288381527

This item was submitted to Loughborough University as a PhD thesis by the
author and is made available in the Institutional Repository

(https://dspace.lboro.ac.uk/) under the following Creative Commons Licence
conditions.
For the full text of this licence, please go to:

http://creativecommons.org/licenses/by-nc-nd/2.5/
Finite element modelling of tennis racket
impacts to predict spin generation
By
David Weir MEng
A Doctoral Thesis submitted in partial fulfilment of the requirements for

the award of Doctor of Philosophy of Loughborough University
© David Weir 2012
i
Abstract
Over the last 20-30 years the subject of spin in tennis has become increasingly important. A
great deal of work has been done to establish the effects which, increased levels of spin
have, on shots. The most prominent effect of increased spin in a tennis shot is the resulting
deviation in flight which allows players to, amongst other things, strike the ball harder with
top-spin in the knowledge that it will still fall inside the court due to the extra aerodynamic
downward force. With such significant advantages available racket manufacturers are
naturally keen to maximise spin generation. That being said, very little research has been
performed into the subject of spin generation in tennis and the affecting factors.
This thesis details the development of a finite element model which is to be used to allow a
greater understanding of spin generation and how varying properties such as string density
(the number of strings in a string-bed), gauge and orientation affect its magnitude. The
primary aim, or goal, of this research is to create an FE model which can be used to model
oblique impacts and measure the resulting spin. Whilst considerable focus was placed on
developing novel, modelling techniques to create the FE model, a great deal of emphasis
was also placed on its validation. The validity of the model was examined under static
loading conditions, such as that experienced during stringing. The dynamic performance was
also validated using a combination of modal analysis and high speed video of dynamic
impacts. Each of the validation methods provided assurance of the models performance, with
all error margins less than 5%.
The two areas of the FE model which required the most attention were the interaction
properties (specifically coefficient of friction (COF)) and material properties. Previous studies
have sought to obtain a single value for the COF of a tennis racket/ball system but this study
examines how the COF varies as the strings interact first, with themselves and secondly with
the ball.
Each of the validation methods (dynamic and static) were deemed successful as they
provided concise data which could be readily compared with the results produced by the FE
model. Having validated the model’s performance, with respect to predicting outbound spin,
a number of oblique impact angles were modelled to allow a greater understanding of how
the mechanisms of spin generation change with the inbound trajectory of the ball. This
analysis showed that for the impact conditions studied the contact time of the impact was
ii
reduced from 6.2 milliseconds to 5.7 milliseconds when the angle was increased from 32
degrees to 40 degrees. Furthermore, a number of novel string-beds were modelled, with
varying string orientations (between 30 degrees and 60 degrees relative to the rackets
frame) and subjected to a similar analysis procedure, with their results providing the
concluding section of the thesis.
iii
Acknowledgements
I would firstly like to thank my supervisors Prof. Roy Jones, Dr. Paul Leaney and Dr.
Andrew Harland for all of their guidance and advice during the course of my PhD. I
would also like to thank the Sports Technology Institute technicians Steve Carr,
Andrew Hallam and Max Farrand for their invaluable contributions and advice.
I also wish to thank Dunlop Slazenger, in particular Martin Aldridge and Dr. David
Barrass, for all the support they have offered, both in terms of materials for testing
and their insightful knowledge of the tennis industry.
Lastly I would like to thank my wife, my parents and my brother Charles for providing
me with the support and security to fully focus on my PhD over the last three years
as well as my Undergraduate degree.
v
“The one thing that cycling has taught me is that if you can
achieve anything without a struggle it’s not going to be
satisfying”
Greg Lemond
vi
Contents
Cover Page .................................................................................................................. i

Abstract ....................................................................................................................... ii
Certificate of Originality ............................................................................................... iii
Acknowledgements .................................................................................................... iv
Contents ..................................................................................................................... vii
Nomenclature ............................................................................................................. xi
Glossary .................................................................................................................... xiv
1 Introduction ......................................................................... 1
1.1 History of tennis ............................................................................................. 1
1.2 Background .................................................................................................... 5
1.3 Aims and objectives ....................................................................................... 6
1.4 Thesis outline ................................................................................................. 8
2 Literature review ............................................................... 10
2.1 Finite element modelling .............................................................................. 10
2.2 Tennis strings ............................................................................................... 17
2.3 Modal analysis ............................................................................................. 22
2.4 Spin generation during tennis impacts ......................................................... 24
3 Generation of the initial finite element model ................ 33
3.1 Equipment and software packages .............................................................. 33
3.1.1 Co-ordinate measuring machine............................................................ 33
3.1.2 Abaqus .................................................................................................. 34
3.1.3 Altair Hypermesh ................................................................................... 34
3.1.4 Siemens NX 5.0..................................................................................... 35
3.2 Racket frame modelling ............................................................................... 35
3.2.1 Modelling of racket frame geometry ...................................................... 35
3.2.2 Racket head mesh................................................................................. 37
3.2.3 Racket Handle Mesh ............................................................................. 43
3.3 Modelling of strings ...................................................................................... 43
3.3.1 Modelling of string geometry ................................................................. 43
vii
3.3.2 String mesh ........................................................................................... 46
3.4 Modelling contact interactions ...................................................................... 50
3.5 Materials properties...................................................................................... 52
3.6 Modelling string tension ............................................................................... 55
3.6.1 Predefined field method......................................................................... 55
3.6.2 Connector element method ................................................................... 58
3.7 Concluding comments.................................................................................. 64
4 Static validation of the racket frame model ................... 65
4.1 Three point bend testing .............................................................................. 65
4.2 Validation of the racket frame and string bed model under tension ............. 67
4.2.1 GOM Aramis photogrammetry system .................................................. 68
4.2.2 Stage 1 .................................................................................................. 70
4.2.3 Stage 2 .................................................................................................. 72
4.2.4 Stage 3 .................................................................................................. 73
4.2.5 Stage 4 .................................................................................................. 74
4.2.6 Stage 5 .................................................................................................. 75
4.2.7 Stage 6 .................................................................................................. 76
4.2.8 Stage 7 .................................................................................................. 77
4.2.9 Stage 8 .................................................................................................. 78
4.2.10 Results/Discussion ............................................................................. 79
4.3 Validation of the string-bed stiffness profile.................................................. 89
4.3.1 Experimental measurement ................................................................... 89
4.3.2 Finite element model ............................................................................. 96
4.4 Discussion.................................................................................................... 98
4.5 Concluding comments.................................................................................. 99
5 Modal analysis ................................................................ 101
5.1 Introduction ................................................................................................ 101
5.2 Methods ..................................................................................................... 102
5.2.1 Stage 1 ................................................................................................ 105
5.2.2 Stage 2 ................................................................................................ 105
5.2.3 Stage 3 ................................................................................................ 105
5.2.4 Stage 4 ................................................................................................ 106
5.2.5 Racket Testing..................................................................................... 108
viii
5.3 Finite element model .................................................................................. 113
5.4 Results ....................................................................................................... 114
5.5 Discussion.................................................................................................. 122
5.5.1 Unstrung racket frames ....................................................................... 122
5.5.2 Strings ................................................................................................. 122
5.5.3 Strung racket frames ........................................................................... 123
5.6 Concluding comments................................................................................ 128
6 Creation of a materials database .................................. 130
6.1 Young’s modulus for various tension ranges ............................................. 130
6.1.1 Equipment ........................................................................................... 131
6.1.2 Methods and Results ........................................................................... 132
6.2 Variation in Young’s modulus due to extension rate .................................. 134
6.2.1 Methods and Results ........................................................................... 135
6.3 Coefficient of Restitution ............................................................................ 138
6.3.1 Methods............................................................................................... 139
7 Modelling of friction ....................................................... 147
7.1 The role of friction in tennis racket impacts ................................................ 147
7.2 Coefficient of sliding friction ....................................................................... 149
7.2.1 Tribology .............................................................................................. 149
7.2.2 Measurement Equipment Set-up ......................................................... 150
7.2.3 Testing procedures and results ........................................................... 159
7.3 Coefficient of sliding friction ....................................................................... 162
7.4 Friction properties of different strings ......................................................... 169
8 Oblique impacts of spheres .......................................... 173
8.1 Dynamics of oblique impacts ..................................................................... 173
8.2 FE modelling of oblique impacts ................................................................ 176
8.2.1 Creation of the model .......................................................................... 176
8.2.2 Results analysis................................................................................... 178
9 Dynamic validation of spin generation ......................... 189
9.1 Normal impacts .......................................................................................... 189
ix
9.1.1 Experimental set-up/method ................................................................ 189
9.1.2 Results/Discussion .............................................................................. 191
9.2 Oblique Impacts ......................................................................................... 194
9.2.1 Experimental set-up/methods .............................................................. 194
9.2.2 Results/Discussion .............................................................................. 195
9.2.3 FE Model ............................................................................................. 196
10 Modelling of different string patterns ........................... 205
10.1 Automation of string-bed geometry creation ........................................... 205
10.1.1 Generating points using Excel® ........................................................ 206
10.2 Spin of different string-beds .................................................................... 211
10.2.1 Different string density...................................................................... 211
10.2.2 Different string gauge ....................................................................... 212
10.2.3 Different string orientation ................................................................ 214
10.2.4 Spin generation of different string materials ..................................... 217
10.3 Results/Discussion ................................................................................. 217
10.3.1 Different string-bed densities............................................................ 217
10.3.2 Different string orientation ................................................................ 220
10.3.3 Different string gauges ..................................................................... 222
10.3.4 Different string material .................................................................... 223
10.4 Concluding comments ............................................................................ 224
11 Conclusions and Further Work ..................................... 225
11.1 Conclusions ............................................................................................ 225
11.2 Future Work ............................................................................................ 229
12 References ...................................................................... 233
13 Appendices ..................................................................... 243
13.1 Appendix 1: ITF regulation for tennis equipment .................................... 243
13.2 Appendix 2: GOM Aramis technical data (GOM, 2012) .......................... 245
13.3 Appendix 3: Abaqus error message........................................................ 246
x
Nomenclature
α = inbound angle (degrees)
αi = angle between points C and P (degrees)
αT = coefficient of thermal expansion
β = racket tilt angle (degrees)
ξ = critical coefficient of damping
μ = coefficient of friction
μs = coefficient of sliding friction
μr = coefficient of rolling friction
ρ = density (kg/m3)
σ = stress (N/m2)
Φ = angle 1 between cross and main strings (degrees)
φ = angle 2 between cross and main strings (degrees)
ν = poisson’s ratio
ω = frequency (Hz)
ωn = natural frequency (Hz)
ω1 = inbound spin (rev/min)
ω2 = outbound spin (rev/min)
λ = rebound angle (degrees)
a = acceleration (m/s2)
A = cross-sectional area (m2)
COF = coefficient of friction
COP = centre of percussion
COR = coefficient of restitution
xi
Cxi = x co-ordinate of point C at time increment i
Cyi = y co-ordinate of point C at time increment i
d = diameter (m)
E = Young’s modulus (N/m2)
Ei = internal Young’s modulus (N/m2)
F = force (N)
FE = finite element
fi = internal Force (N)
k = stiffness (N/m)
kb = ball stiffness (N/m)
ks = string stiffness (N/m)
KEi = elastic component of the stiffness matrix
KGi = geometric component of the stiffness matrix
li = elongated length (m)
l0i = original length (m)
m = mass (kg)
me = effective mass (kg)
mb = mass of the ball (kg)
ms = mass of the strings (kg)
MOI = moment of inertia (kg•m2)
r = radius (m)
Pxi = x co-ordinate of point P at time increment i
Pyi = y co-ordinate of point P at time increment i
R = normal force (N)
td = dwell time (s)
xii
t = time (s)
ti = time at increment I (s)
T = torque (Nm)
vb = velocity of the ball (m/s)
xiii
Glossary
Coefficient of restitution – ratio of outbound velocity to inbound velocity.
Dwell/contact time – The period of time for which the tennis ball is in contact with the
string-bed
Finite element model – a computer generated model which allows the user to
simulate loading of a structure
Oblique impacts – Impacts where the flight path of the ball is not normal to the plane
of the string-bed
Normal impacts – Impacts where the flight path of the ball is normal to the plane of
the string-bed
Racket – A strung racket frame
Racket frame – The frame which, the strings are woven into
Spin – The angular velocity of the ball about its centre
String-bed – The woven section of strings within the racket frame
String-bed density – The number/spacing of the strings within the string-bed
String-gauge – the cross-sectional diameter of the string
String orientation – the angle a string is strung across the racket frame
xiv
String-tension – This term refers to the magnitude of force applied to a string during
stringing of a racket frame and thus dictates the overall stiffness of the string bed
Sweet-spot – There are several technical definitions of this term but from a playing
point of view it is the point of the strings which provides the most pleasant feeling
impact
Top-spin – This term describes the state of motion where a ball spins around its own
centre in the same direction as the translational motion
xv
1 Introduction
1.1 History of tennis
Tennis has been documented, in one form or another, from as early as Egyptian
times, when two players would propel the ball towards one another whilst straddling
the back of a partner. When the players performed a “fault”, they would switch
positions with their respective partners and support them (Clerici, 1976). Exactly what
a “fault” entailed, however, is not entirely clear as the first record of court dimensions
and rules in tennis did not appear until the twelfth century in France (Clerici, 1976).
From the twelfth to the fourteenth century, there are a series of documents which tell
of holy-men playing longue paume or courte paume. The word paume in French
literally translates as palm, with the game later becoming known as jeu de paume,
which translates literally as, “the game of the hand”. Jeu de paume soon became the
game of the upper classes in France and when the area of the Louvre closest to the
Seine was built in the 16th century, Henry II of Valois ordered that it be built with
adequate space for his favourite pass-time (Clerici, 1976).
By the mid-fifteenth century the game of paume was beginning to contradict its own
name, and, the white leather gloves depicted by most artists were being replaced by
wooden paddles and, soon after, strung rackets. The earliest rackets were strung
obliquely, as opposed to the traditional method seen today, whereby the main strings
are placed along the axis of the handle and the cross strings perpendicular to said
axis. Developments in the racket were relatively limited for the next few hundred
years and the only major changes were in the shape of the head. By the early 1900s,
the experimentation with the shape of the racket had all but ceased, with the oval
shaped head becoming widely accepted as standard (Clerici, 1976).
1
By the early 20th century the rules of the game had also traversed through several
versions to become much more recognizable as the game we know today. The size
of the court had been standardized and was now completely free of any walls, unlike
earlier courts which were enclosed and often involved the walls as playing surfaces
(much like in the modern game of squash) (Clerici, 1976).
With the invention of television and the subsequent boom in televised sports, there
was a huge increase in the number of professional tennis players and the level of
skill with which the game was played. Suddenly the margins for error were narrowing
and the cost of being on the wrong side of that margin was increasing. As a result,
the investment into the development of new equipment was substantial and
exploration into alternative materials for the construction of rackets began.
The first commercially successful racket made from a material other than wood was
the Wilson T2000 steel framed racket, produced in the late 1960s and popularized by
American world number one Jimmy Connors. The development of tubular aluminium
rackets allowed for the introduction of the first over-sized rackets, increasing the area
of the head from 65 square inches to 110 square inches and providing the blue-print
for the modern racket. An example of such a racket is shown in Figure 1.1, along with
annotations which describe the tennis racket’s main features.
2
Mains Strings
Cross Strings
Racket Head/String-bed
Yoke
Racket Handle
Figure 1.1: Annotated image of a modern over-sized racket
The increases in head-size was well-received by both professionals and amateurs as

it increased the “sweet-spot” of the racket (there are several definitions for this term
but the simplest is the area which offers the least vibration during ball impact and
results in the most accurate rebound (Brody, 1981)) and reduced the chances of
playing a bad shot or missing the ball altogether. In short, oversized rackets
increased the quality of tennis across the sport.
By the 1980s the focus on racket performance was almost entirely on weight and
racket stiffness, hence, the introduction of carbon-fibre. Carbon-fibre came to the fore
3
of tennis technology when John McEnroe won the 1985 Wimbledon Championships
with the Dunlop Max 200G. Since then there have been a variety of graphite-
composites used including boron, titanium, Kevlar and fibre-glass. Other, more
innovative, approaches were also taken to increase the stiffness, such as changing
the profile of the racket. Wilson, adopted this approach with their “Profile” racket
where they increased the thickness of the racket at the top of the head, the area
which is required to provide the most reaction force when the strings deflect due to
impact, whilst reducing the thickness elsewhere (Cooper, 2003).
Despite the long history of tennis and the conception of rules in the game dating back
to as early as the mid-fifteenth century, it was not until the 1970s (when an East-
German horticulturist designed a new method of stringing rackets) that the ITF were
forced to define the term “Tennis Racket”. The method of stringing that brought about
this change is known as “Spaghetti stringing” and involved three sets of non-
intersecting strings, two sets of mains, which are fed through rollers, and one set of
crosses. The physics behind the increased levels of spin generated by “Spaghetti
Strung” rackets (Goodwill et al., 2002) will be discussed later.
However, the newly introduced regulations on rackets from the ITF were still
extremely lenient. The rules regarding tennis rackets can basically be summed up in
a few words: the racket and its strings must be uniform. The current ITF ruling on
rackets can be viewed in Appendix 1.
4
1.2 Background
Since the inception of metal rackets in the 1970s, and as racket technologies have
advanced, there has been a steady increase in the capabilities of tennis rackets and
subsequently the speed at which the game has played has increased. The main fear
of the tennis industry is that this increased speed may lead to a downturn in spectator
levels and thus, a reduction in the level of income which can be generated through
sponsorship. The governing body of the sport, the International Tennis Federation
(ITF), is extremely concerned about the effect this may have on the game at all levels
and recently commissioned research into the development of a novel test machine
which could “investigate racket performance under realistic service conditions”
(Kotze, 2005).
To attribute the increase in performance solely to improvements in technology would

be dismissive of the improved coaching and conditioning of athletes. These factors
however, are out of the ITF’s remit, leaving them no option but to focus on the
equipment if they wish to control the manner in which the game is played. The
mounting pressure on the ITF to regulate the performance of rackets has not just
been restricted to the issue of racket power. Some leading experts in the game,
including former Wimbledon and US Open champion John McEnroe, have publicly
called on the ITF to impart regulations on the amount of spin which can be generated
by today’s modern rackets. The effects of such regulations could be extremely
harmful to the major tennis manufacturers who would be restricted in terms of
product development in an already near optimized market. Dignall et al. (2004)
showed that an infinitely stiff tennis racket could only increase the power output of
today’s top-end rackets by 1.8%.
5
1.3 Aims and objectives
With spin becoming such an integral part of tennis the primary aim, or goal, of this
research is to create an FE model of the tennis racket which can be used to model
oblique impacts and measure the resulting spin of different string bed configurations.
A secondary aim is use the FE model to model novel string-bed designs and also
obtain a greater understanding of how spin is generated. In order to create such
models, a good database of material and the interaction properties is required and
hence the need to develop appropriate means of acquiring such properties.
Validation of the model is also a fundamental requirement to establish confidence in
results of the models and therefore establishing good validation techniques will form
a significant part of the work. Project goals are presented in tabular form in Table 1.1.
6
Primary aim Secondary aims Tertiary aims
Creation of an FE model
Measuring spin generation of a tennis racket
of novel string-beds (using Obtaining material
an FE model) properties
Investigating spin
generation performance of Obtaining interaction
tennis string-beds through properties
finite element modelling Obtaining a greater

Static validation
understanding of spin
generation of resulting
from oblique impacts Dynamic validation
Table 1.1: Project Goals
7
1.4 Thesis outline
The approach taken to creating an FE model of tennis racket impacts was first to
create an initial model to then build on. The strategies adopted in creating this model
are detailed in Chapter 3.
The next step is to validate the model’s prediction of the racket frame and string-bed
under static loading conditions. The experimental work and corresponding modelling
undertaken to validate the racket’s static performance, including the displacement of
the racket after stringing and the tension profile of the string-bed, is detailed in
Chapter 4.
In order to validate the fundamental model of the frame which is used in modelling
dynamic impacts, a series of modal analysis experiments are performed and these
experiments are detailed in Chapter 5. The experimental results of the modal
analysis of the racket are used to validate the vibrational characteristics of the model.
Chapter 6 outlines the approach used to obtain a comprehensive materials

properties library for a variety of tennis strings. This work is necessary to determine
the most appropriate materials model to apply.
In any dynamic analysis of two contacting interfaces the contact model is crucial to
providing accurate results. Chapter 7 describes the contact mechanisms which occur
during tennis racket impacts, the most suitable way to computationally model them
and the acquisition of the necessary experimental values.
In order to manipulate string-bed configurations such that they will generate an

increased level of outbound spin, a good understanding of the mechanisms which
generate spin is required. Chapter 8 describes a series of FE experiments where a
8
ball is projected at a number of oblique angles and the change in impact mechanism,
due to the inbound angle, is observed.
Once the model has been validated under static conditions the next step is to
examine its dynamic performance. This meant the introduction of a ball to the model
to create dynamic impacts and high speed video of experimental impacts as a
validation tool. The procedure is performed using a variety of ball impact velocities
and impact angles, all of which are examined in Chapter 9.
The concluding chapter in this thesis, Chapter 10 integrates all of the work done in
the previous chapters to realise the overall goals of the project, observing changes in
spin generation due to variations in the string pattern through FE modelling. Such
changes included changes in the string density, orientation and gauge (cross-
sectional area).
9
2 Literature review
2.1 Finite element modelling
There have been several attempts to use finite element analysis (FEA) to model
tennis rackets, balls and their impacts over the years. The first such attempt was by
Bitz-Widing et al. (1989), who used a tennis racket as an example application of
stiffening a frame with tension members. The strings were modelled using
isoparametric non-linear elements and the analysis produced displacements, forces
and stresses for the entire structure. The elements used to model the strings were
referred to by the author as “cable members” and have a stiffness that depends on
their internal force, Fi. The internal force of the members, were found using the
relationship show in Equation 2.1:
𝐹 𝐸 𝐴 Equation 2.1
𝑖= 𝑖 𝑖 (𝑙𝑖 −𝑙0𝑖 )
𝑙 0𝑖
Where Ei is the modulus of the string, Ai is the cross-sectional area, l0i is the original
length and li is the elongated length. The stiffness matrix of the elements consisted of
an elastic component, KEi, (equations 2.2) and a geometrical component, KGi
(equation 2.3).
1 0 0 −1 0 0
⎡0 0 0 0 0 0⎤
⎢ ⎥
𝐸𝐴 0 0 0 0 0 0⎥
𝐾𝐸𝑖 = 𝑙𝑖 𝑖 ⎢ Equation 2.2
0𝑖 ⎢−1 0 0 1 0 0⎥
⎢0 0 0 0 0 0⎥
⎣0 0 0 0 0 0⎦
10
0 0 0 0 0 0
⎡0 1 0 0 −1 0 ⎤
⎢ ⎥
𝐹 0 0 1 0 0 −1⎥
𝐾𝐺𝑖 = 𝑙 𝑖 ⎢ Equation 2.3
0𝑖 ⎢0 0 0 0 0 0⎥
⎢0 −1 0 0 1 0⎥
⎣0 0 −1 0 0 1⎦
The above matrices will combine to produce a complete stiffness matrix, subject to all
translational degrees of freedom, for the element (Equation 2.4). Bitz-Widing et al.
decomposed it into the above components, however, as she reasoned that some
areas of the element would behave elastically and some would not.
The local element stiffness matrix, KLi, is obtained through the summation of these
components and using standard finite element analysis practice (Rockey et al., 1983)
the forces within the element can be calculated by multiplying out the matrix shown in
(𝑗) (𝑗)
Equation 2.5, where 𝐹𝑖 is the Force at node “j” in the “i” direction and 𝑢𝑖 is the
displacement at node “j” in the “i” direction.
𝐸𝑖 𝐴𝑖 −𝐸𝑖 𝐴𝑖
⎡ 𝑙0𝑖 0 0 0 0⎤
𝑙0𝑖
⎢ 𝐹𝑖 𝐹𝑖 ⎥
⎢ 0 𝑙0𝑖
0 0 𝑙0𝑖
0⎥
⎢ 0 0 𝑙0𝑖
0 0 𝑙0𝑖 ⎥
𝐾𝐿𝑖 = ⎢−𝐸𝑖 𝐴𝑖 𝐸𝑖 𝐴𝑖 ⎥ Equation 2.4
⎢ 𝑙0𝑖 0 0 0 0⎥
𝑙0𝑖
⎢ 0 𝑙0𝑖
0 0 𝑙0𝑖
0⎥
⎣ 0 0 𝑙0𝑖
0 0 𝑙0𝑖 ⎦
11
𝐸𝑖 𝐴𝑖 −𝐸𝑖 𝐴𝑖
⎡ 𝑙0𝑖 0 0 0 0⎤
(1) 𝑙0𝑖 (1)
⎡ 𝐹1 ⎤ ⎢ 𝐹𝑖 −𝐹𝑖 ⎥ ⎡𝑢1 ⎤
0 0 0 0 ⎥ (1)
⎢𝐹2 ⎥ ⎢
(1) 𝑙0𝑖 𝑙0𝑖 ⎢𝑢2 ⎥
⎢ (1) ⎥ ⎢ 0 𝐹𝑖 −𝐹𝑖 ⎥ ⎢
(1) ⎥
⎢𝐹3 ⎥ ⎢ 0 0 0
𝑙0𝑖 𝑙0𝑖 ⎥ ⎢𝑢3 ⎥
⎢ (2) ⎥ = ⎢−𝐸𝑖 𝐴𝑖 𝐸𝑖 𝐴𝑖 ⎥ ⎢ (2) ⎥ Equation 2.5
⎢𝐹1 ⎥ ⎢ 𝑙0𝑖 0 0 𝑙0𝑖
0 0 ⎥ ⎢𝑢1 ⎥
⎢𝐹 (2) ⎥ ⎢ ⎥ ⎢ (2) ⎥
0 ⎥ ⎢𝑢2 ⎥
−𝐹𝑖 𝐹𝑖
⎢ 2(2) ⎥ ⎢ 0 𝑙0𝑖
0 0 𝑙0𝑖 (2)
⎣𝐹3 ⎦ ⎢ −𝐹𝑖 𝐹𝑖 ⎥ ⎣𝑢3 ⎦
⎣ 0 0 𝑙0𝑖
0 0 𝑙0𝑖 ⎦
The more advanced materials models used in today’s finite element packages (such
as the Neo-Hookean or Ogden material models available in Abaqus) will use a
similar approach to calculate the stiffness matrix with an increased number of
components.
Bitz-Widing et al. modelled the impact of the ball on the string bed as a static
parabolic load, as shown in Figure 2.1. The displacements, strains and stresses were
then calculated for each element, showing that the majority of stress propagation due
to the ball force occurred in the lower half of the racket
12
Parabolic Load
String-bed
Figure 2.1: Parabolic load representing the impact force of the ball against the
string bed (Bitz-Widing et al., 1989)
Due to the lesser capabilities of FE packages at the time, the slippage between the
strings was not modelled and the strings were considered to be rigidly fixed at their
intersection points.
Another example of FE modelling of a tennis racket is that of Hambli et al. (2006),

who modelled the racket frame as a rigidly clamped structure, around a network of
truss elements, representing the string-bed. The author’s use of truss elements
somewhat limited the performance of the model, since truss elements are unable to
simulate bending. His assumptions with regard to the boundary conditions, that the
racket was best modelled rigidly clamped, is also contrary to the accepted wisdom
within tennis communities that the best representation of a handheld racket is freely-
suspended (Brody et al., 2002). Hambli also neglected the work of Cross (2000a),
which found the coefficient of friction (COF) between tennis balls and strings to be
between 0.27 and 0.43, choosing instead to impose a value of 0.1
The most recent attempt at modelling of a tennis racket string-bed is that of Allen et
al. (2008). The author’s decision to use the relatively geometrically incompatible 8
noded brick elements meant that over 37,000 elements were required to provide a
suitable representation of the racket geometry. The author did not expand on his
reasons for using such a high volume of such computationally expensive elements;
13
therefore one would have to assume it was deemed necessary from an accuracy
point of view. Unlike other authors, Allen et al. neglected to model the string-bed as
part of a racket frame, instead choosing to fix the end of the strings to rigid cylinders.
The author claimed that to model the tensioning of the strings a force of 150 Newtons
was imposed on each cylinder, in order to provide an overall tensioning force of 300
Newtons to each string. Interestingly, the author chose not to validate his model
using rackets strung to this tension, instead using rackets strung to 200 Newtons and
289 Newtons.
FE modelling has also been applied to other areas of sports equipment. For example,
Penrose et al. (1999) used FE modelling in order to study the modal and impact
characteristics of a cricket bat. By increasing the stiffness of the bat model the author
was able to show that the contact time during impact decreased. As stiffness
increased, a similar conclusion was reached by Wiezel et al. (2004), where a
numerical model was used to show that increasing the stiffness of the string-bed in a
tennis racket will increase the contact time. The numerical model was developed
using basic mechanics and yielded Equation 2.6:
𝐾 +𝐾 𝑚 𝑚 1
𝑇𝑑 = 𝜋� 𝐾𝑏 𝐾 𝑠 �𝑚 𝑏+𝑚𝑒 �(1−𝜉2 ) Equation 2.6
𝑏 𝑠 𝑏 𝑒
Where Td is the dwell time, Kb and Ks are the stiffness of the ball and the strings
respectively, mb and me are the ball mass and effective mass of the racket
respectively and ξ is the critical coefficient of damping.
Vedula et al. (2004) used FEA to investigate the “sweet-spot”, which was defined by
Vedula as the position giving maximum rebound velocity, in a baseball bat. Vedula
used modal analysis to obtain the modes of the bat and their corresponding nodes.
14
Pendulum tests were also performed to obtain the moment of inertia (MOI) of the bat
and hence, centre of percussion (COP) using Equation 2.7:
MOI
COP = Equation 2.7
m×r
Where m is the mass of the baseball bat and r is the distance from the axis of
rotation to the centre of gravity. Two models of the bat were created, one wooden
(solid) and one aluminium (hollow). The wooden bat was modelled using eight noded
brick elements whilst, due to its hollow structure, the aluminium bat was modelled
using four noded shell elements. Due to the anisotropic nature of wood it was
necessary to assign it the properties of an orthotropic-elastic material (a material
model which accounts for directionality), whilst the aluminium bat was modelled using
isotropic properties.
By changing the mass distribution of the model, the author found that weighting the
mass towards one end of the bat resulted in the COP moving in the same direction.
Changes in material elasticity had no effect on the natural frequency node or the
COP, however, the natural frequency of the bat did change.
Casolo et al. (2000) also used FEA to model a tennis racket in order to investigate
several features of the tennis racket impact, such as the ball’s rebound velocity, the
impulse force imparted on the hand and the vibration modes and the position of their
nodes. The tennis racket was modelled using simple 2D beam elements which were
assigned material properties obtained from tensile testing of a dissected racket. The
stringing process was actually carried out within the model to evaluate the evolution
of stress and strain. The results showed a variation of 200 Newtons to 260 Newtons
for a specified tension of 240 Newtons. The model of the racket frame was then
validated by a series of modal frequency tests, which showed the natural frequencies
15
of the frame (134 Hertz and 399 Hertz for the first lateral and torsional modes
respectively) correlated well between the model and the experimental analysis.
16
2.2 Tennis strings
Tennis string tensions are traditionally quoted in Pounds-force but in the interests of
consistency all values will be specified in Newtons within this thesis. For ease of
comparison a look-up table comparing the most commonly used string tensions in
Pounds and Newtons can be found below in Table 2.1:
Pounds Newtons
40 178
45 201
50 223
55 245
60 267
65 290
70 312
Table 2.1: Equivalent SI values of pounds-force string tension
The physics surrounding the behaviour of a string-bed during an impact is extremely

complex. The player’s perception of the sweet spot is defined by the “feel” or comfort
of a shot and also the performance of the shot in terms of its translational velocities.
From a performance point of view, the sweet spot is the area of the string-bed which
will provide the best return of energy, thus the most powerful shot. From a comfort
point of view, there are two areas which contribute towards the most pleasant feeling
shot. The first is the centre of percussion, which, when struck avoids any out of
balance rotations of the racket along the handle (see Figure 2.2). The second is
simply the point of impact which will provide the least vibration, the fundamental node
(Brody, 1981).
17
The presence of a sweet-spot is created by the string’s tendency to deflect equally
regardless of where along its length it is hit. The result of this behaviour is that off-
centre shots seem stiffer as the string is deflecting unequally on either side of the
ball, yet the effective stiffness of the shorter side of the string is much greater and
dictates the feel of the shot. It is for this reason that string-beds will appear softer
when the string spacing becomes larger, as the length of the string which is able to
deflect uninhibited is increased. Consequently, in the majority of cases, the sweet
spot of the string bed will be where the longest main string crosses the longest cross
string (Brody, 1981).
(a) (b) (c)
Figure 2.2: Reaction forces when the ball strikes the racket at (a) the COP (b)
above the COP and (c) below the COP
An early study performed by Groppel et al. (1992) into the performance of different
tennis string materials found that synthetic strings (in this case nylon) produced a
lower coefficient of restitution (COR) than natural gut. This is thought to be due to the
increased natural stiffness of synthetic strings relative to gut. Groppel tested strings
at tensions of 178 Newtons, 223 Newtons, 267 Newtons, 312 Newtons and 356
18
Newtons. He found that COR generally decreased with increasing tension (although
the results were by no means linear) with the decrease in COR becoming less
pronounced as the tension was increased. Groppel reasoned that COR decreasing
as string tension increased is due to the increasing deformation of the ball, and
hence increasing contact time, on the stiffer string-bed. Naturally this trend is not
linear across all string tensions as the energy saved through decreasing contact time
will eventually be overcome by the string-bed’s inability to return energy to the ball as
the tension decreases.
The issue of reference stringing tension not being equal to the actual tension was
explored by Cross (2001). It is well known to professional racket stringers that the
reference tension of a string is not the actual tension, since the string will creep after
clamping. This is because tennis strings are viscoelastic and during the tensioning of
the string the elastic fibres will immediately react and stretch to allow the string to
elongate. In the first 20 seconds to 30 seconds after tensioning, however, the viscous
fibres will unfold and elongate in the string and partially unburden the elastic fibres,
thus slowing the rate of elongation.
Cross used load cells attached to strings to investigate the change in tension of
strings which were tensioned to 275 Newtons and 177 Newtons as shown in Figure
2.3. The loss in tension in the first 100 seconds was around 10 Newtons and then the
rate of loss began to slow, showing a linear trend against the logarithm of time.
During the impact testing it was shown that the strings briefly increased in stiffness
during the impact before decreasing to a tension lower than that prior to impact. Of all
the strings tested, gut’s increased elasticity meant that it was able to retain its
stiffness for longer during the impact testing.
19
Figure 2.3: Reduction in string tension over time Cross (2001)
The initial tension loss in strings during stringing can be limited, however, by
increasing the length of time at which the string is clamped before being released.
For instance, a string pulled to 311 Newtons of tension and held for 3 seconds will
have the same tension as one pulled to 223 Newtons or 267 Newtons in tension for
55 seconds, 60 seconds after being released. Cross et al. found that the string
tension of a racket is generally 30% to 40% lower than the reference tension. He also
found that the tension in the mains was generally higher than in the crosses. This
was thought to be due to the fact that the shorter cross strings created a greater
deformation of the racket in their direction than the longer mains. As a result the
racket would be reduced in length in the direction of the crosses and increased in the
direction of the mains, further increasing the tension of the main strings.
Knudson (1991) carried out research into the effect which string tension had on
vertical angle of rebound in tennis ball impacts. Three midsized rackets were strung
at 223 Newtons, 267 Newtons and 311 Newtons using 15L gauge nylon and gut
strings. Twelve balls were fired from a distance of 1.49 metres at a mean velocity of
20
19.1 metres per second and a mean angle of 25.6 degrees to the horizontal. The
nylon strings exhibited little change in rebound angle throughout the range of
stiffness values, whilst the guts strings varied significantly. At a tension of 223
Newtons they provided a more accurate rebound ratio (the ratio of the outbound
angle over the inbound) than nylon (1.4 compared to 1.5). However, whilst nylon
maintained its performance for the higher tensions of 267 Newtons and 311
Newtons, the ratio of rebound for gut deteriorated to 1.6 and 1.7 respectively. The
same trend applied for COR, with gut originally outperforming nylon at 0.51
compared to 0.5 for 223 Newtons but decreasing significantly to 0.43 for 311
Newtons whilst nylon only decreased to 0.49.
Knudson (1993) also investigated the effect which impact location had on rebound
angle accuracy for nylon strings. Central impacts were shown to give a rebound
angle of 1.1 degrees consistently, whilst impacts 80 millimetres off-centre gave
rebound angle ratios varying between 1.25 degrees and 1.65 degrees. This
difference may seem relatively insignificant but when the length of the shot (which
could be up to 24 metres) is taken into consideration such a difference in initial
projection could result in an extra 2 metres in shot length.
Further work on rebound angle accuracy in relation to string tension was carried out
by Bower et al. (1999). The author reasoned that lower string tension’s more
accurate rebound angle in experimental analysis is compromised in practice due to a
longer dwell time. The reason for this being, most ground-strokes are struck off-
centre and when this happens the racket becomes unstable and rotates about the
axis of the handle, altering the angle of the ball’s rebound plane. For looser strings
there is an increase in dwell time, meaning that the effect of the racket’s rotation will
be amplified. Bower quantifies this effect theoretically, stating that a decrease in
string tension of 45 Newtons would increase the dwell time by 0.5 milliseconds and
increase the difference in rebound angle by 1 degree.
21
Despite the broad range of literature showing the profound effect of changes in string
tension Bower et al. (2003) showed that player’s perception of changes in string
tension is not nearly as accurate as they would have academics believe. Bower
strung five identical rackets at 177 Newtons, 206 Newtons, 215 Newtons, 234
Newtons and 261 Newtons and the participants of the study were asked to
differentiate between the tensions of two rackets during play. To reduce the risk of
someone guessing correctly they were tested on two different levels of tension. The
first set of rackets they were tested on had a difference of 49 Newtons (177 Newtons
to 226 Newtons and 226 Newtons to 275 Newtons) and the last a difference of 20
Newtons (206 Newtons to 226 Newtons and 226 Newtons to 246 Newtons). If the
players could not correctly identify the higher tension of rackets with a 49 Newtons
difference they work backwards to a difference of 98 Newtons until they successfully
identified a racket with the higher tension. The results showed that most of the
participants could not differentiate between tension differences less than 49
Newtons, whilst out of 30 participants only 2 were successful in every test. The
author also expresses concerns at the similarity of the results to that of random
guess work, suggesting that the actual results were even poorer than indicated.
2.3 Modal analysis
In the context of this project, modal analysis can be used as a multi-purpose tool. Its
primary function is to act as a validation tool for the finite element model but a
convenient by-product of this validation is an insight into how the vibrational
characteristics vary from racket to racket (Mohr, 2008). The vibrational characteristics
of the racket play a large role in the player’s perception of the shot quality since they
dictate how the shot actually feels to the player. As a result the word “feel” is
commonly used as a descriptive for the quality of a racket, with a racket providing
little vibration termed as having good “feel”.
22
By performing a modal analysis of a tennis racket one obtains the fundamental node
of the racket, which is commonly used as a scientific definition of the area on the
string-bed defined as the “sweet spot”. The sweet spot is a term used by players to
define the area of the string bed which provides most pleasant feeling shot. As
eluded too earlier, there are several definitions commonly used for the sweet spot
(Brody, 1981), the most common of which are:
• The area which of the string bed which gives the highest COR.
• The centre of percussion of the racket.
• And, the one which is of most interest to this study, the fundamental node of
the string bed.
Whilst none of these areas/points fall exactly within each other, they are generally
found in the same area of the string bed, to the point where a player can often
encounter all three within one shot. As a result players identify this area as being a
large single region, rather than the collection of three smaller regions that it is.
In a study examining the sweet spot Hedrick et al. (1979) mounted an accelerometer
on the handle of a freely suspended racket to measure the impulse and oscillation
through the handle when the string bed was struck at various locations. Hedrick
found that, as expected, there was negligible impulse when the racket was struck on
its centre of percussion and there was minimal vibration when the node was struck.
In a study which focused on vibration damping of tennis rackets Brody showed a

higher level of damping when in the hand held position than when freely constrained.
This conclusion was drawn from an experiment where a small vibration sensor was
taped to the throat of the racket to measure the vibrations transmitted to the handle
when the string bed was struck. Brody also showed that, whilst the introduction of a
string damper had very little effect in reducing the vibrations of the racket, it did have
considerable effect in reducing the sound vibrations created by the strings.
23
Numerous authors (Axe et al., 2002, Mohr et al., 2008 and Hocknell et al., 1998)
have used modal analysis as a technique of validating their models and exploring the
changes in vibrational characteristics which result from altering certain parameters of
their model. One such example is that of Penrose et al. (1999) who used the finite
element package ANSYS to model two cricket bats, one representing a traditional
design and the other representing a less conventional design. For the fundamental
frequency the results were not particularly significant with the traditional design
having a frequency of 93.9 Hertz compared to 94.1 Hertz of the unconventional
design.
2.4 Spin generation during tennis impacts
The generation of top-spin in tennis is perhaps the most significant contributor to the
increased pace at which the game is played. Tennis players can generate top spin by
sliding the racket up and over the ball in a diagonal direction to the normal of the
ball’s incoming trajectory. This has the effect of imparting a forward rotational velocity
which, due to an increased drag force on the upper side of the ball, allows the ball to
decrease in height at a much faster rate (as shown in Figure 2.4). Consequently, the
ball can be hit much harder and still fall within the baseline, as demonstrated by the
“high top spin” trajectory in the figure below.
24
Figure 2.4: Ball trajectories for different spin rates
The defining moment of spin quantification in tennis was instigated by the

introduction of the “Spaghetti Strung” (SS) racket (see Figure 2.5). Goodwill et al.
(2002) carried out an investigation into the characteristics of the racket, which
consists of two non-intersecting planes of strings and a third set of strings which are
tied across the main strings (the pink strings in Figure 2.5).
A traditional SS racket was compared to four conventionally strung rackets, strung

with nylon and gut strings at tensions of 178 Newtons and 312 Newtons. A “BOLA”
ball firing machine was used to fire balls at the rackets, which were clamped at an in-
bound angle, α, of 39 degrees to the flight path of the ball as depicted in Figure 2.6.
The inbound ball spins imparted on the ball by the BOLA machine were in the range
of 0 radians per second to 420 radians per second for the conventionally strung
rackets and 0 radians per second to 300 radians per second for the SS racket. The
author was not clear as to why a consistent level of inbound spin was not used for
both racket types. The spin imparted on the ball by the BOLA machine was backspin
(negative angular velocity) whilst the resulting spin after impact was topspin (positive
angular velocity). The inbound and outbound angular and translational velocities of
the balls were measured using high speed video footage, from which the ball’s logo
was used as a reference point for spin measurements.
25
Figure 2.5: Spaghetti strung racket
The rebound angle, λ, of the conventionally strung rackets varied from 23 degrees to
5 degrees for a tension of 312 Newtons, and 23 degrees to 10 degrees for a tension
of 178 Newtons. The SS rackets had a markedly different trend with the rebound
angles varying from -5 degrees to +15 degrees over a smaller range of spins (300
radians per second compared to 420 radians per second). The amount of spin
imparted on the ball was in the region of 100 radians per second to150 radians per
second greater for the SS rackets and, for some inbound spin rates, almost twice as
great. This is thought to be due to the fact that the string bed is not interwoven and,
hence, is allowed to deflect in the direction of the ball before returning the energy in
the reverse direction of the inbound angular velocity.
26
α=39
Positive angle Racket handle
λ
Negative angle
Figure 2.6: Measurement set-up used by Goodwill et al. (2002)
Goodwill et al. (2006) carried out a more recent study into the ball spin generated by
different types of tennis strings. Identical tennis rackets were strung at 270 Newtons
with 30 different types of string and allowed to “settle” for 24 hours. Balls were fired at
each racket at a velocity of 25 metres per second, angles of 40 degrees and 60
degrees to the normal, and spin rates ranging up to 400 radians per second. The
impacts were filmed at 1,000 frames per second and three mutually perpendicular
lines were drawn on the ball to calculate the rate at which the ball was spinning.
It was observed that polyester strings gave higher outbound spin rates for all
impacts at 40 degrees and only at inbound spin rates of 400 radians per second did
polyester give less outbound spin than any other string for 60 degrees. Goodwill
argued that the frictional properties of the strings would have no effect on the level of
spin imparted on the ball since all strings possessed frictional coefficients high
enough to initiate rolling and any increase in the COF above this level would not
increase spin. He also reasoned that the increased stiffness of the polyester strings
would allow them to recover faster from the deflection caused by impact, thus
rebounding earlier and losing less energy than the other strings. This is a statement
which many authors disagree with; including Cross (2000) who stated that the COF
was “an important parameter since it determines the dynamics of the collision in a
27
direction parallel to the string bed and… the amount of spin which can be imparted
on the ball”.
The difference in trend associated with the steeper, 60 degree angle is accredited to
the fact that the ball is more likely to slide along the string bed, resulting in a different
mechanism of spin production. The polyester strings would be less able to deflect in
the direction of the ball as it impacts, hence the ball will be more likely to slip over
them, rather than rebound in the opposite direction, as in the case of more elastic
strings.
Although Goodwill deemed friction to have a negligible effect of the generation of

spin Ashcroft et al. (2002) carried out a number of experiments with the intention of
identifying a relationship between the two parameters. Using high speed video
footage, outbound spin rates were obtained for impacts, without inbound spin, at
angles of either 30 degrees or 35 degrees. Three different racket types, strung at 225
Newtons and 315 Newtons, were clamped between two plates and allowed to travel
only in the plane of the racket itself (i.e. the racket can only move along the axis of its
handle). The average value of the friction in this scenario was found to be 0.498,
compared to a value of 0.39, obtained from dragging a weighted tennis ball along a
string-bed. The author reasoned that the vast difference in these values is likely to be
due to the secondary factor of friction: mechanical interlocking. Ashcroft stated that,
whilst the majority of the friction in a drag test is due to the inherent surface
properties of the materials, the friction resulting from an impact test will be subject to
mechanical interlocking of the materials due to a higher normal force.
However, Bao et al. (2003) carried out a similar study using the same process of
dragging a weighted tennis ball over a surface and deduced the COF from the
required force. Contrary to Ashcroft et al.’s findings, during this experiment Bao found
that the COF reduced as the contact force was increased. This would lead one to
conclude that the weighted sled test is not an accurate enough representation of a
dynamic impact to provide reliable values of the COF.
28
As well as exploring the effect which string tension has on the production of post-
impact angular velocity, Bower et al. (2007) also looked at the role played by racket
stiffness, which was altered by clamping the racket at various positions along its
length. The rackets were strung at tensions of 180 Newtons, 225 Newtons and 270
Newtons using a conventional stringing machine and balls were fired at varying
velocities of 16 metres per second, 20 metres per second and 24 metres per second
at an angle of 45 degrees to the plane of the racket. The angular velocity
measurement system used was similar to that used by Goodwill et al. (2006). The
spin rates of the balls increased with string stiffness and impact velocity but seemed
to be unaffected by the stiffness of the racket. The spin rate also increased as the
impact speed was elevated, with a difference of 34 radians per second between high
(24 metres per second) and medium (20 metres per second) speed impacts and 16
radians per second for the low (16 metres per second) and medium speed impacts.
Bower also identified a small, yet significant relationship between string tension and
the level of spin imparted by the ball.
Contrary to this Goodwill found that there was no relationship between

experimentally obtained results and argued that the difference in spin at higher
tensions must be due to the excess force players exert during the shot to
compensate for lack of power (i.e. the player’s subconsciously feel that a higher
tension will reduce their power and strike the ball harder). However, Goodwill does
observe that dwell time of the ball increases by 20% when the string tension is
increased from 178 Newtons to 312 Newtons and the author assumed that the dwell
time must have some effect on the ability to impart spin on the ball since, reasoning
that the longer the ball lies on the string-bed the more its rebound characteristics are
likely to be effected by the motion of the racket through mechanical interlocking. The
reason for the dwell time of the ball increasing as the tension is increased is that the
strings account for only 2% to 4% of the energy loss in a ball-racket impact (Hatze,
1993) and therefore, although higher string tensions lead to a reduction in string
deformation, they also lead to an increase in contact time by increasing the
deformation of the dominant energy loss component within the system; the ball.
29
In general, most high outbound spin rates are produced as a result of oblique
impacts, the mechanism of which was explored by Cross (2003). Cross found that
most experimental analyses of tennis racket impacts were performed with the racket
fully clamped, effectively rendering the frame infinitely stiff in every plane. This is
especially important when investigating the effects of friction, as a rigidly clamped
racket will not allow for the movement of the racket which occurs as a result of the
frictional force between the racket and ball. To eradicate this effect Cross mounted
the racket on rollers, meaning that the racket is only infinitely stiff in the plane of the
racket, whilst free to deflect in all other directions. The rollers also measured the
perturbation of the racket, which, combined with the mass of the racket, allows the
frictional force exerted by the ball to be calculated. The conclusion of this work was
that for a ball impacting at around 25 degrees a frictional coefficient of 0.43 plus or
minus 0.02 was obtained.
Cross (2005) further examined the role of friction in a tennis racket impact with
particular emphasis on the variation with sliding speed. In order to draw a
comparison between low and high sliding speed, two separate experimental set-ups
were utilised. The first involved a small weighted sled with its underside lined with
tennis cloth being dragged along a smoothed table top at relatively low speeds
(0.001 metres per second to 1 metres per second) (Figure 2.7) whilst the second
involved firing tennis balls at an angle of 17 degrees and speeds ranging from 1metre
per second to 20 metres per second (Figure 2.8). Cross essentially discovered that
for low sliding speeds the COF decreased as mass was added to the sled whilst it
increased with the apparent contact area.
30
Figure 2.7: Variation in COF with sliding speed for different normal forces and
contact areas (Cross, 2005)
Figure 2.8: Variation in COF for a tennis impact at 17 degrees to the normal at
various speeds (Cross, 2005)
31
Similar investigations into the role of friction in the creation of spin have been carried
out in golf by Chou et al. (1994). Initially, Chou created an FE model of a club face
which showed that for club face angles less than 30 degrees spin occurs mainly as a
result of the ball “sticking” to the club face (i.e. mechanical interlocking of the ball and
the club-face). A secondary analysis was carried out involving balls being fired
against a steel block at a variety of angles. Each of the impacts was filmed using a
high speed camera and, using dots on the ball, the spin rates for each impact were
calculated. The results validated the FE model, showing that higher impact angles
resulted in less spin due to the ball slipping across the surface and the interlocking
effect not being initiated.
All of this previous research into the friction associated with tennis racket impacts is
beneficial on two levels. Firstly it demonstrates potential methods for the acquisition
of friction properties which can be used in a finite element model. Secondly, although
each tennis string will have its own COF, previous studies provide an insight into the
range in which one should expect values to fall when carrying out original testing.
This will provide confidence in the results of this testing, assuming the values
acquired are of a similar magnitude.
32
3 Generation of the initial finite element model
Like many processes, the development of a finite element model can often be an
iterative process. This project has taken the approach of developing certain aspects
of the model; such as material properties, interaction properties and energy
dissipation, individually rather than trying to perfect them all at the same time. In
order to do this however, it was necessary to have a base model which can be used
as a platform for assessing the performance of these individual aspects. This chapter
will detail the development of this basic, but crucial, part of the modelling process.
3.1 Equipment and software packages
3.1.1 Co-ordinate measuring machine
To obtain the geometry of the racket, an “LK Ultra” co-ordinate measuring machine
(CMM) was used. The CMM uses a measurement probe mounted on a bridge which
is free to move in Y and Z, whilst the probe can move along the length of the bridge
allowing motion in the X direction. The machine has a measuring volume of 1000
millimetres by 800 millimetres by 600 millimetres (X, Y and Z respectively). The
machine was calibrated to UK Accreditation Services (UKAS) standard and is
accurate to 0.0018 millimetres over 1 metre of measurement length (Singh, 2012).
33
3.1.2 Abaqus
Abaqus is a commercially available suite of finite element analysis softwares. Abaqus

contains two main solver options; Abaqus/Standard and Abaqus/Explicit.
Abaqus/Standard uses an implicit solving method and is used for static or long
duration analyses whereas Abaqus/Explicit uses an explicit solving method and is
more suited to high speed or small duration analyses. Both define equilibrium using
the relationship defined in Equation 3.1, where PF is the external applied force, IF is
the internal element forces, m is the mass matrix and a is acceleration:
m × a = PF – IF Equation 3.1
Both methods solve for the nodal accelerations at each time step but use different
approaches in the way they do so. The method used by Abaqus/Explicit allows it to
calculate the nodal accelerations for large displacement, dynamic analyses in a far
more efficient manor than Abaqus/Standard and, as such, will be the most commonly
used method in this thesis.
Abaqus provide their own "Complete Abaqus Environment" (CAE) environment which
provides everything the user requires to perform an analysis. Using the pre-
processor the user can mesh geometry, define load-cases and submit the model for
analysis. Once complete, the results of the job can then be reviewed in
Abaqus/Viewer; the dedicated post-processing package in CAE.
3.1.3 Altair Hypermesh
Hypermesh is one of many software packages available as part of the finite element
analysis software suite, Altair Hyperworks. Hypermesh allows users to mesh
geometry in a largely similar way to Abaqus CAE. Hypermesh, however, provides a
34
larger number of meshing tools than most other finite element packages, making it
more effective and quicker when it comes to meshing complex geometry. Hypermesh
can also export meshed geometry in various formats, including Abaqus input (.inp)
files, meaning that geometry can be meshed in Hypermesh and then exported to
other softwares, such as Abaqus.
3.1.4 Siemens NX 5.0
NX 5.0 (also known as Unigraphics) is an advanced computer aided design (CAD)

software package. Like most other CAD packages, NX 5.0 allows the user to
generate virtual geometry which can be used in a variety of applications, although it
is predominantly used as a design tool. NX 5.0 allows geometry to be generated and
exported in a file format which is compatible with most finite element software
packages.
3.2 Racket frame modelling
3.2.1 Modelling of racket frame geometry
Although the project is primarily concerned with the modelling of the string-bed, a
satisfactory model of the racket frame is a fundamental requirement. In order to
obtain the geometry of the racket, a co-ordinate measuring machine (CMM) was
used. Using this machine, a series of points from around a tubular section “Dunlop
200” aluminium racket’s outer and inner edges were captured and translated into the
CAD system NX 5.0. The points were measured at five millimetre intervals along the
length of the handle, around the head of the racket and then along the yoke of the
racket. This resulted in a data set of 252 measurement points
From this series of points, a spline was created to represent the outline of the racket.
The racket’s outline was then projected onto a solid block, the thickness of the
35
racket, and then extruded outwards to create a solid three dimensional profile of the
racket.
The cross sectional area of the racket is uniform throughout the majority of the frame
and consequently can be represented by the profile extracted from the block. The
handle however differs from the frame both in magnitude and profile and must be
represented by a separate entity. This entity is created by sketching the profile on the
bottom face of the frame and then extruding this sketch to form the handle (see
Figure 3.1).
Figure 3.1: Creation of the racket handle in NX 5.0
As well as capturing the points which represent the outline of the frame the CMM was
also used to measure the position of the string holes on the racket. The holes were
created using the imported points as their centre points. The extrusion representing
36
the frame was selected as the body from which the holes would be cut and the global
X and Y axis were assigned as the vector along which the holes for the cross and
main strings were created, respectively. Each of the holes were assigned a uniform
diameter of 1.6 millimetres.
3.2.2 Racket head mesh
The hollow cross section of the racket head is not modelled in NX 5.0, instead the
geometry is imported into the meshing package Altair Hypermesh as an “.iges” file
and the solid geometry is modelled as being hollow, using shell elements with a
specified thickness.
Unlike early carbon fibre rackets, which commonly possessed numerous irregularities
in their cross-sections, aluminium is largely consistent throughout and can be
modelled with constant property and a uniform thickness (which can be represented
using shell elements).
The elements used to model the racket head are known as S3Rs, where S denotes
that the elements are shell elements which can be used to model stress and
displacement, 3 denotes the number of nodes within the element and R denotes that
the element is reduced integration.
The introduction of reduced integration within the elements has benefits in terms of
computational efficiency but can also give rise to a phenomenon known in finite
element analysis as “Hour-glassing”. Hour-glassing is a term defined by Abaqus
(2010) and occurs due to only one integration point being used, which can lead to all
strains at said integration point being zero, which in turn leads to excessive distortion
of the element. Hour-glassing can however, be avoided by the distribution of point
37
loads and boundary conditions over a number of nodes, a technique which will be
described in further detail at a later stage.
Typically, shell elements can be used in situations where the wall thickness of the
structure is 1/10th of the overall dimension or less and the stress in the thickness
direction is negligible (Abaqus, 2007). Therefore, by assuming that the stress in the
thickness direction is negligible it is possible to dramatically reduce the number of
nodes per element and, often reduce the number of elements.
Unlike three dimensional continuum elements, however, conventional shell elements

can be used to model displacement and rotational degrees of freedom, i.e. they are
able to simulate rotation of their nodes. The benefit of having elements which can
model rotational degrees of freedom is that a greater number of outputs are available
to the user, e.g. nodal rotation/rotational velocity/rotational acceleration.
Using the “Automesh” feature in the “2D” sub-panel of Altair Hypermesh, an element
size of 0.5 millimetres was specified along with a maximum aspect ratio of 5. This
resulted in the generation of 262,380 S3R elements. The resulting mesh is shown
below in Figure 3.2.
Figure 3.2: Mesh of the racket head (262,380 S3R elements)
38
The main area of difficulty when generating the mesh was the string holes within the
frames. Given their relatively small diameter, capturing their geometry required a
much finer mesh than the rest of the frame. The result of using such a coarse mesh
when modelling the holes can be seen below in Figure 3.3. It was therefore,
necessary to increase the number of nodes and, hence the number of elements
around the hole from 3 to 6. The resulting mesh used in the analysis is shown in
Figure 3.4.
Figure 3.3 Coarsely meshed hole
39
Figure 3.4: Finely meshed hole
The difficulty of meshing arises from the need to generate as uniform a mesh as
possible. A good mesh will consist of uniform elements with consistent angle sizes.
Figure 3.5 shows an exaggerated example of desirable and undesirable elements.
The left-most element, containing similar angle sizes and edge lengths, is the more
desirable of the two, whilst the right most element would be likely to artificially alter
the stiffness of the structure, reduce the overall accuracy of the model and decrease
the stable time increment, thus increasing the computation time of the model.
Figure 3.5: Desirable and undesirable elements
40
In order to avoid the presence of poor quality elements, the quality was checked
using the “check elems” option in the “Tool” sub-panel. The “check elems” feature
allows the user to specify targets for the elements - such as minimum angle or length
– and highlights any elements within the mesh that do not meet these targets. Using
this feature, the model was checked for triangular 2D elements with angles outside
the range of 30 degrees to 130 degrees.
The failed elements were saved and re-meshed using the “elem cleanup” option in
the “2D” sub-panel. The “elem cleanup” option is a predefined subroutine, which
allows users to specify targets for element variables such as minimum/maximum
angle, aspect ratio, element length etc. Once the user has specified the desired
targets, the software will identify any failing elements and refine the mesh
accordingly. This process was repeated several times until all unsatisfactory
elements were eliminated.
When meshing the frame the element type in the automesh panel was set to “trias”,
which generated a group of S3R elements, as shown in Figure 3.5. The element
type was changed to the desired S3R by editing the .inp file, which was exported
from Hypermesh, in a text editor and simply editing the file to read:
“Element type = S3R”
Rather than,
“Element type = S3”
Where S refers to the fact it is a shell element, 3 refers to the number of nodes. By
adding the R to the element the user is specifying that there will be a reduced
number of integration points (or Gauss points) in the elements (Figure 3.6). The
integration points within an element are necessary to perform the integration
quadrature necessary for finite element analysis. By reducing the number of
integration points, the speed of the analysis is improved but the accuracy can, in
some circumstances, be diminished (Abaqus, 2007).
41
Integration Points
Figure 3.6: Position of integration points
42
3.2.3 Racket Handle Mesh
In order to represent the solid geometry of the handle, solid four noded tetrahedral
(C3D4) elements were selected (these elements will be described in more detail in
section 3.2.2). As specified in Section 3.2.1, the geometry of the handle was
represented as a simple solid extrusion. As a result, a far less detailed mesh was
required to capture the geometry of the handle. Within Hypermesh’s “Tetramesh”
sub-panel, the solid handle geometry was selected as the target to be meshed and
an element size of 1 millimetre was specified. The resulting mesh of 65,699
tetrahedral elements can be seen below in Figure 3.7.
Figure 3.7: Mesh of the racket frame (65,699 C3D4 elements)
3.3 Modelling of strings
3.3.1 Modelling of string geometry
As with the acquisition of the frame geometry, the geometrical features of the string-
bed (the points at which the strings intersect) were obtained using the CMM. The
intersection points were obtained by placing the measurement probe at each
intersection which, for a string-bed of eighteen main strings and twenty cross strings,
resulted in a data set of three hundred and sixty points.
43
Two slightly differing approaches were taken to modelling the strings. For initial
analyses the string bed was modelled simply as a 2D set of intersecting lines with no
cross sectional area. For this model the points were imported and used to create a
series of line representing the string-bed, the results of which can be seen in Figure
3.8.
The second model was more complex, in that it was a 3D model which captured the
weave of the strings, and required two sets of points to be imported. The first set of
points was offset to a height of -0.8 millimetres and the second to a height of +0.8
millimetres. The main strings were created using a spline through the points with the
same X value and alternating between -0.8 millimetres and +0.8 millimetres whilst the
cross strings were created using a similar strategy with the points of an equal Y
value.
A series of planes were then created at the left most end of each cross string and the
lower most end of each main string. A sketch of a circle was then created on the
plane with a diameter of 1.27 millimetres (the diameter of a 15 gauge string when
pulled to 223 Newtons tension (Racket Tech Publishing, 2005). A 3D representation
of the string was then created by using the “sweep along guide feature”, with the
spline acting as the guide string and the circle acting as the section string. The
resulting woven string-bed can be seen below in Figure 3.9. As in the case of the
frame an .iges file was then exported to Hypermesh.
44
Figure 3.8: Unwoven String-bed
Figure 3.9: Woven String-bed
45
3.3.2 String mesh
One of the main reasons for initially modelling the string-bed as a 2D structure is that
it allows a CAE user to make use of line elements (essentially one dimensional),
which, are far more convenient at an early stage. There are a variety of line elements
available in Abaqus, the simplest of which is the two noded truss element (Figure
3.10). Truss elements are able to move in 3-space, with each of the nodes having
three degrees of freedom, thus giving the truss element 6 degrees of freedom
overall. The degrees of freedom for a truss element are commonly displayed in
matrix form, as shown in Equation 3.2, where the subscript represents the node
number and the superscript represents the degree of freedom to.
Figure 3.10: 2 noded truss element
(1)
⎡𝑢1 ⎤
⎢𝑢2(1) ⎥
⎢ (1) ⎥
⎢𝑢 ⎥
[𝑢] = ⎢ 3(2) ⎥ Equation 3.2
⎢𝑢1 ⎥
⎢𝑢(2) ⎥
⎢ 2(2) ⎥
⎣𝑢3 ⎦
46
A truss element, however, is only able to sustain axial forces in its local basis and in
order to model the bending of the strings required an element which can be
subjected to transverse forces. This requirement leads us to the beam element which
introduces rotational degrees of freedom. The displacement matrix for a beam
element can be seen below in Equation 3.3, where the u components within the
matrix are translational and the θ components are rotational.
(1)
⎡𝑢1 ⎤
⎢𝜃1(1) ⎥
⎢ (1) ⎥
⎢𝑢2 ⎥
⎢𝜃2(1) ⎥
⎢ (1) ⎥
⎢𝑢3 ⎥
⎢𝜃3(1) ⎥
[𝑢] = ⎢ (2) ⎥ Equation 3.3
𝑢
⎢ 1 ⎥
(2)
⎢𝜃2 ⎥
⎢ (2) ⎥
𝑢
⎢ 1 ⎥
(2)
⎢𝜃2 ⎥
⎢ (2) ⎥
𝑢
⎢ 1(2) ⎥
⎣𝜃3 ⎦
For the purposes of contact, beam elements can be assigned a cross sectional area,
which must be kept constant during the analysis (Abaqus does offer a beam element
which can be assigned a constant change in cross-sectional area but the strain for
the element is constant in all directions). The fundamental assumption which must be
made when using beam elements is, therefore, that the cross sectional area of the
element will remain constant during the analysis.
The beam elements are satisfactory for initial approximations of the model but their
underlying shortcoming is an inability to plot stress contours due to their lack of a
third dimension. For this reason it is therefore necessary to model the structure as a
47
3D interwoven structure in the more advanced analyses. Therefore an appropriate
3D element must be selected. The default element, for geometrical performance as
much as any other reason, within most meshing packages is the four noded
tetrahedral (C3D4) element.
Figure 3.11: Four node tetrahedral element
Despite their ability to easily model a wide range of geometrical shapes, tetrahedral
elements, and indeed their 2D counterparts, can be overly stiff compared to elements
containing a higher number of nodes. However, given the number of elements which
will be required to model the curvature of the strings, artificial stiffness will not be an
issue in this case.
Once again, the 3D “Tetramesh” function was used to generate the mesh of the
strings. Due to the level of detail required to capture the geometry of the woven
string-bed, an element length of 0.2 millimetres was specified. The resulting mesh of
203,682 elements can be seen below in Figure 3.12.
48
Figure 3.12: The string mesh (203,682 C3D4 elements)
49
3.4 Modelling contact interactions
The modelling of contact within finite element simulations is an extremely complex

area. The two main methods of modelling contact with an Abaqus/Standard
simulation are:
• Surface based contact.

• Contact element based contact.
Surface based contact requires the definition of surfaces in either the part of
assembly module of Abaqus/CAE, and can be defined simply by the selection of
elements, either individually or by selecting elements which lie along the same angle.
Contact elements are available for analysis situations where surface contact is not
appropriate. Fortunately, the need for contact elements tends to be restricted to
analyses involving pipelines, axisymmetric simulation or heat transfer, none of which
are features of this study.
A third method of modelling contact, available exclusively in Abaqus/Explicit, is “all

with self” (AWS). AWS contact is the simplest contact simulation to define as it does
not require the definition or selection of any specific surfaces. AWS contact is
however, computationally expensive as it involves the entire model within the
simulation rather than a specified area.
Due to its simplicity, with respect to definition, the AWS contact method was initially
used to define the contact between the strings of the interwoven string-bed. For
unknown reasons however, the system was unable to correct initial element over-
closures, which had arisen from slight distortions of the string geometry during
meshing, and resulted in the strings slipping over one another.
50
For this reason the alternative approach of “Surface-to-Surface” (STS) contact,
where individual contact relationships are defined between two surfaces, was
adopted. Using the “select elements by angle” option, the six faces which define each
string were selected to define its surface. A STS contact interaction is then defined
for each of the 18 main strings with each of the 20 cross strings, resulting in a total of
360 interactions.
In order to define an interaction in Abaqus one must first define the properties of that
interaction. The four main properties of an interaction which can be defined in
Abaqus are:
• Mechanical contact properties

• Thermal contact properties
• Electrical contact properties
• Pore fluid contact properties
Although this study may look at the thermal effects of the contact between the strings
at a later time, at present the only one of the four properties which is of interest is the
mechanical contact.
The default tangential contact formulation within Abaqus is “frictionless” which

prohibits tangential interactions and allows infinite slipping between surfaces.
Conversely, the “rough” formulation does not allow any slip once the points defined in
the contact interaction are in contact. The contact formulation which is used in this
model is the “penalty” formulation. This formulation can be defined in terms of:
• Friction coefficient
• Slip rate
• Contact pressure
51
• Temperature
Although of the above options only the friction coefficient is compulsory. Although
some research has been carried out with respect to the friction coefficient between
the ball and the string-bed (Cross 2000a), extensive research performed during the
initial phase of this thesis did not uncover any work looking into the friction between
the strings. Accurately defining the friction coefficient of different strings could be
particularly crucial to modelling their performance with respect to spin generation and
therefore, research will be carried out to ascertain whether coefficients can be
accurately obtained. Until this research has been completed a default value of 0.1
shall be used for the friction between the strings.
3.5 Materials properties
One of the main parameters easily altered in a tennis racket is the strings’ material
properties. Tennis strings are available in a variety of materials and cross sectional
areas (gauge), both of which are easily altered in a finite element model. The
material properties of strings vary widely. Strings made of materials such as cow
intestine (also known as “natural gut”) and nylon have relatively good elastic
properties which allow them to deform more when the ball impacts, thus providing the
player with more power (since the ball deforms less).
The initial stretch of a nylon string is greater when compared to a gut string but this
value is irrelevant since it is the performance of the string when strung to tension
which is crucial. In this respect gut performs well, as, like nylon, it possesses a
relatively low modulus and, moreover, it retains its tension for longer than other
strings. The third most commonly used material in tennis strings is polyester.
Polyester possesses a similar strength to nylon strings but is much stiffer and also
more brittle. Polyester is, however, less expensive and therefore a better option for
player’s who play less or simply do not hit the ball quite so hard (Brody, 2002).
52
In order to make full use of the finite element model a series of string materials
properties were obtained in Chapter 5 to compare their performance. However, for
the initial stages of the modelling a basic Young’s modulus test was used to obtain a
smaller sample of strings materials.
In order to ascertain the properties of the strings a series of tensile tests (the
experimental method is described in detail in chapter 6) were carried out at the
following speeds:
• 5 millimetres per minute

And for the following string types:
• Gut (Wilson)
• Kevlar/Polyester (Prince Problend)
• Nylon (Prince Tournament)
The results can be viewed below in Figure 3.13.
53
Young's modulus of tennis strings
4500,00
4000,00
3500,00
3000,00
Gut
E (MPa)
2500,00
Kevlar
2000,00 Synthetic
1500,00
1000,00
500,00
0,00
speed speed speed speed Average
5mm/min 25mm/min 50mm/min 100mm/min (MPa)
Figure 3.13: Young’s modulus of tennis strings
The value used for the strings in the model was the average of the gut string, 2.483 ×
103 Megapascals.
For simplicity, within the early stages of modelling the racket frame was modelled as
aluminium. Whilst carbon fibre rackets are designed to have varying material
properties throughout the frame aluminium rackets are considerably more
homogenous. As a result the racket was assigned a single material section with the
following values taken from the Abaqus example problems within the documentation
(Abaqus, 2007):
E = 70 × 103 MPa
ρ = 2700 kg/m3
ν = 0.3
Where E is the Young’s modulus, ρ is the density and ν is the Poisson’s ratio.
54
3.6 Modelling string tension
A number of strategies were explored when modelling the effect of tension within the
stringbed. In the case of the 2D stringbed, each string was attached to the frame
using a “tie” constraint. Tie constraints require the selection of master and slave
nodes, which in this case, were the end node of the string and its corresponding
node on the frame, respectively. Once the master and slave nodes have been
defined the slave node is then constrained to follow all displacements of the master
node during simulation. As a result of utilising the tie constraint method, simply
imposing a force on the end of the strings would not be possible, since any
displacement of the string due to force would be experienced by the frame.
3.6.1 Predefined field method
As a result the stress state of the tension within the strings had to be modelled
without altering the geometry of the strings. This was achieved by creating a
predefined stress field assigned to the string-bed, which was changed to a stress
field using the keywords editor (which is accessed by right-clicking on the current
model in the model tree) of the model. It was necessary to first create the predefined
field in Abaqus CAE as a temperature field and later change it to a stress field using
the model keywords editor as the creation of a stress related predefined field is not
supported in Abaqus CAE. The predefined stress field was assigned a value of 112
Megapascals and “instantaneous” amplitude, meaning the stress would immediately
take effect.
The result of introducing tension can be seen below in Figure 3.14. The contours in
Figure 3.14 (a) and (b) represent the deflection of the string-bed due to the impacting
of the ball and Figure 3.14 (a) shows the more evenly distributed deflection one
would expect from a taut set of strings, whilst the deflection in Figure 3.14 (b) is far
more localised.
55
(a) Ball impacting 2D string-bed with tension
(b) Ball impacting 2D string-bed without tension
Figure 3.14: Effect of adding string tension
Although this technique gives a good representation of the effect the tensioning has
on the strings themselves it does not simulate the strain placed on the racket, due to
the fact there is no geometrical contraction from the strings. As a result an alternative
approach was adopted for the 3D string-bed, where the material property of the string
was assigned a thermal expansion coefficient and a temperature related predefined
field was defined.
56
The coefficient of thermal expansion can be defined as the strain divided by the
temperature drop. Since the diameter, d, is known to be 0.0016 metres, one can
calculate the string area from Equation 3.4:
𝜋𝑑2 𝜋∗0.00162
𝐴= = = 2 × 10−6 𝑚2 Equation 3.4
4 4
Where, if one assumes the strings are tensioned to a force, F, of 250 Newtons, the
stress, σ, can then be calculated from Equation 3.5:
𝐹 250
𝜎 = 𝐴 = 2×10−6 = 1.25 × 102 𝑀𝑃𝑎 Equation 3.5
Since the Young’s modulus, E, of the gut strings is known to be 2.483 × 103
Megapascals the strain, ε, can be calculated from Equation 3.6:
𝜎 1.25×102
𝜀= = = 0.05 Equation 3.6
𝐸 2.483×103
Thus, if a temperature drop of 10 Kelvin is introduced during the simulation, the

desired strain can be achieved from the following coefficient of thermal expansion
(derived using Equation 3.7):
𝜀 −0.05
𝛼= ∆𝑇
= −10
= 0.005 K-1 Equation 3.7
A temperature predefined field is then created with a magnitude of -10 and assigned
to the strings via an assembly set, containing the strings.
57
Although this method did have the desired effect of contracting the strings, it resulted
in a very uneven string pattern. As the strings were pulled to tension uneven gaps
began to open up between the strings. The uneven nature of the strings meant that
there was continuing movement of the strings as they tried to obtain a state of
equilibrium. It was hoped that given a “settling” period that the strings could obtain
this state of equilibrium but this proved not to be the case, as the strings position
simply continued to oscillate. Therefore, it was concluded that this would not be a
suitable method of modelling the string tension accurately.
3.6.2 Connector element method
Having discarded the previous method it was concluded that a method of modelling
the string tension was required whereby the strings could be fixed to the racket whilst
they were being loaded, without the load having the effect of expanding the racket.
With this in mind “connector” elements were considered.
Connector elements can be used in applications where two points are connected in
some way but the relationship is not necessarily linear (Abaqus, 2010). Unlike other
elements, connectors can be assigned behaviour dependencies such as locking or
failure mechanisms whereby certain degrees of freedom will be locked or freed when
a certain criteria is reached.
Rather than define the elements from geometry within a meshing packager as with
previous cases, connector elements are created by assigning a “connector section”
to a “wire” feature, which is defined between two points. The points which were used
to define the wire features representing the connector element geometry were the
centre nodes at each end of the strings and a reference point which was defined at
the centre of the hole through which the string will pass. The connector elements
were fixed to the racket, constraining the reference point, which defined its end point,
to the racket frame using a “coupling” constraint. Coupling constraints are similar to
58
“tie” constraints but they allow a single control point; in this case the reference point
which acts as the end point of the wire feature, to be coupled to a series of points; in
this case the points surrounding the hole.
Defining the wire features relative to a manually defined reference point allows the
length of the connector element to be readily adjusted. Once the wire feature has
been defined it must then be assigned a connector section, which defines the
connector behaviour. The connector behaviour can be defined as the available
degrees of freedom of the connector. For example, an axial connector section has
only one available degree of freedom and is only able to simulate motion along the
1st axis. Conversely, beam connector sections have no available degrees of freedom
and are used to model rigid connections between two points.
In order to model the uni-directional motion which will represent the tensioning of the
strings, “connector elements” were used. Connector elements are one dimensional
line elements that can be used to define a connection between two nodes. They can
be used to transfer displacements, rotations and forces between nodes but cannot be
used to model contact.
There are a number of connector element types within the Abaqus element library,
but in this case the most basic type was used, the axial connector. Axial connectors
can only be used to model displacement along the 1 direction of the element’s local
co-ordinate system.
A beneficial feature of connectors is that they can be rigidified or “locked” when

certain limits such as maximum displacement or force are reached. This
characteristic makes them ideal for modelling string tension as it allows the strings to
be pulled to a certain tension before being “locked” in contact with the racket frame.
59
One end of each string was attached to the racket, using a tie constraint. An axial
connector element was then created between the other end of the string and the
frame.
The connector elements were loaded using a feature known as “connector loads”.
Connector loads behave in the same way as conventional loads and can possess a
magnitude in any direction in 3-space. However, for an axial connector section the
only direction in which a connector load can have any meaningful contribution is the
1 direction, since all other directions are fixed and any loading would be
meaningless.
In order to create symmetric loading around the racket, the end of the string which
was loaded and the end which was assigned a connector was alternated. This
resulted in the loading pattern seen in Figure 3.15.
60
Figure 3.15: Connector loading of strings to simulate tension
A disadvantage of axial connector elements is that displacements and forces can

only be modelled in the 1 direction. In the case of the main strings this is not an issue
since their loading direction is, in fact, the 1 direction. However, in the case of the
cross strings, where the strings are loaded along the 2 axis, it is necessary to define
a local axis for the connector section. The local axis is defined with the 1 axis along
61
the line of the connector element by defining the reference point as the origin and the
end of the string as a point which lies on the 1 axis.
In order to fix the strings in place when they have been loaded to the appropriate
tension, the connector lock application is used. Connector locking criteria are defined
as part of the connector section and can be associated to either the displacement or
force experienced by the connector element. For example, by setting a position
locking criteria of plus or minus 0.1 millimetres, the connector will lock in the specified
degrees of freedom once this displacement is reached. Similarly, in this case, a force
criterion was defined whereby the connector element was locked once a force of 180
Newtons was reached.
In order for the connector to realise this force it was assigned a connector load with a
magnitude of 200 Newtons in the 1 direction and “smooth step” amplitude. The
smooth step amplitude was used, since the default “instantaneous” amplitude would
immediately impart a force of 200 Newtons on the connector and activate the locking
criteria immediately.
For the smooth step amplitude, the magnitude of the force F between two points in
time, t; (Fi, ti) and (Fi+1, ti+1) is calculated from Equation 3.8:
𝐹 = 𝐹𝑖 + (𝐹𝑖+1 − 𝐹𝑖 )𝛿 3 (10 − 15𝛿 + 6𝛿 2 ) Equation 3.8
Where the component δ is given by Equation 3.9:
(𝑡− 𝑡𝑖 )
𝛿= (𝑡𝑖+1 − 𝑡𝑖 )
Equation 3.9
The connector lock status can be verified by creating a history output for the
geometry set which defines the wire. The connector lock status history output is
62
available in each of the three translational and three rotational degrees of freedom.
For the axial connectors the only degree of freedom that is of interest is the 1st
translational degree of freedom. Figure 3.16 below shows the effect of the smooth
step amplitude, with the connector reaching its upper limit of 180 Newtons at 0.78
seconds. The transition from “unlocked” to “locked” then occurs during the next time
increment with the connector reaching lock status at a time of 0.8 seconds (where 1
represents a lock status of “ON”).
Figure 3.16: Connector Lock Status
63
3.7 Concluding comments
A satisfactory representation of the strung racket has been created using finite
element analysis. The model uses a novel method of applying a force between two
structures to model the string tension. In order to validate the model it must be
subjected to several stages of validation. The ultimate purpose of this model is to
study high speed ball-racket impacts but to apply the model to such a study without
first validating its correlation with experimental data under static load would be ill
advised. For this reason the racket will be subjected to a series of static tests which
will be recreated in Abaqus to validate the finite element model.
64
4 Static validation of the racket frame model
Although finite element models often appear realistic, it is far from advisable to simply
assume that the results which they produce are completely accurate. Whilst FEA is a
powerful tool it is often misused in this context, with many users simply creating
impressive animations for marketing purposes which, if taken as an accurate
representation of the system can be dangerous.
Given that this study is concerned with the modelling of dynamic impacts, the first
method of validation should be dynamically impacting a ball against the string bed.
This, however, would be a rather large leap to make without any intermediate steps
and, as a result, a series of static validations of the racket model were carried out to
ensure that the racket model is accurate under stationary loading conditions.
4.1 Three point bend testing
The most common method of quality assurance used by racket manufacturers is the
three point bend test (Woodward, 2010). The position of the load and the support is
often varied to test the bending properties of the racket at various points along its
length. The most common bend test uses supports positioned just inside the string-
bed at either end, with the load applied half way between the two, as shown in Figure
4.1.
65
Figure 4.1: Three point bend test
The same individual aluminium racket frame was tested unstrung and strung to 223
Newtons tension with nylon strings. During the test, a load of 150 Newtons was
applied at a rate of 1,000 millimetres per minute. The strung and unstrung racket was
tested five times each and averaged to give a stiffness of 116.7 (plus or minus 1.7)
kilonewtons per metre in the case of the unstrung racket and 129 (plus or minus 2.0)
kilonewtons per metre in the case of the strung racket, where the experimental error
was assumed to be one standard deviation (unless otherwise status all other
experimental errors in this thesis are calculated as one standard deviation).
The model of the racket was subjected to similar experimental conditions. Two rigid
cylinders were introduced to represent the supports and a third cylinder used to apply
the force. Two concentrated point loads were assigned to the nodes at either end of
the applicator cylinder and given smooth step amplitudes in order that the load would
66
be applied gradually, thus avoiding excessive distortion of the elements. The results
of the FE bend test were 112 kilonewtons per metre and 124 kilonewtons per metre
for the unstrung and strung rackets respectively. Both results were within 5% of the
experimental values. An error margin of this magnitude can be attributed to a number
of factors, such as the variation in the material properties due to the room
temperature at the time of the experiment.
4.2 Validation of the racket frame and string bed model under tension
An effective way to validate the static deformation of the model is by observing the
strung and unstrung states of the racket. When strung, the racket undergoes a
deformation considerable enough that it is visible to the naked eye. Observing the
deformation of the racket frame under string tension allows one to validate two
characteristics of the model:
1. The magnitude of the frame’s deformation under load from the strings
2. The reaction force being imparted to the strings by the frame and the effect
this will have on the overall string-bed tension
Validation of the second point is particularly important since it is the application of a

novel, previously untested method. Validation of FE models of structures under load
has traditionally been performed with the use of strain gauges placed on the
structure’s surface (Carreira et al., 1985, Vollmer et al., 2000, Stolk et al., 2002). A
less intrusive approach is now being adopted by many authors (Mosse et al., 2006,
Ivanov et al., 2009) known as image correlation photogrammetry.
67
4.2.1 GOM Aramis photogrammetry system
The validation of the frame deformation due to string tension was performed using an
“Advanced 3D Image Correlation Photogrammetry System” (ICPS) named “Aramis”,
(GOM, 2010) which, like all ICPSs, tracks the changes in a pattern (in this case a
random micro-pattern) which is applied to the surface of the structure being placed
under strain.
Aramis obtains the change in the surface pattern via camera images captured during
loading. Aramis tracks the deformation of the random micro-pattern on the target’s
surface and processes the information to produce a real time strain contour of the
target during loading.
The system can be used with a variety of cameras but the higher the resolution of the
camera the more accurate the result will generally be. The cameras used in this case
had a resolution of 1,280 pixels by 1,024 pixels. For this measurement set-up the
Aramis system is capable of measuring a strain range of 0.01% to 100% to an
accuracy of plus or minus 0.01% (GOM, 2012). The process flow used in conjunction
with the Aramis system is shown in Figure 4.2 and described in the next section. A
list of the performance specifications for the system configuration used in this
analysis can be viewed in Appendix 2.
68
Stage 2
Preparation of the racket
with speckle paint
Stage 1 Stage 3
Initial Calibration of the String racket to 223
equipment Newtons tension. First with
Cross and Main Strings,
then with Mains only
Stage 4
Place racket in jig, and capture image of racket with strings
and without
Stage 5
Create a new project and specify the facet size and number
(step size)
Stage 6
Order the images such that the strung racket (stressed
state) is the second image and the unstrung racket is the
first image
Stage 7
Mask off any area of the images that are not required for
the analysis
Stage 8
Initiate computation and, once complete review strain plots
generated by Aramis
Figure 4.2: Process flow for measuring strain of the racket frame due to string
tension
69
4.2.2 Stage 1
Before the random pattern was applied the racket was coated with a white Halfords
primer paint over the surface to increase the definition of the pattern. A black spray
paint (CRC Industries, Graphit 33) was then applied randomly to the entire surface of
the racket. The quality of the pattern is particularly important as:
• If the pattern is too sparse, the system will not be able to identify enough
points to track the motion of the surface between images.
• If the pattern is too dense, the system will not be able to distinguish between
the individual speckles.
• If the individual speckles are too large/small then the system will not be able to
identify them.
Examples of speckle patterns which performed poorly and well during the analysis
procedure can be seen in Figure 4.3 and Figure 4.4 respectively.
70
Figure 4.3: Example of a speckle pattern which performed poorly
Figure 4.4: Example of a speckle pattern which performed well

71
Figure 4.3 and Figure 4.4 also show a difference in lighting condition. Figure 4.4
shows the more desirable condition, with good contrast between the racket and its
background. Poor lighting, such as that in Figure 4.3, can lead to the system
becoming confused and recognising features in the background as speckle pattern.
The improved lighting condition in Figure 4.4 was achieved by decreasing the frame
rate from 250 frames per second to 50 frames per second and introducing additional
lighting using an Arri 650 Watt lighting unit. Another method of increasing the light
would have been to increase the aperture but, given the need for a certain level of
focus, a relatively narrow aperture was required. Had a still frame camera been used
the shutter speed could have been altered but because video cameras were more
readily available, they were used; with the first frame of each sequence used.
4.2.3 Stage 2
Two high speed, high definition Photron SA5 cameras were positioned at an angle of
70 degrees with respect to the plane of the string-bed. The orientation of the system
can be seen below in Figure 4.5. The system was calibrated using a calibration board
with a unique sequence of symbols, which were photographed at a series of
orientations. The calibration images were taken with the board a horizontal distance
of 440 millimetres from the beam which supported the two cameras. Once the
system has been initially calibrated at a distance of 440 millimetres from the target
image, the user is then free to adjust the target distance in order to provide a
significant field of view for their measurements. In the case of this study, a distance
of 1,000 millimetres was chosen.
72
Figure 4.5: Orientation of the cameras with respect to the string-bed.
4.2.4 Stage 3
With the system set up and calibrated, the pre-painted racket was strung to a tension
of 223 Newtons and left to “settle” for a period of 24 hours.
73
4.2.5 Stage 4
After this period the racket was placed into a specially constructed steel clamp -
fabricated by the Loughborough University Sports Technology Institute technicians -
to ensure that it was in exactly the same position when both the strung and unstrung
images were captured.
An image of the clamp holding a racket can be seen below in Figure 4.6. The racket
was clamped simply by placing the handle flat against the steel plate and tightening a
fixing plate over the free side of the handle. The fixing plate contained four holes
through which bolts were passed into the main body of the clamp. The plate was then
secured by tightening the nuts shown visible in Figure 4.6.
Figure 4.6: Racket clamp
74
The first image captured was that of the racket in the strung state. Although the
cameras used were video cameras rather than still cameras, the Photron Fastcam
Viewer software which was used allows the user to specify the number of frames he
or she wishes to capture. Therefore, the first and last frame specified for automatic
download was frame 1. The auto download function was used, meaning the image
was instantly downloaded to a specified file the moment it was captured. Following
the successful download of the image of the strung racket, the strings were removed
and the procedure was repeated for the unstrung racket.
4.2.6 Stage 5
Once captured, the images were imported into the Aramis analysis software as part
of a “project”. Upon creating a new project one is prompted to specify the “facet” size
and facet step which will be used when computing the images. A facet is a small area
of the surface which the software will process for each image before stepping on to
another area on the surface. Therefore, for optimum accuracy it is desirable to have
a large facet size, which covers a large area of the surface, and a small facet step,
allowing the system to move slowly over the surface, thus being less likely to lose its
position with regards to the other images.
Naturally, the larger the facet size and smaller the facet step, the longer the
computation time and the more memory is required. However, given that this analysis
only requires 2 images, the luxury of a higher accuracy can be afforded. An example
of different facet size and step values is shown below in Figure 4.7 and Figure 4.8.
For clarity, only one in every five facets is displayed and the size of the facets is not
to scale. It is still clear, however, that there is a much larger overlap for the more
accurate procedure represented in Figure 4.8.
75
Figure 4.7: Example of a relatively small facet size (13) and large step (10)
Figure 4.8: Example of a relatively large facet size (25) and small step (5)
4.2.7 Stage 6
Having settled on facet size and step values 25 and 5 respectively the next stage in
the process was to define a stage using the images that were collected. Due to the
order in which the images were taken, with the strung racket (stressed state)
captured first it was necessary to define two sets of two stages and then delete the
first and last stages. This left the image of the unstrung racket as the first stage and
76
the image of the strung racket as the second stage, which allows the system to
calculate the strain due to the racket being strung and display it as a contour field on
the image of the strung racket.
4.2.8 Stage 7
Having defined the series of images which will form the basis of the analysis, the
area of the images to be analysed was specified. This was achieved using the
software’s “mask” function. Using this software the entire image was masked and
only the area of the image containing the frame and around 5 millimetres either side
was unmasked. The resulting image can be seen below in Figure 4.9.
77
Figure 4.9: Image of the racket with excess area masked
4.2.9 Stage 8
Once the two stages had been created from the image series, the computation of the
project was initiated. After the software has completed the computation it
automatically opens a post-processor window displaying the strain contours. The
78
resulting strain contours for the unstrung and fully strung racket can be seen below in
Figure 4.10 (a) and Figure 4.10 (b).
The same racket was restrung and tested five times, both for the fully strung and
mains only condition. There was no measured difference between either the
magnitude of the strain or the position of the peak values.
4.2.10 Results/Discussion
4.2.10.1 Fully strung racket
(a) (b)
Figure 4.10: Strain contour of the (a) unstrung and (b) fully strung racket
Figure 4.10 shows the strain contour of the fully strung racket, where Figure 4.10 (a)
is the reference unstrung state and Figure 4.10 (b) is the loaded strung state. The
79
highest strain concentration can be seen along the inner edge of the frame and also
along the outer edge of the corner points of the frame.
The corresponding model of the strung racket is shown in Figure 4.11. The image is
similar in terms of the strain levels and distribution to that of the experimental data.
As in the case of the experimental data, the highest strain is along the inside edge of
the racket and the maximum value was 3%. The finite element model did not,
however, display the same level of strain as the Aramis results along the outer edge
of top end the racket. This could be due to a number of factors but given the level of
accuracy achieved throughout the remainder of the racket the model can be
considered satisfactory.
80
Figure 4.11: Finite Element model of the strung racket
4.2.10.2 Racket strung with mains strings only
To further validate the deformation performance of the racket frame model, an image
of the racket, strung solely with vertical main strings was also captured, and the
resulting strain contour can be seen below in Figure 4.12.
81
Figure 4.12: Strain contour of the racket strung with main (vertical) strings only
To validate the deformation of the racket with only the main strings, a model was
created in which the axial connector elements used to model the loading of the
crosses were defined as rigid beam elements, hence removing the tension loading
from the cross strings. The contact interactions between the mains and crosses were
also removed to ensure that the mains strings could elongate freely – as if the cross
strings were not present. The resulting strain contour can be seen below in Figure
4.13.
82
Figure 4.13: Finite element model of the racket strung with mains only
83
It can be seen from Figure 4.10 (b), that not all of the frame surface area has been
captured during the analysis. One of the main reasons for this is the curved, and
relatively small, surface area of the racket and the angle at which the cameras view
the frame from. The set-up is depicted in Figure 4.14 and shows the field of view of
the two cameras. The curvature of the racket means that only a small area of the
racket’s cross-sections is visible to both cameras.
Figure 4.14: Aerial schematic of the cameras’ field of view
84
The colour contour around the left side of the racket in Figure 4.12 shows that, as
with the fully strung racket, a strain of around 3% was experienced. Along the top
and bottom of the racket, however, there is an increased level of strain compared to
the fully strung racket. Whereas, Figure 4.10 (b) displayed a maximum strain of 3%,
an increased level of 7% was observed in Figure 4.12, along the middle of the frame
on both the upper and lower edges.
The level of the maximum strain in the mains only case (Figure 4.12), at 7%, is not
reflected in the finite element model (Figure 4.13). The poor performance of the
mains only model, compared to the fully strung model, could be due to the higher
deformation the mains only model experiences. Whilst, in the case of the fully strung
racket some of the deformation in the direction of the mains’ load is neutralised by
the cross strings’ load, in the case of the mains only strung racket, the racket
experiences a much higher deformation, in the afore-mentioned direction. In the
secondary mains only analysis the model clearly struggles to capture this more
severe loading procedure. The full results of the strain levels at the inner, outer and
mid sections of the racket at its top, bottom and side (described in Figure 4.15) can
be seen in Table 4.1.
85
Section Fully Strung Fully Strung Mains Only Mains Only
Experimental Model Experimental Model
Top-Mid (1) 2% 2% 3% 3%
Top-Outer (2) 3% 3% 7% 5%
Top-Inner (3) 3% 3% - 3%
Bottom-Mid (4) 2% 2% 3% 5%
Bottom-Outer (5) - - 7% 5%
Bottom-Inner (6) 3% 3% 7% 5%
Side-Mid (7) 2% 2% 3% 3%
Side-Outer (8) - - - 2%
Side-Inner (9) 3% 3% - 2%
Table 4.1: Strain Values for the strung racket in FE model and experimental
analysis
86
2
1
3
8
7
9
6
4
5
Figure 4.15: Points from which the most significant racket frame strain
measurements were take
The positions of high strain are perhaps not where one would intuitively expect them
to be found. As the racket is strung, assuming equal deformation in the lateral and
longitudinal directions, the racket would experience the highest levels of strain
around the “corners”. However, due to the non-uniform shape of the racket the
deformation is not equal in both directions, which leads to areas of high strains which
are more difficult to capture. The high levels of strain experienced along the inside
edge of the frame can be explained by two factors. The first is that there are a
greater number of cross strings therefore, they will exert a greater force and, as a
result, deformation on the racket during stringing. The second is that the inner edge
of the frame is smaller than the outer edge and as a result, a greater strain will occur
for the same deformation.
87
Another possible explanation for discrepancies between the experimental and
computational results could be the fact that the experimental method is only capable
of capturing the racket deformation in two dimensions. Therefore, any changes in the
depth of the racket, as shown in Figure 4.16 will be neglected.
y
x
It is not able
The photogrammetry
to capture
system is able to
deformation
capture the deformation
in the Z
of the rackets cross
Figure 4.16: Deformation of the tubular racket frame’s cross section
88
4.3 Validation of the string-bed stiffness profile
4.3.1 Experimental measurement
4.3.1.1 Instron 3365 force measurement machine
The stiffness profile of the string-bed was acquired using an Instron 3365 force
measurement machine which is part of the Loughborough University Sports
Technology Institute equipment pool. The machine was fitted with a circular stainless
steel applicator which was ten millimetres in diameter.
According to the manufacturer, the machine is accurate to within plus or minus

0.25% of the indicated force and plus or minus 0.5% of the indicated displacement
(Instron, 2012). The machine had also been calibrated within the two year period
which the manufacturer deems necessary (Farrand, 2012)
4.3.1.2 Methods
Using the “Instron” force measurement machine, the stiffness profile of the string-bed
was acquired. An adjustable jig, similar to that which may be used in a 3 point bend
test, was constructed to support the racket during testing. A diagram of the fixture is
shown in Figure 4.17.
89
Figure 4.17: Diagram of the supported racket
The jig was bolted on to the base of the Instron machine with the racket positioned
below the compression tool, depending on the area of the string-bed which was
being tested. The steel beam used in section 4.1 was removed and replaced with a
small circular applicator, which was eight millimetres in diameter. The applicator size
was chosen as it was small enough to focus on specific string intersections without
creating stress concentrations which may lead to string failures.
Using Instron’s own software, “Bluehill”, a compression test procedure was defined.
During the procedure, a 150 Newtons load, F, was applied at a rate of 5 Kilonewtons
per minute. This procedure was followed for the intersection of each main and cross
string, resulting in a total of 358 tests. From the test, the Bluehill software produced a
plot of the compression force against displacement.
The maximum value of compressed displacement, d0-150, was taken from the plot
and, along with Equation 4.1, used to calculate the stiffness, K0-150, of that particular
string intersection.
𝐾0−150 = 𝐹/𝑑0−150 Equation 4.1
90
Due to the number of tests to be carried out, there was not sufficient time to perform
multiple tests at each string intersection. As a result the statistical accuracy of the
test was determined by performing five tests for every tenth string intersection. This
approach resulted in the multiple test locations being distributed as shown in Figure
4.18. The results of the multiple tests can be seen in Table 4.2, with the Standard
Deviation less than plus or minus 3% in all cases.
91
320
310
300
280
290
270
260
250
230 240
210 220
200
190
180
170
150 160
130 140
110 120
100
90
80
60 70
40 50
20 30
10
Figure 4.18: Illustration of which intersection were tested multiple times
92
Standard Deviation of K (%)
178 N String 223 N String 267 N String
Intersection
Tension Tension Tension
10 2.46 0 2.05
20 2.66 2.46 2.25
30 2.66 2.46 2.25
40 0 1.48 2.66
50 1.54 0 1.02
60 2.00 1.48 1.38
70 1.55 0 1.18
80 1.71 0 1.38
90 0 0 1.80
100 0 0 0
110 0 1.39 2.75
120 1.74 1.39 2.75
130 1.71 1.48 2.82
140 1.80 1.69 0
150 1.71 1.48 1.15
160 2.36 1.02 0.90
170 2.72 0 1.55
180 2.25 1.79 1.50
190 2.24 2.66 0
200 2.23 2.66 0
210 0 1.36 2.94
220 2.16 1.64 0
230 2.25 0 1.24
240 1.85 2.09 1.85
250 0 0 0
260 2.16 1.64 1.63
270 2.16 1.59 2.59
280 2.01 2.94 1.93
290 1.75 2.16 2.44
300 1.74 1.18 1.93
310 1.71 2.24 2.77
320 0 1.23 2.88
Table 4.2: Standard deviation of intersections subjected to multiple tests
As can be seen in Table 4.2, the sting bed stiffness profile was obtained for string
tensions of 178 Newtons, 223 Newtons and 267 Newtons. This was done in order to
observe the difference in stiffness profile for varying string tensions.
93
For each string tension the strings were left to elongate for 24 hours before being
tested. Once the data had been obtained it was processed in Microsoft Excel and an
“Area” plot of the stiffness for each string-bed was generated.
A number of formatting strategies were explored in order to achieve the best

visualisation of the stiffness contour. The chosen method used increments of 20
kilonewtons per metre along the “depth” axis, as this gave the clearest image of how
the different string tensions varied. The resulting plots for the three string-beds can
be seen below in Figures 4.19, 4.20 and 4.21 respectively.
20-40 40-60 60-80 kN/m
Cross
Strings
Mains lower end

of racket
Strings
Figure 4.19: Stiffness profile of a Racket strung to 178 Newtons tension
94
20-40 40-60 60-80 80-100 kN/m
Cross
Strings
Mains Strings
Figure 4.20: Stiffness profile of a racket strung to 223 Newtons tension
20-40 40-60 60-80 80-100 100-120 kN/m
Cross
Strings
Mains Strings
Figure 4.21: Stiffness profile of a racket strung to 267 Newtons tension
95
4.3.2 Finite element model
To validate the stiffness profile of the racket generated in Chapter 3, a simulation was
created similar to the experimental analysis. In order to carry out the analysis on the
model of the tensioned string-bed (20 crosses and 18 mains) without having to rerun
the analysis, the restart analysis function was used.
The restart function effectively begins the analysis from the end-point of an existing
output database (.odb) file of a previous job, in this case that of the tensioning
analysis. To perform a restart analysis a new step, containing the additional
conditions, must be added to the model. New loads, boundary conditions,
interactions etc. can all be added to the model in this new step but the previous
steps’ attributes must remain constant
To replicate the experimental conditions, a 150 Newtons load was created which was
defined as a concentrated force of 10 Newtons applied to the 15 nodes which
surround the intersection of the strings (see Figure 4.22).
Figure 4.22: Load at string intersection
96
When this function was initially used, an unexpected error message, which
referenced “ErrNodeDomainDecompReStrt” was encountered. The error message can
be viewed in Appendix 3.
Reviewing the set “ErrNodeDomainDecompRestrt” showed that the nodes in

question were those being subjected to the connector lock condition. The domain
decomposition, which the error message refers too, refers to the method used to
break down the model when multiple processors are used to run the simulation. After
consulting the Abaqus support service it was discovered that the error was not a
problem with the model, rather it was a bug in the package. The problem occurs due
to the software’s inability to distinguish the connector lock from boundary conditions.
During the pre-processing stage of the restart analysis the software identifies the
connector locks as boundary conditions and assumes that the user has violated the
procedure by changing the boundary conditions of the previous analysis.
As a result, all jobs intended to be used as part of a restart analysis, and indeed the
restart analyses themselves, had to be performed using a single processor, greatly
increasing the computation time for all jobs. For the restart analyses themselves,
however, there were numerous jobs to be submitted, meaning that several jobs could
be run on a single processor simultaneously. Thus, there was little disruption
imposed on the computational efficiency of the restart analyses through having to run
each job on a single processor.
As for the experimental analysis, each intersection was subjected to a force of 150
Newtons (10 Newtons at 15 nodes in the case of the FE model). To obtain the
deflection at the node a history output request was created for the displacement in
the direction of the force (U3) at the centre node of the intersection. The stiffness
profile of the string-bed strung to 178 Newtons can be seen below in Figure 4.23.
97
10-13 13-16 16-19 19-22
Stiffness (kN/m)
Crosses
Mains handle end of

racket
Figure 4.23: Stiffness profile of the finite element model string bed
4.4 Discussion
The first observation to be drawn from each of the stiffness profiles is that they are all
unique both in shape and magnitude. The profile of the FE model should ideally be
similar to that of the experimentally obtained profile of the racket strung to 178
Newtons. It is clear, however, that it is of a reduced stiffness and also a different
profile. In the case of the FE model, the string stiffness decreases at the opposite
end of the racket from the handle which is not the case for the experimental results.
This could be a result of the short step time in which the loading of the string
simulating tensioning is performed (around 10 milliseconds for the FE model
compared with 6 milliseconds for the experimental analysis).
98
It is, however, interesting to note the continuity with which the experimental profiles
change as string tension is increased. As the tension increases, so too, as one would
expect, does the stiffness. The interesting feature is, however, the way in which the
stiffness if distributed around the string bed. For increasing string tensions the profile
becomes much steeper approaching the racket frame (this is illustrated by the
increasing variety of colours around the edge of the profile and the larger area they
occupy). It could be assumed, therefore, that if the magnitude of stiffness within the
FE string-bed is simply not high enough then an increase in string tension would
have the effect of increasing the stiffness whilst also creating a steeper gradient
profile around the edge of the racket.
The values provided in Table 4.1 show that the magnitude and location of the
significant strain experienced by the racket due to stringing are of a good likeness.
For the fully strung racket the strain magnitudes all correlate to the nearest
percentage point. For the more extreme deformation associated with the mains only
loading the model did not predict the magnitude as accurately (5% was the maximum
for the model compared to a maximum experimental value of 7%) but it did still
predict the location at which the highest strains occur.
A more complete image of the racket’s strain contour would have been desirable but
due to the nature of the target surface this was difficult to obtain. Had more time been
available it may have been possible to gain a more complete contour by performing
several analyses on the racket which focused in on smaller areas of the frame, rather
than trying to capture the entire racket. These individual contours could then be
“stitched” together to form a more comprehensive contour than was possible with a
single experiment.
99
Although the static deformation profiles of the string-bed and racket were not entirely
similar to the experimental, within the central area of the bed the values showed a
level of magnitude similar to the experimental values. Both the experimental and
finite element model results gave a stiffness in the range of 20 plus or minus 5
kilonewtons per metre within the most central area of the racket. Since this study is
predominantly focusing on the most commonly used hitting area, the fact that the
model did not capture the higher stiffness around the edge of the string-bed is not of
great concern. However, if future studies were to look at the performance of rackets
with respect to off-centre impacts, a more sophisticated string model would be
required.
This testing has demonstrated that the behaviour of the FE model of the racket frame
and string-bed are similar to that of a racket under experimental loading. The
experimental testing shows that the FE model predicts the areas of maximum
deflection during loading, which allows confidence to then apply the model to more
dynamic conditions.
100
5 Modal analysis
5.1 Introduction
The vibration characteristics of a tennis racket are of particular importance since they
play a significant role in the sensations experienced by the player during competition.
When a tennis player strikes the ball they will generally aim to do so at the “sweet-
spot” of the racket. There are several definitions of a racket’s “sweet-spot”, the most
scientific definition of which is the area which corresponds to the node of the racket’s
first natural frequency (Brody, 1981). This node is generally located at the centroid of
the string-bed, (Mohanty et al., 2002), which is also the area that offers consistent
outbound flight trajectories. As a result this area will offer the most accurate shot
(since the rebound angles outside of this area vary too widely to be reasonably
controlled by the player) and hence the level of vibration produced in this instance
has positive connotations for the player. Simply stated, rackets which vibrate less
give the player the impression they are playing better.
As well as the connotations of a more effective shot, the lack of vibrations emanating
from the fundamental node also have a more evident advantage, that of comfort. Not
only can these vibrations be uncomfortable, they are also perceived to cause serious
injury in the form of lateral epicondylitis, more commonly known as “tennis elbow”
(Roetert et al., 1995, Pluim, 2000, Jobe et al., 1994).
In addition to providing a guide to the performance of the tennis racket, modal

analysis can also be used as a means for validation of finite element (FE) models.
Modal analysis has been used by various authors to help relate the characteristics of
sports equipment to quantifiable parameters. Brody (1995), Cross (2001b) and
Mohanty et al. (2002) have all used modal analysis as a means of explaining such
characteristics in tennis rackets. In the case of Brody and Mohanty it was the so
called “sweet-spot”, whilst Cross used modal analysis to examine the change in
101
racket stiffness due to stringing. Similar analysis has also been carried out on cricket
bats (Brooks et al. (2006), baseball bats (Noble et al., 1994), golf clubs (Merkel et al.
(1999), Thomas et al. (1995)) and even on soccer and golf balls (Ronkainen et al.
(2007) and Axe et al. (2002), respectively).
Despite modal analysis being used as a means of validating finite element models of
sports equipment, little has been done to understand how vibrational characteristics
vary between different models of rackets and indeed, between rackets of the same
model. The aim of the research work presented here is to offer an insight into how
the vibrational characteristics of aluminium rackets of the same model vary between
two rackets and how the same frame will vary when strung at varying string tension.
Future work may also provide information the variation between different models of
racket.
5.2 Methods
Instead of the more traditional methods used in modal analysis this study has chosen
to use a technique which has been used rarely within the sports industry. With the
exception of Ronkainen et al. (2007), most previous studies have been carried out
using more traditional experimental equipment such as accelerometers and
mechanical exciters. In the work presented here a 3D laser doppler vibrometer (LDV)
is used to provide the signal to drive the excitation of the racket and also obtain the
measurements of the system’s vibration.
The major difference between laser doppler vibrometry and more conventional modal
analysis techniques is that, in conventional modal analysis the amplitude of
displacement and frequency is determined simply by the direct measurement of the
subject’s acceleration from the accelerometer at a specific point. An LDV, however,
will extract the measurements from the Doppler shift of the laser beam frequency
when the laser in question is reflected from the subject. As a data acquisition tool the
102
obvious benefits of LDVs are that, unlike accelerometers, they do not have to be
attached and unattached for each measurement point. Moreover, the more advanced
LDVs are able to scan multiple points rapidly, significantly reducing data acquisition
time.
Another significant advantage of using LDVs is the elimination of mass-loading the

structure during analysis. Attaching accelerometers to the structure can have the
effect of changing the frequency response function (FRF) and continually moving the
accelerometers around the structure will mean that one may not receive the same
structural response for each point. The FRF of a system is represented in Figure 5.1
and can be defined as:
“the ratio of the Fourier transform of an output response (X(ω)) divided by the Fourier
transform of the input force (F(ω)) that caused the output” (Schwarz et al., 1999)
Mechanical
system
𝐹(𝜔) × [𝐻(𝜔)] = 𝑋(𝜔)
Figure 5.1: Transformation of input force to output response for a mechanical

system
One drawback of more basic LDVs is that they only consist of a single scanning head
(i.e. one laser). As a result, they are unable to take measurements in three
dimensions. However, the development of LDVs has led to the introduction of 3
dimensional systems which consist of three scanning heads. As a result the system
is able to acquire the motion of the subject in all 3 axes and can directly acquire the
103
mode shapes of the subject and display them without the need for any intermediate
software.
The system used for this series of experiments was the Polytec PSV-400-3D
Scanning Vibrometer. The system’s performance specifications can be seen below in
Table 5.1 (taken from the manufacturer’s web-site):
Bandwidth 0 kHz – 80 kHz
Velocity range 0.01 µm/s ... 10 m/s
Working distance > 0.4 m
Laser wavelength 633 nm (red)
Laser safety class Class II He-Ne laser, 1 mW per sensor,

eye-safe
Scanner resolution ± 0.002°
Sample size Several mm² up to m² range
Table 5.1: Performance specification of the Polytec PSV-400-3D Scanning

Vibrometer (http://www.polytec-ltd.co.uk/uk/products/vibration-
sensors/scanning-vibrometers/psv-400-3d-scanning-vibrometer)
The generic process for performing a natural frequency extraction using the Polytec
system described above is detailed below and supported by the process flow in
Figure 5.2.
104
5.2.1 Stage 1
The first stage in the test process was to position the three laser heads so that they
are facing the target which is to be measured. Ideally, the set-up should be similar to
that in Figure 5.3, with the central scanning head lower to the ground than the other
heads. Once the scanning heads are in position, the user must then choose a series
of calibration points on the target. The user must then manually position each of the
three laser heads on one of the calibration points before “locking” the point into the
system. Once each of the three heads has been “locked” on a calibration point the
user must then repeat this process for at least two more points.
5.2.2 Stage 2
The second stage in the process is to specify a number of sample points around the
target. The more sample points that are specified, the longer the analysis will take
but the more detail will be provided in the depiction of the mode shapes. To specify
the test points, the user must manually position the laser heads (after calibration the
laser heads will all move in synchronicity so the user need only move the central
head) at each of the desired sample points.
5.2.3 Stage 3
Before extracting the mode shapes for the natural frequencies, the user must first
perform a “white noise” excitation of the system, which is a random excitation, to
identify where along the frequency spectrum the natural frequencies occur. During
this analysis the system will take measurements at each of the specified sample
points and use them to produce a Frequency Response Function (FRF) plot.
105
5.2.4 Stage 4
Having performed the “white noise” excitation, the user must then identify the natural
frequency peaks on the FRF plot. Once the user has identified the natural
frequencies of the desired modes, they must then perform and excitation of the
system at this frequency. During this excitation, the system will again measure the
response at each of the sample points, from which the mode shapes of the natural
frequencies will be generated.
106
Stage 1
Calibrate the system for the
target
Stage 2
Specify the sample points.
Stage 3
Perform a white noise
analysis.
Stage 4
Perform an excitation at the
desired natural frequency to
obtain the mode shape.
Figure 5.2: Process flow for modal analysis using a Polytec 3D LDV
107
5.2.5 Racket Testing
Two tubular aluminium unstrung tennis racket frames (racket frame 1 and 2) of the
same model were analysed in the first instance. Aluminium rackets were chosen due
to the relative ease with which they can be modelled in finite element analysis. The
unstrung analysis was performed on two rackets to identify the variability between
rackets.
Then, in order to assess the effect of string tension on the vibrational characteristics
of the racket, the same racket (racket frame 1) was strung to tensions of 182
Newtons, 223 Newtons and 267 Newtons and analysed. In order to assess the
experimental error each analysis was performed 5 times and the average value
obtained. Given the system is capable of measuring frequencies up to 80 kilohertz
(Table 5.1) it was assumed that any significant error would be identified by the
repeatability testing. Having identified the variability between racket frames in the
unstrung state it was not deemed necessary to analyse each racket in the strung
states, therefore only frame 1 was tested in the strung states. The number of tests
performed is shown in Table 5.2.
Tested Tested TestedTested Number of

Unstrung Strung to Strung to
Strung to 267 tests
Newtons 5
5 times 178 Newtons 223 Newtons
times
5 times 5 times
Racket
Frame 1
    20
Racket
Frame 2
 - - - 5
Table 5.2: Rackets tested
108
During the analysis, an aluminium racket was suspended from a horizontal beam
using string and elastic bands. The elastic bands were attached to the upper edge of
the head and handle at either end of the racket and the string was passed through
the bands to support the racket from above (see Figure 5.4). The inclusion of the
elastic bands minimises any reaction forces created by supporting the racket in this
manner, thus providing a more realistic representation of the freely suspended
condition. This is particularly beneficial as it not only removes the need to model
additional boundary constraints but also provides a better representation of the
racket in the hand held condition (Brody, 1995).
Figure 5.3: Schematic of the LDV experimental set-up
109
Figure 5.4: System used to constrain the racket
The LDV control unit not only controls the data acquisition but also the excitation of
the racket. Initially, a loudspeaker was used to excite the racket and stringbed,
similar to the method used by Ronkainen et al. (2007) but this did not provide an
adequate level of force to excite the natural frequencies. As a result a Bruel and
Kjaer mechanical exciter (model number 4824) with a force rating of 100 Newtons
was chosen and was connected to the control unit via a power amplifier. The exciter
drove an axial stinger which was fixed to the end of the rackets (slightly above centre
to avoid exciting the racket on the torsional node) using a general purpose sealant.
The axial stinger simply consisted of a metal rod around 0.3 millimetres in diameter
and 100 millimetres in length which could be fixed to the exciter at one end and the
racket at the other. The exciter was hung from a gantry using string in order to
reduce any reaction forces from the exciter.
For the unstrung racket, 96 equally spaced data collection points were assigned to
the elliptical frame section. In the case of the strung racket, the strings and the frame
were analysed individually. The data points for the frame were positioned as in the
110
case of the unstrung racket whilst the data points for the string-bed were assigned to
the intersections between the cross and main strings, resulting in the analysis area
shown in Figure 5.5 and a total of 360 points.
Figure 5.5: The area of the string-bed analysed
The natural frequencies of the system were obtained from the frequency response
function (FRF) plot which, was obtained from a white noise excitation (frequency
range 0 Hz - 1,200 Hz).
111
Once the spectrum of the FRF had been obtained the positions of the peaks
representing the natural frequencies were noted and the rackets were excited and
scanned at these frequencies to obtain the mode shapes. The resolution of the
resulting spectrum was primarily dependent on two factors; the sample time (which
itself is dependent on the bandwidth of the analysis and the number of Fast Fourier
Transform lines used) and the sample frequency. The relationship between these
factors is shown below in Equation 5.1 and Equation 5.2.
FFT _ Lines
sample _ time = Equation 5.1
Bandwidth
Sample _ time
resolution = Equation 5.2
Sample _ Frequncy
For the majority of experiments the acquisition variables were as follows:
FFT lines = 1600
Sample frequency = 2.56 Hz
Bandwidth = 1000 Hz
Sample time = 1.6 s
These values were found to give a suitable definition for the spectrum without greatly
increasing the sampling time.
Once the rackets had been re-scanned at the natural frequencies the LDV software
created an animation of their mode shapes, the validity of which is discussed later.
112
5.3 Finite element model
An FE model of the aluminium racket was created by running a “restart” analysis of

the model described in Chapter 3. In Abaqus, a “restart” analysis takes the final state
of a previous analysis and starts a new analysis from that point.
In this case a linear pertubation frequency analysis step was inititaed from the point
at which the string tenisioning process had been completed. Abaqus offers several
forms of frequency analysis, the default of which is “Lanczos” frequency extraction as
it is the quickest to run and, in most cases, the most accurate.
As a result a “Lanczos” frequency extraction step was specified for frequencies

ranging from 0 Hertz to 1,200 Hertz. This frequency range was chosen since it was
shown from the experimental work that all natural frequencies fell comfortably within
1,200 Hertz, Brody (1995).
113
5.4 Results
The experimental results of the unstrung aluminium racket frames 1 and 2 are shown
in Table 5.3, where rackets 1 and 2 are the same model.
Experimental
Lateral Mode Torsional Mode

Racket-Frame (Hz) (Hz)
1 130 +/- 0.4 353 +/- 1.1
2 132 +/- 0.5 352 +/- 0.7
Table 5.3: Natural frequencies of unstrung rackets (experimental)
The natural frequencies for the frame and stringbed for a strung racket (racket 1) are
shown for both the experimental case (Table 5.4 and Table 5.5) and the natural
frequencies of the finite element model (Table 5.6 and Table 5.7).
Experimental
Tension Mode (Hz)
Lateral Torsional
182 N 137 +/- 1.8 369 +/- 2.7
223 N 140 +/- 1.9 367 +/- 1.9
267 N 136 +/- 2.1 366 +/- 1.6
Table 5.4: Average natural frequencies of Racket 1 strung at varying tensions

(experimental)
114
Experimental
Tension Mode (Hz)
1st String Mode 2nd String Mode
178 N 576 +/- 2.6 853 +/- 2.9
223 N 574 +/- 1.8 856 +/- 1.4
267 N 576 +/- 1.3 856 +/- 1.9
Table 5.5: Average natural frequencies of the string-beds of Racket 1 when

strung at varying tensions (experimental)
FE Model
1st Lat 1st Tors
131 Hz 357 Hz
Table 5.6: Natural frequencies of finite element model of an unstrung

aluminium tennis racket
FE Model
1st String 2nd String
560 Hz 768 Hz
Table 5.7: Natural frequencies of a finite element model of a string-bed

tensioned to 178 Newtons
115
(a) 1st Lateral mode of the finite element model (132 Hz)
(b) 1st Lateral mode of the experimental analysis (Racket 1, 130 Hz)
Figure 5.6: Lateral Modes (Unstrung Racket)
116
(a)1st torsional mode of the finite element model (353 Hz)
(b)1st torsional mode of the experimental analysis (357 Hz)
Figure 5.7: Torsional Modes
The lateral mode shapes for the FE model and experimental analyses are displayed
in Figure 5.6 (a) and Figure 5.6 (b) respectively. The figures show similar mode
117
shapes have been achieved for both the model and experimental data. Figure 5.6 (b)
does, however, display a minor discrepency at the tip of the racket in the form of a
slight “kink”. This is also the case for the experimental torsional mode in shown in
Figure 5.7 (b), but again it is otherwise consistent with the finite element model mode
shape shown in Figure 5.7 (a).
118
(a) 1st String mode of the experimental analysis (576 Hz) superimposed onto the
image of the racket
(b) 1st String mode of the finite element mode (560 Hz)
Figure 5.8: 1st string modes
119
(a) 2nd String mode of the experimental analysis (853 Hz)
(b) 2nd string mode of the finite element model (768 Hz)
Figure 5.9: 2nd String Modes
120
Figure 5.8 (a) and Figure 5.9 (a) show the string-bed mode shapes associated with
the natural frequencies of the aluminium rackets obtained via the LDV, whilst Figure
5.8 (b) and Figure 5.9 (b) show their finite element counterparts.
121
5.5 Discussion
5.5.1 Unstrung racket frames
The natural frequencies of the FE model of the aluminium racket frame, displayed in
Table 5.6 showed good correlation with the experimental results shown in Table 5.3,
with the fundamental lateral value of 131 Hertz falling exactly between the
experimental values and the torsional value yielding a 1.5% difference.
The discrepancy at the tip of the racket shown in Figure 5.6 (b) was a common
problem, caused by the calibration of the laser positioning. Each time a new racket
was tested the lasers had to be repositioned at the centre of the racket and each one
calibrated at a series of fixed points around the racket, as described in section 5.2.1.
Naturally the laser positioning at the centre of the racket was easily refined, however,
the positioning of the lasers on measurement points at the extreme ends of the
racket was difficult. It is for this reason that only the head and sections which join the
handle to the racket were included in the LDV scans.
A similar discrepancy to that witnessed in Figure 5.6 (b) can be seen for the torsional
mode in Figure 5.7 (b), where the expected smoothness at the curved tip of the
racket is not apparent. The correlation between the FE and experimental modes is,
however, still very good and further validates the FE Model.
5.5.2 Strings
It is widely accepted (Brody 1995) that the frequency range for a racket-ball impact in
tennis is within the range of 100 Hertz to 200 Hertz. However, the results of the
modal analysis of the tennis strings shows that the value for the fundamental mode is
576 Hertz (see Table 5.5), which, indicates that the role of the natural frequency of
the strings may have little effect on the experience of the player during competition.
122
This can be concluded from the contact time of a typical ball-stringbed interaction,
which lasts around 5 milliseconds Brody (1987) and, hence, results in excitations at a
frequency of around 100 Hertz. In order for the fundamental frequency of the string-
bed to be excited an impact of around 2 milliseconds would be required. The 1st
mode shape, shown in Figure 5.8 is similar to that of a trampoline and is the shape
one would associate with with a clamped circular plate, as shown by Reddy (2007) in
Figure 5.10. Like the first mode, the experimental results of the second mode are
concurrent with the mode shapes obtained from the FE model.
Figure 5.10: 1st Mode shape of a clamped circular plate
5.5.3 Strung racket frames
The analysis of the strung rackets also revealed higher natural frequencies than
those obtained when the frames were scanned unstrung. This was concurrent with
the results of the three point bend tests obtained in section 4.1, where the stiffness of
the strung racket were greater than the unstrung rackets. The ratio of the natural
frequencies of the strung and unstrung rackets can normally be obtained via the
results of a three point bend test using the relationship derived below, where ωn is
the fundamental mode, k is the stiffness and m is the mass (Reddy, 2007).
𝑘
𝜔𝑛2 = 𝑚 Equation 5.3
123
𝑘𝑠𝑡𝑟𝑢𝑛𝑔�
2 𝑚
𝜔𝑛 𝑠𝑡𝑟𝑢𝑛𝑔 ⎛ ⎞
� � =⎜
𝜔𝑛 𝑢𝑛𝑠𝑡𝑟𝑢𝑛𝑔 𝑘𝑢𝑛𝑠𝑡𝑟𝑢𝑛𝑔� ⎟
𝑚
⎝ ⎠
2
𝜔𝑛 𝑠𝑡𝑟𝑢𝑛𝑔 𝑘𝑠𝑡𝑟𝑢𝑛𝑔
� � =� �
𝜔𝑛 𝑢𝑛𝑠𝑡𝑟𝑢𝑛𝑔 𝑘𝑢𝑛𝑠𝑡𝑟𝑢𝑛𝑔
This relationship, of course, assumes that the mass of the strings is negligible. For
the case of the experimental results obtained from this study, this relationship
becomes:
138 2 129
� � ≈� �
131 116.7
1.11≈1.1
124
The mode shapes for the 2 strung aluminium rackets tested were similar to one
another and what one would expect from a structure which can be loosely
approximated as a beam. The mode shapes also showed good similarity to that of
the FE model.
Table 5.4 shows the results of analysis carried out on racket frame 1 strung to
several different string tensions. For each analysis the racket was strung the previous
evening and left to “settle” overnight. The environmental conditions, such as ambient
temperature, humidity etc. were not measured during this analysis so it cannot be
stated whether slight changes in the conditions had any effect on the results. It can
be stated, however, that the analyses, which are discussed in this section were all
performed within a five day window and that any changes in room conditions would
have been relatively small.
The results of this procedure show little difference in the results and do not display
any particular trend. It is more reasonable to assume that the 2 Hertz swing in the
results is due to uncontrollable changes in measurement conditions. It can be
concluded, therefore, that the fundamental frequency is 138 Hertz across all the
tension range used by most players. It is more than likely that the increase in racket
frequency that can be seen between the strung and unstrung rackets occurs
significantly over the lower tension ranges and the the increase in frequency across
the tension range of 178 Newtons to 267 Newtons is relatively negligible, as
indicated in Figure 5.11.
125
140
138
Fundamental Frequency (Hertz)

136
134
132
130
128
126
0 44 89 133 178 223 267
String Tension (Newtons)
Figure 5.11: Racket frequencies for different string tensions
As a result, it can be concluded that changing the string tension, in the range which
most players will string their rackets, has no significant impact on the fundamental
natural frequency of the racket. The change in frequency due to stringing the racket
could be explained by a combination of several factors. The first factor is the
increased stiffness of the racket due to stringing. As stated in Equation 5.3, the
natural frequency, ωn of a system is proportional to the square root of the stiffness, k.
The second factor is the deformation of the racket caused by stringing. When rackets
are strung they tend to deform more under the load of the cross strings than the
mains strings which leads to an increase in the rackets length and reduction in its
width. The bending stiffness, kbend, of the frame is the product of I, the second
moment of area and E, the Young's modulus (Equation 5.4 (Reddy, 2007).
kbend = EI Equation 5.4
126
If we negelect the handle of the racket, which is not subject to deformation during
stringing and can be assumed to remain constant, the unstrung head of the racket
can be treated as a square cross section. Once strung, the racket will become more
oval-like in shape or, adopting a more basic assumption, like a rectanglular cross
secion as shown in Figure 5.12:
Figure 5.12: Second moment of area dimensions
For a rectangular cross section the Second moment of area is represented by

Equation 5.5, Reddy (2007):
bo • ho3 − bi • hi3
I=
12 Equation 5.5
For this system, as b0 and bi decrease and h0 and hi increase, assuming a constant
thickness, I will increase. Hence, if we assume that the tennis racket will behave in a
similar manner then it can also be assumed that the bending stiffness, and as a
result ωn, will increase.
127
A 3D laser doppler vibrometer (LDV) has been used to obtain the natural frequencies
of aluminium tennis rackets, both in their strung and unstrung states. The first two
natural frequencies and modes shapes were obtained for the racket. The values for
the fundamental frequency of the racket were similar to those obtained by Brody
(1995) (between 100 Hertz and 150 Hertz). A comparison could not be drawn on the
values for the string modes, since the extremely small mass of a stringbed had
meant that convenional modal analysis – adding an accelerometer – would have
rendered results meaningless. This non-contact measurement of this study, however,
did not encounter this problem.
The mode shapes observed experimentally are consistent with what would be
expected from basic consideration of vibration theory, with the racket being
represented as a free beam whilst the string bed behaved as a clamped circular
plate.
The experimental variation of the results obtained for the racket frame using the LDV
was less than 3% in all cases. Such a low level of variation gave good confidence in
the results and allowed for their use as a correlation method for the FE models.
Although an accuracy of 3% is sufficient for the validation of the FE model, it was not
accurate enough to identify any variation in the natural frequency of the rackets or
the strings due to variation in string tension. Although it is disappointing not to identify
a relationship between string tension and natural frequency, it is not the main priority
for the use of modal analysis in this study.
The experimental values for the natural frequencies of the aluminium tennis racket
were used to validate the finite element model created for the same racket. The
model showed good similarity with the experimental results (less than three percent
128
variation between the experimental and FE results for both modes), both in terms of
the values of the natural frequencies and their corresponding mode shapes.
The FE model did not perform as well at the higher values of natural frequency found
for the string-bed. The two results were within 3% of one another for the first string
mode but the FE model produced a second string mode with a difference of 10%
when compared to the experimental results. Having said that, it is very ambitious to
expect the FE model to remain accurate at such a high frequency which, furthermore,
is well beyond what is likely to ever be experienced by a tennis player.
The validation of the mode shapes was somewhat subjective and was performed
purely by visual inspection. More robust methods of comparison, such as use of a
Modal Assurance Criteria, are possible through commercially available softwares
(LMS Virtual Lab, 2012). If a higher level of accuracy was required then this would be
an appropriate approach but for the purposes of this study it would be unnecessarily
costly. As a result, the use of the LDV to correlate the model can be considered to be
very successful and the model itself can be deemed fit for further development.
Having validated the dynamic characteristics of the racket, the next step in this
process is to validate its performance in contact with the ball during an impact.
Before this can take place however, it is necessary to obtain the materials propeties
of a number of strings in order that the models performance can be validated for
different string types.
129
6 Creation of a materials database
6.1 Young’s modulus for various tension ranges
Section 2.2 reviewed the laboratory work carried out by various authors with regards
to the material properties of tennis strings. Although a great deal of research has
been carried out into the resultant performance of different string materials when
strung in a racket, relatively little has been done to understand how the fundamental
properties, such as Young’s modulus, affect the performance.
One author, Cross (2000a, 2000b and 2003), has done more than any other to
examine the fundamental properties of tennis strings. In the specified research Cross
focussed mainly on the dynamic stiffness, kd; the stiffness of the string after it has
been tensioned. Cross's main findings were that, when pre-tensioned at 196
Newtons:
• The stiffness of all the tested strings (Natural gut, Nylon, Polyester, Kevlar)
increases.
• The force transmitted through the tennis string to its clamp increases.
The relationship between the Peak force experienced at the clamp and the
displacement of the strings relative to one another is summarised in Figure 6.1.
Cross found that Kevlar transmitted the highest force whilst experiencing the least
displacement whilst Gut absorbed more force while experiencing less displacement.
130
Kevlar
F Polyester
O
R Nylon
C
E Gut
DISPLACEMENT
Figure 6.1: General trend of peak force and displacement experienced by pre-
tensioned tennis strings when struck by a swinging hammer (Cross, 2000b)
Although the dynamic stiffness is interesting in terms of string performance it is not a

parameter which is compatible with the Abaqus materials models. To define a
materials model within Abaqus one requires some form of stress-strain data, the
simplest form of which is the Young’s modulus.
6.1.1 Equipment
The same Instron 3365 force measurement machine was used as described in
Section 4.3.1.1. In this instance the circular compressive applicator and racket
support were replaced with vice grips, as shown in Figure 6.2. The vice grips
contained a textured surface to ensure that slip did not occur during testing.
131
Figure 6.2: Vice grips used to hold the tennis strings in place during tensile
testing
6.1.2 Methods and Results
To obtain the Young’s modulus of a variety of tennis strings, a procedure similar to

that used in Section 3.5 was adopted. In this set of experiments, however, the
Young’s modulus was measured for a variety of string tension ranges rather than the
single range (100 Newtons - 150 Newtons) used in the initial experimentation. The
Young’s modulus of the strings, were measured at a strain rate of 100 millimetres per
minute in the following ranges:
• 178 Newtons - 228 Newtons

132
These tensions were chosen as the lower limits; 178 Newtons, 223 Newtons and 267
Newtons, represent the most commonly used string tensions. The upper limit was
simply an additional 50 Newtons, assumed to represent an impact force similar to
that of a tennis ball impact. The gauge of all the strings tested was 15, meaning a
cross sectional diameter of 1.3 millimetres. The cross sectional diameter of the
strings, post-tensioning was obtained from Racket Tech Magazine (Racket Tech
Publishing, 2005), the final diameter of the string was fed into the Bluehill software to
account for changes in cross sectional area and hence allow a more accurate
modulus to be calculated.
5000
Babolat Pro
Hurricane
4500
Prince
Tournament
Young's modulus (MPa)
Polyester
4000
Prince
Tournament
Nylon
3500 Luxilon Big
Banger
Prince
3000
Synthetigut
2500
181-231 227-277 271-321
Applied Force Range (N)
Figure 6.3: Young’s modulus of tennis string for different tension ranges
Figure 6.3 shows the results of the testing and some interesting trends across the
tension ranges. The first point which is apparent from Figure 6.3 is the variation in the
133
trend between the nylon (Prince Synthetigut, Prince Tournament Nylon) and
polyester (Babolat Pro Hurricane, Prince Tournament Poly Luxilon Big Banger)
strings across the tension range. The Young’s modulus is shown to increase with
tension for the nylon strings as opposed to the polyester strings, which decrease.
It is also apparent that the variation in Young’s modulus across the different tensions
is at least 10% for all strings, which is significant enough to have an effect on an
impact simulation. Therefore, when defining the material data for the finite element
model it would be prudent to adjust the Young’s modulus depending on which string
tension condition is being modelled.
6.2 Variation in Young’s modulus due to extension rate
As with any dynamic analysis, the strain rate of the material data can play a
significant role in the accuracy of the results. In a tennis impact the string-bed will
typically deflect by around 5 millimetres to 10 millimetres in around 5 milliseconds to
8 milliseconds (meaning the maximum deformation occurs at around 2.5 milliseconds
to 4 milliseconds). The length of the elongated string can be calculated using
Pythagoras, as in Equation 6.1, where l0 is the initial string length, l1 is the elongated
length and d is the string deformation.
𝑙 2
((0.5𝑙0 )2 + (𝑑)2 ) = � 1 � Equation 6.1
2
2
𝑙12
((0.5 × 0.2) + (0.01)2 ) =
4
𝑙12 = 4 × 5.05 × 10−3
𝑙1 = 0.201𝑚
This results in a strain of 0.001 and a strain rate of 0.4 metres per second (24,000
millimetres per minute). Recreating strain rates of this magnitude, whilst measuring
134
the string response, was beyond the capabilities of this study. Efforts were taken,
however, to examine the variation in Young’s modulus due to lower strain rates (25
millimetres per minute to 100 millimetres per minute).
6.2.1 Methods and Results
Three strain rates were examined, 25 millimetres per minute, 50 millimetres per
minute and 100 millimetres per minute, at the three previously mentioned tension
ranges for the nylon, titanium and polyester strings. The results of these tests can be
seen below in Figure 6.4, Figure 6.5 and Figure 6.6
178 Newtons - 228 Newtons

6000
5500
5000
4500 Nylon
Titanium
4000 Poly
3500
3000
25 mm/min 50 mm/min 100 mm/min
Strain rate
Figure 6.4: Young’s modulus for 178 Newtons - 228 Newtons at various strain
rates
135
5500
5000
4500
Nylon
Titanium
4000
Poly
3500
3000
Strain rate
rates
136
5500
5000
4500
Nylon
4000 Titanium
Poly
3500
3000
Strain Rate
rates
Reviewing the results of the strain rate experiments shows that the titanium polymer
string’s modulus tends to vary for the lower applied force of 178 Newtons to 228
Newtons but remains relatively constant for the higher forces. Conversely, the nylon
and polyester strings show little variation in modulus across strain rates for the lower
forces but increase with strain rate for the higher force.
Another point which is interesting to note is that whilst the titanium string decreases
in modulus as the strain rate is increased, the polyester and nylon strings increase in
modulus.
It is perhaps, no coincidence that the value of tension (223 Newtons) which the
strings’ modulus is most consistent in, with respect to strain rate, for all the strings is
also the most commonly used string tension. Noting the consistency of the results at
137
this value one can also assume that the value of the modulus is unlikely to be
changed at the higher strain rates achieved in an impact during play. With this in
mind the decision was taken to use 223 Newtons as the default tension when
analysing the change in output variables, such as spin and COR, due to changes in
material properties.
6.3 Coefficient of Restitution
The COR of a tennis racket impact is dependent upon a variety of factors:
• Energy loss through deformation of the ball and strings during impact
• Energy loss through oscillation of the ball post-impact
• Energy transformed into internal vibration of the frame
Hatze (1993) quantified the contribution of each of these factors through a

mathematical model. He stated that in a typical impact with a COR of 0.8, around
2%-4% of the energy would be lost through the strings during impact, 15% would be
lost due to the deformation of the ball whilst the majority of the energy loss (56%-
65%) would take place post-impact in the form of ball oscillation and racket vibration.
As a result, the critical role played by the string-bed in an impact, with respect to
COR, is not the amount of energy it absorbs, it is the speed with which they return to
their original position and the ball rebounds. Therefore, the string’s ability to deflect
and return to its initial position is the most significant factor when considering its
merits.
In order to quantify the string’s contribution to the COR of the impact system the most
significant variable is therefore, the contact time. In order to form a comparison of the
contribution made to the COR by different strings a procedure was put in place to
138
model the impact of a solid tennis ball. The reason for using a solid representation of
the ball was that this would remove the effect of the energy dissipation of the ball and
isolate the effect of the strings.
6.3.1 Methods
To obtain the desired experimental set-up an INSTRON 9250G Drop-test tower,

shown in Figure 6.7, was used. The "G" in the system's title stands for gravity as this
variation of drop-tester uses gravity to generate the impact force. This is achieved by
raising a weighted carriage to a specified height above the test sample and releasing
it. The 9250G is available with a light, medium and heavy carriage. The model used
in this instance contained the medium carriage which was capable of generating
kinetic energy in the range of 4.6 Joules to 300 Joules. At the time these experiments
were performed, the system was less than six months old and the initial calibration
certificate was still valid.
139
Weighted Carriage
Figure 6.7: Intron 9250G Drop-test Tower
The user can define the initial position of the carriage by specifying one of the three
following variables:
• Impact velocity
• Kinetic energy
• Drop height
This action is performed via a control unit (typically a PC) which is installed with
Instron's own interface software, "Impulse". According to the manufacturer the final
impact condition is accurate to within 2% of that specified by the user (Instron, 2011).
140
Since the weight of the carriage can be altered at any point, the system requires the
user to perform a weight calibration before each measurement. This is achieved by
moving the carriage to a position where it is not in contact with the measurement
specimen and selecting the "Weigh" option within the Impulse interface. Until this
action is performed, all measurement options will appear greyed out within the
interface.
Inside the drop-test tower, two light gates are positioned just above the string-bed to
measure the inbound and outbound velocity. The manufacturer states that the light
gates will give a velocity measurement which is accurate to within 0.25% (Instron,
2011).
Once the impact has been performed, the Impulse interface will produce a Force-
Time plot. An example of such a plot is shown in Figure 6.8. Furthermore, the
Impulse interface will also provide the inbound and outbound velocity, measured by
the light gates.
Figure 6.8: Example of the data output from the drop-test for a Polyester string-
bed
141
A half-hemisphere, 67 millimetres in diameter, was machined from aluminium to
represent the shape of a tennis ball, attached to the under-side of the carriage and
positioned above a custom made square string-bed, as can be seen in Figure 6.9.
Figure 6.9: Experimental set-up for string COR testing
The string-bed was fabricated by the technicians within the Loughborough University
Sports Technology Institute, specifically to fit within the drop test tower. It consisted
of ten millimetre diameter solid steel tubing, with holes strung at five millimetre
intervals along the length of the tubing. The frame was sixty millimetres in length and
breadth from the inner edges. The string-bed was constructed along with special
fixtures which allowed it to be strung on a Babolat racket stringing machine, as
142
Figure 6.10: Square string-bed mounted on the Babolat® stringing machine
In order to deflect the string-bed to a similar magnitude as in a ball-bed impact some

general calculations were made to find the energy required. If the kinetic energy, Ek,
is given by Equation 6.2:
1
𝐸𝑘 = 2 𝑚𝑏 𝑣𝑏2 Equation 6.2
1
𝐸𝑘 = ∗ 0.057 ∗ 302 = 25.65 Joules
2
Where, mb and vb are the mass and velocity of the ball, respectively (Matsuhisa,
2004). Initial trials of with this kinetic energy, however, resulted in consistent failure of
the string-bed. This was thought to be due to the absence of ball deformation and an
increased stress concentration in the string-bed (normally the ball would deform and
spread out during contact). Some ballistics experts (Dr. Ron Thomson, Department
of Mechanical Engineering, University of Glasgow) state that the different peak
stresses between a high-mass/low-velocity impact and a low-mass/high velocity
impact may result in material failures even if Ek remains constant.
143
As a result, a reduced energy level of fifteen Joules was specified. This level of
impact will still provide enough deformation of the different string types to highlight
any variation in their characteristics.
This value of potential energy resulted in an impact velocity of 2 metres per second.
For completeness the drop tests for impact velocities of 1 metre per second and 1.5
metres per second were also performed. These drop tests were completed five times
for gut, polyester and nylon string-beds, each of which was strung to a tension of 223
Newtons.
As well as recording the force measured from the load cell and the velocities via the
light gates the impacts were also recorded using a Photron Fastcam SA5 high speed
video camera at a frame rate of four thousand frames per second. The manufacturer
claims that the timing of the system is accurate to within plus or minus 0.005%
(Photron, 2012), which is not a significant enough error to be of concern.
The contact time of each impact was taken from the high speed video by counting
the number of frames from when the impacter first contacted the string bed to when it
rebounds and contact ceases. If we assume a potential error of half a frame due to
contact occurring slightly before it is captured and half a frame due to it ceasing
slightly before it is captured, this gives a potential experimental error of plus or minus
0.00025 seconds.
The average COR and contact time and the corresponding standard deviation were
calculated for each of the string types and each impact velocity. In the case of the
contact time, there was no measured variability detected i.e. the experimental
variation was too small to be identified with a frame rate of four thousand frames per
second. Therefore the potential error for the contact time is 0.00025 seconds. For the
COR the variability was found to be less than 0.5% in all cases. When added to the
144
potential error of the initial velocity measurement this gives a maximum potential
error of 0.75% across all impact conditions.
The results of the experiments for COR and contact time can be seen in Figure 6.11
and Figure 6.12 respectively. Figure 6.11 shows that for each string material the
COR reduces as the impact velocity is increased. Conversely, an opposing trend, in
the case of the contact time can be seen in Figure 6.12.
0.98
0.96
0.94
0.92
0.9
0.88 1m/s
0.86 1.5m/s
0.84 2m/s
0.82
0.8
0.78
0.76
Gut Poly Nylon
Figure 6.11: Coefficient of restitution of various string materials
0.005
0.0048
0.0046
0.0044 1m/s
0.0042 1.5m/s
2m/s
0.004
0.0038
0.0036
Gut Poly Nylon
Figure 6.12: Contact time for various string materials
145
It is interesting to note that, whilst the COR for the different strings shows some level
of variation at the higher impact velocities, there is very little difference between them
for impacts at one metre per second. This may suggest that for lower velocity
impacts, such as drop shots and volleys, there would be a noticeable variation in the
COR for different strings but for high velocity impacts such as serves and baseline
forehands there would be little difference.
However, as previously noted, the most significant contribution of the strings in terms
of the overall COR of the ball racket system is how quickly the strings return to their
original position, which dictates the dwell time of the ball. By looking at the contact
times in Figure 6.12 it can be seen that the contact times of the strings and the rigid
ball actually shows the greatest variation at the higher velocity of two metres per
second. This would suggest that, since the ball will have more time to deform on the
bed, nylon strings would have a reduced COR at higher velocities, relative to the
other strings.
The materials properties testing obtained the desired results in terms of Young’s
modulus required for the finite modelling whilst also providing insight into string
performance. It was shown that the tension which provided the most consistent value
of modulus for a varying strain rate was 223 Newtons. It was also shown that whilst
all the strings performed similarly in terms of COR with an impact with a solid
hemisphere, nylon provided the longest contact time, which when in contact with a
real tennis ball, would lead to a lesser COR as more deformation of the ball, and
hence more energy dissipation, would occur.
146
7 Modelling of friction
The following section deals with the role of friction in oblique tennis impacts and
seeks to compile a library of data which can be used to define the contact between
the ball and strings and strings with themselves, within the finite element model.
7.1 The role of friction in tennis racket impacts
The role of friction in tennis rackets impacts is a complex topic and opinions vary
amongst academics who study the sport. Some authors (Daish, 1972, Goodwill,
2002, Brody, 2002, Allen et al., 2010b) have taken the simplistic view that all tennis
strings have a large enough COF to instigate rolling and that any increase in COF
above this critical point has no effect. Cross (2000), however, disagreed and adopted
a more detailed approach and analysed the COF (μ) as two separate components μs
and μR, the coefficients of sliding and rolling friction respectively.
The underlying fault with the assumption that the COF has no effect on the
generation of spin is that, unlike the work of Cross, it assumes a constant COF
throughout contact. In reality μs will be significantly different from μR and a system
which has a similar value of μ for the rolling section of an impact may vary greatly
during the sliding section. As a result the ball may slide for longer and at a higher
velocity and the resulting level of spin would surely be affected.
Using a series of mechanical formulas Cross calculated the effect which increasing
the value of μs would have over the outbound spin, ω2. The results can be viewed
below in Figure 7.1, where the tilt angle, β, (Figure 7.2) represents the angle of the
head. Cross showed that for a COF equal to 0.2 the spin rate will increase as the tilt
angle of the racket, β, is increased.
147
Figure 7.1: Variation in spin vs head angle for μs = 0.2 (Cross, 2000)
148
β
ω1
ω2
Figure 7.2: Schematic of the ball’s flight path
It seems reasonable to assume, therefore, that if μs has such a profound effect on the
generation of spin then the point at which sliding occurs will also be a significant
factor in the level of spin generation. Whilst sliding motion is relatively simple to
simulate, rolling motion, requires rather more consideration.
7.2 Coefficient of sliding friction
7.2.1 Tribology
Obtaining the COF for a system through experimental data can be difficult since the
magnitude is dictated by a number of factors. COF is very much a function of the
entire system and a value of μ obtained for a ball sliding at 10 metres per second
would not be representative of a system containing a ball spinning at 200 revolutions
per minute. It is therefore necessary to create an experimental setup where the
conditions of the system in question are reproduced as closely as possible.
149
The most commonly used type of apparatus in friction measurement is known as a
tribometer. One of the most prevalent manufacturers of Tribometers, CSM
Instruments (CSM, 2012) describes the basic principle of tribology (the science of
tribology) as follows:
“Tribometers determine the magnitude of friction and wear as two surfaces rub
together. In one measurement method a flat or a spherical probe is placed on the test
sample and loaded with a precisely known weight. The sample is either rotating or
reciprocating in a linear track. The resulting frictional forces acting between the probe
and the sample are measured.”
7.2.2 Measurement Equipment Set-up
7.2.2.1 Tribometer
Cotton (2008) performed significant research into the measurement of sliding friction
of soccer balls against synthetic turf. A schematic of the system used by Cotton is
shown below in Figure 7.3. At one end of the lever system lies a steel drum which is
attached to a motor turning at a constant speed. At the opposite end is a weight stack
which is used to apply a known force to the drum via the sample tray.
150
Figure 7.3: Schematic of the tribometer used by Cotton (2008)
As Cotton was investigating the relationship between soccer balls and turf the drum
was lined with soccer ball panel material approximately one millimetre thick and the
drum was placed above a sample tray containing synthetic turf. Once the motor is up
to full speed the sample tray is pressed against the drum and, since the tangential
resistance applied via the tray is known from the weight stack, the resistance
measured via the drum can be converted to COF.
When the ball impacts the string-bed it meets with a frictional force, F, which is the
product of the COF, μ, and the normal force, R), as shown in Equation 7.1.
F = µR Equation 7.1
151
Since the radius of the drum, r, and the applied normal force, R, are known it is
possible to calculate μ from the difference in torque, T, experienced by the motor
when the force is applied to the rotating drum, as shown in Figure 7.4. Inserting
Equation 7.2 into Equation 7.1 yields Equation 7.3:
𝐹 = 𝑇/𝑟 Equation 7.2
F T/r
µ= = Equation 7.3
R R
T
Direction of
drum motion
r
F
Frictional Force
Normal Force R
Figure 7.4: Free-body diagram of the tribometer in contact with the test surface
As stated previously, in order for this procedure to give as accurate a value of μR as

possible the conditions of the system must be closely replicated during the
experimental analysis. One such condition is the rotational velocity of the drum. A
typical tennis ball impact would occur with an initial rotational velocity of around 600
radians per second. Unfortunately, the drum was only capable of rotating at speeds
of around 150 radians per second, so the maximum conditions experienced during
an impact could not be investigated. It was felt however, that the speeds produced by
152
the drum would offer valuable insight into the behaviour of friction under a variety of
conditions.
The other significant variable in this procedure was the normal force applied via the
sample tray. In order to ascertain the level of applied force required it was first
necessary to obtain the level of deformation experienced by the ball during an
oblique impact.
7.2.2.2 High Speed Camera
The apparatus used to obtain the deformation of the ball was similar to that used by
Cottey (2002) to capture the COR of a tennis racket impact. To fire the balls Cottey
used a pneumatic cannon, which was specially designed and built by the
Loughborough University Sport Technology Institute for the analysis of high speed
ball impacts. The velocity of the balls can be determined by varying the pressure of
the cannon. The relationship between the pneumatic pressure of the cannon and the
initial velocity generated for a tennis ball is shown below in Figure 7.5:
153
Figure 7.5: Variation of ball velocity with air pressure, Cottey (2002)
7.2.2.3 Compressive force experienced by the ball during impact
This setup differed from that used by Cottey, however, in that the high speed video of
the oblique impact was acquired using a Fastcam SA5 camera (Cottey used a Kodak
Ektapro 4540). The impact was filmed at 4,000 frames per second with a pixel ratio
of 1,280 by 768.
154
Figure 7.6: Deformation of the ball captured using the Photron SA5 camera
In order to obtain an impact of as close to thirty metres per second as possible, the
air cannon was set to 0.276 Megapascals. A set of light gates were placed at the exit
point of the cannon to measure the initial velocity and the cannon was fired several
times before an impact of 29.8 metres per second was achieved.
The footage used (shown in Figure 7.6) was of a racket tilted to an angle of 29
degrees (or 61 degrees off normal) relative to the flight path of the ball. As stated, the
ball impacted the racket at a velocity of 29.8 metres per second, with negligible,
assumed zero, angular velocity. The deformation of the ball was measured using the
image processing software Image Pro-Analyzer® and plotted against time. The
results can be viewed below in Figure 7.6.
Given that the deformation of the ball was measured from the images of the ball
during impact it was necessary to ascertain the clarity of the images to understand
the potential error. It is shown below in Figure 7.7 that the surface of the ball can be
155
identified to within plus or minus two pixels. The standard diameter of a tennis ball is
known to be 0.067 metres (Appendix 1) and, using Image Pro-Analyzer, was
measured at 154 pixels. Thus, using Equation 7.4 the potential measurement error
associated with this method is plus or minus 0.0008 metres.
+/- 2 pixels
Figure 7.7: Potential measurement error from the image of a tennis ball
𝑑𝑚𝑒𝑡𝑟𝑖𝑐
𝐸𝑟𝑟𝑜𝑟𝑚𝑒𝑡𝑟𝑖𝑐 = 𝐸𝑟𝑟𝑜𝑟𝑝𝑖𝑥𝑒𝑙𝑠 𝑑𝑝𝑖𝑥𝑒𝑙𝑠
Equation 7.4
0.067
= ± �2 × �
154
= ± 0.0008 𝑚
Using the footage of the impacts, a plot of the ball’s deformation against time was
generated and is shown below in Figure 7.8.
156
14
12
10
Deformation (mm)
8
0
0 2 4 6 8 10 12 14
-2
Time (ms)
Figure 7.8: Ball deformation vs time for an oblique racket impact
The stiffness of the tennis ball was assumed to be 16 kilonewtons per metre, as
stated by Sissler et al. (2010), since the same make and model of ball was used in
both studies. Applying this value of stiffness, K and the maximum deformation, d
(taken from Figure 7.8) to Equation 7.5 resulted in a calculated force, F of 176
Newtons being applied to the drum via the sample tray.
𝐹 =𝐾×𝑑 Equation 7.5
= 16,000 × (11 ± 0.8)
= 176 ± 13.6 𝑁
157
7.2.2.4 Experimental method using Tribometer
Having determined the applied force required, the experiment was carried out and
the results logged via the acquisition unit. Once the drum is in motion, the acquisition
unit gives a read-out of the torque measured from a transducer within the drum. A
torque value, T, was measured by the tribometer during the test and, using the
applied force, F, and drum radius, r, converted into μ.
The British Standard for measuring the friction of rubber, BS ISO 15113:1 1999,
suggests that each sample should be tested five times and each time should consist
of three cycles, as shown in Figure 7.9, the average of which is then taken to be the
true value. As such, the sample was loaded three times and an average of the three
was calculated using Equation 7.6.
0.6
0.5
0.4
0.3
COF
0.2
0.1
0
0 2 4 6 8 10 12 14 16
-0.1
Time s
Figure 7.9: Coefficient of friction vs. time
𝐹1 +𝐹2 +𝐹3
𝑀𝑒𝑎𝑛 (𝜇) = Equation 7.6
3𝑅
158
In order to convert the tribometer to one suitable for testing tennis string/ball cloth
friction the drum was wrapped tennis cloth and the sample tray was replaced with the
small square string-bed previously mentioned.
The variation in COF for several variables was investigated using the Tribometer,
including string gauge, string-bed density and orientation. Each variable was tested
for Babolat Hurricane mono and multi-filament strings.
7.2.3 Testing procedures and results
The conditions altered during testing were:
• String gauge: 1.25 millimetres, 1.30 millimetres and 1.35 millimetres

• String spacing (mains and crosses simultaneously): 5 millimetres, 10
millimetres, 15 millimetres and 20 millimetres (for a string gauge of 1.30
millimetres)
• String orientation: Conventionally strung and diagonally string (45 degrees to
the normal of the frame) (for a string gauge of 1.30 millimetres)
The results were processed as described in the previous section and can be viewed
in Figures 7.10, 7.11 and 7.12. The first conclusion to be drawn from each of the
figures is that the multifilament string has a consistently higher COF.
Figure 7.10 shows that, for both string compositions, there is an increase in COF by
around 7%-8% as the string diameter is increased. This could be explained by the
fact that thinner strings tend to be more elastic and thus would deform more readily
to the shape of the wheel under load.
159
0.6
0.55
0.5
COF
0.45
Multi
0.4
mono
0.35
0.3
1.25 1.3 1.35
String Gauge (mm)
Figure 7.10: COF of different string gauges
The theory that a more deformable string-bed leads to a reduction in COF is

reinforced by the results in Figure 7.11, where the COF for different string densities
are displayed. Naturally, a less densely strung strung-bed would lead to a less
inhibited, more deformable bed which, the results seem to suggest, leads to lower
level of friction.
160
0.6
0.55
0.5
COF
0.45
0.4 mono
tour
multi
0.35
0.3
1 hole 2 holes 3 holes 4 holes
Distance between strings (1 hole = 5mm)
Figure 7.11: COF for different string densities
Although it was difficult to compare the string-densities of the conventional and

diagonally strung beds it was visibly noticeable that the diagonally strung pattern was
more open. It seems logical therefore, to assume that the change in COF due to
string orientation is again due to the change in string-bed stiffness.
161
0.6
0.5
0.4
COF 0.3
mono
0.2 multi
0.1
0
conventional string-bed diagonal string-bed
String-bed type
Figure 7.12: COF for different string orientation
7.3 Coefficient of sliding friction
The mechanism of sliding friction between two contacting interfaces is well

documented in a variety of fields. As earlier stated, the value of sliding COF tend to
be system specific and there have been studies that have attempted to ascertain
values for the sliding phase of a tennis racket impact. Cross (2000) and Bao et al.
(2003) attempted to measure the sliding COF of a tennis ball over a string-bed by
placing weights on half a tennis ball to form a sled and measuring the force required
to drag the sled along the bed at a constant velocity, resulting in COF values of 0.2-
0.49.
Although attempts to quantify the sliding COF between tennis ball cloth and racket
strings have been made, little work has been done with respect to the interaction
between the strings themselves. The interaction between the strings is of importance
because as the ball impact the string-bed there will inevitably be some movement of
the strings relative to one another and the magnitude of this movement will be, to
some extent, dependent upon the COF between the strings.
162
With this in mind, an experimental set-up was created, similar to that of Cross, which
could be used to measure the interaction properties of the strings with themselves as
well as the strings with the ball. The main difference with the experimental set-up
compared to the apparatus used by Cross is that rather than dragging a sled across
a string-bed, a hanging weight stack was pulled between two pulleys, with a known
friction, by tennis string.
To measure the COF of the string against the ball, a platform covered in tennis cloth
was placed under the string. In order to measure the COF for the string against the
string the platform was simply replaced with a tennis string pulled to a tension of 178
Newtons.
The Instron 3365 Force-Displacement machine described in section 4.3.1.1 was

used to control the velocity at which the string was pulled (0.45 metres per second)
and measure the required force. The normal force, N, applied to the tennis cloth and
sting was calculated from the mass of the weight stack, m, using Equation 7.7:
N = (mg) × 0.5 Equation 7.7
Where g is the gravitational acceleration exerted on the weight stack. The factor of
0.5 in Equation 7.7 is a result of assuming that the force created by the weight stack
is shared evenly between the pulley which feeds in the tennis string and the tennis
ball cloth or string being tested.
The mass was varied to provide a normal force of 2.25 Newtons and 75.8 Newtons in
the case of the string against the ball cloth and 35.9 Newtons and 75.8 Newtons in
the case of the string against string. The normal force values were restricted by the
experimental set-up, which was only able to support a certain number of weights.
163
The COF, μ, was calculated using Equation 7.8, where F is the force required to pull
the string across the tennis ball cloth or string.
F
µ=
N Equation 7.8
As with all experimental set-ups measuring frictional force, there will be a component
of the force which should be attributed to the inherent friction within the testing
apparatus. An initial test was performed, therefore, to establish the force required to
freely pull the sled along the testing apparatus without any tennis cloth or string
impeding its progress (Figure 7.13). Having established this force, it was then
subtracted from all subsequent values (Equation 7.9) in order to give the true value of
frictional force for the interaction between the string and the ball (Figure 7.14) and the
strings themselves.
F = Measured Force – Frictional Force of Rig Equation 7.9
164
Instron
Machine
Pulley String
Weight
Stack
Weight Stack Travel
Figure 7.13: Experimental set-up for testing sliding friction of rig
Instron
Machine
Pulley
Weight
Tennis
cloth Stack
Figure 7.14: Experimental set-up for dragging a tennis string over tennis cloth
165
The four test conditions (two string tensions against the tennis cloth and two string
tensions against the tennis string) were performed five times for a mono-filament
(Babolat Hurricane Pro Tour) and multi-filament (Babolat Hurricane Tour). The data
points displayed in Figures 7.15 and 7.16 display a distinct pattern of COF variation
with string travel. For the multi-filament string against cloth (N = 75.8 Newtons) the
standard deviation varied across the five tests as shown in Figure 7.17. For
consistency, the maximum standard deviation from each test condition was assumed
the truest value and presented on Figures 7.15 and 7.16 as such, e.g. the maximum
standard deviation on Figure 7.17 was 8% of the mean average and presented as
the most reliable value across all values in Figure 7.15.
The results of the string against cloth experiments can be seen below in Figure 7.15
along with a tabular comparison of values obtained from the previously mentioned
studies, in Table 7.1.
0.6
0.5
0.4
Multi 75.8N
COF
0.3
Mono 75.8N
Multi 2.25N
0.2
Mono 2.25N
0.1
0
15 20 25 30 35 40 45 50
String Travel (mm)
Figure 7.15: Sliding COF of different string against tennis ball cloth
166
Author COF Pressure Sliding Velocity
Cross, 2000 0.27 – 0.45 30 KPa <1 ms-1
Bao, 2003 0.20 – 0.49 5 – 20 KPa 0.2 ms-1
This study 0.19 – 0.53 n/a 0.45 ms-1
Table 7.1: Comparison of COF for tennis string against tennis ball from various
studies
The main observation to be drawn from Table 7.1, is that the results are in a similar
range to that of Cross’ and Bao’s. Figure 7.15 shows that the multi-filament tour
string has a higher COF than the mono-filament and also suggests that this
difference is amplified at a greater string tension.
0.26
0.24
0.22
0.2
Multi 37.9N
COF
0.18
Mono 37.9N
0.16 Mono 75.8N
Tour 75.8N
0.14
0.12
0.1
1 2 3 4 5 6 7 8
String Travel (mm)
Figure 7.16: Sliding COF of string against string
167
0.6
0.5
0.4
COF
0.3
0.2
0.1
0
10 15 20 25 30 35 40 45 50 55
String Travel (mm)
Figure 7.17: Sliding COF of multifilament string against tennis cloth (N = 75.8
Newtons) with variable standard deviation error bars
In the case of the string against string experiments it was necessary to raise the
lower normal force from 21.6 Newtons to 347.3 Newtons. The upper string tension,
however, remained constant at 347.3 Newtons.
Figure 7.16 shows the sliding COF of the string against string testing for the mono
and multi-filament strings. The results show a similar trend to the string against ball
with the COF reducing as the string tension is increased. The differential between the
mono and multi-filament strings is also amplified at higher normal force.
As suspected, however, there is a significant difference between the sliding COF

results for the two experimental set-ups, with the string against string values proving
to be significantly less in all cases. This reduction in COF is likely to be due to the
string containing significantly less surface asperities than the tennis cloth, thus
168
reducing the level of mechanical interlocking which will take place between the
interfaces during sliding.
Two conclusions can be immediately drawn from viewing these results:
• The magnitude of sliding COF is significantly different between the ball and
the string and the strings themselves and, thus, independent contact
definitions should be assigned to the two interactions.
• The magnitude of COF for the strings sliding against the strings is highly
dependent upon the tension of the string. Thus, assigning a single interaction
property across all values of string-bed tension may not be appropriate.
7.4 Friction properties of different strings
As well as indicating the deviation in the COF across the string-bed, examining the
variation in the contact properties of different strings also allows an insight into the
stringing process. During stringing a uniform tension is applied to all of the strings but
as the stringer reaches the end of the stringing process it becomes more difficult to
weave the cross strings through the mains and, as a result, less extension of the
string is required to pull it to tension (since a greater level of tension already exists
from the string having been woven through a stiffer bed than the previous string).
Therefore, an experiment is required, where the different friction levels experienced

by mains strings during the stringing process is identified. Performing this experiment
will not only provide insight into possible pitfalls of a global interaction property
assignment, it will also indicate if a similar concern exists with regards to material
property assignment.
169
This test was performed in a similar manner to the previous tests with the exception
that the racket was strung to various points and then the next string was pulled to
tension using the Instron 3365 force measurement machine described in Section
4.3.1.1 (as shown in Figure 7.18) and the other end of the string was tied off and
effectively clamped. The same set-up was used as described in Section 6.1.1 but in
this instance the lower vice grip was used to hold the racket in place whilst the upper,
mobile, vice grip was used to elongate and tension the cross string.
The string’s Young’s modulus was calculated from the resulting force displacement
curve. This procedure was carried out for the string without the racket to provide a
reference value and for the 1st, 4th, 8th, 12th and 16th cross strings.
Force
Measurement
Machine
Figure 7.18: Experimental set-up for measuring the variation in frictional

properties/Young’s modulus of different cross strings
170
2200
2150
2100
2050
2000
1950
1900
ref 1 4 8 12 16
String
Figure 7.19: Young’s modulus of different cross strings
The results of the experiment can be seen in Figure 7.19. The results show a very
slight reduction in modulus for the cross strings relative to the reference value
(approximately 6%). Furthermore, a very slight downward trend is present in the
modules of the measured string as the strings are added. This is contrary to the
expected outcome. A possible explanation could be due to the fact that more force is
required to elongate the string than to simply “straighten” it out by displacing the
mains strings. As it is difficult to tell at what point the force is acting to elongate the
string and not simply “straighten” it by displacing the mains, however, it is difficult to
prove or disprove this theory. Given the very slight magnitude of the difference
between the strings (>3% from the 1st to the 16th string) it was assumed that a global
property assignment for the string on string COF would be appropriate.
171
It has been shown that assigning a global interaction property to the model is not
suitable if the different interactions between ball and strings are to be modelled
accurately. The results of the experiments carried out in this section have shown that
the values of COF between the mains and cross strings and the ball and the strings
are significantly different.
It has also been shown that the COF varies with string tension, and as such, when
modelling different string-bed tensions; it would be prudent to alter the interaction
property assignment accordingly. As a result, subsequent models will contain a COF
value for the strings against the strings and the strings against the ball and the values
will be altered depending on the string material being used.
172
8 Oblique impacts of spheres
In the previous chapter, it was shown that the level of friction produced by a string-
bed is dependent upon several characteristics. Friction is of interest to this study
because of the role it plays in spin generation, since high levels of friction will cause
the ball to “bite” sooner and slide less. Although it is accepted that these factors have
an effect on spin generation, little knowledge of the “effect” is available. In this
chapter, a series of simulations will be performed to study the behaviour of the ball
during impacts of different inbound angles. It is hoped that by studying these varying
impacts a better knowledge of the following points can be obtained:
• How the ball behaves during an oblique impact (i.e. does the ball start
spinning as soon as it impacts the string-bed or does it slide first?)
• How this behaviour might change with the inbound angle (i.e. is the potential
sliding phase of the impact lengthened or shortened?)
• What affect this all has on outbound spin.
8.1 Dynamics of oblique impacts
As discussed previously, in Chapter 7, there are two main components to the oblique
impact; the sliding component and the rolling component. Figure 8.1 is taken from
Brody et al. (2002) and depicts the various velocity components of a ball sliding (a-b)
and biting (c-d). In the case of the sliding ball a point at the top of the ball is rolling
around the axis at 6 metres per second while the axis itself is travelling at 6 metres
per second. The result of this motion is an overall forwards velocity of 14 metres per
second for the point at the top of the ball.
Similarly the point at the bottom of the ball is travelling at a resultant forward velocity
of 2 metres per second, as is the case for every point on the bottom of the ball in
173
contact with the string-bed. The ball’s most forward point has a forward velocity of 8
metres per second whilst also rotating downwards at a velocity of 6 metres per
second giving a resultant velocity of 10 metres per second (using Pythagoras 102 =
82 + 62) in a downward/forward direction. The point to the rear has a similar resultant
velocity with an upward/forward direction. When viewing Brody’s graphical
interpretation of this mechanism in Figure 8.1, it is important to remember that the
velocity vectors around the edge of the ball are relative to the centre of the ball,
whereas the velocity vector at the ball’s centre is relative to the surface it is
impacting.
The frictional force experienced by the section of the ball in contact with the string-
bed, however, has the effect of reducing the forwards velocity and increasing the
rotational velocity of the ball. When this happens, a critical point is reached where the
velocity of the bottom of the ball and the centre have equal and opposite velocity
components and the ball is translationally at rest. This is known as the point at which
the ball “bites”. As the ball continues to rotate there is a build-up of energy at the
front, which is returned to the rear and transformed to an upwards translational
velocity, causing the rear of the ball to become “unstuck” from the string-bed, thus
causing it to rebound.
174
Figure 8.1: Motion of a ball when it slides (a-b) or bites (c-d) (Brody)
A simpler way of thinking about oblique impacts is to compare the ball impacting the
string-bed to a car during heavy braking. The frictional force experienced by the front
of the ball can be compared to the braking force applied to the front wheels of the
car. As the car brakes, the car slides forward whilst also rising up at the rear end and
lowering at the front end as the momentum from the rear of the car seeks to
overcome the braking force at the front. The same principal applies to the ball during
impact. As the ball slides along the string bed, the increasing friction force acts like
the braking force on the front wheels of the car and, just like the car, the rear of the
ball will begin to rise up. When the friction force is significant enough the front of the
ball begins to move even slower than the rear of the ball, the ball begins to pivot
around its front and rolling occurs.
175
8.2 FE modelling of oblique impacts
8.2.1 Creation of the model
Although several authors (Ashcroft et al., 2002, Bao, 2003, Cross, 2003) have looked
at the results of changing the ball’s inbound angle, in terms of outbound spin and
velocity, little research has been carried out into the how the mechanism of the
impact changes (i.e. when do the sliding/rolling phases begin/end). In order to obtain
a more complete understanding of how the mechanisms of the impact change with
inbound angle a simple model of a ball obliquely impacting a square string-bed was
created. This model was similar in dimensions to the string-bed mounted on the
tribometer. The strings’ geometry and mesh were generated using the strategy
described in chapter 3. The resulting model can be seen in Figure 8.2.
An initial translational velocity of 30 metres per second (zero rotational velocity) was
assigned to the ball and the trajectory of the ball was defined by altering the Y and Z
velocity components, where Y and Z are the vertical and horizontal directions
respectively. The ball model used was developed by Sissler et al. (2010) and
required the inclusion of an inflation step to obtain a suitable internal pressure before
the application of the velocity. The velocity boundary condition was created in a short
dynamic/explicit step lasting 0.05 milliseconds before being deactivated in the impact
analysis step; a dynamic/explicit (described in Section 3.1.2) step lasting 7
milliseconds. This procedure was repeated for the following impact angles, α (for a
depiction of the impact angle see Figure 8.3):
• 32 degrees
• 34 degrees
• 39 degrees
• 45 degrees
176
The ball was fired against a string-bed smaller than that of a normal racket, in order
to decrease the number of elements used, and hence computation time, whilst
retaining a good geometrical accuracy of the strings.
The impact angle, α, is the angle between the velocity vector and the plane of the
string-bed (see Figure 8.3). These angles were chosen as they are amongst the most
commonly used in experimental examination of oblique tennis impacts (Goodwill et
al., 2002, Goodwill et al., 2004). The material and interaction properties used were
that of synthetic gut, obtained in the previous chapters.
y
x
Figure 8.2: Model used to examine oblique impacts.
177
y
Figure 8.3 : Schematic of an oblique impact
8.2.2 Results analysis
Once the analysis had been completed, a plot of the nodal force (Field output option
NFORC) within the ball experienced during impact was plotted in Abaqus/Viewer.
NFORC is defined in the Abaqus user’s manual (Abaqus, 2010) as “the force at the
node from the regular deformation modes of the element”. These nodal forces are
obtained from an element stiffness matrix similar to that shown in Equation 2.5,
although the matrices generated by a finite element solver such as Abaqus for a
dynamic analysis such as this will be much more complex.
The plot of the nodal forces was achieved using the “create XY data” option in
Abaqus/Viewer (N.B. “XY data” is a term used to refer to the data plot, which has X
and Y axes, and can be used for data of any sort, not just data in the X or Y planes).
The data plot for the 32 degree impact can be seen in Figure 8.4.
178
600
400 Rolling initiated
200
Force (N) 0
-200
-400
-600
Separation
Sliding initiated
-800
0.007
0
0.0007
0.0014
0.0021
0.0028
0.0035
0.0042
0.0049
0.0056
0.0063
0.00035
0.00105
0.00175
0.00245
0.00315
0.00385
0.00455
0.00525
0.00595
0.00665
time (s)
Figure 8.4: Total nodal NFORC experienced by the ball for a 32 degree impact
at 30 metres per second
As described in section 8.1, there are three distinct points during an impact where
sliding and rolling are present. The force contour of the ball at each of the peak
points can be seen below in Figures 8.5-8.7, where the grey area indicates stress
“hot-spots”. Figures 8.5-8.7 show the ball at various stages of contact with the string-
bed during an oblique impact. The images were obtained by “blanking” the string-bed
from the animation such that the deformation created by the strings (the grey area)
was visible. Comparing Figure 8.5 and Figure 8.6 shows the propagation of NFORC
due to the ball deformation during sliding. The grey area on these plots represents
the areas where the NFORC experienced by the ball exceeds the maximum value of
55.62 Newtons on the contour legend.
179
y
x
z
z
Leading edge
Figure 8.5: NFORC contour of the ball at the point where pure sliding is
initiated
y
x
z
z
Figure 8.6: NFORC contour of the ball at the point where rolling is initiated
180
During the impact the ball experiences significant deformation. As the ball travels
through the sliding phase it stretches in the Z direction while compressing in the X
direction. As the frictional force brings the ball to rest translationally the ball begins to
return to its original shape and in doing so initiates the rolling phase of the impact.
This is also the point at which the maximum negative NFORC is experienced, as
y
x
z
z
Figure 8.7: NFORC contour of the ball as it begins to rebound
181
600
400
Rolling Initiated
200
Force (N) 0
-200
-400 Separation
-600
-800 Sliding Initiated
0.007
0
0.0007
0.0014
0.0021
0.0028
0.0035
0.0042
0.0049
0.0056
0.0063
0.00035
0.00105
0.00175
0.00245
0.00315
0.00385
0.00455
0.00525
0.00595
0.00665
time (s)
Figure 8.8: Total NFORC experienced by the ball for a 34 degree impact
600
400
200 Rolling Initiated
0
Force (N)
-200
-400
-600 Sliding Separation

Initiated
-800
0.007
0
0.0007
0.0014
0.0021
0.0028
0.0035
0.0042
0.0049
0.0056
0.0063
0.00035
0.00105
0.00175
0.00245
0.00315
0.00385
0.00455
0.00525
0.00595
0.00665
time (s)
Figure 8.9: Total NFORC experienced by the ball for a 36 degree impact
182
600
400
Rolling Initiated
200
0
Force (N)
-200
-400
-600 Separation
Sliding
-800 Initiated
-1000
0.007
0
0.0007
0.0014
0.0021
0.0028
0.0035
0.0042
0.0049
0.0056
0.0063
0.00035
0.00105
0.00175
0.00245
0.00315
0.00385
0.00455
0.00525
0.00595
0.00665
time (s)
Figure 8.10: Total NFORC experienced by the ball for a 40 degree impact at 30
metres per second
Comparing Figure 8.4, Figure 8.8, Figure 8.9 and Figure 8.10 shows the different
rates at which the NFORC experienced by the ball accumulates for the different
impact angles and the point in time at which it reaches its peak. In each case there is
a visible plateau around the peak negative force. This negative force is generated by
the compression of the ball during impact and the plateau represents the sliding
phase of the impact. The point at which the ball begins to return to its original shape
and rolling is initiated is identifiable from the plots as the point when the force begins
to increase.
The most noticeable change in the NFORC plots is the peak positive value
experienced by the ball. This indicates that for increasingly normal impacts the ball
experiences a greater impact force and hence greater deformation. The second point
of note is that the time between the maximum compressive (negative) force and the
force beginning to return to zero is far greater for the 32 degree impact than the 40
183
degree impact. During this phase, where the NFORC holds relatively constant
around its maximum, the ball experiences its maximum deformation and the
dominant motion is sliding. It should be noted that the NFORC does not actually
cease to increase as the ball begins to slide, its rate of increase simply slows.
The sliding phase of the ball occurs when a force large enough to overcome the
frictional force is achieved and ends, when the elastic energy stored at the rear of the
ball, due to deformation, is transferred to the fore and the ball begins to roll and
rebound. As this rebound phase begins, the NFORC tends towards zero and then
increases in the opposite direction as the ball is propelled away by the strings and its
own elastic energy. The positive peak which can be seen on the NFORC plots is as a
result of the ball expanding back to its original shape and then slightly beyond. Whilst
the other slight dip below zero is simply the ball experiencing a slight contraction in
shape before returning to its original form.
The previously described process can be seen in Figure 8.11 where the left most
images are the front of the ball and the right most images are the rear of the ball. As
the impact develops the difference in NFORC between the front and rear of the ball
can be seen as the ball slides along the strings and stress increases. After the ball
overcomes the frictional force and begins to roll, the transfer of energy to the rear of
the ball can be seen from the changing contours. The same stress “hot-spots” are not
present on the rear of the ball as energy is transferred, rather, there is a much more
even distribution of stress around the rear of the ball as it begins to expand and
separate from the strings.
184
0 ms
2 ms
4 ms The energy
is
transferred
to the rear
of the ball
which
6 ms causes it to
separate,
Energy builds at rebound
the front of the and roll
ball as it slides
and begins to
bite
8 ms
10 ms
(a) (b)
Figure 8.11: NFORC contours of (a) the front and (b) the rear of the ball during
an oblique impact
185
Impact angle
(degrees) 32 34 36 40
Contact time (ms) 6.2 6.0 5.7 5.6
Time to roll (ms) 3.36 2.94 2.73 2.68
Time to peak F (ms) 4.6 4.35 4.1 3.7
Peak F (N) -744.6 -751 -802 -891
Table 8.1: Oblique impact data
Table 8.1 lists some of the key measurables for the different impact angles. In the
case of the 32 degree impact, the NFORC increases relatively gradually before
beginning to slide at around 2.8 milliseconds. Unfortunately, it is not possible to
identify a specific point in time at which the sliding phase is initiated, as the entire
ball’s trajectory does not change simultaneously.
The area of the ball closest to the string-bed will naturally be the first to deviate from
the original flight path as the frictional force is overcome with the string-bed and the
rest of the ball will change its trajectory as the force is increased and the ball deforms
until it is sliding uniformly along the bed. For the 32 degree angle, once the sliding
phase has been fully initiated, it continues for a period of about 1.2 milliseconds.
An annotated example of the NFORC plot for the 34 degree impact angle can be
seen below in Figure 8.12, in which the sliding and post impact oscillation phases are
labelled. It can be seen from the NFORC plots that, as the impact angle is increased
the time taken to initiate sliding is reduced as is the length of the sliding phase; to the
point where it is of negligible length for impact angles of 40 degrees and above.
186
600
400 oscillation phase
200
0
Force (N)
-200
-400
sliding
phase
-600
-800
0.007
0
0.0007
0.0014
0.0021
0.0028
0.0035
0.0042
0.0049
0.0056
0.0063
0.00035
0.00105
0.00175
0.00245
0.00315
0.00385
0.00455
0.00525
0.00595
0.00665
Time
Figure 8.12: Example NFORC plot of a ball during an impact with a string-bed
Another trend visible from the plot is the increasing nodal forces experienced by the
ball post impact. This represents the more severe oscillation experienced by the ball
at more normal impact angles.
By exploring how oblique impacts change with inbound angle, a useful method for
investigating the mechanisms of spin generation has been discovered. It was shown
that for angles lower than forty degrees there are two distinct phases during the
187
impact; the biting phase and the rolling phase. It was also shown that the initiation of
the sliding phase is accelerated as the impact angle is increased and that the peak
force experienced by the ball is increased.
188
9 Dynamic validation of spin generation
Before using the model to predict the performance of novel string-bed designs it must
first be validated against experimental data of an existing string pattern. This section
describes the methods used to validate the model of said existing string pattern.
9.1 Normal impacts
9.1.1 Experimental set-up/method
To validate the normal impacts, a series of impacts were filmed using a Photron
Fastcam SA5 high speed camera, as described in section 7.2.2.3. The tennis racket
was clamped around its handle, and mounted to a frame (similar to the set-up used
by Allen et al., 2010a) and placed within a poly-carbonate case for safety. Unlike the
oblique impact described in Chapter 7 the tennis balls were fired at the racket along a
velocity vector normal to the string-bed.
The same racket frame was used for each analysis and was strung to 223 Newtons
tension with polyester, nylon and natural gut strings. After each stringing procedure,
the racket was left to settle for twenty four hours in order to avoid the immediate drop
in tension found by Cross (2001a). Again, the same pneumatic cannon was used to
impart velocity on the ball but this time the velocity was varied between 15 metres
per second and 30 metres per second. The variation in ball speed was obtained by
adjusting the pressure of the pneumatic cannon as per Figure 7.6, with nine impacts
performed for each string material.
The camera was positioned, such that the line of sight captured was parallel to the
string-bed as shown in Figure 9.1. The camera recording was remotely triggered at
the same time as the ball was fired from the cannon and the high speed videos of
189
each impact were analysed to give the inbound and outbound velocities as well as
the contact time.
The contact times, tc, were obtained simply by counting the number of frames for
which the ball was in contact with the string-bed and multiplying this number by the
length of a frame, in this case, 0.00025 seconds.
Measuring the COR required an inbound and outbound velocity measurement. In

both instances, this was achieved by counting the number of frames (which
subsequently gives a value of time) that it takes the ball to travel the length of its own
diameter. The COR was then calculated using Equation 9.1:
𝑁𝑜. 𝑜𝑓 𝑓𝑟𝑎𝑚𝑒𝑠 𝑖𝑛𝑏𝑜𝑢𝑛𝑑

𝐶𝑂𝑅 = 𝑁𝑜.𝑜𝑓 𝑓𝑟𝑎𝑚𝑒𝑠 𝑜𝑢𝑡𝑏𝑜𝑢𝑛𝑑
Equation 9.1
190
Figure 9.1: Position of the camera with respect to the racket
The contact times for the impacts can be seen in Figure 9.2. As one would expect the
synthetic gut strings, conventionally regarded as the string material with the largest
dynamic stiffness, have the longest contact time, whilst the ball tends to dwell slightly
less on the natural gut and polyester strings. The opposite trend for the COR can be
viewed in Figure 9.3, with the less stiff natural gut and polyester strings providing a
greater outbound velocity relative to the polyester strings.
191
35
30
Velocity (m/s) 25
Synth Gut
20 Poly
Natural Gut
15
10
0.0035 0.004 0.0045 0.005
Contact time (s)
Figure 9.2: Contact time for experimental normal (β=90°) tennis racket impacts
35
30
Velocity (m/s)
25
Synth Gut
20 Poly
Natural Gut
15
10
0.65 0.7 0.75 0.8 0.85 0.9 0.95
COR
Figure 9.3: COR for experimental normal (β=90°) tennis racket impacts
192
Figure 9.4 shows a comparison of how the COR varies with the contact time. The
results show that the stiffer materials produce a longer contact time than less stiff
strings. What is interesting to note from these results, however, is that the synthetic
gut strings produce a lower COR for a given contact time relative to the other strings.
This is evidence of the relatively poor elasticity of the synthetic gut strings since they
are unable to return the same level of energy to the ball during the rebound phase.
The increased elasticity of the natural gut relative to synthetic, however, allows it to
experience greater deformation whilst dissipating less energy.
0.95
0.9
0.85
0.8
COR
Poly
0.75
Synth
SynthGut
0.7
Natural
Gut
0.65
0.6
0.003 0.0035 0.004 0.0045 0.005
Contact time (s)
Figure 9.4: COR versus contact time
Such characteristics can make synthetic strings desirable to players as they allow for
a longer contact time during impact, which in turn leads to increased control for the
player, without losing too much power (outbound velocity).
193
9.2 Oblique Impacts
9.2.1 Experimental set-up/methods
A similar approach to that described in section 9.1 was used to obtain oblique impact
data. The only difference in the two experimental set-ups was that in the case of the
oblique impacts the racket was turned around the axis along its handle to vary the
angle of the ball’s inbound angle, β, relative to the plane of the string-bed, as shown
in Figure 9.5. Also, in the case of the oblique impacts, due to time constraints only
the synthetic gut strings were analysed.
Normal impact
Racket position
Racket Handle
Oblique impact
Racket position
Figure 9.5: Racket position during oblique impact
The different angles tested were similar to those used to examine oblique impacts in
section 8.2, ranging from 29 degrees to 45 degrees rotation around the axis parallel
to the handle. The specific values of the angles were dictated by the nature of the
racket clamp. As in section 9.1, the rackets were strung using natural gut strings to a
tension of 223 Newtons and the inbound velocity was altered by varying the pressure
of the pneumatic cannon.
194
The results of the experiment can be seen below in Figure 9.6, where a decreasing
trend in outbound rotational velocity, as β is increased, is apparent. These results are
in keeping with Cross (2000a) who also found a decreasing level of spin generation
as the inbound angle tended towards normal.
Spin Rate (Revs/min)
15.8m/s
23.3m/s
29.8m/s
Inbound Angle (β) (Degrees)
Figure 9.6: Spin rate for oblique impacts at ball various speeds
Furthermore, the results show that as the inbound velocity is increased, so too is the
resulting outbound spin rate. In order for this greater level of outbound spin to occur,
an increased transformation of energy must take place. This would suggest that
either less energy is lost during the impact (and transformed into rotational velocity)
or a greater level of translational velocity is transformed into rotational velocity; or
indeed a combination of the two.
195
The relationship between inbound translational velocity and outbound rotational
velocity could be investigated (much in the same way as the relationship between
rotational velocity and inbound angle was investigated in Chapter 8) using a finite
element model if the time were available to run all of the necessary simulations.
However, given the time taken to run a simulation of this type (over twenty four
hours) the resources were not available to investigate how the energy transformation
varies as a result of changing inbound velocity.
9.2.3 FE Model
An FE model was compiled to recreate the experimental oblique impacts at 29.8

metres per second. The ball was assigned a velocity vector of 30 metres per second
at an angle of 28 degrees, 32 degrees, 36 degrees, 39 degrees and 40 degrees to
the horizontal. The material and interaction properties used were those of the
polyester multifilament string obtained in Section 7.2.
Quantifying the rotational velocity of the ball is challenging as the ball is constructed
using 3D elements. The, rotational velocity, VR, is not available as an output for 3D
elements since only the translational degrees of freedom are available. Another
method of measuring the ball’s spin rate is therefore required. The rotational velocity
is available for the shell elements used to model the foundation layer of the ball’s
cloth. This is not a good indicator of the spin rate, however, as the ball is deforming
locally and the nodal values of UR at these points will only show how much the
elements are bending at this point.
Although the rotational properties of the ball’s elements are not available, it is still
possible to calculate the rotational velocity of the ball using the Cartesian co-
ordinates of the nodes. The co-ordinates of a point on the ball’s circumference are
sampled at two time increments, t1 and t2 and using the average co-ordinates of the
ball’s centre C as a reference point, the change in angle can be calculated. The
196
changing position is shown below in Figure 9.7, where Pi and Ci are the position of P
(a point on the surface of the ball) and C (the centre of the ball) with respect to ti.
Since the ball rotates predominantly in the XY plane, change in P and C along the Z
axis is omitted from the calculation
P1
C1
P2
C2
Figure 9.7: Position of a point on the ball’s surface during rotation
Once the analysis is complete, the coordinate position of each of the ball’s outer
surface nodes is obtained via the field output option and, using the “Operate on XY
197
Data” option from within the “XY data” menu, these points are averaged to obtain, C,
the coordinates of the ball’s centre. Using the Cartesian co-ordinates of P in the X
and Y direction the distance from C is then calculated.
The angle, αi for a point in time “i”, of the vector can be calculated using Equation
9.2.
𝑃 −𝐶
𝛼𝑖 = tan−1 � 𝑃𝑦𝑖 −𝐶𝑦𝑖 � Equation 9.2
𝑥𝑖 𝑥𝑖
The terms of this equation are depicted in Figure 9.8:
198
P
Py1 – Cy1
Px1 – Cx1
P
Py2 – Cy2
Px2 – Cx2
Figure 9.8: Change in angle of P with respect to C due to rotation
199
Having calculated the angle, α for each time increment, the spin rate can then be
obtained by dividing the resulting change in angle by the respective change in time,
as shown in Equation 9.3.
𝑆𝑝𝑖𝑛 𝑅𝑎𝑡𝑒 = (𝛼2 − 𝛼1 )/( 𝑡2 − 𝑡1 ) Equation 9.3
tan−1��𝑃𝑦2 − 𝐶𝑦2 �⁄(𝑃𝑥2 − 𝐶𝑥2 )� − tan−1 ��𝑃𝑦1 − 𝐶𝑦1 �⁄(𝑃𝑥1 − 𝐶𝑥 1)�

𝑆𝑝𝑖𝑛 𝑅𝑎𝑡𝑒 =
𝑡2 − 𝑡1
In order to eliminate the effect of ball deformation the spin rate is calculated 2
milliseconds after the ball has achieved separation from the string-bed and any
oscillations have been damped.
The above process was carried out for eight sample points around the ball’s
perimeter. The initial results of the FE analysis, shown in Figure 9.9, displayed a
decreased level of spin compared to the experimental results (for the 29.8 metres per
second impacts from section 9.2.2) and also did not follow the same trend; increasing
as the angle incidence is increased as opposed to decreasing. Looking at the FE
impacts objectively, it seems that the prolonged contact time of the shallower angles
is creating a higher level of energy dissipation and thus reducing the energy returned
to the ball and hence, the rotational velocity.
200
6000
5000
Spin Rate (Revs/min)
4000
Experimental
3000
FE Alpha
FE Beta
2000
1000
0
28 30 32 34 36 38 40 42 44
Inbound Angle (Degrees)
Figure 9.9: Spin rates for oblique impacts (experimental and FE)
At this point of the investigation, the material damping being used for the ball is the
same stiffness proportional damping used by Sissler et al. (2010). The most
commonly used method of damping, within Abaqus is Rayleigh damping. Rayleigh
damping is defined within the material model and consequently applied to any
element to which that material has been assigned. As shown in Equation 9.4, the
Rayleigh damping matrix, Cm, is equal to a linear combination of the mass and
stiffness matrices, M and K respectively:
𝐶𝑀 = 𝛼𝑅 𝑀 + 𝛽𝑅 𝐾 Equation 9.4
Where αR is the mass proportional damping factor and βR is the stiffness proportional
damping factor.
201
The non-linear rubber model used by Sissler et al. to represent the core of the ball is
defined in such a way that as the deformation of the ball increases, the stiffness, K,
also increases. As a result the stiffness proportional component, βRK, increases at a
non-linear rate, thus generating an increased level of damping.
As a result the value of beta damping applied (β = 0.0032) by Sissler to the rubber
core was removed and replaced with an alpha damping value of 1,000. This value of
α was obtained by a process of trial and error.
Since the contact time for impacts of smaller inbound angle (e.g. 30 degrees) is
greater, the deformation is far more prolonged. As a result, the level of energy
dissipation is far greater and energy which ought to be transformed into rotational
velocity is lost.
With this in mind, the βR component was removed from the model and replaced with
a value of αR. The αR component operates on the mass matrix which, for an analysis
of this type, remains constant. As a result the energy dissipation stands to be more
representative of experimental analysis and will not increase excessively with
increased contact time.
To alleviate this excessive energy loss the beta damping was removed from the ball
and mass proportional alpha damping was introduced. The results of the
experimental analysis are plotted against the results of the different models
containing alpha and beta damping in Figure 9.9. Whilst the beta damping model
displays the opposite trend previously described, the alpha damping results are much
more in keeping with the experimental results.
A series of experimental impacts were performed for both normal and oblique
inbound trajectory. The relationship between inbound velocity, contact time and COR
202
was investigated for polyester, synthetic gut (nylon) and natural gut. As found by
Brody et al. (2002) the contact time increased with inbound velocity as a result of the
higher deformation associated with a greater impact velocity. It was also shown that
the COR decreased with increasing inbound velocity, due to a greater level of energy
dissipation occurring during the elongated contact time.
A similar method of experimental measurement was then used to obtain the

outbound spin rate of oblique tennis impacts. The outbound spin rate was obtained
for a variety of inbound angles (28 degrees to 40 degrees) and velocities (15.8
metres per second to 29.8 metres per second), with a view to ultimately using the
data to establish the accuracy of oblique impacts simulated using the finite element
model.
The experimental results showed that as the inbound angle tended away from normal
(90 degrees) the outbound spin rate increases. For example, in the case of the 29.8
metres per second impacts the spin rates were 4,000 revolutions per minute and
3,582 revolutions per minute for inbound angles of 28 degrees and 40 degrees
respectively. Furthermore, it was also shown that as the impact velocity was
increased, so too is the outbound spin rate. In the case of the impacts with an
inbound angle of 28 degrees impact velocities of 29.8 metres per second yielded an
outbound spin of 4,000 revolutions per minute compared to just 2,448 revolutions per
minute for a lower impact velocity of 15.8 metres per second.
Initially, when comparing the outbound spin rate experimental results for the inbound
velocity of 29.8 metres per second with that of the model, the opposite trend was
observed. In the case of the model the spin rate was decreasing as the inbound
angle tended away from the normal. Taking into account the behaviour of the ball
during normal impacts, where increased energy dissipation occurs as a result of
increased contact time, it was felt that in the model the increased contact time
associated with more oblique inbound angles could be having a similar effect.
203
As a result, the material model used to represent the damping within the system was
altered from being a stiffness based feature (one which increases non-linearly with
the deformation of the ball) to a mass based feature (one which will remain constant
regardless of deformation). This method of modelling material damping proved to be
far more effective for oblique impacts and resulted in a trend which was much more
in keeping with the experimental results.
Having developed and correlated a model which can be used to simulate outbound
spin rates for oblique impacts, the next step in the project was to use the model to
investigate how spin rate changes as key variables such as string spacing, gauge
and orientation are altered.
204
10 Modelling of different string patterns
This chapter of the thesis brings all of the previous work together in an attempt to
model novel string-bed arrangements and measure their spin generation. In doing so
it is hoped that a greater understanding of how various string-bed characteristics
affect spin can be achieved.
10.1 Automation of string-bed geometry creation
As a number of string-patterns are to be modelled, steps are taken to partially

automate the procedure for generating a string-bed mesh. Although it is not possible
to fully-automate the process, due to the different programs used to create the mesh,
there is vast room for improvement, which will be addressed in this chapter.
In the procedure described in Section 3.3.1, the points on the string-bed where the
strings intersect are measured using a CMM and imported into the CAD package NX
(both of which are described in Section 3.1.1). This method is convenient when
creating geometry of an existing racket but is not viable when generating a novel
design.
The string variables which are altered during the analyses are those which a
manufacturer or indeed a player would be readily able to specify when selecting a
racket, such as:
• String density
• String orientation
• String gauge
• Young’s modulus
205
Therefore, a system is required, whereby a set of points representing the string
geometry can be generated at the touch of a button, whilst allowing each of these
variables to be instantly modified.
10.1.1 Generating points using Excel®
As well as generating splines through a series of specified points, NX also provides

the option of generating splines with “points from file”, where the file is a text file
containing the Cartesian coordinates of a series of points. The task of generating
such files is undertaken with the use of Excel®. Using Excel®, a spread-sheet is
created where lines representing the strings are generated using a series of
formulae.
10.1.1.1 Creating the X and Y position of the strings
One of the first things to consider when defining the geometry of the string-bed is the
X and Y position of the string bed. The spread-sheet was set-up with the cells
containing the main variables – string spacing and angle – position in the top left of
the page. The format of these cells is shown in Table 10.1:
A B C D
1 strings spacing angle deg
2 Mains 20 10 Ф
3 Cross 20 10 φ
Table 10.1: String variable definition cells
The “Y” position of the mains strings is then determined using the formulae shown in
Table 10.2, whilst the “X” position of cross strings is generated using the formulae
displayed in Table 10.3.
206
A B
6 cross 1 =(B7+C$3)
7 cross 2 =(B8+C$3)
8 cross 3 =(B9+C$3)
9 cross 4 =(B10+C$3)
10 cross 5 =(B11+C$3)
11 cross 6 =(B12+C$3)
12 cross 7 =(B13+C$3)
13 cross 8 =(B14+C$3)
14 cross 9 =IF(B3=17,0+D3,IF(B3=16,-(C3/2)+D3,IF(B3=18,(C3/2)+D3,B15+C3)))
15 cross 10 =IF(B3=19,0+D3,IF(B3=18,B14-C3,IF(B3<18,B14-C3,B16+C3)))
16 cross 11 =IF(B3=22,(C3/2)+D3,IF(B3=21,0+D3,IF(B3=20,-(C3/2)+D3,B15-C3)))
17 cross 12 =(B16-C$3)
18 cross 13 =(B17-C$3)
19 cross 14 =(B18-C$3)
20 cross 15 =(B19-C$3)
21 cross 16 =(B20-C$3)
22 cross 17 =IF(B$3<17,"",(B21-C$3))
23 cross 18 =IF(B$3<18,"",(B22-C$3))
24 cross 19 =IF(B$3<19,"",(B23-C$3))
25 cross 20 =IF(B$3<20,"",(B24-C$3))
26 cross 21 =IF(B$3<21,"",(B25-C$3))
27 cross 22 =IF(B$3<22,"",(B26-C$3))
Table 10.2: Formulae used to calculate the “Y” position of the cross strings
D E
6 mains 1 =(E7+C$2)
7 mains 2 =(E8+C$2)
8 mains 3 =(E9+C$2)
9 mains 4 =(E10+C$2)
10 mains 5 =(E11+C$2)
11 mains 6 =(E12+C$2)
12 mains 7 =(E13+C$2)
13 mains 8 =IF(B2=16,IF(((C2/2)+D2)<0,"",C2/2+D2),E14+C2)
14 mains 9 =IF(B2=17,0+D2,IF(B2=18,(C2/2)+D2,IF(B2=16,-(C2/2)+D2,E15+C2)))
15 mains 10 =IF(B2=20,C2/2+D2,IF(B2=19,0+D2,IF(B2=18,-(C2/2)+D2,(E14-C2))))
16 mains 11 =E15-C2
17 mains 12 =E16-C2
18 mains 13 =E17-C$2
19 mains 14 =E18-C$2
20 mains 15 =E19-C$2
21 mains 16 =E20-C$2
22 mains 17 =IF(B2<17,"",E21-C$2)
23 mains 18 =IF(B2<18,"",E22-C$2)
24 mains 19 =IF(B2<19,"",E23-C$2)
25 mains 20 =IF(B2<20,"",E24-C$2)
Table 10.3: Formulae used to calculate the “X” position of the main strings
207
Since the centre of the string-bed is the origin it is necessary to construct the
equations in such a way that the string closest to the origin can be changed
depending on the number of strings (e.g. for a string-bed with 16 cross strings,
strings 8 and 9 would be either side of the origin but if there were 20 cross strings, it
would be strings 10 and 11). For this reason, conditional “IF” statements are included
for the strings which will be at the centre of the string-bed and the equations for the
other strings are defined with respect to the central strings.
A standard scatter plot was created using the values generated from the formulae
listed in Table 10.3. The X value of the intersection was given by the main strings
formulae and the Y value was given by the cross strings formulae. This resulted in a
series of lines similar to that seen below in Figure 10.1:
200
Main1
150
100
50
0
-150 -100 -50 0 50 100 150
-50
-100
Cross 1
-150
-200
Figure 10.1: Scatter plot generated from the formulae in Table 10.2 and Table
10.3
208
10.1.1.2 Creating X and Y positions for angled string-beds
The procedure for generating angled string-beds is largely similar to that described in
the previous section. The main difference is that when a non-zero value of the angle
Ф is specified, the lines generated from the mains formulae are divided by Cos Ф.
Similarly, when a non-zero value of φ is specified the cross strings are divided by
Cos φ. As a result, the angled lines are generated from the scatter plots and using
the “line intercept” function in Microsoft Excel® the intersection points of the string-
bed are extracted.
10.1.1.3 Exporting the string intersection points from Excel®
Having created a spread-sheet which can conveniently create the points for the
strings, the next step in the automation process is exporting the data from Excel® and
into NX. This is achieved, using a “macro” which allows the user to specify a series of
actions which can be re-created at the touch of a button. In this case, the actions are
selecting each of the columns which represented a single string and, individually
writing them to a .dat file (a text file compatible with the “create points from file”
feature in NX). The .dat file produced a number of points to represent the strings, one
for each point where the string met the racket and one for each string intersection.
The string intersection points are offset by the radius of the string in order to create a
3D woven bed without penetration. In order for this to be successful, the user must
first save the spread-sheet to a convenient location, before running the macro, as this
will define the location to which the .dat files (which contain the points define the
strings) are saved when the program is run.
Due to incompatibilities in the software it is not possible to automate the stage of the
process during which the splines are transformed from 1-dimensional lines to 3D
parts. This process must be undertaken manually by sketching a circle at the end of
the spline and using the “sweep along guide” feature within NX (Figure 10.2), where
209
the shape of the circle is swept along the length of the spline to create a solid
geometry. Following this procedure, the user is then left with a 3D solid, as shown in
Figure 10.3, which can be exported into the meshing package, Hypermesh, to be
meshed before being analysed.
Figure 10.2: Splines through offset points to create woven bed
Figure 10.3: Circles swept along splines to create 3D strings
210
10.2 Spin of different string-beds
10.2.1 Different string density
Using the method outlined in section 10.1, a number of different string-beds, with
varying string-density, are generated. The characteristics of the string-beds are
detailed in Table 10.4.
Mains/Crosses String spacing String diameter Modified E (MPa)
String-bed 1 16/16 12 mm 1.38 mm 2.800 × 103
String-bed 2 16/17 12 mm 1.38 mm 2.760 × 103
String-bed 3 17/18 10 mm 1.38 mm 2.625 × 103
String-bed 4 18/20 10 mm 1.38 mm 2.483 × 103
Table 10.4: Characteristics of the string-beds
Initially, the string-beds are all assigned the material properties of natural gut at 223
Newtons tension, the properties of which were as follows:
E = 2.483 × 103 MPa
ρ = 1301 kg/m3
ν = 0.35
An explicit analysis is performed in which the ball impacts the string-beds on a

velocity vector of 30 metres per second, 32 degrees from parallel to the string-bed
(i.e. 58° to normal). An oblique impact is used in order to generate post-impact spin.
As the contact time of an impact is increased, so too is the potential for energy, to be
dissipated through ball deformation (which otherwise could be transformed into
211
rotational velocity). Given that spin, therefore, could be dependent on contact time,
which itself is dependent on the stiffness of the string-bed, a secondary analysis is
performed where the Young’s modulus of the strings is altered such that all the string
patterns are of the same stiffness.
To obtain the global stiffness of the string-bed a concentrated load of 10 Newtons is

applied to the fifteen most central nodes – thus giving a cumulative load of 150
Newtons - within the string-bed along the Z axis (the direction normal to the string-
plane). The stiffness of the bed is then calculated using the Z-displacement, (Abaqus
field output, U3), and the modulus of the each string-bed is adjusted such that all
string patterns gave the same value of U3 under load as the 18 by 20 bed. The
modified values of E can be seen in Table 10.4.
10.2.2 Different string gauge
In terms of the variability of performance characteristics of different string gauges

Brody et al. (2002) states that; since the elasticity of a string is proportional to the
inverse of its cross sectional area thinner strings are more desirable. This is because
thinner strings will deform elastically more under a given load and provide the player
with increased control and power. However, Brody makes little reference to the spin
generation performance of different string gauges.
One would expect that increasing the string diameter of a string-bed would have a
similar effect to increasing the string density, as the gaps between the strings will
decrease and the string-bed will become stiffer. However, as has been observed in
the previous chapter, the relationship between string-bed stiffness and mechanical
interlocking can lead to varying levels of spin. It seems prudent, therefore, to perform
a similar set of experiments to those in the previous section, in which the variation in
spin for the three most conventional string gauges (1.25 millimetres, 1.3 millimetres
and 1.38 millimetres) is examined.
212
The model used to perform the analysis is similar to String-bed 4 used in section
10.2.1, the only variation being the string gauge. This is altered by changing the
cross-sectional area of the strings at the CAD stage of the model development such
that three different string beds are generated for the diameters previously specified.
213
10.2.3 Different string orientation
Having observed the variation in spin due to changes in string density, another
interesting question may be: how is spin generation affected by the string
orientation? It can be assumed that increased mechanical interlocking will result from
the increased surface asperities associated with a higher string density. It is unclear,
however, how spin levels would react to significant changes in the magnitude and
orientation of those surface asperities which would result from the orientation of the
strings being altered.
Using the string-bed generation system described in section 10.1, a series of string
beds are created with different string orientations. The string-beds all contain 20
cross strings and 20 main strings and had a cross sectional diameter of 1.3
millimetres. The orientation of the strings is altered by varying the angle, Φ, of the
cross and main strings relative to one another, displayed in Figure 10.4.
Φ
φ
Figure 10.4: Novel string-bed arrangement
214
In total, seven string-beds (shown in Figure 10.5) are tested, with the value of Φ
varying from 30 degrees to 60 degrees in 5 degree intervals. The string orientations
of each of the string-beds are listed in Table 10.5 and shown in Figure 10.5. As in the
previous section the stiffness of the string-beds are tested prior to performing a full
impact analysis and the results of these analysis can be seen in Table 10.5. In this
case, however, the overall stiffness’s of the string-beds did not vary significantly
(<1%), therefore it was not considered necessary to alter the Young’s modulus to
eradicate the effect of the stiffness may have on spin generation.
Φ (degrees) Stiffness (kN/m)

String-bed 1 60 16.05
Table 10.5: String orientation of the modelled string-beds
215
(1) (2)
(3) (4) (5)
(6) (7)
Figure 10.5: String-beds 1-7
As in previous analyses, a velocity vector with an overall magnitude of 30 metres per

second at an angle of 32 degrees to the plane of the string-bed is imparted on the
ball. Having completed the various simulations, the outbound spin rates are obtained
using the procedure defined in section 9.2.3.
216
10.2.4 Spin generation of different string materials
In order to characterise the spin generation performance of different string properties,

the different material properties obtained earlier for the titanium polymer and the
polyester string are submitted to the model, to be compared with the results obtained
for the synthetic gut string. The ball is subjected to a velocity vector with a magnitude
of 30 metres per second at an incidence angle of 32 degrees perpendicular to the
string bed. The different string materials and their corresponding properties can be
seen below in Table 10.6. All strings had a cross-section diameter of 1.38
millimetres.
Material Young’s modulus (MPa) Density (kg/m3) Poisson’s ratio

Titanium 5.015 × 103 2056 0.35
Polyester 3.886 × 103 1565 0.35
Synth Gut 4.369 × 103 1670 0.35
Table 10.6: Material properties for different strings
10.3 Results/Discussion
10.3.1 Different string-bed densities
4200
4100
4000
Spin Rate (revs/min)
3900
3800
3700
3600
3500
3400
3300
16by17 17by18 20by18
Strings (mains by crosses)
Figure 10.6: Spin rate for different string densities
217
It can be seen from Figure 10.6 that the spin tends to increase with the number of
strings. There are a number of factors which could lead to this increase. Firstly it
could be the result of an increased level of mechanical interlocking between the
strings and the ball due to the increased number of strings.
One of the most interesting points to note from this result is that increasing the main
strings (conventionally considered to be the string with the most significant
contribution to spin generation Brody et al. (2002)) has a lesser effect on the
generation of spin than increasing the cross strings. Figure 10.7 shows that there is
an increase in spin of around 11% when 1 cross string and 1 main string are added
compared to an increase of only 2.5% when 3 main strings are added. This raises
the question of whether there is a critical string spacing at which the increase in spin
- due to reduction in the area between the strings - begins to plateau.
4200
4000
Spin Rate (revs/min)
3800
3600 normal E
increased E
3400
plate
3200
3000
16/16 16/17 17/18 18/20
Strings (mains by crosses)
Figure 10.7: Outbound spin for different string densities
The respective spin rates for the normal and modified values can be viewed in Figure
10.7. It can be seen that the spin rates for the strings with a modified value of E
(detailed in Table 10.4) are reduced relative to the strings with a normal value of E
218
(2.483 × 103 Megapascals). By viewing these results it can be seen that spin rate
does increase with string-bed density.
As a point of reference, the ball is also subjected to a 30 metres per second impact at
an angle of 32 degrees relative to the horizontal, against a rigid plate. This allows us
to see the levels of spin obtained without the contribution of mechanical interlocking
and effect on contact time due to string-bed deformation. Interestingly, the values of
spin for the plate are consistently less than those for the normal values of E but in the
case of the modified values of E, the most open string-bed (16 by 16) is actually less
than the plate.
This raised the question of whether or not the two factors in question (mechanical
interlocking and contact time) are independently variable with regards to the spin or
if, indeed, they interact at some point to increase spin levels, i.e. for a string density
less than 16 by 17 the increased contact time leads to an outright reduction in spin
relative to a solid, rigid surface. This would explain why, for the unmodified value of
E, the 16 by 16 string pattern still generates a greater level of spin than the rigid
plate, since it is less stiff and has a shorter contact time for the reduction in energy
due to mechanical interlocking to propagate.
As was stated in section 8, there are two mechanisms which are critical to the
generation of spin, the sliding phase and the biting phase. If the sliding phase is
prolonged, as would be the case for the rigid surface due to its lack of surface
asperities, greater levels of energy dissipation occur which leads to a reduction in
spin rates. In order to maximise spin, therefore, the sliding phase must be minimised
and the biting phase accelerated. These results show that this is possible with
mechanical interlocking, such as that introduced by a woven string bed. If the
asperities become too large, however, the contact time will be inadvertently
increased and all benefit lost.
219
10.3.2 Different string orientation
The variation in spin rate due to string orientation can be seen in Figure 10.8.
2400
2300
2200
2100
Spin rate (revs/min)
2000
1900
1800
1700
1600
1500
1400
30/60 35/55 40/50 45/45 50/40 55/35
Angle between strings ϕ/φ (degrees)
Figure 10.8: Spin rates for different string orientations
The first conclusion to be drawn from the results is that the spin rates are consistently
lower than that of the conventional string-bed configurations. Initial suspicions that
this reduction in spin could be due to the change in stiffness of the string-beds
relative to the conventionally strung beds were dispelled by performing a stiffness
test similar to that described in section 10.2, which showed a 22% reduction in
stiffness in the case of the novel string arrangements. This reduction in the overall
stiffness of the string-bed is due the fact that the distance between the string
intersections is generally larger for the novel string-beds than for the conventional
string-beds. As a result, deformation of the strings between their intersections (and
therefore across the string-bed as a whole) occurs more freely for the novel string-
beds.
220
Figure 10.8 does, however, display a generally increasing trend as Φ is increased,
peaking at 50 degrees before decreasing slightly. It would seem, therefore, that the
spin tends to increase as the angle of strings opens up to the flight-path of the ball,
e.g. configuration (a) in Figure 10.9 and Figure 10.10 will generate more spin than
configuration (b) (where configuration (a) represents a string configuration with a
greater value of Ф than configuration (b)).
This can be explained by the fact that after the strings are deformed by the ball in the
direction of its flight path, they will return this energy in the opposite direction, thus
imparting rotational energy upon the ball. The reason why configuration (a) is more
able to do this is that the component, L, which performs the role of the mains string –
in terms of imparting rotational velocity on the ball – is greater than configuration (b).
As a result, it is less resistant to deforming in the direction of the ball’s flight and
hence has a greater potential to absorb translational energy as it experiences a
greater level of elongation (Figure 10.10).
L D
(a) (b)
Figure 10.9: Ball impacting different string configurations
221
Greater string elongation of
configuration (a) allows it to
absorb more energy and
return it to the ball as
energy which is
transformed into rotational
velocity.
Un-deformed string
Deformed string
(a) (b)
Figure 10.10: String deformation during impact
This also explains why the conventional configuration produces more spin, since
having the main strings normal to the flight path of the ball provides the optimum
opportunity for the type of string deformation required for spin generation.
10.3.3 Different string gauges
The results of the analysis can be viewed in Figure 10.11. As expected, there is a
slight reduction in spin for the largest diameter (1.38 millimetres), due to the increase
in stiffness it creates. The minimal difference, however, makes it difficult to ascertain
whether the increased surface area of a larger string diameter will have any effect.
Even if one could ascertain the effect a larger diameter has on the difference in spin
generation, the observed effect would likely be so negligible that it would be of little
worth.
222
4500
4400
4300

4200
4100
4000
3900
3800
3700
3600
3500
1.25 1.3 1.38
String guage (mm)
Figure 10.11: Spin rates for different string gauges
10.3.4 Different string material
The results of this analysis can be seen in Figure 10.12. As one would expect, the
stiffest string material, the titanium polymer, gives the lowest level of spin. This is
because, as previously stated, a stiffer string bed leads to higher levels of ball
deformation which, in turn leads to a higher level of energy loss. Conversely, a lower
level of ball deformation occurs on the synthetic string bed and, as a result, increased
levels of spin occur.
223
4150
4100

4050
4000
3950
3900
3850
3800
Titanium Poly Synth Gut
String Material
Figure 10.12: Outbound spin rates for different string materials
In order to allow novel string-bed designs to be generated quickly and efficiently an

automated string-bed generation system was created. Using this system the spin
generation of string-beds with different string orientation, density and gauge were
examined. It was found that the most significant factor in spin generation was the
stiffness of the string-bed and that, generally, any change in spin generation for
differing string patterns is often, in part due to the resultant change in stiffness.
It was shown however, that due to an increased level of mechanical interlocking,

outbound spin tended to increase with string density. It was also shown that for novel
string-bed orientations, the outbound spin increased as the angle opposite the ball’s
flight path increased. None of the novel string-beds tested, however, resulted in
outbound spin greater than that of a conventional string bed.
224
11 Conclusions and Further Work
11.1 Conclusions
The primary objective of this study was to create an FE model of a tennis racket, for
which suitable CAD geometry of the racket was a fundamental requirement. Not only
was this achieved but a partial automation of the CAD generation was also
accomplished. This was achieved through the creation of a Microsoft Excel spread-
sheet based system which allowed the user to specify key parameters of a string-bed
before exporting the fundamental geometry data. The Excel document produces a
2D line representing the string which the user can export to the CAD system NX5.0,
whereupon they are required to create the 3D volume of the string. Some further
work to fully automate the process would, therefore, still be beneficial.
Having created and meshed the racket geometry, several strategies of tensioning the
strings whilst connected to the racket were examined. A method using contracting
“connector” elements, attached to both the racket frame and strings, was adopted as
it allowed the racket and the strings to be loaded simultaneously.
With a satisfactory finite element model of the racket in place, several static
validation procedures were developed for the model before the ball was introduced.
The racket was first validated statically, by testing the bending stiffness of the racket
frame (strung and unstrung) and the stiffness profile of string-bed. The bending
stiffness of the racket was obtained using a traditional 3 point bend test and the
bending stiffness of the strung and unstrung rackets were found to be 117
kilonewtons per metre and 129 kilonewtons per metre respectively. Furthermore, the
model was found to correlate well in both the strung and unstrung states yielding a
variation of less than 5% in both cases.
225
The stiffness profile of the string-beds were obtained by creating a plot of the
individual stiffness values for each intersection of a cross and main string. The
stiffness of each intersection was obtained using a force measurement machine,
after which the plots were created using Microsoft Excel. This was performed for
string tensions of 178 Newtons, 223 Newtons and 267 Newtons with the resulting
plots showing that as the string tension increases, so too does the variation in string
stiffness across the string bed. For the lowest string tension the stiffness was
between 20 kilonewtons per metre and 24 kilonewtons per metre for 95% of the
intersections – with only the intersections closest to the frame outside this range.
A simulation of this testing was carried out with the FE model to enable the
correlation of its string-bed stiffness profile – a method not previously used to validate
finite element model of tennis rackets. The model showed a slightly lower magnitude
of stiffness – 16 kilonewtons per metre to 19 kilonewtons per metre for the majority of
intersections – but was still of a similar enough magnitude to allow the model to be
used for dynamic impacts. To date, this technique has not been used by any other
author for correlating finite element models of tennis rackets.
The technique of photogrammatry was also used to validate the deformation of the
racket frame under the tension of the strings. The GOM Aramis system used for this
work is traditionally used in the automotive industry and has not previously been
applied to research of this kind. Two conditions were tested; the racket fully strung
and the racket strung with mains strings only. The Aramis system captured images of
the racket in its strung and unstrung state and by tracking a random speckle pattern
on the racket’s surface, was able to produce a strain contour projected onto the
image of the strung racket. This was then compared to the strain values given by the
FE model when subjected to similar string loading. The fully strung model predicted
the location and magnitude (2%-3%) of all the highest strains. The model did not
predict the higher strains experienced by the racket when strung with mains strings
only as accurately as in the fully strung racket. Although the model did predict the
location of the highest strains it predicted values of 5% as opposed to the 7% yielded
by the photogrammetry analysis.
226
In order to validate the vibrational properties of the racket 3D laser doppler
vibrometry was used. Previous modal analysis studies have been carried out for
tennis racket frames using contacting measurement techniques (Brody, 1995, Cross,
2001b and Mohanty et al., 2001). The non-contact approach used in this thesis,
however, allowed for the extraction of the natural frequencies of the string-bed as
well as the frame.
Using this technique, the lateral (131 Hertz) and torsional (353 Hertz) frequencies of
the racket frame and their respective natural frequencies were obtained. These
values were compared to those obtained from a simple analysis of the FE model and
found to have a correlation error of less than 5% in both cases. The lateral values of
the strung racket were also obtained for the racket strung at varying tensions of 178
Newtons, 223 Newtons and 267 Newtons and found to show no level of variation
higher than that which, could be attribute to experimental error (<1%). The vibration
modes of the string-bed tensioned to 178 Newtons were also acquired experimentally
(576 Hertz and 853 Hertz) but were slightly less cohesive with the finite element
model (560 Hertz and 768 Hertz), yielding an error of under 5% and 10% for first and
second modes respectively.
Having validated the model’s static performance and vibrational characteristics a

series of tests were performed to acquire and examine the properties of a variety of
different string materials at varying tensions. The strings were subjected to a number
of extension and impact tests to establish how their elongation and energy absorption
properties varied under different loading conditions. It was found that the most
consistent string tension over a variety of different loading speeds was 223 Newtons
whilst nylon displayed the highest dwell time in impact testing which, in practice
would lead to greater energy dissipation and hence a reduction in COR.
Previously, a number of authors have researched the issue of friction between the
ball and the string-bed but to the author’s knowledge a value of the friction coefficient
between the strings themselves has not been obtained. As a result a number of
227
experimental procedures were defined to establish the COF between the strings and
the ball and the strings themselves. It was shown ball/string COF varied significantly
from the string/string value, with the ball/string value producing values in the range of
0.19 to 0.53 depending on string construction, material and normal force, whilst the
string/string values were in the range of 0.12 to 0.23. A tribometer was also
developed in order to more accurately represent the scenario of a rolling ball’s
interaction with the string-bed. Using the tribometer the variation in friction for a
number of different string-bed configurations was explored and found to be at its
highest level for a densely strung, conventional string-bed.
Using the various properties obtained, a model was set-up to examine how the
mechanisms of oblique impacts changed with impact angle. It was shown that three
distinct phases; sliding, biting and rolling, exist within an oblique impact of a certain
angle range but as the impact direction moves towards normal the sliding phase
becomes less obvious and is eventually eradicated above angles of 40 degrees
(where 90 degrees is normal to the string bed).
Contour plots of the nodal forces due to element stress (NFORC) during these
oblique impacts were created. These plots showed how the nodal forces increased at
the front of the ball as the sliding phase began - with a number of nodes at the
central area of the leading edge experiencing forces as high as 66.1 Newtons. The
transfer of this energy was evident from following increments of the analysis, which
showed an immediate reduction of nodal forces at the leading edge of the ball, whilst
nodal forces at the rear of the ball increased to 55.6 Newtons. This transfer of energy
leads to the separation of the ball from the string-bed and the ultimately the initiation
of rotational velocity.
Oblique impacts of the ball impacting the string-bed were performed for varying
inbound angles (ranging from 29 degrees to 40 degrees at 29.8 metres per second)
experimentally and compared to those obtained from a similar simulation using the
FE model. Initially it was found that the FE model showed the opposite trend of the
228
experimental data, with spin decreasing as the angle tended away from normal.
Upon further inspection, it became clear that this was due to the method used to
model energy dissipation (stiffness proportional damping) over-damping the model.
As a result the energy dissipation was modelled using mass proportional damping in
the strings. The beta damping component of 0.0032 applied the ball’s rubber core
was removed and replaced with an alpha damping component of 1,000. This method
of damping yielded computational values which correlated within 10% of the
experimental values. The values of spin produced by the model varied from 4,000
revolutions per minute to 3,000 revolutions per minute for an inbound angle range of
29 degrees to 30 degrees respectively. This was compared to experimental outbound
spin values ranging from 4,200 revolutions per minute to 3,300 per minute for the
same range of inbound angles.
In the final chapter of the thesis the primary objective of modelling novel string
patterns was achieved. Using the string-bed generation technique described earlier,
a number of string-beds were created and using the properties which, had been
validated earlier in the model, a series of analyses were performed. Among the
analyses run were string-beds which varied in string pattern density (i.e. number of
strings), string gauge, string orientation and string material. In the case of the
different string pattern density and material it was found that the densest pattern (20
cross strings and 18 mains strings) and the synthetic gut strings gave the highest
level of spin, whilst the novel string orientations were shown to offer no improvement
on conventionally strung beds.
11.2 Future Work
Although this model achieved the desired goal, in terms of predicting spin generation,
a great deal of work could still be done to reduce the time taken to perform
simulations. The most straightforward way of reducing the computation time of the
229
model would be to refine and reduce the mesh. The mesh used in this study was
chosen as it represented the geometry well and was uniform throughout the racket
and string-bed. It would be possible to reduce the number of elements in the model
by introducing a coarser mesh in less critical areas of the racket (e.g. away from the
impact zone of the string-bed and on the racket handle).
Furthermore, there are a wide variety of interaction models available in Abaqus which
could be used to model the contact properties of a tennis racket impact. The “All With
Self” model was used because of its robustness and ease of application but a less
computationally expensive model - which may be just as accurate - could be
available. Another reason for not exploring other materials models in this thesis is
that they generally require specific interaction property data, which would also give
rise to further experimental work. However, if it was deemed worthwhile, it would be
possible to acquire the necessary data, either through consulting an established
company or by setting up specific equipment such as the Tribometer detailed in
Chapter 7.
Another way in which the model could be improved is with the introduction of Hyper-
elastic material models. Currently, a linear elastic material model is used to represent
the behaviour of the strings. Although this proved effective for impact speeds of up to
30 metres per second it is likely that impacts of a higher velocity would require a non-
linear material curve. The acquisition of such materials data would require specialist
equipment, capable of testing samples at much higher strain rates than were
possible for this study. If such equipment were available then it would be a
worthwhile investment of a research student’s time to investigate how the string
properties change a higher strain rates.
The model could also be improved by introducing non-uniform materials property

assignment to the string-bed. It was shown in Section 4.3 that the stiffness profile of
the model’s string-bed, although of a similar magnitude, differed from the
experimental data. By applying a non-uniform materials assignment the user could
230
reverse-engineer the stiffness profile of the string-bed to match the experimental data
more closely.
There is also potential for further work in the area of Photogrammetry. The strain
contour plots acquired using the GOM system were not as comprehensive as they
could have been, and values were missing for some areas of the racket. By
analysing the sections of the racket which displayed the highest strains, rather than
the entire racket, it may be possible to obtain a more detailed strain contour of the
racket.
Although the values of the natural frequencies obtained using the LDV in Chapter 5
displayed a good repeatability (plus or minus 0.5 Hertz for the fundamental mode of
the racket frame) the animation of the mode shapes did display some discrepancies
at the extreme ends of the racket. These discrepancies could be removed by
performing a series of more focussed analyses on different sections of the racket and
“stitching” them together to form an animation of the whole racket.
The validity of this project has reached a good level in terms of the semi-dynamic (i.e.
static racket and moving ball) validation data, given the validation data currently
available. Spin data from an actual tennis stroke could be acquired but achieving the
repeatability which would be needed to compare the performance of different string
types and bed configurations is still unattainable at this time. Given the ever
increasing technologies available in the sports industry however, it would not be
inconceivable to perform repeatable tennis impacts against a moving racket
mimicking the motion imparted on it by a player. In fact, such a study, relating to the
foot-strike of a running shoe, already exists (Ronkainen et al., 2009) and could
provide a useful template for this work. This would be the natural progression to this
study and would yield further confidence in the spin generation predictions the model
could provide.
231
An area which could potentially affect the performance of tennis rackets - and hence
the reliability of this model - is the environmental conditions in which they are used.
The conditions in which tennis is played can differ significantly, even at the same
tournament (NDTV Sports, 2012), and the ability to predict how this will affect the
performance of a given racket configuration would be greatly beneficial. Future work
in this field could include modelling the effects of the frame and strings subjected to
varying levels of temperature and humidity and comparing the results to experimental
analysis performed within a controlled environment of similar conditions. If the model
were to be used to predict the variation in performance due to environmental
conditions it would be prudent to perform a further comprehensive validation of the
model, given the extra level of uncertainty that varying environmental conditions
would introduce. As a result, the model would need to be validated at various stages
of complexity, as in this thesis, to give confidence and understanding of how the
environment affects the racket frame and strings during play.
232
12 References
Abaqus (2007). “Abaqus Documentation (v6.7-1)”. Providence, RI, USA.: Abaqus,

Inc.
Abaqus (2010). “Abaqus Documentation (v6.10-1)”. Providence, RI, USA.: Abaqus,

Inc.
Allen, T., Goodwill, S.R. and Haake, S.J. (2008). “Experimental Validation of a Finite-
element Model of a Tennis Racket String-bed”. The Engineering of Sport 7, Vol. 1, pp
115-123. Oxford: Blackwell Science.
Allen, T., Hart, J., Spurr, J., Haake, S. J. and Goodwill, S. R. (2010a). “Validated
dynamic analysis of real sports equipment using finite element; a case study using
tennis rackets”. Procedia Engineering, Vol. 2, Iss. 2, pp 3275-3280.
Allen, T., Haake, S. and Goodwill, S. (2010b). “Effect of friction on tennis ball
impacts”. Proceedings of the Institution of Mechanical Engineers, Part P: Journal of
sports Engineering and Technology, Vol. 224 (3) 229-236.
Ashcroft, A.D.C. and Stronge, W.J. (2002): “Putting Spin on a Tennis Ball and Racket
String-Bed”. Engineering of Sport 4, pp 185-191. Oxford: Blackwell Science.
Axe, J.D., Brown, K., Shannon, K. (2002). “The Vibrational Mode Structure of a Golf
Ball”. Journal of Sports Sciences, Vol. 20, Iss. 8, pp 623-627.
233
Bao, L., Sakurai, M. and Nakazawa, M. (2003). “Relationship Between Frictional
Characteristics and the Strings Ability to Make a Tennis Ball Spin”. Textile Research
Journal, Vol. 73, No. 7, pp570-574.
Bitz-Widing, M.A., Moeinzadeh, M.H., (1989). “Tension Members”. Computers and

Structures, Vol. 33, Iss. 1, pp 233-240.
Bower, R. and Cross, R. (2003). “Player Sensitivity to Changes in String Tension in a

Tennis Racket”. Journal of Science and Medicine in Sport 6(1) pp120-131.
Bower, R. and Sinclair, P. (1999). “Tennis Racquet Stiffness and String Tension
Effects on Rebound Velocity and Angle for an Oblique Impact”. Journal of Human
Movement Studies, 37, pp 271-286.
Bower, R and Sinclair, P. (2007). “Tennis Racket Stiffness, String Tension and
Impact Velocity Effects on Post-Impact Ball Angular Velocity”. Sports Engineering,
Vol. 10, No. 2, pp 111-122.
Brody, H. (1981). “Physics of the tennis racket II: The “sweet spot””. American
Journal of Physics, Vol. 49, No. 9, pp. 816-819.
Brody (1987). “Tennis Science for Tennis Players”. Philadelphia: University of

Pennsylvania Press.
Brody, H. (1995). “How Would a Physicist Design a Tennis Racket”. Physics Today,
Vol. 48, Iss. 3, pp 26-31.
Brody, H., Cross, R., Lindsey, C. (2002). “The Physics and Technology of Tennis”.
San Diego: Racquet Tech Publishing.
234
Brooks, R., Mather, J. S. B., Knowles, S. (2006). “The Influence of Impact Vibration
Modes and Frequencies on Cricket Bat Performance”. Proceedings of the Institution
of Mechanical Engineers, Part L: Journal of Materials: Design and Applications, Vol.
220, No. 4, pp 237-248.
Carreira, D.J., Chu, K. (1985). “Stress-Strain Relationship for Plain Concrete in

Compression”. American Concrete Institute Journal Proceedings, Vol. 82, Iss. 6, pp
797-804.
Casolo, F. and Lorenzi, V. (2000). “On tennis equipment, impact simulation and
stroke precision”. Tennis Science and Technology, pp.83-90. Oxford: Blackwell
Science.
Clerici, G. (1976). “Tennis”. London: Octopus Books Ltd.
Chou, P.C., Liang, D., Yang, J. (1994). “Contact Forces, Coefficient of Restitution,
and Spin Rate of Golf Ball Impact”. Proceedings of the World Scientific Congress on
Golf. Illinois: Human Kinetics.
Cooper, J (2003). “An Evolutionary History of Tennis Rackets”.

http://tennis.about.com/od/racquetsballsstringing/a/evolmodracquet.htm. Accessed
June 2009.
Cottey, R.A. (2002). “The Modelling of Spin Generation with Particular Emphasis on
Racket Ball Games”. PhD Thesis. Loughborough University: Loughborough
University.
235
Cotton, R.T. (2008). “Surface Interactions of Soccer Balls”. PhD Thesis.
Loughborough University: Loughborough University.
Covill, D., Farr, J., Katz, T., White, D. (2008). “Use of Static Stiffness Behaviour to
Characterise Field Hockey Sticks”. The Engineering of Sport 7, Vol. 2, pp 239-246. .
Oxford: Blackwell Science.
Cross, R. (2000a). “Effects of friction between the ball and strings in tennis”. Sports
Engineering, Vol 3, Iss. 2, pp 85-97.
Cross, R. (2000b). “Dynamic Properties of Tennis Strings”. Tennis Science and

Technology, pp 119-126. Oxford: Blackwell Science.
Cross, R and Bower, R. (2001a). “Measurements of String Tension in a Tennis

Racket”. Sports Engineering 4, Vol. 3, pp 165-175.
Cross, R. (2001b). “Why Bows Get Stiffer and Racquets Get Softer When the Strings
are Added”. American Journal of Physics, Vol 69, No. 8, pp 907-910.
Cross, R. (2003). “Oblique Impact of a Tennis Ball on the Strings of a Tennis

Racket”. Sports Engineering, Vol. 6 pp 235-254.
Cross, R (2005). “Increase in Friction with Sliding Speed”. American Journal of

Physics, Vol. 73, pp. 330-339.
CSM (2012). “The Tribology Principal”. http://www.csm-

instruments.com/en/Tribometer. Accessed June, 2012.
236
Daish, C. (1972). “The Physics of Ball Games”. London: The English Universities
Press.
Dignall, R.J., Goodwill, S.R., Haake, S.J., Miller, S. (2004). “Tennis GUT- Modelling
the game”. Engineering of Sport 5, Vol. 2, pp 382-388. Oxford: Blackwell Science.
Farrand, M. (2012). Private communication. Max Farrand manages the equipment

pool at the Loughborough University Sports Technology Institute.
Goodwill, S.R. and Haake, S.J. (2002). “Why Were Spaghettis String Rackets
Banned in the Game of Tennis”. Engineering of Sport 4, pp 231-240. Oxford:
Blackwell Science.
Goodwill, S.R. and Haake, S.J. (2004b). “Effect of String Tension on the Impact
Between a Tennis Ball and Racket”. Engineering of Sport 5, Vol. 2, pp 3-9. Oxford:
Blackwell Science.
Goodwill, S.R., Douglas, J., Miller, S. Haake, S. (2006). “Measuring Ball Spin off a
Tennis Racket”. Engineering of Sport 6, Vol.1, pp 379-384. Oxford: Blackwell
Science.
GOM (2010). “Aramis Software” http://www.gom.com/3d-software/aramis-

software.html. Accessed June 2010.
GOM (2012). ”Aramis – Optical 3D Deformation Analysis”,

http://www.gom.com/metrology-systems/system-overview/aramis.html. Accessed
July 2012.
237
Groppel, J.L, Shin, I.S., Thomas, J. and Welk, G.J. (1987). “The Effects of String
Type and Tension on Impact in Midsized and Oversized Racquets”. International
Journal of Biomechanics, Vol. 3, Iss. 1, pp 40-46.
Hambli, R., Chamekh, A. and Hadj Salah, H.B. (2006). “Real-Time Deformation of
Structure Using Finite Elements and Neural Networks in Virtual Reality Applications”.
Finite Elements in Analysis and Design, Vol. 42, Iss. 11, pp 985-991.
Hatze, H. (1993). “The Relationship Between the Coefficient of Restitution and

Energy Losses in Tennis Rackets”. Journal of Applied Biomechanics, Vol. 9, Iss. 2,
pp 124-142.
Hedrick, K., Ramnath, R. and Mikic, B. (1979). “Measurement of Tennis Racket

Properties”. World Tennis, Vol. 27, No. 4, pp 78.
Hocknell, A., Mitchell, S.R., Jones, R., Rothberg, S.J. (1998). “Hollow Golf Club Head
Modal Characteristics: Determination and Impact Applications”. Journal of
Experimental Mechanics, Vol. 38, No. 2 pp 140-146.
Ivanov, D., Ivanov, S., Lomov, S., Verpoest, I., (2009). “Strain Mapping Analysis of
Textile Composites”. Optics and Lasers in Engineering, Vol. 47, Iss. 3, pp 360-370.
Jobe, F.W., Cicotti, M.G. (1994). “Lateral and Medial Epicondylitis of the Elbow”.
Journal of the American Academy of Orthopaedic Surgeons, Vol.2, No. 1, pp 1-8.
Kotze, J. (2005). “Tennis Racket Performance Studies and the Design of a Novel
Test Machine”. PhD Thesis. Loughborough University: Loughborough University.
238
Knudson, D (1991). “Effect of String Type and Tension on Ball Vertical Angle of
Rebound in Static Tennis Impacts”. Journal of Human Movement Studies, Vol. 9, Iss.
20, pp 39-47.
Knudson, D. (1993). “Effect of String Tension and Impact Location on Ball Rebound
Accuracy in Static Tennis Impacts”. Journal of Applied Biomechanics, Vol. 9, Iss. 2,
pp 143-148.
LMS Virtual Labs (2012). “LMS Virtual Labs Correlation Tools”

http://www.lmsintl.com/simulation/correlation/model-updating. Accessed June 2012.
Matsuhisa, H., Sonoda, M., Park, J.G., Utsono, H. (2004). “Energy Analysis of Tennis
Ball-Racket Collision”. Engineering of Sport 5, Vol.2, pp 17-23. Oxford: Blackwell
Science.
Merkel, R.C, Blough, T. (1999). “Dynamic Characterization and Comparison of Golf

Clubs”. IMAC XVII: International Modal Analysis Conference No. 17, Vol. 3727, No.
2, pp 513-517.
Mitchell, S. (1997). “ITF Racket Power Standard”. Technical Report Commissioned

by the ITF. Loughborough University: Loughborough University.
Mohr, S., Cottey, R., Lau, D., Gillet, C., Kotze, J., Jolly, M., Schwenger (2008).
“Dynamics of String-Bed Damper on Tennis Rackets”. The Engineering of Sport 7,
Vol. 2, pp 179-189. Oxford: Blackwell Science.
Mohanty, P. Rixen, D.J. (2002). “Measuring Sweet Spots of Tennis Rackets”. IMAC
XX: Conference on Structural Dynamics No. 20, Vol. 4753, Iss. 2, pp 1539-1545.
239
Mosse, L., Compston, P., Cantwell, W. J., Cardew-Hall, M., Kalyanasundaram, S.,
(2006). “The development of a finite element model for simulating the stamp forming
of fibre–metal laminates”. Composite Structures, Vol. 75, Iss. 1, pp 298-304.
NDTV Sports (2012). “Wimbledon Roof Plays into Djokavic’s Hands”,

http://sports.ndtv.com/wimbledon2012/news/item/192946-wimbledon-roof-plays-into-
novak-djokovics-hands. Accessed July 2012.
Noble, L., Walker, H. (1994). “Baseball Bat Inertial and Vibrational Characteristics
and Discomfort Following Ball-Bat Impacts”. Journal of Applied Biomechanics, Vol
10, Iss. 2, pp 132-144.
Penrose, J.M.T., Hose, D.R. (1999). “An impact analysis of a flexible bat using and
iterative solver”. Journal of Sports Sciences, Vol. 17, No. 8 pp 677-682.
Photron (2012) Private communication. This information was obtained through a

private communication as a result of an enquiry made to Photron via their web-site.
Pluim, B.M. (2000). “Rackets, Strings and Balls in Relation to Tennis Elbow”. Tennis
Science and Technology, pp 321-332.
Racket Tech Publishing (2005). “Racket Tech Magazine”. San Diego: Racket Tech
Publishing.
Reddy, J. N. (2007). “Non-linear finite element analysis”. Oxford: Oxford University

Press.
240
Rockey, K., Evans, H., Griffiths, D., Nethercot, D. (1983). “The Finite Element
Method: A Basic Introduction for Engineers”. St. Albans: Grenada Publishing.
Roetert, E, Brody, H., Dillman, C., Groppel, J., Schulteis, J. (1995). “The
Biomechanics of Tennis Elbow: An Integrated Approach”. Clinical Sports Medicine,
Vol. 14, No. 1, pp 47-57.
Ronkainen, J., Harland, A. (2007). “Soccer Ball Modal Analysis Using a Scanning
Laser Doppler Vibrometer”. Journal of Sports Engineering, Vol. 10, No.1, pp 49-54.
Ronkainen, J.A., Roberts, M., El Kati, R.F.M., Fleming, P.R. and Forrester, S.E.
(2009). “Robotic Simulation of the Ground Contact Phase of Forefoot and Heelstrike
Running”. Footwear Science, Vol.1, Supplement 1, pp.3-5.
Schwarz, B. J. and Richardson, M. H., (1999). “Experimental Modal Analysis”.

Proceedings of the CSI Reliability Week, Orlando, Forida,
Singh, J. (2012). Private communication. Mr. Singh is the technician in charge of the
co-ordinate measuring machine at Loughborough University’s Wolfson School.
Sissler, L., Jones, R., Leaney, P.G. and Harland, A. (2010). “Viscoelcastic Modelling
of Tennis Ball Properties”. IOP Conference Series: Materials Science and
Engineering, Vol. 10, No. 1, pp 1-9.
Stolk, J., Verdonschot, N., Critofolini, L., Toni, A., Huiskes, R. (2002). “Finite element
and experimental models of cemented hip joint reconstructions can produce similar
bone and cement strains in pre-clinical tests”. Journal of Biomechanics, Vol. 35, Iss.
4, pp 499-510.
241
Thomas, G., Deiters, T., Best, C. (1995). “Simulating Golf Club Performance Using
Modal Analysis”. International Modal Analysis Conference 13 Proceedings, Vol.
2460, pp 989-995.
Vollmer, D., Meyer, U., Joos, U., Vegh, A., Piffko, J., (2000). “Experimental and ﬁnite
element study of a human mandible”. Journal of Cranio-Maxillofacial Surgery, Vol.
28, 91–96.
Wiezel, A. and Fajitis, A. (2004). “Calculation of Dwell Time for Tennis Rackets”.
Engineering of Sport 5, Vol.2, pp347-353. Oxford: Blackwell Science.
Woodward, Angela (2010). “Tennis Racket”, http://www.enotes.com/how-products-

encyclopedia/tennis-racket. Accessed June 2010.
242
13 Appendices
13.1 Appendix 1: ITF regulation for tennis equipment
a. The hitting surface, defined as the main area of the stringing pattern bordered by
the points of entry of the strings into the frame or points of contact of the strings with
the frame, whichever is the smaller, shall be flat and consist of a pattern of crossed
strings connected to a frame and alternately interlaced or bonded where they cross.
The stringing pattern must be generally uniform and, in particular, not less dense in
the centre than in any other area. The racket shall be designed and strung such that
the playing characteristics are identical on both faces. The racket shall be free of
attached objects, protrusions and devices other than those utilised solely and
specifically to limit or prevent wear and tear or vibration or, for the frame only, to
distribute weight. These objects, protrusions and devices must be reasonable in size
and placement for such purposes.
b. The frame of the racket shall not exceed 29.0 inches (73.7 cm) in overall length,
including the handle. The frame of the racket shall not exceed 12.5 inches (31.7 cm)
in overall width. The hitting surface shall not exceed 15.5 inches (39.4 cm) in overall
length, and 11.5 inches (29.2 cm) in overall width.
c. The frame, including the handle, and the strings, shall be free of any device which
makes it possible to change materially the shape of the racket, or to change the
weight distribution in the direction of the longitudinal axis of the racket which would
alter the swing moment of inertia, or to change deliberately any physical property
243
which may affect the performance of the racket during the playing of a point. No
energy source that in any way changes or affects the playing characteristics of a
racket may be built into or attached to a racket.
244
13.2 Appendix 2: GOM Aramis technical data (GOM, 2012)
System Configurations High Speed Camera
Frame Rate (Hz) up to 500 (4000)
Camera Resolution (pixel) 1280 x 1024
Measuring Area mm² to > m²
Strain Measuring Range (%) 0.01 up to > 100
Strain Measuring Accuracy (%) up to 0.01
Image Memory uses PC RAM
Tool Free Mounting No
Positioning
1 or 3
Pointers
Illumination external
High-End PC Yes
Notebook N/A
Control Device Sensor Controller
Operating Temperature 5-40°C
Specimen Temperature typ. -100°C up to +1500°C
245
13.3 Appendix 3: Abaqus error message
“The Abaqus/Explicit restart job cannot proceed due to a violation of the domain
decomposition of the original analysis. There are one or more nodes involved in constraint
that were shared nodes in the original analysis but are now part of an implicit constraint
system and hence are non-shared nodes in the restart analysis. This could be due to
application of boundary conditions on these nodes in the restart step. A dummy step may be
introduced in the original analysis with the same set of boundary conditions on these nodes as
defined in the restart step, to prevent this error. A node set name
“ErrNodeDomainDecompReStrt” has been created for use in Abaqus/Viewer to identify these
nodes.”
246

288381527

Uploaded by

Copyright:

Available Formats

288381527

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

288381527

Uploaded by

Copyright:

Available Formats

This item was submitted to Loughborough University as a PhD thesis by the

author and is made available in the Institutional Repository

For the full text of this licence, please go to:

David Weir MEng

A Doctoral Thesis submitted in partial fulfilment of the requirements for

© David Weir 2012

Cover Page .................................................................................................................. i

α = inbound angle (degrees)

αi = angle between points C and P (degrees)

αT = coefficient of thermal expansion

β = racket tilt angle (degrees)

ξ = critical coefficient of damping

μs = coefficient of sliding friction

μr = coefficient of rolling friction

Φ = angle 1 between cross and main strings (degrees)

φ = angle 2 between cross and main strings (degrees)

ωn = natural frequency (Hz)

ω1 = inbound spin (rev/min)

ω2 = outbound spin (rev/min)

λ = rebound angle (degrees)

A = cross-sectional area (m2)

COF = coefficient of friction

COP = centre of percussion

COR = coefficient of restitution

Cyi = y co-ordinate of point C at time increment i

E = Young’s modulus (N/m2)

Ei = internal Young’s modulus (N/m2)

fi = internal Force (N)

kb = ball stiffness (N/m)

ks = string stiffness (N/m)

KEi = elastic component of the stiffness matrix

KGi = geometric component of the stiffness matrix

li = elongated length (m)

l0i = original length (m)

me = effective mass (kg)

mb = mass of the ball (kg)

ms = mass of the strings (kg)

MOI = moment of inertia (kg•m2)

Pxi = x co-ordinate of point P at time increment i

Pyi = y co-ordinate of point P at time increment i

R = normal force (N)

td = dwell time (s)

ti = time at increment I (s)

vb = velocity of the ball (m/s)

Coefficient of restitution – ratio of outbound velocity to inbound velocity.

Racket – A strung racket frame

Spin – The angular velocity of the ball about its centre

String-bed – The woven section of strings within the racket frame

String-bed density – The number/spacing of the strings within the string-bed

String-gauge – the cross-sectional diameter of the string

1.1 History of tennis

Figure 1.1: Annotated image of a modern over-sized racket

The increases in head-size was well-received by both professionals and amateurs as

To attribute the increase in performance solely to improvements in technology would

tennis string-beds through properties

finite element modelling Obtaining a greater

Table 1.1: Project Goals

Chapter 6 outlines the approach used to obtain a comprehensive materials

In order to manipulate string-bed configurations such that they will generate an

2.1 Finite element modelling