
developed for full-scale experiments, with the same objective of translating the pressure from the location
of interest to the scanner that effectively realizes the measurement.

In addition to the general requirements of least airflow disturbance, the pad technology was developed for
limiting the increase of stiffness to sails and to guarantee an adequate layout for differential pressure
measurements. These necessities lead to the use of standard silicon tubing, not directly attached to the sail
to carry the pressure, and of spot components as terminal points of the tubes, namely “pressure pads” (Fig.
32), located on the measurement sections.

Fig. 32 – Pressure pad. To the left, a schematic representation with measures in mm. To the
right, a picture of the actual pad.

26
These pads, which can be individually placed on the sail, contain two main taps on one extremity and a
pressure tube adapter with two metal tubes of 1 mm diameter on the other. In order to measure
differential pressures, one tap faces to the windward and one to the leeward side of the sail. Since the pad
is attached to one side of the sheet only, it is indispensable to make a hole in the sail directly beneath the
respective tap in order to get access to the airflow on the other side. Although very small (0.8 mm
diameter), the holes are permanent on sails; therefore, once instrumented, they remain so for the rest of
their operative life.

The pads are made of a base layer in Mylar or PET with a thickness of 0.5 mm and of a cover layer in
transparent adhesive tape of 0.3 mm, resulting in a maximum height above the sail of 0.8 mm. The
progressive overlapping of these two layers guarantees a smooth transition from the sail surface to the pad
which results in almost no alteration of the airflow (Fig. 32).

4.2.5 Full-scale layout


The positioning of the scanners, pads and tubing on full-scale main and jib has been a laborious process
both in its design and application phases. At a conceptual level, it has been decided to measure pressure on
4 sections at different heights in order to thoroughly screen the airflow over the entire surface. With this
respect, identifying the right vertical positions has been quite a challenge due to the several geometrical
constraints and scientific requirements: on the one hand, each entire section should be located in an area
of the sail surface free from sticks and reinforces; on the other, it was decided to have aligned sections on
jib and main in order to ideally map the pressure on a single streamline along the entire sail plan. Fig. 33
shows the chosen layout, result of these considerations.

The employment of 3 scanners per sail has been judged to be a reasonable compromise between
complexity, costs and measurement completeness. Therefore, given the 16 channels per scanner, the two
lower test sections on each sail have been sampled
with as many pads connected to one scanner,
while the two upper ones are equipped with 8
pads each and share the same scanner. This
feature introduces an additional constraint to the
vertical positioning of the sections: the lowest
distance shall be guaranteed between pressure
taps and scanner for optimal performance also in
dynamic investigations; therefore, the two higher
sections are located fairly close to each other.

The pads are oriented vertically on the sail with


pressure taps pointing downwards and tubing
upwards in order to introduce less disturbance to
the measured flow. The rationale behind this
decision is that streamlines are predominantly
horizontal with a small vertical component
directed upwards due to the boat heel angle;
therefore, the flow disturbance introduced by the
tubes is confined downstream the measuring spots
(Fig. 35). For the same reason, also the customized
sleeves containing the bundle of tubes and the Fig. 33 - Full scale arrangement of test sections on jib and mainsail
shelters for the scanners are positioned higher

27
than the relative test sections.

Some final considerations regard the chord-wise position of the pads, distanced according to an upscaling
of the wind tunnel scheme, and placed alternatively on both sides in order to avoid a preferable sailing side
due to asymmetries in the tubing system (Fig. 34). This choice has been paid in terms of manufacturing
time since all the tubing sleeves have been doubled and holes have been opened in the sail cloths, but
should pay-off in terms of freedom while navigating and result accuracy since the tube bundles are thinner.

To conclude, even though, all the electronic equipment is waterproof, dedicated low-intrusive sleeves have
been arranged to protect both the scanners and the vertical CAN cables which otherwise would risk to be
teared off during trimming operations.

Fig. 34 – Mainsail equipped with pads and scanners during navigation

Fig. 35 – Particular of the pads and scanner arrangement. The tubing from the pads runs
through the blue sleeve until the scanner which is place right above the test section and
protected itself by a dedicated sleeve. On each
28sail, the CAN Bus that connects the three
scanners is sheltered in the white groove.
4.3 Sail Shape Detection
Quantitative monitoring of the shape assumed by sails in navigating conditions is of great utility for two
reasons. First of all, it allows to obtain a precise correlation between the sail plan geometry and the
pressure distribution that develops around it; secondly, that shape provides the input for prospective
numerical simulations.

In common aeronautical applications, the problem of determining the wing geometry is not an issue since
this is fixed and known. Dealing with soft sails instead, introduces a higher level of uncertainty because the
shapes change depending on operating conditions. In principle, there exists a “design” shape, used by sail
makers to produce the sail itself, and a “flying” shape that is the shape assumed while navigating: these
two are usually significantly different. The dissimilarity arises from many factors related to sailing
conditions, such as wind strength and direction that determine specific pressure distributions, but also to
structural properties, consequences of sail trims and forces acting on the rig. Therefore, in order to conduct
a thorough aerodynamic study, it is fundamental to be able to determine the flying shape contemporarily
to the force and pressure measurements.

As already pointed out, in the last years, several strategies have been proposed to face this challenge, some
based on reproducing rigidly the sail geometries and performing tests on them, others adopting shape
detection systems for soft sails. At Politecnico, for these kind of tests, the latter type of solution is preferred
because allows to perform experiments in more realistic conditions, both in terms of geometries and of
possible trim regulations, and because is a method suitable for both wind tunnel and full scale cases.

In the past, the research group has developed a procedure to reconstruct the sail shape with multiple infra-
red cameras able to determine the three spatial coordinates of dedicated markers applied on the sail
surface (Photogrammetry) ([1]). Even though the results were somehow satisfactory, the method is suited
for wind tunnel experiments only, and requires a quite tedious calibration process for the cameras.
Therefore, in the last experimental campaigns and, in particular, in the one object of the present work, a
novel sail flying shape detection system, based on Time of Flight Technology, has been developed ([8],
[43]).

4.3.1 Time of Flight Technology


The Time of Flight Technology (TOF) is an optical active non-contact technique for determining the
geometry of 3D objects. The underlying principle is the emission and reception of a laser pulse by an optical
device, the Laser Scanner. The pulse is sent against a target surface which reflects it and sent back to the
scanner. By measuring the time the signal takes to return, it is possible to establish how distant is the point
of reflection and thus, the position in space of the target surface. The explained process, valid for a single
point, is repeated for a large number of points, through which is possible to reconstruct the entire sail
surface. To this aim, the laser scanner has a built-in rotating prism that deflects the laser light on the plane
perpendicular to the prism rotational axis (tilt angle). In order to extend the scan to the 3D space, it is
necessary to mount the laser on an external driving motor that provides a rotation in the vertical plane
perpendicular to the previous one (span angle). By proper coupling the prism rotation and the action of the
external motor, the entire space around the scanner can be scanned.

There exist several devices based on TOF that differ for operating ranges and accuracy. In the present
study, two custom-made laser scanners developed by Ing. Vandone, PhD have been used to detect mainsail
and jib’s shapes separately (Fig. 37 and Fig. 36). They are made of similar 2D Sick scanners (LMS 111 and
LMS 511) that scan horizontal planes, mounted on external motors that provide the vertical span. The
29
operating range reaches a distance of 80 m and spreads horizontally up to 270° with angular step typically
set at 0.5°. The rotational speed of the external motor is imposed in order to obtain a sufficiently resolved

Fig. 37 – Laser scanners developed and assembled by Ing. Vandone, PhD. Left-hand side: Scanner Sick LMS511. Right-hand side:
Scanner Sick LMS111

Fig. 36 – The two scanners are used both in wind tunnel and at full scale. Each sail is scanned by a dedicated scanner

surface within a reasonable scanning time frame.

The output of the acquisition process is a point cloud that traces the space around the scanner. For each
laser pulse, the position of the related point is univocally determined by its polar coordinates, i.e. tilt angle
α and distance r, and the information regarding the span angle θ, provided by an encoder fitted on the
external motor. In the post-process phase, the sail is reconstruct in the virtual environment Rhinoceros®
starting from such point cloud.

In summary, the adopted laser scanners and the underlying Time of Flight Technology are capable of
providing extremely fine point clouds and proved to be well suited for sailing experimental applications
both in wind tunnel and at full-scale. The absence of contact between the hardware and the measured
object during data acquisition is of paramount importance when dealing with soft and light sails. Moreover,
differently from most of other methods adopted in previous experiences, the combination of the laser
scanner and an external motor defines the sail shape in terms of fully 3D point clouds, which results in a
considerably higher level of accuracy. Last but not least, the possibility of operating in a wide measuring
range and the low sensibility to weather and working conditions, make the laser scanner and TOF one of
the strength points of the present project.
30
5 Data Management
All the devices described in the previous chapter, and consequently the flow of data coming from them,
needed to be thoroughly managed before, during and after acquisitions. In particular, each instrument has
been validated and checked before each test session in order to assess its correct functioning; the stream
of simultaneous outcomes from all the devices were handled by a custom-made acquisition software; in
the post-process phase, the raw data have been managed and organised in order to obtain clear figures
regarding forces and pressures, and useful features for numerical simulations concerning sail shapes.

In this chapter, we will focus on the procedures for generating comprehensive and useful data, on the
software used for managing them and on the post-process phase, essential for presenting the experimental
outcomes in a clear and general form.

Being wind tunnel and full-scale tests conceptually very similar, as we will see, it has been possible to
develop analogous software and follow comparable procedures for the two applications.

5.1 Pre-process
Most of the described hardware and software adopted in the experimental campaigns were developed
before the author of the present work joint the Lecco Innovation Hub research group. Therefore, the
validation procedure of each single component of the measurement chain will not be reported here. For
reference, the reader should consult the detailed descriptions included in the already mentioned PhD
theses ([7], [8]).

Here, we limit to mention that the three measurement systems (force, pressure, sail shape) have gone
through extensive metrological validation processes in order to assess measurement precision and
accuracy.

In particular, concerning load acquisitions, the wind tunnel dynamometric balance has a long history of test
campaigns and therefore did not need any further check; on the other hand, for the newer SYL, a dedicated
calibration methodology was outlined and applied in order to define a calibration matrix and to quantify
the uncertainties on errors ([7]).

The pressure system with CSEM C16 scanners was tested in wind tunnel facilities to determine both static
accuracy and dynamic response. The former was checked against the consolidated high accuracy pressure
scanner system (PSI) already available at Politecnico and regularly involved in wind tunnel tests, whereas
the dynamic behaviour of pressure strips was evaluated with the aid of a truck hooter, used as pressure
wave generator ([7]).

In order to check the capabilities of TOF laser scanners declared by vendors and to unsure their
applicability for sailing applications, several tests have been carried out varying distance, orientation,
material of the planar target surface and investigating environmental effects like brightness ([8]).

Having assessed the general reliability of the single devices, before each experiment, a final check on the
correct functioning of the entire integrated systems was anyhow performed. In particular, each pressure
strip and tap has been verified singularly by applying a known pressure with a calibrator and checking the
correspondent measurement output was congruent (Fig. 38). This verification was useful not so much to
certify the proper operation of the scanner, but for evaluating the reliability of the overall system, including
tubing, connections and so on. In fact, being the tubing permanently applied on sails, they might be
obstructed by dust or badly bended, compromising their proper functioning.

31
Fig. 38 – Left. Pressure Calibration on model scale sails. The same procedure has been applied to full scale sails. Right. Verification of
the load cells output on the SYL. Known loads were applied to the frame and the corresponded output checked

It should be also mentioned that a calibration check of the dynamometric balance on the SYL has been
performed after the substitution of two beams in correspondence of two load cells. The same calibration
procedure originally developed has been adopted: onshore, known loads (both in value and direction, have
been applied at different locations of the internal frame and the output of the dynamometer recorded and
verified (Fig. 38).

Also the capabilities of the boat anemometer have been re-checked against the measurement of a certified
sonic anemometer in order to be sure about the input wind speed needed for calculating coefficients in
post-processing.

In spite of all the described cautions, being the SYL effectively a boat, it is characterised by the typical and
inherent unforeseen that all sailors well know and are used to deal with. Therefore, continuous
maintenance is a must and even so, something not working properly will still be there.

5.2 Acquisition
In order to manage the stream of data during tests, both in wind tunnel and at full-scale, a dedicated
software for the contemporary acquisition of global aerodynamic forces, pressure distributions and sail
shape has been developed from scratches within the Lecco Innovation Hub project ([7], [8]). The purpose
was to provide the Sailing Yacht Lab with a comprehensive software to easily manage the flow of data
during acquisitions while navigating and to integrate the well-established force measurement setup of the
wind tunnel. In the following, a complete description of the system as it was used in full-scale investigations
is proposed. At model-scale, the same programme was adopted to handle pressure and sail shape
measurements only, whereas forces were acquired from the dynamometric balance through the standard
wind tunnel instrumentation.

The scheme of the overall software architecture (Fig. 39) shows how a Master PC manages on one side the
streaming of force, pressure and boat attitude data and, on the other, controls a Slave PC dedicated to sail
shape acquisition. More in details, the user operates, through a graphical interface, on a program written in
C# that controls all the main activities during tests such as starting and stopping the acquisition, supervising
the data flow and naming files.
32
Fig. 39 – Full scale acquisition software architecture ([7])

Focusing on the Forces/Pressure/Trim side of the diagram, the actual management of data coming from all
the devices is done within RTMaps, a virtual environment particularly suited for the contemporary
streaming of several tools thanks to a multi-thread programming approach ([7]). Each connected
instrument on board the SYL has a specific function providing various pieces of information:

 the six load cells are responsible of measuring the overall forces acting instantaneously on the
internal aluminium frame;
 Three scanners per sail acquire the pressure distributions on four horizontal sections
 An inclinometer and an inertial platform trace the roll and pitch rotations and accelerations of the
boat thanks to integrated gyroscopes and accelerometers;
 Common navigation instruments, including GPS and anemometer, provide information about the
environmental conditions during acquisition.

In order to communicate with the Master PC via USB connection, each signal coming from the
instrumentation is digitalized with dedicated converters: a NIDAQ 9178/9205 ADC converter processes the
signals coming from load cells and the inclinometer; the communication standard NMEA from marine
instruments is converted in USB through an embedded device and the pressure outcome is properly read
with a CAN Bus converter provided by CSEM. Once all the outputs are transformed, RTMamps synchronizes
all the different data using a common timestamp given by the Master PC clock. This re-sampling procedure
has the drawback of lowering the cut-off frequency which however, does not represent an issue for the
application considered since phenomena like wind turbulence has rather low frequencies. In particular,
with reference to the pressure scanner acquisition, a sampling rate of 50 Hz has been considered an
acceptable trade-off between acquisition accuracy and CAN Bus flow data handling capabilities.

33
The sail shape acquisition is managed through a Slave PC
that receives a trigger from the Master PC via Ethernet
cable through a TCP/IP protocol. On the Slave PC, scan
parameters, such as angular ranges and velocities of
external motors and laser scanners, are set according to
test conditions, through a software developed in LabView
environment. As already mentioned there are two
scanners that work contemporarily, each dedicated to one
single sail for better description of their geometries. The
output is provided in terms of point clouds in .txt format,
whose preview is available right after the end of the
acquisition (Fig. 40). This aspect is important because
allows to assess straight away if the process has been
successful. If this is not the case, it is possible to launch
another trigger from the Master PC and re-scan the sails
without influencing the contemporary acquisition of all
the other parameters.

All the saved files, both containing test parameters and


point clouds are automatically named using the time at
which the acquisition starts. This allows in post-process
phase to univocally match the information coming from Fig. 40 – Typical output of the main sail scan at full scale
all the devices.

5.3 Post-process
Except for few scripts used to visualize real time figures during acquisitions, the great amount of data
coming from all the connected instruments was managed after tests: the MatLab® environment was used
to post-process forces and pressures, whereas the software Rhinoceros® was adopted for generating sail
surfaces from point clouds.

The aerodynamic phenomena investigated in the present work can be considered, in first approximation,
time independent due to the limited surface areas interested by flow separation and vortex shedding
phenomena. Therefore, in case of limited external dynamic disturbances during experiments, the working
hypothesis of steady state is fully justified. Under this assumption, tests were carried out by measuring the
quantities of interest for a certain time period and the outcome averaged.

In wind tunnel this way of proceeding was rather straightforward since the working conditions could be
easily controlled and kept constant. Each single test lasted for 60 seconds during which the wind blew and
no changes of regulations were applied; everything was apparently static, with the sails assuming a certain
flying shape for the entire duration of the trial.

On the other hand, at full-scale the environment is obviously not as stable and controllable due to external
factors, such as wind speed and angle changes, and human actions, mainly related to the steering of the
boat. During experiments, it has been tried to keep all these elements under control, but the results were
not always satisfactory. In order to deal with these disturbances, when analysing the gathered data, a pre-
selection of “good” figures has been necessary, and represented the first level of post-processing. More in
details, with the aim of time averaging sufficiently constant values, several “steady” segments, with a
duration of at least 10 seconds, have been extrapolated from the acquisitions. To determine which were

34
the constant portions of the recordings, three criteria related to wind speed and angle ranges and wind
speed variation have been taken into consideration as follow:

 Maximum AWA range: 5°


 Maximum Wind Speed variation: ±5% from the range’s mean value
 Maximum Boat Speed variation: ±5% from the range’s mean value

These constraints, together with the carefulness in performing the acquisitions, guarantee a systematic
method to extrapolate steady segments in which all the measured variables can be reasonably considered
time independent and therefore, averaged in time with a certain confidence. Fig. 41 is an example of how
segments are manually extrapolated from full scale acquisitions: the plots of the three variables subjected
to take-out criteria are plot with the same time scale on the x-axis; once the user has identified on the
graphs the ranges of stability, the script evaluates the intersection of the chosen windows by applying the
mentioned criteria. If these are all contemporary satisfied, it is possible to save the steady segment.

Fig. 41 -Example of segmentation of a single full scale acquisition lasted for 100 seconds. Five different segments
have been identified, each of which contemporary satisfies the three criteria

Once the time windows containing “constant values” are identified, all the parameters are averaged in time
obtaining a single value per segment for each quantity, related to a particular wind condition and sail
regulation. At this point, the objective is to make results comparable to each other, i.e. not contingent
dependent. In this way, it is possible to compare the efficiency of sails of different total area at different
conditions of dynamic pressure ([9]). To do so, pressures and forces are transformed in non-dimensional
coefficients as we have seen in the Chapter 2.

5.3.1 Pressures
Regarding pressure, obtaining coefficients related to every single tap is not difficult. In wind tunnel, the
pressure measurement is performed for each tap, both on leeward and windward sides, referring to the
static pressure in the far field (𝑝0 ); therefore, knowing the wind speed of the undisturbed flow (𝑉∞ ) and the
air density (𝜌), the pressure coefficient is calculated as:

35
𝑝 − 𝑝0
𝐶𝑝 =
1 2
2 𝜌𝑉∞
At full-scale, as previously explained, there is no possibility for detecting pressure in the undisturbed flow,
hence pressure measurements are differential between windward and leeward sides. Therefore, for each
point moving along the section chord, a single differential coefficient is computed:
𝑝𝐿 − 𝑝𝑊
∆𝐶𝑝 =
1 2
2 𝜌𝑉∞
It is clear how with ∆𝐶𝑝 part of the information is lost because it is not possible to evaluate separately the
contributions of the suction on the leeward side and the over pressure on the windward. In order to
represent wind tunnel results with in the full-scale form, a simple subtraction of coefficient on windward
and leeward sides is required.

5.3.2 Forces
The procedure for obtaining aerodynamic coefficients out of force measurements is very similar. In this
case though, in order to obtain non-dimensional coefficients, at the denominator there is also the surface
area of the sail plan (𝐴) responsible for the force considered:
𝐹𝑥 𝐹𝑦 𝐿 𝐷
𝐶𝑥 = ; 𝐶𝑦 = ; 𝐶𝐿 = ; 𝐶𝐷 =
1 2 1 2 1 2 1 2
2 𝜌𝐴𝑉∞ 2 𝜌𝐴𝑉∞ 2 𝜌𝐴𝑉∞ 2 𝜌𝐴𝑉∞
Recalling what was explained in Chapter 2, the most interesting force coefficients for our application are
the driving and heeling force coefficients (𝐶𝑥 and 𝐶𝑦 ) together with the lift and drag (𝐶𝐿 and 𝐶𝐷 ). The main
challenge here is not to compute the four loads, which
is done by simply projecting the force along the
respective directions, but rather to identify the portion
of force attributed to sails only: to this aim, the overall
output needs to be treated, extrapolating from the
measures only the aerodynamic contribution developed
by the sail plan.

In wind tunnel, since the entire boat, including hull and


part of the instrumentation, is placed on the
dynamometric balance, it is necessary to evaluate the
share of force related to the aerodynamic contribution
of everything but sails. To do so, for each AWA
considered, the force developed by the model without
sails under the effect of wind, from now on referred as
windage, was recorded and subtracted from each
acquisition in post-processing phase (Fig. 42). With this
procedure, it is ensured to obtain “pure” sail
aerodynamic loads from which coefficients are
calculated.

At full-scale, the force treatment is more complicated


Fig. 42 – The forces developed by the boat model without
sails under the effect of wind were measured for each AWA because the output of the load cells includes in principle
tested, in order to evaluate the windage to be subtracted to aerodynamic loads, self-weight and inertial forces. In
obtain the “pure” aerodynamic loads
36
order to analyse the sail plan contribution only, these two latter terms need to be identified and
subtracted.
The component related to the dynamics of the boat can be neglected considering the calm environment
and the procedure followed in the experiments: on the one hand the still water of the lake does not
excessively excite dynamically the boat motion; on the other, great attention has been addressed to start
and end each acquisition in time periods with as stable conditions as possible. Moreover, as mentioned
before, in post-processing phase the outcomes have been further skimmed out and only the steadiest
portions of acquisitions were considered.
In order to subtract the self-weight of the boat, it is necessary to know at each instant the position of the
Centre of Gravity (CoG) and the direction along which gravity acts in the boat local coordinate system.
These pieces of information are obtained thanks to the instant output of the inclinometer and the inertial
platform. The CoG of the frame and of the entire boat was accurately measured in the design phase as
described in ([7]) and it is assumed to remain constant in the boat reference system. At each instant during
acquisition, by knowing roll and pitch, it is possible to subtract from the load cells output the three
components of weight force and the respective moments, following the procedure described in [38].

At full-scale, the windage is represented by the resistance to the air flow of the structures connected to the
internal frame only, i.e. mast, boom and rigging. The most significant contribution related to the hull,
present in wind tunnel, here is not included in the measurements since the dynamometric balance is
embedded in the hull itself and not fixed in an external absolute reference system. Therefore, at full scale,
the windage has a substantially lower effect on the results; hence, it is assumed negligible. It should be also
mentioned that an accurate evaluation of this effect would be more complicated at full scale since an
artificial wind at different angles should be created, for example moving the boat with an engine, and the
yacht should be heeled somehow for a proper simulation of its attitude in sailing conditions.

Once the pure aerodynamic forces are extrapolated by the raw measurements, the non-dimensional
coefficients of interest can be derived and properly compared.

5.3.3 Sail Shape


The sail shape acquisition system returns as output a point cloud of several thousands of points that traces
a large portion of the space around the scanners. In order to obtain the desired flying shape out of it, the
relevant .txt file is imported in the modelling environment Rhinoceros® and treated manually. The first step
when handling the raw and confused picture is to orient the point cloud in a reference system coherent
with the CAD of the hull and riggings that had been previously prepared. The procedure, called registration,
consists of identifying some details among the points and properly overlapping them on the corresponding
element of the CAD files. In wind tunnel, this operation is slightly easier than at full-scale because there are
more landmarks to work on, such as the horizontal floor and the vertical walls. At full scale instead, the
main fixed reference elements are represented by the hull deck and mast. An example of the registration
phase can be seen in Fig. 44.

Once the point cloud is properly oriented in space, the dots belonging to the sails need to be isolated from
all the others. This process, namely segmentation, is not trivial due to the several objects around the
sheets, such as lines and other trimming equipment, that need to be accurately removed (Fig. 435). This
operation is rather tedious but needs to be performed very carefully because directly impacts on the
quality of sail reconstruction. In addition to that, the sail edges shall be identified in order to clearly define
the border of the shape.

37
Fig. 44 – Point clouds obtained in wind tunnel (Left) and at full scale (Right). The different location of the scanners in
the two environments influences the distribution of points obtained. At full scale for example, the higher portion of the
sail is less well defined because the scanner is placed on deck

Fig. 43 – Elaboration sequence of the acquired point cloud in wind tunnel. After the registration of the clouds, the points belonging
to the sails are identified (segmentation). Then, the surfaces are create starting from the rectangular shapes overlapped on the
point clouds and cut in correspondence of the edges

38
When the points that belong to the sail are extracted, an interpolating surface is created in Rhino® with the
“Patch” command that deforms a rectangular sheet element overlapping on the desired point cloud. From
this deformed element, the proper sail shape is outlined with the aid of the previously identified edges (Fig.
43).

It is worthy to mention that the “Patch” command allows to set some parameters for creating the surface.
The most important is the “stiffness control” which defines how much the interpolating surface can be
deformed and forced to pass through the points: the lower is this value, the higher is the number of
crossed points, but the curler is the surface. This numerical roughness, generated by the light misalignment
of the detected points, is caused by the inherent inaccuracy of the laser scanners that have, even if small,
an uncertainty on measurements. In light of these considerations, a proper value of stiffness shall be set as
a trade-off between smooth surface and perfect overlapping on the point cloud.

In order to verify the quality of the reconstructed sail, i.e. that the reproduced surface does not excessively
move away from the point cloud, the mean deviation between the two is evaluated. For the cases analysed
in the present work, relative to wind tunnel acquisitions only, a mean distance lower than 1 mm have been
always guaranteed, ensuring a good trade-off between point cloud agreement and smoothness of the
surface. Moreover, a further control of the validity of the process is performed by comparing the lengths of
the virtual edges and the real ones. If these checks are not satisfactory, further iterations are performed
until the outcome results acceptable.

The virtual surfaces of main and jib represent, together with hull and mast, the input geometries for the
numerical simulations that will be presented in the next chapter.

As a closing remark, it should be underlined that in literature is hard to find such a detailed description of
the sail shape reconstruction process, mostly due to the inherent difficulties in validating the acquisition
system and the unknown uncertainties in reconstructing the sail geometry. The proposed system and post-
process methods represent benchmark results in this respect, allowing on the one hand to define the sail
by means of a great amount of points with a known uncertainty and on the other, to quantify the disclosure
between the acquired points and the reconstructed sail.

39
6 Numerical Methods
The described experimental set-up has been developed with the explicit aim of obtaining detailed
information about sail plan aerodynamics in navigating conditions. Combined, with a thorough detection of
sail flying shapes, the investigation outcomes represent complete and valuable data for validating
numerical codes.

With the objective of proving that the developed measurement system could be effectively exploited to
confirm CFD results, the two most common numerical methods used for upwind steady sailing
investigations, VLM and RANS-based methods, have been tested. For this purpose, the codes developed in
two previous thesis projects at KTH and Politecnico have been improved and adapted to the scopes of this
analysis. The VLM code is entirely written in Matlab, a tool widely spread in the scientific academic
environment, whereas the open source software OpenFOAM has been used for RANS simulations. This
aspect is noteworthy because potentially broadens the present treatise to a large number of interested
users without the need of expensive tools or particularly specific knowledge in the informatics field.

It has been decided to compare results from both numerical models because they contribute at different
levels to the research activity. VLM codes are based on rather rough assumptions but have the advantage
of returning fairly accurate results about global forces in very short times. Therefore, they are very useful
for preliminary studies and in situations where not particularly detailed information about the flow
patterns are required. On the other hand, RANS-based codes give a more accurate picture of the local
behaviour around the sails, because the underlying model sticks more to reality. The higher level of
accuracy though, is extensively “paid” in terms of time and computational cost. These kinds of simulations
are preferable for detailed studies of upwind sailing conditions and definitely necessary for downwind
sailing, where potential flow theory assumptions do not hold at all.

It should be clear that the aim of this project is not to propose a technical and detailed dissertation about
advanced CFD simulations, but rather to highlight capabilities and drawbacks of the two proposed
mathematical models, understanding their validity and limitations through the comparison with
experimental results. For this reason, only a brief description of the underlying principles of the two
methods is presented here; for more detailed explanations, the reader should refer to the correspondent
reported sources.

6.1 VLM Method


The VLM code used was originally written by Helmstad and Larsson in 2013 ([34]) within their master thesis
project at KTH. It is based on Potential Flow Theory which is valid under the assumptions of incompressible,
inviscid and irrotational fluid. In particular, the negligence of viscosity denies the possibility of resolving the
flow in the boundary layer and in separated regions, where viscosity plays a significant role due to the high
velocity gradients. It follows that the method is suitable for simulating closed-hauled navigation only,
where the contribution to the overall forces of viscous effects is limited.

6.1.1 Theory
The Potential Flow theory, is based on the definition of a potential function, Φ, whose first derivatives
represent the velocity components:
𝜕Φ 𝜕Φ 𝜕Φ
∇Φ𝑥,𝑦,𝑧 = , , = 𝑢, 𝑣, 𝑤
𝜕𝑥 𝜕𝑦 𝜕𝑧
40
For incompressible fluids, the continuity equation can be written as the sum of velocity component
derivatives, i.e. as the Laplacian of the potential function:

𝜕𝑢 𝜕𝑣 𝜕𝑤
+ + = ∇2 Φ = 0
𝜕𝑥 𝜕𝑦 𝜕𝑧

This expression, known as Laplace equation, constitutes the governing equation on which Potential Flow
theory is built. The solution of this expression coupled with proper boundary conditions yields to the
definition of the velocity components of the modelled flow field. The boundary conditions are related to
the geometry of the problem analysed and shall be expressed as a combination of specific fluid elements or
singularities defined in potential flow theory, such as free-stream, sources, sinks and vortices. The linear
nature of the Laplace equation allows to find a solution to several singularities and add them together in
order to obtain the velocity at each location in the flow even around bodies of complex geometry. In
particular, for modelling sails that produce lift and drag when immersed in a fluid vein, the most common
and convenient approach is to adopt a distribution of vortex elements for their characteristic of deviating
the flow non-symmetrically.

More in details, a vortex element influences the free-stream flow by generating a field of tangential velocity
(Fig. 45) that linearly decreases in intensity as the distance 𝑟 from the origin increases and that depends on
the vortex strength, Γ, as described in [39]:
𝚪
𝑞𝜃 (𝑟) =
2𝜋𝑟
Since the circulation Γ has a prescribed direction, the stream-flow is altered in a non-symmetrical way,
leading to the production of a force, 𝐹, expressed by the Kutta-Joukowski theorem as:

𝑭 = 𝜌𝒒× 𝚪
where 𝐹 is a force per unit length, 𝜌 is the fluid density and 𝒒 is the velocity vector at the location of
interest, i.e. the sum of free stream velocity 𝑄∞ and the velocity induced by the vortex 𝑞𝜃 .

The mathematical model built on the mentioned


equations is very well suited for simulating the
behaviour of thin lifting surfaces like sails immersed in
fluid veins with small angles of attack, where viscosity
plays only a limited role.

In order to get to a 3D problem from the two-


dimensional point vortices elements described, it is
necessary to define “vortex lines” features on which
vortex elements are distributed such that the induced
tangential velocity is orthogonal to the line. These Fig. 45 – Point Vortex positioned in a free-stream
vortex lines are then composed together to obtain
either ring or horseshoe elements in order to model the geometry of the considered surface (Fig. 46). Both
these two types of arrangements comply with the 3 theorems, stated by Hermann von Helmholtz, that
govern the vortex elements behaviour in order to properly model the lifting force production mechanism in
real world ([39]):

41
1. The strength of a vortex element is constant along its length;
2. A vortex element cannot start or end in a fluid (it must be closed path or extend to infinity);
3. Vortex elements will remain vortex elements with time.

In the Vortex Lattice Method implemented by Helmstad and Larsson, sail surfaces are discretized into
quadrilateral panels on which vortex ring elements are placed with their leading bound vortex at a distance
of 1⁄4 of the panel chord form the panel leading edge.

Fig. 46 – Two possible arrangements to discretize panels in space. Each side is a vortex line on which vortex elements are distributed

In order to calculate the vortex strength of each element, a linear system of equations is solved imposing
the non-penetrating condition at the surface. For ring elements, this is equivalent to impose that the
normal component of the velocity at the control point of each panel considered, located at 3⁄4 of the
chord from the panel leading edge and in the middle of the span, equals zero.

𝑞𝑐,𝑖 ∙ 𝑛𝑖 = 0

Knowing the strength of all vortices, the velocity field around the sail plan is obtained and aerodynamic
forces are calculated according to Kutta-Joukowski theorem.

A further condition must be imposed at the sail trailing edge, ensuring that the flow leaves the surface
parallel to it. This behaviour, known as Kutta condition, must be imposed “manually” because is due to
effects of viscosity that are not modelled in potential flow theory. This is done by enforcing zero circulation
between the last panel on the sail surface and the first panel of the wake, which is thus discretised as well,
by setting the vortex strength of this latter element equal and opposite to the former.

6.1.2 Implementation
The inputs required for the VLM computations are the problem geometry and the main environmental
variables. Considering that the purpose of VLM simulations is a preliminary overall indication of the forces
developed by the sail plan, the problem geometry is limited to the two sail shapes implemented by means
of organized point clouds obtained from the output of the laser scanners. These are treated by the code by
generating quadrilateral panels on which vortex ring elements are virtually placed. The number of panels is
decided by the user by setting the discretization points, both longitudinally and vertically, forming an
equally spaced grid as shown in Fig. 47. In the present study, each sail has been discretized with a 20x10
panel grid in order to sufficiently well describe the sail curvatures and, at the same time, keeping the
computational time very limited. Regarding the environmental conditions, the only variables needed are
the apparent wind speed (AWS), apparent wind angle (AWA) and air density.

42
An important choice related to the VLM implementation is related to the wake modelling. As previously
mentioned, the Kutta condition must be guaranteed at the trailing edge of the sail surface and, to do that,
at least an additional row of panels must be added to cancel out the circulation of the relevant trailing
bound vortices. Conceptually, the wake needs to be discretized into panels in order to guarantee the Kutta
condition, but at the same time it should be ensured that the wake does not produce any force because it
does not happen so in reality. Therefore, each panel placed on the wake should be aligned with the flow
velocity so that no lifting force is generated. The simplest way to model this behaviour is to assume that the
flow behind the sails is mainly aligned with the free-stream. Hence, the wake can be modelled by a single
row of panels that extends for several meters downstream, oriented as the free-stream velocity which is
known and constant. A more precise model would take into consideration that the velocity field in the

Fig. 47 – Example of discretised sail plan. The continuous acquired sail surfaces (Left) are discretised with an equally
distributed grid of panels (Right). The arrangement of the chosen grid allows to accurately represent the sail geometry
without weighting down the simulations

wake is influenced by the sails, in particular at the tips where large vortices are shed, deviating locally the
flow from the free-stream direction. A completely free-force wake model ensures that each discretization
panel is oriented with the local velocity, not generating any lift. This latter model is conceptually more
correct than the former one but requires more computational time since the local velocities in the wake are
calculated iteratively until the right alignment between panels and local flow is reached.

From Helmstad and Larsson’s work, the results obtained with the two models present only minor
differences in terms of aerodynamic coefficients. Therefore, considering the scope of preliminary indication
attributed to the VLM in the present project, the fixed-wake simple model was judged to give sufficiently
good results and to be extremely convenient for the very short computational time required. Thus, it was
preferred to the more elaborate free-wake model.
43
6.1.3 Post-process
By solving the linear system, whose rows express the non-penetrating condition at the control point of each
panel, the circulation strengths of the bound vortex of all panels are calculated. Each vortex influences the
velocity field around it, inducing a certain tangential velocity that depends both from the vortex strength
and from the distance of the point considered. With these information, for each panel’s control point on
which the non-penetrating condition is imposed, the velocity is obtain as the sum of the free-stream
velocity and the velocity induced by all the other vortex rings on the considered panel. Finally, knowing all
the circulation strengths, Γ𝑖 , and velocities, q𝑖 , the force acting on each panel is computed with the formula
expressed in the Kutta-Joukowski theorem.

The total aerodynamic forces acting on the sails are obtained as the sum of all forces, whereas the pressure
acting on each panel results from the ratio between the normal component of force acting on the panel
and its area. As previously mentioned, in order to compare the results from different environmental
conditions and sail geometries, aerodynamic coefficients are computed dividing forces and pressures by the
dynamic pressure.

6.2 RANS Equations


The most innovative element of the experimental set-up of the present project is arguably the
contemporary measurement of forces, pressures and sail shapes. These pieces of information, available all
together, allow to accurately point out the correlation between sail regulations and flow patterns,
representing meaningful validation parameters for detailed numerical simulations.

Nowadays, a good trade-off between accuracy of results and computational cost can be achieved with
simulations based on RANS equations. Therefore, these have been chosen for predicting the flow field
around the boat and the sail plan in the present investigation. In particular, the numerical set up has been
developed in the open source environment OpenFOAM, starting from the bases laid by a previous thesis
work at Politecnico ([35]).

6.2.1 Theory
The physical description of the fluid motion is based on the combination of the equations of motion and
continuity equation that together take the name of Navier-Stokes equations. In steady state conditions and
under the assumption of incompressible fluid, valid for air in the present study due to the low velocities
considered, they are written as:
∇𝑝
(𝒖 ∙ ∇)𝒖 − 𝜈∇2 𝒖 + +𝒇 = 0
{ 𝜌
∇∙𝐮=0
Where 𝜌 is the fluid density, 𝜈 the cinematic viscosity, 𝐮 is the velocity vector, 𝑝 is the pressure and 𝒇
gathers all the possible external forces acting on the fluid volume considered.

The steady state Navier-Stokes equations, together with proper boundary conditions, describe the motion
of a fluid in a certain volume, but only in rare cases they can be applied in their pure form to each single
fluid particle. In almost every aeronautical application, given the relatively high Reynolds numbers, the flow
is mainly turbulent, and therefore, resolving the movement of each particle would demand an unaffordable
computation power. Nowadays, the most common approach to overcome the problem is to consider all the
variables that appear in the equations composed by the sum of a mean and a fluctuating value. This

44
operation, called Reynolds average, whose detailed treatise can be found in [40], greatly reduces the
computational demand because many averaged terms are cancelled out.

Following this approach, the Navier-Stokes equations become effectively Raynolds Averaged Navier-Stokes
equations, which in scalar form are written as:
1 𝜕𝑃
̅̅̅̅̅̅
𝑑𝑖𝑣(𝑈𝑼) + 𝑑𝑖𝑣(𝑢 ′ 𝒖′ ) = − + 𝜈𝑑𝑖𝑣(𝑔𝑟𝑎𝑑(𝑈)) − 𝑓𝑥
𝜌 𝜕𝑥
1 𝜕𝑃
̅̅̅̅̅̅
𝑑𝑖𝑣(𝑉𝑼) + 𝑑𝑖𝑣(𝑣 ′ 𝒖′ ) = − + 𝜈𝑑𝑖𝑣(𝑔𝑟𝑎𝑑(𝑉)) − 𝑓𝑦
𝜌 𝜕𝑦
1 𝜕𝑃
𝑑𝑖𝑣(𝑊𝑼) + 𝑑𝑖𝑣(𝑤̅̅̅̅̅̅
′ 𝒖′ ) = − + 𝜈𝑑𝑖𝑣(𝑔𝑟𝑎𝑑(𝑊)) − 𝑓𝑧
𝜌 𝜕𝑧
{ 𝑑𝑖𝑣(𝑼) = 0
The terms are the same as in the pure form of the N-S equations except for the second terms of the left-
hand sides. These elements, that represent the product of velocity fluctuations, are associated with the
transfer of quantity of motion due to turbulent vortices and, multiplied by the density, assume the
dimensions of stresses. In particular, they represent the tangential components of the 6-component stress
tensor, named Reynolds stress tensor:

𝜏𝑥𝑥 𝜏𝑥𝑦 𝜏𝑥𝑧 −𝜌𝑢̅̅̅̅


′2 ̅̅̅̅̅̅
−𝜌𝑢 ′𝑣 ′ ̅̅̅̅̅̅
−𝜌𝑢 ′𝑤 ′

𝜏
𝝉 = [ 𝑦𝑥 𝜏𝑦𝑦 𝜏𝑦𝑧 ] = [ −𝜌𝑣̅̅̅̅̅
′𝑢 −𝜌𝑣̅̅̅̅
′2 ̅̅̅̅̅̅
−𝜌𝑣 ′𝑤′] ; 𝜏
𝑥𝑦 = 𝜏𝑦𝑥 ; 𝜏𝑥𝑧 = 𝜏𝑧𝑥 ; 𝜏𝑦𝑧 = 𝜏𝑧𝑦
𝜏𝑧𝑥 𝜏𝑧𝑦 𝜏𝑧𝑧 ̅̅̅̅̅̅ ̅̅̅̅̅̅
′ 𝑢′ −𝜌𝑤 ′𝑣′ ̅̅̅̅̅
′2
−𝜌𝑤 −𝜌𝑤

These constitute six further unknowns of the problem and therefore need as much equations in order to
solve the system. For this purpose, RANS equations need an additional model for the treatise of turbulent
stresses.

A detailed review of all the possible turbulence models exceeds the scope of the present work, where only
a brief introduction about the underlying hypotheses and principles will be proposed. In general,
turbulence models are based on the Boussinesq hypothesis which considers the anisotropic component of
Reynolds stress tensor proportional to the mean rate of strain tensor, by means of a proportionality
constant 𝜐𝑡 > 0, called turbulence eddy viscosity. Even though, this assumption is not fully acceptable from
a physical point of view due to a non-exact description of the energy dissipation process related to
turbulence, the Boussinesq hypothesis is widely employed for its simplicity and effectiveness. Essentially,
the consequences of turbulence are taken into account within the eddy viscosity, 𝜐𝑡 , which is added to the
kinematic viscosity, 𝜐, to give an overall effective viscosity, 𝜐𝑒𝑓𝑓 :

𝜐𝑒𝑓𝑓 (𝑥⃗) = 𝜐 + 𝜐𝑡 (𝑥⃗)

The newly defined effective viscosity, introduced in the RANS equations, reduces the unknowns to the sole
𝜐𝑡 which, in turn, can be expressed as the product of velocity and length turbulence scales, 𝑢𝑚 and 𝑙𝑚
respectively:

𝜐𝑡 (𝑥⃗) = 𝑢𝑚 𝑙𝑚
Therefore, the aim of turbulence models is to properly relate these scales, in order to come up with a
suitable turbulence eddy viscosity. This is done by introducing a number of equations that can vary from
zero to two: in 0-equation models, the eddy viscosity is modelled based on empirical assumptions, whereas
1-eq and 2-eq models derive 𝜐𝑡 as function of turbulent quantities that are related to each other by
transport-diffusive equations.

45
Following the examples available in literature, the present study focuses on the latter category and, in
particular, on the three most adopted models for external flow applications:

- Spalart-Allmaras (1992): 1 equation


- 𝑘 − 𝜀 (1974): 2 equations
- 𝑘 − 𝜔 𝑆𝑆𝑇 (1994): 2 equations

Since a “best” model for any case does not exist, a preliminary study to investigate which model was more
suited for our type of analysis has been performed on a test case representative of the different conditions
under investigation. As it will be presented more in details further on, the Spalart-Allmaras turbulence
model proved to best predict the major characteristics of the flow around the sail plan, resulting in
pressure distributions and forces in good accordance with experimental observations. Despite the lower
computational cost and limited applicability compared to the other models, this outcome was not
surprising considering that the Spalart-Allmaras model has been specifically developed for external
aerodynamic calculations. In summary, it defines the turbulent eddy viscosity 𝜐𝑡 with one single transport
equation and specifies the length scale by means of an algebraic formula, that is certainly not suited in
situations with complex geometries where the length scale is difficult to define, but works well on smooth
and thin surfaces like sails.

6.2.2 Implementation
In general, Computational Fluid Dynamic simulations consist in solving the equations belonging to a chosen
mathematical model in a region of interest. This operation however, cannot be performed for each single
point in the space; instead, a discretisation grid is needed to define the position and number of spatial
nodes on which the equations are solved. The overall number of cells is most of the time a critical
parameter in simulations and shall be carefully decided considering the trade-off between accuracy of
results and computational cost which, for 3D simulations, might be particularly demanding.

Geometry
When setting the computational domain, the first preliminary step to limit the resources requested, is to
“clean” as much as possible the geometry of the problem, including only the features that have significant
effects on the flow, disregarding all those particulars whose contribute is negligible. With this aim in mind,
the wind-tunnel boat model, object of the study, has been
virtualised in terms of .stl files containing the shapes of hull,
mast and sails only (Fig. 48). The mast, placed upstream of the
main sheet, is a key responsible of the peculiar pressure
distribution described in the previous chapter and observed
experimentally; therefore, it needs to be included in the
simulation and accurately placed in the space guaranteeing the
correct relative position with respect to the sails. The hull has
been modelled because it is thought to impact the incoming
flow, generating disturbance especially in the lower parts of the
sails which contribute significantly to the overall aerodynamic
forces given the larger extension of surfaces. On the other hand,
the boom, all the riggings and the instrumentation placed on the
deck during wind tunnel tests, have been excluded from
numerical simulations since they would have greatly increased
the complexity of the geometry and consequently the
computational cost, with only minor effects on the flow. Fig. 48 – Boat model used in RANS simulations of
wind tunnel cases
46
It should be underlined that, in order to properly generate the mesh around the sails, an artificial thickness
of 2mm has been added to the surfaces generated from the point clouds obtained during experiments; in
this way, the sail flying shapes have been transformed in solid features, that are better handled by the
mesh generator.

In the simulations, all these components are considered as “solid walls” and therefore have fixed shapes
through which the fluid cannot pass. In fact, the flow motion is simulated around these objects in a “virtual
box” representing the testing room of the wind tunnel (Fig. 49). In order to get a proper analogy between
experiments and numerical simulations, the dimensions of the domain have been chosen equal to the
effective wind tunnel measures: 36x14x4 meters.

Fig. 49 – The computational domain resembles the wind tunnel testing room

Mesh generation
Starting from the “box”, the mesh is generated in terms of structured hexahedral cells that, from a rather
big edge size (1,1 m), are progressively refined getting closer to the boat model as the flow gradients to be
resolved get steeper, downsizing up to 2 mm. In order to define these refinement regions, smaller
concentric boxes are placed around the features of interest and the cell dimensions can be individually
defined in order to guarantee a smooth transition of cell size between regions (Fig. 50).

Fig. 50 – Mesh refinement in the overall computational domain. Downstream the boat model, the mesh is refined in order to
properly describe the wake

47
Even though cells with edges as small as 2 mm seem to form a rather fine discretisation within an overall
domain of more than 2.000 m3, these dimensions to not allow to properly resolve the boundary layer
regions that develop around the wetted surfaces, in particular sails and mast, due to the imposed no-slip
condition. At the local Reynolds numbers of the cases tested, 𝑅𝑒𝑥 ~150.000, the boundary layer has an
overall thickness of maximum 17 mm; this means that for a proper resolution of all the flow phenomena
inside it, in particular the viscous sublayer, a mesh size in the order of the tenth of millimetre is needed
closed to the surface. Since this option would drastically increase the overall number of cells, making the
simulations not manageable by the computers available, a different solution has been adopted. The flow
has not been “directly” resolved up to the viscous sublayer, but instead, the numerical solution has been
conveyed from the region attached to the surface to the outer flow with the aid of specific wall-functions.
Essentially, these are empirical laws that replicate the behaviour of the variables of interest in the first layer
of cells, simulating the effects of the viscous sublayer on the outer flow. This expedient substantially
reduces the computational cost but, on the other hand, introduces a further level of approximation that
influences the accuracy of results, particularly in those regions interested by flow separation where the
fluid motion is more chaotic.

Fig. 51 – Detailed refinements around the yacht model. In particular, the sail plan has dedicated concentric refinement regions in
order to guarantee a smooth transition in terms of grid sizing, keeping as limited as possible the number of cells

It should be noted that there is not a universal recipe for the number of cells to be included in a numerical
simulation, even though this parameter is one of the keys for satisfactory computations. In CFD, before
discussing any kind of result, it is paramount to assess whether the outcomes properly represent the actual
physics of the problem, without being influenced by the contingent numerical set-up. In order to do that,
convergence studies shall be performed: in the present work, the same pilot case used for choosing
turbulence model has been investigated also with six different meshes with variable grid sizes, ranging from
2 million cells for the coarsest configuration to around 16 million for the finest. As it will be shown in the
following, increasing the number of cells, the numerical results approach the target experimental
outcomes; therefore, with the aim of limiting the computational costs, a compromise configuration that
guarantees reasonably good results and manageable amounts of data, have been chosen and adopted for
all the simulations reported in the project. In particular, roughly 6 million cells with maximum and
minimum sizes of respectively 1,1 m and 0,002 m, have been adopted (Fig. 51).
48
Boundary Conditions
The computational domain described above represents the space through which the fluid, in this case air,
can move. In order to solve the equations that describe the flow motion, the user has to specify some
values from which the calculations start. These boundary conditions need to be imposed for all the
variables that appear in the mathematical model (velocity, pressure and turbulence quantities) on each
single “border” of the domain which, in the present simulations, are the faces of the box, i.e. the walls of
the wind tunnel, and all the contour surfaces of sails, mast and hull (Fig. 52). Moreover, an initial value of
those variables shall be imposed also on the internal field in order to give to the solver a starting point for
the iterations.

Fig. 52 – Names of Boundary Conditions in the Computational Domain

Considering only the Spalart-Allmaras turbulence model, the 4 variables that need to be specified are:

1. 𝑼: velocity vector
2. 𝑝: pressure
3. 𝜈𝑡 : turbulent eddy viscosity
4. 𝜈̃: modified turbulent viscosity

The values defined on the internal field and on single boundaries are reported in Tab. 3.
Tab. 3 - Boundary conditions specified in OpenFOAM on each wall and in the internal field

Inlet Outlet
Variable Type Value Unit Variable Type Value Unit
𝑼 Fixed Value 4.5,0,0 m/s 𝑼 Zero Gradient - m/s
𝑝 Zero Gradient - m2/s2 𝑝 Fixed Value 0 m2/s2
𝜈𝑡 Free Stream 1e-06 m2/s 𝜈𝑡 Free Stream 1e-06 m2/s
𝜈̃ Free Stream 0.006 m2/s 𝜈̃ Free Stream 0.006 m2/s

Front / Back / Upper Lower / Hull


Variable Type Value Unit Variable Type Value Unit
𝑼 Symmetry Plane - - 𝑼 Slip - -
𝑝 Symmetry Plane - - 𝑝 Slip - -
𝜈𝑡 Symmetry Plane - - 𝜈𝑡 Slip - -
𝜈̃ Symmetry Plane - - 𝜈̃ Slip - -

49
Jib / Main / Mast Internal Field – Initial conditions
Variable Type Value Unit Variable Type Value Unit
𝑼 Fixed Value 0,0,0 m/s 𝑼 Fixed Value 4.5,0,0 m/s
𝑝 Zero Gradient - m2/s2 𝑝 Fixed Value 0 m2/s2
𝜈𝑡 Wall Function 0 m2/s 𝜈𝑡 Fixed Value 1e-06 m2/s
𝜈̃ Wall Function 0 m2/s 𝜈̃ Fixed Value 0.006 m2/s

The values of velocity and pressure are set according to the experimental conditions of wind tunnel during
tests, whereas 𝜈𝑡 and 𝜈̃ are estimated following guidelines for Spalart-Allmaras model available in
literature. The types of boundaries determine whether the specified values are fixed numbers or a
gradients and specify how the walls shall be treated. In the present study, we are exclusively interested in
the interaction between air and the sail plan; therefore, the non-slip condition is imposed only on sail and
mast surfaces and, as it was previously mentioned, the flow local phenomena are resolved with the aid of
specific wall functions, named NutUSpalding. All the rest of boundary surfaces are considered either no-slip
walls or symmetrical planes since there, an accurate description of the flow is out of the interest of this
work.

Numerical schemes and solving algorithm


The simplified RANS equations to be solves in our simulations, contain differential operators such as
gradients and Laplacians that need to be solved on discrete grids; therefore, it is necessary to define
numerical schemes to properly approximate derivatives. Without going too deeply into details, in general it
is preferable to use second order schemes because introduce less numerical diffusive errors compared to
first-order methods. However, second order schemes suffer more steep gradients if the mesh is not fine
enough to properly resolve them, leading to instability and consequently divergence of the numerical
solution. In the present simulations, second order linear methods have been adopted for discretizing
gradients, whereas divergence and Laplacian operators have been treated with second order, upwind-
biased schemes in order to guarantee the proper convergence of solutions.

Being the discretisation techniques defined, starting from the specified initial and boundary conditions, the
RANS equations are solved iteratively with the SIMPLE algorithm (Semi-Implicit Method for Pressure Linked
Equations), commonly used for steady-state ([23]), incompressible fluid flows, whose detailed explanation
can be found in OpenFOAM guide ([41]). The iterations stop either when specified values of residuals is
matched or the maximum number of iterations is reached. The choice of residuals on the computed
variables, is again a trade-off between accuracy of results and computational time. In any case, the trend of
these parameters shall be always monitored during simulations and checked at the end, since they give an
important preliminary idea on the validity of results.

6.2.3 Post-process
The outcome of simulations is expressed in terms of velocity and pressure values for each single cell in the
fluid domain: from these quantities all the other variables of interest can be derived. As previously
discussed, our purpose is to obtain overall aerodynamic forces and local pressure values acting on the sail
surfaces in order to compare numerical results and experimental outcomes. The first step to extrapolate
these quantities from the simulation results has been to specify proper OpenFOAM built-in libraries that
saved the values of interest in dedicated directories; afterwards, these data have been handle with proper
MatLab functions in order to obtain and represent the desired figures.

50
Considering the aerodynamic forces, only the loads acting on the sail surfaces have to be extracted,
coherently with the fact that in the post-process phase of wind tunnel results, the windage had been
subtracted from the overall forces measured by the balance. In OpenFOAM, this operation has been rather
straightforward as the desired sail surfaces could be easily selected through their .stl files. It shall be noted
that, since the sails had been actually implemented as solid features, both windward and leeward surfaces
have to be selected and then added together to obtain the total force attributed to the sail both for jib and
main. In Matlab then, the forces have been added together and projected both along the boat principal
axes to obtain thrust and heeling coefficients, and along the parallel and perpendicular flow directions to
get lift and drag coefficients.

Not as smooth has been the treatment of pressure outcomes. Extracting suitable values that could be
properly compared with experiments have been quite a challenge because the process for the definition of
the probes’ locations on virtual sails could not be fully automatized. The position of pressure taps,
preliminary measured on the sails used in wind tunnel, had to be accurately reproduced on the virtualized
sail flying shapes, that for each case considered, were positioned differently in space and exhibited slight
discrepancies in terms of characteristic dimensions. For these reasons, the procedure of defining the
probes’ location has been performed manually in Rhino, re-scaling, if necessary, the relative position of the
virtual probes in order to define them as accurately as possible with respect to the real ones (Fig. 53).

Fig. 53 – Virtual probes located on the sails of the CFD boat model

In Matlab, the extracted values needed to be distinguished between windward and leeward and ordered
along the sail section chords. Only after these operations, it was possible to compare numerical and
experimental pressure coefficients and discuss results.

Besides providing these important pieces of information, numerical simulations play a fundamental role in
aerodynamic studies because allow to visualise the flow velocity and pressure fields at any location in the
computational domain; this gives an overall picture of the situation around the sails, which is hardly
achievable in experiments. All the fluid visualizations proposed in the following chapter, have been
produced with the open-source application Paraview, expressly suited for handling OpenFOAM results.

51
6.2.4 Convergence Study
Before presenting any result, a proper definition of the numerical set-up is necessary to ensure the
reliability of the outcome. In the present study, this preliminary investigation encompasses both the choice
of turbulence model that allows to properly describe the main fluid physical phenomena, and the number
and size of cells used to discretise the computational domain. These two aspects have been tried out in
combination resulting in a matrix of different configurations. Among these, a single “best” arrangement has
been chosen and adopted for performing all the following simulations.

Considering the case of AWA 27.5° in upright condition, the three turbulence model, Spalart-Allmaras, k-ε
and k-ω SST have been tested with different meshes. For each combination, the four aerodynamic
coefficients (Cx, Cy, Cl, Cd) have been extracted and evaluated against the correspondent values of wind
tunnel. Tab. 4 summarizes the relative errors on the coefficients between numerical result and
experimental outcome for each case. The same information is represented also graphically, in order to
better interpreting the figures (Fig. 54).

Tab. 4 – Combination of turbulence models and mesh refinement tried out on the configuration AWA 27,5° - Upright in order to
understand the best numerical set-up for all the RANS simulations. Relative errors on coefficients between numerical and
experimental outcomes are reported; where coefficients are missed, the numerical simulations did not converge

Mesh (n° cells) Spalart-Allmaras K-ω SST K-ε


Cx: 1.6% Cd:-0.6% Cx: 16% Cd: -1% Cx: 11% Cd: 6%
Very Fine (15.9 mln) Cy: 0.9% Cl: 1% Cy:10% Cl: 11% Cy: 9.2% Cl: 9.5%
Cx: 1.9% Cd:-0.07% Cx: 13.8% Cd: -1.9% Cx: 9.7% Cd: 4.9%
Fine (10.8 mln) Cy: 1.3% Cl: 1.4% Cy: 8.5% Cl: 9.3% Cy: 8% Cl: 8.3%
Cx: 2.5% Cd:-0.03% Cx: 15.8% Cd: -4% Cx: 9.5% Cd: 3.7%
Med Fine (6.3 mln) Cy: 1.5% Cl: 1.7% Cy: 9.3% Cl: 10% Cy: 7.6% Cl: 7.9%
Cx: 2.9% Cd:-0.03% Cx: 15.5% Cd: -4.7% Cx: 9.2% Cd: 1.6%
Med Coarse (3.7 mln) Cy: 1.8% Cl: 2% Cy:8.7% Cl: 9.8% Cy: 6.6% Cl: 7%
Cx: 3.6% Cd:-0.5% Cx: - Cd: - Cx: 9.7% Cd: 0.6%
Coarse (2.5 mln) Cy: 2.2 % Cl: 2.4% Cy: - Cl: - Cy: 6.7% Cl: 7.1%
Cx: 4.4% Cd:-0.4% Cx: - Cd: - Cx: 10% Cd: -0.9%
Very Coarse (1.9 mln) Cy: 2.8% Cl: 3% Cy: - Cl: - Cy: 6.3% Cl: 6.9%

52
Cx
0,52
0,5
0,48 Wind Tunnel
Cx [-]
0,46
0,44 SA
0,42
k-epsilon
0,4
0 5 10 15 20 k-omega
N° Cells [mln]

Cy
1,45
1,4
Wind Tunnel
Cy [-]

1,35
SA
1,3
k-epsilon
1,25
0 5 10 15 20 k-omega
N° Cells [mln]

Cd
0,24
0,23
Wind Tunnel
Cd [-]

0,22
SA
0,21
k-epsilon
0,2
0 5 10 15 20 k-omega
N° Cells [mln]

Cl
1,55
1,5
1,45 Wind Tunnel
Cl [-]

1,4 SA
1,35
k-epsilon
1,3
0 5 10 15 20 k-omega
N° Cells [mln]

Fig. 54 – Trends of coefficients varying with number of cells for the tested turbulence models

53
The Spalart-Allmaras model presents the most coherent trends: the coefficients tend towards the values
obtained in experiments as the number of cells increases, i.e. as the computational grid gets finer, as one
could expect. On the other hand, the behaviour of k-ε and k-ω SST models might seem somehow surprising
considering that they should represent more accurately the physical phenomena, being composed by two
transport equations. In fact, it seems that the results do not depend on the grid resolution or even that the
trend is opposite to the one expected. A possible explanation for this behaviour might be related to the use
of wall-functions which need specific sizes of cells close to the walls of interest. In particular, it should be
guaranteed a certain thickness of the first layer (𝑦 + > 30) that, in our case would have prevented to place
a minimum number of cells within the boundary layer, resulting in a grid too coarse for properly refining
other important flow phenomena. It is reasonable to think that this aspect has more influence on
turbulence models finer and more sensitive such as k-ε and k-ω SST. Moreover, the quality of Spalart-
Allmaras results is justified by the fact that the configurations simulated in the present study reflect quite
precisely the applications for which the model was developed: external flows, low angles of attack and
limited areas interested by flow separation.

For these reasons, the Spalart-Allmaras model was adopted and a suitable grid resolution chosen as a
compromise between accuracy of results and manageable computational loads and times. In particular, the
configuration with around 6 million cells guaranteed a good trade-off between these two aspects,
generating, for the test case, relative errors lower than 2.5% compared to experimental coefficients and
reasonable computational times of approximately 5 hours with 32 processors running in parallel. Therefore,
this set-up was selected as grid structure for all the simulations.

Within this convergence study, together with the global aerodynamic coefficients, also a check on pressure
distributions was performed to verify the agreement between numerical and experimental results even at a
local level. The best agreement between computed and measured pressures was found again for the
Spalart-Allmaras model, confirming the reliability of the numerical apparatus involved and the
correspondence between force and pressure calculations.

Once the numerical framework was set, all the simulations have been performermed with the same overall
set-up. From one case to the other, in order to replicate the experimental activity, the only things that
changed were the orientation of the boat model, that was moved and heeled like in the wind tunnel, and
the sail shapes, imported and “hoisted” in the virtual environment for the correspondent case considered.
The mesh refinement regions were adjusted each time around the boat model, mantaining the selected
sizes and ratios between sub-domains.

54
7 Results
Following the procedures described in this report, a great amount of data was obtained during the
experimental campaigns and numerical simulations. In this chapter, only the most relevant outcomes will
be reported, in order to focus on the features of interest for our scope. The overall objective is to
demonstrate the effectiveness of the developed experimental and numerical set-ups in investigating
upwind steady sail aerodynamics.

The controlled environment and the possibility to adjust sail trim through precise regulations allowed to
obtain more systematic and deterministic results from wind tunnel rather than form full-scale experiments.
This aspect, together with project time constraints, suggested to simulate numerically only the most
relevant model-scale cases in order to assess the correspondence between empirical observations and
numerical predictions.

The model-scale results will be presented firstly in terms of global force and pressure coefficients for the
most representative configurations investigated in wind tunnel. Then, the numerical simulation outcomes,
obtained with the detected and virtualised sail shapes, will be analysed against experimental data; as
mentioned before, for RANS and VLM numerical simulations, only the cases of max power for each
Apparent Wind Angle in both upright and heeled configurations have been considered.

After having presented the capabilities and levels of accuracy of the two numerical models, some examples
of the information that can be obtained from virtual simulations will be proposed. These will not be
discussed in depth but represent a solid starting point for further dedicated analyses.

Finally, an extract of the full-scale outcome will be presented in terms of experimental forces and pressure
distributions. These will be analysed and compared with wind tunnel data evaluating similarities and
differences between the two research methods and operating conditions.

55
7.1 Wind Tunnel Measurements
7.1.1 Aerodynamic force coefficients
The overall outcome of the wind tunnel campaign is summarized in Fig. 55 and Fig. 56, where the depower
curves for all the upright and heeled configurations tested are drawn. Each curve refers to the depowering
process for the same Apparent Wind Angle, reporting the thrust and heeling coefficients at the various
trims examined.

Fig. 55 – Depower curves drawn for wind tunnel tests with the boat model in upright conditions. Each curve
connects the points that represent the different sail regulations tested for the indicated AWA

Fig. 56 - Depower curves drawn for wind tunnel tests with the boat model in heeled conditions. Each curve
connects the points that represent the different sail regulations tested for the indicated AWA

Considering a single curve at a time, the highest point corresponds to the trim that produces the maximum
thrust, whereas the one further to the right generates the highest heeling force. This top-right region is the
starting point of the depowering procedure described in previous chapters and is achieved by properly

56
tightening the sails. From this stage, if the trimmer further pulled in the main, the heeling force would have
increased and the thrust force would drop due to flow separation in the higher part of the sail. From a
sailing point of view, this effect is not desirable, since the boat would slow down and excessively heel. On
the other hand, it is interesting to investigate how the forces change when the main sail and the traveller
are progressively eased, i.e. when the angle of attack is reduced and the twist increased. The effect of these
regulations is shown in the graphs moving to the left along the curves: both the aerodynamic forces
decreases even though not at the same rate: Cy diminishes more than Cx because the twist of the main
allows the flow to stay attached to the sail for a larger portion which mainly affects the side force.
From a sailing point of view, this behaviour underlies the importance of a good twist of the main sail that
allows to maintain a good thrust force, limiting at the same time the side load: this is paramount for
maximizing the boat speed that depends not only on the thrust but also on the boat attitude.

Considering the curves with respect to each other, is clear that the thrust force increases with the wind
angle. This behaviour is not related to an increased efficiency of the sail plan but rather to the more
favourable directions of projection of the forces. In fact, the lift force whose contribution is predominant
compared to the drag, even remaining almost constant (𝐶𝑙 ≈ 1.5) for max power configurations of all the
AWA considered (Fig. 57), it gets more aligned with the forward direction of the hull, thus generating a
larger contribution to the thrust force, which is even able to counteract the larger drag.

Fig. 57 Lift and drag coefficients for the sole upright configurations at different AWA. The max power
configurations for each wind angle is the point further to the right-hand side of the plot

If we finally look at the coefficients arising from the same wind angle in upright and heeled conditions (Fig.
55 and Fig. 56), we can observe that in the latter cases the values of both aerodynamic forces are lower.
This fact is probably related to the flow main direction around the sail plan and particularly around the jib,
where the forces are oriented more towards the forward direction: in fact, in heeled conditions is expected
to have a flow with a larger vertical velocity component that interferes negatively on the development of
aerodynamic forces. In the next paragraph, the pressure distributions of the upright and heeled cases will
be compared, clarifying this aspect.
In summary, the depowering curve graphs denote the effectiveness of the wind tunnel experimental
apparatus and procedure. Besides assuming values coherent with the ones presented in literature, the
coefficient trends meet all the preliminary expectations.

57
7.1.2 Pressure distributions
The kind of information obtained from pressure measurements in wind tunnel is illustrated in Fig. 59: at
each wind angle, for every trim tested, the local pressure distribution was evaluated along 3 sections (25%,
50% and 75% of sail height) of mainsail and jib. The figure shows the detected sail shapes relative to the
max power configuration at 27.5° AWA on which the leeward and windward distributed pressures are
represented with blue and red arrows applied in correspondence of the probes.

Fig. 59 – Representation of the outcome from Fig. 58 – Expected pressure distributions on the mid
pressure measurements: distributions on both sections of the two sails
leeward (blue) and windward (red) sides of 3
sections

For a preliminary qualitative analysis, the schematic pressure drawings published in [20] (Fig. 58) can be
used as reference: the obtained distributions especially on the middle sections of main and jib present
remarkable agreements with the expected trends. In general, the aerodynamic loads on the headsail are
greater and oriented more towards the forward direction compared to the main sail; this characteristic is
due to the different shape of the two sails and their each other interaction. Recalling what was explained in
Chapter 2, the main sail affects positively the jib by increasing its effective angle of attack (up-wash effect)
and accelerating the flow on the leeward side thanks to the suction region created; to the contrary, the
headsail acts oppositely on the main (down-wash effect) which therefore is subjected to lower aerodynamic
loads.

The other aspect that leaps out from the picture is that the major contribution to the force generation is
given by the suction on the leeward side rather than by the over pressure on the windward, in accordance
with the expectations. This means that effectively the boat is not pushed from the wind but rather pulled
by the under-pressure regions created by the sails. This concept, well-known in aeronautical applications,
might result quite unintuitive to less experienced readers.

In the next paragraphs, for a more thorough analysis of the local pressure characteristics, the sail sections
will be presented by means of two-dimensional scaled graphs. Firstly, will be discussed the differences of
load distributions on sails due to the depowering procedures; secondly, the influence of the variation of
wind and heel angles will be evaluated.

58
Effects of Depowering process
In order to present the similarities and differences in pressure distributions arising from the various trim
regulations, three upwind cases at low, medium and high wind angles have been selected as examples of
the entire test spectrum (Fig. 61, Fig. 62 and Fig. 63). Each figure illustrates the pressure distributions
recorded during the consecutive depowered configurations: the dots with the same colors are related to a
single point in the depower curves previously shown and reported in Fig. 60. It shall be remembered that
the depowering procedure has been performed by regulating the main sail only, while the jib has been kept
at a fix trim for each wind angle. The increasing numbers in the legends indicate the progressive easing of
the sheets (n°1 represents the most tightened trim)

From a first look, the trends obtained present characteristics in accordance to the reference studies present
in literature ([9], [22]). Moreover, the different flow features described in Chapter 2, such as leading edge
separation bubble, curvature-related suction peaks and trailing edge pressures, are all clearly visible both
on main and jib experimental distributions. This is particularly true for the middle sections where the flow is
not disturbed by tip effects.

Fig. 60 – Depower curves of the three cases considered for illustrating pressure distributions. The
numbers represent the consecutive trims analysed and are used as reference for the reported pressures

59
Fig. 61 – Distributed pressures on the 3 sections of sails for the configuration AWA 22,5° - Upright

Fig. 62 - Distributed pressures on the 3 sections of sails for the configuration AWA 27,5° - Upright

60
Fig. 63 - Distributed pressures on the 3 sections of sails for the configuration AWA 35° - Upright

Main - Leeward
Starting from the main sections, as expected, the mast has great influence on the pressure distribution and
interestingly, the effect varies along the sail height and with trim. In addition to the plots presented so far,
Fig. 64 and Fig. 65 relate the developed pressures to the correspondent sail shapes on the three sections
considered for the cases of max power at 27,5° and 35° AWA, respectively.

In the middle section, the pressure plateau at the leading edge indicates the presence of a large separation
bubble that extends up to 10% of the chord in the more powerful configurations. As the main is eased, the
flow is less accelerated and consequently the depression region assumes less negative values. In these
cases, the greater sail curvature leads to higher (more negative) pressure gradients that allow to reach
similar values at the middle of chord; from that location, the downstream distribution remains almost
identical within the same depower curve. At the trailing edge, the pressure tends to constant values
indicating partial flow detachments that prevent the pressure to fully recover.

At the lowest section, the flatter shape of the sail reflects on the pressure distribution: at the leading edge,
since the flow is less accelerated, the pressure recovers faster leading to smaller separation bubbles; along
the chord, the limited curvature even for depowered trims, only partially increases the flow speed resulting
in lower (less negative) pressures without showing any peak, but rather a smooth trend. The positive
pressure gradient at the leading edge suggests that the flow leaves the sail without separating, which is
coherent with having a flat shape.

61
Fig. 64 – AWA 27,5°. Distributed pressures on detected sail Fig. 65 – AWA 35°. Distributed pressures on detected sail
shape sections shape sections

The highest section is extremely interesting as clearly shows the effect of twist. For all the three angles
considered and especially at 22.5°, the most tightened trims (lower numbers) present high angles of attack
that produce large leading edge bubbles and flat shapes that induce the flow to detach quite upstream: the
combination of these two effects results in an almost monotonic increasing pressure trends, as described in
[37]. When the traveler is eased and the main twisted, the sail curvature increases and the angle of attack
decreases, leading to pressure distributions more similar to the other sections. In real sailing conditions, for
achieving a good trim, this kind of regulation is essential because allows to depower the highest sections,
thus reducing their aerodynamic contribution which is mainly addressed to heel the boat and is responsible
of generating strong tip vortices, i.e. induced drag.

Headsail - Leeward
Regarding the headsail, it shall be remembered that its shape was not changed during the depowering
procedure; therefore, the pressure distribution differences between the various trims must be attributed to
the changes of main sail’s influence only. In particular, a more powerful suction on the main’s leeward side,
correspond to a stronger up-wash effect which, in turn, results in greater induced angles of incidence. With
reference to the 22.5° case, it is evident how the changing angle influences the jib’s leeward pressure
distribution, especially at the leading edge. More specifically, when the up-wash effect is very strong, i.e.
the angle of incidence very high (trims 1 and 2), the pressure at the leading edge reaches considerably low
values, the separation bubble enlarges and consequently the positive pressure gradient related to the
reattachment decreases. At the same time, the flat pressure distribution close to the trailing edge indicates
that the flow detaches prematurely due to the high curvature and large velocities. This finding, clearly
visible also on L1 of Fig. 65 confirms the distributed pressure described in [37] in case of very high angles of
attack.
62
As the main is eased, the up-wash induced angle of attack of the jib decreases and the leading edge bubble
gets smaller. In some cases, we can only grasp its presence from the positive pressure gradients that
indicate flow reattachment. The following suction peak is related to sail curvature that appears in
correspondence of the point of maximum camber of the sail. At the trailing edge, the low pressures suggest
that the flow leaves the surface almost with no separation.

The pressure distribution differences between the three sections are mainly related to the sail leading edge
orientation with respect to the incoming flow, which result in higher or lower angles of attack. As it has
been described, in the former cases, the pressure increases almost monotonically, whereas in the latter,
two suction peaks can be identified, one at the leading edge and one related to the sail curvature. L1 and L2
do not present a LE suction peak indicating that the angle of attack there is close to the ideal one, i.e. the
surface is almost aligned with the incoming flow. It is evident here that the ideal AoA is only a geometric
condition and does not implies achieving optimum performance [37].

Windward pressure distributions


The windward pressure distributions both on main and jib confirm the main features presented in Chapter
2. On the jib, the values closed to the leading edge are almost 1 in correspondence of the ideal stagnation
point. The value 1 is never reached since the flow is three dimensional and thus always have a vertical
component of velocity. The pressure then remains almost constant till nearly 80% of the chord, where due
to the effect of viscosity, the influence of the leeward side starts to be felt by air particles. In order to leave
the surface smoothly the flow velocity of leeward and windward side needs to be equal and, for this
reason, the flow on the windward side is accelerated. Consequently, the pressure decreases.

The situation on the mainsail is similar and reflects the three zones, VII, VIII and IX described by Wilkinson,
especially on L2 where the flow is less subjected to 3D effects (Fig. 66 and Fig. 67): the constant region
close to the leading edge depicts the presence of the separation bubble generated by the presence of the
mast, the following positive pressure gradient is due to the effect of the curvature (opposite to what
happens on the leeward side) and the final pressure decrease is related to the alignment between
windward and leeward flow veins.

Fig. 66 – Distributed pressure on mainsail at Fig. 67 – Schematic representation of mainsail


AWA 27,5° - upright - max power pressure distribution from Wilkinson’s study on 2D
profiles
63
Effects of AWA
Some interesting considerations can be drawn also comparing similar trims at different wind angles. Fig. 68
and Fig. 69 report the max power configurations of all the cases tested with the upright and heeled model,
respectively. The first aspect that leaps out from the upright cases is the similarity of pressure distributions
on the jib for all the cases, which was already documented in [9]: regardless of the wind angle, the values
assumed along the chord are almost identical, indicating that exists an optimum shape that maximizes the
aerodynamic efficiency of the sail. Nevertheless, at increasing wind angles, this same regulation results in
growing angles of attack that lead to strong leading edge negative peaks followed by monotonic pressure
distributions as illustrated in the previous section. The L1 pressure distribution of the heeled case at 35°
AWA, moves away from the other cases probably because of an unnatural shape assumed by the jib due to
the presence of tubing and scanners that somehow lay on the sail in the heeled configuration.

Some differences, especially in terms of values can be noticed on the main sail’s distributions. The different
orientation of the mast, which has an elliptical section, together with the different sail curvatures for the
angles considered, generates suction peaks of different intensity especially at the leading edge. At 35° the
mast remarkably accelerates the flow, resulting in a separation bubble with high mean velocity; therefore,
the pressure reaches and maintain for the length of the recirculation bubble considerably low values. As
the AWA decreases, the angle of attack of the mast reduces weakening the flow acceleration and,
consequently, the intensity of the depression. The curvature related suction peak is generally slightly
greater for higher angles of attack as expected, since in these conditions the main is trimmed with more
curvature. On the higher level, the behavior perfectly reflects the differences of angles of attack: as the
angle increases, the pressure distribution passes from a double peak trend to a monotonic increasing
tendency. Similar considerations apply for the heeled configurations as well (Fig. 69).

Fig. 68 – Distributed pressure of max power configurations of all the upright cases tested in wind tunnel

64
Fig. 69 - Distributed pressure of max power configurations of all the heeled cases tested in wind tunnel

Effects of Heel
Following the discussion started previously about the influence of the boat heel angle on force coefficients,
the pressure distributions illustrated in Fig. 70 clarify the relative load contributions from main and jib. The
plots refer to the cases of Max power at 25° wind angle both in upright and heeled conditions. Analysing
the figures, emerges that the pressure distributions on the three levels of the main sail are very similar,
presenting only minor differences regarding the curvature related suction peak, most probably caused by a
slightly different sail shape assumed in the two cases. On the other hand, the pressures on the headsail are
quite different both on leeward and windward sides. In particular, the heeled configuration presents higher
pressures (less negative) on the leeward side, especially at the leading edge, and less positive on the
windward, resulting ultimately in lower ∆𝐶𝑝 . These features can be explained considering that in heeled
conditions the flow is more three dimensional: the inclination of the sails induces the flow to move
upwards along the surfaces enhancing the vertical component of velocity; this interferes with the 2D flow
that develops along the sail chord increasing the velocity on the windward side and reducing the effective
angle of attack, which in turns weakens the suction region on the leeward side. This phenomenon is more
evident on the jib rather than on the main because of the sails shapes (the headsail has greater curvature)
and sail location (the flow is accelerated vertically in the first portions of the chord and then becomes more
horizontal following the surfaces); nevertheless, the depowering effect is almost equally evident on both
thrust and heel forces since the reduction is not limited to the leading edge areas, but rather interests the
entire jib section chords (Fig. 71).

65
Fig. 70 - Comparison between distributed pressures at 25° wind angle in upright and heeled conditions

Fig. 71 – Thrust and heeled coefficients of upright (0°) and heeled (30°) configurations.

66
7.2 Comparison of Wind Tunnel Measurements and Numerical Simulations
The scopes of this work regarding numerical simulations are to assess the effectiveness of the presented
experimental set-up in validating mathematical models and to investigate their capabilities in predicting the
relevant phenomena concerning upwind sailing steady aerodynamics. With these aims in mind, in this
section the numerical outcomes from RANS and VLM simulations will be presented, analyzing the quality of
the predictions and possible disagreements from experiments. For this purpose, comparisons of global
aerodynamic coefficients and pressure distributions will be discussed together with the visualization of the
most relevant flow patterns.

7.2.1 Aerodynamic coefficients


The numerical results are analysed in comparison with the wind tunnel experimental outcomes which are
taken as reference values. In order to give a comprehensive picture of the simulations performed, Tab. 5
summarizes the values of the four coefficients 𝐶𝑥 , 𝐶𝑦 , 𝐶𝑑 , 𝐶𝑙 relevant to the max power configurations
considered, for AWA ranging from 20° to 35° with both upright and heeled boat model. The three columns
gather separately the results form Wind Tunnel, RANS and VLM simulations so that each coefficient can be
compared for the three methods.

Tab. 5 – Global aerodynamic coefficients obtained experimentally and numerically in the max power configurations of the cases
tested

AWA Heel Wind Tunnel RANS VLM


Cx: 0,34 Cd: 0,15 Cx: 0,31 Cd: 0,17 Cx: 0,32 Cd: 0,16
20° 0°
Cy: 1,39 Cl: 1,42 Cy: 1,35 Cl: 1,38 Cy: 1,35 Cl: 1,37
Cx: 0,45 Cd: 0,20 Cx: 0,44 Cd: 0,20 Cx: 0,47 Cd: 0,20
25° 0°
Cy: 1,44 Cl: 1,50 Cy: 1,42 Cl: 1,47 Cy: 1,48 Cl: 1,54
Cx: 0,50 Cd: 0,22 Cx: 0,49 Cd: 0,22 Cx: 0,54 Cd: 0,21
27,5° 0°
Cy: 1,43 Cl: 1,50 Cy: 1,41 Cl: 1,47 Cy: 1,49 Cl: 1,57
Cx: 0,53 Cd: 0,26 Cx: 0,51 Cd: 0,27 Cx: 0,59 Cd: 0,22
30° 0°
Cy: 1,45 Cl: 1,53 Cy: 1,43 Cl: 1,49 Cy: 1,47 Cl: 1,57
Cx: 0,62 Cd: 0,28 Cx: 0,63 Cd: 0,31 Cx: 0,78 Cd: 0,28
35° 0°
Cy: 1,37 Cl: 1,47 Cy: 1,44 Cl: 1,54 Cy: 1,60 Cl: 1,76
Cx: 0,23 Cd: 0,20 Cx: 0,24 Cd: 0,19 Cx: 0,31 Cd: 0,15
20° 30°
Cy: 1,38 Cl: 1,43 Cy: 1,39 Cl: 1,44 Cy: 1,47 Cl: 1,54
Cx: 0,33 Cd: 0,20 Cx: 0,36 Cd: 0,21 Cx: 0,41 Cd: 0,17
25° 30°
Cy: 1,31 Cl: 1,40 Cy: 1,42 Cl: 1,52 Cy: 1,44 Cl: 1,56
Cx: 0,55 Cd: 0,32 Cx: 0,54 Cd: 0,42 Cx: 0,74 Cd: 0,28
35° 30°
Cy: 1,47 Cl: 1,68 Cy: 1,64 Cl: 1,83 Cy: 1,70 Cl: 2,00

For a better evaluation of the agreement between the values obtained experimentally and numerically,
also a graphical representation of these results is proposed: Fig. 72 refers to the upright configurations
whereas Fig. 73 illustrate 𝐶𝑥 and 𝐶𝑦 for the cases with heeled model. In fact, 𝐶𝑑 , 𝐶𝑙 represent only different
projections of the same forces; therefore, they will not be reported graphically.

67
Cx - Upright
0,80
Cx - WT
0,60
Cx [-]
Cx - RANS
0,40 Cx - VLM
0,20
20 22 24 26 28 30 32 34
AWA [°]

Cy - Upright
1,80

1,60
Cy [-]

Cy - WT
1,40
Cy - RANS
1,20 Cy - VLM
20 22 24 26 28 30 32 34
AWA [°]

Fig. 72 – Thrust and side coefficients measured and computed for upright configurations

Cx -Heeled
0,80
0,70
0,60
Cx [-]

0,50 Cx - WT
0,40
0,30 Cx - RANS
0,20 Cx - VLM
20 22 24 26 28 30 32 34
AWA [°]

Cy -Heeled
1,80

1,60
Cy [-]

Cy - WT
1,40
Cy - RANS
1,20 Cy - VLM
20 22 24 26 28 30 32 34
AWA [°]

Fig. 73 - Thrust and side coefficients measured and computed for heeled configurations

68
In general, the agreement between experimental and numerical coefficients is more satisfactory for the
upright configurations. This behaviour might be related to the more complex flow patterns arising from the
greater three-dimensional nature of the flow, which are more difficult to correctly simulate numerically. As
we have seen before, the leading edge pressure distribution is very sensitive to the angle of attack and the
heeled configuration introduces a higher level of uncertainty. Moreover, with the boat heeled, comparing
coefficients coherently is more difficult as the measured and simulated forces need to be oriented also in
the vertical plane; this might explain the higher errors on 𝐶𝑦 rather than on 𝐶𝑥 .

As expected, in every case, the RANS simulations better predict the wind tunnel coefficients compared to
the Vortex Lattice method. In particular, the thrust coefficient calculated with the viscous model accurately
follows the experimental one, remaining within 5% relative error for almost every configuration
considered. The heel coefficient is predicted correctly as well, particularly in upright conditions where the
errors stay below 6%; the heeled configurations present worse results for the reasons stated above but the
relative errors only slightly exceed 10%. Therefore, these figures confirm that the Spalart-Allmaras
turbulence model is very well suited for this kind of aeronautical applications, where flow separation
phenomena are limited to small areas. As it will be presented later on when discussing pressure
distributions, the major source of discrepancies between numerical and experimental outcomes originates
from an imperfect description of the flow around the leading edge area, interested by high gradients and
flow separation and reattachment phenomena.

The VLM code confirmed to give fairly good results regarding global coefficient especially at lower wind
angles, where viscosity has minor effects. Meeting the expectations, the errors on both coefficients get
larger as the wind angle increases: the lift is progressively overpredicted since the model completely
neglects the incipient flow separation. In upright conditions, the error on 𝐶𝑙 grows from 3% at 20° AWA to
almost 20% at 35° AWA and a similar trend is followed by all the other coefficients. In summary, the two
main sources of error of potential flow theory are related to the neglection of viscosity: on the one hand,
the resistance to the incoming air is underestimated because only the form portion of the drag is
considered; on the other, the lift potentially grows to infinity as the angle of attack increases, since no flow
separation is considered. Therefore, is very important to apply the Vortex Lattice method in circumstances
where the model underlying assumptions hold; for sail aerodynamics, this means at small angles of attack.

In the next paragraph, we will analyse more in details the origin of the highlighted discrepancies on the
forces by comparing the measured and computed pressures on the sail surfaces. This will be done firstly for
the results from the RANS simulations from which both leeward and windward pressures could be
extrapolated; afterwards, a comparison of the two numerical methods with the wind tunnel outcomes will
be proposed in terms of ∆𝐶𝑝 , that is the local pressure difference between the two side of the sails. The
reason for this lays in the adoption, in the present work, of the VLM code written by Larsson and Helmstad
([34]) that was not meant to provide pressure information from the two sail sides separately.

7.2.2 Pressure distributions


WT - RANS
Given the very good accordance found on global forces from wind tunnel and RANS simulations, it shall not
be surprising to obtain fine agreements also regarding pressure distributions. Nevertheless, being able to
cross the two information represents the real added value of the present work.

Fig. 74, Fig. 75 and Fig. 76 show the pressure distributions obtained experimentally and numerically for
three of the upright configurations considered. In particular, they report the comparison of pressure trends
along the main and jib’s sections at 25%, 50% and 75% height, referring to the cases of maximum thrust for
25°, 30° and 35° wind angles. The pressure value obtained in wind tunnel from each tap is directly
69
compared to the value calculated numerically at the same chordwise location on each section; in this
manner, results very straight forward to assess the differences between the two methods.

Fig. 74 – Wind tunnel – RANS comparison of pressure Fig. 75 – Wind tunnel – RANS comparison of pressure
distributions (leeward and windward) for max power distributions (leeward and windward) for max power
configuration at AWA 25° (upright hull) configuration at AWA 30° (upright hull)

Fig. 76 – Wind tunnel – RANS comparison of pressure


distributions (leeward and windward) for max power
configuration at AWA 35° (upright hull)

70
In general, the pressures measured experimentally are very well replicated by the RANS simulations that
are able to correctly follow the evolution of the flow along the entire chord of all the three sections.

The plots relative to the main are very satisfactory on the leeward side where the separation bubble due to
the mast and the following reattachment is detected accurately both regarding intensity and length. Along
the chord of L1 and L2, the curvature related depression is in every case optimally reproduced confirming,
among the rest, the high quality of the sail shape detection system adopted in wind tunnel. Some small
disagreements are observed on the top section where the influence of jib’s tip vortices increases the
complexity of the flow patterns. Nonetheless, the leading edge pressure fluctuations are correctly
identified according to the different angles of attack. At the trailing edge, even though the general trend is
properly reproduced, the Spalart-Allmars model slightly underestimates the flow separation that occurs at
the very end of the chord; this issue was expected knowing the formulation of the chosen turbulence
model, but does not represent a major problem for upwind investigations.

Interestingly, the most significant disagreements between experiments and simulations are observed on
the windward side of the sail. Specifically, on the main surface, the pressure is quite largely overpredicted,
indicating that the simulated flow is slower than in wind tunnel. This behavior, somehow surprising because
numerically the flow speed should be higher since the wind shear layer is not modelled, is most probably
related to the poor simulation of the interaction between mast and main sail. On the other hand, the sail
surface is reconstructed from a point cloud and therefore, the sail leading edge might slightly differ from
the real one, and the generated sail plan is placed manually on the virtual boat model. The combination of
these two procedures, inherently characterized by uncertainties, affects the sensitive interaction between
mast and main sail. This explanation is coherent also with the fairly accurate agreement on the windward
side of the headsail which is only marginally influenced by the forestay, present in the wind tunnel and not
modelled in OpenFOAM.

On the other hand, the leeward pressure distribution of the jib, especially close to the leading edge, is the
most critical to reproduce numerically since the flow separates and abruptly reattaches downstream. These
phenomena are very sensitive to small differences of angles of attack and local interactions and therefore
represent the most difficult areas to investigate ([23]). Despite these common obstacles however, the
results obtained are still satisfactory. For all the three wind angles considered, the lowest section present a
monotonic increasing pressure, evidence of high angle of attack. This feature is very well reproduced
numerically along the entire chord, starting from the LE sever suction peak till the flow separation region in
the rear part. Less accurate is the agreement on the two higher sections where the LE suction peak is
considerably underestimated. This behavior is most probably related to the lower angle of attack between
the sail surface and the incoming flow: contrarily to what happens on L1 where the sail is forced to assume
a certain shape by the tight sheeting, resulting in a very high angle of attack, moving upwards, the jib
surface becomes more free to follow the flow direction, tending naturally to reduce it. In this situation, the
flow is more sensitive to small variations of angle, hence an accurate accordance between experiments and
numerical simulations is more difficult.

In heeled conditions (Fig. 77, Fig. 78, Fig. 79) despite the wind tunnel pressure trends are less smooth, the
numerical results still reflect with reasonable accuracy the major phenomena. Regarding the main sail, the
heeled configuration complicates the description of the interaction between mast and sail. If in upright
conditions, the major disagreements were localized on the windward side, now the errors propagate to the
leeward side as well, resulting in overpredicted LE pressure peaks. In addition to all the uncertainties
already mentioned, this outcome might be partially attributed to the poor description of the vertical flow
movements, particularly relevant with the sail surfaces heeled. This aspect could be improved with a
specific optimization of the computational grid, that shall have cells as aligned as possible to the flow
direction.

71
Fig. 77 – Wind tunnel – RANS comparison of pressure Fig. 78 – Wind tunnel – RANS comparison of pressure
distributions (leeward and windward) for max power distributions (leeward and windward) for max power
configuration at AWA 20° (heeled hull) configuration at AWA 25° (heeled hull)

Fig. 79 – Wind tunnel – RANS comparison of pressure


distributions (leeward and windward) for max power
configuration at AWA 35° (heeled hull)

Similar considerations apply to the headsail. Even though the general tendencies are correctly reproduced,
the LE pressure is poorly predicted: on the leeward side the suction peak is overpredicted almost in every
case due to the difficulties in reproducing accurately the complex patterns of flow deflections; on the
windward side, the disagreement between experimental and numerical values propagates from the leading
to the trailing edge, resulting in the general overprediction of forces previously presented.

72
WT - VLM
In this section, the results from WT and RANS simulations are presented in a different way, in order to
compare them with the VLM outcome (Fig. 81, Fig. 80, Fig. 82). It shall be remembered that the code based
on potential theory does not include neither the mast nor the hull in the computations. Beyond these
differences, the graphs, which report the ∆𝐶𝑝 on sections, highlight the expected difficulties of VLM in
accurately predict local phenomena. Due to the marginal effects of viscosity, up to 30° AWA, the values of
the two lower sections both on jib and main reasonably correspond to the outcomes of the other methods,
resulting penalized mainly at the leading edge of the jib and on the lowest section of the main. There, the
pressure is underpredicted due to the strong influence of the numerical tip vortices generated at the
bottom of both sails. This effect is even more remarked at 35° where the pressure at middle of the chord
on main L1 is almost 50% lower than for the other methods. On the other hand, in the same configuration,
the lift on the jib is considerably higher since the mechanisms of separation and reattachment are
completely neglected: in fact, in potential flow theory, the higher angle of attack simply results in larger
aerodynamic loads. The same, behaviour is observed on the top sections of all the configurations
considered, where in reality, the flow partially detaches due to the high angles of attack and the influence
of tip vortices.

These considerations about the capabilities of VLM codes are in line with the other works present in
literature, where the inaccuracy in computing the pressure distributions close the leading edge and near
the head of the sails is widely documented ([20], [23]).

Fig. 81 – Distributed pressures of max power Fig. 80 – Distributed pressures of max power
configuration at AWA 20° (upright hull) from wind configuration at AWA 25° (upright hull) from wind tunnel
tunnel measurement, RANS and VLM simulations. The measurement, RANS and VLM simulations. The presented
presented coefficients are differential between leeward coefficients are differential between leeward and
and windward sides windward sides

73
Fig. 82 – Distributed pressures of max power configuration
at AWA 35° (upright hull) from wind tunnel measurement,
RANS and VLM simulations. The presented coefficients are
differential between leeward and windward sides

74
7.3 Numerical Results
A direct comparison of local information obtained experimentally and numerically is of paramount
importance in every research field as it allows to establish the level of accuracy of the simulations
performed. In this work, through the observation of forces and pressure distributions along sail sections, it
has been able to determine the trustfulness of two different numerical codes in predicting the behavior of
a sail plan in upwind steady conditions.

Being this clarified, the great usefulness of numerical simulations is the possibility to handle many
information about the problem of interest that are not available in experiments. Regarding sail
aerodynamics, this means mainly to be able to determine how the loads are distributed on the surfaces and
to visualize how the flow develops around the sail plan at each location. For these purposes, the RANS
simulations certainly offer a more accurate and comprehensive description of the situation, since the entire
space around the yacht model is discretized and calculated. With respect to the VLM code, only a general
understanding of how the flow develops around the sail plan is obtained through the calculation of the
strength of the vortices distributed on the sails.

In order to better understand the usefulness of numerical simulations, in this paragraph some examples of
distributed pressures and streamlines obtained from both codes are presented. It shall be remembered
that, due to time constraint, only the max power configurations of the cases tested in wind tunnel have
been simulated numerically.

7.3.1 Pressure coefficient


So far, the pressure distributions on sections have been presented in terms of point values at fixed
chordwise locations in order to compare experimental and numerical results (Fig. 83); however, these
represent only an extrapolation of the entire field computed. As an example, Fig. 84 reports the trend of
the pressure coefficient around the three
sections tested. Besides recognizing the
discussed features such as separation
bubbles, leading edge and curvature related
suction peaks, is clearly visible how the
pressure develops in the space around the
sails. Moreover, it is straight forward to
relate the pressure trends with the sail
geometry, such that optimizations of the
shapes and their interactions can be easily
studied.

Looking in details at the example proposed,


it is plausible that the main discrepancies
between experimental and numerical results
shall be attributed to a poor reconstruction
of the sail shape at that location, regarding
the overestimated curvature related suction
peak on the top section of the main sail, and
to the artificial thickness of 2 mm of the sails,
concerning the poor description of the jib’s
Fig. 83 – AWA 30° - Upright. Comparison of extrapolated
leading edge pressure peak. pressures from RANS simulations and wind tunnel outcome

75
Fig. 84 – AWA 30° - Upright. Pressure distributions around the three sail sections at 25%, 50%, 75% of the mast height. The patchy
representation has been selected on purpose in order to better identify the various pressure regions

76
Furthermore, 3D numerical simulations are necessary to study the overall pressure distribution around the
sail plan in its entirety analysing the effects of all the factors even between different planes. For example,
with respect to Fig. 85, it is possible to observe that the on the lower part of the jib the flow “feels” the
presence of the hull that accelerates the fluid vein. This kind of considerations, immediate with numerical
simulations, can only be supposed from experimental data. In addition, once the numerical set-up has been
defined, the performance of different sail regulations or even geometries is very straight-forward.

Fig. 85 – AWA 30° - Upright. Distribution of Cp on both leeward and windward sides obtained with RANS simulations

For preliminary analyses, VLM codes are very valuable, since they offer a general picture of how the loads
are distributed indicating where to focus on more detailed investigations. As shown in Fig. 86, the adopted
code correctly captures the effect of the increasing angles of attack, showing progressively higher loads
going moving from 25° to 35° wind angles. It shall be noted that in this case, it is plotted the ∆𝐶𝑝 , that is the
coefficient calculated with the differential pressure between leeward and windward sides. The middle
picture refers to the casa AWA 30°, thus corresponding to the case in Fig. 85. The depression peaks on both
jib and main sail are correctly described, whereas the influence of the hull is neglected in VLM simulations.

Fig. 86 – From Left to Right: AWA 25°, AWA 30°, AWA 35° - Upright obtained with VLM code. The pictures show differential pressure
and the scale colour is inverted compared to the RANS pictures: red indicates negative pressures and blue positive

77
7.3.2 Streamlines
Numerical simulations are valuable not only for the indications about the distribution of loads, but also for
the information regarding how the flow develops around the sail plan. To this end, the streamlines, curves
tangent to the local velocity vectors, give a clear picture. Fig. 87, which refers to the three sections of the
case AWA 30° - Upright, shows of how 2D streamlines look like around sails.

Fig. 87 – AWA 30° - Upright. 2D streamlines around the sail plan. The red lines indicate regions of two-dimensional high speed flow,
whereas blue lines represent zero 2D velocity. The areas without lines constitute regions in which the flow is three dimensional: the
streamlines in those regions originate from other planes and therefore are not included in 2D representations
78
This kind of representation shows the development of streamlines originating from a defined line upstream
the boat model. The streamlines displayed for each section lay all on the same horizontal plane, whereas
the regions without lines shall be filled with streamlines originating from other planes. Therefore, they
represent areas of strong 3D flow that can be cased either by flow separation, like downstream the mast
and at the trailing edge, or by the problem geometry, such as in the lower section of the jib. From the
picture, it can be also appreciated the up-wash effect of the main sail on the headsail, with the latter
immersed in the region of accelerated flow of the former.
Additional considerations, could be drawn from the observation of 3D streamlines and their development
downstream the boat (Fig. 88). In particular, these are interesting to understand the effect of the sail
geometry at the tips in order minimize the induced drag caused by the shedding of large vortices.

Fig. 88 – Three dimensional streamlines developed from a single horizontal line source. The progressive curling of the lines indicates
the shedding of vortices downstream the yacht

These are only few examples of the possible investigations that can be carried out with the proposed
numerical simulations. The scope of the present work with these was to demonstrate the vast capabilities
of numerical codes and to offer an effective validating procedure indispensable to claim the truthfulness of
the results.

79
7.4 Full-Scale Measurements
In this final section, some of the data gathered during the numerous acquisition sessions on the SYL are
presented and discussed mainly in relation to the wind tunnel outcomes. The results will be presented
similarly to the previous paragraphs, firstly discussing the force coefficients and then the distributed
pressures on sail sections.

At full-scale, as previously described, the tests were not as stable and systematic as in wind tunnel due to
several inherent environmental disturbances and uncertainties. Even though the measurement sessions
have been carried in the finest weather conditions and with the support of professional sailors, an intensive
post-process of the results was necessary to identify stable ranges within the acquisitions. Therefore, if in
wind tunnel each couple of Cx-Cy and each pressure distribution refers to a specific acquisition, at full-scale
each set of data is relative to an identified window within a longer acquisition, that satisfies the decided
criteria:
 Min Time frame: 10 s
 Max AWA range: 5°
 Max Wind Speed variation: ±5% from the range’s mean value
 Max Boat Speed variation: ±5% from the range’s mean value

In order to come up with both force and pressure coefficients, all the parameters relevant to each range
were then averaged in time.

7.4.1 Aerodynamic force coefficients


Following the examples of Bayati ([7]), the force coefficients Cx and Cy are presented firstly against the
Apparent Wind Angle (Fig. 92 and Fig. 91) and then grouped in depower curves. Two different graphs are
reported: one relative to left and one to right sheets. From the figures, leaps out the difference of values
assumed by Cy that, in the right sheet case, are almost half than for the left side. Moreover, compared to
the values obtained by Bayati in previous experimental campaigns with the same measurement system
([7]), both Cx and Cy this time are considerably lower.

Analysing the data coming from the load cells and inclinometers, they do not seem to present any anomaly;
therefore, the asymmetry of the results is most probably related to a misalignment of the internal frame
with respect to the hull. A recalibration of the entire system will be necessary for future force acquisitions
to clarify this point.

From the graphs, not much can be said quantitatively, whereas from a qualitative point of view, can be
observed that Cx increases with the wind angle, coherently to what was found in wind tunnel, and that the
ratio Cy/Cx assumes values in line with the other reported experiences in literature.

In order to compare these data with the ones obtained in wind tunnel, the coefficients have been grouped
in depower curves for the angles considered (Fig. 89 and Fig. 90). Since, it was not possible to maintain the
same AWA for the entire depowering procedure, the acquisitions have been grouped in ranges of angles.
The trends found are analogue to the model scale ones: along the curves, Cy decreases faster than Cx,
indicating that the optimal trim is not the one that maximizes the thrust force but, rather, a slightly
depowered configuration that allows to keep a good propulsion while considerably reducing the heeling
loads. As all experienced sailors knows, lateral forces shall be kept as low as possible since they represent,
for various reasons, large sources of hydrodynamic drag.

In addition, as expected and found in wind tunnel, the curves present higher values of Cx as the AWA
increases, more than doubling the max thrust passing from the 20°-25° range to 35°-40°.

80
Fig. 92 – Thrust and side coefficients recorded during Fig. 91 – Thrust and side coefficients recorded during
right sheet navigation left sheet navigation

Fig. 89 – Full scale depower curves drawn for ranges Fig. 90 – Full scale depower curves drawn for ranges
of apparent wind angles (right sheet) of apparent wind angles (left sheet)

7.4.2 Pressure distributions


During the acquisition sessions on the SYL, besides the global force coefficients, also the pressure
distributions on sails were measured. In particular, following the same procedure as in wind tunnel, for
each apparent wind angle, different sail trims have been tested in order to evaluate the impact of the
regulations on the flow stream around the surfaces. Differently from wind tunnel though, at full-scale the
pressure measurements were differential between leeward and windward sides; therefore, the outcome is
81
presented in terms of ∆𝐶𝑝 along 8 or 16 taps on 4 sections. In order to cover the entire spectrum of wind
angles tested, several pressure distributions of depowering processes from 25° to 41° are reported (Fig. 93
to Fig. 97).

In general, the results are very satisfactory as they qualitatively match with the outcome of a similar study
performed by Viola and Flay on a 24-foot sailing yacht ([21]) and confirm the general trends found in wind
tunnel. Even from a quantitative point of view, the values correspond to the obtained model scale results
and full-scale literature references: on the headsail, the pressure coefficients are higher than on the main
due to the up-wash effect and all the relevant flow features assume coherent values.

Fig. 93 – Differential pressure distributions from the Fig. 94 – Differential pressure distributions from the
depowering process of AWA range: 25°-27°. The dark depowering process of AWA range: 31° - 33°. The dark
blue dots refer to the most tightened regulations; the blue dots refer to the most tightened regulations; the
progressive lighter colours correspond to depowered progressive lighter colours correspond to depowered
regulations of the mainsail regulations of the mainsail

82
Fig. 95 – Differential pressure distributions from the Fig. 96 – Differential pressure distributions from the
depowering process of AWA 34°. The dark blue dots depowering process of AWA 35°. The dark blue dots
refer to the most tightened regulations; the refer to the most tightened regulations; the
progressive lighter colours correspond to depowered progressive lighter colours correspond to depowered
regulations of the mainsail regulations of the mainsail

Fig. 97 – Differential pressure distributions from the


depowering process of AWA range: 38°-41°. The dark blue
dots refer to the most tightened regulations; the
83
progressive lighter colours correspond to depowered
regulations of the mainsail
Main
With reference to all the apparent wind angles considered, the pressure trends on the main sail show
similar features to the wind tunnel, yet with some interesting differences. At lower AWA (Fig. 93, Fig. 94),
the leading edge suction peak behind the mast is limited to the lower sections only; moving upwards along
the sail and easing the sheets, it further progressively decreases, with pressures that assume in some cases
even positive values. This behaviour denotes low angles of attack, due to the limited wind angles, that
further gradually decrease because of the easing of the traveller and, moving upwards, due to the
increasing sail twist. The combined effect of these three factors results even in negative angles of incidence
that generate positive pressures on the leeward side of the main, close to the leading edge.

As the angle to the wind increases, the lift force is increasingly oriented towards the forward direction of
the vessel, thus contributing less to heel the boat and more to the thrust. Therefore, it becomes convenient
to increase the angle of attack of the entire sail, by sheeting in the main, in order to generate more lift. This
increment reflects in more remarked leading edge peaks along the four sail sections and, in general, in
increasing aerodynamic loads at higher wind angles.

Interestingly, except for the 38° - 41° case, all configurations present a pressure plateau at the trailing edge
of L1 and L2; the diffused behaviour suggests that probably the sail was trimmed with an excessive tension
of the leach that induced the flow to detach prematurely.

Headsail
Very similar considerations about the angle of attack applies to the jib. As the wind angle increases, the
angle of incidence of the sail grows and consequently the aerodynamic loads get higher. It is curious to
note for AWA between 25° and 35°, the pressures never present leading edge peaks, whereas at higher
angles the distribution becomes monotonic, indicating very high wind incidence. Since during navigation,
the jib was tightened as much as possible for all the wind angles until 35°, the results obtained reveal that
the angles in this range are too low for the tested boat with the regulations available on board. In fact, in
order to generate higher angles of incidence, the jib should have been further trimmed in, but it was not
possible with the available equipment.

Moreover, contrarily to the wind tunnel cases, at full-scale the pressure trend did not change during the
depower process, most likely because the main sail was less eased from one regulation and the other,
resulting in minor differences of up-wash effects, which thus remained almost constant throughout the
depowering.

84
7.5 Comparison of Full Scale and Wind Tunnel Measurements
If we compare these results with the wind tunnel outcome, beyond the discussed discrepancies about the
low angle of attack of the jib, it is interesting to note that the main sails are trimmed very differently at full
and model scale.

Fig. 98 and Fig. 99 show the pressure distributions of the full and model-scale “max power” configurations
and the “min power” tested in wind tunnel at 25° AWA. The full-scale pressure presents more similar values
to the wind tunnel depowered configuration rather than to the max power one. At full-scale, where the
crew has the direct feeling of the behaviour of the boat, the sails are regulated in order to best perform
within real limitations; therefore, the sheets and traveller are tensioned reasonably for not heeling too
much and for maximizing velocity. On the other hand, in wind tunnel, the effects of reality are much less
appreciable from the trimmer, whose primarily aim is to maximize the aerodynamic loads. This results in
much more tightened shapes, indicated by the pronounced leading edge suction peak that extend all the
way up to the sail and that even increases moving upwards along the sail luff. This quite unrealistic
behaviour, leads to the conclusion that the more depowered configurations tested in wind tunnel are the
ones that better resemble the real navigating conditions, whereas the trims that produce higher
aerodynamic forces are often not achievable in reality.

Among the rest, this consideration highlights the paramount importance of having available full-scale
observations to judge the range of validity of model-scale experiments.

Fig. 98 – Full Scale, AWA 25°, max power configuration

85
Fig. 99 – Wind Tunnel, AWA 25°. To the left, pressures obtained in the max power configuration; to the right, distributions
measured with depowered sails

86
8 Conclusions
In the present study, the upwind sail aerodynamics have been extensively investigated with the most
recent technologies available both at model and full scale. The integration of experimental outcomes and
numerical simulations allowed to cover the subject from the broadest perspective.

In the experiments, the contemporary measurement of global forces, local pressure distributions and sail
flying shapes proved to give a comprehensive and thorough characterization of the aerodynamics of the sail
plan under the effect of different wind angles and boat attitudes. Forces have been measured by means of
various load cells distributed beneath the yacht model in wind tunnel and within the hull, at full scale.
Pressure distributions were measured along three or four horizontal sections thanks to dedicated systems
developed within the joint project among Politecnico di Milano, North Sails and CSEM. Sail shapes were
detected in real-time through customized laser scanners based on the Time of Flight technology.

The data gathered in wind tunnel experiments have been employed to validate two types of numerical
codes based on Vortex Lattice Method and RANS equations. Through the comparison of the experimental
outcome and virtual simulations, it has been possible to demonstrate not only the validity and ranges of
applicability of the numerical methods, but also the effectiveness of the measured quantities and
instrumentation in representing the point of contact between experiments and simulations.

Experimental tests have been performed both at model and full scale investigating forces and pressure
distributions with similar instrumentations. This combined study allowed to assess the validity of wind
tunnel investigations as a truthful representation of real sailing conditions; therefore afterwards, they were
singularly analysed in details and discussed in light of different regulations, wind and heeling angles. The
depowering curves of the tested cases, drawn with the measured forces as the standard procedure to
investigate yacht aerodynamics, were used to relate the regulation of the sails with the distributed
pressures on their sections. This correlation was indispensable for commenting in details on the obtained
distributions. In general, the measured trends are coherent with the other results present in literature and
can be explained in terms of conventional thin airfoil theory; nevertheless, the present work did not limit to
observe the characteristic features of pressure distributions described in other studies, but proposed a
systematic investigation of the influence of sail regulation, wind and heel angles on the flow characteristics
around the sail plan. This aspect is of great importance considering that the wind tunnel tests should
represent real navigating conditions, inherently subjected to several kinds of changes and disturbances.

The pressure distributions measured at full scale were the first results obtained on the SYL with the newly
developed system. The outcomes highlighted the importance of verifying the model scale figures with real
data since, for example, the twist of the main on regular boats is significantly higher than in wind tunnel. In
literature is rare to find such a broad and systematic treatise of pressure distributions at full scale.

Numerically, the objective of assessing the suitability of the two tested codes has been fulfilled comparing
both forces and pressures. This was possible primarily thanks to the accurate sail shape detection system
and the consequent virtualization process, that allowed to precisely reproduce the sail flying shape without
introducing excessive uncertainties on the input geometries for the simulations; this made possible to
determine the effective capabilities of the codes in simulating the wind tunnel situation. The Vortex Lattice
Method gave reasonably good results regarding global force for the upright model at lower wind angles and
presented coarser errors at higher angles and in heeled conditions. The RANS simulations, presented very
good results in upright conditions both regarding forces and pressures, where only minor discrepancies
were found in the regions of the leading edge, interested by very steep gradients. In heeled conditions the
agreement was less precise due to the more three-dimensional flow, more complicated to be simulated
numerically.

87
The presented numerical results and flow visualizations shall be intended as examples of the great level of
understanding that can be achieved through virtual simulations, once their validity has been proved with
experimental observations. These achievements constitute the baseline for further and more focused
numerical investigations.

In summary, the present work represents a benchmark for future studies regarding experimental
instrumentation and techniques and virtual simulations for the study of sail yacht aerodynamics. In
addition, the results gathered and reported within this project could be used, in future, as powerful
validating tools for computational codes of different kinds.

The main areas for forthcoming developments interest both experiments and simulations: the full scale
force measurement system needs a complete recalibration and the RANS based code shall be integrated
with a full refinement of the areas close to the walls and with algorithms that allow to run time-dependent
simulations. These improvements will certainly help to extend the study of sail aerodynamics to the
downwind conditions as well, a subject still open to vast areas of exploration.

88
9 References
[1] F. Fossati, Aero-Hydrodynamics and the Performance of Sailing Yachts, 2009.

[2] C. A. Marchaj, Sailing Theory and Practice, 1964.

[3] F. Fossati, I. Bayati, F. Orlandini, S. Muggiasca, a Vandone, G. Mainetti, and R. Sala, “A novel
full scale laboratory for yacht engineering research”, Ocean Eng., vol. 104, pp. 219–237,
2015.

[4] J. H. Milgram, D. B. Peters, and D. N. Eckhouse, “Modelling IACC Sail Forces by Combining
Measurements with CFD”, Elev. Chesap. Sail. Yacht Symp., pp. 65–73, 1993.

[5] Y. Masuyama “The work achieved with the sail dynamometer boat ‘fujin’, and the role of
full scale tests as the bridge between model tests and CFD”, Ocean Eng., vol. 90, pp. 72–83,
2014.

[6] K. Hochkirch and H. Brandt, “Fullscale hydrodynamic force measurement on the Berlin
sailing dynamometer”, 14th Chesap. Sail. Yacht Symp., no. October, pp. 33–44, 1999.

[7] I. Bayati, “Aerodynamics of Sailing Yachts: A New Generation Full Scale Dynamometric Test
Rig”, Politecnico di Milano, 2016.

[8] A. Vandone, “Measurement and Analysis of free-form objects: development of a solution


for flying sail shape reconstruction”, Politecnico di Milano, 2016.

[9] F. Fossati, I. Bayati, S. Muggiasca, A. Vandone, G. Campanardi, T. Burch, “Pressure


Measurements on Yacht Sails: Development of a New System for Wind Tunnel and Full
Scale Testing”, 22nd Chesap. Sail. YACHT Symp., no. March, 2016.

[10] R. G. J. Flay, I. J. Vuletich, “Development of a wind tunnel test facility for yacht
aerodynamic studies”, Jnl. Wind Eng. Ind. Aerodyn., vol. 58, pp. 231-258, 1995.

[11] R. G. J. Flay, “A twisted flow wind tunnel for testing yacht sails”, Jnl. Wind Eng. Ind.
Aerodyn., vol. 63, pp. 171-182, 1996.

[12] H. Hansen, P. J. Richards, K. Hochkirch, “Advances in the Wind Tunnel Analysis of Yacht
Sails”, 26th Symp. Yacht Des. Constr., pp. 1–27, 2005.

[13] H. Hansen, P. J. Richards, P. Jackson, "Enhanced Wind Tunnel and Full Scale Sail Force
Comparison", 16th Chesap. Sail. YACHT Symp, 2007.

[14] G. F. Clauss, W. Heisen, "CFD analysis on the flying shape of modern yacht sails", 11th Int.
Cong. Maritime Association of the Mediterranean, 2006.

[15] Y. Masuyama and T. Fukasawa, “Tacking Simulation of Sailing Yachts with New Model of
Aerodynamic Force Variation During Tacking Maneuver”, 3rd High Perform. Yacht Des.
Conf., no. March, pp. 1–34, 2011.

89
[16] Y. Tahara, Y. Masuyama, T. Fukasawa, and M. Katori, “Sail Performance Analysis of Sailing
Yachts by Numerical Calculations and Experiments”, Intechopen.Com, 2008.

[17] Y. Masuyama, Y. Tahara, T. Fukasawa, and N. Maeda, “Database of sail shapes versus sail
performance and validation of numerical calculations for the upwind condition”, J. Mar.
Sci. Technol., vol. 14, no. 2, pp. 137–160, 2009.

[18] J. H. Milgram, "Fluid mechanics for sailing vessel design." Annual Review of Fluid
Mechanics, pp. 613-653, 1998

[19] S. Wilkinson, “Static pressure distributions over 2D mast/sail geometries”, Mar. Technol.,
vol. 26, no. 4, pp. 333–337, 1989.

[20] I. M. Viola and R. G. J. Flay, “Sail pressures from full-scale, wind-tunnel and numerical
investigations,” Ocean Eng., vol. 38, no. 16, pp. 1733–1743, 2011.

[21] I. Maria Viola and R. G. J. Flay, “Full-scale pressure measurements on a Sparkman and
Stephens 24-foot sailing yacht”, J. Wind Eng. Ind. Aerodyn., vol. 98, no. 12, pp. 800–807,
2010.

[22] I. M. Viola, J. P. Pilate, and R. G. J. Flay, “Upwind sail aerodynamics: a pressure distribution
database for the validation of numerical codes”, Trans RINA, vol. 153, 2011.

[23] I. M. Viola, P. Bot, and M. Riotte, “Upwind sail aerodynamics: A RANS numerical
investigation validated with wind tunnel pressure measurements”, Int. J. Heat Fluid Flow,
vol. 39, pp. 90–101, 2013.

[24] D. Le Pelley, D. Morris, P. Richards, and D. Motta, “Aerodynamic force deduction on yacht
sails using pressure and shape measurements in real time”, Trans. R. Inst. Nav. Archit. Part
B Int. J. Small Cr. Technol., vol. 157, pp. 65–78, 2012.

[25] F. Fossati, F. Martina, and S. Muggiasca. "Experimental database of sail performance and
flying shapes in upwind conditions", Proc. Int. Conf. Innovation in High Perf. Sailing Yachts,
2008.

[26] K. Graf and O. Müller, “Photogrammetric Investigation of the Flying Shape of Spinnakers in
a Twisted Flow Wind Tunnel”, Proc. 19th Chesap. Sail. Yacht Symp., no. March, pp. 97–108,
2009.

[27] D. Le Pelley and O. Modral, “V-SPARS: a combined sail and rig shape recognition system
using imaging techniques”, 3rd High Perform. Yacht Design Conf., pp. 57–66, 2008.

[28] Y. Masuyama, “The work achieved with the sail dynamometer boat ‘fujin’, and the role of
full scale tests as the bridge between model tests and CFD”, Ocean Eng., vol. 90, pp. 72–83,
2014.

[29] B. Augier, P. Bot, F. Hauville, and M. Durand, “Journal of Wind Engineering Experimental
validation of unsteady models for fluid structure interaction: Application to yacht sails and
rigs”, Jnl. Wind Eng. Ind. Aerodyn., vol. 101, pp. 53–66, 2012.
90
[30] K. L. Hedges, P. J. Richards, and G. D. Mallison, “Computer modelling of downwind sails,” J.
Wind Eng. Ind. Aerodyn., 1996.

[31] H. Miyata and Y.-W. Lee, “Application of CFD simulation to the design of sails,” J. Mar. Sci.
Technol., vol. 4, no. 4, pp. 163–172, 1999.

[32] B. Krebber and K. Hochkirch, “Numerical Investigation on the Effects of Trim for a Yacht
Rig”, pp. 14–16, 2006.

[33] Y. Roux, M. Durand, A. Leroyer, P. Queutey, M. Visonneau, J. Raymond, J. M. Finot, F.


Hauville, and A. Purwanto, “Strongly Coupled Vpp and Cfd Ranse Code for Sailing Yacht
Performance Prediction,” 3rd High Perform. Yacht Des. Conf. Auckland, 2008.

[34] A. Helmstad and T. Larsson, “An Aeroelastic Implementation for Yacht Sails and Rigs”, KTH
Royal Institute of Technology, 2013.

[35] G. Castellano, “Simulazioni RANS dell’interferenza tra pian velici per lo sviluppo di un
programma di supporto alla tattica di regata”, Politecnico di Milano, 2015.

[36] M. J. Crompton and R. V Barrett, “Investigation of the separation bubble formed behind
the sharp leading edge of a flat plate at incidence”, Proc. Inst. Mech. Eng. Part G J. Aerosp.
Eng., vol. 214, pp. 157–176, 2000.

[37] I. M. Viola and R. G. J. Flay, “Sail aerodynamics: Understanding pressure distributions on


upwind sails”, Exp. Therm. Fluid Sci., vol. 35, no. 8, pp. 1497–1504, 2011.

[38] J. H. Stack, “A Sail Force Dynamometer: Design, Implementation and Data Handling”, Yale
University, 1985.

[39] J. Katz, A. Plotkin, Low Speed Aerodynamics. From Wing Theory to Panel Methods. 1991

[40] P. K. Kundu, I. M. Cohen, and D. R. Dowling, Fluid Mechanics. 2011.

[41] OpenFOAM User Guide, //cfd.direct/openfoam/user-guide/

[42] F. Fossati, S. Muggiasca, I. M. Viola, and A. Zasso, “Wind tunnel techniques for investigation
and optimization of sailing yachts aerodynamics”, 2nd High Perform. Yacht Des. Conf., no.
February, pp. 105–113, 2006.

[43] F. Fossati, G. Mainetti, M. Malandra, R. Sala, P. Schito, and a Vandone, “Offwind sail flying
shapes detection”, 5th High Perform. Yacht Des. Conf., pp. 48–59, 2015.

91
0001234156

You might also like