Monopile Foundations Under Complex Cyclic Lateral Loading

Monopile foundations under complex
cyclic lateral loading
Iona A. Richards
St. Peter’s College
University of Oxford
A thesis submitted for the degree of

Doctor of Philosophy
Trinity 2019
Abstract
Monopile foundations under complex cyclic lateral loading
Iona A. Richards
St Peter’s College, University of Oxford
A thesis submitted for the degree of Doctor of Philosophy
Trinity 2019
The vast majority of offshore wind turbines are supported on monopile

foundations. These foundations experience lateral loading which is both cyclic and
complex: continuously varying in amplitude, direction, frequency and eccentricity
with the environmental and turbine operation conditions. The effect of cyclic loading
on the accumulated rotation (ratcheting), strength and stiffness of the foundation must
be considered, while advanced structural dynamic modelling requires prediction of the
cyclic hysteretic response. However, there are currently no commonly accepted design
methods for predicting the cyclic response of monopile foundations, particularly under
the complex loading to which they are exposed.
This thesis explores the response of monopile foundations to cyclic lateral loading
through laboratory-scale physical modelling in dry sand, principally at 1g. The
development of novel laboratory apparatus allows the application of complex,
continuously varying cyclic loading. Regular, unidirectional tests complement previous
studies and explore the impact of load amplitude and asymmetry in detail. Regular,
multidirectional tests provide novel insight into the response to loading with multiple
direction components. And irregular, multi-amplitude and multidirectional tests reveal
the response to realistic loading, highlighting the dominance of large load events.
Throughout, the monopile response is characterised in terms of ratcheting, evolution
of hysteresis loop shape and the post-cyclic reloading response. Behaviour in very
loose and dense sand is found to be qualitatively similar.
A complementary study using centrifuge modelling explores the impact of stress-
level on the monopile response, to establish the relevance of 1g modelling and inform
comparison of monopile responses observed at different stress-levels. Qualitatively
similar behaviour is observed at multiple stress-levels, although the rate of ratcheting
and stiffening reduce with increasing stress-level.
Together, the physical modelling results inform the design of monopile foundations
under cyclic lateral loading. In particular, the results facilitate demonstration and
development of numerical models in the Hyperplastic Accelerated Ratcheting Model
(HARM) framework, which can capture ratcheting and evolution of the hysteretic
response, and can respond to arbitrary loading. Models in the HARM framework are
shown to capture key features of the response to complex — irregular, multi-amplitude
and multidirectional — cyclic loading in dry sand of various densities and at multiple
stress-levels. The results build confidence in the application of models in the HARM
framework for full-scale monopile design.
Acknowledgements
First and foremost, I would like to thank my supervisors Byron Byrne and Guy Houlsby.
This thesis would not have been possible without their careful guidance and
enthusiasm. In particular, I thank Guy for patiently accommodating many an
impromptu meeting, and Byron for the encouragement over many years, from my time
at Catz as an undergraduate. Special thanks also go to Clive Baker. From machining
bespoke components to advising on the intricacies of lifting equipment, Clive’s
technical expertise, support and good humour facilitated all aspects of the
experimental work at Oxford.
It has been a pleasure to be part of the Civil Engineering Group at Oxford,
surrounded by so many cheerful, helpful people. In particular, my gratitude goes to
Christelle Abadie, whose support and generous advice was invaluable in the
development of this project, and to Toby Balaam for numerous fruitful discussions,
particularly regarding our cyclic definitions work. I am also grateful to Tom Adcock and
Mark McAllister for facilitating access to wave loading data from the DeRisk project,
and providing much-appreciated technical input. For advice and assistance in the lab
and beyond, I thank Jonathan White, Russell Mayall, Brian Sheil and Ross McAdam;
and for all the help with purchasing, I thank Alison May. Special thanks must also go to
the REMS contingent and the ever-evolving Wallis room lunch group, for all the laughs,
and to the civil climbing group, for the support on and off the wall.
At UWA, I am very grateful to Fraser Bransby and Christophe Gaudin for supporting
my visit in 2018, and for many insightful discussions. The experimental work at UWA
would not have been possible without the support of the excellent technical team, led
by John Breen; thank you all for going the extra mile. I am also grateful to Manuel
Herduin for allowing me to use his laboratory apparatus, and to Juliano Nietiedt for the
assistance in the laboratory (as well as all the lifts to the office). I must also thank
everyone at UWA who made me so welcome, and especially those who showed me a
little of Western Australia.
Beyond the Jenkin building (Oxford) and the IOMRC (Perth), I thank Liz and Nigel
for providing a home-from-home during my visits to Cranfield, Em and Freddie for the
adventures, and Renée for all the support. Fearghus, thanks for being my (often rather
distant) rock. Finally, I thank my parents and Al for, ultimately, making this possible.
Contents
1 Introduction 1
1.1 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Offshore wind power . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Offshore wind turbine support structures . . . . . . . . . . . . . . 2
1.2 Design of monopile foundations . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Illustrative monopile response . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Design requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.4 Monotonic design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.5 Cyclic design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Relevant research contributions . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Physical modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Mechanistic studies . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.3 Proposed cyclic design methods . . . . . . . . . . . . . . . . . . . 22
1.4 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.1 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.2 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2 Design of novel laboratory apparatus for 1g monopile testing 29

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Model pile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Loading and pile measurement apparatus . . . . . . . . . . . . . . . . . . 31
2.3.1 Design overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.2 Tank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3.3 Pile cap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3.4 Actuators and load application . . . . . . . . . . . . . . . . . . . . 34
2.3.5 Position measurement . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 Data acquisition, kinematics and control software . . . . . . . . . . . . . 36
2.4.1 Software overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.2 Actuator communication and data acquisition . . . . . . . . . . . 37
2.4.3 Kinematics calculations . . . . . . . . . . . . . . . . . . . . . . . . 37
iv
Contents v
2.4.4 Adjustment for transducer friction . . . . . . . . . . . . . . . . . . 40

2.4.5 Load control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.4.6 Data logging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 Sample preparation apparatus . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6 Pile installation apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3 1g test procedure, monotonic responses and cyclic definitions 46

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2 Sand sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.1 Sand properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.2 Sample preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.3 Impact of stress-level . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.4 Cone penetration tests . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3 Set-up and test procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4 Monotonic response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.4.1 Moment-rotation response . . . . . . . . . . . . . . . . . . . . . . 55
3.4.2 Maximum stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4.3 Normalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.5 Definitions for cyclic loading and the cyclic response . . . . . . . . . . . . 60
3.5.1 Cycle definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.5.2 Cyclic load characterisation . . . . . . . . . . . . . . . . . . . . . . 62
3.5.3 Ratcheting definition . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.4 Secant stiffness definition . . . . . . . . . . . . . . . . . . . . . . . 64
3.5.5 Energy dissipation definition . . . . . . . . . . . . . . . . . . . . . 64
3.5.6 Application to multi-amplitude loading . . . . . . . . . . . . . . . 67
3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4 Regular cyclic loading response at 1g 69

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Test programme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 Unidirectional hysteretic response . . . . . . . . . . . . . . . . . . . . . . 74
4.4 Unidirectional high cycle response . . . . . . . . . . . . . . . . . . . . . . 77
4.4.1 Ratcheting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.4.2 Secant stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.3 Energy dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4.4 Uplift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.5 Multidirectional hysteretic response . . . . . . . . . . . . . . . . . . . . . 91
4.6 Multidirectional high cycle response . . . . . . . . . . . . . . . . . . . . . 92
4.6.1 Perpendicular cyclic loading . . . . . . . . . . . . . . . . . . . . . . 92
Contents vi
4.6.2 Fan-type cyclic loading . . . . . . . . . . . . . . . . . . . . . . . . . 101

4.7 Post-cyclic reloading response . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5 Application of realistic storm loading at 1g 113

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.2 DeRisk wave tank tests (Bredmose et al., 2016) . . . . . . . . . . . . . . . 113
5.3 Wave load processing and test programme . . . . . . . . . . . . . . . . . . 115
5.3.1 Application of transfer function . . . . . . . . . . . . . . . . . . . . 115
5.3.2 Addition of wind loading . . . . . . . . . . . . . . . . . . . . . . . . 117
5.3.3 Projection of loads to constant eccentricity . . . . . . . . . . . . . 118
5.3.4 Scaling wave loads . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.3.5 Test programme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.4 Storm loading response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.4.1 General observations . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.4.2 Impact of wind loading . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.4.3 Impact of storm multidirectionality . . . . . . . . . . . . . . . . . 123
5.4.4 Application of consecutive storms . . . . . . . . . . . . . . . . . . 125
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6 Investigating the effect of stress-level 130

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.2 Experimental set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.2.1 Model pile, loading and instrumentation . . . . . . . . . . . . . . 133
6.2.2 Sample preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.2.3 Test procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.2.4 Resolution of pile displacement and applied loads . . . . . . . . . 137
6.3 Test programme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.4 Monotonic response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.4.1 Normalisation approaches . . . . . . . . . . . . . . . . . . . . . . . 141
6.4.2 Maximum stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.5 Unidirectional cyclic response . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.5.1 Load-displacement response . . . . . . . . . . . . . . . . . . . . . 144
6.5.2 Ratcheting response . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.5.3 Secant stiffness response . . . . . . . . . . . . . . . . . . . . . . . . 148
6.5.4 Energy dissipation response . . . . . . . . . . . . . . . . . . . . . . 150
6.5.5 Reloading response . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.5.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.6 Multidirectional cyclic response . . . . . . . . . . . . . . . . . . . . . . . . 154
6.6.1 Displacement and ratcheting response . . . . . . . . . . . . . . . . 154
Contents vii
6.6.2 Secant stiffness response . . . . . . . . . . . . . . . . . . . . . . . . 155

6.6.3 Energy dissipation response . . . . . . . . . . . . . . . . . . . . . . 157
6.6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7 Modelling the response of a monopile to complex cyclic loading 162

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.2 Key behaviours observed experimentally . . . . . . . . . . . . . . . . . . . 163
7.3 Modelling basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.3.1 Hyperplasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.3.2 Multi-surface kinematic hardening models . . . . . . . . . . . . . 165
7.3.3 HARM framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
7.3.4 Modelling choices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.4 Capturing the hysteretic response . . . . . . . . . . . . . . . . . . . . . . . 168
7.4.1 Unidirectional hyperplastic kinematic hardening model . . . . . 169
7.4.2 Bi-directional hyperplastic kinematic hardening model . . . . . . 170
7.4.3 Calibration of kinematic hardening models . . . . . . . . . . . . . 171
7.4.4 Computation of hysteretic response . . . . . . . . . . . . . . . . . 175
7.5 Capturing the high-cycle response . . . . . . . . . . . . . . . . . . . . . . 177
7.5.1 Unidirectional hyperplastic ratcheting model . . . . . . . . . . . . 178
7.5.2 Bi-directional hyperplastic ratcheting model . . . . . . . . . . . . 179
7.5.3 Calibration of ratcheting models . . . . . . . . . . . . . . . . . . . 181
7.5.4 Evolution functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.5.5 Evolution function parameters . . . . . . . . . . . . . . . . . . . . 188
7.5.6 Computation of regular cyclic loading response . . . . . . . . . . 192
7.5.7 Computation of response to storm loading . . . . . . . . . . . . . 200
7.6 Prediction of prototype-scale response to storm loading . . . . . . . . . . 202
7.6.1 Generation of prototype-scale model . . . . . . . . . . . . . . . . . 203
7.6.2 Response to storm loading . . . . . . . . . . . . . . . . . . . . . . . 204
7.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
8 Conclusions 209
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.2 Key contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.3 Key implications for design . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.4 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
8.5 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
References 216
Nomenclature
Greek alphabet variables
Empirical value in dilatancy expressions (Bolton, 1986; Bolton, 1987)

α Internal variables (hyperplasticity)
Ratcheting power-law exponent
αr Ratcheting strain (HARM)
Scalar hardening parameter (HARM)
β
Secant stiffness power-law exponent
Energy loss power-law exponent
γ
Shear strain
γ0 Soil effective unit weight
Strain (may represent pile rotation θ or pile displacement u for macro pile
ε
response)
εa Axial strain
ζb Cyclic load magnitude characterisation ζb = σe /σR
ζc Cyclic load asymmetry characterisation ζc = σr /σe
Energy loss factor
η Pile aspect ratio
Stress-dependent stiffness exponent
θ Pile rotation
κ Increase in surface strengths Km (HARM evolution functions)
κ∗g Parameter for evolution of Km (HARM evolution functions)
κ∗m Parameter for evolution of Km (HARM evolution functions)
λ Plastic multiplier (plasticity)
ξ Damping ratio (structural dynamics)
ρκ Parameter for evolution of Km (HARM evolution functions)
τ Shear stress
Stress (may represent moment M or horizontal load H applied to pile for
σ
macro pile response)
σ0 Vertical effective stress
σc0 Effective confining stress
viii
Contents ix
0
σREF 0
Reference vertical effective stress σREF = 0.7Lγ 0
φ0 Angle of friction
φ0c Critical angle of friction
φ0p Peak angle of friction
Half internal spread angle (fan-type loading)
Φ Standard deviation of wrapped normal spreading distribution (wave
loading, e.g. Adcock and Taylor, 2009)
χ Generalised stress (hyperplasticity)
χ̄ Dissipative generalised stress (hyperplasticity)
ψ0 Angle of dilation
ω Angular velocity (centrifuge)
Latin alphabet variables
an Point on cyclic response at mean stress on loading (cyclic definitions)

Area beneath backbone curve (HARM evolution functions)
A
Ratcheting power-law coefficient
bn Point on cyclic response at mean stress on unloading (cyclic definitions)
B Secant stiffness power-law coefficient
BL Secant stiffness logarithmic coefficient
c Constraint (hyperplasticity)
cv Coefficient of consolidation
C Energy loss power-law coefficient
Dissipation function (hyperplasticity)
d
Pile pivot point depth
d10 , d30 ,
Soil particle sizes
d50 , d60
Parameter for backbone correction (HARM calibration)
D
Pile outside diameter
DR Relative density (%)
Dt Tank diameter
Error in process variable (control)
e
Soil void ratio
ẽ Dimensionless loading eccentricity
en Point on cyclic response at extreme stress (cyclic definitions)
Parameter for backbone correction (HARM calibration)
E
Young’s modulus
EE Elastic energy (per cycle)
EH Hysteretic energy loss (per cycle)
Contents x
Ei Initial stiffness (monotonic response)

Ep Ip Pile bending stiffness
ESL Soil’s Young’s modulus
Frequency
f Helmholtz free energy function (hyperplasticity)
Ratio of prototype-scale to laboratory-scale parameter
f0 Natural frequency (structural dynamics)
fL Data logging frequency
F Friction (in displacement transducers)
Fβ ∗ Factor (HARM evolution functions)
FE Excitation force (structural dynamics)
FT Transmitted force (structural dynamics)
Acceleration due to gravity
g
Gibbs free energy (hyperplasticity)
GM AX Maximum shear modulus
GS Soil specific gravity
Drop height (sand raining)
h
Pile loading eccentricity
ht Tank depth
Horizontal load applied to pile
H
Surface modulus (kinematic hardening model)
HB Horizontal response at pile base (PISA monotonic method)
k Secant stiffness
kM AX Maximum stiffness (not secant)
Gain (control)
K
Surface strength (kinematic hardening model)
Kb , Kc Empirical functions for secant stiffening
Kp Rankine’s passive earth pressure coefficient
KR Pile relative stiffness (Poulos and Hull, 1989)
li Magnitude of linkage vector Li (kinematics)
L Pile embedded length
Li Linkage vector (kinematics)
Distributed moment soil reaction (PISA monotonic method)
m Pile mass
Surface index (kinematic hardening model)
mkg Parameter for evolution of Km (HARM evolution functions)
mkm Parameter for evolution of Km (HARM evolution functions)
mr Parameter for evolution of R (HARM evolution functions)
Contents xi
ms Parameter for evolution of R (HARM evolution functions)

mσ Power-law exponent for variation of Tb with ζb
Maximum number of surfaces (kinematic hardening model)
M
Moment applied to pile (at mudline)
MB Moment response at pile base (PISA monotonic method)
Cycle number
n Experimental g-level
Number of transducers aligned with loading direction
N Maximum cycle number
p Distributed lateral soil reaction (p-y and PISA monotonic methods)
p0 Mean effective stress
pa Atmospheric pressure
P Normalised excess pore pressure (Li et al., 2019)
Pc Monotonic conic function parameter
Pi Coordinates of pivot point on pile (kinematics)
Pj Monotonic Jeanjean parameter
Pp Monotonic power-law parameter
q Deviator stress
qc CPT cone resistance
Empirical value in dilatancy expression (Bolton, 1986)
Q
Flow rate (sand raining)
rn Point on cyclic response at reversal stress (cyclic definitions)
Empirical value in dilatancy expression (Bolton, 1986)
R
Ratcheting parameter (HARM)
R0 Parameter for evolution of R (HARM evolution functions)
Ra Pile roughness
Re Centrifuge effective radius
Rn Normalised roughness (Ra /d50 )
∗
RR Parameter for R (HARM evolution functions)
s Pile pose (kinematics)
Pile wall thickness
t
Time
tp Wave loading period
T̂ Transformation matrix (kinematics)
Tb , Tc Empirical functions for ratcheting response
Td Derivative time (control)
Ti Integral time (control)
Contents xii
Ti Coordinates of pivot point on transducer (kinematics)

TR
i Coordinates of pivot point on retracted transducer (kinematics)
Tp Normalised wave loading period (Li et al., 2019)
Controller output (control)
u Excess pore pressure
Pile displacement (at load application point)
V Vertical (axial) force on pile
V Vector of linkage lengths (kinematics)
W Work done on pile-soil system
Lateral pile displacement (p-y and PISA monotonic methods)
y
Yield surface (hyperplasticity)
z Vertical pile displacement
Common subscripts, superscripts, prefixes and diacritics
Ẋ Time derivative of X
X̄ Mean value of X
X̃ Dimensionless parameter (Leblanc et al., 2010a)
∆X Accumulated value of X (ε, u or θ relative to monotonic response)
Indication of special case of X
X∗
Specific value of X (HARM evolution functions)
X0 Initial value of X
XAV Average value of X
Xc Common value of X
XCY C Cyclic amplitude of X
Error value
Xe Experimentally-determined value
Extreme value of X (cyclic definitions)
XH Relating to horizontal load H
Indicating parameter is defined in both the x- and y-direction
Xi
General index
Xl Value of X on cyclic loading
XL At laboratory-scale
Xm Surface index (kinematic hardening model)
Xm0 Value of X at σe on cycle n = 0
XM Relating to moment M
XM AX Maximum value of X
XM IN Minimum value of X
Contents xiii
Xn Value of X at cycle n
Xp Peak value of X
XP At prototype-scale
Parameter corresponding to ratcheting (hyperplasticity)
Xr
Reversal value of X (cyclic definitions)
Xr0 Value of X at onset of reloading
XR Reference value of X
Relating to pile displacement u
Xu
Value of X on cyclic unloading
Xx Relating to x direction
Xy Relating to y direction
Test name abbreviations
L Very loose sand (OU)

D Dense sand (OU)
1G 1g (UWA)
9G 9g (UWA)
80G 80g (UWA)
M Monotonic loading
TW Two-way unidirectional cyclic loading (ζc = −1)
OW One-way unidirectional cyclic loading (ζc = 0)
PT Partial two-way unidirectional cyclic loading (−1 ≤ ζc ≤ 0)
PO Partial one-way unidirectional cyclic loading (0 ≤ ζc ≤ 1)
SP Multidirectional spiral loading
T T-shaped multidirectional cyclic loading
L L-shaped multidirectional cyclic loading
TH High amplitude T-shaped cyclic loading
F Fan-type multidirectional cyclic loading
UD Unidirectional irregular storm loading
MD Multidirectional irregular storm loading
Other common abbreviations
CPT Cone Penetration Test

FE Finite Element
FLS Fatigue Limit State
Contents xiv
HARM Hyperplastic Accelerated Ratcheting Model

OU Oxford University
OWT Offshore Wind Turbine
RNA Rotor Nacelle Assembly
SF Superfine (silica sand)
SLS Serviceability Limit State
ULS Ultimate Limit State
UWA The University of Western Australia
YLB Yellow Leighton Buzzard (sand)
Publications
Richards, I. A., Byrne, B. W. and Houlsby G. T. (2018). “Physical modelling of monopile

foundations under variable cyclic lateral loading”. Proceedings of the 9th International
Conference on Physical Modelling in Geotechnics (ICPMG). London, UK, pp. 737–742.
Richards, I. A., Houlsby G. T. and Byrne, B. W. (2019). “Exploring the response of a

monopile to storm loading”. Proceedings of Coastal Structures 2019. Hannover,
Germany.
Richards, I. A., Byrne, B. W. and Houlsby G. T. (2019). “Monopile rotation under

complex cyclic lateral loading in sand”. Géotechnique (Ahead of Print). DOI:
10.1680/jgeot.18.p.302.
Richards, I. A., Bransby, M. F., Byrne, B. W., Gaudin, C. and Houlsby G. T. (2019). “The
effect of stress-level on the response of a model monopile to cyclic lateral loading in
sand”. Submitted.
Chapter 1
Introduction
1.1 Application
1.1.1 Offshore wind power
The historic Paris Agreement (United Nations, 2015), signed by 196 countries, has
the long-term goal of limiting the global average temperature rise to below 2° above
pre-industrial levels, and to pursue efforts to limit the increase to 1.5°. To achieve
this goal, countries have pledged to reach “net-zero” emissions by the second half
of the century. An urgent, global transition from fossil-fuels to renewable energy is
therefore necessary (EWEA, 2015).
Offshore wind power is playing a significant role in this energy transition. Figure
1.1 shows how offshore wind installations in Europe have grown significantly over the
last decade. The UK currently has the largest offshore wind capacity with over 8 GW
installed (Wind Europe, 2019), supplying almost 10% of the UK electricity demand
(Renewable UK, 2019). Growth of the industry is expected to continue worldwide, with
70 GW installed capacity expected in Europe by 2030 (Wind Europe, 2017), and markets
in China, Japan, Taiwan and the USA expected to grow rapidly. Indeed, IEA (2017)
suggests that offshore wind could represent 4.5% of the global electricity output by 2050.
Compared to onshore wind, offshore wind takes advantage of higher and more
consistent wind speeds offshore. With fewer logistical and planning constraints, larger
wind farms made up of bigger, more efficient turbines can also be developed. Offshore
Wind Turbines (OWTs) have grown in size significantly over the last decade, as shown in
1
1. Introduction 2
Figure 1.1: Annual offshore wind installations by country and cumulative capacity (Wind
Europe, 2019)
Figure 1.2. This trend is expected to continue, with a 12 MW capacity turbine with 220 m
rotor diameter already in the late stages of development (GE Renewable Energy, 2019).
The growth of the offshore wind industry has been coupled with a significant
reduction in the cost. Economies of scale have been realised as farms and turbines
have increased in size, and greater wind speeds have been exploited by developing
further offshore, typically in deeper water (Wind Europe, 2018). Technological
advancements, including in foundation design, have also contributed to
cost-reduction. In the UK, the cost of offshore wind has reduced by 50% since 2015,
and is now cheaper than new gas or nuclear power (Renewable UK, 2019). Meanwhile,
in Germany and the Netherlands the latest offshore wind auctions were won at
“zero-subsidy” (Windpower, 2019).
1.1.2 Offshore wind turbine support structures
The components of an OWT required for electricity generation and control operations
are contained within the Rotor Nacelle Assembly (RNA), which is almost universally
supported atop a tubular steel tower. However, there is some variation in the choice of
support structure (and foundation) for the RNA and tower beneath the water-level.
Figure 1.3 illustrates a number of possible support structures for OWTs: monopiles,
1. Introduction 3
Figure 1.2: Yearly average of newly installed offshore wind turbine rated capacities in Europe
(modified from Wind Europe, 2019)
Figure 1.3: Illustration of possible support structures for OWTs
gravity bases, jacket structures on either piles or caissons, tripods and floating
spar-type structures.
The vast majority of OWTs are supported on monopiles — large-diameter open-
ended steel piles. In Europe, monopiles represent 81.5% of installed OWT support
structures, while jackets, gravity bases, tripod foundations and floating structures
represent 8%, 6%, 4% and 0.3% of installed structures respectively (Wind Europe, 2019).
With support structures accounting for approximately 20% of capital expenditure for
an offshore wind farm (EWEA, 2009), their selection and design is critical.
The current prevalence of monopiles is explained by their simple design, established
supply chain and robust installation procedures (typically impact driven). Although
1. Introduction 4
initially considered to be a solution only for shallow water sites, the proven performance
of these foundations, coupled with ongoing optimisation of their design, has made
monopiles an attractive option for turbines of increased size and farms in deeper water.
The dominance of the monopile is expected to continue, although the market-share
of alternative grounded support structures may increase as offshore wind expands
outside of Europe, where different soil conditions and environmental loading exist.
Meanwhile, floating OWTs present an opportunity for development of offshore wind
in deeper water (> 50 m) (IRENA, 2017).
Figure 1.4 illustrates a monopile foundation and highlights key dimensions:
embedded length L and diameter D. The monopile wall thickness t, which may vary
along the pile length, is also a key dimension. Modern monopiles have diameters up to
8 m (Sørensen et al., 2017) with length (L) to diameter (D) ratios (L/D) of 3 − 6
(Sørensen et al., 2017; Schroeder et al., 2015) and wall thicknesses around 100 mm
(Schroeder et al., 2015). The next generation of monopiles might be 10 m in diameter.
Monopiles are typically connected to the tower with a transition piece located at the
water-level, while scour protection is typically installed to minimise erosion at the
seabed. Figure 1.5 shows a fabricated monopile before installation.
1.2 Design of monopile foundations

1.2.1 Loading
Lateral loading dominates the design of monopile foundations for OWTs, in contrast
with the design of piles for conventional applications (e.g. supporting offshore jacket
structures or high-rise buildings onshore), where axial loading is the key concern. A
combination of wind, waves, current and turbine operation loads act on the OWT
structure. These loads, which also interact with the dynamic response of the structure,
result in combined moment (M ) and horizontal loading (H) on the monopile
foundation, as indicated in Figure 1.6. All load components vary temporally and
spatially; waves are usually the dominant cyclic component with a frequency of around
0.1 Hz. The combined loading experienced by the monopile is therefore cyclic and
1. Introduction 5
Figure 1.5: Fabricated monopile for Veja Mate offshore

wind farm (D = 7.8 m) (Windindustrie in Deutschland,
2016)
Figure 1.4: Schematic of an OWT

on a monopile foundation
Figure 1.6: Schematic indicating lateral loading on an OWT on a monopile foundation

1. Introduction 6
Limit state Moment amplitude /MU LS No. of cycles N
Ultimate Limit State (ULS) 1.00 1

Serviceability Limit State (SLS) 0.64 102
Fatigue Limit State (FLS) 0.39 107
Table 1.1: Example distribution of cyclic load amplitudes experienced by monopile foundation
over lifetime (Leblanc et al., 2010b)
Figure 1.7: Example 30 second moment loading on an OWT monopile in the North Sea in
idling conditions (derived from strain measurements at mudline) (Page et al., 2019)
complex: continuously varying in amplitude, direction, frequency and eccentricity
(M/H).
An example distribution of approximate cyclic moment amplitudes experienced
by a monopile foundation (relative to the Ultimate Limit State (ULS) design moment
MU LS ) is presented in Table 1.1 (following Leblanc et al., 2010b). The distribution
highlights the variation in load amplitude and the small number of larger-amplitude
load cycles experienced over the foundation lifetime. Meanwhile, Figure 1.7 shows the
moment loading experienced at the mudline of an OWT monopile (in a single direction)
over 30 seconds, in idling conditions (Page et al., 2019). This measured data shows the
type of irregular, multi-amplitude cyclic loading monopiles experience in the field.
Figure 1.8 presents the significant spatial variation of wind and waves for an
example site offshore the Netherlands. Although the spatial variation of loading is
site-specific, all sites are exposed to multidirectional environmental loading.
Wind-wave misalignment also occurs as weather systems change direction (Van
Vledder, 2013). Figure 1.9 shows, for a site in the Irish Sea, recorded wind-wave
misalignment of up to approximately 90°.

1. Introduction 7
Figure 1.9: Example wind-wave

misalignment at Walney Offshore Wind Farm
in the Irish Sea (Koukoura, 2014)
Figure 1.8: Frequency of occurrence of wave
direction and wind direction for a site
offshore the Netherlands (Bierbooms, 2003)
1.2.2 Illustrative monopile response
Monopile foundations exhibit a non-linear response under lateral loading. Figure 1.10a
illustrates a typical monopile response under continuously increasing (monotonic)
loading. From the monotonic response, the initial (maximum) stiffness kM AX and
the ultimate response at large loads are of interest. The ultimate response is often
defined at a reference rotation or displacement.
Under cyclic loading, monopiles may accumulate pile displacement u and/or pile
rotation θ, and the shape of the hysteresis loop may evolve; both behaviours are of
concern in monopile design. Figure 1.10b illustrates a typical response of a monopile
in sand to simplified (unidirectional, regular) biased cyclic loading to demonstrate
ratcheting (the accumulation of u and/or θ) and evolution of the hysteresis loop shape,
which may be characterised in terms of stiffness and energy dissipation metrics. As
discussed in Section 1.2.1, monopiles in the field are exposed to complex cyclic loading.
The response that might be expected under example irregular, multi-amplitude
loading is illustrated in Figure 1.10c.
The illustrative responses in Figure 1.10 are presented arbitrarily in terms of either
1. Introduction 8
(a) Monotonic loading (b) Regular cyclic loading
(c) Irregular multi-amplitude cyclic loading
Figure 1.10: Illustration of typical monopile response in sand under various types of
unidirectional lateral loading
moment-rotation (M −θ) or horizontal load-displacement (H −u). Both work-conjugate
pairs have been used to present results in previous work, and both pairs are used in
this thesis. An approximate relationship between the responses can be obtained by
considering the total work rate for a rigid monopile under lateral loading:
Ẇ = H ∗ u̇∗ = H u̇ + M θ̇ (1.1)
Where Figure 1.11 illustrates the terms. Assuming a small rotation angle, the total work
1. Introduction 9
H* u*
M=hH*
H=H* u
d θ
Figure 1.11: Illustration of moment M , rotation θ, horizontal load H and displacement u for
rigid monopile under lateral loading
rate can be expressed in terms of either H ∗ − u∗ or M − θ as:
d+h

∗ ∗
Ẇ = H u̇ ≈ M θ̇ (1.2)
h
Where h is the loading eccentricity and d is the depth of the point of pile rotation, which
is expected to remain approximately constant (Abadie, 2015). With d ≈ 0.7L (Abadie,
2015) and e.g. h = 2.5L, the total work rate is approximately 28% greater than that
suggested by responses presented in M − θ space.
1.2.3 Design requirements

Rotation criteria
The response to lateral loading is a crucial element of monopile design. The response is
typically specified in terms of rotation at the mudline, rather than in terms of pile
displacement. The DNVGL-ST-0126 standard Support structures for wind turbines
specifies the calculation of monopile rotation under both ULS and Serviceability Limit
State (SLS) conditions (DNV GL, 2016): the maximum rotation response under ULS
conditions shall be determined (7.6.2.5), as well as the permanent (long-term) rotation
under SLS conditions (7.6.2.7). DNVGL-ST-0126 also stipulates that the effects of cyclic
loading on the strength and stiffness under ULS and SLS conditions should be assessed
(7.4.4.4), and that the effects of pore-pressure build-up under cyclic loading should be
considered (7.4.4.2).
Monopile rotation under ULS conditions may be estimated with a monotonic design
method (discussed in Section 1.2.4) which is able to capture the ultimate response of
1. Introduction 10
the foundation, as illustrated in Figure 1.10a. The ultimate response may be defined as
at yield, or using a rotation (or displacement) criteria if yield is expected at excessively
large rotations, as is often the case for piles in sand. A ULS criteria of 2◦ pile rotation or
0.1D displacement at the mudline appears to be conventional (e.g. Byrne et al., 2017).
To predict permanent accumulated monopile rotation under SLS conditions —
where loading is cyclic — and to estimate the effect of cyclic loading on the SLS and
ULS response, cyclic design methods are required. DNVGL-ST-0126 suggests that the
allowable permanent accumulated rotation of the monopile under SLS conditions
might be 0.25°. However, it also states that these tolerances are typically derived from
visual requirements or requirements for turbine operation, and are therefore expected
to be project-specific.
Dynamic criteria
The monopile’s dynamic response is important, as it determines the dynamic response
of the combined OWT structure, and hence the structural fatigue damage (e.g. Aasen
et al., 2017). Indeed, many monopile designs may be dominated by dynamic
considerations. DNVGL-ST-0126 specifies the evaluation of dynamic load effects using
an integrated analysis of the OWT structure which includes the response of the
foundation (7.4.5.1).
To minimise dynamic load effects, the OWT structure’s natural frequencies must
avoid excitation at i) wind loading frequencies (although these are typically very low), ii)
wave loading frequencies, iii) rotor frequencies (1P), and iv) blade-passing frequencies
(3P). Monopile structures are typically designed as soft-stiff structures, with the natural
frequency of the first tower bending mode (f0 ) in the narrow band between the 1P
and 3P excitation frequencies, as indicated in Figure 1.12. As turbines increase in size
the rotor frequency (1P) reduces, increasing the intersection of the soft-stiff frequency
band and the wave frequency distribution. Non-negligible excitation from wave loading
may therefore occur, the effects of which can be minimised with the addition of active
damping (e.g. Brodersen et al., 2017).
The dynamic response may be quantified in terms of stiffness and energy
dissipation. Monopile stiffness impacts the OWT structure’s natural frequencies, while
1. Introduction 11
Figure 1.12: Typical frequency spectrum for dynamic loads on an OWT structure (after
Bhattacharya, 2014)
monopile energy dissipation contributes to the total structural dissipation, which
moderates dynamic load amplification and will inform the requirement for any active
damping. In an integrated dynamic analysis the monopile response is often
represented by a series of non-linear soil reaction curves, which capture an initial
stiffness kM AX (employing e.g. the p-y method or PISA method, discussed in Section
1.2.4). However, when implemented directly as non-linear springs, these
representations neglect energy dissipation. Single-element (macro) models may also
be used to represent the monopile response. Spring models, which include lateral,
rotational and potentially cross-coupling components (e.g. Arany et al., 2016), may be
employed with or without dissipation. Macro plasticity models have also been
developed, which are able to capture the non-linear foundation response including
dissipation (Page et al., 2018).
Additional considerations
Pile driveability is an essential design requirement. As monopiles increase in size to
support larger turbines in deeper water, larger hydraulic hammers are required and
significant noise pollution is generated, often requiring noise-mitigation. Buckling
during driving can also be a concern (e.g. Randolph, 2018). Alternative installation
methods for so-called XL monopiles include vibratory installation systems (Carbon

1. Introduction 12
Trust, 2014) and systems which drive the monopile with water rather than steel rams
(Carbon Trust, 2019).
The effect of scour is an important design consideration, and the installation of
preventative scour protection is now commonplace. The influence of scour and scour
protection on the structure’s natural frequency and lateral response should be
considered (Mayall, 2019).
Another important consideration is the ability of the supply chain to fabricate,
handle and transport the monopiles. Indeed, these practical considerations present
a very significant challenge as monopiles increase in size.
1.2.4 Monotonic design
Monotonic design methods allow prediction of the rotation (and displacement)
response under ULS conditions and are often a pre-requisite for cyclic design methods.
Also, the initial stiffness predicted by the monotonic response (kM AX ) is often a key
parameter for assessment of the structure’s fatigue life through integrated dynamic
analyses.
Monotonic design of piles under lateral loading has conventionally been conducted
using the p-y method, where the monopile is modelled as an elastic beam and the
distributed lateral response of the soil is represented by a series of independent non-
linear springs, defined by p-y curves (Reese and Van Impe, 2011). The p-y curves
for cohesionless soil presented in design standards (e.g. DNV GL, 2016; API, 2011)
originate from the work of Reese et al. (1974) based on field tests conducted by Cox et al.
(1974), and employ a continuously differentiable function proposed by Murchison and
O’Neill (1984). However, the standard p-y method and the conventional p-y curves —
developed for slender, small-diameter piles — have been found to be deficient for the
design of large-diameter monopiles with L/D < 6 (Alderlieste et al., 2011; Doherty and
Gavin, 2012). Indeed, DNVGL-ST-0126 cautions against the use of the conventional p-y
curves for piles with D > 1 m, without validation by e.g. FE analysis (F.2.4.1).
Recent work as part of the Pile Soil Analysis (PISA) project has developed a new
design methodology for monopiles under monotonic lateral loading (e.g. Burd et al.,
2019). Like the p-y method, the PISA design model is a one-dimensional (1D)
1. Introduction 13
Figure 1.13: Illustration of PISA design model (a) idealisation of the soil reaction components
acting on the pile (b) 1D finite element implementation of the model showing soil reactions
(Burd et al., 2019)
representation of the pile-soil system with the soil response represented by
independent non-linear springs. However, the PISA design model includes distributed
moment m and horizontal HB and moment MB responses at the pile base, alongside
the distributed lateral response p, as indicated in Figure 1.13. These additional soil
reaction components (m, HB , MB ) become increasingly significant as L/D is reduced
(Byrne et al., 2015). Four-parameter conic functions are used to define the soil
reactions within the PISA design model. Parameters for an exemplar sand are
presented by Burd et al. (2019), but the intention is that the parameters are calibrated
to site-specific 3D finite element (FE) analyses for application in different soil
conditions.
Although it is practical to perform a limited number of 3D FE computations for

1. Introduction 14
calibration of a design tool to a site, there is a significant advantage in performing the
majority of the design calculations with faster 1D models, particularly when location-
specific designs are required across perhaps 100 turbine locations. The PISA design
method allows for improved prediction of the monotonic response with a fast 1D model,
and has contributed to optimised design and significant capital expenditure savings,
as demonstrated by Manceau et al. (2019).
1.2.5 Cyclic design
While monotonic design methods allow prediction of the ULS response and the initial
stiffness kM AX for some dynamic calculations, cyclic design methods are necessary
to predict the following three aspects of the monopile response:
1. Permanent monopile rotation (ratcheting) under SLS conditions.
2. The effects of cyclic loading on the strength and stiffness under SLS and ULS
conditions.
3. The hysteretic response under cyclic loading for advanced dynamic modelling.
Cyclic design methods should be able to perform under realistic (irregular,
multi-amplitude and multidirectional) cyclic loading conditions, and similar to
monotonic design methods, they should be fast, given the vast number of design
calculations necessary across a wind farm.
The conventional p-y curves presented in e.g. DNVGL-ST-0126 include empirical
coefficients which reduce the p-y curve capacity under cyclic loading by a constant
factor (Reese et al., 1974). Long and Vanneste (1994) also modified static p-y curves to
account for cyclic loading, as a function of cycle number n instead. However, both
these methods are based on limited tests on slender piles, with relatively few cycles
(packets with N < 100) of high-amplitude cyclic loading applied (Cox et al., 1974; Little
and Briaud, 1988). Moreover, the degraded curves obtained with these cyclic
degradation approaches bear no resemblance to the continuous pile response
observed under cyclic loading.

1. Introduction 15
Figure 1.14 compares the response of a (model-scale) monopile in (dry) sand to
100 cycles of loading with the type of responses that may be obtained using the cyclic
degradation approaches presented by Long and Vanneste (1994) and Reese et al. (1974)
(where it is assumed that the macro response of the foundation reflects the constituent
p-y responses). Although the cyclic degradation approach of Long and Vanneste (1994)
has the ability to predict the rotation response at peak load after n cycles (an idealised
prediction is shown in Figure 1.14), neither approach can predict the permanent pile
rotation on unloading or the hysteretic response, the evolution of which is indicative of
the impact of cyclic loading on the foundation’s strength and stiffness. It is also unclear
how these methods can be adapted for multi-amplitude and multidirectional loading.
No further guidance is provided for cyclic design in standards such as
DNVGL-ST-0126, and indeed no commonly accepted design methods currently exist
for predicting the response of monopile foundations to cyclic lateral loading. This
disparity between the requirements for design and the available design methods has
motivated much research in this area over the last decade, as summarised in Section
1.3. Progress has been made in understanding the monopile response to regular
(constant amplitude), unidirectional cyclic lateral loading and a number of design
methods have been proposed, principally for predicting the permanent rotation
response under SLS conditions. However, few studies have explored the monopile
response under irregular, multi-amplitude and multidirectional loading, and few
design methods are able to capture all aspects of the cyclic response necessary for
design.
Cyclic design becomes increasingly important as conservatism in ULS design is
reduced (Manceau et al., 2019) and OWT structures become exposed to greater cyclic
wave loading at sites with more demanding environmental conditions. Predicting
the permanent rotation under SLS loading and the effect of cyclic loading on the
response is also increasingly important as turbines near the end of their design life
and lifetime extension is considered.

1. Introduction 16
Figure 1.14: Comparison of typical response to unidirectional cyclic loading in drained sand
with responses that may be obtained with cyclic degradation approaches
1.3 Relevant research contributions
This thesis focuses on the response of monopiles in sand. As such, this summary of
relevant research contributions is also generally restricted to sand.
1.3.1 Physical modelling
Many recent studies have explored the response of monopiles in sand to cyclic loading
through small-scale physical modelling, to gain insight into the monopile response and
ultimately inform cyclic design. The response of rigid piles with low L/D ratios has
been explored at both 1g (e.g. Leblanc et al., 2010a) and in the centrifuge at elevated
stress-levels (e.g. Klinkvort and Hededal, 2013); Table 1.2 summarises a selection of
these studies. All reported studies have explored the cyclic response in dry or drained
(water-saturated) sand. Although numerical modelling has demonstrated that partial
drainage and accumulation of pore pressure may be expected for large-diameter
monopiles, particularly under large-amplitude cyclic loading (Li et al., 2019; Peralta
et al., 2017), important insights can be gained from physical modelling which isolates
aspects of the full-scale problem. Physical modelling of pore pressure accumulation
under representative cyclic loading would be of great value, but would require a
viscous pore fluid to capture the drainage response (e.g. Dewoolkar et al., 1999).
1. Introduction 17
Key reference g-level Sand Pile Max. ζb ζc

properties L/D cycles N
Li et al. (2010) 100g Dry, dense 5 1000 N/A 0

Leblanc et al. 1g Dry, loose & 4.5 65000 0.2–0.5 −1.0–0.5
(2010) dense
Cuéllar et al. 1g Saturatedi , ∼4 5000000 0.2, 0.3 < 0, 0
(2012) dense
Klinkvort and 16g–75g Saturatedi , 6 10000 0.08–0.3 −1.0–0.5
Hededal (2013) dense
Nicolai and Ibsen 1g Saturatedi , 5 50000 0.1–0.4 −0.9–0
(2014) dense
Rudolph et al.ii 1g, 200g Saturatedi & 5 30000 N/A 0
(2014) dry, dense
Abadie et al. 1g Dry, very 4.7 100000 0.2–0.7 −0.6–0.5
(2019) loose
Arshad and 1g Dry, dense 10 6000 0.3–0.4 −1.0–0.5
O’Kelly (2017)
Truong et al. 60g, 250g Dry & 6, 11.4 1500 0.4–1.0 −1.1–0.7
(2019) saturatedi ,
dense
i
Water-saturated.
ii
Multidirectional cyclic loading applied, all other studies apply unidirectional loading only.
Table 1.2: Selection of relevant physical modelling studies in sand
Regular, unidirectional cyclic loading
The majority of studies have focused on understanding the permanent rotation (or
displacement) response of monopiles under regular, unidirectional cyclic loading, as
illustrated in Figure 1.10b. Cyclic loading is typically characterised in terms of
parameters ζb (Leblanc et al., 2010a) and ζc (Long and Vanneste, 1994; Leblanc et al.,
2010a), which are defined here in terms of maximum and minimum applied moments
MM AX , MM IN across a cycle:
MM AX
ζb = (1.3)
MR
MM IN
ζc = (1.4)
MM AX
These parameters could equally be defined in terms of maximum and minimum
horizontal loads HM AX , HM IN . Load asymmetry is defined by parameter ζc , with
descriptions summarised in Table 1.3, while cyclic load amplitude is together defined
1. Introduction 18
Constant loading ζc = 1
1-way loading ζc = 0
2-way loading ζc = −1
Partial 2-way loading −1 < ζc < 0
Partial 1-way loading 0 < ζc < 1
Table 1.3: Load asymmetry descriptions
Figure 1.15: Characterisation of cyclic loading in terms of ζb and ζc (after Leblanc et al., 2010a)
by ζb and ζc , as illustrated in Figure 1.15. Definition of an (arbitrary) reference load MR
is also required, which varies between studies, in general.
Permanent rotation (or displacement) has been found to accumulate as a
power-law (Leblanc et al., 2010a; Klinkvort and Hededal, 2013; Nicolai and Ibsen, 2014;
Abadie et al., 2019b; Truong et al., 2019) or logarithmically (Li et al., 2010) with cycle
number. Consistent ratcheting behaviour was also observed during large-scale
(D = 0.762 m, D = 2.0 m) cyclic loading tests performed in dense marine sand as part
of the PISA project (Beuckelaers, 2017). Ratcheting increases significantly with load
amplitude (characterised by ζb for constant ζc ), while load asymmetry (characterised by
ζc ) has also been found to impact the ratcheting response (Leblanc et al., 2010a;
Klinkvort and Hededal, 2013; Nicolai and Ibsen, 2014; Truong et al., 2019). Most studies
have focused on understanding the macro rotation (or displacement) response of the
monopile, although Truong et al. (2019) was also able to observe evolution of the
distribution of pile displacement and variation in residual (locked-in) net lateral
stresses during cyclic loading, using strain gauges on the pile surface.
1. Introduction 19
Some previous studies have also explored the evolution of secant stiffness per cycle,
which helps quantify the impact of cyclic loading on the foundation’s strength, stiffness
and dynamic response. Leblanc et al. (2010a) and Klinkvort and Hededal (2013) found
secant stiffness to evolve logarithmically with cycle number under a range of loading
conditions, while Abadie et al. (2019b) observed a faster increase in secant stiffness
for the first 50 cycles, and a logarithmic trend thereafter. Abadie et al. (2019b) also
explored evolution of the initial (maximum) stiffness with cyclic loading, but found
no variation over up to 100000 cycles.
The hysteretic response under cyclic loading has been reported in various studies
(e.g. Niemann et al., 2018; Beuckelaers, 2017), but Abadie et al. (2019b) and Abadie
(2015) specifically explored the symmetric hysteretic response to demonstrate
approximate adherence to the extended Masing rules (Masing, 1926; Pyke, 1979),
which have important implications for numerical modelling. Abadie et al. (2019b) and
Abadie (2015) also report an exponential decrease in per-cycle energy dissipation with
cyclic loading (in very loose sand), which is relevant for dynamic design.
Multi-amplitude cyclic loading
A few studies have applied successive packets of regular cyclic loading at different
amplitudes, as illustrated in Figure 1.16a, to explore the response to multi-amplitude
cyclic loading (Leblanc et al., 2010b; Abadie et al., 2019b). These studies showed
behaviour consistent with that observed under constant amplitude cyclic loading and
demonstrated a modest impact of packet order. They also emphasised the
disproportionately large contribution to ratcheting made by the larger load cycles,
given the non-linearity of the monopile response. Exploration of the response to
irregular multi-amplitude loading, representative of that experienced by monopiles in
the field, would complement these packeted multi-amplitude studies.
Multi-amplitude tests involving regular cycling followed by application of a large
monotonic load, as illustrated in Figure 1.16b, have also been conducted to explore
the effect of cyclic loading on the ultimate response (Nicolai and Ibsen, 2014; Nicolai
et al., 2017b; Abadie et al., 2019b; Truong et al., 2019). In general, the post-cyclic
responses were found to re-join or exceed the monotonic response, at odds with the
1. Introduction 20
(a) Regular, multi-amplitude (b) Regular cyclic loading (c) Regular, fan-type
packeted cyclic loading followed by reload multidirectional loading
Figure 1.16: Illustration of multi-amplitude and multidirectional loading applied in previous

physical modelling studies
cyclic degradation approaches. The results also suggest that it may be necessary to
consider the interaction of cyclic loading with the ULS response to accurately predict
pile rotation at peak load. Nicolai et al. (2017b) were able to quantify the observed
increase in monopile capacity in terms of cycle number and loading characteristics,
given that their tests reached yield.
Multidirectional cyclic loading
A limited number of studies have explored the response of monopiles to
multidirectional cyclic lateral loading. Peralta (2010) applied successive packets of
regular cyclic loading at 0°, 90°, 180°, etc. The pile was observed to move in the direction
of the applied load regardless of the loading history, and some hardening (reduction of
displacement) was observed with application of successive load packets.
The studies of Dührkop and Grabe (2008), Rudolph and Grabe (2013), Rudolph
et al. (2014b), and Rudolph et al. (2014a) involved application of more representative
loading, with the cyclic loading direction varying continuously in a fan-type shape,
across a specific spread angle, as illustrated in Figure 1.16c. These studies reveal the
important result that multidirectional cyclic loading can lead to greater ratcheting
than equivalent unidirectional loading, although there is inconsistency in the most
damaging spread angle observed across the studies (Rudolph et al., 2014a). The impact
of multidirectionality on secant stiffness and energy dissipation has not been explored.
1. Introduction 21
Figure 1.17: Particle migration at soil surface and evidence of convective region following
post-test excavation (arrows indicate cyclic loading direction) (modified from Cuéllar et al.,
2012)
Figure 1.18: Soil velocity vectors showing total displacement (normalised by pile diameter)
after 1000 loading cycles obtained through PIV analysis (Nicolai et al., 2017a)
1.3.2 Mechanistic studies
The mechanistic response of monopiles to cyclic lateral loading in sand has been
explored variously using Particle Image Velocimetry (PIV) on model-scale centrifuge
tests (Nicolai et al., 2017a), using coloured particles to track grain migration around a
model pile at 1g (Cuéllar et al., 2012), and through Discrete Element Modelling (DEM)
simulations (Cui and Bhattacharya, 2016; Duan et al., 2017).

1. Introduction 22
Figure 1.17 shows how Cuéllar et al. (2012) used coloured particles to track sand
migration on the soil surface and reveal convective regions within the soil volume, while
Figure 1.18 shows the soil velocity vectors after 1000 load cycles determined from an
example PIV analysis by Nicolai et al. (2017a). The studies are consistent, and suggest
the presence of conical convective regions either side of the pile, in-line with the loading
direction and extending to ∼ 1.5D depth. The convective particle motion is explained
by the cyclic variation of stress in-line with loading and the increasing confining stress
with depth (Cuéllar et al., 2012). Local subsidence — inferring local densification — was
also observed during the 1g tests presented by Cuéllar et al. (2012). Local densification
is also supported by DEM simulations (Cui and Bhattacharya, 2016; Duan et al., 2017)
and PIV results (Nicolai et al., 2017a). The asymmetry of both the convective regions
and the densification was found to mirror the asymmetry of the applied cyclic loading.
Ratcheting behaviour may, therefore, be understood to be driven by particle
rearrangement within convective regions, as well as any asymmetry in local
densification. Meanwhile, evolution of the foundation’s hysteretic response is
understood to be related to local densification. These mechanisms are, however,
expected to be coupled.
1.3.3 Proposed cyclic design methods
A number of methods for predicting the response of monopile foundations under cyclic
lateral loading have been proposed, a selection of which are summarised in Table
1.4. Most methods represent the monopile as a single-element (macro) or 1D model,
consistent with the requirement for fast design calculations, and many methods have
focused on predicting only the ratcheting response. In general, validation of these
cyclic design methods has been limited.
Informed by small-scale physical modelling, Leblanc et al. (2010a) and Klinkvort
and Hededal (2013) presented empirical relationships for ratcheting and evolution
of secant stiffness as a function of cycle number. The ratcheting relationships have
also been shown to approximately predict the response to packets of multi-amplitude
loading when coupled with a strain superposition method (Leblanc et al., 2010b; Abadie
1. Introduction 23
Method Key reference(s) Behaviour Modelling details

captured
Empirical Leblanc et al. Ratcheting, Empirical relationships for

relationships (2010), evolution of macro-response, informed by
Klinkvort and secant stiffness physical modelling
Hededal (2013)
Degradation Achmus et al. Ratcheting Degradation of backbone
Stiffness Model (2009) modulus within 3D FE model,
informed by element testing and
3D FE modelling
High Cycle Niemunis et al. Ratcheting Accumulation of strain within 3D
Accumulation (2005) FE or 1D model, informed by
Model element testing
Cyclic contour Andersen (2015), Ratcheting Methods for interpreting cyclic
diagrams Zhang et al. (2016), response from measured
Bayton et al. element testing or physical
(2018) modelling data
Hysteretic Page et al. (2018) Hysteretic Multi-surface kinematic
macro-model response hardening (MSKH) macro-model,
informed by 3D FE modelling
HARM Houlsby et al. Ratcheting, Hyperplastic model based on
(2017) hysteretic MSKH model with ratcheting
response, element, implemented as macro-
evolution of or 1D-model, informed by
loop shape physical modelling
Table 1.4: Selection of proposed design methods for monopiles under cyclic lateral loading
et al., 2015). However, these simple relationships are not adaptable for complex cyclic
loading and calibration for full-scale design has not been explored.
Achmus et al. (2009) incorporated an empirical expression for degradation of the
backbone secant modulus into an elasto-plastic soil model implemented as an FE
model. The resulting Degradation Stiffness Model (DSM) has been shown to predict
ratcheting responses under regular, 1-way cyclic loading reported by Leblanc et al.
(2010a) with reasonable accuracy (Achmus et al., 2011). This approach has the
advantage of being based on element testing results, which may facilitate calibration.
However, it is not clear how multi-amplitude loading can be accounted for.
The High Cycle Accumulation (HCA) model was developed to capture the general
response of coarse sand (Niemunis et al., 2005), and has been applied in 1D models
for application to monopile foundations (Wichtmann et al., 2017). The HCA model
uses a hypoplastic model (Niemunis and Herle, 1997) or elastic model (Wichtmann
1. Introduction 24
et al., 2017) to capture the continuous response for the first cycles and an empirical
model to capture the ratcheting response under many cycles. The model has not been
validated against measured monopile responses, but — like the DSM model — has
the advantage of being derived from element testing results.
Cyclic contour diagrams, which compile the measured cyclic responses from a
number of element tests (Andersen, 2015), have been used for cyclic design of other
offshore structures (e.g. gravity bases) for many years, principally in soft clays. For
application to piles in clay, Zhang et al. (2016) obtained degraded cyclic p-y curves
by scaling the DSS response inferred from Direct Simple Shear (DSS) cyclic contour
diagrams. Meanwhile Bayton et al. (2018) generated contour diagrams to capture the
macro response of a model monopile in dry sand. These approaches do not depend on
empirical relationships, but do require extensive element testing or model testing to
generate the contour diagrams, which may be prohibitively expensive for design.
A cyclic pile design method has also been proposed as part of the SOLCYP joint
industry project, although this project focused on the response of flexible piles with
L/D ≈ 16 (Garnier, 2013). Similarly to the approach of Long and Vanneste (1994), the
SOLCYP approach degrades p-y curves with empirical expressions in terms of cycle
number and cyclic amplitude. Although based on more extensive testing to higher cycle
numbers, this approach — like the cyclic degradation method of Long and Vanneste
(1994) — is only able to capture pile rotation at peak load.
While the models presented so far have focused on predicting the ratcheting
response, the multi-surface kinematic hardening (MSKH) macro model implemented
by Page et al. (2018) is able to capture the foundation’s hysteretic response and can
respond to any arbitrary load history, but cannot capture ratcheting. This model has
been validated against the hysteretic response observed in field-sale tests as part of the
PISA project (Byrne et al., 2017).
Models in the Hyperplastic Accelerated Ratcheting Model (HARM) framework
(Houlsby et al., 2017) are also based on MSKH models, and therefore also capture a
hysteretic response and respond to any arbitrary loading. Importantly, these models
are also able to capture ratcheting and evolution of hysteresis loop shape with loading
1. Introduction 25
history, through the inclusion of a ratcheting element and the selection of empirical
evolution functions (Houlsby et al., 2017). This combination of features allows these
models to capture all three cyclic design aspects detailed in Section 1.2.5, and makes
them very attractive for use in design. Models in the HARM framework have been
implemented as macro models (Abadie, 2015) and 1D models (Beuckelaers, 2017) and
have been shown to predict the ratcheting and hysteretic response under regular,
unidirectional cyclic loading at model-scale in sand (Abadie et al., 2019a) and
field-scale in sand and clay (Beuckelaers, 2017).
1.4 Thesis structure

1.4.1 Outline
Section 1.2.5 outlined the need for cyclic design methods for monopile foundations to
predict the accumulation of permanent rotation under SLS conditions, the effects of
cyclic loading on the strength and stiffness of the foundation response, and the
hysteretic response. Given the complex cyclic loading to which monopiles are exposed,
design methods should perform under irregular, multi-amplitude and multidirectional
cyclic loading conditions.
Much insight into the response of monopiles to cyclic lateral loading has been
gained through physical modelling studies, the majority of which have focused on
understanding the response to regular, unidirectional cyclic loading (Section 1.3.1). In
parallel, a number of design methods have been (and continue to be) developed to
capture the response to cyclic loading, in particular the permanent rotation under SLS
conditions (Section 1.3.3). Models in the HARM framework show particular promise:
able to predict both ratcheting and the hysteretic response, as well as respond to
any arbitrary loading. Demonstration of the performance of models in the HARM
framework under irregular, multi-amplitude and multidirectional cyclic loading is now
necessary to build confidence in their application for full-scale design.
This thesis employs laboratory-scale physical modelling to i) provide fundamental
insight into the foundation response and ii) facilitate demonstration and inform
development of models in the HARM framework. Regular, unidirectional cyclic tests

1. Introduction 26
complement previous experimental studies (Section 1.3.1), explore the impact of cyclic
load amplitude and load asymmetry in detail, and facilitate the interpretation of more
complex tests. Next, regular multidirectional tests systematically explore the response
to cyclic loading with multiple direction components. The response to realistic,
irregular, multi-amplitude and multidirectional storm loading is then explored, and
parallels are drawn between these novel responses and those observed under regular
cyclic loading.
The majority of the testing was performed at 1g at Oxford University (OU), with
complementary centrifuge testing performed at The University of Western Australia
(UWA) to explore the effect of stress-level on the cyclic response. The results from UWA
help to establish the relevance of 1g physical modelling and inform comparison of
monopile responses observed at various stress-levels.
This thesis forms part of a wider coordinated programme of research at Oxford
University, the overall aim of which is to understand monopile performance and
develop practical design methods for full-scale monopile foundations under cyclic
lateral loading. Related research activities include cyclic element testing, theoretical
and numerical model development, and large-scale field testing.
1.4.2 Chapter summary
Figure 1.19 presents the structure of this thesis, highlighting the testing performed and
the key interactions between Chapters. Each Chapter is briefly summarised below:
Chapter 1 provides context and motivation for exploration of the response of
monopiles to complex cyclic lateral loading. Previous work, on which this thesis builds,
is also summarised.
Chapter 2 describes the design of novel computer-controlled experimental
apparatus which allows application of continuously-varying, multi-amplitude and
multidirectional cyclic loading, at constant eccentricity, to a model monopile under
load control. Ancillary apparatus for sample preparation and pile installation is also
described. This laboratory apparatus is used to carry out the 1g tests at OU, presented
in Chapters 3, 4 and 5.
1. Introduction 27
Figure 1.19: Thesis structure

1. Introduction 28
Chapter 3 lays the groundwork for the cyclic results presented in later Chapters.
The test procedure for the 1g tests at OU is outlined, followed by presentation of the
monotonic responses, which underpin the cyclic responses on which this thesis is
focused. Normalisation approaches are discussed and consistent definitions for cyclic
loading and the cyclic response are presented.
The response of the monopile to regular cyclic loading is explored in detail in
Chapter 4, with tests in both very loose and dense sand. The impact of cyclic load
amplitude, load asymmetry and load directionality is examined, with the cyclic
response characterised in terms of ratcheting and evolution of secant stiffness and
energy dissipation per cycle. The post-cyclic reloading response is also explored.
Chapter 5 explores the monopile response to realistic multi-amplitude and
multidirectional storm loading by application of loading derived from independent
model-scale wave basin tests. Parallels are drawn between the responses observed
during regular cyclic loading in Chapter 4 and those observed under storm loading.
A key limitation of 1g physical modelling is the inability to simulate prototype-
scale stress-levels. Chapter 6 presents the results of a study conducted at UWA to
isolate and investigate the effect of stress-level on the response of a monopile in dense
sand to cyclic lateral loading. Tests were conducted with an identical set-up at 1g
and in the centrifuge at 9g and 80g.
Chapter 7 draws on the experimental observations from Chapters 3, 4, 5 and 6 to
demonstrate and inform development of numerical models, formulated in the HARM
framework (Houlsby et al., 2017), to capture key features of the cyclic response of
monopile foundations. Computations are conducted for each of the five datasets
presented in this thesis (OU very loose, OU dense, UWA 1g, UWA 9g and UWA 80g).
Finally, Chapter 8 summarises the key contributions of this thesis and provides
suggestions for future work which have arisen from the research presented herein.
Chapter 2
Design of novel laboratory

apparatus for 1g monopile testing
2.1 Introduction
This Chapter describes the design of laboratory apparatus to explore the response of a
model monopile to regular and complex cyclic lateral loading at OU at 1g, as reported
in Chapters 3, 4 and 5. 1g laboratory testing requires significantly less resource than
centrifuge testing or large-scale field testing, and therefore allows a greater number
of tests to be performed. More complex systems can also be developed at 1g, without
the size and acceleration-level constraints of centrifuge testing or the practical and
environmental constraints of field testing.
Many previous studies have used mechanical loading systems to apply cyclic
loading to model monopiles (e.g. Leblanc et al., 2010a; Nicolai and Ibsen, 2014; Abadie,
2015; Arshad and O’Kelly, 2017). Mechanical systems are reliable, but are generally
limited to application of unidirectional loading at a constant amplitude and frequency,
at least within individual load packets. A computer-controlled loading system is
developed here, in order to apply more complex loading. Computer-controlled loading
systems have also been used by e.g. Dührkop and Grabe (2008), Rudolph et al. (2014b),
and Su et al. (2014).
This Chapter first describes the design of the model foundation, before describing
in detail the loading apparatus and associated software, which is capable of applying
continuously varying, multidirectional and multi-amplitude cyclic loading under
29
2. Design of novel laboratory apparatus for 1g monopile testing 30
External diameter D 80 mm
Embedded length L 320 mm
Aspect ratio η 4 -
Wall thickness t 5 mm
Loading eccentricity h 800 mm
Dimensionless loading eccentricity ẽ 2.5 -
Material Aluminium
Young’s modulus E 70 GPa
Mass m 3.53 kg
Table 2.1: Model pile properties
accurate load control. Sample preparation and pile installation apparatus are also
presented.
2.2 Model pile
The model pile foundation was first defined, as this controls the size of the tank, loading
apparatus and actuator capacity. Table 2.1 summarises the chosen model monopile
properties; the diameter D = 80 mm scales approximately 1 : 100 relative to current
full-scale monopiles (Sørensen et al., 2017). This size of pile is small enough to enable
experiments to be easily managed by one person, but large enough to avoid grain
size effects (Klinkvort et al., 2013; Abadie, 2015) with Yellow Leighton Buzzard 14/25
sand (properties in Table 3.1, Section 3.2.1).
The choices of pile length L and loading eccentricity h = M/H were informed by
the dimensionless framework derived by Leblanc et al. (2010a) for laterally loaded
monopiles. The dimensionless load eccentricity ẽ and dimensionless pile aspect ratio η
(Equations 2.1 and 2.2, Leblanc et al., 2010a) were chosen to represent full-scale
monopiles.
h M
ẽ = = (2.1)
L HL
L
η= (2.2)
D
In the field, ẽ varies continuously with environmental conditions, but the impact
of variable loading eccentricity was not explored in this work. A constant eccentricity
ẽ = 2.5 was therefore chosen, giving h = 800 mm. This eccentricity may represent
conditions where thrust loading at the rotor (with eccentricity e.g. 4.4L) is 80% of
wave loading (with eccentricity e.g. L); this eccentricity may occur during operational
conditions, but lower eccentricities are likely during storm loading, when the rotor is
parked. A pile aspect ratio η = 4 was chosen, in-line with aspect ratios for modern
full-scale monopiles (Sørensen et al., 2017; Schroeder et al., 2015), giving L = 320 mm.
Pile rigidity is also an important consideration for the model monopile. The pile
relative stiffness KR and thresholds derived by Poulos and Hull (1989) are often used to
classify pile rigidity:

(
Ep Ip < 0.0026 flexible piles
KR = 4 (2.3)
L ESL > 0.208 rigid piles
Where ESL is the soil’s Young’s modulus at the pile tip and Ep Ip is the pile bending
stiffness. Abadie (2015) demonstrated that most monopiles installed at sand sites
pre-2015 have an intermediate rigidity, but future monopiles are expected to behave
rigidly. An aluminium model pile with wall thickness t = 5 mm was chosen. This pile
is classified as fully rigid (Poulos and Hull, 1989) in loose sand, where ESL < 27 MPa.
In dense sand, where e.g. ESL = 60 MPa (Geotechdata, 2013) the pile is close-to-rigid
(KR = 0.093). Although the pile rigidity is representative, the ratio of pile diameter
to wall thickness D/t = 15.9, is smaller than for typical full-scale monopiles, where
D/t ≈ 100 (University of Strathclyde, 2015); this is likely to affect installation effects,
but not the lateral response directly.
The pile has a normalised mean roughness Rn = Ra /d50 ≈ 0.001 (relative to Yellow
Leighton Buzzard 14/25 sand) measured using a hand-held stylus profilometer (Uesugi
and Kishida, 1986). The pile is therefore classified as smooth, in contrast to full-scale
monopiles which are typically rough. However, surface roughness has been shown to
have only a small impact on the monopile’s lateral response (Klinkvort, 2012).
2.3 Loading and pile measurement apparatus

2.3.1 Design overview
Figure 2.1 shows the Computer Aided Design (CAD) model of the loading and pile
measurement apparatus, and Figure 2.2 shows the apparatus in the laboratory. The
Pile cap Displacement

transducers
Actuator
Pile, D = 80 mm
Base plate
Tank, Dt = 800 mm
Figure 2.1: CAD model of laboratory apparatus (displacement transducers shown in all
possible positions, connections to pile exploded)
apparatus centres on a rigid cylindrical steel tank with upper and lower flanges, which
holds the soil sample and doubles as the reaction frame. A steel base plate bolts to
the top flange of the tank, on which removable steel brackets are attached to support
installation apparatus, actuators and displacement measurement transducers. The
modular design ensures that the apparatus is flexible.
Biaxial loading is applied to the pile with two perpendicular computer-controlled
actuators, while six displacement measurement transducers track the pile’s position
in space. A six Degree of Freedom (DoF) pile measurement system is necessary as the
pile is not restrained in any direction. A global coordinate system is defined with x- and
y-axes aligned with the line of action of the actuators and displacement transducers
(see Figure 2.2a). The z-axis is vertical, and the origin is at the sample surface.
Displacement
transducers
Actuator
Pile cap
Pile
x z
Base plate
y
Tank
(a)
Pile cap
Flexible couplings
Load cells
(b)
Figure 2.2: Photographs of laboratory apparatus (electrical and communication connections

omitted for clarity)
2.3.2 Tank
The internal tank dimensions were chosen to balance minimising boundary effects and
the volume of soil required. Numerical investigations by e.g. Achmus et al. (2007) state
that lateral boundary effects are avoided with a soil volume of diameter 12D, while
previous experimental work has used soil samples with (effective) diameter e.g. 7D
(Leblanc et al., 2010a; Abadie, 2015) and 10D (Albiker et al., 2017). Here, a cylindrical
tank with diameter Dt = 10D = 800 mm was chosen. Little work has investigated
the impact of the bottom boundary, however, 2D to 3D space beneath the pile tip is
typical. A tank depth ht = 800 mm ensures sufficient space for piles up to 480 mm
long (L/D = 6). The design assumes that the soil surface is located 100 mm below the
tank top flange, which helps avoid sample disturbance during set-up. The tank was
mounted on four soft rubber feet to isolate the system from external vibrations.
2.3.3 Pile cap
A pile cap was designed to transfer load to the pile and allow straightforward connection
of the upper displacement transducers to the pile. The cap sits on top of the pile, secured
by grub screws, and is attached following pile installation. The steel pile cap weighs
1.49 kg and includes a vertical shaft to hold 1 kg − 5 kg ring-shaped masses to apply
additional static vertical loading if required.
2.3.4 Actuators and load application
The actuators were sized to achieve 2° pile rotation in dense sand. Basic monotonic
analyses were performed (following the p-y approach described in Section 1.2.4) and
the model-scale results of Abadie (2015) and Leblanc et al. (2010a) were consulted.
Zaber BAR-E200 linear actuators were chosen, with specification summarised in Table
2.2. High resolution of motion is essential for accurate control at low load amplitudes.
The actuators are supported on brackets at h = 800 mm above the sample surface.
Figure 2.3 shows the components in-line with each actuator. Fatigue-resistant, HUCO
polymer couplings provide a relatively flexible link between each actuator and the
pile which masks any changes in the stiffness of the pile-soil system during loading
and improves actuator control, while low backlash universal joints allow the pile to
Travel 203 mm
Maximum thrust 540 N
Motor type Stepper
Resolution of motion 0.248 µm
Maximum speed 65 mm/s
Table 2.2: Zaber BAR-E200 linear actuator specification summary
Flexible coupling
Universal joint (UJ-A) Load cell Universal joint (UJ-P)
Actuator
Pile cap
Loading linkage length
(unloaded 125 mm)
Figure 2.3: Components in-line with actuator (configuration used for testing presented herein)
rotate and settle without applying moment. 200 N capacity S-beam load cells were
chosen to measure the in-line load applied to the pile. The load cells and associated data
acquisition system (described in Section 2.4.2) have a noise amplitude of around 0.07 N.
2.3.5 Position measurement
Temposonic ER-type magnetostrictive displacement transducers were chosen for
measurement of the pile position. Compared to standard linear variable differential
transformers (LVDTs) these sensors do not require signal processing or regular
recalibration. 50 mm travel was chosen to capture the pile’s response up to 2° rotation,
without compromising on accuracy during low-amplitude cyclic loading. The
transducers and associated data acquisition system (described in Section 2.4.2) have a
noise amplitude of around 1.5 µm.
Universal joints
Displacement
Actuator
transducer
Rigid rod Magnetic
Pile cap
connector piece
Figure 2.4: Components in-line with displacement transducer

Testing mode
Manual mode Data acquisition
Display loads, displacements Kinematics calculations
Initialise and actuator positions Load adjustment for transducer Close
Send commands to actuators friction
Take transducer zeroes Actuator load control
Data logging
Figure 2.5: Overall structure of Labview software
Six transducers are supported on three rigid brackets, appropriately orientated to
ensure resolution of the six degrees of freedom. Typically, one transducer is located
vertically above the pile; three transducers are located on the bracket aligned with
the x-direction, at two levels (800 and 400 mm above the sample surface); and two
transducers are located on the bracket aligned with the y-direction, at the same two
levels. The transducers attach to the brackets with sliding clamps, which allow the
transducer position in-line with the measurement direction to be varied.
Rigid aluminium rods link the transducers to the pile, with low-backlash universal
joints at either end to accommodate pile movement, as shown in Figure 2.4. Magnetic
connectors provide a simple and secure connection between the rigid rods and either
the pile cap (upper transducers) or small bosses on the pile wall (lower transducers).
2.4 Data acquisition, kinematics and control software

2.4.1 Software overview
Software was developed in the National Instruments Labview environment to perform
the data acquisition, kinematics calculations and control processes necessary to set-up
and run tests with the apparatus described in Section 2.3. The software runs on a
standard PC and includes a user interface to accept commands and display key data.
The software operates in one of two modes: Manual or Testing, between Initialise
and Close operations, as indicated in Figure 2.5. The Manual mode acquires and
displays load and displacement measurements and actuator positions, enables manual
commands to be sent to the actuators and records transducer zeroes before testing. This
simple mode is primarily used for test set-up, and its operation is not described further.
loop frequency ~50 Hz

Read load demand
from input file Output: actuator-1
velocity
Acquire
displacement Compute
Output: actuator-2
measurements pile pose
Adjust load for velocity
transducer friction
x
PID controller 1
Acquire load Resolve loads x Input: error in x-load
measurements Filtering
in x and y
y
PID controller 2
y Input: error in y-load
data logged at frequency fL<50 Hz
Figure 2.6: Structure of Testing mode within Labview software
The Testing mode is designed for running tests: performing data acquisition,
kinematics calculations, load demand adjustments, actuator load control and data
logging, as summarised in Figure 2.6. Sections 2.4.2 to 2.4.6 describe the key operations
performed within the Testing mode, which loops continuously at 50 Hz.
2.4.2 Actuator communication and data acquisition
Communication with the actuators is through a Zaber X-MCB2 dual-axis controller
which connects to the PC via USB. The Binary communication protocol was chosen, as
this increases communication speed by around 10 times, compared to default ASCII
communication. Fast actuator communication is essential for successful load control.
A CompactDAQ system from National Instruments is used for data acquisition and
signal conditioning. This modular system consists of a chassis with Ethernet
communication and two data acquisition modules. A 16-bit analogue input module
(NI 9205) acquires data from the displacement transducers, while a 24-bit
bridge-specific module (NI 9237 ) provides voltage excitation, signal conditioning and
analogue input for the load cells.
2.4.3 Kinematics calculations
Kinematics calculations are necessary to i) calculate the pile’s position in space (pose)
from the six displacement measurements and ii) calculate the position of the load cells
and resolve measured in-line loads to x- and y-components. To calculate the pile pose
Pi
Pile or
Ti pile cap
Displacement mm)
transducer, i l i (150.5
Figure 2.7: Schematic illustrating parameters for forward kinematics calculations (Pi , Ti , li )
from transducer measurements an iterative forward kinematics scheme is necessary;
the method described by Byrne and Houlsby (2005) and Byrne and Houlsby (2010)
for pivoting transducers is adapted here for fixed transducers with pivoting linkages.
The load cell positions and resolved loads can then be found directly using more
straightforward inverse kinematics calculations. The forward and inverse terminology
comes from robotics, where forward calculations find the pose of the end piece (here the
pile) from joint (here the transducer or load cell) positions, while inverse calculations
do the reverse (Kucuk and Bingul, 2006).
The pile’s pose is defined by the vector s = [u v w α β γ]T where u, v, w are pile
displacements in the x, y, z directions and α, β, γ are rotations about the x, y, z axes
respectively. The global coordinate system is defined in Section 2.3.1, and the pile is
treated as a rigid body. For each transducer i = 1, 2...6 the coordinates of the associated
pivot point on the pile or pile cap Pi = [xi yi zi 1]T , pivot point on the transducer
Ti = [xi yi zi 1]T and the linkage vector Li = [xi yi zi 0]T , with magnitude li = |Li |, are
defined. Figure 2.7 indicates Pi , Ti , li schematically for transducer i. In this set-up
the linkage vector magnitude is fixed, li = 150.5 mm.
The transformation matrix T̂ translates between global and local coordinate
systems, identified with subscripts G and L, respectively. For example:
PiG = T̂PiL (2.4)
The transformation matrix adopted here follows Byrne and Houlsby (2010):
 
cos β cos γ sin α sin β cos γ − cos α sin γ cos α sin β cos γ + sin α sin γ x
 
 cos β sin γ sin α sin β sin γ − cos α cos γ cos α sin β sin γ + sin α cos γ y 
T̂ =   (2.5)
 
 − sin β sin α cos β cos α cos β z
 
0 0 0 1
As matrix multiplication is not commutative, the transformation matrix is not uniquely
defined, but depends on the order in which transformations are considered to be
applied. However, for small displacements the impact of the choice of any
transformation matrix is negligible (Byrne and Houlsby, 2010).
The iterative forward calculation requires global transducer pivot coordinates TiG ,
local pile pivot coordinates PiL and linkage lengths li — which make up a vector of
linkage lengths V = [l1 , l2 , ..., l6 ]T — as inputs. TiG are determined from the
coordinates of the transducer pivot points when fully retracted TR

iG and the transducer
measurements ∆Ti . For a transducer parallel to the x-axis:
TiG = TR
iG + [∆Ti 0 0 0]
T
(2.6)
Both TR
iG and PiL were determined with the use of a CNC (Computer Numerical
Controlled) coordinate measuring machine. Coordinates of etched points on the base
plate and brackets were (CNC) measured relative to the global coordinate system, with
nine points measured close to each transducer location. The coordinates of TR

iG were
then determined (with a spreadsheet) from these coordinates and calliper
measurements between these points and the transducers, following transducer
positioning (which may vary between tests). Local coordinates on the pile and pile cap
were (CNC) measured at the magnetic connection points, which are straightforwardly
related to the pile pivot points PiL by the length of the connector pieces. The pile’s
local coordinate system was defined to be aligned with the global coordinate system for
a perfectly vertical pile installation to 320 mm embedment.
The iterative calculation is set up in terms of a change in vector of linkage lengths
∆V = [∆l1 , ∆l2 , ..., ∆l6 ]T . As the linkage lengths are fixed in this set-up, a solution is
obtained when ∆V → 0. It is assumed that the linkage lengths li can be linearised with
respect to the pile pose following Lazarevic (1997). This is expressed as:
dli
∆V = = Ri,j ∆s (2.7)
dsj
dli dLi
The chain rule can be used to express the matrix components as Ri,j = · . Noting
dLi dsj
dli 1
that Li = PiG −TiG = T̂PiL −TiG and = Li the matrix components Ri,j become:
dLi li
Li d(T̂PiL − TiG ) Li dT̂

Ri,j = · = · PiL (2.8)
li dsj li dsj
Having obtained R, the change in pile pose ∆s can then be calculated as:
∆s = R−1 ∆V (2.9)
Starting with an initial guess for s, the pose is updated iteratively using the solution
from the previous iteration. The scheme converges within a few iterations, although
10 iterations are used in the software, as the kinematics calculations do not limit the
loop speed in Testing mode (actuator communication speed does).
Calculation of the pile pose allows the position of the universal joints on the pile
cap (UJ-P, Figure 2.3) to be determined. The positions of the universal joints attached
to the actuators (UJ-A, Figure 2.3) are found independently using encoders in each
actuator. With the position of each universal joint known, the load cell angles and
resultant loads in the x- and y-direction are found using standard geometry (inverse
kinematics), which is not discussed further.
After e.g. loading the pile in the y-direction to 1° rotation, the resolved load in the
x-direction is approximately 83% of that measured in-line by the corresponding load
cell, highlighting the need to adjust for load cell angle. This discrepancy is dependent
on the length of the loading linkage, which is 125 mm for this set-up, when unloaded
(as indicated in Figure 2.3). Both the forward and inverse kinematics calculations are
embedded within the Testing loop and performed on-line at a rate of 50 Hz.
2.4.4 Adjustment for transducer friction
The displacement transducers (described in Section 2.3.5) have sliding frictional
resistance F̄ ≈ 0.22 N. The net force required to overcome the transducer friction is
therefore non-negligible when very low amplitude cyclic loading (e.g. < 10 N) is
applied to the pile. Approximate load adjustment is included in the Testing mode, as
indicated in Figure 2.6, to reduce conflation of the frictional response of the
transducers with the pile response.
The net friction magnitude is nF̄ , where n is the number of transducers aligned
with the load direction. Typically n = 3 in the x-direction and n = 2 in the y-direction.
Assuming load demand is being accurately tracked, the friction force Fi depends on
dHDi
the rate of change of load demand in the i-direction as:
dt
dHDi


 nF̄ >0



 dt





dHDi
Fi = −nF̄ <0 (2.10)


 dt






0 dHDi
=0

dt
0 = H +F ,
To account for transducer friction the load demands HDi are adjusted to HDi Di i
while the loads recorded as applied to the pile HRi are adjusted from the measured
loads HM i as HRi = HM i − Fi . To smooth abrupt changes in load demand and improve
load control performance, a moving average filter is also applied to Fi , which would
otherwise be a square wave for constant amplitude cyclic loading. Although this method
reduces the effect of transducer friction, it is not perfect, as F̄ is not constant and there
is always some error in load demand tracking.
2.4.5 Load control
Accurate load control is central to the operation of the Labview software in Testing mode.
Two PID (Proportional Integral Derivative) controllers (inbuilt Labview subroutine)
perform load control; Figure 2.6 shows how the controllers integrate into the Testing
structure. The controllers compare the measured x- and y-direction loads HM i to the
0 , and output velocities v for actuator 1 and 2, respectively.
(adjusted) load demands HDi i
0 are specified in a text file with x- and y-component loads defined

Load demands HDi
at the loop frequency (50 Hz). Actuator 1 and actuator 2 are initially aligned with the
x- and y-axes, but as the pile moves the load application axes become misaligned and
(e.g.) actuator 1 may apply some y-direction load. However, with a fast loop rate and
relatively small displacements, this coupling does not generate control issues.
PID controllers determine their output u(t) from the error in the Process Variable
(PV) e(t) as (Cannon, 2012):

Z t
1 de

0 0
u(t) = K e(t) + e(t )dt + Td (2.11)
Ti 0 dt
Where the controller variables are gain K, integral time Ti and derivative time Td . In this
case the output is actuator velocity u(t) = vi (t) and the PV is measured load HM i (t); the
0 (t) for axis i. A second order low-pass
error in the PV is therefore e(t) = HM i (t) − HDi
Butterworth filter (inbuilt Labview subroutine, cut-off frequency 15 Hz) is applied to
the PV to remove high frequency components and reduce actuator wear, although
this does introduce a small phase lag.
The controllers are tuned manually to achieve a fast response, without overshoot, to
a step-change in load applied to the pile-soil system. The flexible couplings (described
in Section 2.3.4) introduce a near-constant system compliance which leads to good
control performance with a constant gain and no integral and derivative components
(K ≈ 5000, Ti = ∞, Td = 0, effectively a simple Proportional controller) in both loose
and dense sand, and at variable loading rate. Gain amplification is also included to
increase K by factor 1.6 close to zero load (−1 < H < 1 N). Small but finite backlash in
the universal joints in-line with the actuators introduces additional system compliance
close to zero load, and increasing K here improves controller performance.
The control system has been tested under sinusoidal cyclic loading with maximum
loading rates up to 2πHCY C f = 11 N/s, where HCY C is the cyclic load amplitude and f
is the cyclic frequency. At these loading rates |e(t)| < 1 N , although this instantaneous
error is mostly due to the phase lag introduced by filtering.
2.4.6 Data logging
Throughout testing raw measurement data, pile pose, resolved loads and load demand
are logged to text files at a frequency fL suitable for the test, where (fL < 50 Hz).
2.5 Sample preparation apparatus
Apparatus was developed to prepare dense, dry sand samples in the tank described in
Section 2.3.2 (very loose samples were prepared manually, and both sample preparation
procedures are described in Section 3.2.2). Vibratory methods and air pluviation (sand
raining) methods have been used previously to prepare dense samples. Air pluviation
was chosen here as it may simulate a soil fabric similar to that found in natural deposits
formed by sedimentation (Rad and Tumay, 1987); is also known to be difficult to
achieve repeatable samples with vibratory methods. Air pluviation typically involves
pouring sand from a hopper at height h above the sand sample through a flow control
mechanism to achieve an effective unit weight γ 0 . The flow rate Q controls the ability of
particles to settle to a close-packed state and γ 0 reduces with increasing Q. Drop height
controls the particle velocity and γ 0 increases with increasing h, although only to the
point at which the sand particles reach terminal velocity (Rad and Tumay, 1987).
Various sand raining devices have been developed. For example, Schnaid (1990)
used a large fixed hopper at a high drop height and prepared samples in a single pour,
while Peralta (2010) maintained a constant drop height by pouring locally from a flexible
hose. Flexible, robotic sand rainers have also been developed by e.g. Zhao et al. (2006)
and Gaudin et al. (2018). A simple fixed device was developed here, as shown in Figure
2.8. A hopper, capable of holding around 0.3 m3 sand, is supported 750 mm above the
maximum soil sample height. The resulting minimum drop height h is assumed to
be large enough for most particles to reach terminal velocity, following the approach
of Schnaid (1990). The hopper is supported on a flanged cylinder, which contains
any dust generated by the raining process, but is also likely to introduce boundary
effects. Two shutter plates with matching hole patterns are positioned beneath the
hopper. The plates are initially misaligned, but moving the lower plate (by impact
with a rubber mallet) aligns the holes and allows sand to flow. A diffuser mesh is
positioned 150 mm beneath the shutter plates.
The shutter plate and diffuser mesh together control the sand flow rate Q and
therefore effective unit weight γ 0 . 20 mm diameter shutter plate holes on a 80 mm
equilateral triangular grid with a 2.54 mm square weldmesh diffuser sieve were found
to generate repeatable samples with Yellow Leighton Buzzard 14/25 sand (properties in
Table 3.1, Section 3.2.1) with average γ 0 = 16.2 kN/m3 and relative density DR = 60%.
Changing the diffuser mesh to a 1.27 mm square weldmesh and 5 mm diameter circular
Hopper
Shutter plates
Mesh level
Tank
Figure 2.9: Photograph of residual sand in

hopper following raining — highlighting
shutter grid pattern
Figure 2.8: Photograph of sand raining

apparatus
perforated plate with 6 mm pitch generated DR ≈ 70% and DR ≈ 80%, respectively.
However, sand accumulated on these meshes during raining, and so preparation of
DR = 60% samples was preferred. The shutter grid pattern is highlighted by the
residual sand in the hopper in Figure 2.9.
2.6 Pile installation apparatus
Apparatus was designed to install the model pile by driving. Figure 2.10 shows the
apparatus in the laboratory, with the actuator and transducer brackets omitted for
clarity. The pile is installed manually by raising and dropping the hammer mass (1.4 kg
or 2.8 kg for very loose and dense sand respectively) onto the pile. The Perspex hammer
guide ensures the mass drops squarely onto the pile sleeve from a constant height
(≈ 240 mm). The hammer mass and drop height were chosen to ensure pile installation
is controlled but not overly slow. Approximately 120 blows and 200 blows were required
for installation of the pile in very loose and dense sand respectively. The pile guide
holds the pile vertically at the tank centre during installation. It is designed to be easily
removed after installation without disturbing the pile.

Pulley
Pull
cord
Hammer
guide
Hammer
mass
Pile
sleeve
Pile
Hammer guide
frame
Figure 2.10: Photograph of pile installation apparatus
2.7 Summary
This Chapter has described the design of novel apparatus to explore the response of a
model pile with diameter D = 80 mm to cyclic lateral loading under load control. The
loading apparatus facilitates the application of complex — multidirectional,
multi-amplitude, continuously varying — cyclic loading, which would not have been
possible with the mechanical systems used in many previous studies. The
accompanying Labview software, which performs data acquisition, load control and
kinematics calculations, is integral to the performance of this loading apparatus and
has been described in detail. Development of ancillary apparatus for sample
preparation and pile installation have also been discussed.

Chapter 3
1g test procedure, monotonic

responses and cyclic definitions
3.1 Introduction
This Chapter describes the sample preparation and test procedures adopted for the
tests at OU (presented in this Chapter and Chapters 4 and 5). The monotonic
responses, which underpin the cyclic responses on which this thesis focuses, are
presented. The effect of loading rate and direction is explored, and normalisation
approaches are discussed. Consistent definitions for cyclic loading and the cyclic
response are also presented.
3.2 Sand sample

3.2.1 Sand properties
Yellow Leighton Buzzard (YLB) 14/25 sand was used for the tests conducted at OU. This
sand has been previously used for physical modelling in the geotechnical group at OU
(e.g. Villalobos, 2006; Leblanc et al., 2010a; Abadie et al., 2015). In parallel with the
work presented in this thesis, White (2020) has conducted extensive laboratory testing
to explore the response of this sand at low confining stresses.
Yellow Leighton Buzzard 14/25 sand is a coarse, uniformly graded silica sand. Figure
3.1 presents the measured particle size distribution, which approximately corresponds
to the lower bound of Fraction B, as defined in BS 1377 : Part 4 (British Standard, 1998).
The 0.6 − 1.18 mm sieve fractions (where the majority of this sand lies) correspond
46
3. 1g test procedure, monotonic responses and cyclic definitions 47
Specific gravity (Schnaid, 1990) Gs 2.65 -

Particle size d10 , d30 , d50 , d60 0.56, 0.69, 0.81, 0.87 mm
3
Minimum dry unit weight 0
γM IN 14.43 kN/m
3
Maximum dry unit weight 0
γM AX 17.64 kN/m
Maximum void ratio eM AX 0.80 -
Minimum void ratio eM IN 0.47 -
Critical friction angle (Schnaid, 1990) φ0c 34.3 °
Table 3.1: Yellow Leighton Buzzard 14/25 sand properties
to British Standard Sieve Series Mesh No. 14-25, hence common reference to this
sand as Yellow Leighton Buzzard 14/25.

0
Minimum and maximum dry unit weights (γM 0
IN , γM AX ) were obtained following
the methods described in BS 1377 : Part 4 (Sections 4.4 and 4.2) (British Standard, 1990),
0
and are reported in Table 3.1. However, methods for obtaining γM 0
IN and γM AX differ
between standards and organisations. Blaker et al. (2015) explored method dependency
0
for five different sands (including Yellow Leighton Buzzard 14/25), determining γM IN
0
and γM AX following four different standards and using in-house methods from three
organisations. Vibrating hammer, discrete hammer and mould vibration methods were
0
used to determine γM 0
AX , while methods for γM IN included placement with funnels,
tubes and scoops, and agitation by inversion of a cylinder. Significant variation in both
0 0 0
γM IN and γM AX was observed; for Yellow Leighton Buzzard 14/25, 14.12 ≤ γM IN ≤
15.13 kN/m3 and 16.45 ≤ γM

0 3
AX ≤ 17.85 kN/m (Blaker et al., 2015).
0
Variation in values of γM 0
IN and γM AX leads to variation in reported relative density
DR values. Figure 3.2 plots lower and upper bounds for relative density DR as a
function of unit weight γ 0 , using the ranges of γM

0 0
IN and γM AX reported by Blaker et al.
0
(2015). In this work, DR is obtained using the measured values of γM 0
IN and γM AX , but
Figure 3.2 highlights how comparisons to other data sets should be made with caution,
considering the significant method dependency reported (Blaker et al., 2015).
3.2.2 Sample preparation
The tests at OU were conducted at two densities: very loose and dense. The very loose
samples were built up in three layers by repeated manual pouring from a very low
Figure 3.1: YLB 14/25 particle size Figure 3.2: Relative density DR of YLB 14/25
distribution as a function of effective unit weight γ 0 (lower
and upper bounds use values of γM 0
AX and
0
γM IN from Blaker et al. (2015))
drop height (< 70 mm) using a shovel with approximate volume 0.0015 m3 . A constant
sample depth was achieved by careful pouring to the required depth, indicated by a
guide. The average unit weight across 80 samples (determined from global mass and
0
volume measurements) was γAV = 14.46 ± 0.08 kN/m3 , corresponding to an average
relative density at model-scale of DR = 1%. These samples were straightforward to
prepare and are aligned with the very loose samples used by Leblanc et al. (2010a) and
Abadie et al. (2015), but do not represent field conditions.
The dense samples were prepared using the sand raining device described in
Section 2.5. Due to lifting restrictions, the sample was prepared in two stages (rains). A
constant sample depth was achieved by vacuuming down to the required level with the
0
vacuum nozzle positioned in a jig. An average unit weight of γAV = 16.20 ± 0.06 kN/m3
was achieved across 10 samples (the unit weight was not recorded for all dense
samples, given the confidence in the sample preparation technique), corresponding to
an average relative density at model-scale of DR = 60%. These samples are more
representative of conditions off the UK coast, where sand deposits have relative
densities up to DR = 100% (e.g. Lunne, 2012).
Pile installation did not cause visible grain crushing, but did cause sample
densification close to the pile in both the very loose and dense samples. The reported
densities must therefore be considered as pre-installation values.
All tests were performed in dry sand to simulate fully drained conditions. Drained
conditions may be representative for small diameter monopiles in high permeability
sands, but they are less representative for silty sands and large diameter monopiles. Li
et al. (2019) proposed a preliminary criterion for assessment of the drainage conditions
around a monopile under cyclic loading, derived from a 2D finite element model
with isotropic linear elastic soil (only modelling horizontal drainage). The variation
of normalised excess pore pressure P = u/p (where u is excess pore pressure and p is
the average bearing pressure on the pile) with normalised loading period Tp = tp cv /D2
(where tp is the cyclic loading period and cv is the coefficient of consolidation) and
distance from the pile was explored, and the results are presented in Figure 3.3.
An 8 m diameter monopile in coarse clean sand (cv ≈ 10 m2 /s) exposed to wave
loading with period tp = 10 s gives Tp = 1.56, with the response predicted to be partially
drained. This is in agreement with the work of Peralta et al. (2017), who used analytical
solutions for consolidation around a laterally loaded pile from Osman and Randolph
(2012) to show that soil close to the monopile (in the same system) would behave as
partially drained. The magnitude of pore pressure accumulation also depends on the
load amplitude, as indicated by the pore pressure normalisation chosen by Li et al.
(2019), and so the effects will be more pronounced under large amplitude storm loading.
Partial drainage and the effect of pore pressure build-up during cyclic loading is
an important area of research. However, the response of monopiles to cyclic loading
is compound, and there is a need to isolate behaviour to properly understand the
response mechanisms. Understanding the drained response facilitates assessment
of more complex partially drained behaviour.
3.2.3 Impact of stress-level
The behaviour of sand is inherently stress-level dependent: typically peak friction
angle φ0p increases logarithmically with decreasing stress-level (Bolton, 1986), while
maximum shear modulus GM AX increases approximately with the square root of stress-
level (Hardin, 1965). To understand what soil state the low-stress laboratory sample
may represent at full-scale, either stiffness, friction angle or state parameter (Altaee and
Figure 3.3: Variation of normalised excess pore pressure P at peak cyclic load with normalised
distance from pile L/D and normalised loading period Tp (Li et al., 2019)
Fellenius, 1994) may be matched (it is not possible to match all these parameters with
laboratory-scale testing). Here, expressions for peak friction angle φ0p at laboratory-scale
and full-scale were equated to determine the full-scale relative density simulated in the
laboratory, consistent with previous work (e.g. Leblanc et al., 2010a; Zhu et al., 2017).
Bolton (1986) collated data on the strength and dilatancy of 17 sands, measured
in triaxial and plane strain conditions, and proposed empirical relationships to relate
peak friction angle φ0p with stress-level and relative density. The relationships can be
combined into the following expression for φ0p :
φ0p = φ0c + α DR Q − ln p0 − 1 (3.1)

Where p0 is the mean effective stress and α, Q and R are empirical values (α = 3 for
triaxial strain and α = 5 for plane strain, Q = 10 and R = 1). Limited data were
available at low stress levels and a dilation limit was therefore proposed, equivalent to
constraining φ0p − φ0c ≤ 12° in triaxial strain (Bolton, 1986). In a discussion on this work,
Tatsuoka (1987) presented additional data at low confining stresses 5 ≤ σc0 ≤ 50 kPa
which led to a revision of the empirical relations to limit dilation more strongly (Bolton,
Figure 3.4: Graphical representation of Equation 3.2 (Bolton, 1987), with approximate
representative stress-levels at laboratory (3 kPa) and full-scale (220 kPa) indicated
1987): 
φ0 + α D 5 − ln p0

c R − 1 for p0 > 150 kPa
φ0p = 150
(3.2)
φ0 + α (5DR − 1) for p0 < 150 kPa
c
Chakraborty and Salgado (2010) also analysed data for Toyoura sand at confining
stresses 2 ≤ σc0 ≤ 197 kPa and proposed varying empirical factor Q with stress level.
The experimental investigations of White (2020), which explore the behaviour of
Yellow Leighton Buzzard 14/25 sand at low confining stresses, show a reduction in
dilation at low stress levels and support the use of Equation 3.2. Figure 3.4 presents
Equation 3.2 graphically (for triaxial conditions), and indicates approximate
representative stress-levels σREF 0 at laboratory and full-scale (σREF 0 ≈ 3 kPa,
σREF 0 ≈ 220 kPa, respectively). σREF 0 is defined here as the vertical effective stress at
70% pile embedment (σREF 0 = 0.7Lγ 0 ), and is used as a proxy for mean effective stress.
Using Equation 3.2 and matching φ0p at laboratory- and full-scale suggests that the very
loose and dense samples represent relative densities of DR = 1.1% and DR = 65% at
full-scale, respectively. Classification of the samples as very loose and dense is therefore
not affected by stress-level.
3.2.4 Cone penetration tests
Cone penetration tests (CPTs) were performed on samples of Yellow Leighton Buzzard
14/25 sand prepared to various densities to calibrate a new laboratory CPT device for
the experimental studies of Mayall (2019), as described in Mayall et al. (2019). The
CPT has cone diameter 8 mm and angle 60°, with resistance measured locally at the
pile tip. As part of the CPT calibration work, three CPTs were performed on very loose
and dense samples prepared as described in Section 3.2.2, in the test tank described in
Section 2.3.2. All CPTs were performed at the centre of the sample – the sample’s radial
homogeneity was therefore not explored. However, monotonic tests in various loading
directions (presented in Section 3.4.1) demonstrate invariance to loading direction.
Figure 3.5 shows the measured cone resistance qc , while Figure 3.6 shows the
associated relative density DR profiles, obtained by application of the relationships
presented by Mayall (2019) for Yellow Leighton Buzzard 14/25 sand. Data is presented
for one very loose sample and two dense samples.
The CPT device was sized for dense sand, and so there is significant noise in the
very loose qc and associated DR profiles. Nevertheless, two regions of locally higher
resistance/density are observable at around −500 mm and around −260 mm. These
regions approximately correspond to the intersections of the three sand layers
deposited; similar layer depths were used to ensure repeatability. A region of higher
resistance/density can also be observed in the dense sand at around −400 mm,
corresponding to the intersection of the two rained layers. In this case, the layer was
positioned around 1D beneath the pile tip to minimise the impact on the pile response.
The dense samples exhibit an increase in DR with depth. The significant increase in
the first 80 mm embedment is attributable to the change in CPT bearing behaviour
(Gui et al., 1998), but the continued increase beyond 80 mm is likely to be an artefact of
the sample preparation technique. However, the consistency of the two profiles —
obtained from virgin samples — gives confidence in the repeatability of the dense
samples, particularly when coupled with the consistency in global unit weight
measurements (see Section 3.2.2).
3.3 Set-up and test procedure
The following procedure was adopted for the tests at OU. All tests were conducted
under load control using the apparatus described in Chapter 2.

Figure 3.5: CPT resistance qc for one very loose and two dense samples (data presented on two
qc scales for clarity)
Figure 3.6: Relative density profiles determined from CPT measurements for one very loose
and two dense samples (employing relationship between qc and DR from Mayall, 2019)
1. Preparation of sample. Either very loose or dense samples were prepared
following the methods described in Section 3.2.2. Sample preparation was
performed with the base plate and brackets unattached.
2. Location of base plate and brackets. The base plate — with brackets,
displacement transducers and actuators attached — was then located on the tank
and secured with bolts, before making the necessary power and data
communication connections.
3. Installation of pile. The pile installation apparatus described in Section 2.6 was
then attached to the base plate and the pile was installed to the target embedment
depth (320 mm) by manual hammering. Following pile installation, the pile
hammer and hammer frame were removed to create space for the remaining test
set-up.
4. Location of displacement transducers. The vertical displacement transducer
bracket was positioned above the pile and the displacement transducers were
located in positions appropriate for the particular test. The positions of the
transducers relative to the brackets were measured and used to determine the
coordinates of TR
iG (defined in Section 2.4.3) using a spreadsheet. These values
were input to the Labview software to allow determination of the pile pose during
testing.
5. Attachment of transducers and load lines. Next, the pile cap was positioned on
the pile and secured with grub screws, four transducers were connected to the
pile cap and two were connected to bosses on the pile wall, and the actuators
were jogged carefully into an initial position to minimise load applied to the pile
upon connection. The loading lines were then connected using grub screws at the
join between the flexible coupling and universal joint. The pile pose was recorded
throughout this process and care was taken to avoid disturbing the pile.
6. Running test. Before running the test, transducer zero readings were taken using
the Manual mode within the Labview software (Section 2.4), the load demand
text file was loaded, and an appropriate data logging rate fL was chosen. When
running the test, software operation moved to the Testing mode, where load
control, kinematics calculations and data logging were performed.
3.4 Monotonic response
Table 3.2 summarises the monotonic tests reported in this thesis. Tests were performed
in very loose and dense sand samples, in various loading directions and at various
loading rates. The test names indicate sequentially: sand density, test type (Monotonic),
loading rate, loading direction and repeat number (where relevant).

Test name Loading direction Loading rate [N/s]
Very loose sand samples
L.M.01.x x 0.1
L.M.01.y y 0.1
L.M.01.45 45° to x 0.1
L.M.1.x x 1.0
L.M.001.x y 0.01
L.M.Var.x y 1 & 0.01
Dense sand samples
D.M.02.x.1 x 0.2
D.M.02.x.2 x 0.2
D.M.02.x.3 x 0.2
Table 3.2: Monotonic test programme
3.4.1 Moment-rotation response
Figure 3.7 presents the monotonic response of the model monopile. As monopile design
criteria are typically defined in terms of foundation rotation (e.g. 3.10.2.2 DNVGL-ST-
0126 (DNV GL, 2016)), the OU responses are presented in this thesis in moment-rotation
(M − θ) space. The upper plots in Figure 3.7 present the full monotonic response, while
the lower plots focus on the low amplitude region relevant for cyclic loading.
Monotonic tests were conducted in three different directions in very loose sand,
following the x- and y-axes of the apparatus, and at 45° to the axes (L.M.01.x, L.M.01.y,
L.M.01.45). The consistency of these responses demonstrates invariance of the
apparatus, software and sand sample to load direction, which is essential for
investigation of the pile’s multidirectional response. The impact of loading rate was
also explored in the very loose samples, with tests at 0.01, 0.1 and 1.0 N/s, and with rate
changing between 0.01 and 1 N/s four times within a test (L.M.1.x, L.M.01.x, L.M.001.x,
L.M.Var.x). No discernible rate dependency was observed, as expected in fully drained
or dry sand. In general, the monotonic responses demonstrate excellent repeatability
at both densities.
Mean monotonic responses are determined from the tests summarised in Table
3.2. These mean monotonic responses or mean backbone curves provide a baseline
(a) Very loose sand (b) Dense sand
Figure 3.7: Monotonic response (lower plots show low amplitude region relevant for cyclic
testing, note different scales)
for interpretation of the response to cyclic loading, and are included in various plots
in Chapters 4 and 5. The initial monotonic loading responses for the unidirectional
cyclic loading tests summarised in Tables 4.1 and 4.2 (Section 4.2) are also presented
in Figure 3.7, and exhibit behaviour consistent with the monotonic tests.
3.4.2 Maximum stiffness
The maximum stiffness of the monopile kM AX is an important parameter in monopile
design as it controls the foundation’s (initial) dynamic response, and is a key input for
integrated dynamic analyses. In this work, kM AX is used as a reference value for
presentation of stiffness data (e.g. Section 4.4.2) and as an input parameter for
numerical modelling (Section 7.4.3). The maximum stiffness kM AX is most
straightforwardly estimated from the initial portion of the monotonic response, but
can also be obtained from the first unloading portion during cyclic tests. The initial
Figure 3.8: Initial portion of monotonic tests and first unloading from cyclic tests for estimation
of maximum stiffness kM AX (grey shaded region shows range of estimates for kM AX )
maximum stiffness is undisturbed, but may be subject to bedding-in effects which
would not be expected to affect the first unloading maximum stiffness.
Figure 3.8 presents the initial portion of the M − θ response for the monotonic tests
summarised in Table 3.2 alongside reflected unloading responses for the unidirectional
cyclic loading tests summarised in Tables 4.1 and 4.2 (Section 4.2). The monotonic
responses are zeroed from the onset of pile movement, while the cyclic tests are zeroed
from the point of maximum load on the 0th cycle (see Section 3.5.1). At very low
rotations (θ < 0.003°) there is little variation in the initial and unloading responses in
very loose sand, but in dense sand the unloading responses are a little stiffer than the
initial responses. No purely elastic region is observed at either sand density, i.e. the
response is non-linear even at very small rotations. Upper, lower and mean estimates
for kM AX are obtained by manual fitting to both the initial and unloading data at
θ < 0.003°, and are summarised in Table 3.3.
The mean maximum stiffness values (936 N m/° and 979 N m/° for very loose and
dense sand respectively) are less than 1/6 of the equivalent stiffness of the monopile
stick-up (modelled as a cantilever beam), confirming that it is appropriate to treat the
monopile as rigid.
Max. stiffness kM AX Reference rotation Reference moment

Sample
θR [°] MR [N m]
Mean [N m/°] Range [N m/°]
Very loose 936 557–1315 2 26

Dense 979 548–1410 2 95
Table 3.3: Reference values for monopile response
3.4.3 Normalisation
To facilitate comparison of experimental data, it is typical to present foundation
responses in a normalised form. Normalisations can either be conducted using
reference values or using dimensionless groups, which should capture the physics of
the system.
Leblanc et al. (2010a) and Kelly et al. (2006) develop dimensionless frameworks
for monopiles and caissons, respectively, which aim to capture the key behaviour of
the foundation-soil system and allow translation between scales. Both frameworks
incorporate stress-level dependent stiffness, with shear modulus G ∝ p0η . The exponent
is chosen as η = 0.5, which is generally accepted at small strains (GM AX ∝ p00.5 ), but at
larger strains η → 1 (Oztoprak and Bolton, 2013). Kelly et al. (2006) used this framework
to successfully compare tests on caissons conducted in the laboratory and field. The
results presented in Section 6.4.1 show that the similar framework of Leblanc et al.
(2010a) is able to account for stress-level effects under monotonic loading at load
amplitudes of interest. The dimensionless parameters for vertical load, horizontal
load, moment and rotation presented by Leblanc et al. (2010a), and the displacement
parameter presented by Abadie et al. (2015), are summarised in Table 3.4.
The dimensionless framework (Table 3.4) is used to discuss realistic cyclic load
amplitudes in Section 4.2, to scale between model and prototype-scale in Sections 5.3.4
and 7.6 and to interpret stress-level effects in Section 6.4.1. However, for standard
comparisons of the monopile response, data in this thesis is generally normalised
using reference values, rather than the dimensionless framework, as this is deemed to
be more intuitive.
V
Vertical load Ṽ =
L2 Dγ 0
H
Horizontal load H̃ =
L2 Dγ 0
M
Moment M̃ =
L3 Dγ 0
r
u pa
Displacement ũ =
D Lγ 0
r
pa
Rotation θ̃ = θ
Lγ 0
Table 3.4: Key dimensionless parameters (Leblanc et al., 2010a; Abadie et al., 2015)
Monopiles in sand do not typically reach well-defined failure, and instead show
continued hardening to large rotations. It is therefore necessary to arbitrarily define a
reference rotation (or displacement) value, from which a reference moment (or
horizontal load) can be determined. A variety of different reference values have been
used by previous researchers in this area. For example, Leblanc et al. (2010a) used a
dimensionless reference rotation of 4°, Abadie et al. (2015) used a ground level
displacement of 0.1D at model-scale, Arshad and O’Kelly (2017) used a rotation of 1.5°
at model-scale, while Bayton et al. (2018) defined a reference rotation of 0.25° on
unloading corresponding to the SLS criteria suggested by DNV GL (2016). A reference
rotation of θR = 2° at model-scale was defined for this work, broadly consistent with
Abadie et al. (2015) and Arshad and O’Kelly (2017). It is emphasised that the reference
rotation is arbitrary and is not intended to represent a design rotation.
Table 3.3 presents reference moment MR values for each sand density,
approximately corresponding to θR = 2°, while Figure 3.9 presents the normalised
mean monotonic responses in very loose and dense sand. Employing this
normalisation, the response in dense sand has a lower rate of change of normalised
Figure 3.9: Mean monotonic responses in very loose and dense sand normalised by reference
values in Table 3.3
tangent stiffness – and therefore appears more linear – than the response in very loose
sand.
3.5 Definitions for cyclic loading and the cyclic response
The definitions presented in this Section were developed in collaboration with T. D. Balaam.
Various definitions for cyclic loading and the cyclic response have been used in
previous physical modelling studies (see Table 1.2, Section 1.3.1), which hinders the
comparison of behaviour between studies. Cyclic definitions used in physical
modelling studies also differ from those used in cyclic element testing, which
complicates the comparison of behaviour observed at element and macro level; such
comparisons might inform numerical model calibration (Balaam, 2020). This Section
presents a consistent and rigorous framework, which brings together approaches from
previous studies, and proposes new definitions where inconsistencies arise. This
framework is adopted in the following Chapters.
The framework is presented generally in terms of stress σ and strain ε to emphasise
that the framework may be applied to both physical modelling of pile-soil systems and
to cyclic element testing. When considering the macro-response of a monopile σ and ε
may correspond to either applied moment M and pile rotation θ or applied load H and
σ Cycle 1 Cycle n Partial
2-way ζc=-1
1
e0 e1 en 2-way
σe
=0
-1<ζc<0
c
ζ
ay
Cycle 0
0.75
w
1-
ζ b=
1
σCYC/σREF
Partial
0.50
1-way
ζ b=
0<ζc<1
0.
5
0.25
ε
Constant ζc=1
σr 0.25 0.50 0.75 1
r1 r2 rn
σAV/σREF
Figure 3.10: Cycle definitions Figure 3.11: Graphical representation of
relationship between ζb , ζc , σAV , σCY C
pile displacement u (as discussed in Section 1.2.2). For element testing, σ and ε may
correspond to deviator stress q and axial strain εa or shear stress τ and shear strain γ.
The cyclic response is characterised in terms of ratcheting, secant stiffness and
energy dissipation, following previous studies (e.g. Leblanc et al., 2010a; Klinkvort,
2012; Abadie et al., 2019b) and in-line with the cyclic design considerations for
monopiles discussed in Section 1.2.5. Focus is primarily placed on definitions for
regular, unidirectional cyclic loading, although definitions for multi-amplitude,
unidirectional cyclic loading are briefly discussed. Extension of this framework for
multidirectional loading is discussed in later Chapters, as required.
3.5.1 Cycle definition
A loading cycle is defined as a load-unload loop from the reversal stress σr to the extreme
stress σe and back to the reversal stress σr (points rn → en → rn+1 in Figure 3.10). This
definition is consistent with typical definitions used in physical modelling studies (e.g.
Klinkvort, 2012), but in element testing, cycles are typically defined as starting and
finishing at the average stress. The physical modelling approach is preferred as it allows
straightforward extension to half cycles for multi-amplitude loading.
The stress bounds (σe , σr ) are rigorously defined to account for loading in either
“positive” or “negative” stress directions (as may occur for element tests in extension):
(
σAV + σCY C if σAV ≥ 0
σe = (3.3)
σAV − σCY C if σAV < 0
(
σAV − σCY C if σAV ≥ 0
σr = (3.4)
σAV + σCY C if σAV < 0
Note, however, that for applications such as lateral loading of piles, the definition of a
“positive” or “negative” stress direction is arbitrary. In this case σe and σr correspond
to the more commonly used values σM AX and σM IN , respectively.
Cycle 1 is defined as the second load-unload loop to pass through σe . Cycle 0 is
defined as all loading that occurs before Cycle 1, as indicated in Figure 3.10. The cyclic
response is typically presented for cycles 1 ≤ n ≤ N .
3.5.2 Cyclic load characterisation
In physical modelling, cyclic loading is typically characterised by parameters ζb (Leblanc
et al., 2010a) and ζc (Long and Vanneste, 1994; Leblanc et al., 2010a), as presented in
Section 1.3.1. These parameters are defined here in terms of σe and σr :
σe
ζb = (3.5)
σR
σr
ζc = (3.6)
σe
Load asymmetry is defined by ζc , while load amplitude is together defined by ζb and ζc .
Definition of an (arbitrary) reference stress σR is also required, for example, as
discussed in Section 3.4.3.
In contrast, in element testing (e.g. Andersen, 2015), cyclic loading is typically
characterised by parameters σCY C and σAV :
σe − σr
σCY C = (3.7)
2
σe + σr
σAV = (3.8)
2
Parameter σCY C defines the cyclic amplitude while σAV defines the mean cyclic load or
load bias. These parameters may also be normalised by σR for consistency with
parameters ζb and ζc .
All four parameters (ζb , ζc , σAV , σCY C ) are useful in characterising the cyclic load,
but parameters ζb and ζc are principally used in this thesis, for consistency with
previous physical modelling studies. Equations 3.9 and 3.10 express ζb and ζc in terms
of σCY C and σAV , while Figure 3.11 presents the relationship between these
parameters graphically for σAV > 0.
σAV + σCY C
ζb = (3.9)
σR
2σCY C
ζc = 1 − (3.10)
σAV + σCY C
3.5.3 Ratcheting definition
Ratcheting is the accumulation of permanent strain during cyclic loading, and therefore
requires a measure of strain per cycle εn . In physical modelling studies, εn has typically
been defined at the point of extreme load en (e.g. Leblanc et al., 2010a; Abadie, 2015;
Klinkvort and Hededal, 2013; Truong et al., 2019). However, this definition is at odds
with ratcheting being a measure of permanent strain, and conflates changes in stiffness
with ratcheting behaviour. In element testing, εn is typically defined as the mean of
strains at points en and rn (e.g. Andersen, 2015).
In physical modelling studies, it is typical to report permanent strain as an
accumulated strain per cycle ∆εn , relative to the equivalent monotonic strain
component (at n = 0); this helps to decouple ratcheting behaviour from the monotonic
response. Although the element testing definition for εn is more aligned with our
understanding of ratcheting as a permanent strain, and minimises conflation with
stiffness change, it does not have a well-defined equivalent monotonic strain, given
that r0 does not necessarily exist.
A new definition for permanent strain per cycle εn is therefore proposed, as the
mean value of strain at σAV on loading (at point an ) and at σAV on unloading (at point
bn ), as indicated in Figure 3.12. This definition is aligned with the element testing
definition, but also allows straightforward definition of accumulated strain. Equation
3.11 and Equation 3.12 define εn and ∆εn , in terms of points an and bn , respectively.
1
εn = (ε(an ) + ε(bn )) (3.11)
2
σ Δεn σ
εn en
σe σe
kn
b0 an bn
σAV a0 σAV
ε ε
σr σr
rn rn+1
Figure 3.12: Definition of permanent strain Figure 3.13: Definition of secant stiffness per
per cycle cycle
1
∆εn = ((ε(an ) + ε(bn )) − (ε(a0 ) + ε(b0 ))) (3.12)
2
3.5.4 Secant stiffness definition
Secant stiffness per cycle kn has previously been defined in physical modelling studies
as a loading stiffness between rn and en (Klinkvort and Hededal, 2013; Abadie, 2015)
and as an unloading stiffness between en and rn+1 (Leblanc et al., 2010a). However,
both these definitions conflate stiffness change with ratcheting behaviour.
A new definition for kn is proposed which minimises conflation of stiffness change
with ratcheting. Visually, this stiffness is at the cycle centre, as indicated in Figure 3.13.
It is defined by Equation 3.13 in terms of points en and rn , and is the inverse of the
average of the loading and unloading flexibilities.
σ(en ) − σ(rn )
kn = (3.13)
ε(en ) − 12 (ε(rn+1 ) + ε(rn ))
3.5.5 Energy dissipation definition
Cyclic energy dissipation is generally quantified for a symmetric closed hysteresis loop
by a metric proportional to the ratio of hysteretic energy loss EH to maximum stored
elastic energy EE . The areas representing these energies are shown in Figure 3.14.
Løvholt et al. (2020) and Inman (2014) use an energy loss factor η equal to the energy
lost per radian divided by the elastic energy (Equation 3.14). However, Kramer (1996),
σ
σe k
EH
EE
εe ε
Figure 3.14: Definition of EH and EE for a closed hysteresis loop
σ σ
σ
EH/2
EE
ε ε
ε
EH/2
EH
EE
EE
(a) Taborda et al. (2016) on (b) Taborda et al. (2016) on
(c) Abadie (2015)
loading unloading
Figure 3.15: Definitions of EH and EE for open loops proposed by previous studies
Abadie (2015) and Taborda et al. (2016) use a factor equal to half this value. Equation
3.14 is preferred as it is consistent with the quality factor often used to describe resonant
systems (Green, 1955). This energy dissipation measure is referred to as an energy loss
factor rather than a damping ratio to avoid confusion with the structural dynamics
term for the ratio of viscous damping to critical damping for a single degree of freedom
system (ξ). For a closed loop, η is bounded: 0 ≤ η ≤ 4/π.
1 EH

η= (3.14)
2π EE
EH and EE are straightforwardly defined for a symmetric closed loop with Equations
3.15 and 3.16, respectively. However, adaptation is needed for non-closing loops, which
occur when ratcheting occurs or multi-amplitude loading is applied.
1 σ2
EE = (σe εe ) = e (3.15)
2 2k
I
EH = σ(ε) dε (3.16)
Taborda et al. (2016) propose a method to calculate energy loss factor (presented as
“damping”) for non-closing loops caused by multi-amplitude loading, where the energy
loss is calculated incrementally from the previous load reversal to the current strain
(Figure 3.15a,b); this method is aimed at application to half-cycles. Meanwhile, Abadie
(2015) proposes an energy loss factor (also presented as “damping”) to account for
open loops due to ratcheting behaviour. The energy loss is equal to the area enclosed
by the hysteresis loop (closed at the load reversal), while the elastic energy is calculated
using a stiffness equivalent to kn (Figure 3.15c). This method was also used by
Beuckelaers (2017). However, this geometric definition of energy loss is only
appropriate for 1-way loading.
The hysteretic energy loss EH can be more generally defined by Equation 3.17.
This definition is equivalent to the approach of Abadie (2015) for 1-way loading, but
is also applicable to partial 2-way and partial 1-way loading, where the net energy
loss will not equate to the loop area.
Equation 3.18 defines the elastic energy EE , which follows the approach of Abadie
(2015) and is consistent with the definition for closed loops (Taborda et al., 2016). The
proposed energy loss factor per cycle ηn is then defined by Equation 3.19, following
Equation 3.14, where ηn is bounded as 0 ≤ ηn ≤ 4/π for closed loops. Figure 3.16
visualises the calculation of ηn for an example partial 2-way loading case.

Z en Z rn+1
EH = σ(ε) dε + σ(ε) dε (3.17)
rn en
(σe − σr )2
EE = (3.18)
8kn
Z en Z rn+1
4kn

ηn = σ(ε) dε + σ(ε) dε (3.19)
π(σe − σr )2 rn en
This definition of ηn is equivalent to Method 0 presented by Løvholt et al. (2020).
Løvholt et al. (2020) also present more complex methods for calculation of η which i)
σ σ
en en
σe σe
positive EH Loading
negative EH half-cycle
EE Unloading
bn half-cycle
an
σAV
ε
ε
rn+1
σr σr
rn rn+1 rn
Figure 3.16: Visualisation of general EH and Figure 3.17: Definitions for multi-amplitude
EE for partial 2-way loading cyclic response
attempt to remove the contribution of ratcheting to damping, and ii) relate the energy
loss factor to the phase angle between stress and strain in the frequency domain.
3.5.6 Application to multi-amplitude loading
For extension of the cyclic response definitions to multi-amplitude loading, cycles are
divided into alternate loading and unloading half-cycles between alternate extreme
and reversal points. The extreme points are defined as having greater mean stress
across the signal than the reversal points. Extreme and reversal points (en , rn ) and
extreme, reversal and average stresses (σe , σr , σAV ) are defined for each loading
half-cycle in Figure 3.17.
Strain per cycle can be defined at the half-cycle average stress on either loading (at
point an , εna ), or unloading (at point bn , εnb ) as indicated in Figure 3.17. For
multi-amplitude loading, evolution of strain is generally presented in terms of strain
per cycle (εna or εnb ) rather than an accumulated strain per cycle (∆εn ), as the
equivalent monotonic strain component necessary to obtain ∆εn is poorly defined
under multi-amplitude loading.
Stiffness may be calculated for each half-cycle following Abadie (2015) on loading
(knl , Equation 3.20) and Leblanc et al. (2010a) on unloading (knu , Equation 3.21),
although these definitions conflate ratcheting with stiffness change.
σ(en ) − σ(rn )
knl = (3.20)
ε(en ) − ε(rn )
σ(en ) − σ(rn+1 )
knu = (3.21)
ε(en ) − ε(rn+1 )
The energy loss factor may also be calculated per half-cycle, using the definitions
for EE and EH proposed by Taborda et al. (2016) (see Figure 3.15a,b); however, this
approach is not entirely consistent with the approach presented in Section 3.5.5, which
includes dissipation due to ratcheting.
3.6 Summary
This Chapter has laid the groundwork for exploration of the cyclic response of monopile
foundations in the following Chapters. The sample preparation and test procedures
used for testing at OU were described and the monotonic responses at 1g in very
loose and dense sand were presented. Normalisation approaches were discussed and
consistent definitions for cyclic loading and the cyclic response were proposed; these
definitions are adopted in the following Chapters.

Chapter 4
Regular cyclic loading response at

1g
4.1 Introduction
This Chapter presents the response of a model monopile to regular, unidirectional
and multidirectional cyclic lateral loading in very loose and dense dry sand at 1g. All
tests were conducted using the apparatus presented in Chapter 2 following the test
procedures outlined in Chapter 3. The tests build in complexity: from unidirectional,
symmetric tests with a few cycles to multidirectional tests with continuously varying
loading direction and 1000 cycles. The tests provide fundamental insight into the
response of a monopile to cyclic lateral loading and inform development of models
within the HARM framework (Houlsby et al., 2017), as described in Chapter 7.
4.2 Test programme
Figure 4.1 illustrates the six test types which constitute the regular cyclic loading test
programme. These six test types allow exploration of the hysteretic (few cycle)
response and high cycle response, under unidirectional and multidirectional loading.
The hysteretic responses reveal fundamental behaviour and adherence to Masing rules
(Masing, 1926), while the high cycle responses show evolution of pile rotation
(ratcheting), secant stiffness and energy dissipation with cycle number. Unidirectional,
1-way tests were conducted to 10000 cycles but many other tests were conducted to
1000 cycles given the diminishing returns associated with performing longer-term tests.
The unidirectional tests complement previous test campaigns (e.g. Leblanc et al.,
69
4. Regular cyclic loading response at 1g 70
Unidirectional Multidirectional
Hysteretic response
Multidirectional,
spiral tests
θy
Unidirectional,
symmetric 2-way tests
M
θx
Multidirectional,
Unidirectional, perpendicular tests
1-way tests θy
High cycle response
θx
Unidirectional, Multidirectional,
partial 1 & 2-way tests fan-type tests
My
M
Mx
θ
Figure 4.1: Illustration and categorisation of regular cyclic loading test types
2010a; Abadie et al., 2015; Truong et al., 2019) and exhibit behaviour consistent with
previous observations, while the multidirectional tests are more novel. Tests of each
type are performed in both very loose and dense sand, and the results are presented
side-by-side, although the test programmes are not identical. The very loose and dense
test programmes are summarised in Tables 4.1 and 4.2 respectively.
The test names indicate sequentially: sand density, load type (2-way (TW), 1-way
(OW), partial 2-way (PT), partial 1-way (PO), spiral (SP), T-shape (T), L-shape (L),
high mean T-shape (TH), fan-type (F)), and load amplitude or direction. The cyclic
loading was applied using sinusoidal waveforms, and post-cyclic reloading to 0.8MR
was conducted as part of many tests to explore the impact of cyclic loading on the
response under large loads. To ensure accurate load control, tests were conducted
at a range of frequencies 0.025 ≤ f ≤ 0.2 Hz depending on the cyclic amplitude. As
shown in Section 3.4.1, rate effects are not observed.

The cyclic amplitudes (and corresponding ζb values) were selected considering a)
experimental limitations, and b) representative cyclic loading. The minimum cyclic
amplitude was set by the amplitude of transducer friction (nF̄ = 0.44 or 0.66 N, Section
2.4.4). Given the approximations made in adjusting for the transducer friction, it was
not deemed sensible to run tests where the moment caused by transducer friction was
greater than around 15% of the cyclic amplitude. This restricted cyclic amplitudes to
MCY C > 3.5 N m or ζb > 0.13 and ζb > 0.04 for 1-way cyclic loading in very loose and
dense sand, respectively.
Representative SLS cyclic load amplitudes may be estimated following the
dimensionless framework of Leblanc et al. (2010a) (Table 3.4, Section 3.4.3), by
equating θ̃ at laboratory- and prototype-scale. For a prototype-scale monopile with
diameter D = 8 m and L/D = 4 in dense, saturated sand with γ 0 = 10 kN/m3 , the ratio
of prototype-scale rotation (θP ) to laboratory-scale rotation (θL ) fθ = θP /θL ≈ 8.3.
Prototype-scale rotation under SLS conditions of 0.25° (DNV GL, 2016) therefore
corresponds to approximately 0.03° rotation at laboratory-scale in both very loose and
dense sand. The associated moments under monotonic loading are 7.3 N m and
8.5 N m in very loose and dense sand respectively. This analysis suggests that cyclic
loading with ζb > 0.28 in very loose sand and ζb > 0.085 in dense sand may not
represent SLS conditions, as it would lead, upon initial monotonic loading, to θ > 0.25°
at prototype-scale.
To best represent SLS conditions, tests were conducted at cyclic amplitudes which
were as low as possible, while satisfying the restriction due to transducer friction. For
the unidirectional, 1-way tests – where the effect of cyclic amplitude was explored –
tests were conducted at ζb = 0.2, 0.3, 0.4 in very loose sand and ζb = 0.05, 0.1, 0.2, 0.3
in dense sand. Other test types were conducted at fewer, but often corresponding,
cyclic amplitudes. While many of the tests were conducted at amplitudes which may
be greater than those expected under SLS conditions, given the non-linear monotonic
response (Figure 3.7, Section 3.4.1) there is no reason to expect a step-change in the
behaviour at lower cyclic load amplitudes. Exploration of the response under realistic
loading is considered in Chapter 5.

Loading Cycle Reload

Test name ζb ζc
direction Number N
Unidirectional, symmetric 2-way tests
L.TW.Var x 0.2, 0.4, 0.6 -1 3.5 No

L.TW.04 x 0.4 -1 2000 Yes
Unidirectional, 1-way tests
L.OW.02 x 0.2 0 10000 No

L.OW.03 x 0.3 0 10000 Yes
L.OW.04 x 0.4 0 10000 Yes
Unidirectional, partial 1-way and 2-way tests
L.PT.260 x 0.2 -0.60 1000 Yes

L.PT.460 x 0.4 -0.60 1000 Yes
L.PO.633 x 0.6 0.33 1000 No
Multidirectional, spiral test
Spiral
L.SP.1 <0.60 - 5 No
(x − y)
Multidirectional, perpendicular tests
x 0.2 1
L.T.02 1000 No
y 0.2 -1
x 0.2 1
L.L.04 1000 No
y 0.4 0
x 0.4 1
L.TH.04 1000 No
y 0.2 -1
Multidirectional, fan-type tests
L.F.15 ±15° fan 0.4 0 1000 Yes

L.F.30 ±30° fan 0.4 0 1000 Yes
L.F.45 ±45° fan 0.4 0 1000 Yes
L.F.60 ±60° fan 0.4 0 1000 Yes
L.F.90 ±90° fan 0.4 0 1000 Yes
L.F.120 ±120° fan 0.4 0 1000 Yes
L.F.150 ±150° fan 0.4 0 1000 Yes
Table 4.1: Regular cyclic test programme in very loose sand

Loading Cycle Reload

Test name ζb ζc
direction Number N
Unidirectional, symmetric 2-way tests
0.08, 0.16, 0.24,

D.TW.Var x -1 5.5 No
0.32, 0.40
D.TW.008 x 0.08 -1 1000 Yes
D.TW.016 x 0.16 -1 1000 Yes
D.TW.032 x 0.32 -1 1000 Yes
D.TW.04.W0 x 0.4 -1 1000 No
D.TW.04.W2 x 0.4 -1 1000 No
D.TW.04.W5 x 0.4 -1 1000 No
Unidirectional, 1-way tests
D.OW.005 x 0.05 0 10000 Yes

D.OW.01 x 0.1 0 10000 Yes
D.OW.02 x 0.2 0 10000 Yes
D.OW.03 x 0.3 0 10000 Yes
Unidirectional, partial 1-way and 2-way tests
D.PT.160 x 0.1 -0.60 1000 Yes

D.PT.260 x 0.2 -0.60 1000 Yes
D.PT.245 x 0.2 -0.45 1000 Yes
D.PT.275 x 0.2 -0.75 1000 Yes
D.PO.333 x 0.3 0.33 1000 No
Multidirectional spiral test
Spiral
D.SP.1 <0.45 - 4 No
(x − y)
Multidirectional, perpendicular tests
x 0.1 1
D.T.01 1000 No
y 0.1 -1
x 0.1 1
D.L.02 1000 No
y 0.2 0
x 0.2 1
D.TH.02 1000 No
y 0.1 -1
Multidirectional, fan-type tests
D.F.30 ±30° fan 0.2 0 1000 Yes

D.F.90 ±90° fan 0.2 0 1000 Yes
Table 4.2: Regular cyclic test programme in dense sand

Figure 4.2: Response to first four cycles of unidirectional 2-way cyclic loading (mean
monotonic response shown in grey)
Test Mean ηn value
L.TW.04 0.56
D.TW.008 0.36
D.TW.016 0.38
D.TW.032 0.42
Table 4.3: Mean energy loss factors ηn for first four cycles
4.3 Unidirectional hysteretic response
Figure 4.2 presents the response of the model monopile to constant amplitude, 2-way
(TW) cyclic loading. These symmetric tests minimise ratcheting and allow focus to
be placed on the underlying hysteretic response. Only the first four loading cycles
are presented in Figure 4.2; evolution of secant stiffness and energy dissipation under
many cycles of 2-way loading is discussed in Section 4.4.
The shape of the hysteretic response is related to the foundation’s energy dissipation,
which is quantified by the energy loss factor ηn (defined in Section 3.5.5). Table 4.3
summarises the mean ηn values for the data presented in Figure 4.2. The mean energy
loss factor ηn increases with cyclic amplitude MCY C (or ζb ) and is greater for the test in
very loose sand, which exhibits a more convex response and no gapping-type behaviour.
In Figure 4.2b, the inflexion in the response around zero load for the dense sand
Figure 4.3: Hysteretic response at n = 4 in Figure 4.4: Illustration of Masing rules for
dense sand normalised by maximum single hysteresis loop following Puzrin and
moment and rotation to highlight variation Shiran (2000)
in shape of hysteretic response
tests, observed most clearly for test D.TW.032, is indicative of gapping-type behaviour:
the tangent stiffness reduces as the pile traverses a gap or region of low-stress. Figure
4.3 highlights the increase in gapping-type behaviour with cyclic amplitude MCY C (or
ζb ), by presenting the hysteretic response at n = 4 normalised by maximum moment
and rotation values (MM AX , θM AX ) for each test (note that θM AX occurs at n = 1). As
no physical gap was observed during these tests, it cannot be determined whether
either a very small gap or a region of low stress was the cause of the gapping-type
response. Gapping-type behaviour is not generally expected in dry sand, however,
electrostatic forces, ambient moisture or particle interlocking may provide enough
effective cohesion to generate a gap or low-stress region under cyclic loading.
Gapping-type behaviour is not observed in the very loose sand responses, perhaps
due to a reduced tendency for particle interlocking or reduced electrostatic forces
given the different sample preparation method. In the tests at UWA, gapping-type
behaviour is also observed under symmetric 2-way loading in dense sand at 1g and 9g
(reported in Section 6.5.1). In field tests as part of the PISA project, gapping-type
responses were recorded in both sand and clay, and gapping was also observed on site
(Beuckelaers, 2017).
The hysteretic responses also allow exploration of adherence of the system to
Masing rules (Masing, 1926). The Masing rules allow definition of the cyclic response
given definition of an initial loading (backbone) curve ε = f (σ), as shown in Figure 4.4,
and describe behaviour which complies with kinematic hardening models (Pyke, 1979).
Masing (1926) proposed the following rules, which were initially concerned with
modelling the behaviour of brass:
• The shape of the unloading or reloading curves is the same as that of the backbone
curve, with the scale enlarged by a factor of 2.
• The initial tangent modulus on each loading reversal is the same as the initial
tangent modulus on the backbone curve.
Pyke (1979) later extended the Masing rules for application to irregular cyclic loading:
• The unloading and reloading curves should follow the backbone curve if the
previous maximum stress is exceeded.
• If the current loading or unloading curve intersects the curve described by a
previous loading or unloading curve, the stress-strain response follows that of the
previous curve.
The extended Masing rules have been found to capture the key behaviour of many
materials and systems, including: soil behaviour in cyclic element tests (e.g. Puzrin
and Shiran, 2000), the lateral response of laboratory-scale caissons in dense sand at
1g (Byrne, 2000; Villalobos, 2006), and the first few cycles of the lateral response of a
laboratory-scale monopile in very loose sand at 1g (Abadie, 2015). The extended Masing
rules also provided a basis for modelling the response of large-scale monopile tests in
sand and clay, as part of the PISA project (Beuckelaers, 2017).
Figure 4.5 presents the response of the monopile to 2-way loading with ζb
increasing on each successive cycle. These tests are similar to those conducted by
Byrne (2000) and Abadie (2015), and explore adherence to the extended Masing rules.
The upper plots show the experimental response and the lower plots show the
response predicted by the extended Masing rules. In very loose sand, the response is
Figure 4.5: Response to 2-way cyclic loading with increasing amplitude alongside prediction of
the response following extended Masing rules (mean monotonic response shown in grey in
upper plots)
captured by the extended Masing rules with good accuracy. In dense sand, the
response broadly adheres to the extended Masing rules but exhibits some hardening,
and the aforementioned gapping-type behaviour, which are not captured by Masing
rules. The greater number of cycles in dense sand may contribute to the increased
hardening (or increase in secant stiffness) observed.
4.4 Unidirectional high cycle response
This Section explores the response of the monopile to at least 1000 regular,
unidirectional loading cycles. The response is characterised by accumulation of pile
rotation (ratcheting) and evolution of stiffness and energy dissipation.
4.4.1 Ratcheting
Although the monopile response approximately adheres to the extended Masing rules
under symmetric 2-way loading, accumulation of rotation (ratcheting) occurs under
biased loading. Figure 4.6 shows an example of the response to four cycles of 1-way
Figure 4.6: Example response to first four cycles of 1-way loading in dense sand to highlight
ratcheting (test D.OW.03)
loading in dense sand. Rotation accumulates in the direction of the load bias, as
evidenced by the non-closing hysteresis loops.
Figure 4.7 presents the accumulated rotation per cycle ∆θn (defined in Section
3.5.3) for the 1-way (OW), partial 1-way (PO) and partial 2-way (PT) tests summarised
in Tables 4.1 and 4.2 (the ratcheting response under 2-way loading is not of interest).
The magnitude and rate of ratcheting increases with cyclic amplitude MCY C (or ζb ,
given constant ζc ), while ζc impacts the shape of evolution of ratcheting. Ratcheting
continually slows with cycle number but does not stop accumulating, at least for the
cycle numbers tested here. Similar behaviour is observed at both densities, although
the response of test D.PT.275 is anomalous, with no data plotted at n = 1. For this test,
the combination of a slower initial ratcheting rate due to the loading asymmetry and
the development of a concave (gapping-type) hysteretic response between n = 1 and
n = 2 (like seen in e.g. Figure 4.2b) led to a negative value of ∆θn at n = 1.
To assess the impact of ζb and ζc on the ratcheting response, a power-law is fitted
to the evolution of accumulated rotation ∆θn with cycle number n:
∆θn
= Anα (4.1)
θR
The coefficient A controls the initial magnitude of ∆θn while the exponent α controls
the evolution of ∆θn with n. Power-laws capture the reduction in ratcheting rate with
Figure 4.7: Unidirectional ratcheting response with power-law fit (Equation 4.1) using
common exponent αc = 0.25 shown dashed
cycle number and have been used by various authors to approximate the ratcheting
response of a pile (e.g. Leblanc et al., 2010a; Klinkvort and Hededal, 2013; Abadie et al.,
2019b; Truong et al., 2019). Leblanc et al. (2010a) assumed a constant exponent α and
made A a function of load characteristics and relative density as A = Tb (ζb , DR )Tc (ζc ),
while Truong et al. (2019) made A equal to the measured ratcheting accumulated on
the first cycle ∆θ1 and let α vary with DR and ζc .
First, the empirical approach of Leblanc et al. (2010a) is followed. The dashed lines
in Figure 4.7 show the result of fitting Equation 4.1 to the data in Figure 4.7 with a
common (fitted) exponent αc . With αc = 0.25, a very good fit is obtained to the 1-way
and partial 1-way data, and a reasonable fit is obtained to the partial 2-way data for
n > 100. The reported exponent α depends on the chosen strain variable. For example,
with accumulated rotation at cycle extreme as the strain variable instead, αc = 0.30,
aligned with the results of Leblanc et al. (2010a) and Abadie (2015); with total rotation
at cycle extreme as the strain variable, αc = 0.17.
Figure 4.8 presents the variation of Tb with ζb and DR obtained from power-law
fits to the 1-way data with common exponent αc . Tb increases with DR and varies
with ζb as a power-law (shown dashed) with exponent mσ = 3.1 for very loose sand
and mσ = 2.3 for dense sand. This behaviour is similar to that reported by Leblanc
et al. (2010a) and Abadie (2015), although Abadie (2015) found Tb to vary with ζb as
a power-law with exponent mσ = 4.
Figure 4.9 plots the variation of Tc with ζc obtained from power-law fits with
common exponent αc . Figure 4.9 also presents data from other studies (Leblanc et al.,
2010a; Nicolai and Ibsen, 2014; Albiker et al., 2017), where tests were conducted to a
range of cycle numbers 1000 ≤ N ≤ 100000. Despite some scatter, there is a clear trend
across the datasets for greatest Tc values under partial 2-way loading at ζc ≈ −0.6, with
Tc reducing as ζc → −1 and ζc → 1. The impact of ζc is significant, with 1.3 ≤ Tc ≤ 4 at
ζc ≈ −0.6, and with the highest Tc values occurring in looser sand samples. Like for α,
the choice of strain variable impacts the Tc values, and the choice of accumulated or
total strain is particularly significant (Albiker et al., 2017). All studies presented in
Figure 4.9 use an accumulated rotation value (∆θ) as the strain variable, and are
therefore approximately comparable.
Alternatively, the effect of ζc on the ratcheting response can be explored by letting
α vary with ζc and DR , following the approach of Truong et al. (2019). Equation 4.1
is fitted independently to each test shown in Figure 4.7, which leads to better fits,
particularly for the partial 2-way and 2-way tests. The α values obtained are plotted in
Figure 4.10 against ζc , and with distinction between sand densities. There is a trend
for decreasing α with increasing ζc for −0.75 ≤ ζc ≤ 0, but for 0 ≤ ζc ≤ 0.33 there is
no clear variation in α. The impact of sand density is not strong, although there is
a weak tendency for greater α values in the looser sample. The impact of density is
therefore less pronounced than that observed by Truong et al. (2019), although the
impact of ζc is broadly consistent with Truong et al. (2019).
Both empirical approaches demonstrate the importance of load asymmetry (ζc ) in
controlling ratcheting: the approach of Leblanc et al. (2010a) allows the impact of ζc
Figure 4.8: Variation of Tb with ζb and density

following the empirical approach of Leblanc
Figure 4.9: Variation of Tc with ζc following
et al. (2010a) (power-law fit shown dashed)
the empirical approach of Leblanc et al.
(2010a) including comparison to other
studies
on the initial magnitude of ratcheting to be assessed (Figure 4.9), while the separate
approach of Truong et al. (2019) assesses the impact of ζc on ratcheting evolution
(Figure 4.10). The empirical approach of Leblanc et al. (2010a) also provides insight
into the impact of ζb , and indicates that the initial magnitude of ratcheting increases
as a power-law (exponent 2.3 − 3.1) with ζb (Figure 4.8).
Qualitatively, the impact of ζc on ratcheting may be understood as a competition
between (a) an increasing cyclic amplitude MCY C with increasingly negative ζc , which
increases pile movement and therefore the potential for soil particle rearrangement
and (b) a decreasing mean load MAV with increasingly negative ζc . Greater particle
rearrangement is likely to lead to greater ratcheting, which is understood to occur
through a combination of densification and convective mechanisms (Cuéllar et al.,
2012; Nicolai, 2017, Section 1.3.2), while a reduction in mean load will reduce ratcheting
behaviour as the mechanism becomes more symmetric. There is no such competition
when considering the impact of ζb on ratcheting, as both the cyclic amplitude MCY C
and mean load MAV increase with increasing ζb . This interpretation supports the
Figure 4.10: Variation of α with ζc following the empirical approach of Truong et al. (2019)
use of parameters MCY C and MAV to describe cyclic loading, as is commonplace in
element testing (e.g. Andersen, 2015).
4.4.2 Secant stiffness
During cyclic loading in dry sand, an increase in secant stiffness kn (defined in Section
3.5.4) tends to be observed. Figure 4.11 highlights how kn changes between cycle n = 1
and n = 10 for example 1-way cyclic loading in dense sand. Figure 4.12 presents the
evolution of secant stiffness kn with cycle number n under 2-way (TW), 1-way (OW),
partial 1-way (PO) and partial 2-way (PT) loading. An increase in kn is observed during
all tests, with a mean ratio of secant stiffness at n = 1000 (kn=1000 ) to initial stiffness
kn=1 (kn=1000 /kn=1 ) of approximately 3. However, the shape of evolution varies
significantly with the loading asymmetry (ζc ). The stiffness response of test D.OW.005
appears to be anomalous, with stiffness plateauing for 100 ≤ n ≤ 1000 and increasing
again for n ≥ 1000.
Leblanc et al. (2010a) and Klinkvort and Hededal (2013) both proposed logarithmic
relationships to capture the evolution of secant stiffness kn with cycle number n under
cyclic loading in dry sand. The expressions used by Leblanc et al. (2010a) and Klinkvort
and Hededal (2013) differ slightly, but have the same general form, which is expressed
here as:
kn kn=1
= + BL ln(n) (4.2)
kM AX kM AX
Figure 4.11: Example response to 1-way loading in dense sand highlighting secant stiffness at
cycles n = 1 and n = 10 (test D.OW.03)
Figure 4.12: Secant stiffness evolution under unidirectional cyclic loading with individual
logarithmic fits (Equation 4.2) shown dashed and range of fits to 1-way and partial 1-way data
with common coefficient BLc indicated by blue shaded region
Coefficient BL controls the rate of change of kn . Leblanc et al. (2010a) found BL to be
independent of DR and loading type (ζb , ζc ) within the tested range, while Klinkvort
and Hededal (2013) let BL vary with ζb and ζc . Both authors let the initial stiffness
kn=1 vary with the loading characteristics with an expression of the form kn=1 /kM AX =
Kb (ζb )Kc (ζc ), with Kc = 1 at ζc = 0.
Figures 4.13 and 4.14 plot the variation of Kb and Kc with ζb and ζc respectively,
obtained from the initial stiffness values kn=1 . Despite some scatter, there is a clear
trend for increasing Kb with reducing ζb and increasing Kc with increasing ζc . The
variation of kn=1 , and therefore Kb and Kc , can also be approximated analytically
by application of the extended Masing rules (Section 4.3). Given a function for the
backbone curve ε = f (σ), kn=1 may be approximated as:
MR ζb (1 − ζc )
kn=1 = Kb (ζb )Kc (ζc )kM AX ≈ (4.3)
MR
2f 2 ζb (1 − ζc )
This analytical approximation is presented in terms of Kb and Kc in Figures 4.13 and
4.14 for very loose sand, where the backbone curve is approximated with a Ramburg-
Osgood (Ramberg and Osgood, 1943) or power-law expression with exponent 3.4 (see
Section 7.4.3). The approximate analytical expression does not account for the stiffness
reduction due to ratcheting in the first cycle, but nevertheless captures the variation
of Kb and Kc for very loose sand with good accuracy.
Leblanc et al. (2010a) and Abadie (2015) both used a constant coefficient BL to
approximate the evolution of stiffness under various cyclic loading conditions. A
constant coefficient (BLc = 0.037) is able to approximately fit the evolution of kn for
the 1-way and partial 1-way tests presented here, as indicated by the blue shaded
areas in Figure 4.12, but cannot adequately capture the response to partial 2-way and
particularly 2-way cyclic loading. Following the approach of Klinkvort and Hededal
(2013) Equation 4.2 is fitted to each test independently to obtain the fits shown dashed
in Figure 4.12. However, even with independent fitting, the logarithmic expression
is unable to adequately capture the shape of evolution of kn under partial 2-way and
2-way loading. A power-law of the form:
kn
= Bnβ (4.4)
kM AX
Figure 4.13: Variation of Kb with ζb for 1-way Figure 4.14: Variation of Kc with ζc following
data following empirical approach of empirical approach of Klinkvort and Hededal
Klinkvort and Hededal (2013) and Leblanc (2013) and Leblanc et al. (2010a)
et al. (2010a)
is able to capture a faster rate of change of stiffness at higher cycles and provides a
better fit to the 2-way and partial 2-way data, at least for n > 10, as shown in Figure 4.15.
Secant stiffness change is understood to be caused by local densification (discussed
in Section 1.3.2), which relies on particle rearrangement. It is conceivable that 2-way
and partial 2-way loading (which passes through zero load, ζc < 0) leads to greater
particle rearrangement, and therefore a greater rate of stiffening, than 1-way and
partial 1-way loading (ζc ≥ 0). Greater particle movement is certainly likely to
accompany the gapping-type behaviour which was observed under larger amplitude
2-way loading in dense sand (Section 4.3).
4.4.3 Energy dissipation
Figure 4.16 highlights how the hysteretic response of the monopile changes under
1-way cyclic loading in dense sand, with the loop area (representing the hysteretic
energy loss EH , defined in Section 3.5.5) shaded at cycle n = 1 and n = 10. Figure 4.17
presents the evolution of hysteretic energy loss EH with cycle number n under 1-way
(OW), partial 1-way (PO), 2-way (TW) and partial 2-way (PT) cyclic loading. Hysteretic
energy loss EH generally reduces with cycle number n in-line with the increase in
secant stiffness kn with n; indeed, Abadie (2015) showed that EH can be approximated
Figure 4.15: Secant stiffness evolution under partial 2-way and 2-way loading with individual
power-law fits (Equation 4.4) shown dashed
Figure 4.16: Example response to 1-way loading in dense sand highlighting energy dissipation
with shaded areas equal to hysteretic energy loss EH at cycles n = 1 and n = 10 (test D.OW.03)
as a linear function of 1/kn . EH will also decrease with reducing ratcheting rate, which
is, in general, aligned with increasing secant stiffness.
The evolution of hysteretic energy loss EH with n at high cycle numbers is greater
under 2-way and partial 2-way loading than under 1-way and partial 1-way loading,
consistent with secant stiffness observations. However, for many of the 1-way tests
EH plateaus at high cycle number, which is not consistent with the observed increase
in secant stiffness and decrease in ratcheting rate with cycle number. This behaviour
is understood to be caused by the contribution from transducer friction, which is

Figure 4.17: Evolution of hysteretic energy loss EH with cycle number n under unidirectional
cyclic loading
reduced but not eradicated by the control system (Section 2.4.4). Figure 4.18 presents an
example hysteresis loop for test D.OW.01 at n = 1000, to highlight the square response
at load reversals — and systematic increase in recorded hysteretic energy loss EH —
caused by the transducer friction.
Energy dissipation is quantified by the energy loss factor per cycle ηn (defined in
Section 3.5.5) which is proportional to the hysteretic energy loss per cycle EH
normalised by the elastic energy per cycle EE , which varies with secant stiffness kn .
Figure 4.19 presents the evolution of energy loss factor ηn with cycle number n. Under
2-way and partial 2-way loading, and for 1-way and partial 1-way loading for n ≤ 100,
ηn tends to reduce with cycle number as EH does. However, where EH plateaus due to
transducer friction contributions, ηn increases: as the secant stiffness kn increases and
therefore EE decreases. In general, Figure 4.19 shows the significant impact cyclic
loading can have on energy dissipation. For example, during tests L.OW.04 and
Figure 4.18: Hysteresis loop for test D.OW.01 at n = 1000 to highlight the impact of transducer
friction on the hysteresis loop shape (squared response at load reversals)
Figure 4.19: Evolution of energy loss factor ηn with cycle number n under unidirectional cyclic
loading
D.OW.03 ηn reduces by a factor of 3 over 1000 cycles.
Abadie (2015) explored the evolution of energy loss factor (referred to as “damping”,
but proportional to ηn here for 1-way loading) with cycle number n and fitted a power-
law with constant exponent to its evolution under 1-way and partial 1-way loading. A
power-law would also approximately capture the evolution of ηn reported here under 2-
way and partial 2-way loading, but it would not be appropriate to fit a power-law to the 1-
way data distorted by contributions from transducer friction. Empirical expressions for
the evolution of ηn with n are not considered further here, in part due to the distortion
of the 1-way data by transducer friction, and in part because ηn is considered to be
a secondary parameter that is largely dependent on secant stiffness and ratcheting
magnitude, the evolution of which have been explored in previous Sections.
4.4.4 Uplift
Significant vertical uplift was observed during test D.TW.04.W0, which was conducted
in dense sand with large amplitude (ζb = 0.4) 2-way loading. Similar uplift behaviour
was observed by Niemann et al. (2018), however, minimal vertical pile movement
(|z| < 1 mm) was observed for the other tests presented here, in very loose sand or at
smaller cyclic amplitudes. Observation of uplift prompted the performance of tests
D.TW.04.W2 and D.TW.04.W5 with the same loading, but with additional 1.7 kg and 4.8
kg vertical mass applied, respectively. Figure 4.20a shows the vertical displacement of
the monopile during these three tests. More than 25mm vertical uplift is observed at n =
1000 without additional vertical load applied, but as expected, increasing the vertical
load suppresses the uplift, and with 4.8 kg additional mass 1.6mm downward settlement
is observed. Figure 4.20b shows the impact of uplift on the secant stiffness kn . For n ≤
100, kn is comparable across the three tests but kn reduces rapidly for test D.TW.04.W0
as the cycles progress and the pile moves upwards — reducing pile embedment and
increasing the loading eccentricity. Two potential mechanisms which may have driven
the uplift behaviour are discussed below and shown schematically in Figure 4.21:
Climbing mechanism As the pile rotates, the horizontal stresses on the pile’s passive
side will be greater than on the active side. This could cause the pile’s rotation
(a) Vertical displacement (b) Secant stiffness
Figure 4.20: Response to 2-way cyclic loading at ζb = 0.4 in dense sand with additional vertical
masses (Wn)
point to move laterally away from the pile centreline, allowing an upwards
climbing motion to develop during cycling. A similar mechanism was suggested
by Villalobos (2006) for uplift observed in tests on laboratory-scale caissons.
Unfortunately, noise levels in the measured pile pose were too large to determine
and track the pile’s instantaneous rotation point with confidence.
Sand migration Migration of sand particles beneath the pile tip could also force the
pile upwards. A low-stress region is likely to be generated beneath the pile base on
the active side as the pile rotates. Particles may migrate to this low-stress region,
incrementally increasing the soil volume beneath the pile and forcing it upwards.
Cuéllar et al. (2012) and Nicolai et al. (2017a) both observe sand particles moving
downwards at the pile face, but neither focus on particle migration at the pile
base.
The vertical load applied during test D.TW.04.W0 (and all other tests at OU except
D.TW.04.W2 and D.TW.04.W5) was equal to the weight of the pile and pile cap: 49.2 N
(5.02 kg). However, the weight of these components was largely incidental to other
design considerations. It is therefore worth considering what vertical load may be
representative for a monopile in the field. Assuming a prototype-scale RNA weight of
4.7 MN (Wind Turbine Models, 2015), a tower weight of 5.6 MN (6 m mean diameter, 40
2
1
Climbing mechanism Sand migration
Figure 4.21: Potential uplift mechanisms
mm mean wall thickness, 95 m height, steel) and a buoyant monopile weight of 8.3 MN
(8 m diameter, 80 mm mean wall thickness, 60 m total length, steel), the
prototype-scale vertical load is estimated as VP = 18.6 MN. Employing the
dimensionless framework of Leblanc et al. (2010a) (Table 3.4, Section 3.4.3) then gives a
representative laboratory-scale vertical load of VL = 27.9 N. The typical vertical load
applied in the laboratory is around 80% larger than VL ; applying additional vertical
loading (as done for tests D.TW.04.W2 and D.TW.04.W5) is therefore expected to reduce
the representativeness of the vertical loading further.
Although uplift was observed in the laboratory with greater vertical loading than
expected for equivalent full-scale monopiles, no known uplift has been reported in the
field. The rotations generated under loading to ζb = 0.4 in the laboratory are
significantly larger than the equivalent rotations expected for full-scale monopiles (as
discussed in Section 4.2). Given that the suggested uplift mechanisms will become
more pronounced at greater pile rotations, this may explain the presence of the
observed behaviour in the laboratory. Villalobos (2006) also suggested that the higher
stress-level in the field may increase wall friction and reduce uplift, relative to
laboratory tests at low stress-level.
4.5 Multidirectional hysteretic response
Figure 4.22 presents the load path and monopile response under multidirectional spiral
loading centred on zero load (L.SP.1, D.SP.1). Under spiral loading, the load magnitude
q
|M | = Mx2 + My2 increases at a constant rate while the resultant loading direction
moves around 360° per cycle. In very loose sand 5 cycles were applied up to ζb = 0.60,
while in dense sand 4 cycles were applied up to ζb = 0.45. These tests are not intended
to represent loading experienced by a monopile, but instead present a challenge for
multidirectional numerical modelling.
Qualitatively, the response is very similar at both sand densities, with the rotation
responses also tracing spiral shapes, as expected. The response in the dense sample is
slightly elongated in the x-direction, indicating the presence of some sample
inhomogeneity. The moment-rotation responses in the x- and y-directions resemble
those under unidirectional 2-way loading with increasing load amplitude, which
broadly adhere to the extended Masing rules (Figure 4.5). However, the interaction
between load components here generates a smooth response at the load reversals in
both the x- and y-directions.
4.6 Multidirectional high cycle response
This Section explores the response of the monopile to regular, multidirectional loading
cycles. Novel perpendicular loading tests provide fundamental insight into the
monopile response, while fan-type tests allow systematic investigation of the impact of
the spread of cyclic loading direction. The tests were chosen to elucidate
multidirectional behaviour rather than represent realistic loading, although the
fan-type loading may represent a simplified storm with varying direction and the
perpendicular tests may represent misaligned (approximately constant) wind and
(cyclic) wave loading.
4.6.1 Perpendicular cyclic loading
Figure 4.23 describes the T-shape and L-shape perpendicular loading applied, relative
to corresponding unidirectional loading. T-shape tests have mean load perpendicular
to the cyclic loading direction, while L-shape tests have equal mean load applied both
perpendicular to and aligned with the cyclic loading direction. Load is first increased
in the x-direction before cyclic loading is applied in the y-direction.

(a) Very loose sand
(b) Dense sand
Figure 4.22: Multidirectional spiral loading and response (backbone curve shown in grey)
My My My
Mx Mx Mx
Unidirectional T-shape L-shape
Standard mean
MCYC MAV
High mean
Figure 4.23: Description of perpendicular cyclic loading
Sand Unidirectional T-shape L-shape

Description MAV /MR MCY C /MR
density test test test
Very Standard
0.2 0.4 L.OW.04 L.T.02 L.L.04
loose mean (SM)
Very High mean
0.4 0.4 L.PO.633 L.TH.04 -
loose (HM)
Standard
Dense 0.1 0.2 D.OW.02 D.T.01 D.L.02
mean (SM)
High mean
Dense 0.2 0.2 D.PO.333 D.TH.02 -
(HM)
Table 4.4: Summary of unidirectional and perpendicular tests designed for comparison
Exploration of the impact of ζb and ζc on the unidirectional ratcheting response and
consideration of the mechanisms driving ratcheting in Section 4.4.1 suggest that the
cyclic amplitude MCY C and mean amplitude MAV play an important role in
controlling the cyclic response. To explore this further, tests at the same MCY C and
MAV are compared here. The impact of mean load MAV is also explored with
unidirectional and T-shape tests conducted at two different MAV values at each
density: high mean (HM) and standard mean (SM). Table 4.4 summarises the sets of
relevant tests. The responses to unidirectional loading (L.OW.04, L.PO.633, D.OW.02,
D.PO.333) are presented throughout this Section for comparison, but were previously
presented in Section 4.4.
Figure 4.24 presents the continuous rotation responses for the unidirectional and
perpendicular tests summarised in Table 4.4. The pile rotates broadly in the direction
of MAV regardless of the cyclic loading direction: the unidirectional tests move in the
Figure 4.24: Total rotation response under unidirectional and perpendicular cyclic loading
(note different scales; response to first 1000 cycles shown)
y-direction, the T-shape tests move in the x-direction and the L-shape tests move at
approximately 45° to the axes. The impact of load bias is also clear in Figure 4.24, with
the HM tests accumulating significantly more rotation than the SM tests.
Ratcheting
Figure 4.25 presents the ratcheting response of the monopile in the x-direction for the T-
shape tests and in both the x- and y-directions for the L-shape tests. The corresponding
unidirectional responses are also plotted. The ratcheting magnitude and evolution for
the L-shape and T-shape tests is similar to that for the corresponding unidirectional
tests, suggesting that — to a first approximation — ratcheting is independent of the
cyclic loading direction for a given MCY C and MAV . More subtle variation in the
ratcheting response is explored in Figures 4.26 and 4.27.
Following the approach of Truong et al. (2019), power-laws (Equation 4.1) are fitted
to the evolution of ∆θn for each individual test, as shown dashed in Figure 4.25. The
power-law expression captures the ratcheting response accurately under L-shape
loading in both directions and for T-shape loading for n > 10. The shape of ratcheting
evolution under T-shape loading is similar to that under partial 2-way loading (Figure
4.7); these loadings both pass through zero load (ζc < 0).
Figure 4.25: Ratcheting response under perpendicular cyclic loading with individual
power-law fits (Equation 4.1) shown dashed
The variation of power-law exponent α with load type is presented in Figure 4.26,
with α normalised by the power-law exponent for the corresponding unidirectional test
αU D . The exponents obtained in very loose sand (1.04 ≤ α/αU D ≤ 1.47) are a little
higher than those in dense sand (0.90 ≤ α/αU D ≤ 1.24). It is also significant that the
exponents under L-shape loading in the x-direction are smaller than the exponents in
the y-direction. Figure 4.27 provides complementary information on the impact of
load type on ratcheting, with accumulated rotation ∆θn normalised by accumulated
rotation for the corresponding unidirectional test ∆θn,U D . The smaller accumulated
rotation observed in the x-direction (perpendicular to cycling) under L-shape loading,
coupled with the observation of smaller α values in this direction, is indicative of
non-linear dependency of ratcheting on load magnitude (although the mean loads are
the same in the x- and y-directions, the peak load in the y-direction is twice that in the
x-direction). This observation is in-line with the observed power-law variation of
ratcheting magnitude coefficient Tb with cyclic amplitude for unidirectional, 1-way
loading ζb (Figure 4.8).
Figure 4.24 highlighted the significant impact of MAV on the rotation response. To
further assess the impact of MAV , a ratio of accumulated rotation for the HM tests
∆θn,HM to that for the SM tests ∆θn,SM is plotted in Figure 4.28. Given that MAV,HM =
Figure 4.26: Variation of power-law exponent α with perpendicular load type
Figure 4.27: Accumulated rotation under perpendicular cyclic loading normalised by

accumulated rotation for corresponding unidirectional tests
2MAV,SM , values of ∆θn,HM /2∆θn,SM may be expected to equal unity if ratcheting is
proportional to MAV . Figure 4.28 shows how ratcheting may be proportional to MAV
in very loose sand at high cycle numbers, but in dense sand the impact of MAV is
less significant. The variation with density might be explained by the difference in
linearity of the backbone curves: the ratio of rotation generated under monotonic
loading up to MAV under HM loading to that under SM loading (θAV,HM /θAV,SM ) is
around 6 in very loose sand but only around 3 in dense sand. As previously discussed,
greater rotations lead to a greater likelihood of particle rearrangement and therefore

Figure 4.28: Accumulated rotation for high mean (HM) tests normalised by accumulated
rotation for standard mean (SM) tests
ratcheting. To provide further insight, these tests could be repeated with a constant
ratio of θAV,HM /θAV,SM . It should be noted that the comparisons of unidirectional tests
L.OW.04 & L.PO.633 and D.OW.02 & D.PO.333 are included in Figure 4.28 given the
focus on the impact of MAV and MCY C in this Section.
Secant stiffness
The evolution of secant stiffness kn /kM AX with cycle number n under perpendicular
loading is plotted in Figure 4.29 (stiffness is only relevant in the cycling direction, y).
Stiffness evolves approximately logarithmically (following Equation 4.2) under T-shape
and L-shape loading, as shown by the dashed logarithmic fits in Figure 4.29. The
addition of mean load MAV perpendicular to the cyclic loading direction appears to
have little impact on the stiffness: the L-shape secant stiffness follows the
corresponding unidirectional stiffness closely at both densities, while the stiffness for
the HM T-shape tests is similar to that for the SM tests. It is interesting that the stiffness
under T-shape loading evolves logarithmically, as a power-law better captured the
evolution of kn under 2-way unidirectional loading, which is equivalent to T-shape
loading in the y-direction.

Figure 4.29: Secant stiffness evolution under perpendicular cyclic loading with logarithmic fits
(Equation 4.2) shown dashed
Energy dissipation
The evolution of energy loss factor ηn with cycle number n is shown for the
perpendicular tests in Figure 4.30 (ηn is only relevant in the cycling direction, y). The
values of ηn for n > 20 are remarkably similar for the L-shape and T-shape tests at both
mean load MAV amplitudes, consistent with the similarity of secant stiffness kn values.
For n > 20, ηn for the perpendicular tests is greater than for the equivalent
unidirectional tests, despite similar values for ratcheting and secant stiffness. For
example, ηn=1000 for the perpendicular tests is at least 1.6 times larger than ηn=1000 for
the corresponding unidirectional tests. Figure 4.31 shows the hysteretic response at
n = 1000 for an example unidirectional test and an equivalent L-shape test; the loop
area, equivalent to hysteretic energy loss EH , is evidently larger for the L-shape test.
The increase in EH and therefore ηn under perpendicular cyclic loading may be
explained by the additional soil volume disturbed, which would increase dissipation.
Overview
The broad similarity in the cyclic response under unidirectional, T-shape and L-shape
loading at a given MCY C and MAV demonstrates the insensitivity of the response to the
direction of cyclic loading (relative to mean load direction), whilst also highlighting the
Figure 4.30: Evolution of energy loss factor under perpendicular cyclic loading
Figure 4.31: Example hysteretic response in dense sand under unidirectional and L-shape
loading at n = 1000 to highlight difference in hysteretic loop shape
importance of MCY C and MAV in controlling the cyclic response. Although
characterisation of cyclic loading in terms of ζb and ζc is useful, using these parameters
alone does not reveal the full picture. For example, using ζb and ζc does not reveal the
similarities in loading between a T-shape test (ζc,y = −1) and a corresponding
unidirectional test (ζc = 0). Comparison of SM and HM tests reveals that MAV affects
the magnitude of ratcheting, but does not significantly affect stiffness or energy
dissipation.
4.6.2 Fan-type cyclic loading
Figure 4.32 describes the fan-type loading applied in Mx − My space, which allows
exploration of the impact of the spread of cyclic loading direction. The loading traces the
sector of a circle with radius ζb MR and half internal spread angle Φ. 1-way cyclic loading
is applied within this sector at a frequency f , while the loading direction changes
sinusoidally at a frequency f /100, so that 1000 cycles corresponds to 10 complete
sweeps of the sector. Loading starts and finishes in the x-direction. Equations 4.5 and
4.6 define Mx (t) and My (t) for a maximum load ζb MR and cyclic loading frequency
f . This loading closely follows the multidirectional loading applied by Dührkop and
Grabe (2008), Rudolph et al. (2014b), and in other related work.
ζb MR 2πf t

Mx (t) = (1 + sin (2πf t)) cos Φ sin (4.5)
2 100
ζb MR 2πf t

My (t) = (1 + sin (2πf t)) sin Φ sin (4.6)
2 100
Seven fan-type tests were performed in very loose sand with 15° ≤ Φ ≤ 150°, while
two tests were performed in dense sand with Φ = 30° and Φ = 90° (as summarised in
Tables 4.1 and 4.2). In practice, loading directionality is site-specific and depends on
the time-scale considered. For a single storm with a following sea state, a spreading
angle of around 25° may be typical. However, loading direction will vary more
significantly over the structure’s lifetime, as shown for an example site offshore the
Netherlands in Figure 1.8 (Chapter 1).
All loose tests were conducted at ζb = 0.4 and all dense tests at ζb = 0.2; the fan-type
tests are therefore comparable to unidirectional tests L.OW.04 and D.OW.02
respectively. Figure 4.33 shows the rotation response for an example fan-type test in
very loose sand with Φ = 60°. Rotation accumulates in the x-direction, but significant
transient rotation also occurs in the y-direction as the loading direction changes over
each sweep of the sector.

My
Φ
ζbMR Mx
Figure 4.32: Description of fan-type

loading
Figure 4.33: Example rotation response

under fan-type loading (test L.F.60)
(b) Dense sand
(a) Very loose sand
Figure 4.34: Ratcheting response under fan-type cyclic loading with individual power-law fits
Figure 4.35: Variation of α with spread angle Φ under fan-type loading
Ratcheting
Figure 4.34 presents the accumulated rotation ∆|θ|n response under fan-type loading
alongside the response to corresponding unidirectional loading. The modulus notation
indicates that the data presented has been obtained from the absolute moment-rotation
q q
response (|M | = Mx2 + My2 , |θ| = θx2 + θy2 ), which is appropriate given that the
absolute load |M | has constant amplitude throughout the tests. The secant stiffness
|k|n and the energy loss factor |η|n are also determined from the absolute response.
As the loading direction changes, the rate of evolution of ratcheting varies — seen as
ripples in Figure 4.34. The magnitude of these ripples increases with increasing spread
angle Φ, and for Φ ≥ 90° significant negative ratcheting is observed transiently where
the load in the x-direction goes negative (σAV,x < 0). The spread angle also impacts the
overall ratcheting evolution, and is assessed by fitting power-law expressions (Equation
4.1) to each fan-type test, as shown dashed in Figure 4.34. The coefficients A are set
equal to the values for the corresponding unidirectional tests as the initial loading cycle
is unidirectional. The variation of exponent α is plotted in Figure 4.35 normalised by
the corresponding unidirectional exponent αU D . For 0° ≤ Φ ≤ 90°, α increases with
spread angle Φ and in very loose sand at Φ = 90° , α/αU D = 1.46. However, α decreases
at larger Φ values (observed in very loose sand only).

Cycle Strain Worst Multidirectional

Study Test conditions
No. N metric ε angle Φ∗ factor εN,F /εN,U D
1g, medium-
Dührkop and dense sand 50000 Pile head 45° 13
Grabe (2008) displacement
1g, dense sand 50000 45° 2−3
1g, medium-
10000 90° 1.2
dense sand Displacement
Rudolph et al.
1g, dense sand 10000 at lowest 120° 1.4
(2014a)
200g, medium- LVDT
3000 30° 1.7
dense sand
200g, dense
3000 90° 1.5
sand
1g, very loose

sand
1000 Accumulated 90° 2.3
This study
pile rotation
1g, dense sand 1000 90° 1.8
Table 4.5: Comparison of worst-case spread angles and multidirectional factors for ratcheting
response under fan-type loading (this study and similar studies)
The impact of spread angle Φ on ratcheting can also be assessed with the ratio of
strain at cycle N under fan-type loading εN,F to that under corresponding
unidirectional loading εN,U D . This multidirectional factor is summarised for this study
and similar studies by Dührkop and Grabe (2008) and Rudolph et al. (2014a) in Table
4.5. Whilst an increase in strain under fan-type loading relative to unidirectional
loading is observed in all cases, there is significant variation in both the spread angle
which causes the greatest ratcheting (Φ∗ ) and the multidirectional factor (εN,F /εN,U D ).
Ratcheting behaviour is likely to vary with the specific load regime (e.g. rate of change
of load direction), as reported by Rudolph et al. (2014b). Application of a blanket
multidirectional factor on ratcheting is therefore not recommended, but these results
together highlight the need to account for changing cyclic loading direction to ensure
that ratcheting is not underestimated.
In a similar way to the impact of ζc on ratcheting under unidirectional cyclic
loading, the impact of spread angle Φ on ratcheting may be understood as a
competition between (a) increasing pile and therefore soil movement with increasing
Φ, and (b) decreasing mean load in the dominant loading direction with increasing Φ.
Figure 4.36: Secant stiffness evolution under fan-type cyclic loading with logarithmic fits
Secant stiffness
The evolution of secant stiffness of the absolute response |k|n is plotted in Figure
4.36 for the fan-type and corresponding unidirectional cyclic tests. Secant stiffness
varies with the loading direction, causing ripples similar to those observed in the
ratcheting response. These ripples become very large for Φ ≥ 90°, as |k|n passes
through theoretically infinite stiffness and becomes negative when the direction of
ratcheting accumulation reverses. The responses for Φ ≥ 90° therefore cannot be
presented on the axes used in Figure 4.36.
Logarithmic fits (Equation 4.2) approximately capture the overall evolution of secant
stiffness for Φ < 90°, as shown in Figure 4.36. The impact of spread angle Φ is assessed
by plotting the variation of coefficient BL (which controls the rate of change of stiffness,
see Equation 4.2), normalised by the unidirectional coefficient BL,U D , against Φ in
Figure 4.37. There is a clear trend for increasing BL , and therefore rate of stiffening,
with spread angle Φ. This trend may be expected: as Φ increases, the volume of soil
likely to densify under cyclic loading also increases.
Energy dissipation
The energy dissipation under fan-type loading is presented in Figure 4.38, with the
energy loss factor of the absolute response |η|n plotted against cycle number n. Data
Figure 4.37: Variation of BL with spread angle Φ under fan-type loading
Figure 4.38: Evolution of energy loss factor under fan-type cyclic loading
for Φ ≥ 90° is not presented, given that |η|n → 0 for these tests when the direction of
ratcheting accumulation reverses. Again, ripples in the |η|n response are observed as
the loading direction changes. The overall energy dissipation follows a similar trend to
that for the corresponding unidirectional tests, although |η|n is consistently larger at
given cycle number n for tests with larger spread angles. This trend is explained by the
increase in soil volume disturbed as spread angle Φ increases, which is likely to
increase dissipation. A similar observation on energy dissipation was made for
perpendicular loading.
Figure 4.39: Post-cyclic reloading responses (mean monotonic response shown in grey)
4.7 Post-cyclic reloading response
Post-cyclic reloading was performed as part of many of the cyclic tests, as identified
in Tables 4.1 and 4.2. The post-cyclic response is analogous to the response which
might be expected during the ULS event, and can also inform interpretation of irregular
multi-amplitude loading. Figure 4.39 presents the reloading responses with the cyclic
responses omitted for clarity. In many cases the reloading responses exceed the mean
monotonic response (backbone curve), indicating an increase in capacity. In other
cases the reloading responses approach the backbone curve. The greatest exceedance of
the backbone curve occurs for large amplitude 2-way tests (L.TW.04, D.TW.032), where
significant secant stiffening but relatively little ratcheting occurred. The reloading
responses are also plotted in Figure 4.40 for M/MR < 0.5, with the rotation zeroed
from the onset of reloading (at θ = θr0 ). This plot highlights the increase in stiffness
relative to the backbone curve, particularly for M/MR < ζb , which is consistent with
Figure 4.40: Post-cyclic reloading responses with rotation zeroed from the onset of reloading at
θr0 (mean monotonic response shown in grey)
the observed increase in secant stiffness under cyclic loading.
As discussed throughout this Chapter, ratcheting and secant stiffening often occur
in parallel under cyclic loading in dry sand. Ratcheting moves the reloading responses
beneath the backbone curve and increases pile rotation on reloading, while the
densification driving secant stiffening also increases stiffness beyond the region
strained under cyclic loading, leading to a decrease in pile rotation on reloading. These
processes are therefore in competition, as discussed by Abadie et al. (2019b). Under
some conditions, the balance of ratcheting and stiffening processes may lead to no
significant change in the response at large loads, as observed by Abadie et al. (2019b)
and for many of the 1-way tests presented here. Where stiffening processes dominate,
such as for the large amplitude 2-way tests presented here, an increase in post-cyclic
monotonic capacity can be observed. Truong et al. (2019) and Nicolai et al. (2017b) also
report an increase in post-cyclic monotonic capacity, and Nicolai et al. (2017b) are
further able to quantify the increase in capacity in terms of cycle number, ζc and ζb ,
given that their reloading tests reach clear yield.
The conventional cyclic degradation approaches apply reduction factors to p-y
curves to reduce the pile’s lateral capacity (Section 1.2.5), consistent with the idea that
cyclic loading is an intrinsically damaging process. Whilst this may be true for clays,
the current experimental evidence does not support a decrease in post-cyclic capacity
in (fully drained) sands, and under some conditions an increase in capacity is observed
which may be exploited to optimise monopile design.
4.8 Summary
This Chapter has explored the response of a model monopile foundation to regular,
unidirectional and multidirectional cyclic lateral loading in dry sand at 1g. The following
points summarise the key observations, which are, in general, applicable to both very
loose and dense sand.
Unidirectional hysteretic response (Section 4.3)
• The hysteretic response under 2-way loading broadly adheres to the extended
Masing rules for the first few (n < 4) loading cycles.
• Behaviour indicative of gapping is observed under high amplitude 2-way loading
in dense sand.
Unidirectional high cycle response (Section 4.4)
• Accumulation of pile rotation (ratcheting) occurs under biased cyclic loading and
tends to evolve as a power-law with cycle number.
• The magnitude of ratcheting increases as a power-law with cyclic amplitude
(characterised by ζb for constant ζc ).
• The ratcheting shape, rate and therefore magnitude is impacted by load
asymmetry. Ratcheting increases under partial 2-way loading where loading
passes through zero load (ζc < 0).

• Secant stiffness increases under cyclic loading. A logarithmic relationship
captures the evolution of stiffness under 1-way and partial 1-way loading, but a
power-law captures the behaviour better under 2-way and partial 2-way loading,
where loading passes through zero load (ζc < 0).
• Energy dissipation varies significantly with cyclic loading as the secant stiffness
and ratcheting rate change.
• Uplift is possible under large amplitude cyclic loading, even with larger than
representative vertical loading.
Multidirectional hysteretic response (Section 4.5)
• The hysteretic response under multidirectional spiral loading in either the x-
or y-direction resembles that under equivalent unidirectional loading (which
adheres to the extended Masing rules) although some interaction between loading
components is observed.
Multidirectional high cycle response (Section 4.6)
• The response to unidirectional, T-shape and L-shape cyclic loading at a given
cyclic amplitude MCY C and mean load MAV is similar, demonstrating insensitivity
of the response to the cyclic loading direction (relative to the mean load direction)
and highlighting the importance of parameters MAV and MCY C in controlling the
cyclic response.
• Ratcheting occurs in the direction of the mean load, regardless of the cyclic
loading direction.
• The spread of loading directions has a significant impact on the evolution of
ratcheting and stiffening under fan-type loading. For the most damaging spread
angle, accumulated pile rotation at n = 1000 is 2.3 and 1.8 times larger than the
rotation under corresponding unidirectional loading, in very loose and dense
sand respectively.
• Energy dissipation under perpendicular loading and fan-type loading is greater
than under corresponding unidirectional loading.

Post-cyclic reloading response (Section 4.7)
• The reloading response is understood to be controlled by the ratcheting and
stiffening processes which occur under cyclic loading in dry sand and have
opposing effects on the reloading response.
• The inferred post-cyclic capacity is equal to or greater than the monotonic
capacity for all applied load cases, at odds with the cyclic degradation approach
(see Section 1.2.5).
• The reloading response tends to the backbone curve following 1-way cyclic
loading.
In summary, the monopile response adheres to the extended Masing rules for the
first few cycles, but evolution of secant stiffness and associated energy dissipation
are observed under many cycles. Ratcheting also occurs under many cycles of biased
loading. The behaviour observed under unidirectional loading is generally consistent
with observations from previous studies (e.g. Abadie et al., 2019b; Truong et al., 2019),
while multidirectional tests provide new insights. Fan-type tests highlight the significant
impact of the spread of loading direction on both ratcheting and stiffening behaviour,
while perpendicular tests reveal the insensitivity of the cyclic response to cyclic load
direction (relative to the mean load direction) and confirm that ratcheting occurs
in the direction of the mean load.
The energy dissipation evolution and reloading responses are mostly explicable
in terms of ratcheting and stiffening behaviour, and are therefore considered to be
secondary, but no less important, phenomena. The observation that the post-cyclic
capacity is equal to or greater than the monotonic capacity has important implications
for ULS design, and contrasts with the conventional cyclic degradation approach.
Throughout, the observed behaviour was interpreted in terms of the densification
and convective mechanisms understood to be driving stiffening and ratcheting at the
macro scale, as discussed in Section 1.3.2. In general, increased ratcheting (magnitude
and evolution) and stiffening (rate) coincided with an increased potential for soil
particle rearrangement, which is necessary for both densification and convective
mechanisms. Increased soil particle rearrangement is likely with increased cyclic load
amplitude, increased loading multidirectionality, and where loading passes through
zero load.
As well as providing fundamental insight into the response of a monopile to cyclic
lateral loading, the observations in this Chapter i) provide a basis for interpretation of
the response of the monopile to realistic multi-amplitude, multidirectional storm
loading in Chapter 5, and ii) inform development and calibration of models in the
HARM framework in Chapter 7.

Chapter 5
Application of realistic storm

loading at 1g
5.1 Introduction
Regular cyclic loading tests, such as those presented in Chapter 4, allow systematic
investigation of the cyclic response of monopile foundations. But in reality, monopiles
supporting OWTs are exposed to continuously varying cyclic loading. This Chapter
explores the response to more realistic, multi-amplitude and multidirectional, storm
loading. The results provide novel insight into the response of monopiles and allow
parallels to be drawn with the regular cyclic loading response. The results also provide
valuable data for validation of numerical models, such as presented in Chapter 7.
The tests were conducted in dry sand at 1g using the apparatus described in Chapter
2, with the test procedures outlined in Chapter 3. The applied loads were derived from
wave basin tests performed as part of the DeRisk project (Bredmose et al., 2016). The
processing steps required for appropriate application of these loads to the geotechnical
model monopile are discussed in detail. Understanding the characteristics of realistic
storm loading is also an important aspect of this work.
5.2 DeRisk wave tank tests (Bredmose et al., 2016)
The DeRisk project aimed to reduce the risk associated with predicting ULS wave loads
on OWT structures (Bredmose et al., 2016). The multi-centre project was funded by
Innovation Fund Denmark and conducted over the period 2015 to 2019. As part of the
113
5. Application of realistic storm loading at 1g 114
Figure 5.1: High speed photograph from DeRisk wave basin testing showing model monopile
and wave (Bredmose et al., 2016)
project, wave loading tests were performed on model monopiles in a shallow water
basin at the Danish Hydraulic Institute (DHI), as shown in Figure 5.1.
An approximately rigid cylinder with diameter 140 mm was used to represent a 7 m
diameter prototype monopile structure in the water column (1 : 50 scale). The cylinder
was instrumented with (amongst other instruments) four load cells for resolution of
the applied horizontal and moment load in two orthogonal directions (x, y). Various
sea states were generated, representing medium to severe storms in the North Sea. The
sea states chosen for deriving loading for the geotechnical model monopile were in a
prototype water depth of 33m, with peak spectral period 15s and significant wave height
9.5 m, approximately corresponding to a 100-year return period. Loads were derived
from both a unidirectional (UD) and multidirectional (MD) sea state. The waves in the
unidirectional sea state were all aligned with the x-direction. Meanwhile, the waves
in the multidirectional sea state varied in direction (around x) with spreading angle
Φ = 22°, where Φ is the standard deviation of a wrapped normal spreading distribution
(e.g. Adcock and Taylor, 2009). This spreading angle may be typical for a following
sea state, where one weather system is present.
Deriving loading from measured wave data introduces non-Gaussian skewness in
loads that is representative of loading in the field. Wave skewness affects the mean and
peak wave loads, both of which play an important role in determining the monopile
response. This compares to the use of simplified wave models which underestimate
wave skewness (e.g. Wang, 2018), and linear wave models which predict symmetric
waves with zero mean load.
5.3 Wave load processing and test programme
The wave loads were provided by the DeRisk team at prototype-scale with moment
M and horizontal load H components in the x-direction for the UD sea state and in
the x- and y-directions for the MD sea state. Storms of up to 70 hours in length (at
prototype scale) were generated in the wave basin, but only the first 6 hours of data
were used here. A number of processing steps were necessary before application of
the wave loads to the geotechnical model:
1. Application of a transfer function (TF) to approximate the dynamic response of
an OWT structure.
2. Addition of constant wind loading in the x-direction to better model combined
environmental loading.
3. Projection of loads to a constant loading eccentricity to allow application with the
apparatus described in Chapter 2.
4. Scaling of the prototype loads for application to the laboratory-scale geotechnical
model.
Figure 5.2 shows the effect of processing steps 1-3 on the first 6 hours of prototype
wave loads for the MD sea state.
5.3.1 Application of transfer function
Offshore wind turbine structures on monopile foundations typically have first natural
frequencies close to the frequency content of waves (Bachynski et al., 2017), whereas
the natural frequency of the approximately rigid cylinder used in the wave basin
experiments was much higher (f0 ≈ 5.7 Hz). Dynamic amplification of loads will
therefore be significant for the real structure but negligible for the rigid cylinder. To
account for dynamic amplification of loads, the OWT structure was approximated as a
single degree of freedom system with a natural frequency f0 and damping ratio ξ. The
Figure 5.2: Impact of wave processing steps 1) to 3) on prototype-scale loads for the MD sea
state (upper plots show first six hours of the sea state, lower plots highlight arbitrary 100 s of
loading indicated by grey shaded region in upper plots)
ratio of transmitted force FT to excitation force FE can then be found with the transfer
function (TF):
FT 1
= s (5.1)
FE 2 2 2
f
1− f0 + 2ξ ff0
Dynamic amplification of loads will also occur at higher frequency modes (e.g. second
tower bending mode and blade bending modes) but this simple transfer function
captures the key dynamic behaviour of the OWT structure, and has been used by e.g.
Arany et al. (2017).
The first natural frequency of the prototype structure was estimated as f0 = 0.26 Hz,
between the 1P and 3P blade passing excitation frequencies for a Vestas V164-8.0MW
turbine (University of Strathclyde, 2015). Given that the sea states used represent
extreme storm conditions, the turbine is assumed to be parked. In parked conditions,

Figure 5.3: Impact of TF on frequency Figure 5.4: Illustration of projection of MD

content of MD moment loads in the loads in x-direction to constant loading
x-direction eccentricity
where aerodynamic damping is negligible, the total damping is estimated as ξ = 0.65%,
in-line with Kementzetzidis et al. (2019) for dense sand.
The transfer function was approximated as an arbitrary magnitude filter and applied
to loads in both directions in the frequency domain. Figure 5.3 demonstrates the impact
of the transfer function on the frequency content of the MD moment loads in the x-
direction, showing clearly the amplification of the signal around the natural frequency
of the structure. Figure 5.2(1) shows the impact of the transfer function on the MD
moment loads in the time domain. The close proximity of the structure’s natural
frequency f0 to the frequency content of the waves, coupled with the low damping
ratio ξ, leads to significant dynamic amplification of wave loads. The peak moment
amplitudes approximately double and the number of cycles increases by around 50%
(for both the UD and MD cases). The impact of this transfer function is consistent with
the experimental observations of Bachynski et al. (2017), who used an appropriately
flexible monopile to explore dynamic amplification of wave loads.
5.3.2 Addition of wind loading
Wind loading may be added to the prototype wave loads to better represent combined
environmental loading. A wind load of 1.4 MN acting at a height 85.5 m above the
mudline was estimated by assuming appropriate prototype OWT dimensions (water

depth 33 m, tower height 105 m, mean tower diameter 6 m), with the turbine parked and
blades feathered, and employing a 50-year design wind speed of 50 m/s, consistent with
the extreme storm conditions represented by the waves. The relatively low-frequency
fluctuations in wind amplitude were neglected, with the wind load approximated as a
constant load. Wind loading was aligned with the dominant wave loading direction (x).
Figure 5.2(2) shows how the addition of wind loading adds a constant load bias
in the x-direction and increases the overall mean load M̄ and peak moment load Mp .
Throughout, the monopile response is explored with and without wind loading.
5.3.3 Projection of loads to constant eccentricity
The measured wave loads have a variable eccentricity (ẽ = M/HL). However, for
loads to be applied using the loading system described in Chapter 2, the loads must be
projected to a constant eccentricity. Loads were projected to a load line in M − H space,
with a gradient representative of that applied by the laboratory apparatus (ẽ = 2.50), in
a direction parallel to an approximate monopile yield surface. The x- and y-direction
loads were projected independently.
By assuming a distributed lateral load per unit pile length of DKγ 0 z, assuming
K = 3Kp (Broms, 1964) and linearising the yield surface in the region of interest (where
M/H is positive), an approximate yield surface can be defined by:
M H
+ 0.75 = 0.29Kp (5.2)
L3 Dγ 0 L2 Dγ 0
Figure 5.4 illustrates the process of projecting the x-direction loads for the MD sea state
(assuming example prototype parameters D = 7 m, L = 28 m, γ 0 = 10 kN/m3 , φ0 = 40°)
while Figure 5.2(3) shows the impact on the moment loads for the MD case. The
eccentricity of the measured wave loads is smaller than the eccentricity applied by the
laboratory apparatus, thus the amplitude of moment loads increases upon projection.
5.3.4 Scaling wave loads
Lastly, the prototype-scale wave loads were scaled down to model-scale using the
dimensionless framework of Leblanc et al. (2010a) (Table 3.4, Section 3.4.3), by
equating expressions for H̃ and M̃ at laboratory- and prototype-scale. Parameters L, D

and γ 0 therefore required specification at prototype-scale. Although the wave basin
model represented a prototype monopile with diameter 7 m in the water column, from
a geotechnical design perspective the monopile size would vary with sand density.
Therefore, a prototype pile diameter of 6.9 m was chosen for tests in dense sand and a
prototype pile diameter of 9.7 m was chosen for tests in very loose sand; the aspect
ratio η = 4 at both densities.
The dimensionless framework accounts for the effect of stress-level on stiffness,
but does not account for the variation of friction angle with stress-level and density.
Peak friction angle φ0p is therefore matched between prototype- and laboratory-scale
using the empirical relations presented by Bolton (1987) (discussed in Section 3.2.3) to
determine DR and hence γ 0 at prototype-scale for each density. Table 5.1 summarises
the prototype parameters used to scale the wave loads at each density and the resulting
ratio of prototype-scale moment to laboratory-scale moment (fM = MP /ML ). The
sand is assumed to be saturated at prototype-scale.
3
Density Pile diameter D [m] Pile length L [m] Unit weight γ 0 [kN/m ] fM
Very loose 9.7 38.8 4.7 69.7 × 106

Dense 6.9 27.6 6.5 22.2 × 106
Table 5.1: Prototype-scale parameters used for scaling wave loads
5.3.5 Test programme
Table 5.2 summarises the realistic storm loading tests performed. The loading is
summarised in terms of the peak moment Mp and overall mean moment M̄ across the
test, in the x- and y-directions. The test names indicate sequentially: sand density,
source sea state, addition of wind loading, and loading direction details where
necessary. To obtain the cycle number N , extreme and reversal points were defined
where the peak prominence of M exceeds MR /100; cycles were then defined between
these points following Section 3.5.6.
The loading frequency was reduced relative to prototype-scale to ensure accurate
load control, and a greater reduction was necessary for the tests in dense sand, where
the load amplitudes were greater. To avoid tests running for an excessively long time
Test Source sea state Cycle Loads [/MR ]

Wind
name and details No. N
Mp,x M̄x Mp,y M̄y
Very loose sand samples
L.UD.0 UD No 5400 0.40 0.004 0 0

L.UD.W UD Yes 5400 0.46 0.070 0 0
L.MD.0 MD No 5400 0.49 0.002 0.12 0.002
MD, applied twice
L.MD.W.x Yes 10400 0.55 0.065 0.12 0.002
in x-direction.
MD, applied in x-direction 0.55 0.065 0.12 0.002
L.MD.W.45 Yes 10400
then at 45◦ to x (italic loads) 0.39 0.046 0.38 0.049
MD, only x-direction
L.MD.0.I No 5400 0.49 0.002 0 0
component applied
L.MD.W.I Yes 5400 0.55 0.065 0 0
component applied
Dense sand samples
D.MD.0 MD No 2000 0.42 0.002 0.1 0.002

D.MD.W MD Yes 2000 0.48 0.058 0.1 0.002
D.MD.W.I Yes 2000 0.48 0.058 0 0
component applied
Table 5.2: Realistic storm loading test programme at 1g
(> 18 hr), the first 6 prototype hours of sea state data were used for the very loose tests
and the first 3.4 prototype hours of data were used for the dense tests. Changing the
loading frequency was not expected to impact the results, given that significant rate
effects were not observed in the monotonic tests (Section 3.4.1).
Tests were conducted with (.W) and without (.0) wind loading to explore the effect
of the increase in load bias (see variations in M̄ and Mp ) caused by the wind load. Tests
identified with .I were derived from MD sea states, but applied loading in only the
x-direction; comparison of these tests with the equivalent multidirectional tests
therefore allows exploration of multidirectionality effects. Moment loading for example
tests L.MD.0 and L.MD.0.I is shown in Figure 5.5. Tests L.MD.W.x and L.MD.W.45
involve application of the MD sea state aligned with the x-direction, followed by
application of the same loading either aligned with (.x) or at 45◦ to (.45) the x-direction.
These tests allow exploration of the impact of consecutive storms and the impact of
changing storm direction.

Figure 5.5: Applied loading for multidirectional test L.MD.0 compared with loading for
corresponding unidirectional test L.MD.0.I
Figure 5.6 presents the distribution of applied MCY C /MR and MAV /MR (and
equivalently ζb and ζc , as shown in Figure 3.11, Section 3.5.1) in the x-direction for key
tests presented in Table 5.2. The distributions are presented in terms of the sum of
number of half-cycles ( n0.5 ) in a given bin (size 0.01M/MR ), on a logarithmic scale.

P
The figure shows how the addition of wind loading shifts the distribution on the
MAV /MR axis, increasing M̄ and leading to more 1-way and partial 1-way loading. The
largest half-cycles tend to be partial 2-way. Across the tests, the vast majority of loading
occurs at ζb < 0.25 and much of the loading occurs at ζb < 0.15, despite these tests
representing extreme storm conditions. Presentation of realistic multi-amplitude
loading in this way can help inform the design of focused physical modelling
programmes.
5.4 Storm loading response

5.4.1 General observations
The response of the monopile to multi-amplitude storm loading is presented in
moment-rotation space in the left and mid plots in Figures 5.7, 5.8 and 5.9. The
continuous responses show the shape of the hysteretic loops and may be compared to
the backbone curve (shown in grey in the plots). In general, the hysteretic responses
are consistent with those observed for regular cyclic loading (in Section 4.3), with some
Figure 5.6: Distribution of applied MCY C /MR and MAV /MR in the x-direction for
multi-amplitude storm loading tests (generated with software from T. D. Balaam)
gapping-type behaviour (inflexion in the moment-rotation responses around zero
load) observed under large amplitude loading in dense sand (Figures 5.7c and 5.8c).
Upon applying loads greater than those previously applied, the responses
approximately follow the backbone curve. This behaviour is consistent with one of the
two Masing rules extensions proposed by Pyke (1979) and with many of the reloading
responses observed following regular 1-way cycling, as shown in Section 4.7.
The evolution of pile rotation with cycle number is presented in the right-most
plots in Figures 5.7, 5.8 and 5.9. As described in Section 3.5.6, pile rotation per cycle
may be defined for multi-amplitude loading on either the loading or unloading

half-cycles, at the half-cycle average load. The rotation per cycle on loading (θna ) is
used here, although the observed behaviour is not affected by the choice of either
parameter. The rotation on loading θna responses are dominated by rotation
accumulated during the large load events, in agreement with observations from
Leblanc et al. (2010b) and Abadie et al. (2019b) under regular multi-amplitude cyclic
loading. However, rotation also accumulates between the large load events. This
behaviour is consistent with the ratcheting observations in Chapter 4: ratcheting
occurs under biased cyclic loading at all explored amplitudes, but its magnitude and
evolution is rather dependent on the loading characteristics.
Accumulated rotation (∆θn ) is not presented here, as the rotation during Cycle 0
depends on the (variable) amplitude of Cycle 0. It should also be noted that pile rotation
is plotted on linear scales in this Chapter, in contrast to the logarithmic scales used
in Chapters 4 and 6. The evolution of secant stiffness and energy loss factor are not
presented for these tests, as any changes in these parameters with cyclic loading are
conflated with the more significant impact of cyclic load amplitude on them, which
varies from cycle to cycle. However, the application of consecutive storms does provide
some insight into the evolution of secant stiffness and energy dissipation.
5.4.2 Impact of wind loading
Figure 5.7 specifically explores the impact of additional wind loading on the x-direction
response. Wind load increases M̄ and Mp but leaves MCY C unchanged. As Mp increases,
the peak rotations increase to an even greater degree, given the non-linearity of the
backbone curve. The ratcheting rate between large load events also increases, given
the increase in M̄ , and consistent with observations under regular cyclic loading. For
the tests presented, pile rotation on loading at cycle N (θN a ) increases by between 2.5
and 3.5 times with the addition of wind loading.
5.4.3 Impact of storm multidirectionality
Figure 5.8 highlights the impact of storm multidirectionality by presenting the response
to multidirectional tests derived from MD sea states with spreading angle 22◦ alongside
the response to corresponding unidirectional tests (indicated by .I). Slightly greater pile
(a) UD sea state, very loose sand
(b) MD sea state, very loose sand
(c) MD sea state, dense sand
Figure 5.7: Impact of additional wind loading on the x-direction response (mean monotonic
response shown in grey)
* First half of test L.MD.W.x presented
rotations are observed in the x-direction under multidirectional loading, compared to
under corresponding unidirectional loading. However, the difference in pile rotation is
small, particularly in the dense sand, and could be attributed to experimental variability.
Minimal pile rotation is observed in the y-direction.
The observations made here are consistent with the results from regular fan-type
loading tests (Section 4.6.2). The fan-type results suggest that for half internal spread
angles < 30° (where the vast majority of this multidirectional loading occurs) a small
increase in ratcheting rate may be observed, with the ratio of multidirectional
ratcheting exponent to unidirectional ratcheting exponent expected to be
1 < α/αU D < 1.24 (see Figure 4.34, Section 4.6.2).
5.4.4 Application of consecutive storms
Figure 5.9 presents the response to tests L.MD.W.x and L.MD.W.45, where two
consecutive storms were applied. The second storm was either aligned (.x) or
misaligned at 45◦ (.45) to the first. The small difference in response for n < 5200, where
identical loading was applied, is attributed to experimental variability.
The black loops in the moment-rotation plot for test L.MD.W.x highlight the
response during application of the same large amplitude cycle in each of the two
storms (n = 2050, 7250). This cycle occurs after the peak cyclic load Mp and therefore
allows the change in secant stiffness and energy loss factor to be explored without
conflation with Masing behaviour. The loading secant stiffness knl increases by a factor
of 1.23 (from knl /kM AX = 0.26 to knl /kM AX = 0.31), while the energy loss factor on
loading ηnl decreases very slightly (from ηnl = 0.52 to ηnl = 0.51) over 5200
multi-amplitude cycles. The factor of increase in stiffness is less pronounced than
across the same number of regular cycles (Section 4.4.2), which may be explained by
the low average cyclic amplitude for these tests (mean MCY C /MR = 0.046) compared
to the regular cyclic tests (MCY C /MR ≥ 0.2 in very loose sand for 1-way loading). The
minimal change in energy loss factor is consistent with observations under 1-way
regular cyclic loading beyond n = 100.
The pile begins to rotate in the y-direction when the storm is applied at 45◦ to the
axes during the second half of test L.MD.W.45. The y-direction hysteretic response
(a) Without wind load, very loose sand
(b) With wind load, very loose sand
(c) With wind load, dense sand
Figure 5.8: Impact of multidirectionality on the response (mean monotonic response shown in
grey)
* First half of test L.MD.W.x presented
Figure 5.9: Response to consecutive MD storms in very loose sand aligned and misaligned at
45◦ to each other (mean monotonic response shown in grey)
follows the backbone curve on reloading, similarly to in the x-direction. The rotation
accumulated in the y-direction exceeds that accumulated in the x-direction during the
second half of the test, despite the loading in the x- and y-directions being very similar.
The smaller rotation accumulated in the x-direction in the second half of test L.MD.W.45
may be explained by greater densification caused by previous cycling in this direction.
Densification is expected to vary spatially with the loading direction, as also indicated
by the ripples in the secant stiffness response under fan-type loading (Section 4.6.2).
The magnitude of pile rotation accumulated during the second storm is
approximately 30% of the rotation accumulated during the first storm, in both the
aligned and misaligned case.

5.5 Summary
The experimental results presented in this Chapter reveal the response of monopile
foundations to realistic multi-amplitude and multidirectional storm loading in dry
sand. The preliminary load processing also highlighted the importance of the
structural dynamic response in determining the cyclic load amplitude and number of
load cycles experienced by monopile foundations. The list below summarises the key
experimental observations, which are valid for both very loose and dense sand unless
stated otherwise:
• Upon applying loads greater than those previously applied, the response
approximately follows the backbone curve.
• Pile rotation is dominated by large load events, although ratcheting is also
observed during application of smaller-amplitude cyclic loading between the
large load events.
• The addition of wind loading, and the associated increase in load bias,
significantly increases the magnitude of pile rotation.
• Typical storm multidirectionality (spreading angle 22◦ ) does not have a significant
impact on pile rotation in the dominant loading direction.
• Storm misalignment has no significant impact on the magnitude of pile rotation,
for a misalignment angle of 45◦ .
• There is evidence of an increase in equivalent secant stiffness under
multi-amplitude loading (in very loose sand).
In general, the response to multi-amplitude and multidirectional storm loading is
consistent with many of the observations for regular cyclic loading in Chapter 4:
ratcheting occurs under biased loading and increases with load bias and load
amplitude, and secant stiffness increases under general cyclic loading. The effects of
multidirectionality are small or negligible for the realistic spreading angle and storm
misalignment angle explored here. Importantly, the tendency for the backbone curve
to be followed when applying loads greater than those previously applied suggests that
the maximum rotation during a short storm may be approximated by the monotonic
response. As well as providing novel insight into the response of monopiles to realistic
cyclic loading, these results constitute a valuable set of validation data for use in model
development.
Chapter 6
Investigating the effect of

stress-level
6.1 Introduction
As it is not practical to test monopiles at full-scale, given their large size (e.g. D = 8 m,
Sørensen et al., 2017), the response may be explored through physical modelling at
reduced-scale. Various physical modelling approaches exist:
1. Laboratory-scale testing at 1g, as employed in Chapters 3, 4 and 5, and in many
previous similar studies (see Section 1.3.1);
2. Laboratory-scale testing at elevated stress-level using a centrifuge, as conducted
by e.g. Klinkvort and Hededal (2013);
3. Medium- to large-scale testing in natural deposits in field conditions, as
performed as part of the PISA project (Byrne et al., 2015) and by e.g. Cox et al.
(1974) for the development of conventional p-y curves.
Laboratory-scale testing at 1g allows the development of complex systems and requires
less resource than the other two physical modelling approaches, facilitating more
extensive and longer-term testing. However, full-scale stress-levels cannot be simulated
with 1g laboratory testing, and it is difficult to simulate natural soil deposits. Centrifuge
testing involves subjecting a geotechnical model to accelerations n times Earth’s gravity
(ng) to simulate effective unit weights and therefore stress-levels n times larger than
would be experienced on the laboratory floor. With the geotechnical model spun in a
centrifuge at angular velocity ω, the acceleration experienced at effective radius Re is
130
6. Investigating the effect of stress-level 131
ng = Re ω 2 . Centrifuge testing allows full-scale stress-levels to be simulated, but the
high acceleration levels and space restrictions can constrain the model geometry and
complexity, compared to 1g testing (Byrne, 2014). Medium- and large-scale field testing
allows exploration of the response in natural deposits, and approximates the ratio of
foundation to soil grain size better than in 1g or centrifuge laboratory-scale testing.
However, field testing is costly and logistically challenging.
In practice, a combination of these modelling approaches might be used to most
efficiently explore the response of monopile foundations, but stress-level varies
significantly between them, with only centrifuge testing able to simulate full-scale
stresses. Understanding the impact of stress-level on the response of monopile
foundations is therefore an important step to i) inform comparison of observations at
1g, in the centrifuge and at medium to large-scale in the field, and ii) establish the
relevance of physical modelling at low stress-levels.
The fundamental behaviour of sand depends on the stress-level. For example,
dilatancy and peak friction angle φ0p increase approximately logarithmically with
reducing stress-level (at least at relatively high stress-levels (Bolton, 1987)), while the
maximum shear modulus GM AX increases as a power function of stress-level, with the
exponent typically 0.5 (Hardin, 1965; Oztoprak and Bolton, 2013). This fundamental
stress-level dependent behaviour is reflected in the global foundation response, as
observed by e.g. Ovesen (1975) for footings, Kelly et al. (2006) for caissons, and
Klinkvort (2012) for monotonically laterally loaded monopiles.
Very few studies have explored the impact of stress-level on the response of
monopiles to cyclic lateral loading. Rudolph et al. (2014b) and Rudolph and Grabe
(2013) report a decrease in accumulated displacement for centrifuge tests compared to
1g tests, while Nicolai et al. (2017b) compare post-cyclic behaviour between 1g and
centrifuge tests. However, the set-up and sand type varied between 1g and centrifuge
tests in these studies, and so stress-level effects were not completely isolated.
To specifically explore issues relating to stress-level, this Chapter presents a series of
monotonic, unidirectional cyclic and multidirectional cyclic tests, performed at three
different stress-levels using the 5 m radius beam centrifuge at UWA (Figure 6.1). By
Figure 6.1: 5 m radius beam centrifuge at UWA
Figure 6.2: Location of model tests at ng, simulated prototypes at 1g and current full-scale
monopiles
varying the centrifuge velocity ω, and therefore g-level, stress-level can be controlled;
and with an identical set-up at each g-level, stress-level effects can be isolated.
g-levels of 1g, 9g and 80g were chosen, corresponding to normalised reference

0
stress-levels (defined as the vertical effective stress at 70% pile embedment, σREF /pa =
0.7Lγ 0 /pa ) of 0.02, 0.18 and 1.60 for a 42 mm diameter pile with L/D = 4, in dense,
dry sand (γL0 = 17 kN/m3 ). These tests simulate monopiles with diameters between
71 mm and 5.7 m in dense, saturated sand (γP0 = 10 kN/m3 ), as shown in Figure 6.2.
Although real monopiles are typically even larger, this range of stress levels is deemed
sufficiently large to observe relevant stress-level effects. The logarithmic variation of
g-level was chosen to be in-line with the expected logarithmic variation of dilatancy
and friction angle with stress-level.
So-called modelling of models test campaigns are more common than investigations
External diameter D 42 mm
Target embedded length L 170 mm
Aspect ratio η 4 -
Wall thickness t 3.2 mm
Loading eccentricity h 424 mm
Dimensionless loading eccentricity ẽ 2.5 -
Material Mild steel
Young’s modulus E 200 GPa
Mass m 1.69 kg
Table 6.1: Model pile properties
into stress-level effects (e.g. Ovesen, 1975; Dewoolkar et al., 1999; Klinkvort et al., 2013),
and typically involve tests at different g-levels and model sizes which all simulate the
same stress-level and prototype. Modelling of models tests lie along diagonal lines in
Figure 6.2, as illustrated with the example tests of Klinkvort et al. (2013). Modelling
of models campaigns can help identify scale effects specific to centrifuge testing (e.g.
grain-size effects) and provide verification of centrifuge modelling techniques, but do
not provide independent information on stress-level effects.
6.2 Experimental set-up

6.2.1 Model pile, loading and instrumentation
Tests were performed on sandblasted mild steel piles with properties summarised
in Table 6.1. The L/D ratio, h/L ratio and smooth surface (Rn = Ra /d50 ≈ 0.016,
measured using a stylus profilometer) are consistent with the pile used for 1g testing
at OU. The pile has a closed-end (appropriate given the wished in place installation
method) and is assumed to behave rigidly.
Figure 6.3 shows the loading and instrumentation apparatus, which is modified
from that designed by Herduin (2019) for multidirectional loading of anchors for wave
energy devices. Actuators sit perpendicular to platform A and apply load to the pile via
cables which travel around pulleys on platform C and attach to 1 kN capacity inline load
cells at the base of the pile stick-up. Pile displacements are measured with six string
potentiometers. Three are positioned on platform B (253 mm above the load application
point) and three are positioned on platform C (30 mm above the load application point).
Figure 6.3: Schematic and photograph of loading and instrumentation apparatus
Each triplet of string potentiometers is arranged in a 120° star to minimise the net load
applied to the pile by these sensors, as each applies a tensile load of around 1 N.
The pile stick-up allows straightforward connection of the string potentiometers
and load lines to the pile. The stick-up comprises an insert, which is fixed inside
the pile head with two bolts and two grub screws, and a threaded rod to which the
potentiometers and load lines are attached.
The set-up shown in Figure 6.3 is for multidirectional loading, with three actuators
and associated load lines positioned 120° apart. For unidirectional loading two
actuators are used, positioned 180° apart, to simplify the set-up.

Specific gravity Gs 2.67 -

Particle size d10 , d50 , d60 0.12, 0.18, 0.19 mm
Minimum effective unit weight 0
γM IN 14.69 kN/m3
Maximum effective unit weight 0
γM AX 17.40 kN/m3
Maximum void ratio eM AX 0.78 -
Minimum void ratio eM IN 0.51 -
Critical friction angle φ0c 31.9 °
Table 6.2: UWA SF sand properties (Chow et al., 2018)
6.2.2 Sample preparation
UWA superfine (SF) silica sand was used for these tests, with properties summarised in
Table 6.2. Soil samples were prepared by air pluviation using an automatic sand raining
device into a square strongbox with base 996 × 996 mm and height 500 mm. Three
0
dense samples were prepared to an average unit weight γAV = 17.00 ± 0.20 kN/m3
(DR,AV = 87.4 ± 5%). The samples were prepared dry to simulate a fully-drained
response. Figures 6.4 and 6.5 show the sand raining procedure and prepared sample.
To avoid the introduction of complex stress-fields and local density changes through
in-flight (or 1g) installation, the piles were wished-in-place during the sand raining
process. Sand was first rained to the depth of the pile tip, before hanging the piles in
position and raining further sand around them. The sand surface was then vacuumed
to achieve an average installed pile embedment LAV = 167 ± 3 mm (L/D ≈ 4). Nine
piles were installed per strongbox, with a minimum centre-to-centre spacing of 7.4D.
Cone penetration tests (CPTs) were performed using a 7 mm diameter 60° cone
to characterise each soil sample. For sample S1, used for 1g testing, two CPTs were
performed at each stage: a) pre-testing and b) post-testing. For samples S2 and S3,
used for testing at 9g and 80g respectively, one CPT was performed at each stage:
a) at 1g pre-spinning, b) at ng pre-testing, c) at ng post-testing, and d) at 1g post-
spinning. Figure 6.6 presents the CPT cone resistance qc profiles measured at each
g-level. At 1g, the S1 CPTs and the pre-spinning S2 and S3 CPTs show good consistency,
in-line with the small variation in global unit weight measured across the samples.
The consistency of the pre- and post-testing S2 and S3 CPTs also gives confidence in
Figure 6.5: Prepared sample with nine piles wished in

Figure 6.4: Sample preparation place
with sand raining device
Figure 6.6: CPT profiles in SF sand (DR,AV = 87.4%, note different qc scales, T=testing,
Sp=spinning)
the homogeneity of the sample and insensitivity to repeated centrifuge spin cycles
(the centrifuge was spun down after each test).
The 1g post-spinning S2 and S3 CPTs show increased CPT resistance, although bulk
mass and volume measurements imply a minimal increase in density post-spinning (a
0.45% and 0.57% increase in γ 0 following 9g and 80g testing respectively). This result is
therefore attributable to overconsolidation effects, also observed by e.g. Gui et al.
(1998) and Roy et al. (2019). However, no monopile tests were performed on these
overconsolidated samples.
6.2.3 Test procedure
9g and 80g tests were performed in the 5 m radius, 240g-tonne capacity beam centrifuge
at UWA (Gaudin et al., 2018), while the 1g tests were conducted with the same set-up
on the laboratory floor. For the centrifuge tests the effective radius Re , at which the
nominal g-level is achieved, was chosen to be at 1/3 pile embedment (Re ≈ 4.66 m).
Given the size of the centrifuge relative to the model, the variation of acceleration in
the radial direction and associated stress error is negligible (Schofield, 1980).
Monotonic tests were performed under displacement control with a single load line,
whereas cyclic tests were performed under load control with sinusoidal waveforms,
using either two or three load lines. Actuator control and data-acquisition was
performed using UWA’s in-house software (PACS and DigiDac).
To ensure accurate load control with the (flexible) load lines, the lines were kept
in tension during cyclic testing, with one or two load lines holding constant load.
Appropriate choice of load demands for each line allows application of cyclic loading,
including symmetric 2-way and multidirectional cyclic loading. Before application
of cyclic loading, loads in the lines were increased simultaneously to an appropriate
pre-tension. Data was not re-zeroed following this procedure, as negligible net load
was applied to the pile and negligible pile movement was recorded.
To ensure accurate load control, a cyclic frequency of 0.15 − 0.20 Hz was used for
the 9g and 80g tests. To achieve accurate load control at 1g, where the load cells were
significantly over-sized, the cyclic frequency was reduced (to 0.01 − 0.05 Hz). The
pile response is not expected to be rate dependent, given the tests were performed
in dry sand (see also Section 3.4).
6.2.4 Resolution of pile displacement and applied loads
Time limitations did not allow development of software to perform kinematics
calculations (resolution of pile pose and load line angles) on-line, as described for the
OU apparatus in Section 2.4. Instead, the pile displacement and applied loads were
resolved during post-processing. The actual applied loads therefore differ slightly from
the nominal load demands under multidirectional loading, where the load line angles
change as the pile moves. This is discussed further in Section 6.6.
To resolve pile position, vertical displacements were neglected and the planar
position of the pile at the level of each triplet of string potentiometers was calculated.
Under unidirectional loading, pile displacement was obtained directly from the string
potentiometer aligned with the loading direction and net loads were obtained directly
from the load measurements aligned with loading. Under multidirectional loading,
measurements from all three potentiometers were used to calculate pile displacement,
and to account for measurement redundancy the error between actual and measured
length was assumed to be equal for all three string potentiometers. The net loads were
then resolved, accounting for the instantaneous pile position and load line angle.
6.3 Test programme
Tables 6.3 and 6.4 summarise the tests completed as part of this investigation into
stress-level effects. The test programme was informed by testing conducted at 1g at
OU. Monotonic tests provide a baseline for interpretation of cyclic tests; unidirectional,
symmetric 2-way cyclic tests explore the hysteretic response; unidirectional 1-way
cyclic tests explore ratcheting, stiffening and evolution of dissipation at various load
amplitudes; and multidirectional, perpendicular cyclic tests explore the impact of cyclic
loading direction, following the T- and L-shaped loading described in Section 4.6.1. The
cyclic tests were conducted at values of ζb comparable to those used at OU, with the
minimum cyclic amplitude dictated by the accuracy of the control system at 1g.
The results presented in Chapter 4 show how key cyclic behaviour is observable
within 1000 cycles. As such, tests were conducted to 1000 cycles here, except where load
application issues occurred. The post-cyclic response was also explored with reloading
to 0.8HR following many unidirectional cyclic tests. Figure 6.7 shows the sequence of
testing within each sample; the loading directions are also indicated.
Results are presented in terms of both g-level n and normalised reference stress-level
0
σREF /pa . Table 6.5 expedites translation between these two measures of stress-level,
and presents the (prototype) monopile diameter DP simulated at each g-level.

No. of Reload
Test name g-level ζb ζc
cycles N
Monotonic tests
1G.M.1 1g - - - -
9G.M.1 9g - - - -
9G.M.2 9g - - - -
80G.M.1 80g - - - -
Symmetric 2-way cyclic tests
1G.TW.04 1g 0.4 -1 500 No

9G.TW.04 9g 0.4 -1 1000 Yes
80G.TW.04 80g 0.4 -1 1000 Partial
1-way cyclic tests
1G.OW.02 1g 0.2 0 1000 Yes

1G.OW.03 1g 0.3 0 770 Yes
1G.OW.04 1g 0.4 0 1000 Yes
9G.OW.02 9g 0.2 0 1000 Yes
9G.OW.03 9g 0.3 0 1000 Yes
9G.OW.04 9g 0.4 0 1000 Yes
80G.OW.02 80g 0.2 0 1000 Yes
80G.OW.04 80g 0.4 0 1000 Yes
Table 6.3: Unidirectional test programme investigating stress-level effects
Test Loading ζb ζc No. of

g-level
name direction cycles N
Nominal Applied Nominal Applied
x 0.35 0.37–0.39 1 0.80–1.05

1G.T.02 1g 250
y 0.2 0.19–0.22 -1 -(1.20–0.80)
x 0.35 0.36–0.39 1 0.98–1.08
1G.L.04 1g 153
y 0.4 0.37–0.38 0 -(0.07–0.00)
x 0.2 0.20–0.23 1 0.95–0.99
9G.T.02 9g 1000
y 0.2 0.19–0.20 -1 -(1.03–0.99)
x 0.2 0.21–0.24 1 0.93–0.97
9G.L.04 9g 1000
y 0.4 0.38–0.39 0 -(0.06–0.00)
x 0.2 0.20–0.22 1 0.95–1.00
80G.T.02 80g 1000
y 0.2 0.20–0.20 -1 -(1.00–0.95)
x 0.2 0.20–0.22 1 0.95–1.00
80G.L.04 80g 1000
y 0.4 0.39–0.39 0 -(0.02–0.00)
Table 6.4: Multidirectional test programme investigating stress-level effects

Figure 6.7: Sequence of testing within each sample, with arrows indicating loading direction
g-level, n 1 9 80
0
σREF /pa 0.02 0.18 1.60
DP [m] 0.071 0.64 5.7
Table 6.5: Corresponding g-levels, normalised stress levels and simulated prototype monopile
diameters
Figure 6.8: Monotonic responses (data presented on three different H-axes for clarity)
6.4 Monotonic response
Figure 6.8 presents the response for the monotonic tests alongside the responses for
the initial loading portion of the unidirectional cyclic tests. At 9g, where two
monotonic tests were performed, the average monotonic response is presented. Figure
6.8 highlights the significant variation in load amplitude with stress-level, as well as
consistency of behaviour at each stress-level.
Data is presented in terms of applied horizontal load H and pile displacement at
the load application point u (both at model-scale). The imperfectly rigid connection
between the pile and pile stick-up introduced error into the resolved pile rotation and
prevents presentation of the response in terms of applied moment M and pile rotation
θ, as used in previous Chapters. However, conclusions on stress-level effects are not
expected to be affected by the choice of either work-conjugate pair.
6.4.1 Normalisation approaches
Casting the monotonic response in a dimensionless form facilitates comparison
between tests and helps provide insight into stress-level effects. Three different
normalisation approaches are considered: i) Klinkvort et al. (2013), ii) Klinkvort (2012),
and iii) Leblanc et al. (2010a), as summarised in Table 6.6. The approach of Leblanc
et al. (2010a) was also presented in Table 3.4, Section 3.4.3.

i) Klinkvort et al. (2013) ii) Klinkvort (2012) iii) Leblanc et al. (2010a)
H H H
Load
normalisation D3 γ 0 Kp D 3 γ 0 L2 Dγ 0
u u η−1
Displacement u pa
D D (η = 0.5)
normalisation D Lγ 0
Table 6.6: Comparison of normalisation approaches
(a) (b)
Figure 6.9: Normalisation of monotonic responses; (a) following (i) Klinkvort et al. (2013), (b)
following (iii) Leblanc et al. (2010a) (lower plots show low amplitude region highlighted by grey
box in upper plots)
Figure 6.9a shows the result of applying normalisation (i), which accounts for
variation of γ 0 , to the monotonic responses. This normalisation does not collapse the
responses to one curve, but rather highlights stress-level effects: the decrease in
normalised initial stiffness and inferred normalised capacity with increasing
stress-level are indicative of stress-dependent stiffness and dilatant behaviour.
While exploring stress-level effects Klinkvort (2012) proposed a normalisation
method (ii) which introduces Rankine’s passive earth pressure coefficient Kp to
account for stress-dependent dilatancy. Klinkvort (2012) used this normalisation to
successfully collapse the response of a model monopile at stress-levels from around

70 kPa to around 350 kPa, with φ0p values obtained from complementary triaxial testing.
Complementary triaxial tests are not available for this study, and moreover,
obtaining definitive φ0p values at stress-levels corresponding to the 1g tests is likely to
be very challenging. There are a few approaches for estimating φ0p at low stress-levels
(as discussed in Section 3.2.3), although recent work by White (2020) supports the use
of relationships proposed by Bolton (1987). If the framework of Bolton (1987) is
employed to estimate φ0p for normalisation (ii), then Kp changes negligibly and the plot
resembles that in Figure 6.9a.
In contrast to the approach of Klinkvort (2012), Leblanc et al. (2010a) incorporate
stress-dependent stiffness into normalisation (iii), but do not consider dilatancy (see
Section 3.4.3). Figure 6.9b shows the result of normalising with approach (iii). The
normalisation does a good job of collapsing the results to a single curve, particularly
at small displacements, where the cyclic tests are conducted.
The normalisation proposed by Klinkvort et al. (2013) indicates the presence of
stress-level dependent stiffness and dilatant behaviour. However, it is not possible to
decouple these phenomena and draw stronger conclusions, particularly without
measured φ0p values at the stress-levels of interest. However, the normalisation
approach of Leblanc et al. (2010a), which does not account for stress-dependent
dilatancy, collapses the monotonic responses well, building confidence in the use of
this normalisation framework for translation between scales.
6.4.2 Maximum stiffness
The monopile’s maximum stiffness kM AX is expected to vary with stress-level in the
same way as the soil’s maximum shear modulus GM AX does; indeed, this assumption
is incorporated into the normalisation approach presented by Leblanc et al. (2010a).
As discussed in Section 3.4.2, kM AX can be obtained from the initial loading or first
unloading response. There is more variability in the initial loading response for these
tests than for the 1g tests at OU, probably due to more significant bedding-in effects
given the wished-in-place installation method. Therefore, only the stiffness on first
unloading are used to estimate the maximum stiffness of the monopile kM AX .

Figure 6.10: Initial portion of first unloading responses showing upper and lower estimates for
maximum stiffness kM AX with grey shaded region (note different Hm0 − H scales)
Figure 6.10 presents the first unloading portion for all relevant tests, with the
response re-zeroed from the point of maximum load on the 0th cycle (subscript m0).
Upper and lower estimates of kM AX are obtained by manual fitting to the initial
loading portion (um0 − u < 0.05 mm), and the maximum stiffness bounds are indicated
by the grey shaded regions in Figure 6.10.
Figure 6.11 presents the variation of kM AX with stress-level, with the range and
mean values of kM AX indicated. An exponential function fits the data well, as shown
by the dashed line in Figure 6.11. However, the exponent obtained from least-squares
fitting is η = 0.31, somewhat lower than η = 0.5 implicit in the Leblanc et al. (2010a)
normalisation and also obtained for UWA SF sand from triaxial testing (Chow et al.,
2018). This discrepancy may be explained by experimental variability, the impact
of stiffening during the initial loading portion, or three-dimensional effects when
considering the monopile system. Interestingly, however, using η = 0.31 in place of
η = 0.5 in the normalisation approach of Leblanc et al. (2010a) generates a poorer result.
6.5 Unidirectional cyclic response

6.5.1 Load-displacement response
Figure 6.12 presents the load-displacement (H −u) responses for the unidirectional tests
summarised in Table 6.3, alongside the corresponding monotonic responses (nG.M).
The full response is shown for the 1-way tests, while only the first five cycles of the 2-way
Figure 6.11: Variation of maximum stiffness kM AX with stress-level
Reference displacement Reference horizontal load Mean max. stiffness

uR [mm] HR [N] kM AX [N/mm]
1g 17.8 15.7 57.3

9g 17.8 99.5 127.8
80g 17.8 597 220.8
Table 6.7: Reference values for monopile response
tests are shown, to highlight the shape of the hysteretic response. The results show
good repeatability at each stress-level and qualitatively similar behaviour across the
three stress-levels. The responses are normalised by load and displacement reference
values (HR , uR ) as summarised in Table 6.7, where uR was chosen to approximately
correspond to θR = 2° (as used in previous Chapters). This normalisation approach
allows data at all three stress-levels to be presented on comparable axes, but is not
directly intended to elucidate stress-level effects.
Figure 6.12 shows how the backbone linearity (or dominance of elastic behaviour)
increases with stress-level. This variation in linearity is linked to stress-dependent
stiffness and dilatant behaviour, as previously discussed, but also depends on the load
amplitude. To better match the linearity of the cyclic responses across the stress-levels,
the reference pile displacement uR – which determines HR and therefore the cyclic
amplitudes through ζb – may be adjusted with stress-level, perhaps employing the
dimensionless framework of Leblanc et al. (2010a) (Table 3.4, Section 3.4.3). However,
Figure 6.12: Unidirectional load-displacement response (note different scales on lower row)
the variation in backbone linearity is a stress-level effect, and it is not clear whether
it is appropriate to minimise or eliminate this effect by modifying uR .
Figure 6.12 also reveals gapping-type behaviour in the 2-way tests at 1g and 9g,
consistent with the observations in dense sand at 1g in Section 4.3. The inflexion in
the load-displacement response around zero load is indicative of gapping, as the pile’s
tangent stiffness reduces while traversing a gap or region of low-stress. These tests show
gapping-type behaviour increasing with reducing stress-level, which may be explained
by a reduction in the tendency for sand particles to move into a gap as stress-level
reduces. For gapping to occur at all, some cohesion is required, which may be caused by
particle interlocking, ambient moisture or electrostatic effects, as discussed in Section
4.3. The wished-in-place installation method will lead to lower horizontal stresses
adjacent to the pile, increasing the likelihood of gap formation.

6.5.2 Ratcheting response
Figure 6.13 presents the evolution of normalised accumulated displacement ∆un /uR
with cycle number n for the 1-way unidirectional cyclic tests. A power-law (Equation
6.1) is fitted to each test and shown dashed. Various studies have shown ratcheting
to evolve as a power-law with cycle number n (Leblanc et al., 2010a; Klinkvort, 2012;
Abadie et al., 2019b; Albiker et al., 2017; Truong et al., 2019), and it was also found to
provide a good fit to the 1g results in Section 4.4.1.
∆un
= Anα (6.1)
uR
No dependence of the power-law coefficient A on stress-level is observed, although
Figure 6.14 shows how A varies with ζb as a power-law with exponent mσ = 3.4 (fit shown
dashed in Figure 6.14), consistent with the observed variation of equivalent parameter
Tb in Figure 4.8, Section 4.4.1, and with the results of Abadie (2015). Conversely, no
dependence of α on ζb is observed (in-line with previous studies), but there is a trend
for decreasing α with increasing stress-level, as shown in Figure 6.15. To facilitate
prediction of behaviour at other stress-levels, the ratcheting exponent α is presented in

0
Figure 6.15 against normalised reference stress-level σREF /pa , rather than g-level.
To quantify the variation of α with stress-level, a logarithmic trend line is fitted to
the data by least squares regression, and shown dashed in Figure 6.15. The line is
defined by Equation 6.2.

0
σ

REF
α = 0.127 − 0.022 ln (6.2)
pa
This trend line can be used to inform comparison of behaviour at different stress-
levels. For example, it suggests that the ratcheting exponent for an equivalent full
size monopile in dense, saturated (fully drained) sand (D = 8 m, γ 0 = 10 kN/m3 ,

0
σREF /pa = 2.2) may be as low as α = 0.11, or approximately half of the value in the 1g
model tests. This very important observation highlights how ratcheting may be less
pronounced at full-scale than observed in 1g laboratory-scale physical modelling.
The value of α varies with the chosen strain variable (as discussed in Section 4.4.1)
and it is therefore difficult to compare α values from independent test campaigns to

Figure 6.13: Accumulation of displacement ∆un with unidirectional cyclic loading
Figure 6.14: Variation of ratcheting Figure 6.15: Variation of ratcheting

power-law coefficient A with ζb power-law exponent α with stress-level for
unidirectional cyclic loading
build confidence in this conclusion on stress-dependency. However, the results
presented here are consistent with the increase in normalised pile displacement for 1g
tests compared to centrifuge tests reported by Rudolph et al. (2014b) and Rudolph et al.
(2014a).
6.5.3 Secant stiffness response
Figure 6.16 presents evolution of the secant stiffness of the monopile kn with cycle
number n for all unidirectional cyclic tests. Stiffness is normalised by the mean
maximum stiffness, kM AX (defined in Section 6.4.2). An increase in secant stiffness is

observed for all tests, with kn /kM AX plateauing at high cycle number in all tests except
nG.TW.04.
Evolution of secant stiffness with cycle number has previously been described by
logarithmic functions (Klinkvort and Hededal, 2013; Leblanc et al., 2010a; Abadie,
2015). For the OU tests in Section 4.4.2 logarithmic fits were found to be appropriate
for 1-way and partial 1-way loading, but power-laws fitted the response to 2-way and
partial 2-way loading better. Here, a power-law function (Equation 6.3) is preferred and
fitted to all data for consistency of interpretation with ratcheting. However, neither a
power-law nor a logarithmic function captures the response particularly well for n < 10.
Power-law fits are shown dashed in Figure 6.16.
kn
= Bnβ (6.3)
kM AX
The normalised secant stiffness for the first cycle (kn=1 /kM AX ) is plotted in Figure
6.17. There is no strong dependence of first cycle secant stiffness values on stress-
level, however, the spread of stiffness values decreases with increasing stress-level,
consistent with the increasing linearity of the responses. At each g-level, the first cycle
stiffness values generally decrease with cyclic amplitude HCY C , as expected, given the
non-linear load-displacement response. Plotting the power-law coefficient (or y-axis
intercept) B instead would have produced a very similar plot.
The impact of stress-level on the evolution of stiffness is assessed by plotting the
power-law exponent β against stress-level in Figure 6.18. The exponents for tests
1G.TW.04 (β = 0.304) and 9G.TW.04 (β = 0.110) are not presented on this plot, being
considerably greater than the other exponents. These high values of β are thought to be
linked to the gapping-type behaviour observed for these tests. If a gap is opening, the
possibility of grain migration, and therefore densification close to the pile, will be
increased.
A logarithmic trend line is fitted to the variation of β with stress-level, neglecting
the 2-way tests, and shown dashed in Figure 6.18. The trend line is described by
Equation 6.4 and can be used to inform comparison of monopile behaviour at different
stress-levels. For example, it suggests that for an equivalent full size monopile the
Figure 6.16: Change in secant stiffness kn with unidirectional cyclic loading
Figure 6.17: Variation of normalised stiffness Figure 6.18: Variation of stiffness power-law
for the first cycle kn=1 /kM AX with exponent β with stress-level for
stress-level for unidirectional cyclic loading unidirectional cyclic loading
1-way stiffness exponent may be as low as β = 0.011, or 25% of the stiffness exponent
value in the 1g model tests.

0
σ

REF
β = 0.017 − 0.0076 ln (6.4)
pa
6.5.4 Energy dissipation response
Figure 6.19 shows the evolution of unidirectional energy loss factor ηn with cycle
number n. Although an empirical fit was not made to the OU 1g data in Section 4.4.2, a
power-law fit (Equation 6.5) is used here to quantify the impact of stress-level on ηn .
The power-law provides a reasonable fit to all data, at least for n > 10, with the
resulting fits shown dashed in Figure 6.19.
ηn = Cnγ (6.5)
As for ratcheting and stiffness evolution, the variation of the energy loss factor power-
law exponent γ with stress-level is plotted in Figure 6.21. However, no dependence of
γ on stress-level is observed. Instead, the energy loss factor for the first cycle ηn=1 is
found to decrease with increasing stress-level, as shown in Figure 6.20. A similar result
would be obtained if the power-law coefficient (or y-axis intercept) C had been plotted
instead. The anomalously low value of initial energy loss factor for test nG.TW.04 is
probably linked to the significant gapping-type behaviour observed, which reduces
the hysteretic energy loss EH , compared to no gapping.
A logarithmic function is fitted to the variation of first cycle energy loss factor ηn=1
with stress-level. The trend line, shown dashed in Figure 6.20, is defined by Equation 6.6.
0
σ

REF
ηn=1 = 0.228 − 0.134 ln (6.6)
pa
This trend line suggests that an equivalent full size monopile may have an energy loss
factor for the first cycle of around ηn=1 = 0.12 , which is 16% of the value for the model
tests at 1g.
6.5.5 Reloading response
Post-cyclic reloading was performed after 1000 loading cycles for the majority of the
unidirectional cyclic tests. The upper row of Figure 6.22 shows the reloading responses
with the cyclic responses omitted, while the lower row shows the reloading responses
re-zeroed from the onset of reloading (at u = ur0 ). Qualitatively similar behaviour is
observed at all stress-levels.
The reloading responses following 1-way cyclic loading tend to re-join or exceed the
backbone curve at all stress-levels, while a significant increase in capacity is observed
under 2-way loading at 9g, where little ratcheting or drift occurred. These observations
are consistent with those made in Section 4.7 and support the interpretation of the
reloading response as a competition between ratcheting and stiffening processes.

Figure 6.19: Change in energy loss factor ηn with unidirectional cyclic loading
Figure 6.20: Variation of energy loss factor Figure 6.21: Variation of energy loss factor
for the first cycle ηn=1 with stress-level for power-law exponent γ with stress-level for
unidirectional cyclic loading unidirectional cyclic loading
Re-zeroing the reloading responses highlights the significant increase in stiffness
relative to the monotonic response for H/HR < ζb , consistent with the observed
increase in secant stiffness with cycling.
6.5.6 Discussion
Qualitatively, the unidirectional cyclic responses are similar across the three stress-
levels investigated, with similar reloading responses and with power-law expressions
approximately capturing the evolution of ratcheting, stiffness and energy loss factor.
But, quantitatively, the results reveal some important stress-level effects.

Figure 6.22: Reloading response following unidirectional cyclic loading (cyclic response
omitted and response re-zeroed from ur0 on lower row, note different scale at 80g)
The decrease of first cycle energy loss factor ηn=1 with stress-level (Figure 6.20) can
be directly linked to the linearity of the load-displacement response. The response
linearity increases with stress-level, and is linked to stress-dependent stiffness and
dilatant behaviour, but also depends on the load amplitude. However, the logarithmic
decrease in ratcheting and stiffness exponent with stress-level (Figure 6.15 and Figure
6.18) appears to be independent of load amplitude and therefore linearity of the
load-displacement response.
The various studies from Cuéllar et al. (2012), Nicolai (2017) and Cui and
Bhattacharya (2016) revealed the presence of both convection and densification
mechanisms under cyclic loading in dry sand (see Section 1.3.2). The increase in
secant stiffness is probably driven by a local densification mechanism, while the
ratcheting behaviour is probably caused by both a convection mechanism and any
asymmetry in local densification. Given that both mechanisms require particle
rearrangement, it is explicable that the associated ratcheting and stiffening behaviour

will be affected by increasing stress-level, where (as highlighted by Cui and
Bhattacharya (2016)) there is increasing constraint on particle rearrangement.
6.6 Multidirectional cyclic response
Table 6.4 characterises the multidirectional tests in terms of nominal and applied ζb
and ζc load values in the x- and y-directions. The nominal values were those demanded
of the control system during testing, and do not account for pile movement. The actual
applied values were computed post-testing, accounting for the instantaneous pile
position and load line angle. Pile movement has a non-negligible effect on the applied
loads: the peak cyclic loads vary by up to 10% and the load biases vary by up to 20%
from the nominal values. However, the phenomenological observations made in this
Section are not expected to be affected by these load variations.
The multidirectional T- and L-shaped tests at 9g and 80g correspond to
unidirectional tests nG.OW.04 – having the same cyclic load amplitude HCY C and
average load amplitude HAV . However, the multidirectional 1g tests have a load bias in
the x-direction 75% larger than 1G.OW.04 and so comparisons at 1g are made with
caution.
6.6.1 Displacement and ratcheting response
Figure 6.23 presents the displacement responses for the multidirectional tests nG.T.02
and nG.L.04, alongside the corresponding unidirectional tests nG.OW.04. The
displacements at cycles 1, 10, 100, (1000) are marked. Figure 6.23 shows how the
monopile moves broadly in the direction of the load bias at all stress-levels, although
there is some deviation in tests 9G.T.02 and 80G.T.02. The deviation may be caused by
systematic error in experimental set-up or boundary effects (test 9G.T.02 deviates
towards the strongbox wall while test 80G.T.02 deviates towards the strongbox corner,
see Figure 6.7). Figure 6.23 also highlights the increase in amplitude of displacement
across each cycle with increasing stress-level, consistent with a decrease in secant
stiffness kn normalised by HR /uR (kn uR /HR ) with increasing stress-level.
Figure 6.24 presents the accumulated rotation responses for the multidirectional
tests alongside the corresponding unidirectional tests. For the T-shaped tests significant
Figure 6.23: Displacement response for multidirectional tests (markers indicate location of
cycle 1, 10, 100, (1000))
ratcheting only occurs, and is only reported, in the x-direction. For the L-shaped tests
ratcheting is reported in both the x- and y-directions. A power-law (Equation 6.1) is
fitted to the evolution of ∆un /uR with cycle number n and shown dashed in Figure 6.24.
The power-law generally captures the evolution of ratcheting well, but tends to over-
predict ratcheting for n < 10 for the T-shaped tests. The response of test 80G.L.04(x),
particularly for n > 100, may be anomalous.
Figure 6.25 presents the variation of ratcheting power-law coefficient A with test type
and stress-level. No clear dependence of A on either test type or stress-level is observed.
The ratcheting power-law exponent α does vary with stress-level, and is plotted in
Figure 6.26, accompanied by the dashed trend line obtained for the unidirectional tests
(Equation 6.2). This trend line also fits the multidirectional data well. In general, there
is no clear dependency of ratcheting behaviour on test type (at least for n > 10).
6.6.2 Secant stiffness response
Figure 6.27 presents the evolution of secant stiffness kn for the multidirectional tests
and corresponding unidirectional cyclic tests. For multidirectional loading, evolution
of secant stiffness is only relevant in the direction of cycling. For n > 10, the response is
described well with a power-law (Equation 6.3), which is fitted and shown dashed in
Figure 6.27.
Figure 6.24: Accumulation of displacement ∆un with multidirectional cyclic loading
Figure 6.25: Variation of ratcheting Figure 6.26: Variation of ratcheting

power-law coefficient A with power-law exponent α with stress-level for
multidirectional test type multidirectional cyclic loading
The variation of normalised secant stiffness for the first cycle (kn=1 /kM AX ) is shown
in Figure 6.28. As for the unidirectional tests, there is no clear dependence on stress-
level; there is also no dependence on test type. The variation in power-law stiffness
exponent β is plotted in Figure 6.29, with the trend line obtained for unidirectional
loading (Equation 6.4) shown dashed. The values of β for the multidirectional tests at
1g depart from the trend line, though there is inconsistency in applied mean load. The
multidirectional tests at 9g and 80g are aligned with the unidirectional trend.
Figure 6.27: Change in secant stiffness kn with multidirectional cyclic loading
Figure 6.28: Variation of normalised stiffness Figure 6.29: Variation of stiffness power-law
for the first cycle kn=1 /kM AX with exponent β with stress-level for
stress-level for multidirectional cyclic multidirectional cyclic loading
loading
6.6.3 Energy dissipation response
The evolution of energy loss factor ηn with cycle number n is shown in Figure 6.30, with
a power-law (Equation 6.5) fitted and shown dashed. The multidirectional behaviour is
consistent with that observed for the unidirectional tests: there is no clear dependence
of the energy loss factor power-law exponent γ on stress-level (Figure 6.32), but the
energy loss factor for the first cycle ηn=1 decreases logarithmically with stress-level, as
shown in Figure 6.31. The trend line obtained for unidirectional loading (Equation
6.6) also fits this data well, as shown by the dashed line in Figure 6.31. Values of
Figure 6.30: Change in energy loss factor ηn with multidirectional cyclic loading
Figure 6.31: Variation of energy loss factor Figure 6.32: Variation of energy loss factor
for the first cycle ηn=1 with stress-level for power-law exponent γ with stress-level for
multidirectional cyclic loading multidirectional cyclic loading
ηn=1 and γ are similar for unidirectional and L-shaped tests, but there is a tendency
for the T-shaped tests tend to have lower ηn=1 values and higher γ values. This is
probably associated with the variation in response linearity over the cyclic loading
region with test type, at a given stress-level.
6.6.4 Discussion
In general there is little variation in the magnitude and evolution of ratcheting, stiffness
and energy loss factor with multidirectional test type, at all stress-levels. This implies
that, for a given mean load HAV and cyclic load amplitude HCY C , the direction of cyclic
loading (relative to the mean load) has an insignificant impact on the cyclic response.
Ratcheting also occurs in the direction of the mean load at all stress-levels, regardless
of the cyclic loading direction. These observations are in line with those reported
for 1g testing at OU in Chapter 4, and have important implications for modelling the
multidirectional response. Ratcheting in the direction of the applied load is a key
feature of models in the HARM framework (Houlsby et al., 2017), and may be included
as a feature of other ratcheting models.
In general, the stress-level effects observed in the unidirectional tests are also
observed in the multidirectional tests. The trend lines obtained from the unidirectional
results for variation of ratcheting exponent, stiffness exponent and first cycle energy
loss factor with stress-level are also, in general, appropriate for the multidirectional
results.
6.7 Summary
This Chapter has explored the effect of stress-level on the response of monopile
foundations to lateral loading. Monotonic, unidirectional cyclic and multidirectional
cyclic loading tests were performed on a model monopile in dry, dense sand at three
different g-levels. With the same experimental set-up at each g-level, stress-level effects
were isolated. The following key observations were made:
• The monotonic responses exhibit stress-level effects which can be explained by
stress-dependent stiffness and dilatancy, although decoupling these phenomena
is difficult.
• The monotonic results support the use of the normalisation approach proposed
by Leblanc et al. (2010a), although the observed variation of maximum foundation
stiffness with stress-level is not entirely consistent with this approach.
• Gapping-type behaviour reduces with increasing stress-level.
• Qualitatively, the cyclic response is similar across the three stress-levels, with
power-law functions providing a reasonable fit to evolution of ratcheting, secant
stiffness and energy loss factor with cycle number.

• The ratcheting and stiffness exponents, which control the rate of change of these
parameters with cycle number, both decrease logarithmically with increasing
stress-level. The trends suggest that, for an equivalent full size monopile (D =
8 m), the ratcheting exponent may be half the value at 1g and the stiffening
exponent may be one quarter the value at 1g.
• The reduction in ratcheting and stiffening rate with stress-level is explained by the
increase in confinement with stress-level, which inhibits particle rearrangement
understood to cause ratcheting and stiffness change.
• The energy loss factor for the first cycle reduces with increasing stress-level. This
behaviour depends on the linearity of the backbone curve, which is linked to
stress-dependent stiffness and dilatant behaviour, but also varies with load
amplitude.
• The post-cyclic reloading response shows no clear stress-level dependency.
• The multidirectional tests exhibit stress-level effects consistent with the
unidirectional tests, while the multidirectional observations are consistent with
those made for OU 1g tests in Chapter 4.
Together, these results provide new insight into the impact of stress-level on the
response of monopile foundations, and help inform comparison of monopile behaviour
at different stress-levels. The observed impact of stress-level on the rate of change of
ratcheting and stiffening with cycle number is particularly important, as it suggests
the impact of cyclic loading will be less pronounced at full-scale than observed in
small-scale 1g physical modelling.
The qualitative similarities in response at 1g, 9g and 80g demonstrate the insight
that can be gained from 1g testing, while the observed stress-level effects highlight the
need to simulate full-scale stress-levels to thoroughly understand foundation behaviour.
This supports the suggestion that an efficient physical modelling campaign might use a
combination of modelling approaches. As conducted in Chapters 3, 4 and 5, laboratory-
scale 1g testing may be used to explore a wide range of behaviour, establish trends
and inform initial development of models for design. Focused centrifuge testing can
then assess the applicability of 1g observations at higher (ideally full-scale) stress-levels,
and inform scaling relationships. Large-scale field testing may be used for validation
of design methods in natural deposits.
The results in this Chapter cannot be quantitatively compared to the OU results
(reported in Chapters 3 and 4), given the difference in installation method, sand type,
stress and strain parameters. However, the behaviour is qualitatively similar: with
ratcheting, stiffening and changes in dissipation with repeated cyclic loading, a
tendency for the reloading response to approach or exceed the backbone curve, and
invariance of the magnitude and evolution of the cyclic response to cyclic load
direction.
Chapter 7
Modelling the response of a

monopile to complex cyclic loading
7.1 Introduction
This Chapter draws on the experimental observations from previous Chapters to
demonstrate and inform development of macro models for monopile foundations
under cyclic lateral loading in the Hyperplastic Accelerated Ratcheting Model (HARM)
framework. The hyperplastic modelling framework and the underlying multi-surface
kinematic hardening models are first presented, before the introduction of
unidirectional and bi-directional models which are able to capture the high-cycle
response, including ratcheting and stiffness evolution.
New evolution functions, which describe how ratcheting and stiffness evolve with
cyclic loading, are proposed. Focus is placed on functions which do not significantly
distort Masing behaviour, are able to capture the post-cyclic reloading response, and
depend on parameters which have a clear impact on the response. The models are
calibrated to five datasets (OU very loose and OU dense datasets presented in Chapters
3, 4 and 5, and UWA 1g, UWA 9g and UWA 80g datasets presented in Chapter 6) and
computations are compared to a representative selection of experimental results. The
computations in this Chapter explore the ability of models in the HARM framework to
capture the response to unidirectional and multidirectional regular cyclic loading, as
well as the response to realistic multi-amplitude storm loading. Predictions are also
made at prototype-scale.
162
7. Modelling the response of a monopile to complex cyclic loading 163
7.2 Key behaviours observed experimentally
Key behaviour Key evidence
1 The monopile exhibits a non-linear monotonic response. Section 3.4.1

A, B
2 The extended Masing rules are approximately adhered to during Section 4.3
the first few loading cycles. A
3 Ratcheting occurs i) under biased cyclic loading and ii) in the i) Section 4.4.1
direction of the mean load. ii) Section 4.6.1
i) A, B
4 Ratcheting evolves as a power-law with cycle number. Section 4.4.1
A, B
5 The magnitude of ratcheting increases as a power-law with cyclic Section 4.4.1
amplitude (for constant ζc ). A, B
6 Load asymmetry (ζc ) affects the shape, rate and therefore Section 4.4.1
magnitude of ratcheting.
7 Secant stiffening occurs under cyclic loading and evolves Section 4.4.2
logarithmically or as a power-law, depending on the load A
asymmetry.
8 The cyclic response is broadly independent of the cyclic loading Section 4.6.1
direction, relative to the mean load direction.
9 Multidirectional fan-type loading can lead to greater ratcheting Section 4.6.2
and stiffening than unidirectional loading.
10 The inferred capacity on post-cyclic reloading is equal to or Section 4.7
greater than the monotonic capacity.
Table 7.1: Key behaviours observed experimentally in this work and by Abadie (2015) (A) and
Beuckelaers (2017) (B) and sought to be captured in modelling
The combined experimental observations from Chapters 3, 4, 5 and 6 are used to
determine the key behaviours which are sought to be captured in modelling, as outlined
in Table 7.1. Evolution of energy loss factor is not directly sought to be captured, since it
is understood to depend largely on the evolution of stiffness and ratcheting. No attempt
is made to capture the gapping-type behaviour observed under high-amplitude 2-way
loading in dense dry sand at low stress-levels, given that it occurs only at large rotations
and low stress-levels, which are not representative of field conditions.
Table 7.1 indicates which behaviours were also observed by Abadie (2015) (A),
as these directly informed the development of the HARM framework presented by
Houlsby et al. (2017). Behaviours observed by Beuckelaers (2017) (B) at field-scale,
in both sand and clay, are also indicated.

7.3 Modelling basis

7.3.1 Hyperplasticity
Models in the HARM framework are formulated in the hyperplasticity framework
which is comprehensively described by Houlsby and Puzrin (2006). The hyperplasticity
framework allows compact, systematic development of constitutive models which are
guaranteed to obey the Laws of Thermodynamics. The method is rooted in the work of
Ziegler (1977), and was first presented (for rate-independent materials) by Houlsby
(1981). The framework is applicable to both rate-independent and rate-dependent
materials or systems, and uses internal variables (α, often plastic strains) to describe
the loading history. In general, models are presented in terms of conjugate stress σ and
strain ε variables. The variables σ, ε, α and related variables may be scalars, vectors or
tensors depending on the model dimensionality.
Models in the hyperplasticity framework are fully specified in terms of two potential
functions: one describing stored energy and one describing dissipation. This contrasts
with the conventional plasticity approach, where specification of an elastic constitutive
relationship, yield criterion, plastic potential and hardening law is required. Stored
energy may be defined in terms of the Helmholtz free energy f (ε, α) or the Gibbs free
energy g(σ, α). For rate-independent materials, dissipation may be defined in terms of
the dissipation function d(ε, α, α̇) or with a yield surface y(ε, α, χ); similar functions are
defined for rate-dependent materials. The functions f and g are related through a
Legendre transform and functions d and y are related through a degenerate Legendre
transform.
The constitutive and incremental behaviour of a model is derived from the
potentials with standardised procedures. With a model described by Helmholtz free
energy f and dissipation d functions the behaviour is derived from:
∂f ∂f ∂d
σ= χ̄ = − χ= χ̄ = χ (7.1)
∂ε ∂α ∂ α̇
Where χ̄ is the generalised stress and χ is the dissipative generalised stress, conjugate
to the internal variable α. The condition χ̄ = χ is an expression of Ziegler’s
orthogonality principle (Ziegler, 1977), which is central to the hyperplasticity

framework. This principle assumes that energy is dissipated at the maximal rate, and
allows derivation of the constitutive response without any additional assumptions.
Where additional kinematic constraints are applied c(α̇) = 0, the dissipation
function is augmented with a Lagrangian multiplier Λ:
d∗ = d(ε, α, α̇) + Λc(α̇) (7.2)
With the generalised stress given by:
∂d∗ ∂d ∂c
χ= = +Λ (7.3)
∂ α̇ ∂ α̇ ∂ α̇
The Lagrangian multiplier is eliminated by the use of the constraint c(α̇) = 0.
7.3.2 Multi-surface kinematic hardening models
Models in the HARM framework are based on multi-surface kinematic hardening
models, which adhere to the extended Masing rules. These models are commonly
referred to as Iwan models following the publication of the general characteristics of
these models by Iwan (1967) for application to metals. Conceptually, these models are
made up of M elastoplastic elements which each consist of a linear spring of stiffness
Hm and a rate-independent cohesive (Coulomb) slider of strength Km . Figure 7.1 shows
how the elastoplastic elements (along with an additional solitary spring) are arranged
for unidirectional models in either series or parallel configurations; the interpretation
of parameters Km and Hm differs between the configurations. Bi-directional models are
more easily represented as a series of yield surfaces (circular with radius Km for isotropic
models) which define regions of constant stiffness Em (Em = f (Hm ) which varies
between series and parallel models), as shown in Figure 7.2. The models generate multi-
linear stress-strain (σ−ε) responses, as indicated in Figure 7.2 for loading and unloading
in the y-direction, and automatically respond to any arbitrary, continuous loading.
As well as forming the basis of the ratcheting models presented by Houlsby et al.
(2017), multi-surface kinematic hardening models have been used to capture the cyclic
response of soils (Prévost, 1977; Mrŏz et al., 1978) and the macro response of caisson
foundations (Nguyen-Sy and Houlsby, 2005; Skau et al., 2018) and monopile
foundations (Page et al., 2018). Multi-surface kinematic hardening models are

Parallel model
H M+1
Series model KM HM
HM H2 H1
αM
H0 σ σ
KM K2 K1
K2 H2
αM α2 α1
α2
εε
K1 H1
α1
ε
Figure 7.1: Conceptual models of unidirectional multi-surface kinematic hardening models in

series and parallel configurations
Inital loading Unloading y-direction response
σy σy σy
σx σx εy
Figure 7.2: Representation of isotropic bi-directional multi-surface kinematic hardening

models showing response to loading and unloading in the y-direction
presented here in hyperplasticity theory, but can also be described in conventional
plasticity theory.
7.3.3 HARM framework
The Hyperplastic Accelerated Ratcheting Model (HARM) framework presented by
Houlsby et al. (2017) was developed to capture the response of monopile foundations
to cyclic lateral loading, but is applicable to other materials or systems which exhibit
Masing behaviour augmented by ratcheting. Rate-dependent models of this sort were
initially developed by Abadie (2015), and Houlsby et al. (2017) expanded the framework
Parallel HARM model

H M+1
Series HARM model KM HM

HM H1 R
αM
H0 σ σ
KM K1
R
αr αM α1
K1 H1
εε
αr α1
ε
Figure 7.3: Conceptual models of unidirectional HARM models in series and parallel
configurations (Houlsby et al., 2017)
to present both rate-dependent and rate-independent models using both series and
parallel configurations, with extensions to two-dimensional and tensorial models.
These models are able to capture the system’s continuous response to any arbitrary
loading, but can also be accelerated to capture the total ratcheting response across a
packet of regular cyclic loading with explicit calculation of a single cycle (Houlsby et al.,
2017).
Models in this framework are based on multi-surface kinematic hardening models
but include additional ratcheting elements, as indicated in Figure 7.3. Ratcheting strain
αr accumulates as a proportion R of plastic strain, where R may vary with both loading
history and surface m (R = Rm in general). Stiffness evolution can be captured if the
strengths Km are also made functions of the loading history. These models have been
shown to capture the response of monopiles to regular, unidirectional cyclic loading
at laboratory-scale in loose, dry sand (Abadie et al., 2019a) and at field-scale at both
sand and clay sites (Beuckelaers, 2017). Balaam et al. (2020) has also demonstrated
the ability of these models to capture the cyclic response of clay in cyclic simple shear
apparatus to irregular multi-amplitude loading.
7.3.4 Modelling choices
Models in the HARM framework, and kinematic hardening models in general, can be
formulated in either series or parallel, may be rate-independent or rate-dependent,

and may be implemented as macro models (Abadie, 2015; Abadie et al., 2019a), 1D
Winkler models (Beuckelaers, 2017) or as tensorial models in 3D finite element
software (Houlsby et al., 2017).
Series models are employed here as they are more suited to computation of strain
given a stress input, and simulations are conducted here with stress input to be
consistent with the load-controlled experiments. Series and parallel models can
produce identical responses under unidirectional loading, but produce slightly
different responses under multidirectional loading (Beuckelaers et al., 2018).
Unfortunately, the suitability of either model formulation cannot be determined from
the multidirectional data as the differences in model computations are obscured by
differences in response due to experimental variability (see Figure 7.7, Section 7.4.4).
Rate-independent models are presented here, given the rate-independence
observed experimentally. However, models which capture the rate effects observed in
clays and saturated sands have been developed in the HARM framework by
Beuckelaers (2017).
Macro models for the global pile response are used here. These single-element
models represent the integrated response of the pile-soil system, which may be
represented with greater fidelity using a 1D Winkler model or a tensorial model. The
conjugate stress σ and strain ε variables therefore represent applied moment M and
pile rotation θ (OU datasets) or applied horizontal load H and pile displacement at the
load application point u (UWA datasets).
7.4 Capturing the hysteretic response
This Section explores the ability of kinematic hardening models, without ratcheting
elements, to capture the hysteretic response of monopile foundations observed in the
first few cycles. In particular, the suitability of a bi-directional kinematic hardening
model is explored. The hyperplastic model formulations are first presented.

7.4.1 Unidirectional hyperplastic kinematic hardening model
The model is defined in terms of the Helmholtz free energy function f and the
dissipation function d:
M
!2 M
H0 X X Hm 2
f= ε− αm + α m (7.4)
2 m=1 m=1
2
M
X
d= Km |α̇m | (7.5)
m=1
The internal variables may be interpreted as the plastic strains αm , while Hm , Km and
H0 values are used to calibrate the model, as discussed in Section 7.4.3. The model’s
constitutive behaviour is derived as:

M
!
∂f X
σ= = H0 ε − αm (7.6)
∂ε m=1
M
!
∂f X
χ̄m =− = H0 ε − αm − Hm αm = σ − Hm αm (7.7)
∂αm m=1
∂d α̇m
χm = = Km (7.8)
∂ α̇m |α̇m |
α̇m
χm = χ̄m ⇒ σ − Hm αm = Km (7.9)
|α̇m |
The incremental total strain response can then be derived by differentiation and
rearrangement of Equation 7.6:

M
dσ X
dε = + dαm (7.10)
H0 m=1
The yield functions ym are implicit in Equation 7.9. At yield, when α̇m 6= 0:
ym = |σ − Hm αm | − Km = |χm | − Km = 0 (7.11)
Finally, the plastic strain increment is found by differentiation of the yield functions
ym with respect to time, where the consistency condition enforces ẏm = 0:
dσ
dαm = (7.12)
Hm
In numerical implementation, the yield conditions are determined using αm from the
previous increment. However, the expression for plastic strain increment is exact.
7.4.2 Bi-directional hyperplastic kinematic hardening model
The unidirectional model presented in Section 7.4.1 is extended here for bi-directional
loading. It is a simplified version of the bi-directional ratcheting model presented
by Houlsby et al. (2017). The model response is independent of loading direction, as
expected for piles under lateral loading in a transversely-isotropic soil. The model is
developed in terms of x- and y-components of stress σ and strain ε:

   
σx εx
σ=  ε=  (7.13)
σy εy
The model is defined in terms of the Helmholtz free energy function f and the
dissipation function d:

M
!2 M
!2  M
H0  X X X Hm 2 2

f= εx − αmx + εy − αmy + α mx + αmy (7.14)
2 m=1 m=1 m=1
2
M
X q
d= 2 + α̇2
Km α̇mx my (7.15)
m=1
The model’s constitutive behaviour is derived as:
M M
! !
∂f X ∂f X
σx = = H0 εx − αmx σy = = H0 εy − αmy (7.16)
∂εx m=1
∂εy m=1
M
!
∂f X
χ̄mx =− = H0 εx − αmx − Hm αmx = σx − Hm αmx
∂αmx m=1
M
!
∂f X
χ̄my =− = H0 εy − αmy − Hm αmy = σy − Hm αmy (7.17)
∂αmy m=1
∂d α̇mx ∂d α̇my
χmx = = Km q χmy = = Km q (7.18)
∂ α̇mx α̇ + α̇2
2 ∂ α̇my α̇ + α̇2
2
mx my mx my
α̇mx
χmx = χ̄mx ⇒ σx − Hm αmx = Km q
2 + α̇2
α̇mx my
α̇my
χmy = χ̄my ⇒ σy − Hm αmy = Km q (7.19)
2 + α̇2
α̇mx my
The incremental total strain response is derived by differentiation and rearrangement
of Equation 7.16:
M M
dσx X dσy X
dεx = + dαmx dεy = + dαmy (7.20)
H0 m=1
H0 m=1
The yield functions ym can be obtained by summing the squares of Equation 7.19 in
the x- and y-direction. At yield, when α̇mx

2 + α̇2
my 6= 0:
q q
ym = (σx − Hm αmx )2 + (σy − Hm αmy )2 − Km = χ2mx + χ2my − Km = 0 (7.21)
The following relationship between generalised stresses and plastic strain rates is also
obtained from Equation 7.19:
χmx α̇mx
= (7.22)
χmy α̇my
Substituting this ratio into the consistency condition (ẏm = 0) yields expressions for
the plastic strain increments:
χmx χmy
dαmx = λ dαmy = λ (7.23)
Hm Hm
Where λ is conventionally termed the plastic multiplier, determined as:
χmx dσx + χmy dσy

λ= 2
(7.24)
Km
In numerical implementation, χmx , χmy are determined using αm from the previous
increment. The solution for plastic strain increment dαmx , dαmy is therefore not exact,
however, instabilities are avoided with sufficiently small load increments.
7.4.3 Calibration of kinematic hardening models
The kinematic hardening model parameters Km and Hm are obtained for a series model
from M points on a backbone curve (εm , σm ) using the following relationships:
σm = Km , m = 1...M (7.25)
m−1
X Km − Ki
εm = , m = 1...M, K0 = 0 (7.26)
i=0
Hi
Any monotonically increasing backbone curve can be captured, satisfying key
behaviour 1 (Table 7.1). The slider strengths are typically arranged in increasing order
(Ki+1 > Ki ), and the number of surfaces M is chosen to fit a given backbone curve
with the required accuracy. Here, M = 100 surfaces were typically used, with σm evenly
distributed from 0 to σR .
In practice, the backbone curve is an essential output of a monotonic monopile
design, which is a pre-requisite for cyclic design. Various monotonic design
approaches may be taken: conducting 3D finite element simulations, applying the
numerical or rule-based PISA method (Byrne et al., 2017), or using established p-y
methods (as presented in e.g. DNV GL, 2016). Here, the unidirectional backbone curves
obtained experimentally are used for calibration of parameters Km and Hm , as
monotonic design is not the focus of this work.
Although points (εm , σm ), and therefore parameters Km and Hm , could be obtained
directly from the experimental backbone curves, specification of analytical functions
for the backbone curves ensures smoothness of the model parameters and facilitates
scaling. Three analytical functions are considered:
1. A Ramberg-Osgood (Ramberg and Osgood, 1943) or power-law relationship, as
used by Abadie et al. (2019a) to capture the response of a model monopile in loose,
dry sand:
Pp
σ σR σ

ε= + εR − (7.27)
Ei Ei σR
2. An expression presented by Jeanjean et al. (2017) for the response of clays,
modified to include an elastic component:

 2
σ
σ σR  arctanh tanh(Pj )

σR
ε= + εR −  (7.28)
Ei Ei Pj
3. A conic function, as used to represent the soil reactions as part of the PISA project
(e.g. Burd et al., 2019):

σ 2εpu C
ε= + √ (7.29)
Ei −B + D
2
Where εpu = εR − σERi ; A = Pc σR
Ei εpu ; B = −2A σσR +(1−Pc ) 1 + σR
Ei εpu
σ
σR −1 ;
2
σ
; D = max B 2 − 4AC, 0 .

C=A σR
Figure 7.4 shows the fits obtained to the mean experimental backbone curves, for
each of the five datasets, using each of the analytical functions. The fits are presented
in three ways, in terms of i) the moment-rotation or load-displacement response, ii) the
tangent stiffness Kt response, and iii) the fitting error θe or ue across the response. The
(a) OU very loose (b) OU dense
(c) UWA 1g (d) UWA 9g
Figure 7.4: Fitting analytical functions to mean experimental backbone curves

(e) UWA 80g
Figure 7.4: Fitting analytical functions to mean experimental backbone curves
Parameter OU very loose OU dense UWA 1g UWA 9g UWA 80g
σR 26 N m 95 N m 15.7 N 99.5 N 597 N

εR 2° 2° 17.8 mm 17.8 mm 17.8 mm
Ei 936 N m/° 979 N m/° 57.3 N/mm 127.8 N/mm 220.8 N/mm
Analytical function Power-law Jeanjean Jeanjean Jeanjean Power-law
Fitting parameter Pp = 3.56 Pj = 0.865 Pj = 0.800 Pj = 0.939 Pp = 1.82
2
R 0.997 0.998 0.999 0.991 0.999
Table 7.2: Summary of parameters for analytical functions fitted to mean experimental
backbone curves to facilitate calibration of parameters Km and Hm
initial stiffness Ei values were set equal to the maximum stiffness values Ei = kM AX ,
the reference stress values σR were set to either MR or HR and the reference strain
values εR were set to either θR or uR , as defined experimentally in Sections 3.4.3 and
6.5.1. The remaining fitting parameters (Pp , Pj , Pc for Equations 7.27, 7.28, 7.29) were
then obtained by regression analyses with θ or u as the dependent variable.
The power-law (Equation 7.27) provides a very good fit to the OU very loose data
and UWA 80g data, while the Jeanjean expression (Equation 7.28) provides the best fit
to the OU dense, UWA 1g and UWA 9g data. Table 7.2 summarises the chosen analytical
function, fitting parameters and goodness of fit metric (R2 ) for each dataset.
7.4.4 Computation of hysteretic response
Figure 7.5 presents computations of the unidirectional hysteretic responses using the
unidirectional model described in Section 7.4.1, calibrated to each dataset as described
in Section 7.4.3. Increasing amplitude 2-way tests demonstrate the model performance
for the OU datasets, while the first four cycles of symmetric 2-way tests are used for the
UWA datasets. The kinematic hardening model complies with the extended Masing
rules, which the experimental responses approximately follow for the first few cycles
(key behaviour 2, Table 7.1). The model therefore approximately captures the hysteretic
response for all datasets. The model does not capture gapping, which is most marked
in the UWA 1g data, and does not capture the increase in secant stiffness observed
in the OU dense, UWA 9g and UWA 80g data.
Figure 7.6 shows the performance of the bi-directional model described in Section
7.4.2, calibrated following Section 7.4.3, under multidirectional spiral loading in OU
very loose and dense sand. Although spiral loading is not realistic, these data present a
robust test of the bi-directional model formulation. The OU very loose data is captured
with excellent accuracy, while the OU dense data is captured with reasonable accuracy.
The OU dense test was thought to be affected by sample inhomogeneity (discussed in
Section 4.5), which may explain the poorer model performance. The results indicate
that the bi-directional model formulation proposed by Houlsby et al. (2017) is
appropriate for capturing the response of a monopile to multidirectional lateral
loading.
As discussed in Section 7.3.4, series and parallel kinematic hardening models
produce slightly different responses under multidirectional loading (Beuckelaers et al.,
2018). To explore this, computations are presented in Figure 7.7 using i) the series
model used throughout this Chapter, and ii) an equivalent parallel model (as presented
in e.g. Houlsby et al., 2017). A range of rotation responses are presented for each model,
corresponding to a ±5% variation in the value of σR used to define the backbone curve;
(c) UWA 1g (d) UWA 9g
(e) UWA 80g
Figure 7.5: Computation of unidirectional hysteretic responses using unidirectional kinematic

hardening model
( model computations, experimental data)
Figure 7.6: Computation of multidirectional hysteretic responses using bi-directional

kinematic hardening model
( model computations, experimental data)
the range of responses therefore account for typical experimental variability. The
difference in the computations is sufficiently subtle that, when experimental variability
is accounted for, there is significant overlap in the predicted responses. The series
model appears to perform better for the OU very loose dataset, but no such distinction
can be made for the OU dense dataset; in general it is not possible to establish a
preferred model on the basis of these tests. Indeed, additional trial computations
suggest that it may not be possible to devise a multidirectional test which is able to
establish a preferred model.
7.5 Capturing the high-cycle response
This Section explores the ability of models in the HARM framework to capture the
high-cycle response, which includes ratcheting and stiffening. Unidirectional and
bi-directional hyperplastic ratcheting models (Houlsby et al., 2017) are first presented,
before discussing evolution functions and calibration, and finally computing the
responses to regular cyclic loading and multi-amplitude storm loading.

Figure 7.7: Comparison of series and parallel bi-directional kinematic hardening model
computations with ±5% variation in σR
( experimental data)
7.5.1 Unidirectional hyperplastic ratcheting model
The unidirectional ratcheting model builds on the unidirectional kinematic hardening
model presented in Section 7.4.1, and was presented and demonstrated by Houlsby
et al. (2017). The Helmholtz free energy function f and dissipation function d both
include a contribution from ratcheting strain αr , which plays the role of an additional
internal variable:
M
!2 M
H0 X X Hm 2
f= ε− αm − αr + α m (7.30)
2 m=1 m=1
2
M
!
X
d= Km |α̇m | + σ α̇r (7.31)
m=1
Where a constraint defines the ratcheting strain αr :

M
σ X
c = α̇r − Rm |α̇m | = 0 (7.32)
|σ| m=1
The model’s constitutive behaviour is derived as:

M
!
∂f X
σ= = H0 ε − αm − αr (7.33)
∂ε m=1
M
!
∂f X
χ̄m =− = H0 ε − αm − αr − Hm αm = σ − Hm αm (7.34)
∂αm m=1
M
!
∂f X
χ̄r = − = H0 ε − αm − αr =σ (7.35)
∂αr m=1
∂d∗ α̇m σ α̇m
χm = = Km − Λ Rm (7.36)
∂ α̇m |α̇m | |σ| |α̇m |
∂d∗
χr = =σ+Λ (7.37)
∂ α̇r
χr = χ̄r ⇒ Λ=0 (7.38)
α̇m
χm = χ̄m ⇒ σ − Hm αm = Km (7.39)
|α̇m |
The incremental total strain response is derived from Equation 7.33:
M
dσ X
dε = + dαm + dαr (7.40)
H0 m=1
While the constraint (Equation 7.32) defines the increment of ratcheting strain:
M
σ X
dαr = Rm |α̇m | (7.41)
|σ| m=1
This expression is consistent with key behaviour 3 (Table 7.1), given that dαr is aligned
with the sign of the incremental load. The yield functions and incremental plastic
strain are the same as for the kinematic hardening model, and are implicit in Equation
7.39. At yield, when α̇m 6= 0:
ym = |σ − Hm αm | − Km = |χm | − Km = 0 (7.42)
dσ
dαm = (7.43)
Hm
7.5.2 Bi-directional hyperplastic ratcheting model
The unidirectional ratcheting model presented in Section 7.5.1 is extended here for
bi-directional loading, as proposed by Houlsby et al. (2017). This model builds on the
bi-directional kinematic hardening model presented in Section 7.4.2, which was shown
to capture the multidirectional hysteretic response in Section 7.4.4. The model is
developed in terms of x- and y-components of stress σ and strain ε and is defined in
terms of the Helmholtz free energy function f , dissipation function d and constraints
cx , cy :

M
!2 M
!2  M
H0  X X X Hm 2 2

f= εx − αmx − αrx + εy − αmy − αry + α mx + αmy
2 m=1 m=1 m=1
2
(7.44)
M
!
X q
d= Km 2
α̇mx + 2
α̇my + σx α̇rx + σy α̇ry (7.45)
m=1
M
σx X q
cx = α̇rx − q 2 + α̇2 = 0
Rm α̇mx my (7.46)
σx2 + σy2 m=1
M
σy X q
cy = α̇ry − q 2 + α̇2 = 0
Rm α̇mx my (7.47)
σx2 + σy2 m=1
The model’s constitutive behaviour is derived here in the x-direction only, for brevity:
M
!
∂f X
σx = = H0 εx − αmx − αrx (7.48)
∂εx m=1
M
!
∂f X
χ̄mx =− = H0 εx − αmx − αrx − Hm αmx = σx − Hm αmx (7.49)
∂αmx m=1
M
!
∂f X
χ̄rx =− = H 0 εx − αmx − αrx = σx (7.50)
∂αrx m=1
∂d∗ α̇mx σx α̇mx

χmx = = Km q − Λq Rm q (7.51)
∂ α̇mx α̇2 + α̇2 σx2 + σy2 2 + α̇2
α̇mx
mx my my
∂d∗
χrx = = σx + Λ (7.52)
∂ α̇rx
χrx = χ̄rx ⇒ Λ=0 (7.53)
α̇mx
χmx = χ̄mx ⇒ σx − Hm αmx = Km q (7.54)
2
α̇mx 2
+ α̇my
The incremental total strain response is derived from Equation 7.48 (the equivalent
y-direction expression is also presented):

M M
dσx X dσy X
dεx = + dαmx + dαrx dεy = + dαmy + dαry (7.55)
H0 m=1
H0 m=1
While the constraints (Equation 7.46 and 7.47) define the increments of ratcheting
strain:
M
σx X q
2 + dα2
dαrx = q Rm dαmx my
σx2 + σy2 m=1
M
σy X q
dαry = q 2 + dα2
Rm dαmx my (7.56)
σx2 + σy2 m=1
These expressions are consistent with key behaviour 3 (Table 7.1), with dαr aligned
with the incremental load direction. The yield functions ym are the same as for the
bi-directional kinematic hardening model, and are implicit in Equation 7.54. At yield,
when α̇mx
2 2
+ α̇my 6= 0:
q q
ym = (σx − Hm αmx )2 + (σy − Hm αmy )2 − Km = χ2mx + χ2my − Km = 0 (7.57)
χmx χmy
dαmx = λ dαmy = λ (7.58)
Hm Hm
χmx dσx + χmy dσy
λ= 2
(7.59)
Km
7.5.3 Calibration of ratcheting models
To capture ratcheting behaviour which decreases in rate with loading history, the
ratcheting factor Rm is made a function of a scalar hardening parameter β. Changes
in stiffness are also captured by making Km vary with either the same, or a different,
hardening parameter. It is more straightforward to evolve stiffness by evolving the
slider strengths Km than the spring stiffness Hm , as the derivatives of the Helmholtz
free energy functions f would change if Hm = f (β) (Houlsby et al., 2017). Selection
of the evolution functions Rm = f (β) and Km = f (β) is discussed in detail in Section
7.5.4, and calibration of these functions is discussed in Section 7.5.5.
The parameters Hm and Km0 (the initial values of Km ) are initially calibrated
following the approach for kinematic hardening models outlined in Section 7.4.3.
However, the inclusion of ratcheting and stiffening causes the initial loading
computation to depart from the backbone curve. It is therefore necessary to modify the
initial parameters as part of evolution function calibration to ensure the backbone
curve is followed on initial loading. Here, backbone correction was achieved by
adjusting the Hm values, keeping the Km0 values unchanged. A two-parameter
function was fitted to the variation of Hm with surface m (Hm = DmE ), and the
parameters (D, E) were optimised to minimise the error between the backbone curve
and the initial loading computation affected by ratcheting and stiffening.
7.5.4 Evolution functions
This Section describes selection of the hardening parameter β and evolution functions
Rm = f (β) and Km = f (β), which control the computed ratcheting, stiffening, and
related energy dissipation and post-cyclic response.

Hardening parameter β
The hardening parameter β should increase monotonically with loading history, but
can take various forms. Houlsby et al. (2017) suggest both strain-type values (β̇ = |α̇r |,
PM PM
β̇ = m=1 |α̇m |) and work-type values (β̇ = σ α̇r , β̇ = m=1 Km |α̇m |). The accumulated
ratcheting strain β̇ = |α̇r | was used by Abadie (2015), Abadie et al. (2019a) and
Beuckelaers (2017). However, a work-type hardening parameter is used here as this
makes the evolution functions and parameters comparable between series and parallel
model formulations. The hardening parameter was also chosen to evolve with plastic
strain αm , rather than ratcheting strain αr , as this eases calibration of the ratcheting
evolution function Rm and allows stiffness evolution to be implemented
independently of ratcheting behaviour. Local or surface-specific (βm ) and global (β)
hardening parameters are defined here for bi-directional loading:

q
2 + α̇2
Km α̇mx my
β̇m = (7.60)
σR
 q 
M M 2 + α̇2
Km α̇mx
X X my
β̇ = β̇m =   (7.61)
m=1 m=1
σR
The hardening parameters are independent of strain direction, allowing the evolution
functions to evolve independently of cyclic loading direction to capture key behaviour
8 (Table 7.1).
Function for Rm
The evolution function for Rm should capture the observed power-law evolution of
ratcheting with cycle number n (key behaviour 4, Table 7.1). Abadie et al. (2015), Abadie
et al. (2019a) and Beuckelaers (2017) captured the evolution of ratcheting with
functions of the form:

−mr
β

Rm = R0 · f (Km , σ, ...) (7.62)
β0
Where β0 is the initial value of hardening parameter β, and is specified as an arbitrary
small value. Although functions of this form are able to capture ratcheting under
biased loading, the rate of change of Rm at small β is very large, which can significantly
Figure 7.8: Example of distortion of Masing behaviour when using power-law expression for
Rm
Rm
R0
R*=f(σ) * mr
log scale
log scale
β* β
Figure 7.9: Illustration of proposed evolution function for Rm
distort Masing behaviour. Figure 7.8 exemplifies this effect with two computations for
test D.TW.04: one using a kinematic hardening model and one using a ratcheting
model employing the example evolution functions from Houlsby et al. (2017), which
follow Equation 7.62.
Instead, a modified power-law function for Rm is proposed, which reduces
distortion of Masing behaviour:

1 !!−mr
β R0

mr
Rm = R0 1+ ∗ −1 (7.63)
β R∗
At large values of β this function resembles a power-law, but at small values of β the
rate of change of the function is limited. Parameter R0 is the value of the function
at β = 0, exponent mr is positive and controls the rate of change of the function

at large β, and the pair of parameters β ∗ and R∗ (R∗ < R0 ) control the magnitude
of the function (by pinning it to point ∗). As the parameters β ∗ and R∗ are coupled,
one is arbitrary. The function is illustrated in Figure 7.9 and may also be expressed
in the following symmetric form:

1 ! 1 !
1 R0 1 R0

mr mr
−1 = ∗ −1 (7.64)
β Rm β R∗
The function for Rm should also capture the power-law increase in magnitude of
ratcheting with cyclic amplitude (key behaviour 5, Table 7.1). Here, stress-dependency
is captured by making R∗ a function of stress σ. For bi-directional loading:

q ms
σx2 + σy2
R∗ = RR
∗   (7.65)
σR
q
∗ is the value of R∗ at
Where RR σx2 + σy2 = σR and the exponent ms is positive. The
resulting combined expression for Rm does not vary with surface m and is expressed
here as R:
   1 −mr
mr
 β  R0 
(7.66)

R = R0 1 + √ − 1
  
ms 
β∗
 2 2
σx +σy 
R∗
  
R σR
q
Figure 7.10 plots the variation of R with β and σx2 + σy2 /σR = |σ|/σR , using example
parameter values. The pinned point (β ∗ , RR

∗ ) is indicated (∗). The parameter β ∗ is
defined in terms of a global work metric in Equation 7.73.
Function for Km
To capture the change in secant stiffness kn under cyclic loading the slider strengths
Km are made a function of the hardening parameter β and expressed in terms of
κ = Km /Km0 . In order to use experimental secant stiffness kn values to inform the
choice of evolution function for Km (or equivalently κ), the relationship between κ
and secant stiffness kn is first explored.
Figure 7.11 shows how κ scales the computed backbone curves, for the example
case of κ = 2. The increase in secant stiffness (between σ1 and σ2 , where σ2 > σ1 ) for
a given κ may be approximated with the Masing rules, given an analytical expression
Figure 7.10: Variation of R (Equation 7.66) with β and |σ|/σR

(R0 = 1, mr = 0.85, β ∗ = 1, RR
∗
= 0.2, ms = 3)
for the backbone curve (ε = f (σ)). For a backbone curve defined with a power-law
(Equation 7.27) secant stiffness approximately varies with κ as:

Pp
σ1 −σ2 σR σ1 −σ2
kn Ei + 2 εR − Ei 2σR
= Pp (7.67)
k0

σ1 −σ2 σR σ1 −σ2
Ei + 2κ1−Pp εR − Ei 2σR
Where k0 is the initial secant stiffness. Meanwhile, for a backbone defined with the
Jeanjean expression (Equation 7.28):

2
σ1 −σ2 σR σ1 −σ2
kn Ei + 2 εR − Ei arctan 2σR tanh(Pj ) /Pj
= 2 (7.68)
k0

σ1 −σ2 σR σ1 −σ2
Ei + 2κ εR − Ei arctan 2κσR tanh(Pj ) /Pj
Figure 7.12 plots kn /k0 for the OU dense backbone curve to demonstrate the
non-linear dependency of kn /k0 on both σ and κ; the dependency is similar for the
other backbone curves.
Accounting for this non-linear dependency, approximate experimental values for κ
(κe ) are determined from measured secant stiffness kn values by employing Equation
7.67 or 7.68 with parameter values from Table 7.2. These values are plotted against
cycle number n in Figure 7.13 for exemplar 1-way and 2-way tests from each dataset.
Neglecting ratcheting, cycle number n is approximately proportional to β for regular
cyclic loading. There is some variation in the shape of evolution of κe between 1-way
and 2-way loading, consistent with the observed variation of stiffness evolution with
κ=1
κ=2
kn k0
σ2
σ1
ε1 ε2 ε0,1 ε0,2
Figure 7.11: Illustration of impact of Figure 7.12: Approximate variation of secant

κ = Km /Km0 on secant stiffness kn stiffness kn /k0 with κ and (σ2 − σ1 )/σR for
OU dense backbone curve
Figure 7.13: Variation of experimental value for κ (κe ) with cycle number N (approximately
proportional to β)
load type. The magnitude of κe also varies between datasets. However, overall, the
data plots approximately linearly in log-log space and supports the use of a power-law
to capture the evolution of κ with β. Power-law functions were also used by Abadie
(2015), Houlsby et al. (2017) and Abadie et al. (2019a).
Modified power-laws with a similar functional form to that used for R are proposed
to capture the evolution of stiffness. These functions approximate power-laws at large β
and ensure κ = 1 at β = 0. Stiffening can be applied globally to all surfaces (as ratcheting
is, employing global hardening variable β) or locally at each surface m (employing local
κg , κm
log scale
mkm mkg
κg*
*
κm*
* * *
1
log scale
β*1 β*2 β*m β* β, βm
Figure 7.14: Illustration of proposed evolution functions for κm = Km /Km0 and κg = Km /Km0
hardening variable βm ). Local stiffness evolution is represented by κm (Equation 7.69)
and global stiffness evolution by κg (Equation 7.70):
Km

βm ∗(1/mkm ) mkm
κm = = 1 + ∗ κm −1 (7.69)
Km0 βm
Km

β ∗(1/mkg ) mkg
κg = = 1 + ∗ κg −1 (7.70)
Km0 β
These functions are illustrated in Figure 7.14. At βm = 0 or β = 0, the functions are
equal to unity, while at large βm or β the functions evolve as power-laws with
exponents mkm or mkg . Stiffening behaviour is captured with positive exponents, while
softening behaviour can be captured with negative exponents (Balaam, 2020). The
∗ , κ∗ ) or (β ∗ , κ∗ ). Although
function magnitude is controlled by pairs of parameters (βm m g
parameters κ∗m and mkm control local stiffness evolution they are assumed not to vary
with surface m.
The local stiffening expression (Equation 7.69) is able to capture changes in secant
stiffness kn (key behaviour 7, Table 7.1), while the global stiffening expression
(Equation 7.70) is also able to capture changes in stiffness beyond the strained region,
as observed during post-cyclic reloading (key behaviour 10, Table 7.1). Neither
equation alters the initial stiffness H0 . Combined local and global stiffening may be
captured with the following expression (Balaam, 2020), which partitions local (κm ) and
global (κg ) stiffening with a factor ρκ :
Km
κ= = κρmκ κ1−ρ
g
κ
(7.71)
Km0
To ensure consistency of evolution of κm across the surfaces the local reference

∗ is set as proportional to a surface-specific work quantity.
hardening parameter βm
Here, the following expression is used:
∗ Ku Km0
βm = Fβ ∗ (7.72)
Hm
∗ = F ∗ K 2 /H may be used, which leads
Where Ku = Km0 at m = M . Alternatively, βm β m0 m
to very similar results. The global reference hardening parameter β ∗ may then be set
as a corresponding global parameter. Here, the following expression is used:
β ∗ = Fβ ∗ 2A (7.73)
Where A is the area beneath the backbone curve (εm , σm ) to Ku .
7.5.5 Evolution function parameters
Table 7.3 summarises the necessary parameters for calibration of the evolution
functions for R (Equation 7.66) and Km (Equations 7.69, 7.70 and 7.71). Four
parameters calibrate the evolution of R while five parameters calibrate the evolution of
Km . However, the local and global stiffening parameters are chosen to be equal here
(mkg = mkm , κ∗g = κ∗m ), for simplicity. Additionally, parameter Fβ ∗ controls the value of
the reference hardening parameters β ∗ and βm

∗ . With F ∗ = 5 the reference hardening
β
parameters tend to locate where the functions approximate power-laws, which
increases parameter independence and therefore eases calibration; larger values of Fβ ∗
would also be appropriate.
The evolution function parameters were manually chosen to approximately fit the
regular, unidirectional, 1-way cyclic loading data. Focus was placed on capturing the
ratcheting and stiffness evolution at high cycle numbers — and in combination — the
post-cyclic reloading response. The resulting parameters are summarised in Table 7.4
for each dataset. Calibration methods are not explored in detail here, although the
sensitivity of the computations to the parameters is explored. Analytical calibration
methods have been presented by Abadie et al. (2019a) (for alternative evolution
functions), while Balaam (2020) is exploring methods for calibration of HARM models
from element testing data.

R evolution function parameters (Equation 7.66)
R0 Value of R at β = 0 and |σ| = σR .

∗
RR Value of R at β = β ∗ and |σ| = σR , controls magnitude of R.
ms Positive exponent which controls the stress-dependency of the magnitude of R.
mr Positive exponent which controls the rate of reduction of R with β at large β.
Km evolution function parameters (Equations 7.69, 7.70, 7.71)
κ∗g Value of κg at β = β ∗ , controls the magnitude of κg .

κ∗m Value of κm at βm = βm
∗
, controls the magnitude of κm . Assumed not to vary with
surface m.
mkg Exponent which controls the rate of change of κg with β at large β, positive for
stiffening.
mkm Exponent which controls the rate of change of κm with βm at large βm , positive
for stiffening. Assumed not to vary with surface m.
ρκ Factor which partitions local and global stiffness change, ρκ = 1 for solely local
stiffness change, ρκ = 0 for solely global stiffness change.
Fβ ∗ Factor which controls the value of β ∗ and βm

∗
.
Table 7.3: Description of parameters for calibration of evolution functions
Parameter OU very loose OU dense UWA 1g UWA 9g UWA 80g
R0 1 1 1 1 1
∗
RR 0.08 0.08 0.09 0.03 0.01
ms 2.8 2.2 3.2 3.2 3.2
mr 0.65 0.65 0.8 0.9 1.0
κ∗g = κ∗m 2 2 2 2 2
mkg = mkm 0.1 0.3 0.3 0.1 0.06
ρκ 0.7 0.7 0.7 0.7 0.7
Fβ ∗ 5 5 5 5 5
Table 7.4: Summary of evolution function parameters manually calibrated to fit the regular,
unidirectional, 1-way cyclic loading responses
Figure 7.15 explores the sensitivity of the computations to the parameter values for
an example 1-way test in the UWA 9g dataset (9G.OW.03). The impact of each
parameter on either ratcheting, secant stiffening or reloading is explored in turn,
keeping the remaining parameters constant and equal to the values chosen for the
UWA 9g dataset and presented in Table 7.4. ∗ , m , m and

For parameters RR s r
mkg = mkm , sensitivity is explored across the range of parameter values obtained by
manual calibration to each of the five datasets. For parameters R0 , κ∗g = κ∗m and ρκ ,
where the parameter values do not vary across the manual calibrations, sensitivity is
explored across an arbitrary parameter range.
Figure 7.15b, c, d, f shows that the computations are sensitive to parameters

∗ , m , m and m
RR s r kg = mkm across the range of values obtained in manual calibration.
∗ and m affect ratcheting magnitude as expected, but also impact

Parameters RR s
ratcheting rate. Meanwhile, parameter mr only subtly affects ratcheting rate across the
range of values obtained in manual calibration, but also impacts ratcheting magnitude.
Parameter mkg = mkm affects the rate of stiffness evolution, as expected.
Figure 7.15a shows that the computations are not sensitive to R0 across 0.1 ≤ R0 ≤
10, which justifies the choice of a constant value for all computations. R0 = 1 was
chosen as ratcheting strain dαr tends to be small compared to the associated plastic
PM
strain m=1 dαm (this is not, however, a theoretical requirement).
A constant value of parameter κ∗g = κ∗m was found to be suitable across the datasets,
however, Figure 7.15e demonstrates that the computations are sensitive to this

parameter across the arbitrary range 1.5 ≤ κ∗g = κ∗m ≤ 5. A constant value of ρκ was
also found to be suitable across the datasets. Indeed, including some global stiffening
(0 ≤ ρκ < 1.0) ensures that Ki < Ki+1 as Km evolves and avoids the computation of an
elastic hysteretic response (which would occur if Ki ≥ Ki+1 in the loading region).
Figure 7.15g demonstrates the impact of ρκ on the reloading response over its defined
range 0 ≤ ρκ ≤ 1.
Some trends can be observed across the parameters obtained by manual calibration.
∗ and m
In particular, parameters RR kg = mkm decrease with increasing stress-level
across the UWA datasets, consistent with the experimental observations in Chapter 6.
The parameter ms is also found to be approximately equal to experimental exponent
mσ , as highlighted in Figure 7.16. The experimental exponent mσ captures the power-
law variation of ratcheting magnitude with cyclic amplitude under 1-way loading (see
Figure 4.8, Section 4.4.1; Figure 6.14, Section 6.5.2; Leblanc et al. (2010a) and Abadie
(2015)). The similarity of these parameters may therefore be expected, as both capture
the dependence of ratcheting magnitude on load. It would be appropriate to take
ms = mσ for initial calibration of this model to other datasets.

∗
(a) R0 (b) RR
(c) ms (d) mr
(e) κ∗g = κ∗m (f ) mkg = mkm
Figure 7.15: Exploration of sensitivity of computations to evolution function parameter values

for example test in UWA 9g dataset (9G.OW.03)
( parameter values obtained in manual calibrations, arbitrary parameter values)
(g) ρκ
Figure 7.15: Exploration of sensitivity of computations to evolution function parameter values

for example test in UWA 9g dataset (9G.OW.03)
( parameter values obtained in manual calibrations, arbitrary parameter values)
Figure 7.16: Correlation between experimental exponent mσ and evolution function parameter
ms
7.5.6 Computation of regular cyclic loading response
Figures 7.17 to 7.21 present computations of the regular cyclic response for each of
the five datasets, employing either the unidirectional ratcheting model described in
Section 7.5.1 or the bi-directional ratcheting model described in Section 7.5.2, with the
evolution functions presented in Section 7.5.4. Section 7.4.3 describes calibration of the
backbone parameters Km and Hm , while Table 7.4 summarises the evolution function
parameters, which were manually calibrated to unidirectional, 1-way responses. For

the UWA datasets, computations are presented for all tests conducted, as summarised
in Tables 6.3 and 6.4, Section 6.3. For the OU datasets, where a greater number of
tests were conducted, computations are presented for a representative selection of
the tests summarised in Tables 4.1 and 4.2, Section 4.2.
Each Figure (7.17 to 7.21) presents a unidirectional hysteretic response, the
evolution of ratcheting, secant stiffness and energy loss factor under various
unidirectional and multidirectional cyclic loadings, and post-cyclic reloading
responses; a multidirectional hysteretic response is also presented for the OU datasets.
The corresponding experimental responses are presented (dashed) alongside the
computations, to facilitate direct comparison. Some aspects of the responses are
captured very well and others are captured less well, but together, Figures 7.17 to 7.21
demonstrate the ability of the calibrated models to capture the overall patterns of
behaviour observed experimentally.
The inclusion of ratcheting and stiffening precludes exact adherence of the model
computations to the extended Masing rules. Nevertheless, Masing behaviour is not
significantly distorted and the unidirectional hysteretic responses are generally
captured with good accuracy; the OU very loose case is an exception, where pile
rotation is significantly underestimated upon unloading to M/MR = 0.6. In contrast,
the multidirectional hysteretic responses (for the OU datasets) present a challenge for
the model. Pile rotation is underestimated at both densities, perhaps due to too rapid
an increase in Km . However, the inability of the models to capture this response is not
a key concern, as this loading is not representative of that experienced by monopiles.
The magnitude and evolution of ratcheting under unidirectional, 1-way cyclic
loading is captured accurately at various load amplitudes across the five datasets;
indeed, these data were used for calibration of the evolution function parameters.
Under perpendicular cyclic loading ratcheting is overestimated for the UWA 1g dataset
but is captured well for the remaining four datasets: ratcheting under L-shape loading
is captured accurately in both directions, while ratcheting under T-shape loading tends
to be slightly underestimated. Ratcheting under moderate (Φ = 30°) fan-type loading is
captured with reasonable accuracy for the OU datasets, although the models cannot
(a) Unidirectional hysteretic response (b) Multidirectional hysteretic response
(c) Ratcheting response (d) Secant stiffening
(e) Evolution of energy loss factor (f ) Post-cyclic reloading
Figure 7.17: Computation of regular cyclic responses for OU very loose dataset using
ratcheting models
(E.g. model computations, experimental data)
(a) Unidirectional hysteretic response (b) Multidirectional hysteretic response
(c) Ratcheting response (d) Secant stiffening
(e) Evolution of energy loss factor (f ) Post-cyclic reloading
Figure 7.18: Computation of regular cyclic responses for OU dense dataset using ratcheting
models
(a) Hysteretic response
(b) Ratcheting response (c) Secant stiffening
(d) Evolution of energy loss factor (e) Post-cyclic reloading
Figure 7.19: Computation of regular cyclic responses for UWA 1g dataset using ratcheting
models
models
models
capture the significant increase in ratcheting observed under fan-type loading at
greater spread angles. The OU datasets also highlight the inability of the models to
capture the increase in ratcheting that tends to be observed under partial 2-way
loading.
The evolution of secant stiffness kn under unidirectional, 1-way loading is captured
reasonably well across the datasets. However, there is some variation in the magnitude
of secant stiffness — which will also depend on the monotonic calibration — for the
UWA 1g and OU very loose datasets. Under multidirectional (T-shape, L-shape and
moderate fan-type) loading, the evolution of secant stiffness is generally captured as
well as under unidirectional loading, while the stiffness evolution under 2-way and
partial 2-way loading is often underestimated.
The magnitude and evolution of the energy loss factor ηn under unidirectional and
multidirectional loading is captured very well for the UWA 1g dataset, reasonably well
for the UWA 9g dataset and overestimated for the UWA 80g dataset. This trend is at
odds with the increasing accuracy of the computed ratcheting and stiffness values for
the UWA datasets as stress-level increases, and suggests that evolution of the energy
loss factor may not be wholly explained by evolution of secant stiffness and ratcheting.
For the OU datasets, energy loss factor is significantly and systematically
underestimated where the experimental data is distorted by transducer friction
contributions (at high cycle number and low cyclic amplitude). This represents a
limitation of the experimental data, rather than the numerical model.
The post-cyclic reloading responses are captured approximately for the UWA 1g
dataset and reasonably well for the remaining four datasets. However, the inferred
capacity on reloading tends to be underestimated where secant stiffness evolution is
underestimated (often under 2-way or partial 2-way loading).
Together, Figures 7.17 to 7.21 demonstrate the ability of the calibrated ratcheting
models to capture ratcheting, evolution of stiffness and (in some cases) energy loss
factor under unidirectional and multidirectional (perpendicular and moderate fan-
type) cyclic loading at various cyclic amplitudes, in three sand densities and at multiple
stress-levels. Distortion of Masing behaviour is limited and the post-cyclic reloading

responses are captured with reasonable success. The accuracy of the computations
may also be improved by optimisation of the evolution function parameters.
The ability of the models to capture the response to perpendicular cyclic loading
builds confidence in the bi-directional ratcheting model formulation. However, model
development is necessary to capture the increase in ratcheting and stiffening under fan-
type loading where Φ > 30°. The models are also unable to capture the full impact of
load asymmetry: ratcheting under partial 2-way loading is underestimated and stiffness
evolution under 2-way and partial 2-way loading is often underestimated. This is of
concern as much of the loading experienced by a monopile is expected to be partial
2-way (see e.g. Figure 5.6, Section 5.3.5). Development of a model which better captures
the impact of load asymmetry would therefore be of great value.
7.5.7 Computation of response to storm loading
Figures 7.22 and 7.23 present computations for two example multidirectional storms
using the ratcheting model presented in Section 7.5.2 with the evolution functions
outlined in Section 7.5.4. The results for both the OU very loose and dense datasets are
presented, in terms of the moment-rotation (M − θ) response and rotation per cycle
on loading (θna ). The corresponding experimental responses are plotted alongside the
computations for comparison, and the results are presented in the dominant loading
direction (x) only. The computations used the same backbone parameters (Km , Hm ,
Section 7.4.3) and evolution function parameters (Section 7.5.5) used to compute the
regular cyclic loading responses in Section 7.5.6. The storm responses did not inform
model calibration, but the responses were known at the time of model computation;
the computations are therefore Class C1 predictions (Lambe, 1973).
In general, the computations capture the key features of the monopile response to
storm loading: ratcheting is observed during application of small-amplitude cyclic
loading but the response is dominated by large load events. The peak rotations are
predicted particularly accurately in all cases. However, the OU very loose computations
over-predict the rotation per cycle on loading (θna ) for both tests, principally because
the rotation accumulated across the peak load cycle (at n ≈ 1000) is significantly
over-predicted. This behaviour is related to the poor prediction of the hysteresis loop
Figure 7.22: Computation of example storm responses for OU very loose dataset
model computations,
* First half of test L.MD.W.x presented ( experimental data, peak
load response)
shape, which is also observed in computation of the unidirectional hysteretic response
for the OU very loose dataset (Figure 7.17a). The hysteretic response and evolution of
θna can be better predicted by reducing the evolution of Km (e.g. with mk = 0.01),
however, this would be at odds with the observed increase in secant stiffness under
regular cyclic loading.
The hysteresis loop shapes are predicted more accurately for the OU dense
computations, as too are the accumulated rotations across the peak loads (at n ≈ 750).
However, the rate of ratcheting under small amplitude cycling is slightly
under-predicted, causing an under-prediction of the rotation per cycle on loading (θna )
across the tests. This behaviour can also be observed in the OU very loose
computations, and may be explained by the inability of the presented models to

Figure 7.23: Computation of example storm responses for OU dense dataset

( model computations, experimental data, peak load response)
capture the increase in ratcheting under partial 2-way loading, which constitutes the
majority of the loading in these example storms.
7.6 Prediction of prototype-scale response to storm loading
In this Section, a prototype-scale model is developed to allow exploration of the
response of a monopile to storm loading at prototype-scale. This brief study reflects
the overall aim of the coordinated programme of research to which this thesis
contributes: to develop practical design methods to capture the response of full-scale
monopiles to realistic cyclic lateral loading.

7.6.1 Generation of prototype-scale model
A prototype-scale ratcheting model is generated here by scaling the bi-directional
ratcheting model (Section 7.5.2) for the UWA 80g dataset using the dimensionless
framework presented by Leblanc et al. (2010a). An equivalent prototype-scale
kinematic hardening model (Section 7.4.2) is also generated for comparison. The
models are scaled to represent a prototype monopile with diameter 6.9 m and L/D = 4;
the same size prototype was used to scale down the prototype-scale DeRisk loads to
laboratory-scale for application in dense sand (see Section 5.3.4). There is confidence
in the use of the dimensionless framework of Leblanc et al. (2010a) to account for the
impact of foundation size and stress-level on the monotonic response (Kelly et al.
(2006); Section 6.4.1), but the framework does not account for the impact of stress-level
on ratcheting and stiffness evolution revealed in Chapter 6. The dimensionless
framework is therefore used to scale the UWA 80g backbone curve, while the evolution
function parameters calibrated to the UWA 80g dataset are used directly at
prototype-scale (for the ratcheting model), given that the prototype-scale stress-level is
close to that simulated in the UWA 80g tests.
Scale Pile diameter Pile length Unit weight Normalised reference

D [m] L [m] 3
γ 0 [kN/m ] stress-level σREF
0
/pa
UWA 80g 0.042 0.170 17 × 80 = 1360 1.60

Prototype 6.9 27.6 7.12 1.36
Table 7.5: Parameters used for scaling UWA 80g models to prototype-scale
The laboratory- and prototype-scale parameters necessary for scaling the backbone
curve are summarised in Table 7.5. The prototype-scale unit weight was determined by
matching φ0p between laboratory- and prototype-scale using the relations presented by
Bolton (1987); the prototype soil is assumed to be saturated. The scaling factors on
horizontal load fH and pile displacement fu are determined from the expressions
presented in Table 3.4, Section 3.4.3. Employing the values in Table 7.5, the scaling
factors become:
fH = HP /HL = 22.67 × 103 (7.74)

fu = uP /uL = 151.5 (7.75)
Table 7.6 summarises the resulting prototype-scale backbone parameters, where the
backbone is defined by a power-law (Equation 7.27). The prototype-scale parameters
Km and Hm are then obtained using Equations 7.25 and 7.26.
σ R = HR εR = uR Ei Pp
13.53 MN 2.70 m 33.04 MN/m 1.82
Table 7.6: Prototype-scale backbone curve parameters (backbone defined by a power-law,

Equation 7.27)
7.6.2 Response to storm loading
Figure 7.24 presents the response of the prototype-scale models to an example storm.
The responses are presented in the dominant loading direction (x) only, in terms of
the load-displacement (H − u) response at ẽ = h/L = 2.5 and the pile displacement
per cycle on loading (una ). The loading was derived from the DeRisk wave basin tests
(described in Section 5.3) and corresponds to a multidirectional sea-state with a 100-
year return period acting on a 7 m diameter cylinder, with additional constant wind
loading (MD.W). The kinematic hardening and ratcheting models predict very similar
pile displacement at peak load (≈ 1.8 m), approximately corresponding to 1.16° pile
rotation (assuming the model monopile used for the UWA 80g tests was perfectly
rigid). These displacements are towards the upper limit of what may be tolerated in
ULS design, but are not unrealistic.
For both models, the hysteretic responses are more linear than observed under
equivalent loading for the OU very loose and dense datasets, given the higher stress-
level. For the ratcheting computation, the linearity of the cyclic response, coupled with
evolution parameters which capture the (lower) rate of ratcheting at approximately
prototype-scale stress-level, leads to prediction of negligible ratcheting between peak
load cycles. However, the ratcheting model does predict greater accumulation of pile
displacement across the peak load cycles than the kinematic hardening model. Some
stiffening can be observed in the hysteretic response computed using the ratcheting
model, although the change in loop shape is not significant.

Figure 7.24: Computation of response of prototype-scale monopile to example storm (MD.W)

( kinematic hardening model, ratcheting model, response to peak load)
These computations suggests that a kinematic hardening model may
approximately capture the peak response and hysteretic response over a short storm,
but may underestimate the accumulation of pile displacement. Although the
difference in the pile displacement computed using the two models is small for the
short (6 hour) storm computed here, the contribution from ratcheting may become
large over the 20 − 25 year lifetime of the structure.
These computations demonstrate how numerical models, informed by
experimental testing, can be used to explore prototype-scale responses when coupled
with appropriate scaling techniques. However, validation of the scaling techniques
(particularly the use of evolution function parameters calibrated to centrifuge tests at
approximate prototype stress-level) through e.g. large-scale field testing may be
necessary before such scaled models can be applied with confidence for full-scale
design.
7.7 Summary
Key behaviour Capturing key behaviour
1 The monopile exhibits a non-linear Underlying kinematic hardening model is

monotonic response. able to capture any monotonically
increasing non-linear response.
2 The extended Masing rules are Underlying kinematic hardening model
approximately adhered to during the first adheres to extended Masing rules,
few loading cycles. evolution functions and parameters
chosen to minimise distortion of Masing
rules.
3 Ratcheting occurs i) under biased cyclic The ratcheting strain increment occurs in
loading and ii) in the direction of the the direction of the incremental applied
mean load. load (Equations 7.41 and 7.56).
4 Ratcheting evolves as a power-law with Evolution function for R (Equation 7.66)
cycle number. evolves as a power-law with hardening
parameter β at large β.
5 The magnitude of ratcheting increases as The magnitude of R (Equation 7.66)
a power-law with cyclic amplitude (for varies as a power-law with incremental
constant ζc ). load magnitude.
6 Load asymmetry (ζc ) affects the shape, Behaviour not fully captured: more
rate and therefore magnitude of ratcheting observed than predicted.
ratcheting.
7 Secant stiffening occurs under cyclic Evolution function for Km captures
loading and evolves logarithmically or as power-law increase in local Km at large
a power-law, depending on the load βm , but impact of load asymmetry not
asymmetry. captured.
8 The cyclic response is broadly Evolution of R and Km depend on
independent of the cyclic loading hardening parameter β (Equation 7.61)
direction, relative to the mean load which is direction-independent.
direction.
9 Multidirectional fan-type loading can Behaviour not fully captured: more
lead to greater ratcheting and stiffening ratcheting and stiffening observed than
than unidirectional loading. predicted.
10 The inferred capacity on post-cyclic Evolution function for Km can capture
reloading is equal to or greater than the power-law increase in all (global) Km at
monotonic capacity. large β.
Table 7.7: Description of whether and how the key behaviours observed experimentally are
captured with the presented ratcheting models
This Chapter has demonstrated the ability of constitutive models in the HARM
framework to capture many key features of the response of a monopile to complex
cyclic lateral loading in dry sand. Multi-surface kinematic hardening models were able
to capture the hysteretic response for the first few cycles of loading, but ratcheting
models were necessary to capture the high-cycle response. Both unidirectional and
bi-directional models were presented, formulated as rate-independent, series models
and implemented as macro (single-element) models.
New evolution functions were proposed to capture the evolution of ratcheting and
stiffness under cyclic loading. These functions avoid significant distortion of the
Masing rules, are able to capture the post-cyclic reloading response and are calibrated
with parameters which all have an interpretable impact on the computed cyclic
response. The evolution functions are applicable to equivalent rate-dependent and
parallel models, and are expected to be suitable for application in 1D Winkler models.
The ratcheting models were calibrated to the monotonic responses and the
ratcheting and stiffening behaviour under regular, unidirectional, 1-way cyclic loading
for five datasets (OU very loose, OU dense, UWA 1g, UWA 9g and UWA 80g).
Computation of the cyclic response under a variety of regular cyclic loadings
demonstrates the ability of the models to capture the majority of the key behaviours
outlined in Table 7.1. Table 7.7 summarises whether and how each of the key
behaviours are captured. A key limitation is the inability of the models to capture the
full impact of load asymmetry and multidirectional fan-type loading (for Φ > 30°) on
the cyclic response.
The bi-directional ratcheting model, calibrated to regular, unidirectional, 1-way
data, was also shown to capture the key features of the response to realistic, irregular,
multi-amplitude and multidirectional storm loading. This result builds confidence in
the use of models developed in the HARM framework for design in realistic loading
scenarios. Models were also scaled to explore the potential response at prototype-scale,
reflecting the overall aim of this research.
The development of practical calibration methods for the evolution function
parameters, derived from standard element testing results, is essential if these models
are to be used for design. Balaam (2020) is exploring the correlation between
parameters calibrated to cyclic element tests using a single-element model and
parameters calibrated to the response of monopile systems using a 1D Winkler model.

There is confidence that correlations between these parameters can be observed, given
the similarity in the cyclic response observed in element tests and in model pile tests in
similar soils (Truong et al., 2019).

Chapter 8
Conclusions
8.1 Introduction
This thesis has explored the response of monopile foundations in sand to complex
cyclic lateral loading through physical modelling. Regular cyclic loading tests allowed
systematic investigation of monopile behaviour, while the application of storm loading
and exploration of stress-level effects has provided insight into the response under
realistic loading and at full-scale. The physical modelling data has also facilitated
demonstration of the ability of models in the HARM framework to capture key features
of the monopile response to complex cyclic lateral loading. Summary sections at the
end of each Chapter have presented detailed conclusions; this Chapter distils the key
contributions and implications for design, and makes suggestions for future work.
8.2 Key contributions

Development of novel laboratory apparatus for 1g monopile testing
New laboratory apparatus was developed to apply complex lateral loading to a model
monopile. The loading and pile measurement apparatus, with associated Labview
software, is capable of applying continuously varying, multi-amplitude and
multidirectional loading under accurate load control, and was used to perform the 60
1g tests at OU presented in Chapters 3, 4 and 5.
Regular cyclic loading response at 1g
Regular cyclic loading tests at 1g in very loose and dense sand revealed i) approximate
adherence to the extended Masing rules over the first few loading cycles (Section 4.3),
209
8. Conclusions 210
ii) accumulation of pile rotation (ratcheting) under many cycles of biased loading (e.g.
Section 4.4.1), and iii) evolution of the hysteretic response (characterised in terms of
secant stiffness and energy loss factor) under many cycles of general cyclic loading
(e.g. Sections 4.4.2 and 4.4.3), consistent with previous studies (e.g. Abadie, 2015). The
results also highlighted the significant impact of load amplitude and load asymmetry.
Novel perpendicular multidirectional tests revealed that ratcheting occurs in the
direction of mean load and that the cyclic response is insensitive to cyclic loading
direction (Section 4.6.1), with implications for numerical modelling (Section 7.2).
Meanwhile, spread fan-type loading tests demonstrated that an increase in ratcheting
and secant stiffening is possible under multidirectional loading, compared to
equivalent unidirectional loading (Section 4.6.2).
The post-cyclic reloading responses showed no decrease in capacity relative to the
monotonic response, at odds with the conventional cyclic degradation approach
(Section 4.7). In general, the behaviour in very loose and dense sand was qualitatively
similar.
Application of realistic storm loading at 1g
Irregular, multi-amplitude and multidirectional storm loading was derived from wave
tank tests performed as part of the DeRisk project (Section 5.2); careful processing of
the loading data was necessary, particularly to ensure structural dynamic effects were
accounted for. Application of these realistic loading signals to the model monopile led
to behaviour consistent with that observed under regular cyclic loading, and highlighted
the dominance of large load events (Section 5.4). The results suggested that, for a short
storm, peak monopile rotation may be approximated by the monotonic response.
Investigating the effect of stress-level
Exploration of stress-level effects — through 21 monotonic, unidirectional and
multidirectional cyclic loading tests — revealed qualitatively similar responses at
g-levels from 1g to 80g, building confidence in the applicability of observations made
at 1g to full-scale. However, a logarithmic reduction in the exponents controlling the
rate of ratcheting and secant stiffness evolution with stress-level was observed
8. Conclusions 211
(Sections 6.5.2 and 6.5.3). This suggests that the effects of cyclic loading will be
significantly less pronounced at full-scale than observed in small-scale 1g physical
modelling. These results help inform comparison of behaviour observed at different
stress-levels, and support the dimensionless framework proposed by Leblanc et al.
(2010a) for scaling of the monotonic response of the monopile (Section 6.4.1).
Modelling the response of a monopile to complex cyclic loading
Together, the physical modelling results facilitated demonstration of the ability of
models in the HARM framework to capture key features of the monopile response to
complex cyclic loading in (dry) sand at multiple stress-levels and in various densities,
following calibration to regular, unidirectional data. Multi-surface kinematic hardening
models were able to approximate the hysteretic response (Section 7.4.4), but ratcheting
models were necessary to capture ratcheting and evolution of hysteresis loop shape
under many cycles (Section 7.5.6).
The physical modelling results also informed new functions for capturing the
evolution of ratcheting and hysteresis loop shape (Section 7.5.4). In particular, these
functions limit distortion of Masing behaviour and improve the ability to capture the
post-cyclic reloading response, analogous to the ULS response.
8.3 Key implications for design

Approximating the response to large loads
The monotonic response is likely to provide a conservative estimate for monopile
rotation under ULS loading for drained conditions (Section 4.7 and 6.5.5), given that
the inferred post-cyclic capacity on reloading was found to be equal to or greater than
the monotonic capacity. The monotonic response may also be used to approximate
the peak foundation rotation during a short storm (Section 5.5).
Assessment of loading conditions
Greater ratcheting and stiffening behaviour was observed under partial 2-way loading
and spread multidirectional loading (Chapter 4). This highlights the need to accurately
assess loading asymmetry and directionality before conducting cyclic design for
monopile foundations.
8. Conclusions 212
Considering stress-level effects
The logarithmic reduction in the exponents controlling ratcheting and stiffening rate
with stress-level (Sections 6.5.2 and 6.5.3) suggests that cyclic loading effects will be
less pronounced at full-scale than observed in small-scale 1g physical modelling. As a
result, direct application of empirical functions for ratcheting and evolution of stiffness
with cycle number, derived from small-scale 1g physical modelling (e.g. Leblanc et
al., 2010a), are not recommended for full-scale design, although the expressions for
ratcheting are likely to be conservative.
Modelling the cyclic response
Multi-surface kinematic hardening models may be appropriate for some dynamic
design calculations (Section 7.4), but ratcheting models are necessary for predicting
permanent monopile rotation and the effects of cyclic loading on foundation strength
and stiffness. Models in the HARM framework are able to capture many aspects of the
response to complex cyclic loading, but development of practical calibration methods
for the evolution functions and parameters is required (Section 7.5). Accurate
prediction of the monotonic response is a pre-requisite for models in the HARM
framework (Section 7.4.3), and cyclic design in general.
Noting the stress-level effects discussed above, for preliminary full-scale cyclic
design in sand, the evolution functions and parameters for the UWA 80g dataset might
be employed, as this dataset most closely represents full-scale stress-levels. The
limitations of the presented models to fully capture the impact of load asymmetry and
spread multidirectional loading should, however, be considered.
8.4 Future work

Physical modelling
The physical modelling presented herein simulated fully-drained conditions, which
may not be representative of field conditions (Section 3.2.2). Exploration of pore-
pressure accumulation and the effect on the cyclic response under partial drainage
conditions, in soils of representative permeability and under realistic loading rates,

8. Conclusions 213
would therefore be of great value. Exploration of the monopile response under complex
cyclic loading in cohesive and layered soils would also complement this work.
Monopiles in the field are exposed to lateral loading with continuously varying
loading eccentricity (Section 1.2.1), but a single loading eccentricity was used for the
physical modelling presented herein. It would therefore be of interest to explore the
effect of loading eccentricity, across a representative range, on the monotonic and
cyclic response.
Exploration of stress-level effects
In exploration of stress-level effects in Chapter 6, a constant dimensional reference pile
displacement uR was employed (Section 6.5.1). As a result, the linearity of the
responses (at a given normalised load) varied with stress-level. Although this variation
in linearity may be interpreted as a stress-level effect, complementary insight into the
effect of stress-level may be gained by elimination of the variation in linearity through
selection of a dimensionless reference parameter e.g. ũR , following the dimensionless
framework of Leblanc et al. (2010a). Ideally, a future study into stress-level effects
would involve tests at a number of cyclic amplitudes such that the results could be
interpreted in terms of i) a constant dimensional reference parameter e.g. uR , and ii) a
constant dimensionless reference parameter e.g. ũR , which minimises the variation in
response linearity.
The centrifuge study presented herein was limited to simulation of monopiles 5.7 m
in diameter (Section 6.1). Extension of this study to simulate prototype monopiles with
diameters of 8–10 m would build confidence in the applicability of the presented trends
at full-scale. Future testing may also consider using a driven installation method.
Loading investigations
Chapter 4 showed how spread multidirectional loading can lead to greater ratcheting
and stiffening than equivalent unidirectional loading; for example, accumulated
rotation after 1000 cycles was found to approximately double with a 90° half internal
spread angle (Section 4.6.2). However, application of example multidirectional storm
loading in Chapter 5 showed only a small impact of multidirectionality (Sections 5.4.3

8. Conclusions 214
and 5.4.4). Future work may investigate typical distributions of cyclic load direction in
detail, to provide more informed inputs for physical and numerical modelling. Further
exploration of typical distributions of cyclic load amplitude and asymmetry would also
be of value.
Numerical modelling
Although able to capture many of the key features of the monopile response to complex
cyclic loading, the models in the HARM framework presented in Chapter 7 are unable
to capture fully the effect of load asymmetry and spread multidirectional loading.
Refinement of the models to capture these effects is likely to be important, but must be
informed by further investigations into typical loading (discussed above). This work
employed a series, rate-independent model, but future work might explore other
model variants.
The development of practical calibration methods for models in the HARM
framework will also be essential if these models are to be used in design. Mapping
between parameters calibrated to cyclic element tests and parameters calibrated to the
response of monopile systems (using 1D Winker models) may be possible.
Alternatively, element testing may inform calibration of 3D FE models, from which
parameters for 1D Winkler models may be obtained. Validation of these models and
associated calibration methods against e.g. large-scale responses in natural deposits
may also be necessary to build confidence in their application for design.
Pragmatic cyclic design
The computed response of a prototype-scale monopile to realistic storm loading
(Figure 7.24, Section 7.6) exhibited negligible ratcheting between peak load events,
illustrating the potential combined effects of i) the reduction in ratcheting and
evolution of hysteresis loop shape at large-scale stress-levels, and ii) the dominance of
large amplitude loads in controlling the rotation response under multi-amplitude
loading. Future work could therefore explore the extent to which the ratcheting
response can be approximated by computation of the response to only the largest load
events (following the work of Abadie, 2017).

8. Conclusions 215
8.5 Overview
As conservatism in the monotonic design of monopile foundations reduces, OWTs are
developed at sites with more demanding environmental conditions, and lifetime
extension is considered, the cyclic design of monopile foundations is increasingly
critical. However, no commonly accepted design methods exist for predicting the
response of monopiles to cyclic lateral loading. This thesis forms part of a coordinated
programme of research at Oxford University which aims to understand monopile
performance and develop practical cyclic design methods for the next generation of
monopiles. Through laboratory-scale physical modelling, this thesis has provided
important insights into the response of monopile foundations in sand under regular
and complex — irregular, multi-amplitude and multidirectional — cyclic loading.
These insights inform monopile design and, in particular, build confidence in the use
of models in the HARM framework for the design of monopiles under realistic loading
and at full-scale.
References
Aasen, S., Page, A. M., Skau, K. S., and Nygaard, T. A. (2017). “Effect of foundation modelling on
the fatigue lifetime of a monopile-based offshore wind turbine”. Wind Energy Science 2,
pp. 361–376. DOI: 10.5194/wes-2-361-2017.
Abadie, C. N. (2015). “Cyclic Lateral Loading of Monopile Foundations in Cohesionless Soils”.
DPhil thesis, University of Oxford.
Abadie, C. N. (2017). Personal communication.
Abadie, C. N., Byrne, B. W., and Levy-Paing, S. (2015). “Model pile response to multi-amplitude
cyclic lateral loading in cohesionless soils”. Proceedings of the 3rd International Symposium
on Frontiers in Offshore Geotechnics (ISFOG). Oslo, Norway, pp. 681–686.
Abadie, C. N., Houlsby, G. T., and Byrne, B. W. (2019a). “A method for calibration of the
Hyperplastic Accelerated Ratcheting Model (HARM)”. Computers and Geotechnics 112,
pp. 370–385. DOI: 10.1016/j.compgeo.2019.04.017.
Abadie, C. N., Byrne, B. W., and Houlsby, G. T. (2019b). “Rigid pile response to cyclic lateral
loading: laboratory tests”. Géotechnique 69.10, pp. 863–876. DOI: 10.1680/jgeot.16.P.325.
Achmus, M., Abdel-Rahman, K., and Kuo, Y. S. (2007). “Numerical Modelling of Large Diameter
Steel Piles under Monotonic and Cyclic Horizontal Loading”. Proceedings of the 10th
International Symposium on Numerical Models in Geomechanics (NUMOG). Rhodes,
Greece.
Achmus, M., Kuo, Y. S., and Abdel-Rahman, K. (2009). “Behavior of monopile foundations
under cyclic lateral load”. Computers and Geotechnics 36.5, pp. 725–735. DOI:
10.1016/j.compgeo.2008.12.003.
Achmus, M., Albiker, J., and Abdel-Rahman, K. (2011). “Investigations on the behavior of large
diameter piles under cyclic lateral loading”. Proceedings of the 2nd International
Symposium on Frontiers in Offshore Geotechnics (ISFOG). Perth, Australia, pp. 471–476.
Adcock, T. A. A. and Taylor, P. H. (2009). “Estimating ocean wave directional spreading from an
Eulerian surface elevation time history”. Proceedings of the Royal Society A 465,
pp. 3361–3381. DOI: 10.1098/rspa.2009.0031.
Albiker, J., Achmus, M., Frick, D., and Flindt, F. (2017). “1g Model Tests on the Displacement
Accumulation of Large-Diameter Piles Under Cyclic Lateral Loading”. Geotechnical Testing
Journal 40.2, pp. 173–184. DOI: 10.1520/GTJ20160102.
Alderlieste, E. A., Dijkstra, J., and van Tol, A. F. (2011). “Experimental investigation into pile
diameter effects of laterally loaded monopiles”. Proceedings of the 30th International
Conference on Ocean, Offshore and Arctic Engineering (OMAE), pp. 985–990.
Altaee, A. and Fellenius, B. H. (1994). “Physical modeling in sand”. Canadian Geotechnical
Journal 31.3, pp. 420–431. DOI: 10.1139/t94-049.
216
References 217
Andersen, K. H. (2015). “Cyclic soil parameters for offshore foundation design”. Proceedings of
the 3rd International Symposium on Frontiers in Offshore Geotechnics (ISFOG). Oslo,
Norway, pp. 3–82.
API (2011). Geotechnical and Foundation Design Considerations ANSI/API RP 2GEO. 1st Ed.
Arany, L., Bhattacharya, S., Macdonald, J. H. G., and Hogan, S. J. (2016). “Closed form solution
of Eigen frequency of monopile supported offshore wind turbines in deeper waters
incorporating stiffness of substructure and SSI”. Soil Dynamics and Earthquake
Engineering 83, pp. 18–32. DOI: 10.1016/j.soildyn.2015.12.011.
Arany, L., Bhattacharya, S., Macdonald, J., and Hogan, S. J. (2017). “Design of monopiles for
offshore wind turbines in 10 steps”. Soil Dynamics and Earthquake Engineering 92,
pp. 126–152. DOI: 10.1016/j.soildyn.2016.09.024.
Arshad, M. and O’Kelly, B. C. (2017). “Model Studies on Monopile Behavior under Long-Term
Repeated Lateral Loading”. International Journal of Geomechanics 17.1. DOI:
10.1061/(ASCE)GM.1943-5622.0000679.
Bachynski, E. E., Kristiansen, T., and Thys, M. (2017). “Experimental and numerical
investigations of monopile ringing in irregular finite-depth water waves”. Applied Ocean
Research 68, pp. 154–170. DOI: 10.1016/j.apor.2017.08.011.
Balaam, T. D. (2020). “Calibration of cyclic loading models for monopile foundations”.
Forthcoming DPhil thesis, University of Oxford.
Balaam, T. D., Houlsby, G. T., Page, A. M., Jostad, H. P., and Byrne, B. W. (2020). “Predictions of
multi-amplitude laboratory tests using hyperplasticity models”. Forthcoming Proceedings of
the 4th International Symposium on Frontiers in Offshore Geotechnics (ISFOG). Austin, USA.
Bayton, S. M., Black, J. A., and Klinkvort, R. T. (2018). “Centrifuge modelling of long term cyclic
lateral loading on monopiles”. Proceedings of the 9th International Conference on Physical
Modelling in Geotechnics (ICPMG). London, UK, pp. 689–694.
Beuckelaers, W. J. A. P. (2017). “Numerical Modelling of Laterally Loaded Piles for Offshore Wind
Turbines”. DPhil thesis, University of Oxford.
Beuckelaers, W. J. A. P., Houlsby, G. T., and Burd, H. J. (2018). “A comparison of the series and
parallel Masing-Iwan model in 2D”. Proceedings of the 9th European Conference on
Numerical Methods in Geotechnical Engineering (NUMGE). Porto, Portugal.
Bhattacharya, S. (2014). “Challenges in Design of Foundations for Offshore Wind Turbines”.
Engineering & Technology Reference, IET, pp. 1–9. DOI: 10.1049/etr.2014.0041.
Bierbooms, W. (2003). “Wind and wave conditions, DOWEC (Dutch Offshore Wind Energy
Converter Project)”. Technical Report, Delft University of Technology.
Blaker, Ø., Lunne, T., Vestgården, T., Krogh, L., Thomsen, N. V., Powell, J . J. M., and Wallace, C. F.
(2015). “Method dependency for determining maximum and minimum dry unit weights of
sands”. Proceedings of the Third International Symposium on Frontiers in Offshore
Geotechnics (ISFOG). Oslo, Norway, pp. 1159–1166.
Bolton, M. D. (1986). “The strength and dilatancy of sands”. Géotechnique 36.1, pp. 65–78. DOI:
10.1680/geot.1986.36.1.65.
Bolton, M. D. (1987). “Reply to discussion on: The strength and dilatancy of sands, Bolton
(1986)”. Géotechnique 37.2, pp. 219 –226. DOI: 10.1680/geot.1987.37.2.219.
Bredmose, H., Dixen, M., Ghadirian, A., Larsen, T. J., Schløer, S., Andersen, S. J., Wang, S.,
Bingham, H. B., Lindberg, O., Christensen, E. D., Vested, M. H., Carstensen, S.,
Engsig-Karup, A. P., Petersen, O. S., Hansen, H. F., Mariegaard, J. S., Taylor, P. H.,
Adcock, T. A. A., Obhrai, C., Gudmestad, O. T., Tarp-Johansen, N. J., Meyer, C. P.,
References 218
Krokstad, J. R., Suja-Thauvin, L., and Hanson, T. D. (2016). “DeRisk - Accurate Prediction of
ULS Wave Loads. Outlook and First Results”. Energy Procedia 94, pp. 379–387. DOI:
10.1016/j.egypro.2016.09.197.
British Standard (1990). Soils for civil engineering purposes BS 1377-4 : 1990.
British Standard (1998). Testing Concrete BS 1881 : Part 131 : 1998.
Brodersen, M. L., Bjørke, A., and Høgsberg, J. (2017). “Active tuned mass damper for damping of
offshore wind turbine vibrations”. Wind Energy 20, pp. 783–796. DOI: 10.1002/we.2063.
Broms, B. B. (1964). “Lateral Resistance of Piles in Cohesionless Soils”. ASCE Journal of Soil
Mechanics and Foundation Division 90, pp. 123–156.
Burd, H. J., Taborda, D. M. G., Zdravković, L., Abadie, C. N., Byrne, B. W., Houlsby, G. T.,
Gavin, K., Igoe, D., Jardine, R. J., Martin, C. M., McAdam, R. A., Pedro, A. M. G., and
Potts, D. M. (2019). “PISA Design Model for Monopiles for Offshore Wind Turbines:
Application to a Marine Sand”. Submitted.
Byrne, B. W. (2000). “Investigations of suction caissons in dense sand”. DPhil thesis, University
of Oxford.
Byrne, B. W. (2014). “Laboratory scale modelling for offshore geotechnical problems”.
Proceedings of the 8th International Conference on Physical Modelling in Geotechnics
(ICPMG). Perth, Australia, pp. 61–74.
Byrne, B. W. and Houlsby, G. T. (2005). “Investigating 6 degree-of-freedom loading on shallow
foundations”. Proceedings of the International Symposium on Frontiers in Offshore
Geotechnics (ISFOG). Perth, Australia, pp. 477–482.
Byrne, B. W. and Houlsby, G. T. (2010). “Development of a Multi-Axis Loading System for the
Testing of Shallow Foundations”. Unpublished report.
Byrne, B. W., McAdam, R., Burd, H. J., Houlsby, G. T., Martin, C. M., Zdravkovic, L.,
Taborda, D. M. G., Potts, D. M., Jardine, R. J., Sideri, M., Schroeder, F. C., Gavin, K.,
Doherty, P., Igoe, D., Muir Wood, A., Kallehave, D., and Skov Gretlund, J. (2015). “New design
methods for large diameter piles under lateral loading for offshore wind applications”.
Proceedings of the 3rd International Symposium on Frontiers in Offshore Geotechnics
(ISFOG). Oslo, Norway, pp. 705–710.
Byrne, B. W., McAdam, R. A., Burd, H. J., Houlsby, G. T., Martin, C. M., Beuckelaers, W. J. A. P.,
Zdravković, L., Taborda, D. M. G., Potts, D. M., Jardine, R. J., Ushev, E., Liu, T., Abadias, D.,
Gavin, K., Igoe, D., Doherty, P., Skov Gretlund, J., Pache, S., and Plummer, M. A. L. (2017).
“PISA: New design method for offshore wind turbine monopiles”. Proceedings of the 8th
International Conference on Offshore Site Investigation and Geotechnics (SUT OSIG).
London, UK, pp. 142–161.
Cannon, M (2012). Introduction to Control Theory. Lecture Notes for A2 Engineering Science
course, Oxford University.
Carbon Trust (2014). New project underway using vibration to install monopiles to reduce costs
of offshore wind energy. URL: https://www.carbontrust.com/news/2014/04/new-project-
underway-vibration-to-install-monopiles-reduce-costs-offshore-wind-energy/.
Carbon Trust (2019). Blue Pilot. URL:
https://www.carbontrust.com/offshore-wind/owa/demonstration/blue-pilot/.
Chakraborty, T. and Salgado, R. (2010). “Dilatancy and Shear Strength of Sand at Low Confining
Pressures”. Journal of Geotechnical and Geoenvironmental Engineering 136.1, pp. 527–532.
DOI : 10.1061/(ASCE)GT.1943-5606.0000237.
References 219
Chow, S. H., Roy, A., Herduin, M., Heins, E., Bienen, B., O’Loughlin, C. D., Gaudin, C., and
Cassidy, M. J. (2018). “Characterisation of UWA superfine silica sand”. Internal report, The
University of Western Australia.
Cox, W. R., Reese, L. C., and Grubbs, B. R. (1974). “Field Testing of Laterally Loaded Piles In
Sand”. Proceedings of the Offshore Technology Conference (OTC). Houston, Texas,
pp. 459–464. DOI: 10.4043/2079-MS.
Cuéllar, P., Georgi, S., Baeßler, M., and Rücker, W. (2012). “On the quasi-static granular
convective flow and sand densification around pile foundations under cyclic lateral
loading”. Granular Matter 14.1, pp. 11–25. DOI: 10.1007/s10035-011-0305-0.
Cui, L. and Bhattacharya, S. (2016). “Soil-monopile interactions for offshore wind turbines”.
Proceedings of the Institution of Civil Engineers - Engineering and Computational Mechanics
169.4, pp. 171–182. DOI: 10.1680/jencm.16.00006.
Dewoolkar, M. M., Pak, R. Y. S., and Ko, H.-Y. (1999). “Centrifuge modelling of models of seismic
effects on saturated earth structures”. Géotechnique 49.2, pp. 247–266. DOI:
10.1680/geot.1999.49.2.247.
DNV GL (2016). Support structures for wind turbines DNVGL-ST-0126.
Doherty, P. and Gavin, K (2012). “Laterally loaded monopile design for offshore wind farms”.
Proceedings of the Institution of Civil Engineers - Energy 165.1, pp. 7–17. DOI:
10.1680/ener.11.00003.
Duan, N., Xu, X., and Cheng, Y. P. (2017). “Distinct-element analysis of an offshore wind turbine
monopile under cyclic lateral load”. Geotechnical Engineering 170.6, pp. 517–533. DOI:
10.1680/jgeen.16.00171.
Dührkop, J. and Grabe, J. (2008). “Monopilegründungen von Offshore-windenergieanlagen -
Zum Einfluss Einer Veränderlichen Zyklischen Lastangriffsrichtung”. Bautechnik 85.5,
pp. 317–321. DOI: 10.1002/bate.200810024.
EWEA (2009). The Economics of Wind Energy. Ed. by S. Krohn.
EWEA (2015). “EWEA report on COP21 outcome”. URL:
https://windeurope.org/wp-content/uploads/files/policy/topics/climate-
change/EWEA-report-on-Paris-Agreement.pdf.
Garnier, J. (2013). “Advances in lateral cyclic pile design : Contribution of the SOLCYP project”.
Proceedings of the 18th International Conference on Soil Mechanics and Geotechnical
Engineering (TC 209 Workshop, Design for cyclic loading: Piles and other foundations). Paris,
France, pp. 59–68.
Gaudin, C., Loughlin, C. D. O., and Breen, J. (2018). “A new 240 g-tonne geotechnical centrifuge
at the University of Western Australia”. Proceedings of the 9th International Conference on
Physical Modelling in Geotechnics (ICPMG). London, UK, pp. 501–506.
GE Renewable Energy (2019). Haliade-X Offshore Wind Turbine Platform. URL: https:
//www.ge.com/renewableenergy/wind-energy/turbines/haliade-x-offshore-turbine.
Geotechdata (2013). Soil elastic Young’s modulus. URL:
http://www.geotechdata.info/parameter/soil-young’s-modulus.html.
Green, E. I. (1955). “The story of Q”. American Scientist 43, pp. 584–594.
Gui, M. W., Bolton, M. D., Garnier, J., Corte, J. F., Bagge, G., Laue, J., and Renzi, R. (1998).
“Guidelines for cone penetration tests in sand”. Proceedings of the International Conference
on Centrifuge Modelling (Centrifuge 98). Rotterdam, The Netherlands, pp. 155–160.
References 220
Hardin, B. O. (1965). “Dynamic versus static shear modulus for dry sand”. Materials Research
and Standards 5.5, pp. 232–235.
Herduin, M. (2019). “Multi-directional loading on shared anchors for offshore renewable
energy: Definition and preliminary investigation into soil behaviour and anchor
performance”. PhD Thesis, The University of Western Australia. DOI:
10.26182/5d2fd93d88ce7.
Houlsby, G. T. (1981). “A study of plasticity theories and their applicability to soils”. PhD Thesis,
University of Cambridge.
Houlsby, G. T. and Puzrin, A. M. (2006). Principles of Hyperplasticity: An Approach to Plasticity
Theory Based on Thermodynamic Principles. Springer-Verlag.
Houlsby, G. T., Abadie, C. N., Beuckelaers, W. J. A. P., and Byrne, B. W. (2017). “A model for
nonlinear hysteretic and ratcheting behaviour”. International Journal of Solids and
Structures 120, pp. 67 –80. DOI: 10.1016/j.ijsolstr.2017.04.031.
IEA (2017). World Energy Outlook 2017. URL: https://www.iea.org/weo2017.
Inman, D. J. (2014). Engineering vibration. 4th ed. Pearson. Chap. 2.
IRENA (2017). Electricity storage and renewables: costs and markets to 2030. URL:
https://www.irena.org/publications/2017/Oct/Electricity-storage-and-
renewables-costs-and-markets.
Iwan, W. D. (1967). “On a Class of Models for the Yielding Behavior of Continuous and
Composite Systems”. Journal of Applied Mechanics 34.3. DOI: 10.1115/1.3607751.
Jeanjean, P., Zhang, Y., Zakeri, A., Andersen, K. H., Gilbert, K., and Senanayake, A. I. M. J. (2017).
“A framework for monotonic p-y curves in clays”. Proceedings of the 8th International
Conference on Offshore Site Investigation and Geotechnics (SUT OSIG). London, UK,
pp. 108–141.
Kelly, R. B., Byrne, B. W., and Houlsby, G. T. (2006). “A comparison of field and laboratory tests of
caisson foundations in sand and clay”. Géotechnique 56.9, pp. 617–626. DOI:
10.1680/geot.2006.56.9.617.
Kementzetzidis, E., Corciulo, S., Versteijlen, W. G., and Pisanò, F. (2019). “Geotechnical aspects
of offshore wind turbine dynamics from 3D non-linear soil-structure simulations”. Soil
Dynamics and Earthquake Engineering 120, pp. 181–199. DOI:
10.1016/j.soildyn.2019.01.037.
Klinkvort, R. and Hededal, O. (2013). “Lateral response of monopile supporting an offshore
wind turbine”. Proceedings of the Institution of Civil Engineers - Geotechnical Engineering
166.2, pp. 147–158. DOI: 10.1680/geng.12.00033.
Klinkvort, R., Springman, S., and Hededal, O. (2013). “Scaling issues in centrifuge modelling of
monopiles”. International Journal of Physical Modelling in Geotechnics 13.2, pp. 38–50. DOI:
10.1680/ijpmg.12.00010.
Klinkvort, R. T. (2012). “Centrifuge modelling of drained lateral pile-soil response”. PhD Thesis,
DTU.
Koukoura, C. (2014). “Validated Loads Prediction Models for Offshore Wind Turbines for
Enhanced Component Reliability”. PhD Thesis, DTU.
Kramer, S. L. (1996). Geotechnical Earthquake Engineering. Prentice Hall.
Kucuk, S. and Bingul, Z. (2006). “Robot Kinematics: Forward and Inverse Kinematic”. Industrial
Robotics: Theory, Modelling and Control. Ed. by S. Cubero. IntechOpen. DOI: 10.5772/44.
References 221
Lambe, T. W. (1973). “Predictions in soil engineering”. Géotechnique 23.2, pp. 149–202. DOI:
10.1680/geot.1973.23.2.151.
Lazarevic, Z. (1997). “Feasibility of a Stewart Platform with Fixed Actuators as a Platform for
CABG Surgery Device”. Master’s Thesis, Columbia University.
Leblanc, C., Houlsby, G. T., and Byrne, B. W. (2010a). “Response of stiff piles in sand to long-term
cyclic lateral loading”. Géotechnique 60.2, pp. 79–90. DOI: 10.1680/geot.7.00196.
Leblanc, C., Byrne, B. W., and Houlsby, G. T. (2010b). “Response of stiff piles to random two-way
lateral loading”. Géotechnique 60.9, pp. 715–721. DOI: 10.1680/geot.09.T.011.
Li, S., Zhang, Y., and Jostad, H. P. (2019). “Drainage conditions around monopiles in sand”.
Applied Ocean Research 86, pp. 111–116. DOI: 10.1016/j.apor.2019.01.024.
Li, Z., Haigh, S. K., and Bolton, M. D. (2010). “Centrifuge modelling of mono-pile under cyclic
lateral loads”. Proceedings of the 7th International Conference on Physical Modelling in
Geotechnics (ICPMG). Zurich, Switzerland, pp. 965–970. DOI:
10.1680/ijpmg.2010.10.2.47.
Little, R. L. and Briaud, J.-L. (1988). “Full scale cyclic lateral load tests on six single piles in
sand”. Miscellaneous Paper GL-88-27, Geotechnical Div., Texas A&M University.
Long, J. H. and Vanneste, G. (1994). “Effects of Cyclic Lateral Loads on Piles in Sand”. Journal of
Geotechnical Engineering 120.1, pp. 225–244. DOI:
10.1061/(ASCE)0733-9410(1994)120:1(225).
Løvholt, F., Madshus, C., and Andersen, K. H. (2020). “Intrinsic Soil Damping from Cyclic
Laboratory Tests with Average Strain Development”. Geotechnical Testing Journal 43.1. DOI:
10.1520/GTJ20170411.
Lunne, T. (2012). “The Fourth James K. Mitchell Lecture: The CPT in offshore soil investigations
- a historic perspective”. Geomechanics and Geoengineering 7.2, pp. 75–101. DOI:
10.1080/17486025.2011.640712.
Manceau, S., McLean, R., Sia, A., and Soares, M. (2019). “Application of the Findings of the PISA
Joint Industry Project in the Design of Monopile Foundations for a North Sea Wind Farm”.
Proceedings of the Offshore Technology Conference (OTC). Houston, Texas.
Masing, G. (1926). “Eiganspannungen und verfestigung beim messing”. Proceedings of the 2nd
International Congress of Applied Mechanics.
Mayall, R. O. (2019). “Monopile response to scour and scour protection”. DEng thesis,
University of Oxford.
Mayall, R. O., McAdam, R. A., Whitehouse, R. J. S., Burd, H. J., Byrne, B. W., Heald, S. G.,
Sheil, B. B., and Slater, P. L. (2019). “Flume tank testing of offshore wind turbine dynamics
with foundation scour and scour protection”. Submitted.
Mrŏz, Z, Norris, V. A., and Zienkiewicz, O. C. (1978). “An isotropic hardening model for soils and
its application to cyclic loading”. International Journal for Numerical and Analytical
Methods in Geomechanics 2, pp. 203–221.
Murchison, J. M. and O’Neill, M. W. (1984). “Evaluation of p-y relations in cohesionless soils”.
Proceedings of the ASCE symposium on analysis and design of pile foundations. San
Francisco, USA, pp. 174 –191.
Nguyen-Sy, L. and Houlsby, G. T. (2005). “The theoretical modelling of a suction caisson
foundation using hyperplasticity theory”. Proceedings of the International Symposium on
Frontiers in Offshore Geotechnics (ISFOG). Perth, Australia, pp. 417–423.
References 222
Nicolai, G. (2017). “Cyclic behaviour of laterally loaded monopiles in sand supporting offshore
wind turbines”. PhD Thesis, Aalbog University.
Nicolai, G. and Ibsen, L. B. (2014). “Small-Scale Testing of Cyclic Laterally Loaded Monopiles in
Dense Saturated Sand”. Journal of Ocean and Wind Energy 1.4, pp. 240–245.
Nicolai, G., O’Loughlin, C. D., White, D. J., Cassidy, M. J., and Ibsen, L. B. (2017a). “Centrifuge
study with PIV analysis of monopiles in dense sand under cyclic lateral loading”. DCE
Technical Report No. 206, Deprtment of Civil Engineering, Aalborg University.
Nicolai, G., Ibsen, L. B., O’Loughlin, C. D., and White, D. J. (2017b). “Quantifying the increase in
lateral capacity of monopiles in sand due to cyclic loading”. Géotechnique Letters 7.3,
pp. 1–8. DOI: 10.1680/jgele.16.00187.
Niemann, C., Reul, O., Tian, Y., O’Loughlin, C. D., and Cassidy, M. J. (2018). “Centrifuge tests on
the response of piles under cyclic lateral 1-way and 2-way loading”. Proceedings of the 9th
International Conference on Physical Modelling in Geotechnics (ICPMG). London, UK,
pp. 731–736.
Niemunis, A. and Herle, I. (1997). “Hypoplastic model for cohesionless soils with elastic strain
range”. Mechanics of Cohesive-Frictional Materials 2, pp. 279–299.
Niemunis, A., Wichtmann, T., and Triantafyllidis, Th. (2005). “A high-cycle accumulation model
for sand”. Computers and Geotechnics 32.4, pp. 245–263. DOI:
10.1016/j.compgeo.2005.03.002.
Osman, A. S. and Randolph, M. F. (2012). “Analytical Solution for the Consolidation around a
Laterally Loaded Pile”. International Journal of Geomechanics 12.3, pp. 199–208. DOI:
10.1061/(ASCE)GM.1943-5622.0000123.
Ovesen, N. K. (1975). “Centrifugal testing applied to bearing capacity problems of footings on
sand”. Géotechnique 25.2, pp. 394–401. DOI: 10.1680/geot.1975.25.2.394.
Oztoprak, S. and Bolton, M. D. (2013). “Stiffness of sands through a laboratory test database”.
Géotechnique 63.1, pp. 54–70. DOI: 10.1680/geot.10.P.078.
Page, A. M., Grimstad, G., Eiksund, G. R., and Jostad, H. P. (2018). “A macro-element pile
foundation model for integrated analyses of monopile-based offshore wind turbines”.
Ocean Engineering 167, pp. 23–35. DOI: 10.1016/j.oceaneng.2018.08.019.
Page, A. M., Næss, V., De Vaal, J. B., Eiksund, G. R., and Nygaard, T. A. (2019). “Impact of
foundation modelling in offshore wind turbines: comparison between simulations and field
data”. Marine Structures 64, pp. 379–400. DOI: 10.1016/j.marstruc.2018.11.010.
Peralta, P (2010). “Investigations on the Behaviour of Large Diameter Piles under Long-Term
Lateral Cyclic Loading in Cohesionless Soil”. PhD thesis, Leibniz Universitat Hannover.
Peralta, P., Ballard, J. C., Rattley, M., and Erbrich, C. E. (2017). “Dynamic and cyclic pile-soil
response curves for monopile design”. Proceedings of the 8th International Conference on
Offshore Site Investigation and Geotechnics (SUT OSIG). London, UK, pp. 1054–1061.
Poulos, H. G. and Hull, T. S. (1989). “The role of analytical geomechanics in foundation
engineering”. Proceedings of the ASCE Foundation Engineering Congress (Foundation
Engineering: Current Principals and Practices). Evanston, US.
Prévost, J. (1977). “Mathematical modelling of monotonic and cyclic undrained clay
behaviour”. International Journal for Numerical and Analytical Methods in Geomechanics 1,
pp. 195–216. DOI: 10.1002/nag.1610010206.
Puzrin, A. M. and Shiran, A. (2000). “Effects of the constitutive relationship on seismic response
of soils . Part I . Constitutive modeling of cyclic behavior of soils”. Soil Dynamics and
Earthquake Engineering 19.5, pp. 305–318. DOI: 10.1016/S0267-7261(00)00027-0.
References 223
Pyke, R. (1979). “Nonlinear soil models for irregular cyclic loadings”. Journal of the Geotechnical
Division, Proceedings of the ASCE 105.6, pp. 715–726.
Rad, N. S. and Tumay, M. T. (1987). “Factors Affecting Sand Specimen Preparation By Raining”.
Geotechnical Testing Journal 10.1, pp. 31–37. DOI: 10.1520/GTJ10136J.
Ramberg, W. and Osgood, W. R. (1943). “Description of stress-strain curves by three parameters
(Technical Note 902)”. National Advisory Committee For Aeronautics, Washington, US.
Randolph, M. F. (2018). “Potential Damage to Steel Pipe Piles During Installation”. IPA News
Letter 3.1, pp. 3–10.
Reese, L. C. and Van Impe, W. (2011). Single piles and pile groups under lateral loading. 2nd ed.
CRC Press/Balkema.
Reese, L. C., Cox, W. R., and Koop, F. D. (1974). “Analysis of Laterally Loaded Piles in Sand”.
Proceedings of the Offshore Technology Conference (OTC). Houston, Texas, pp. 473–483. DOI:
10.4043/2080-MS.
Renewable UK (2019). Wind Energy. URL: https://www.renewableuk.com/page/WindEnergy.
Roy, A., Chow, S., O’Loughlin, C. D., and Randolph, M. F. (2019). “Effect of stress history and
shallow embedment on centrifuge cone penetration tests in sand”. Proceedings of the 39th
International Conference on Ocean, Offshore and Arctic Engineering (OMAE). Glasgow,
Scotland.
Rudolph, C. and Grabe, J. (2013). “Untersuchungen zu zyklisch horizontal belasteten Pfählen
bei veränderlicher Lastrichtung”. Geotechnik 36.2, pp. 90–95. DOI:
10.1002/gete.201200025.
Rudolph, C., Grabe, J., and Bienen, B. (2014a). “Drift of piles subjected to cyclic lateral loading
from a varying direction: system vs. soil element behaviour”. Proceedings of 33rd
International Conference on Ocean, Offshore and Arctic Engineering (OMAE). San Francisco,
USA.
Rudolph, C., Bienen, B., and Grabe, J. (2014b). “Effect of variation of the loading direction on
the displacement accumulation of large-diameter piles under cyclic lateral loading in sand”.
Canadian Geotechnical Journal 51.10, pp. 1196–1206. DOI: 10.1139/cgj-2013-0438.
Schnaid, F. (1990). “A study of the cone-pressuremeter test in sand”. DPhil thesis, University of
Oxford.
Schofield, A. N. (1980). “Cambridge Geotechnical Centrifuge Operations”. Géotechnique 30.3,
pp. 227–268. DOI: 10.1680/geot.1980.30.3.227.
Schroeder, F. C., Merritt, A. S., Sørensen, K. W., Muir Wood, A., Thilsted, C. L., and Potts, D. M.
(2015). “Predicting monopile behaviour for the Gode Wind offshore wind farm”.
Proceedings of the 3rd International Symposium on Frontiers in Offshore Geotechnics
(ISFOG). Oslo, Norway, pp. 735–740.
Skau, K. S., Grimstad, G., Page, A. M., Eiksund, G. R., and Jostad, H. P. (2018). “A macro-element
for integrated time domain analyses representing bucket foundations for offshore wind
turbines”. Marine Structures 59, pp. 158–178. DOI: 10.1016/j.marstruc.2018.01.011.
Sørensen, S. P. H., Augustesen, A. H., Leth, C. T., Østergaard, M. U., and Møller, M. (2017).
“Consequences of p-y curve selection for monopile design for offshore wind turbines”.
Proceedings of the 8th International Conference on Offshore Site Investigation and
Geotechnics (SUT OSIG). London, UK, pp. 1062–1069.
Su, D., Wu, W. L., Du, Z. Y., and Yan, W. M. (2014). “Cyclic Degradation of a Multidirectionally
Laterally Loaded Rigid Single Pile Model in Compacted Clay”. Journal of Geotechnical and
Geoenvironmental Engineering 140.5. DOI: 10.1061/(ASCE)GT.1943-5606.0001084.
References 224
Taborda, D. M. G., Potts, D. M., and Zdravković, L. (2016). “On the assessment of energy
dissipated through hysteresis in finite element analysis”. Computers and Geotechnics 71,
pp. 180–194. DOI: 10.1016/j.compgeo.2015.09.001.
Tatsuoka, F. (1987). “Discussion on: The strength and dilatancy of sands, Bolton (1986)”.
Géotechnique 37.2, pp. 219–226. DOI: 10.1680/geot.1987.37.2.219.
Truong, P., Lehane, B. M., Zania, V., and Klinkvort, R. T. (2019). “Empirical approach based on
centrifuge testing for cyclic deformations of laterally loaded piles in sand”. Géotechnique
69.2, pp. 133–145. DOI: 10.1680/jgeot.17.p.203.
Uesugi, M. and Kishida, H. (1986). “Frictional resistance at yield between dry sand and mild
steel”. Soils and foundations 26.4, pp. 139 –149. DOI: 10.3208/sandf1972.26.4_139.
United Nations (2015). “Paris Agreement”. Framework Convention on Climate Change.
University of Strathclyde (2015). XL Monopiles. URL:
http://www.esru.strath.ac.uk/EandE/Web{\_}sites/14-
15/XL{\_}Monopiles/technical.html.
Van Vledder, G. Ph. (2013). “On Wind-Wave Misalignment, Directional Spreading and Wave
Loads”. Proceedings of the 32rd International Conference on Ocean, Offshore and Arctic
Engineering (OMAE). Nantes, France. DOI: 10.1115/OMAE2013-11393.
Villalobos, F. (2006). “Model Testing of Foundations for Offshore Wind Turbines”. DPhil thesis,
University of Oxford.
Wang, S. (2018). “Assessment of offshore wind turbines in extreme weather conditions”. PhD
Thesis, DTU Wind Energy.
White, J. R. F. (2020). “A laboratory investigation into the behaviour of sand at low confining
stresses”. Forthcoming DPhil thesis, University of Oxford.
Wichtmann, T., Triantafyllidis, T., Chrisopoulos, S., and Zachert, H. (2017). “Prediction of
Long-Term Deformations of Offshore Wind Power Plant Foundations Using
Engineer-Oriented Models Based on the High Cycle Accumulation Model”. International
Journal of Offshore and Polar Engineering 27.4, pp. 346–356. DOI:
10.17736/ijope.2017.fv05.
Wind Europe (2017). “Wind energy in Europe: Scenarios for 2030”. URL:
https://windeurope.org/wp-content/uploads/files/about-wind/reports/Wind-
energy-in-Europe-Scenarios-for-2030.pdf.
Wind Europe (2018). “Offshore Wind in Europe: Key trends and statistics 2017”. URL:
https://windeurope.org/wp-content/uploads/files/about-
wind/statistics/WindEurope-Annual-Offshore-Statistics-2017.pdf.
Wind Europe (2019). “Offshore Wind in Europe: Key trends and statistics 2018”. URL:
https://windeurope.org/about-wind/statistics/offshore/european-offshore-wind-
industry-key-trends-statistics-2018/.
Wind Turbine Models (2015). “Vestas V164-8.0”. URL:
https://en.wind-turbine-models.com/turbines/318-vestas-v164-8.0.
Windindustrie in Deutschland (2016). 1300 tonnes – world´s largest monopile moved on
self-propelled transporters. URL:
https://www.windindustrie-in-deutschland.de/fachartikel/1300-tonnes-worlds-
largest-monopole-moved-on-self-propelled-transporters/trackback/.
Windpower (2019). Vattenfall proves offshore wind can be profitable without subsidies. URL:
https://www.windpowerengineering.com/business-news-projects/vattenfall-
proves-offshore-wind-can-be-profitable-without-subsidies/.
References 225
Zhang, Y., Andersen, K. H., Klinkvort, R. T., Jostad, H. P., Sivasithamparam, N., Boylan, N. P., and
Langford, T. (2016). “Monotonic and Cyclic p-y Curves for Clay Based on Soil Performance
Observed in Laboratory Element Tests”. Proceedings of the Offshore Technology Conference
(OTC). Houston, Texas. DOI: 10.4043/26942-MS.
Zhao, Y., Gafar, K., Elshafie, M. Z. E. B., Deeks, A. D., Knappett, J. A., and Madabhushi, S. P. G.
(2006). “Calibration and use of a new automatic sand pourer”. Proceedings of the 6th
International Conference on Physical Modelling in Geotechnics (ICPMG). Hong Kong,
pp. 265 –270.
Zhu, F. Y., O’Loughlin, C. D., Bienen, B., Cassidy, M. J., and Morgan, N. (2017). “The response of
suction caissons to long-term lateral cyclic loading in single-layer and layered seabeds”.
Géotechnique 67.11, pp. 1–13. DOI: 10.1680/jgeot.17.P.129.
Ziegler, H. (1977). An Introduction to Thermomechanics. 2nd ed. North-Holland, Amsterdam.

Monopile Foundations Under Complex Cyclic Lateral Loading

Uploaded by

Copyright:

Available Formats

Monopile Foundations Under Complex Cyclic Lateral Loading

Uploaded by

Document Information

Original Description:

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

Monopile Foundations Under Complex Cyclic Lateral Loading

Uploaded by

Copyright:

Available Formats

What is the thesis exploring?

What is the thesis exploring?

What types of cyclic loading tests were conducted?

What types of cyclic loading tests were conducted?

Monopile foundations under complex

cyclic lateral loading

A thesis submitted for the degree of

The vast majority of offshore wind turbines are supported on monopile

2 Design of novel laboratory apparatus for 1g monopile testing 29

2.4.4 Adjustment for transducer friction . . . . . . . . . . . . . . . . . . 40

3 1g test procedure, monotonic responses and cyclic definitions 46

4 Regular cyclic loading response at 1g 69

4.6.2 Fan-type cyclic loading . . . . . . . . . . . . . . . . . . . . . . . . . 101

5 Application of realistic storm loading at 1g 113

6 Investigating the effect of stress-level 130

6.6.2 Secant stiffness response . . . . . . . . . . . . . . . . . . . . . . . . 155

7 Modelling the response of a monopile to complex cyclic loading 162

Greek alphabet variables

Empirical value in dilatancy expressions (Bolton, 1986; Bolton, 1987)

Latin alphabet variables

an Point on cyclic response at mean stress on loading (cyclic definitions)

Ei Initial stiffness (monotonic response)

ms Parameter for evolution of R (HARM evolution functions)

Ti Coordinates of pivot point on transducer (kinematics)

Common subscripts, superscripts, prefixes and diacritics

Test name abbreviations

L Very loose sand (OU)

Other common abbreviations

CPT Cone Penetration Test

HARM Hyperplastic Accelerated Ratcheting Model

Richards, I. A., Byrne, B. W. and Houlsby G. T. (2018). “Physical modelling of monopile

Richards, I. A., Houlsby G. T. and Byrne, B. W. (2019). “Exploring the response of a

Richards, I. A., Byrne, B. W. and Houlsby G. T. (2019). “Monopile rotation under

of the century. An urgent, global transition from fossil-fuels to renewable energy is

therefore necessary (EWEA, 2015).

further offshore, typically in deeper water (Wind Europe, 2018). Technological

advancements, including in foundation design, have also contributed to

“zero-subsidy” (Windpower, 2019).

1.1.2 Offshore wind turbine support structures

Figure 1.3: Illustration of possible support structures for OWTs

The vast majority of OWTs are supported on monopiles — large-diameter open-

The current prevalence of monopiles is explained by their simple design, established

The dominance of the monopile is expected to continue, although the market-share

of alternative grounded support structures may increase as offshore wind expands

Meanwhile, floating OWTs present an opportunity for development of offshore wind

in deeper water (> 50 m) (IRENA, 2017).

Figure 1.4 illustrates a monopile foundation and highlights key dimensions:

(Schroeder et al., 2015). The next generation of monopiles might be 10 m in diameter.

water-level, while scour protection is typically installed to minimise erosion at the

seabed. Figure 1.5 shows a fabricated monopile before installation.

1.2 Design of monopile foundations

result in combined moment (M ) and horizontal loading (H) on the monopile

Figure 1.5: Fabricated monopile for Veja Mate offshore

Figure 1.4: Schematic of an OWT

Figure 1.6: Schematic indicating lateral loading on an OWT on a monopile foundation

Limit state Moment amplitude /MU LS No. of cycles N

Ultimate Limit State (ULS) 1.00 1

complex: continuously varying in amplitude, direction, frequency and eccentricity

An example distribution of approximate cyclic moment amplitudes experienced