PISA I - Final Report

This report is Confidential to the PISA Project Industry Partners
PISA
PROJECT
PISA Final Report

October 2016
Report produced by members of the PISA Academic Work Group (University of Oxford, Imperial
College London, University College Dublin) and the PISA representatives of DONG Energy.
Revision History
Rev Date Ver No Doc No Description
A 03/02/2016 2450183A 2450183 Draft for comment by DONG Energy and PISA Project Partners.
B 13/05/2016 2450183B 2450193 Final report first revision.
C 20/07/2016 2450183C 2509550 Minor revision to the front page, header, footer plus an acknowledgement of
the funding provided by the Carbon Trust on Page 4.
D 14/10/2016 2450183D 2595138 Correction of typographical errors, including in sand soil reaction parameter
table, updated figures and addition of analysis of a water filled gap for clay.
The sole responsibility for the content of this report lies with the PISA Project Academic Work Group. This
comprised of Byron Byrne, Oxford University, as the Principal Investigator and then in no particular order (a)
Oxford University: Harvey Burd, Guy Houlsby, Chris Martin, Ross McAdam (b) Imperial College: Lidija
Zdravkovic, David Taborda, David Potts, Richard Jardine, (c) University College Dublin: Ken Gavin, David
Igoe.
DONG Energy Document Information:

Doc. no. 2595138
Ver. no. 2450183D
Case no. 030-16-0023
The PISA Project is a joint industry research project supported by:
(Lead Project Partner)
© Copyright PISA Project Academic Work Group: This report is provided to the industry partners of the PISA Project and remains confidential to
them. The contents may be used internally by the partners but not provided to any third parties nor be used for unauthorised purposes.
Page 2/272
Executive summary
This report presents a new design methodology for laterally loaded piles for offshore wind turbines.
The work has been developed by an Academic Work Group (AWG) during the PISA project, and has
been described, as it has developed, through a series of interim reports, reviewed by the project
steering group as well as an independent review committee. This final report covers all work to date
and includes developments based on comments received during the various reviewing stages as
well as from workshops conducted during the project.
To provide the basis for the new design method, the AWG completed a series of state-of-the-art
finite element analyses of laterally loaded piles under a loading similar to that applied to typical
offshore wind foundations. These analyses have been performed for both clay and sand soils,
considered to be representative materials for a number of North Sea wind farm sites. The results
indicate that current design methods consistently under-predict both the ultimate capacity and initial
stiffness for the examined clay till. For sand soil the results indicate that existing methods are not
able to predict accurately the initial stiffness.
Using the computational results a new design method has been derived, drawing on the existing p-y
approach, but with updated soil response curves, that are a significant departure from the existing p-
y curves, and additional components of resistance that have been found to be important. This
process has demonstrated that for long thin piles, a (modified) p-y approach may well be suitable for
design, however for large diameter relatively short piles, typical of the offshore monopile, the p-y
approach alone will never be able to successfully predict the response. Three additional components
to account for the shear and moment at the pile base, and the distributed moment along the pile
have been identified. If these are used to extend the conventional p-y approach, the pile response
can be more accurately computed.
A campaign of medium scale field trials was carried out to test the new design methods. The tests
were carefully planned, based on numerical analyses calibrated to the site soil profiles, and covering
a range of loading conditions appropriate to offshore wind turbines. The field tests were carefully
supervised during execution with the close attention to detail leading to very high quality results. The
field tests demonstrate a close match with the finite element calculations, providing high confidence
that the approach can be used to predict the performance of full scale large diameter monopiles. The
field tests demonstrate that significant savings can be realised through the use of site specific design
calculations.
The AWG propose a design procedure based on a 1D finite element model of the monopile. This 1D
model is, essentially, an extension of the p-y approach in which additional features are included
representing a distributed moment along the pile and a horizontal force and moment at the pile base.
Two alternative approaches are proposed for determining the various soil reaction curves that are
required in the model. The first approach is to apply a ‘rule based method’ that will use lookup tables
to determine the appropriate parameters for input into the model, determined by basic strength and
stiffness parameters. The second, more detailed approach, involves the application of bespoke finite
element modelling to determine specific soil response functions for the particular site of interest. It
should be noted that the soil reaction curves provided in this report only apply to the two sites and
the pile configurations studied during the project, but a database of tables will be accumulated as
additional finite element designs are carried out. This approach offers a significant step forward in
the optimisation of design.
Page 3/272
Table of Contents
Nomenclature ....................................................................................................................................... 5
1 Introduction and project overview .......................................................................................... 7
2 Overview of site selection and characterisation .................................................................. 16
3 Numerical 3D finite element modelling ................................................................................. 24
4 Field testing ............................................................................................................................. 28
5 1D ‘soil reaction curve’ analysis procedure ......................................................................... 39
6 Achievements of the Project and Conclusions .................................................................... 62
Appendix A Site investigation, laboratory testing and derivation of soil parameters for
numerical modelling ............................................................................................................... 64
Appendix B Analysis of monotonic field test measurements ...................................................... 92
Appendix C Development and validation of 3D finite element models for pile behaviour
prediction ............................................................................................................................... 119
Appendix D Detailed design method application to a stiff over-consolidated clay site .......... 141
Appendix E Detailed design method application to a dense sand site ..................................... 187
Appendix F Analysis tools.............................................................................................................. 222
Appendix G Additional areas of interest and future work ........................................................... 239
References ....................................................................................................................................... 270
Funding Acknowledgement: The PISA project was funded by DECC and the PISA Industry Partners
under the umbrella of the Offshore Wind Accelerator (OWA) programme. OWA is a world leading
industry‐led collaborative programme designed and managed by the Carbon Trust, between DECC,
the Scottish Government, Carbon Trust and nine major offshore wind farm developers representing
72% of the licensed capacity in UK waters (DONG Energy, E.ON, Mainstream Renewable Power,
RWE Innogy, Scottish Power Renewables/Iberdrola, SSE Renewables, Statkraft, Statoil and
Vattenfall). The multi‐million pound programme has been running since 2008 with the objective of
bringing down the costs of low carbon electricity produced by offshore wind farms.
Page 4/272
PISA Project – Final Report Section 1 - Introduction and project overview
Nomenclature
2
A Area (m )
L Buried length of pile (m)
Cc Coefficient of compressibility
D Pile outer diameter (m)
e Voids ratio
E Young’s modulus of elasticity (MPa)
g Stress ratio at critical state
G Shear modulus (MPa)
h Height of lateral load application (m)
H Lateral load (in the y-direction) (kN)
4
I Section second moment of area (m )
IR Rigidity index (G0/Su)
′ ⁄
IS Sand stiffness coefficient (G0/pa = IS√𝜎𝑣𝑖 𝑝𝑎 )
J API/DNV empirical constant for clay or deviatoric stress (kPa)
k Parameterised soil response initial stiffness parameter
K Coefficient of lateral earth pressure or bulk modulus
L Pile length (m)
LL Liquid limit
m Distributed moment (kNm/m)
M Bending moment (kNm)
n Parameterised soil response curvature parameter
p Lateral distributed load (kN/m)
p’ Mean effective stress (kPa)
PL Plastic limit
PI Plasticity index
su Undrained shear strength (kPa)
t Pile wall thickness (mm)
v Displacement (m or mm) or specific volume
V Section shear force (kN)
w Weighting factor
W Work done or virtual work (Nm)
Coordinate in direction perpendicular to pile loading (m) (lower case - 1D model, upper
x, Y
case - 3D model)
Coordinate in direction of pile loading (m) (lower case - 1D model, upper case - 3D
y, X
model)
z Depth below ground (m) (1D model)
Z Depth above ground (m) (3D model)
3
γ Unit weight (kN/m )
ε Strain
θ Pile neutral-axis rotation (radians)
ψ Pile cross-section rotation (radians)
φ Azimuthal angle about the z-axis (radians), or angle of shearing resistance (radians)
Page 5/272
 Quality of fit (%)

ν Poisson’s ratio
σ Stress (kPa)
ζc Cyclic load characteristic parameter
γ Shear strain
κ Shear area coefficient
Subscripts
0 Initial value
50 Half of the ultimate value
0.1D Value at a deflection of 10% of pile diameter
B With reference to the pile base
G With reference to ground level
H With reference to base shear curves
i Initial quantity
int Quantity within the pile section
m With reference to distributed moment curves
M With reference to base moment curves
p With reference to distributed lateral load curves
raw Direct measurement from instrument (usually in mV)
sd Value at a defined small displacement (generally D/10,000, but D/1,000 for field tests)
sec Secant value
TXC Triaxial compression
TXE Triaxial extension
v In a vertical direction
u Value at ultimate response
Superscripts
‘ Effective quantity (after the subtraction of pore pressure)

̅ Normalised quantity
Page 6/272
1 Introduction and project overview
The PISA project is focused on the investigation and development of improved design methods for
laterally loaded piles, specifically tailored to the offshore wind sector. It is a joint industry project run
as a discretionary project through the Carbon Trust’s Offshore Wind Accelerator programme, and
led by DONG Energy. The academic contributions are provided by an Academic Work Group (AWG)
comprising Oxford University (OU), Imperial College London (ICL) and University College Dublin
(UCD). The project has used state of the art numerical modelling, which has been validated through
a campaign of large scale field tests, to establish new design methods. The numerical simulations
and field testing have been carried out for profiles of stiff over-consolidated low plasticity clay till and
of dense marine sand. Figure 1 below highlights the main elements of the project.
Figure 1: Overview of project approach and scope of work
1.1 Aim and objectives
This report is the final deliverable by the Academic Work Group for the PISA project, covering the
key achievements from the different work packages, followed by the presentation of the main design
recommendations. The report is structured with a front section that details;
 An overview of the project and approach taken;
 The numerical analysis that was undertaken to support the new design methods;
 The field testing completed to provide a validation of the new design approach; and
 The new design methodology that is recommended.
The front section is followed by a number of detailed Appendices that provide the underpinning
scientific work supporting the overall recommendations of the report including:
 A detailed summary of the soil characterisation for both the clay and sand sites;
 Analysis of the interpreted field test data used to validate the design method;
 An analysis of the 3D finite element modelling used to develop the design methodology;
 A design example applying the methodology to a stiff clay site;
 A design example applying the methodology to a dense sand site;
 A detailed description of the analysis tools used throughout the work; and
 A summary of the additional benefits and future exploitation opportunities from the project.
Further details of the development of the design methodologies and of the outcomes from the field
testing can be found in the previous reports submitted by the AWG during the course of the project
(AWG, 2013, 2014a, 2014b, 2015). As agreed with the partners at the outset of the project, the work
covered in PISA is limited to monotonic loading and to soil conditions comparable to those found at
the two field test sites. Engineering judgment will be required to use these methods in practice,
particularly in the application of the methods more widely, as is the case with the application of any
design method.
Page 7/272
1.2 Problem definition and current modelling approaches
More than 1000 offshore wind turbines have been installed around European coastlines; further
installations are planned for the next decade. Most existing offshore wind turbines are founded on
single, large diameter, driven piles, termed monopiles. Early designs were based on monopiles with
a diameter in the region of 4m. In more recent installations the typical diameter is in the region of
6m; future designs are likely to be based on monopiles with diameters of up to 10m or more. These
piles are subjected to substantial overturning moments from the action of the wind on the turbine
and tower, as well as wave and current loading on the monopile and transition piece. A schematic
showing the design problem is shown in Figure 2, along with relevant dimensionless groups,
developed following the work of Kelly et al. (2006) and Le Blanc et al. (2010).
H Key dimensionless groups
Embedded length 𝐿/𝐷

Wall thickness 𝐷/𝑡
h Height of resultant ℎ/𝐷
Clay Sand
𝐻 𝐻
Horizontal load ′ 2
𝑠𝑢 𝐷2 𝜎𝑣𝑖 𝐷
𝑣⁄ 𝐼 𝑣 𝑝
Lateral deflection 𝐷 𝑅 𝐼 √ 𝑎⁄𝜎 ′
𝐷 𝑠 𝑣𝑖
𝑝
L Rotation 𝜃𝐼𝑅 𝜓𝐼𝑠 √ 𝑎⁄ ′
𝜎𝑣𝑖
𝐺0⁄ 𝐺0⁄ ′
D 𝐼𝑅 = 𝑠𝑢 𝑝𝑎 = 𝐼𝑠 √𝜎𝑣𝑖 ⁄𝑝𝑎
Figure 2: Definitions for monopile embedded in clay soil for the PISA project
The method normally used to design the piles under these lateral and overturning loads is based on
a Winkler modelling approach, commonly termed the p-y method. This approach, recommended in
many of the offshore design codes (e.g. DNV, 2014; API, 2010), assumes that the pile acts as a
beam supported by a series of uncoupled springs, each of which represents the local lateral soil
reaction, similar to that shown in Figure 3(a). These springs are normally described by non-linear
functions (p-y curves) to define the soil reaction, p, at a given depth, as a function of the lateral
displacement, y, as shown in Figure 3(b). There is a long history to the development of the p-y
method, dating back to the 1950s and 1960s, when it was first applied to pile design for offshore oil
and gas structures. The original p-y curves were based on the results of field tests on long slender
piles, with diameters around 610mm and a length to diameter ratio of 34 (Matlock and Reese 1956).
The use of the p-y approach for oil and gas structures, for which design is principally concerned with
avoiding ultimate collapse, is considered routine practice. However, the p-y approach has
increasingly been used, in recent years, for the design of offshore wind turbine monopiles, for which
the pile diameters (and diameter to length ratios) are significantly larger than those on which the p-y
method is based. Questions arise as to whether extrapolating the p-y method to large diameter
monopiles is justified. Moreover, the design conditions that are relevant to offshore monopiles depart
Page 8/272
significantly from those that are typically applied in the design of the relatively slender piles used for
oil and gas structures.
MG
HG
(a) (b)
Figure 3: (a) Current p-y modelling approach, and, (b) typical p-y curve for an element of the pile
1.3 Shortcomings of current p-y approaches applied to wind turbine monopiles
There are three particular areas of concern and uncertainty when the current p-y approaches are
applied to the design of wind turbine monopiles:
a) Uncertainties in current p-y curve formulations: The formulations for the p-y curves specified in
the current API/DNV guidelines are of a generic form. Uncertainties typically exist in the choice of
appropriate parameters to calibrate these models for particular applications. In addition, it is unclear
how soil constitutive data obtained using modern methods of in situ testing (e.g. small strain stiffness
data obtained using seismic cone tests) and laboratory testing (e.g. bender element testing and
locally instrumented stress path tests) may be used to assist in the specification of appropriate p-y
curves for a particular offshore site. Furthermore, in contrast to the oil and gas industry, the design
of a support structure for an offshore wind turbine requires a significant number of load cases to be
investigated, including time-domain analyses for dynamic response during the operation of the wind
turbine, as well as pseudo-static push-over analyses. These calculations focus not only on the
ultimate limit state but also the serviceability and fatigue limit states, with a significant emphasis on
design against fatigue. The accumulation of fatigue damage within the foundation and support
structure is strongly conditioned by the dynamic performance of the system, which in turn, depends
on the stiffness of the foundation. In particular, the assessment of the structural natural frequencies
forms an important part of the design process. Uncertainties exist on the extent to which current p-y
formulations are able to provide realistic estimates of foundation stiffness for use in dynamic design
calculations.
b) The conventional p-y approach neglects identified soil/pile interaction mechanisms: Previous
researchers (e.g. Davidson 1982, Lam and Martin 1986, Gerolymos and Gazetas 2006 and Lam
2013) have noted that for laterally-loaded caisson and drilled shaft foundations, with values of L/D
that correspond to typical offshore monopile geometries, the behaviour of the caisson under lateral
loading is influenced by further soil reactions, in addition to the lateral load resistance generated by
Page 9/272
the soil. These include the vertical shear stresses induced on the external perimeter of the
foundation and a moment and a horizontal force that is developed on the base of the foundation.
These additional soil/pile interaction mechanisms appear to become increasingly significant as the
caisson becomes more stocky (i.e. the value of L/D becomes smaller). These additional soil/pile
interaction mechanisms (omitted in the API p-y approach) have been found to have a significant
influence on the behaviour of monopile foundations for offshore wind applications.
c) Extensions of current p-y method to cyclic loading are rudimentary and simplistic: Environmental
loads applied to offshore wind turbine structures are typically cyclic in nature. The effect that load
cycling might have on the long-term performance of a monopile foundation is a key consideration
during the design process. The approach adopted in the current API procedure is to modify the static
p-y curves by using appropriate empirical factors effectively to reduce stiffness. While this provides a
pragmatic way of extending the static p-y method, it does not reflect the detailed processes that are
associated with the degradation of foundation performance as a consequence of cyclic loading. In
particular, it takes no account of the magnitude of cyclic load or the number of applied cycles.
Furthermore, the approach offers no prediction of the accumulation of permanent deflection under
cyclic loading.
In the most recent update to DNV (2014) these shortcomings were recognised in the specific note
F2.4.1 as reproduced in Figure 4.
F.2.4.1 The nonlinear p-y curves recommended in [F.2.2] and [F.2.3] are meant primarily for analysis of piles
for evaluation of lateral pile capacity in the ULS. These p-y curves have been calibrated for long slender jacket
piles with diameters of up to 1.0 m. They have not been calibrated for monopiles with larger diameters and are
in general not valid for such monopiles. P-y curves to be used for monopile design should be validated for such
use, e.g. by FE analysis.
Figure 4: Amendment made to DNV (2014) recommended practice relating to monopile design
The design approach proposed from the PISA project addresses the issues outlined in a) and b)
above. Extensions to cyclic loading (issue c)), however, fall outside of the scope of the project and
this report.
1.4 Overview of project scope
Table 1 outlines different design methods that could be adopted for lateral pile design methods. The
historic (and current) approach, the p-y method, appears to offer many advantages, not only
because it is well understood by the industry, but also because it is easy to embed into existing
design methods, representing a balance between computational efficiency and problem complexity.
This approach is further developed in the PISA project, but extended to include additional
components of soil reaction, which are demonstrated to be important for wind turbine monopiles.
The development of the design method is based on detailed three dimensional finite element
analyses. In this way the benefits of numerical analysis for accurately modelling complexity can be
combined with the computational benefits of the existing, conventional p-y framework. The approach
is structured so that, as more analyses are completed, the methodology can be updated and
improved, with the design method encompassing an entire process, rather than simply prescribing
equations. The design method is set out in such a way as to be wholly compatible with current
calculation approaches used by structural engineers, enabling the geotechnical response to be
accurately captured in the structural analysis.
Page 10/272
Table 1: Lateral pile design approaches
3D FE Winkler-type Model Surface Spring Model

decreasing computational complexity
Model the entire soil Model soil response by Model entire pile-soil
continuum and capture series of springs down the interaction by representative
complete pile-soil interaction pile length springs at the soil surface
Computationally intensive Computationally fast – a Computationally fast but
(getting faster) and (probably) compromise between (probably) least accurate
most accurate accuracy and time
Model detailed soil stratigraphy, Discretise soil layers, but no Difficult to calibrate for piles in
and complex pile-soil interface interaction between layers layered soil and in-
response homogeneous strength profiles
Any foundation geometry Common pile design tool Used for shallow foundations
(the p-y model) (e.g. spudcans) and pipelines
The development of the new design method was executed as shown in Figure 5. At the outset it was
necessary to select two reference soil materials to constrain the scope of the work, and to focus
attention on likely field testing sites. The two materials chosen, stiff clay and dense sand, led to the
selection of Cowden and Dunkirk as the chosen test sites. A wide range of finite element modelling
was undertaken to aid the development of the new design methods and to explore the design of the
field tests. This finite element modelling made use of both historic soil characterisation data and,
latterly, the more recently acquired soil characterisation data collected during the PISA project. This
included a range of in situ testing and laboratory element testing. The design methodology
development involved detailed interrogation of the numerical data combined with a rational
engineering interpretation of the design problem. Given the timescale of the project the numerical
modelling, development of the design method and the design of the field testing were run in parallel,
but not independently. The field testing provided a clear validation of the numerical modelling, which
then formed the basis for the development of the wider design method for large scale monopiles.
Finite element Site

modelling investigation
Calibrated /
validated design
Existing Full scale
Calibration methods
knowledge of simulations
test sites
FE / design
method
Design of field
Test simulations Field tests Full scale
tests
Knowledge of prediction
potential wind Parameterised
farm soils Draft design method
Parameterised
parameterised Validation
simulations
method
Figure 5: Detailed schematic showing project execution
Page 11/272
1.5 Project structure
The PISA project has involved a large consortium of industrial and academic partners, as shown by
the organogram in Figure 6. More than 100 people have been involved in the work during the course
of the project under the direction of the lead Partner and main contractor, DONG Energy. The work
described in this report is that principally completed by the Academic Work Group (AWG), whose
contributions to the project have been led and managed by Oxford University, with Byron Byrne and
Ross McAdam providing the principal interface with DONG Energy. The Academic Work Group’s
function has been to provide the scientific advice to the project, particularly to direct and execute the
work required to develop new design methods for laterally loaded piles.
The detailed structure of the Academic work group is shown in Figure 7. During the project Oxford
took specific responsibility for the development of the design methods led by Harvey Burd along with
the design and completion of the field testing led by Byron Byrne, as well as the overall co-ordination
of the work led by Byron Byrne. Ross McAdam was the lead Research Assistant (RA) on the project.
The finite element studies, with analysis and derivation of soil parameters, were led and formulated
at Imperial College London (ICL) under the direction of Lidija Zdravkovic and input from David
Taborda, with the Geotechnical Consulting Group (GCG) contracted to carry out initial field test and
full scale parametric simulations. Laboratory element testing of samples obtained from the two sites
was carried out at ICL under the direction of Richard Jardine. The field testing campaign was
designed, supervised and analysed by the three partners of the Academic Work Group, with detailed
supervision by post-doctoral researchers from Oxford University and University College Dublin. The
four research students (Beuckelaers, Ushev, Liu, Abadias) have made important contributions to the
work.
All contracting was completed by DONG Energy. However the AWG contributed significantly to the
development of the detailed specification and scope of work for many of the sub-contracts that were
let to Assisting Consultants and to the Testing Contractors.
Page 12/272
Figure 6: Organogram for PISA project
Page 13/272
Test Contractor
ESG
Lead Partner
DONG Energy
Carbon Trust / OWA /
Project Partners /
Independent Technical
Field Tests Review Panel
Overall Project Co-ordination and

Interface with Industry partners
Oxford University
Byrne (PI), McAdam (RA)
Element Testing / Field Testing

Imperial College London Numerical Analysis
Jardine GCG
Ushev (RS), Liu (RS) Schroeder
Hypothesis, Design
methodology, Field Soil Parameter Derivation and
Field Testing, Instrumentation Testing, Interpretation Numerical Analysis
and Data Analysis and Reporting Imperial College London
University College Dublin PISA AWG Zdravkovic, Taborda, Potts,
Gavin, Igoe (RA)
Abadias (RS)
Field Testing, Instrumentation, Development of Modelling

Data Analysis and Oversight Approach
Oxford University Oxford University
Byrne, McAdam (RA), Burd, Burd, Byrne, McAdam (RA),
Beuckelaers (RS) Houlsby, Martin
Figure 7: Structure of contributions of the PISA Academic Work Group
1.6 Project timeline
The Academic Work Group tender for the PISA project was advertised by DONG Energy and the
Carbon Trust in March 2013. The consortium led by Oxford University with Imperial College London
th
and University College Dublin submitted the successful tender on 30 April 2013, following revisions
st
after interview. The project formally commenced on 1 August 2013. Figure 8 provides the timeline
of the Academic Work Group element of the work, tracking the deliverables that report on the
development of the new design methods.
Field testing
Project kick-off
meeting Draft methodology Final report
for clay delivered draft delivered
Project plan Draft methodology Factual report
delivered for sand delivered delivered
Jul 13 Oct 13 Jan 14 Apr 14 Jul 14 Oct 14 Jan 15 Apr 15 Jul 15 Oct 15 Jan 16
Figure 8: Timeline of PISA AWG deliverables
Page 14/272
The key highlights in the timeline include:

st
 The project plan (AWG 2013), delivered on 1 September 2013 identified the deficiencies in the
existing design methods and outlined a programme of work to develop a new design
methodology, including the requirements of numerical analysis and a suggested field test
campaign.
 The field testing campaign initiated in November 2013, with the release of a tender for test
contractors. There was a period of learning during the field test period with the initial testing at
Cowden taking longer than the testing at Dunkirk.
st
 The draft methodologies in clay (AWG 2014a) and sand (AWG 2014b), submitted on 1 May
st
2014 and 1 August 2014 respectively, proposed new design methodologies for the target
materials, including a process for obtaining design equations.
st
 After a successful test campaign, the final pile was tested at Cowden on 21 July 2015. The
th
factual report (AWG 2015) of these tests was submitted on 6 November 2015, along with digital
files of the test data.
 The draft final report summarises the work carried out on the project and outlines a
recommended design approach for laterally loaded piles.
1.7 Meetings and workshops
During the project the Academic Work Group met every second month (approximately) to discuss
progress, to develop the work and to co-ordinate activity across the different institutions. These
meetings were routinely attended by all academic participants, including research students.
Additional meetings between members of the AWG were organised as required. In the phase of
work leading up to and including the field testing OU met with DONG Energy on a monthly basis.
To facilitate the dissemination of the work to the wider PISA Project technical community a number
of meetings and workshops were organised. These allowed an open discussion of the work,
nd
providing a forum for feedback to the AWG. A kick-off meeting was held on 2 July 2013 at DONG
Energy’s office (London) to agree the scope of work, and importantly to define the limitations of the
project. A Partner/ITRP workshop was held on 14 May 2014 at the Burnham Beeches Hotel, where
the draft design (clay) methodology and proposed field testing were presented. A further meeting
was held with DONG Energy / ITRP on 16 September 2014 at St Catherine’s College in Oxford to
discuss the draft sand design methodology. Following both meetings the AWG addressed feedback
and comments from DONG Energy, Project Partners and the ITRP, resulting in minor amendments
to the design methodologies. Following the completion of the field testing, a Partner/ITRP workshop
was held at St Catherine’s College, on 4 September 2015, to allow a detailed presentation of the
field test results. The final Partner/ITRP workshop was held on 16 February 2016 at Egrove Park in
Kennington near Oxford. This workshop focused on the presentation of the recommended design
methodology, the supporting evidence and the achievements of the project. This series of
workshops has ensured a full and open intellectual engagement between the AWG and the Partners
/ ITRP. Each workshop has benefited from a generous social program.
Page 15/272
PISA Project – Final Report Section 2 - Overview of site selection and characterisation
2 Overview of site selection and characterisation
The aim of the PISA project is to provide a new design methodology that can be applied to a variety
of sand and clay soils. To constrain the parameter space for developing the new method the project
partner technical experts agreed, early in the project, to focus research on two reference materials
representing stiff low plasticity clay and dense sand. Two onshore test sites were chosen for the field
testing to represent these materials: (a) the Cowden clay site in the North-East of England; and (b)
the Dunkirk sand site in northern France. These sites were deemed to adequately represent (a)
glacial, ductile, low plasticity stiff clay and (b) dense sand ground conditions, encountered in some
sectors of the North Sea. Both onshore test sites have extensive histories of pile testing activities,
and consequently the soils at the sites have been reasonably characterised through field testing and
laboratory experiments. This makes them excellent reference sites, with the existing knowledge
base feeding into the preliminary phase of advanced 3D finite element (FE) modelling, and
additional site characterisation, commissioned during the project, feeding into the final phase of
numerical modelling. This section provides a brief overview of the two sites and a high level
characterisation of the key soil parameters. Detailed accounts of laboratory and field testing
programmes and derivation of soil constitutive parameters are summarised in Appendix A.
2.1 Cowden
The site location map and satellite image for Cowden are shown in Figure 9 and Figure 10. This has
been the test site of the Building Research Establishment (BRE) since 1976 and most of the historic
data of field and laboratory investigations on this site, carried out pre-1990’s, are summarised in
Powell & Butcher (2003). These data were used in the derivation of the parameters for the
constitutive model employed to simulate the Cowden till behaviour in the preliminary 3D FE analyses
of the PISA monopiles. Additional advanced experimental programmes were commissioned from
summer 2014 to early 2015 as part of the PISA project, in order to bridge deficiencies in the historic
data and improve the derivation of soil parameters for final numerical analyses.
Figure 9: Location of Cowden test site (Google maps)
Page 16/272
PISA test site
Previous BRE test site
Figure 10: Satellite image of Cowden test site (Google Earth)
The initial ground conditions at Cowden are presented in Table 2 and Figure 11(a). The ground
water table is located at 1m below the ground surface and measurements indicate an under-drained
pore water pressure profile in the top 12m. A hydrostatic distribution exists in deeper deposits.
(a) Ground and ground water (b) K0
Figure 11: Profiles of Cowden Till from Powell & Butcher (2003)
The historic, predominantly oedometer, test data in Figure 11(b) indicate high initial earth pressure
coefficient, K0, in the clay till deposit to 5m depth. These values are considered unrealistic, as the
one-dimensional nature of an oedometer tests is not representative of processes induced by glacier
Page 17/272
advances. As a consequence, the value of K0 adopted for numerical modelling is reduced to 1.5 in
the top 4m.
Table 2: Stratigraphy at Cowden
Depth Material Description

0-5m Weathered till Weathered brown stiff stony clay till. Some fissures.
5-40m Unweathered till Unweathered dark grey brown stiff stony clay till. Becoming less
stony with depth. Stone mainly chalk. With silty sand/gravel bands at
approximately 12 and 18m depth, as detected by CPT profiling.
40m + Chalk -
The key input parameters for engineering design of short term loading cases in clay soils are the
undrained shear strength and the shear modulus. For early stages of monopile loading, the shear
stiffness at small strains is particularly important. The profiles of undrained triaxial compression
(TXC) strength and the maximum shear modulus, G0 (as a function of the mean effective stress, p’),
are shown in Figure 12(a) and (b) respectively, while the normalised (by p’) secant shear stiffness
degradation with strain is presented in Figure 13.
(a) undrained strength su in triaxial (b) maximum shear modulus G0

compression (TXC)
Figure 12: Profiles of Cowden Till properties obtained using in situ testing, laboratory testing
and historical data from Powell & Butcher (2003)
Page 18/272
Figure 13: Small strain stiffness of Cowden till

Parameters derived from the historic data as input into preliminary 3D FE analyses are shown as
grey solid lines. The undrained TXC strength profile adopts a high value of 200kPa in the top 2m of
the deposit which then reduces to 5m depth and subsequently follows an almost linear strength
increase with depth. The derivation of the shear modulus profile, G0, made use primarily of
laboratory triaxial and resonant column experiments, as in situ crosshole and downhole shear wave
data showed significant scatter (historic data are not shown in Figure 12(b) for clarity). The reported
shear stiffness degradation curves lacked data at strains smaller than 0.01%, which reflects
insufficient resolution and accuracy of triaxial testing instrumentation at the time when those
experiments were performed.
Figure 14: CPT profiles from PISA site investigation at Cowden

As part of the SI campaign for the PISA project, individual CPTu profiles were determined at each
pile location, as well as other locations around the site. A summary of the CPT data is presented in
Figure 14, showing the average (mean), maximum and minimum recorded values for all pile
locations. The CPT cone resistance traces (qc) were very consistent across the site, falling in a
narrow band and indicating that the superficial deposit has much lower undrained strength than
originally assumed. At the same time new SCPT profiles, interpreted in Figure 12(b), indicated
higher values of the maximum shear modulus G0. As a consequence, new laboratory testing
Page 19/272
programmes were commissioned at ICL and GEOLABS, using advanced triaxial apparatus and
intact Cowden till specimens sampled on site, to further examine strength and stiffness of the
deposit and aid the derivation of the final set of constitutive model parameters for Cowden till.
The new soil data shown in Figure 12(a), from intact triaxial samples (TXC) and hand shear vane
(HSV) tests, confirmed lower undrained TXC strength in the top 2m, but good agreement with the
rest of the originally derived profile. The black solid line shows the profile adopted for final numerical
analyses. The interpretation of the maximum shear modulus from the triaxial small strain
measurements in Figure 12(b) agrees well with the originally derived profile of 800p’ in deeper
deposit, although higher G0 is indicated in superficial layers. However, G0 profiles interpreted from
new bender element shear wave velocity measurements on the same triaxial samples agree well
with the interpretations of G0 from the new in situ dynamic measurements (perhaps suggesting
anisotropy in elastic stiffness). The G0 profile of 1100p’ is adopted for the final 3D FE analyses,
being approximately an upper boundary of G0 interpretation from triaxial small strain measurements
and a lower boundary of G0 interpretation from dynamic measurement techniques. The clay till
deposit is modelled as a single layer, ignoring the presence of the two sand layers detected from
CPT traces and depicted in Figure 11(a). This is justified by the fact that the maximum depth of the
test piles is 10.5m, which is 1.5m above the shallower sand inclusion at 12m depth. Further details
about model parameter and ground conditions can be found in Appendix A.
2.2 Dunkirk
A location map and satellite image of the Dunkirk test site are given in Figure 15 and Figure 16. The
site is part of the extended beach area, situated on land owned by the Port Authority of Dunkirk. This
site has been used as a field testing site for a number of research projects on structures, particularly
piles, in marine sands since the 1980s. These projects include:
 the French CLAROM/IFP project;
 the UK’s Imperial College and Building Research Establishment tests in the mid-1990s; and
 the GOPAL project in the late 1990s.
In addition, the French SOLCYP project in the late 2000s and still on-going has used a test site very
close to the PISA Dunkirk site at Loon-Plage.
Figure 15: Location of Dunkirk test site (Google Maps)
Page 20/272
PISA test site
Figure 16: Satellite image of Dunkirk test site (Google Earth)
The available historic data that was suitable to calibrate the constitutive model employed to simulate
the Dunkirk sand behaviour in the preliminary 3D FE analyses was found in Chow (1997) and
Kuwano (1999). The former reference contains some field investigation data by BRE during the pile
testing programme on the behaviour of displacement piles conducted by ICL in mid-1990s. The
latter reference presents a programme of laboratory experiments on Dunkirk sand conducted at ICL,
focussed on behaviour at very small strains. Additionally, principally for the derivation of critical state
parameters for Dunkirk sand, use was made of three drained triaxial tests which were taken to
ultimate failure, carried out at ICL by Aghakouchak (2015) as part of his PhD research unrelated to
the PISA project.
The ground conditions at Dunkirk are summarised in Table 3:
Table 3: Stratigraphy at Dunkirk

0-3m Hydraulic Fill Sand fill that was dredged from the offshore Flandrian deposits, and
placed to raise the ground level. No compaction or surcharging has
taken place.
3-30m Flandrian Sand Marine sand deposited during three local marine transgressions.
These sands are often separated by organic layers which
accumulated between transgressions. A 600mm thick organic layer
is found at around 8m depth, separating the Flandrain sand into
upper and middle units.
30m + Yprésienne Clay An Eocene marine clay (also known as London Clay and Argile de
Flandres) which extends beneath the southern North Sea.
Page 21/272
Similar to the Cowden site, the new SI campaign for the PISA project determined individual CPTu
profiles at each pile location at Dunkirk. These are summarised in Figure 17, showing the average
(mean), minimum and maximum recorded values for all pile locations. From the new pore water
pressure measurements shown in Figure 18, the ground water level is estimated at around 5.4m
depth, which is about 1.4m deeper compared to the historic data. With no further site investigation
performed, the hydraulic conditions above the water table remained uncertain, as this part of the
deposit is likely to be unsaturated. The observed creation of stable gaps to sizeable depths, with a
vertical soil face around the laterally loaded test piles, clearly confirmed the existence of
considerable effective stresses (potentially suctions) above the water level, or a certain degree of
cementation which could not be confirmed or quantified. The new suction profile depicted in Figure
18 was adopted for numerical modelling.
Figure 17: CPT profiles from PISA site investigation at Dunkirk

The new CPT data was also utilised in the interpretation of the initial density, Dr, of the sand deposit.
Application of different correlations, as summarised in Appendix A, confirmed 100% density of the
hydraulic fill, but resulted in very different (up to 20%) density profiles in the natural sand. Combining
the new and historic CPT interpretations, the adopted initial density in the natural sand remained at
75%, as in the preliminary analyses.
The change in the ground water table meant that the initial effective stresses were higher than
originally assumed. Consequently, the shear modulus at very small strains, G 0, which depends on
the meant effective stress, p’, is also higher than assumed in the preliminary analyses (Figure 19).
The new laboratory testing of Dunkirk sand, performed as part of the PISA project, involving a range
of initial densities and stress levels, enabled better modelling of sand’s volumetric response and
stiffness degradation. However, the strength parameters in terms of the critical state line remained
unchanged compared to the original model calibration.
Further details about model parameters and ground conditions can be found in Appendix A.
Page 22/272
Figure 18: Interpretation of the initial pore water pressure profile from new CPT data
Figure 19: Interpretation of the maximum shear modulus profile from new data
Page 23/272
PISA Project – Final Report Section 3 - Numerical 3D finite element modelling
3 Numerical 3D finite element modelling
The 3D finite element (FE) modelling on the PISA project was aimed at: (i) establishing the basis for
developing a new design methodology for laterally loaded piles; (ii) identifying mechanisms of pile-
soil interaction; and (iii) informing the design of test piles and field testing programmes at the chosen
clay (Cowden) and sand (Dunkirk) test sites. The accuracy of the developed numerical model is
assessed through comparison of the predicted and measured responses of the test piles, which is
discussed in Appendix C. This section presents the development of the numerical model for laterally
loaded monopiles that has been employed throughout the project using the finite element software
ICFEP (Potts & Zdravkovic, 1999, 2001).
3.1 Model development
A numerical model of any geotechnical problem under investigation is required to represent realistic
geometry, ground conditions, boundary conditions and material behaviour, to be able to produce
accurate predictions of its response to prescribed actions.
3.1.1 Geometry
The problem geometry in the PISA project is that of a laterally loaded monopile, with geometric
characteristics shown in Figure 2: the pile diameter, D, its embedded length, L, the height of the
stick-up, h, and the pile wall thickness, t. The geometries of the test piles are summarised in Table 4.
The horizontal loading of a monopile implies a plane of symmetry in the problem geometry and it is
therefore sufficient to discretize only half of the geometry into a finite element mesh. Figure 20
shows an example of a developed FE mesh for test pile T6 in clay.
Figure 20: Typical FE mesh for the analyses of test piles
The FE mesh contains three different materials: the soil, the pile and the pile-soil interface, which
are all discretised with high-order displacement based isoparameteric finite elements. The soil
makes use of 20-noded hexahedral elements, the pile uses 8-noded shell elements (Schroeder et
al., 2007) and the interface is discretised with special 16-noded zero-thickness interface elements
(Day & Potts, 1994). The interface elements are introduced around the outside of the pile to allow
Page 24/272
appropriate constitutive modelling of the pile-soil interface and to allow separation between the pile
and the soil (primarily opening of a gap around the pile, which is the likely consequence of its lateral
loading). The number of elements and mesh geometries developed for test piles are presented in
Appendix C1.1.
Table 4: Geometry of test piles

Analysis D h h/D L L/D t D/t Equivalent Pile
(m) (m) (m) (mm) Cowden Dunkirk
T2 0.76 10 13.2 2.3 3 10 76 CM2 DM7
T4 0.76 10 13.2 4.0 5.25 13 / 14 58 / 54 CM9 DM4
T6 0.76 10 13.2 7.6 / 6.1 10 / 8 25 30 CM3 DM3
T7 2.0 10 4.8 10.5 5.25 25 / 38 80 / 52 CL2 DL2
*Dual numbers represent Cowden / Dunkirk geometric value
3.1.2 Boundary conditions

It is necessary to prevent rigid body movements of the FE mesh and this is achieved by setting to
zero all three displacement components in the three coordinate directions (X, Y and Z) at the base of
the mesh. In addition, the displacements normal to the vertical cylindrical boundary are also set to
zero, together with zero forces in the vertical Z-direction and directions tangential to this boundary.
To ensure that the X-Z plane at Y=0 is a plane of symmetry, the displacements normal to this plane
(i.e. in the Y-direction) are set to zero, as are the forces in the X- and Z-directions.
The horizontal load at the pile top, at Z = h, is applied in a displacement-controlled manner, by

prescribing increments of uniform displacement in the X-coordinate direction around the pile
perimeter. The horizontal load, H, at pile top is obtained as a reaction to the applied horizontal
displacements.
3.1.3 Ground conditions

From the onset of the project a decision was made not to consider pile installation effects on initial
ground conditions. It was considered that such effects are difficult to verify against realistic field
conditions and would therefore present unknowns which are hard to quantify and take proper
account of in a numerical analysis. Consequently, the piles were modelled as “wished in place”, in
initially undisturbed ground conditions.
Setting the realistic initial ground conditions for an FE analysis requires the specification of the
distributions of the vertical and horizontal effective stresses, over-consolidation ratio, pore water
pressure, void ratio, undrained strength and the maximum shear modulus. Usually a combination of
field and laboratory ground investigations needs to be assessed for accurate derivation of initial
ground conditions. Field testing techniques, such as CPT, could be inaccurate for profiling the
strength in shallow depths of the deposit, where the investigation may be complemented by shallow
trial pits and extraction of soil samples for triaxial testing in the laboratory. The maximum shear
modulus profile could be interpreted from dynamic field techniques such as SCPT, or from the
laboratory dynamic (Bender Element) and static (small strain instrumentation) measurements,
normally using a triaxial apparatus. Derivations of the initial ground conditions at both pile test sites
are described and discussed in Appendix A, with brief extracts shown in Section 2 above.
Page 25/272
3.1.4 Material behaviour

Three different materials are modelled in the FE analyses of PISA monopiles: the soil (over-
consolidated glacial till and dense sand), the steel (pile) and the pile-soil interface.
For modelling monopiles under the monotonic lateral loading, the soil constitutive model is required
to accurately reproduce its small strain stiffness and conditions at failure. The former requirement is
important for realistic simulation of the pile response at early stages of loading (i.e. the initial
gradient of the load-displacement curve). The latter must account for realistic failure loads. The
available site investigation and laboratory experimental data were insufficient for assessing the
effect of soil anisotropy on its strength and stiffness and consequently the soil is modelled as an
isotropic material.
The clay till material is modelled with an extended generalised Modified Cam Clay (MCC) model of
the type described in Tsiampousi et al. (2013). The behaviour of dense sand is reproduced with a
bounding surface plasticity type model originally proposed by Manzari & Dafalias (1997), using here
the modified version of Taborda et al. (2014). Brief descriptions of the models and derivation of
model parameters from the available soil data are presented in Appendix A. Both models are
formulated within the critical state framework and can therefore reproduce conditions at failure. They
can also accurately simulate the small strain nonlinearity of soil behaviour and can account for the
variation of soil strength in the deviatoric plane. The clay model adopts the Hvorslev surface on the
dry side of the critical state to account for realistic strengths of overconsolidated clays, which are
otherwise overpredicted with the basic MCC model. The sand model accounts for realistic volumetric
deformations of sands, contraction and dilation, which are dependent on the void ratio and stress
level. A summary of parameters for each of the models, and their derivation from the available soil
data, are presented in Appendix A.
The pile-soil interface is simulated with elasto-plastic constitutive models which have zero strength if
loaded in tension and assume the compressive strength of the surrounding soil if loaded in
compression. The former characteristic enables the opening of a gap around the pile during lateral
loading. Parameters chosen for the interface models are discussed in Appendix C.
The pile is considered linear elastic and its behaviour is described by the Young’s modulus of
200GPa and a Poisson’s ratio of 0.3.
3.2 Numerical predictions
This section shows the behaviour of test piles predicted from 3D FE analyses. Only brief results from
Cowden test pile analyses are presented here, as the mechanisms of pile response at Dunkirk are
very similar. More detailed comparison between numerical predictions and the pile test data from
both sites is discussed in Appendix C.
Load-displacement curves from 3D FE analyses of piles installed at the Cowden test site are shown
in Figure 21. The horizontal displacement of the pile is taken at mudline level and the horizontal
force is applied at pile top (Z=10m). The comparison of curves for piles T2, T4 and T6, all with
D=0.76m, shows clear evidence of the effect of pile length on its capacity. The dominant effect of
pile diameter on lateral capacity is also evident from the load-displacement curve for pile T7 (D=2m).
Figure 22 shows horizontal displacement profiles of the test piles at two stages of loading, when
displacement at mudline is approximately 0.1% (early loading) and 10% (nominal failure) of pile
diameter. For ease of comparison, the pile depth, z, is normalised by the length of that pile, L. The
Page 26/272
shortest pile T2 rotates as a rigid body from early stages of loading. The remaining profiles indicate
some flexing at the beginning of pile loading, but at the stage of nominal failure only pile T6, with
L/D=10, retains some flexure, whereas T4 and T7 show deflection profiles that are predominantly
associated with rigid body rotation.
Figure 21: Load-displacement curves from Cowden test pile 3D FE analyses
(a) vG = 0.001D (b) vG = 0.1D
Figure 22: Deflection profiles of Cowden test piles
Page 27/272
PISA Project – Final Report Section 4 - Field testing
4 Field testing
It was established within the PISA project plan (AWG 2013) that a field testing campaign would be
conducted at a clay site and a sand site to benchmark and validate the 3D finite element modelling,
which has subsequently been used to develop the recommended design methodology. A detailed
description of the field test setup and factual results can be found in the Field Test Factual Report
(AWG 2015). The primary objectives of these field tests were to:
 examine in detail the load-deflection relationships for the pile-soil system, especially in areas
important to design such as the pile-soil stiffness response and ultimate capacity;
 determine the load distribution in the pile-soil system;
 explore the effect of variations within a realistic geometric parameter space; and
 examine the effects of cyclic loading.
In total 28 piles were tested with varied diameter, length and wall thickness with both monotonic and
cyclic loads and with up to 130 simultaneous instrument measurements of pile and soil response.
These field tests therefore represent a new industry standard database against which design models
in clay and sand may be compared and validated.
4.1 Preparation
The first design of the test program, including a set of desired pile geometries, site locations, test
layout, loading regimes, testing schedule and instrumentation setup was outlined by the AWG in
September 2013 (AWG 2013). Procurement for the field test campaign was initiated in November
2013 by DONG Energy with the distribution of the test contract tender, as shown in Figure 23.
Contracting Manufacture Testing

Installation Cowden Dunkirk
Test contract 1st plates 1st pile Last Cowden Cowden
tender released rolled vibro’d pile installed 2.0m diam
ESG Piezos
selected installed Rogue
bolt
Last plate Bomb

ESG fire
delivered Last Dunkirk
Steel contract pile installed Dunkirk
signed 2.0m diam
1st piles delivered
to Cowden
Oct 13 Jan 14 Apr 14 Jul 14 Oct 14 Jan 15 Apr 15 Jul 15 Oct 15
Figure 23: Timeline of field testing, events shown as vertical lines

The design of the field tests was further refined through analysis of 3D finite element simulations
(described in more detail in Appendix C) and dialogue with the bidding test contractors during the
contracting and pile manufacture phases of the project. Input from the AWG, test contractors, piling
Page 28/272
contractor and funding partners was collated by Atkins to produce the Project Testing Manual, which
provided a unified reference of detailed guidance for contractors on all phases of the test campaign
including:
 health and safety plans;
 responsibilities and lines of communication;
 site plans, rules and SI;
 test assemblies, instrumentation and expected behaviour; and
 method statements.
Prior to the beginning of pile testing extensive preparatory site works were required to ensure
smooth testing operations and the availability of data for pre and post-processing of the results.
These works included:
 an initial SI campaign including borehole sampling, magnetometers, CPTs and SCPTs;
 site development including site clearing, installation of piling mats and drainage,
establishment of welfare facilities and site security; and
 the installation of preparatory instrumentation including ground water level sensors and
embedded pile response piezometers.
4.2 Test specification
Ideally, full scale offshore tests of 6 – 10m diameter piles would be used to benchmark the new
design methodologies. However, due to the high cost of testing offshore and the technical
constraints of the equipment that could be mobilised onshore, it was accepted that testing would be
performed onshore at reduced scale. Scaled pile geometries and loading regimes were adopted
which were representative of offshore wind foundations, as shown in Figure 24 and include:
 pile diameters of 0.273m, 0.762m and 2.0m;
 embedded lengths between 1.43m and 10.5m, providing a range of normalised length
3 ≤ 𝐿⁄𝐷 ≤ 10;
 wall thicknesses 7mm to 38mm, providing a range of normalised thickness 30 ≤ 𝐷 ⁄𝑡 ≤ 80;
 load application heights 5m and 10m, providing load eccentricities 5 ≤ 𝑀𝐺 ⁄𝐻𝐺 𝐷 ≤ 18.3; and
 predominantly monotonic load tests, supplemented with 1-way and 2-way cyclic loading.
(a) 0.273m diameter (b) 0.762m diameter (c) 2.0m diameter
Figure 24: Photographs during tests of piles of varied diameter

Note that Figure 24(a) and (b) show test piles loaded against larger reaction piles, whilst Figure
24(c) shows the simultaneous testing of two 2.0m diameter piles, where a larger reaction pile would
not be feasible.
Page 29/272
The final selection of instrumentation was made in consultation with the main test contractor to
capture high quality data at the necessary range, resolution and sampling rate when measuring
small and large displacements, as well as high frequency cycling. These instruments were chosen to
measure both the above ground load-displacement response as well as the embedded pile-soil
behaviour. A schematic of a fully instrumented 0.762m diameter pile test is shown in Figure 25,
including:
 Microelectromechanical system (MEMS) inclinometers to measure above ground rotation;
 Displacement transducers to measure above ground deflection;
 Load cells to measure the applied load and moment;
 Fibre optic strains gauges from which the embedded section bending moments are
interpreted;
 Retrievable extensometers to act as a back-up and validation of the fibre optic gauges;
 Retrievable inclinometers from which the embedded pile deflection is interpreted;
 Pore pressure transducers to measure dissipation prior to testing and response during
testing; and
 Temperature for calibration of temperature sensitive instruments.
RAMSTROKE
LOAD CELL - LCL
STICK UP
INCLINOMETERS - SUI
DISPLACEMENT
TRANSDUCERS - LVDT
AH PH
AL PL
EXTENSOMETERS - IPE z
INCLINOMETERS - IPI
FIBRE OPTICS - FOS
PIEZOMETERS - PZT
Figure 25: Fully instrumented 0.762m diameter pile test

Within the design of the field tests, key tests were duplicated to provide redundancy in the event of a
test failure. With the success of the main tests, the redundant piles were available to explore
additional phenomena, such as the effect of a varied loading rate. These additional tests are
described in more detail in Appendix G.
4.3 Delivered output
The Field Test Factual Report (AWG 2015) provides a detailed description of the site
characterisation, test setup and loading regimes as well as a reference guide to the accompanying
digital datafiles. The datafiles, which are presented in a clear hierarchical structure, provide single
datafiles of the measured and interpreted data from all instruments for each pile test. An overview of
the additional analysis of the field test data is provided in Appendix B and Appendix C.
One of the most successful features of the field testing campaign was the use of fibre optic strain
gauges to monitor both the installation and loading response of the driven test piles. The excellent
resolution, stability and sensitivity of these instruments has allowed the depth-wise variation of
Page 30/272
embedded pile response to be accurately captured and either integrated or differentiated to obtain
other relevant quantities, using the methodology described in Appendix F2. Figure 26 shows the
measured and interpreted embedded response for pile CM3 (clay site, D = 0.762m, L/D = 10), where
the expression for the bending moment is differentiated to calculate the equivalent distributed lateral
load (previously termed ‘p-y’ curves). Note that due to the slender geometry of pile CM3 the
measured bending moment is predominantly caused by the distributed lateral load. However, for
shorter piles it is necessary to include the effects of the other soil reactions to obtain an accurate
relationship between the measured bending moment and the applied distributed load.
(a) displacement profile (b) rotation profile (c) bending moment
(c) section shear force (d) distributed load (e) soil reaction curves
Figure 26: Embedded pile response for CM3 (D = 0.762m, L/D = 10) (HG = 425kN, vG = 99.3mm)
Page 31/272
4.4 Comparison of measured and predicted monotonic response in clay
As shown in Figure 27, measurements of the pile load-displacement response demonstrate that the
3D finite element modelling is well suited to predicting the response of monotonic laterally loaded
monopiles in clay of varied geometry.
(a) CM2 (D = 0.762, L/D = 3)
(b) CL2 (D = 2.0, L/D = 5.25)
Figure 27: Comparison of field measured response, 3D FE and existing API/DNV methods in
clay
Page 32/272
To quantify the accuracy with which the measured and computed load-displacement responses are
correlated a ‘quality of fit’ metric has been adopted, based on the nomenclature in Figure 28. The
quality of fit metric is defined as,
𝐴𝑟𝑒𝑓 − 𝐴𝑑𝑖𝑓𝑓
𝜂=
𝐴𝑟𝑒𝑓 Equation 1
Separate metrics have been determined for the performance up to ultimate response (𝜂0.1𝐷 ) and the
small displacement performance (𝜂𝑠𝑑 ). The reference small displacement threshold for the field test
results is D/1,000. For the determination of the field test accuracy metrics, the reference load-
displacement curve is derived from the envelope of the field test load steps, as described in
Appendix F3.
Aref H
H Reference Aref
Adiff Reference
Adiff
Other
Other
v v
0.1D D/1,000
(a) Ultimate response (η0.1D) (b) Small displacement response (ηsd)
Figure 28: Illustration of area calculation for accuracy metrics
For the four field test piles which were simulated using 3D finite element modelling, the average
quality of fit metric is 88% and 86% for the ultimate and small displacement responses respectively.
This shows a significant improvement over the ability of the existing API/DNV approach, with quality
of fit metrics of 49% and 21% respectively.
Note that two ranges have been adopted in the calculation of the small displacement quality of fit ηsd,
in this report. When assessing the quality of fit for numerical analyses (section 5, Appendix D and
Appendix E) the range for comparison for ηsd is D/10,000. This is anticipated to be a representative
displacement for the range of stiffness that is most significant to the pile natural frequency. When
assessing the quality of fit for field test measurements (section 4 and Appendix C) the range for
comparison for ηsd is D/1,000, which is the smallest measured displacement that is common
between all field tests. The previous Field Test Factual Report (AWG 2015) adopted a small
rd
displacement measure of the 3 applied load step. This measure was adopted at an early stage of
analysis, has since been superseded and is not used in this report.
Page 33/272
(a) DM7 (D = 0.762, L/D = 3)
(b) DL2 (D = 2.0, L/D = 5.25)
Figure 29: Comparison of field measured response, 3D FE and existing API/DNV methods in
sand
Figure 29 shows a comparison of the measured field test response at the Dunkirk sand test site to
predictions using 3D finite element modelling and the existing API/DNV approach. The complexity
involved in predicting the response of piles in very dense sands means that the accuracy of
response prediction is lower than that seen for the clay site. On average the quality of fit metric for
the ultimate and small displacement responses are 81% and 72% respectively, which still represents
Page 34/272
an improvement in accuracy and consistency over the API/DNV approach, with average quality of fit
metrics of 70% and 42% respectively.
The field test measurements allow the accuracy of the 3D finite element modelling not only to be
demonstrated for the integrated ground level load-displacement response, but also for the
embedded pile-soil load distribution at corresponding ground-level displacements, as shown in
Figure 30. The excellent correlation between the measured pile response and the 3D finite element
modelling indicate that the underlying physics of the pile behaviour is well simulated for varied pile
lengths and diameter and that the existing API/DNV approach is not suitable for predicting the
response of monopiles in clay soil similar to the Cowden Till and of these geometries.
(a) CM2 (D = 0.762m, L/D = 3) (b) CL2 (D = 2.0m, L/D = 5.25)

(vG = 71.9mm) (vG = 208mm)
Figure 30: Comparison of measured and predicted embedded pile response
4.5 Analysis of response to cyclic loading
During the field testing campaign a single pile at each test site was dedicated to examining the
effects of long term cyclic loads over 5 days. A more detailed analysis of the results from the cyclic
testing is provided in Appendix G. Whilst it is accepted that the prediction of pile behaviour under
cyclic loads is a complex problem, the field tests carried out during the PISA project provide
researchers and designers with the tools to initiate the development of robust design approaches
and to maximise the value of future test campaigns. As shown in Figure 31(a) these piles featured a
pulley backstay system, which allowed good control of sinusoidal loads from 0kN to a target load.
The load levels were varied, as shown in Figure 31(b), with up to 11,000 cycles at each load step.
Page 35/272
80
60
HG (kN)
40
20
0
Time
(a) Photograph of test assembly (b) Load time series
Figure 31: Cyclic testing in Dunkirk DM2 (D = 0.762m, L/D = 5.25)

During cyclic tests in clay it was observed that for small loads, the accumulated rotation is small and
tends towards a repeated accommodated response, as shown in Figure 32(a). However, as the load
amplitude is increased, the response does not settle into an accommodated pattern, but instead we
see continuous ratcheting that would eventually result in failure of the structure. Using this and future
analysis it may be possible to establish a safe threshold of load amplitude under which the
accumulated deflection under cyclic loads may be neglected.
(a) 0 – 20kN (b) 0 – 60kN
Figure 32: Accumulated rotation during cyclic loading of pile CM5 (D = 0.762m, L/D = 5.25)
An analysis of the field test cyclic response is given in Appendix G2 and summarised here. Figure
33 shows a comparison of the cyclic and monotonic response for identical geometries in clay. This
figure demonstrates that at a load amplitude of 20kN the accumulated displacement is small and
accommodated, but on increasing the load amplitude to 60kN the accumulated displacement is
significant, continuous but at a reducing rate. The threshold for accommodation of the cyclic
response for this geometry is therefore likely to lie between 20-60kN load amplitude. However, for
subsequent cycles of lower amplitude the displacement does not accumulate past that of the last
60kN cycle, as highlighted in light grey on Figure 33(a).
Page 36/272
(a) overall response (b) small displacements
Figure 33: Comparison of cyclic (CM5 in grey) and monotonic (CM9 in black) response
(D = 0.762m, L/D = 5.25)
Figure 34(a) shows sample load-displacement response curves in clay, during packets of varied
load amplitude, which have been overlaid at the plot origin. This plot shows that as the applied load
is increased there is a change in stiffness and load shape, which is not recovered when the load
amplitude is subsequently decreased (note cycle packets 2 and 4). The path of the change in
stiffness is demonstrated in Figure 34(b), which suggests that the changes induced by a storm event
may result in permanent changes to the subsequent lower amplitude response. It is likely that this
response is related to the on-set of gapping on the active side of the pile. This phenomenon is
further illustrated in Figure 34(c), where subsequent load cycles map on top of one another, both in
terms of stiffness and accumulated displacement.
Page 37/272
(a) Load curve shape (b) Cycle stiffness
(c) post-storm cycles
Figure 34: Evolution of soil response during CM5 test (D = 0.762m, L/D = 5.25)
A detailed analysis of the 2-way small amplitude cyclic testing is developed in the report on the
‘Damping Project’ (AWG, 2016).
Page 38/272
PISA Project – Final Report Section 5 - 1D ‘soil reaction curve’ analysis procedure
5 1D ‘soil reaction curve’ analysis procedure
5.1 Formulation of the method
5.1.1 Structural modelling procedure
The 1D analyses procedure developed during the PISA project retains many of the assumptions that
are fundamental to the conventional ‘p-y’ approach (e.g. the adoption of the Winkler assumption to
specify the soil/structure interaction behaviour and the representation of the pile as a series of 1D
beam elements). It is extended, however, to include additional soil/pile interaction components
(represented in the model by mathematical functions referred to as ‘soil reaction curves’) that have
been found to be significant for monopile foundations with relatively low length-to-diameter ratios.
New forms of mathematical function to represent the soil reaction curves have been developed. The
current modelling procedure is limited to monotonic loading, although it is capable of being extended
to model soil damping (for dynamic analyses) and cyclic loading.
The proposed design approach retains many of the advantages of the traditional ‘p-y’ method (e.g.
fast computation time) while incorporating enhancements to improve the performance of the
modelling approach when used for the design of low length-to-diameter ratio monopiles. A criticism
of the current ‘p-y’ approach is that the fundamental equations (the ‘p-y’ curves) have become
deeply embedded within the various design specifications, with the consequence that the method
has essentially become ‘static’. In developing an improved approach, a key principle that has been
adopted is that the method should be capable of being further refined in the future, as further
experience is gained on site investigation procedures, numerical analysis techniques and the
observed behaviour of installed monopiles.
The proposed 1D analysis procedure is illustrated in Figure 35, for the case where the monopile is
loaded by horizontal force 𝐻𝐺 and moment 𝑀𝐺 applied at the ground level. Figure 35(a) indicates the
assumed components of soil reaction acting on the pile. Figure 35(b) illustrates the 1D structural
model that is used as the basis of the design method; the indicated directions of the displacement,
moments and forces correspond to the sign conventions adopted in this report. The components
p(z,v) and m(z,) (where is the local rotation of the pile cross-section, clockwise positive), denote
the distributed lateral and moment soil reactions respectively. The horizontal force and moment at
the base of the pile are HB and MB respectively.
Page 39/272
Tower
MG
MG
HG HG
Mudline
z v
Distributed
lateral load Lateral soil
reaction
p(z,v)
Monopile
Distributed
moment
m(z,)
Vertical shear
stresses at pile/
soil interface Base shear
force HB(vB) Base moment
MB(B)
Shear force and

moment applied
at the pile base.
(a) assumed soil reactions acting on a monopile (b) analysis model
Figure 35: Key features of the proposed 1D analysis model

In the conventional p-y approach, the soil/pile interaction is represented by a distributed lateral load
applied along the length of the pile. A lateral load soil reaction curve, 𝑝(𝑧, 𝑣) (units of 𝑝: force/length)
which has a similar function to a conventional p-y curve, is adopted in the current model. The
proposed model also includes a distributed moment 𝑚(𝑧, 𝜓) (units of 𝑚: force) acting along the
embedded length of the pile; this distributed moment arises as a consequence of the vertical shear
stresses set up on the pile perimeter by local pile rotation, as shown in Figure 36(a). It is also noted
that significant vertical shear stresses are likely to be generated on the passive side of the pile, for
cases where the pile is loaded close to failure, as a consequence of the wedge-type mechanism that
is expected to develop near to the ground surface (Figure 36(b)). A horizontal force, 𝐻𝐵 (units: force)
developed at the base of the pile and a base moment, 𝑀𝐵 (units force x length) are also included in
the model.
Page 40/272
Rotation,
The vertical shear stress t

f Centre of
varies around the pile
perimeter, leading to a rotation Flow
distributed moment, m. past
dz
Flow
Pile past
Base
shear
(a) Associated with pile rotation (b) Associated with wedge failure
Figure 36: (a) Vertical shear stresses assumed in the model; (b) vertical shear stresses
associate with wedge-type failure mechanism (after Randolph and Gourvenec 2011)
The pile is represented in the 1D model as a line of beam finite elements, based on Timoshenko
beam theory. The local soil deformations (rotation and displacement) are prescribed to conform to
the local pile displacements and rotations along the embedded length of the pile. (It should be
carefully noted that two distinct rotation variables are considered in the Timoshenko beam theory
used to model the pile, namely the overall rotation of the pile, 𝜃, and the rotation of the pile cross-
section 𝜓. In the current approach the local soil rotation is prescribed to conform to the local value of
pile cross-section rotation 𝜓). The soil response is incorporated within the analysis, on the basis of
the Winkler assumption, using appropriate mathematical models for the soil reaction curves
representing the four separate components of soil reaction indicated in Figure 35.
5.2 Formulation of the soil reaction curves
5.2.1 Normalised variables
The development of soil reaction curves for use in the 1D model requires the use of an appropriate
set of normalised variables. Soil reaction curves that are expressed in terms of these normalised
variables can be subsequently scaled for use in a specific design task. The normalised variables
employed in the current modelling procedure are listed in Table 5.
Page 41/272
Table 5: Normalisation of pile reaction components. Note that (i) the values of undrained
shear strength 𝒔𝒖 employed in the normalisation corresponds to the value measured in
triaxial compression and (ii) 𝝈′𝒗𝒊 is the local value of the initial vertical effective stress.
Component Clay normalisation Sand Normalisation

𝑝 𝑝
̅
Distributed load, 𝒑 ′
𝑠𝑢 𝐷 𝜎𝑣𝑖 𝐷
𝑣 𝑣 𝑝𝑎 𝑣 𝐺
̅
Lateral displacement, 𝒗 𝐼 𝐼𝑠 √ ′ [= ′ ]
𝐷 𝑅 𝐷 𝜎𝑣𝑖 𝐷 𝜎𝑣𝑖
𝑚 𝑚
̅
Distributed moment, 𝒎
𝑠𝑢 𝐷 2 𝑝𝐷
𝑝𝑎 𝐺
̅
Pile rotation, 𝝍 𝜓𝐼𝑅 𝜓𝐼𝑠 √ ′ [= 𝜓 ′ ]
𝜎𝑣𝑖 𝜎𝑣𝑖
𝐻𝐵 𝐻𝐵
̅𝑩
Base shear load, 𝑯 ′ 2
𝑠𝑢 𝐷 2 𝜎𝑣𝑖 𝐷
𝑀𝐵 𝑀𝐵
̅𝑩
Base moment, 𝑴 ′ 3
5.2.2 Mathematical form of the soil reaction curves
The soil reaction curves are based on the use of a conic function of the following general form
(where 𝑥̅ refers to a normalised displacement variables in Table 5 and 𝑦̅ is the corresponding load
variable):
𝑦̅ 𝑥̅ 2 𝑦̅ 𝑥̅ 𝑘 𝑦̅
−𝑛 ( − ) + (1 − 𝑛) ( − ) ( − 1) = 0 Equation 2
𝑦̅𝑢 𝑥̅𝑢 𝑦̅𝑢 𝑦̅𝑢 𝑦̅𝑢
for which the real positive roots of 𝑦̅ can be described by:
2𝑐
𝑦̅ = 𝑦̅𝑢 ; 𝑥̅ ≤ 𝑥̅𝑢
−𝑏 + √𝑏 2 − 4𝑎𝑐 Equation 3
𝑦̅ = 𝑦̅𝑢 ; 𝑥̅ > 𝑥̅𝑢
where:
𝑎 = 1 − 2𝑛 Equation 4
𝑥̅ 𝑥̅ 𝑘
𝑏 = 2𝑛 − (1 − 𝑛) (1 + ) Equation 5
𝑥̅𝑢 𝑦̅𝑢
𝑥̅ 𝑘 𝑥̅ 2
𝑐= (1 − 𝑛) − 𝑛 2 Equation 6
𝑦̅𝑢 𝑥̅𝑢
Page 42/272
This function requires the specification of four separate parameters; 𝑥̅𝑢 , 𝑦̅𝑢 , 𝑘 and 𝑛, where 0 ≤ 𝑛 ≤
1 and 𝑥̅𝑢 > 𝑦̅𝑢 ⁄ 𝑘. The general form of the function is plotted in Figure 37(a). As shown in Figure
37(b), a bilinear form of the equation is obtained for a curvature parameter n = 0.
y y
yu yu n=0
k n k
n=1
xu x xu x
(a) General form (b) Extreme values of n
Figure 37: Function forms for soil reaction curves
Note that Equation 3 calculates a positive normalised load based on a positive normalised
displacement. If the input displacement is negative or if the sign convention for a given reaction
dictates that the signs of the displacement and load are different, then it will be necessary to
calculate the load using a positive displacement and to correct the sign of the load according to the
sign of the input displacement.
5.3 Design procedures and guidance
5.3.1 The ‘rule-based’ and ‘numerical-based’ approaches
Appropriate soil reaction curves for the application of the 1D model to a particular design task can be
determined using one of the two alternative procedures indicated in Figure 38. In the ‘rule-based
method’, soil reaction curves are generated using pre-defined mathematical functions, with
parameters that are determined straightforwardly from relevant site investigation data on soil shear
strength and stiffness. Note that the soil reaction functions presented in this report have been
determined from numerical analyses for a specific site, which features a specific soil profile and pile
geometries. The accuracy of the response prediction, when using these soil reaction functions for a
new site, will be dependent on the similarity of the soil profiles and pile geometries. An alternative –
more detailed - approach, referred to as the ‘numerical-based method’, involves the use of 3D finite
element analysis procedures to establish bespoke soil reaction curves for a particular offshore site. It
is anticipated that the rule-based method will be adopted for preliminary design activities. The
‘numerical-based’ method is likely to be applicable to more detailed analyses conducted at later
stages of the design process.
Page 43/272
Detailed strength and

Basic strength and stiffness stiffness parameters from SI
Soil classification
parameters from SI
Soil constitutive model

calibration and FE analysis
Lookup table of parameters

Pile geometry for given soil classification
Soil reaction extraction Array geometry
and parameterisation
Soil reaction curves Loads

Soil reaction curves Loads
Lateral response
prediction Lateral response
prediction
(a) Rule-based method (b) Numerical-based method
Figure 38: Rule-based and numerical-based design procedures

A set of initial rule-based equations for a clay site and a sand site are presented later in this section.
To apply the numerical-based method, a procedure is envisaged in which results obtained from a
suite of detailed 3D finite element calibration analyses of monopile foundation behaviour is used to
calibrate (or ‘train’) the (simpler) 1D finite element model; the 1D model is then used to conduct the
detailed design calculations. The finite element calibration analyses are based on detailed strength
and stiffness data obtained from the site investigation process. These data are used to specify an
appropriate soil constitutive model for the site; the model needs to be sufficiently sophisticated to be
able to reproduce the soil response over an appropriate range of strains. The resulting 3D model is
used to conduct a set of calibration analyses which are based on a set of representative pile
dimensions and loading conditions. These calibration analyses are required to span the design
parameters of interest; the relevant parameters are pile length, 𝐿, pile diameter 𝐷, and load
eccentricity, ℎ. The influence of pile wall thickness, 𝑡, is included explicitly within the 1D model and
variations in this parameter do not need to be explored within the parametric study. Soil reaction
curves are extracted from the results of the calibration analyses; these data are then normalised and
parameterised (i.e. they are fit to the conic function in section 5.2.2). The parameterised forms are
incorporated within the 1D finite element model; the resulting 1D model is used to conduct the
detailed calculations that are required by the design process for geometric and loading variations
that fall within the parameter space used to specify the 3D calibration analyses.
It is important to emphasise the design philosophy that underpins the proposed numerical-based
approach. In the conventional p-y method, the p-y curves are presented as a set of equations within
a design guidance document. This approach means that the form of the curves cannot easily be
developed as new soil constitutive models are devised or new site investigation procedures are
Page 44/272
developed. In contrast, the numerical-based approach provides a procedure in which the soil
reaction curves employed in the 1D model are determined, directly, on a site-specific basis. This
means that the method can evolve with future developments in constitutive modelling and finite
element analysis. Design calculations can be conducted, rapidly and simply, using the calibrated 1D
model; the accuracy of the 1D design calculation is comparable to the accuracy obtained from more
detailed 3D finite element analyses. The proposed design approach comprises a process, and not
just a prescriptive set of equations.
5.3.2 Example soil reaction curves for clay
A set of soil reaction curves (e.g. for use in the rule-based approach) determined for an example
clay site have been developed on the basis of the functional forms set out in Section 5.2.2. A
detailed description of the approach used to determine these curves is given in Appendix D.
5.3.2.1 Calibration analyses.
The soil reaction curves have been determined on the basis of a calibration study using the pile
dimensions and loading conditions specified in Table 6, which were selected as described in
Appendix D4.
Table 6: Geometry and loading adopted in the 3D finite element calibration

Analysis D h L t
h/D h/L L/D D/t
reference (m) (m) (m) (mm)
C1 10 50 5 20 2.5 2 91 110
C2 10 150 15 20 7.5 2 91 110
C3 10 50 5 20 2.5 2 125 80
C4 10 50 5 60 0.83 6 91 110
C5 10 150 15 60 2.5 6 91 110
C6 5 25 5 10 2.5 2 45 110
C7 5 25 5 10 2.5 2 83 60
C8 5 25 5 30 0.83 6 45 110
C9 5 75 15 30 2.5 6 45 110
C10 7.5 37.5 5 15 2.5 2 68 110
C11 7.5 37.5 5 45 0.83 6 68 110
5.3.2.2 Soil reaction curves for clay
Site-specific soil reaction curves have been determined from a 3D finite element calibration exercise
conducted for an example Bolders Bank clay site, as described in Appendix D2. The soil profiles
adopted in this calibration process are indicated in Figure 39 and Figure 40.
Page 45/272
(a) (b)
(c) (d)
Figure 39: Soil profiles for clay calibration analyses
Page 46/272
(a) (b)
Figure 40: Assumed maximum stiffness profile and stiffness degradation curve for the clay
The resulting parameters for the soil reaction curves are specified in Table 7 (in terms of normalised
depth 𝑧⁄𝐷).
5.3.3 Consideration of gap opening for piles in stiff clay
The interface elements employed in the finite element model at the pile/soil boundary allow the
possibility of gap formation on the active side of the pile, near to the ground surface. Since the
detailed performance of the pile is strongly influenced by near surface soil/structure interaction
effects, inclusion of the possibility of gapping was a key element of the modelling process.
In the development of finite element models used for assisting the design of the test piles, any gap
formed around the pile was initially assumed to be filled with air (on the basis that the tests were
conducted onshore with the water table at some depth below the surface). This assumption was also
employed for the initial calibration analyses, and used as a basis for the initial soil reaction curve
parameters. The process of deriving these parameters and the validation of the soil reaction curves,
when interpolating within the calibration parameter space is demonstrated in Appendix D.
However, in an offshore environment, any gap on the active side of the pile will fill with water and, as
a consequence, additional lateral loading will be applied to the monopile. As an interim step the air-
gap analyses were corrected manually to account for a water filled gap. A second phase of
numerical analyses were then conducted for all geometries of the calibration study, with hydrostatic
pressure applied to the faces of the gap opening up between the pile and soil, analogous to a water-
filled gap. This approach is adopted at the end of Appendix D, to provide soil reaction curve
parameters for a water filled gap, for offshore problems; these are listed as the ‘water-filled gap’
expressions in Table 7. There was little difference between the air-gap corrected analyses and the
true water filled gap analyses.
Page 47/272
5.3.4 Clay till parametric curve expressions
Table 7: Parameters for normalised soil reaction curves in an idealised clay till as defined in
section 5.3.2, calibrated within the parameter space defined in Table 6 (𝟐 < 𝑳⁄𝑫 < 𝟔, 𝟓 < 𝐃 <
𝟏𝟎, 𝟓 < 𝒉⁄𝑫 < 𝟏𝟓, 𝟔𝟎 < 𝑫⁄𝒕 < 𝟏𝟏𝟎). Equations are valid for 𝟎 < 𝒛⁄𝑫 < 𝟔.
Soil reaction Parameter Air-filled gap Water-filled gap

component expression expression
(Table 22) (Table 25)
Distributed Ultimate strain, 𝑣̅ pu 200 200
lateral load, 𝒑 𝑧 𝑧
Initial stiffness, kp −1.11 + 8.17 −1.10 + 8.12
𝐷 𝐷
𝑧 𝑧
Curvature, np −0.07 + 0.92 −0.05 + 0.92
𝐷 𝐷
𝑧 𝑧
Ultimate reaction, 𝑝̅u 11.66 − 8.64𝑒 (−0.37𝐷) 10.21 − 7.22𝑒 (−0.33𝐷)
Distributed Ultimate rotation, 𝜓̅mu 10 10
moment, 𝒎 𝑧 𝑧
Initial stiffness, km −0.12 + 0.98 −0.11 + 0.97
𝐷 𝐷
Curvature, nm 0 0
𝑧 𝑧
Ultimate moment, 𝑚
̅u −0.05 + 0.38 −0.04 + 0.38
𝐷 𝐷
Base shear, 𝑯𝑩 Ultimate strain, 𝑣̅ Hu 300 300
𝐿 𝐿
Initial stiffness, kH −0.32 + 2.58 −0.32 + 2.56
𝐷 𝐷
𝐿 𝐿
Curvature, nH −0.04 + 0.76 −0.03 + 0.74
𝐷 𝐷
𝐿 𝐿
̅ Bu
Ultimate reaction, 𝐻 0.07 + 0.59 0.07 + 0.60
𝐷 𝐷
Base Ultimate rotation, 𝜓̅Mu 200 200
moment, 𝑴𝑩
𝐿 𝐿
Initial stiffness, kM −0.002 + 0.19 −0.003 + 0.20
𝐷 𝐷
𝐿 𝐿
Curvature, nM −0.15 + 0.99 −0.16 + 1.01
𝐷 𝐷
Ultimate reaction, 𝐿 𝐿
̅ Bu −0.07 + 0.65 −0.08 + 0.65
𝑀 𝐷 𝐷
5.4 Validation of the design approach for the assumed clay profile
5.4.1 Comparison of 1D model with the 3D finite element calibration data
An exercise has been conducted to compute the response of each of the calibration analyses
specified in Table 6 using the 1D model (using the ‘air-filled gap’ soil reaction curves specified in
Table 7). This provides an indication of the ability of the 1D model to reproduce the calibration data
on which it is based. Comparisons between the performance of the C1 (short) and C4 (long) piles
computed using the 3D finite element model and the calibrated 1D model are shown in Figure 41.
Also shown is the performance computed using the API p-y method (described in Appendix F4).
Page 48/272
Whilst observable error occurs in the predictions of the short C1 pile response determined using the
1D model, the overall behaviour is well represented with accurate predictions of the initial stiffness,
ultimate response and intermediate curvature. This pile provides a stern test of the 1D modelling
process, because the ultimate lateral soil reaction is reached over a significant length of the pile and
therefore a good approximation of the reaction curvature is required to obtain an accurate prediction
of the shape of the overall load/displacement response. The 1D model reproduces closely the
response computed by the 3D model for the (longer) C4 pile. In all cases the calibrated 1D model is
seen to provide a closer match to the 3D finite element results than the API/DNV approach.
(a) C1 (D = 10m, L = 20m) ultimate response (b) C1 (D = 10m, L = 20m) small displacement
(c) C4, (D = 10m, L = 60m) ultimate response (d) C4, (D = 10m, L = 60m) small displacement
Figure 41: Comparison of ground load-displacement response determined from the 3D

calibration analyses and the corresponding results obtained when adopting parameterised
expressions in the 1D model.
The quality of fit metric (described in section 4.4) is shown in Figure 42 for both the ultimate
response and the small displacement response. Note that the small displacement threshold for
Page 49/272
th
numerical simulations is D/10,000, which corresponds to approximately 1/200 of the ultimate load,
and was selected to provide a measure of the response stiffness at displacements relevant to the
dynamic response under fatigue and service loading regimes. The average accuracy of prediction at
large displacements of η0.1D = 94%, is significantly higher than that achieved by the API/DNV
approach (41%). The API/DNV approach is especially poor in the simulation of shorter piles. For the
small displacement response, the mean quality of fit metric for the calibrated 1D model is ηsd = 90%.
This exceeds by a very considerable margin the quality of fit that is achieved using the API/DNV
approach.
D
L/D
M/HD
D/t
(a) ultimate response (up to vG = 0.1D)
D
L/D
M/HD
D/t
(b) small displacement response (up to vG = D/10,000)
Figure 42: Accuracy metrics comparing load-displacement shape accuracy of the 1D model
using parameterised soil reaction curves and API/DNV soil reaction curves, when compared
to 3D finite element simulations
Page 50/272
5.4.2 Predictive capability of the 1D model
The 1D model is intended for use as a predictive tool. Once the model has been trained (using the
3D calibration data), it can be used to determine the performance of a monopile for arbitrary values
of the geometric and loading parameters that lie within the parameter space employed in the 3D
finite element calibration analyses.
To test the predictive capability of the 1D model, two additional test analyses (D1 and D2) have
been conducted using the 3D finite element model. The geometric configurations adopted for these
tests cases are specified in Table 8 and were selected to fall within the parameter space adopted for
the initial calibration analyses, as shown in Figure 43. The analyses were conducted for the
representative clay profiles in Figure 39 and Figure 40.
Table 8: Geometry and loading adopted in the 3D finite element design geometries
Analysis D h L t
h/D L/D D/t
D1 7.5 37.5 5 22.5 3 68 110
D2 8.75 87.5 10 35 4 91 97
(a) MG/HGD (b) MG/HGL
Figure 43: Parametric geometry space for the field test, full scale calibration analyses and
design scenarios
The load-displacement responses computed using the 3D finite element model and the calibrated
parametric 1D model are shown in Figure 44 and Figure 45. The results indicate a close fit between
the 1D and 3D finite element data over the full range of applied loading (Figure 44(a) and Figure
45(a)). The small displacement response computed using the 1D model is seen to be almost
identical to the response computed using the 3D finite element model (Figure 44(b) and Figure
45(b)). Also shown for comparison are the corresponding results obtained using the API p-y method.
The close fit that is obtained between the 1D and 3D finite element results confirms the ability of the
calibrated 1D model to accurately predict pile behaviour for geometries within the parameter space.
Page 51/272
(a) Overall response (b) Small displacements
Figure 44: Comparison of load-displacement response from 3D FE, 1D model simulations using
parameterised soil reactions and the existing API/DNV approach for the D1 design case (D =
7.5, L/D = 3)
8.75, L/D = 4)
Page 52/272
5.5 Example soil reaction curves for sand
A set of soil reaction curves (e.g. for use in the rule-based approach) determined for an example
dense sand site (with relative density Dr = 75%) have been developed on the basis of the functional
forms set out in Section 5.2.2. A detailed description of the approach used to determine these
curves is given in Appendix E.
5.5.1 Calibration analyses.
The soil reaction curves have been determined on the basis of a calibration study using the same
pile dimensions and loading conditions as those described in section 5.3.2.1 and specified in Table
6.
5.5.2 Soil reaction curves for sand
Site-specific soil reaction curves have been determined from a 3D finite element calibration exercise.
The derived constitutive model is based on the Dunkirk site model, provided in Appendix A4, and
modified to represent an idealised and fully submerged site, as described in Appendix E2.
The resulting parameters for the soil reaction curves are specified in Table 9 (in terms of normalised
depth 𝑧⁄𝐷).
Page 53/272
Table 9: Parameters for normalised soil reaction curves in Dunkirk sand, calibrated within the
parameter space defined in Table 6 (𝟐 < 𝑳⁄𝑫 < 𝟔, 𝟓 < 𝐃 < 𝟏𝟎, 𝟓 < 𝒉⁄𝑫 < 𝟏𝟓, 𝟔𝟎 < 𝑫⁄𝒕 < 𝟏𝟏𝟎)
Equations are valid for 𝟎 < 𝒛⁄𝑫 < 𝟔.
Soil reaction component Parameter Expression

Distributed lateral load, 𝒑 Ultimate strain, 𝑣̅ pu 53.1
𝑧
Initial stiffness, kp −0.85 + 7.46
𝐷
Curvature, np 0.944
𝑧
Ultimate reaction, 𝑝̅u −10.18 + 21.61
𝐿
Distributed moment, 𝒎 Ultimate rotation, 𝜓̅mu 20
Initial stiffness, km 20
Curvature, nm 0
𝑧
̅u −0.05 + 0.21
𝐿
Base shear, 𝑯𝑩 𝐿
Ultimate strain, 𝑣̅ Hu −0.29 + 2.31
𝐷
𝐿
Initial stiffness, kH −0.38 + 3.02
𝐷
𝐿
Curvature, nH −0.05 + 0.94
𝐷
𝐿
̅ Bu
Ultimate reaction, 𝐻 −0.07 + 0.62
𝐷
Base moment, 𝑴𝑩 Ultimate rotation, 𝜓̅Mu 50
Initial stiffness, kM 0.29
Curvature, nM 0.89
𝐿
̅ Bu
Ultimate reaction, 𝑀 −0.05 + 0.38
𝐷
5.6 Validation of the design approach for the assumed sand profile
5.6.1 Comparison of 1D model with the 3D finite element calibration data
In a similar fashion to the clay analysis, an exercise has been conducted to compute the response of
each of the calibration analyses specified in Table 6 using the 1D model (with the soil reaction
curves specified in Table 7). This provides an indication of the ability of the 1D model to reproduce
the calibration data on which it is based.
Comparisons between the performance of the C1 (short) and C4 (long) piles computed using the 3D
finite element model and the calibrated 1D model are shown in Figure 46. Also shown is the
performance computed using the current API p-y method (described in Appendix F4).
Page 54/272
For both the short C1 and long C4 geometries, the 1D parametric approach accurately predicts the
overall pile behaviour up to the ultimate response as well as at small displacements. In all cases the
calibrated 1D model is seen to provide a closer match to the 3D finite element results than the
API/DNV approach, despite the fact that the API/DNV approach for this dense sand site achieves
significantly higher accuracy than that seen for the stiff clay analysis.

The quality of fit metric (described in section 4.4) is shown in Figure 47 for both the ultimate
response and the small displacement response. Note that the small displacement threshold for
th
numerical simulations is D/10,000, which corresponds to approximately 1/200 of the ultimate load,
and was selected to provide a measure of the response stiffness at displacements relevant to the
dynamic response under fatigue and service loading regimes. The 1D parametric model consistently
Page 55/272
predicts the shape and magnitude of the ultimate response with an average accuracy of η0.1D = 96%,
which represents a small improvement over the existing API/DNV approach (η0.1D = 88%). There is a
greater disparity in the prediction of the small displacement response, with accuracy metrics of ηsd =
92% and ηsd = 43% for the 1D parametric and API/DNV approaches respectively. Whilst the 1D
parametric approach sees a marginal reduction in small displacement accuracy for the shorter pile
geometries, the API/DNV approach sees a significant reduction, suggesting that the existing design
methods are not well conditioned for predicting the dynamic pile response for shorter geometries.
D
L/D
M/HD
D/t
D
L/D
M/HD
D/t
Page 56/272
5.6.2 Predictive capability of the 1D model
In a similar exercise to that carried out for the clay analysis, two additional test analyses (D1 and D2)
have been conducted using the 3D finite element model. The geometric configurations adopted for
these tests cases are specified in Table 8 and were selected to fall within the parameter space
adopted for the initial calibration analyses, as shown in Figure 43.
parametric 1D model are shown in Figure 48 and Figure 49. The results indicate a close fit between
the 1D and 3D finite element data over the full range of applied loading (Figure 48(a) and Figure
49(a)). The small displacement response computed using the 1D model is seen to be almost
identical to the response computed using the 3D finite element model (Figure 48(b) and Figure
49(b)). Also shown for comparison are the corresponding results obtained using the API p-y method.
7.5, L/D = 3)
Page 57/272
8.75, L/D = 4)
Page 58/272
5.7 Routes for application
5.7.1 Application of the rule-based method
The rule-based equations given in the previous section are presented in a non-dimensionalised
form. To apply them to a specific design task, data are needed on the variation with depth of the
small-strain shear modulus 𝐺𝑜 at the site of interest. In addition, for a clay soil, values of triaxial
compression shear strength, 𝑠𝑢 , are required. From the constitutive formulation of the sand material
studied in this project, the G0 profile also reflects the initial density profile of the deposit and a further
input is needed on the variation of in situ vertical effective stress with depth. However, this approach
could be expanded to using the peak or critical state angle of shearing resistance, depending on the
chosen constitutive formulation.
Values of 𝐺𝑜 are conveniently measured using seismic cone tests. Data on undrained shear strength
may be determined either using cone penetration testing (based on appropriate values of 𝑁𝑘𝑡 ) or by
triaxial testing conducted on undisturbed samples removed from the site. Ideally, a profile of
undrained shear strength at the site of interest is developed on the basis of a combination of data
from triaxial testing and CPT data. Data on in situ vertical effective stress may be estimated on the
basis of soil unit weights that are based on assessments of relative density (e.g. determined from
the results of cone penetration tests).
5.7.2 Application of the numerical-based method
The numerical-based method requires the use of a 3D finite element model of the monotonic
behaviour of a monopile foundation subjected to a lateral load. 3D finite element analyses conducted
for the purpose of determining soil reaction curves should make use of the symmetry of the problem
to minimise the size of the mesh. Boundaries should be placed sufficiently far from the monopile to
ensure that they do not affect the results. (It may be appropriate to conduct some initial numerical
studies to investigate the influence of the proximity of the mesh boundaries on the computed
results). The pile should be modelled using shell elements with zero-thickness interface elements
placed between the soil and the pile. The interface elements provide a means of specifying the
behaviour of the soil/pile interface as well as providing a means of extracting the soil reaction curves
from the analysis.
Careful selection is needed of the soil constitutive model if reliable results are to be obtained. It may
be necessary (in cases where an advanced soil model is selected) to employ a finite element
platform (such as ABAQUS or Plaxis) in which user-specified constitutive models can be employed.
For clay, the soil is expected to behave in an undrained manner; soil models in this case can be
based either on a total stress approach (in which the site-specific variation of undrained shear
strength is incorporated directly in the analysis) or an effective stress approach (in which the
effective stress soil parameters need to be defined, together with appropriate representations of the
stress history and the horizontal effective stresses). For sands, an effective stress model is required
in which the volumetric characteristics of the soil (dilation / contraction) are represented at an
appropriate level of detail.
For the design of a monopile to achieve a specified stiffness (e.g. to determine the natural
frequencies of a wind turbine at low levels of excitation) it may be sufficient to adopt an elastic model
for the soil. A simple approach might be to assume that the soil is linearly elastic with a variation of
shear modulus with depth determined from the results of seismic cone tests. A more sophisticated
Page 59/272
approach would be to assume a non-linear elastic model incorporating an appropriate degradation of

shear modulus with shear strain level. In this latter case, triaxial testing incorporating local strain
measurement is required to determine the appropriate stiffness degradation curve.
For cases where design calculations are required in which the monopile is loaded to failure, the 3D
calibration analyses need to be based on an appropriate elastic-plastic model for the soil. For clay,
the generalised Modified Cam clay model in Tsiampousi et al. (2013) provides one possible
modelling approach. For sand, the bounding surface model in Papadimitriou & Bouckovalas (2002)
and Taborda et al. (2014) could be used. The use of advanced soil models of this sort inevitably
means that further detail on soil behaviour is required for calibration purposes. Variations with depth
of 𝐾𝑜 and pore pressure are invariably required. In addition, further detailed laboratory tests (e.g.
triaxial extension tests) may be needed to determine the required model parameters. Prior to
conducting detailed 3D calibration analysis, it is advisable to conduct some relatively simple
analyses (e.g. a laterally-loaded monopile using a relatively simple built-in soil model)
Books such as Potts and Zdravkovic (1999 and 2001), Brinkgreve (2013) provide background
information on the various modelling approaches that are available.
5.7.3 Design procedures
An approach for applying the new design method in design is shown in Figure 50. Firstly this
illustrates that the “rule-based” method could be used for feasibility design, in a similar way to the
use of current design approaches. Preliminary site investigation data would be gathered to feed into
the concept design, again using the “rule-based” method. This would signal the start of the
geotechnical laboratory testing program which would provide input to the development of
appropriate soil constitutive models for use in the 3D finite element studies.
The concept design and site investigation data would identify the parametric space that the
numerical simulations should span, from which would be drawn the more detailed design
expressions. These expressions would allow the definition of a range of pile responses by
interpolation within the parameter space. The design examples shown in Appendices D and E
provides a detailed description of this process, showing that the approach results in a simplified
model that is both quick to run while reproducing very closely the likely pile response. It is noted that
design process for an actual wind farm will benefit from a much tighter definition of the parameter
space (i.e. more limited range of pile diameters), and so should be even more reliable. These site
specific design expressions applied to the simplified 1D model would allow the site specific
optimisation of each monopile at each location.
Finally, it may be appropriate to check one or two final designs with a specific finite element
calculation, to fully close out the design loop. Logically this approach allows the complexity of the
design approach to develop as the geotechnical knowledge of the site increases.
It should be noted that the evolution of the design procedures from the existing ‘p-y’ method to the
more robust and complete design approach described in this report, will have differing levels of
impact on the ULS, SLS and FLS design cases. Careful thought should therefore be given to the
confidence intervals of soil properties and partial factors applied in each case, to ensure a safe, but
not over-conservative design.
Page 60/272
Status of site knowledge Design procedure Outcome

Design using rule based
Desktop study Feasibility design
method
Rule-based
method Preliminary geotechnical survey Design using rule based
Preliminary concept design
and ground modelling method
High quality sampling and Identify example design

laboratory testing positions that typify site and
foundation geometric
envelope.
Soil constitutive model

Advanced laboratory testing
calibration
3D FEA of design positions and

expected geometric envelope
Numerical-based Calibrated soil reaction curves

Soil reaction extraction and
for site geology and expected
method parameterisation
foundation envelope
Design optimisation for loads,

geometry using soil reaction Detailed concept design (FEED)
curves
Final geotechnical survey at Final design for all position

Final design
turbine locations using soil reaction curves
Increasing knowledge of 3D FEA check of a number of

site geotechnical design positions Increasing design
properties refinement
Figure 50: Application of the new methods in the design process
5.8 Limitations
There are limitations to the application of the design approaches reported in this document. At the
outset the project partners agreed the following:
(a) To consider two specific soil profiles, as representative soil profiles for offshore wind farm
sites. Application to other soil profiles, of the specified “rule based” methods, will require
further analyses and engineering judgment to be applied. Additional work will be required to
further develop the “numerical based” methods for other soil profiles, with this report
outlining a structured design process through which the detailed design equations can be
obtained.
(b) To consider monotonic loading only. This provided a significant focus of the work through
the project with the result that the initial stiffness and ultimate capacity have been much
more clearly defined. Although some of the field testing focused on cyclic loading there has
been limited interpretation of the cyclic response, as this is the focus of on-going research
efforts.
(c) To neglect the effects of installation on the pile response. The new design method is based
on the results from an application of the finite element method, which represents the current
state-of-the-art in numerical modelling. However, modelling of installation effects still
represents a considerable challenge, and is beyond the scope of the project. Differences
between field test results and the numerical calculations could highlight what such effects
might be, and whether they are significant. Based on the work completed it appears that
installation in sandy soils requires further consideration.
These limitations were necessary in order to focus the academic work on delivering engineering
solutions to the partners in the timescale required by the partners. Engineering judgment will be
required to use these methods in practice, as is the case with the application of any design method.
Page 61/272
PISA Project – Final Report Section 6 - Achievements of the Project and Conclusions
6 Achievements of the Project and Conclusions
The aim of the PISA project was to investigate and develop improved design methods for laterally
loaded piles, specifically tailored to the offshore wind sector. The approach adopted by the
Academic Workgroup (AWG) was to validate 3D numerical models using medium scale field tests.
These models were then used to investigate the performance of the existing p-y design method and
to form the basis for developing an improved method capable of accurately capturing the behaviour
of laterally loaded piles under load cases critical to design.
The work of the PISA project supports the following high level findings:
 The API/DNV ‘p-y’ method provides an unsatisfactory basis for the design of wind turbine
monopile foundations, especially those with low L/D ratios and for stiff low plasticity clay
soils where the stiffness and capacity of realistic geometries appears under-predicted. The
stiffness of response for monopiles in sand is not adequately predicted by existing methods.
 A new approach is presented in this report in which the traditional p-y approach is enhanced
by the inclusion of additional soil reaction curve components (distributed moment, horizontal
base force and base moment). A process is outlined in which 3D finite element analyses are
used to calibrate the soil reaction functions, which can then be applied, within a simplified
1D model, to pile geometries and soil conditions that fall within the calibrated parameter
space. This approach benefits from the computational speed of the traditional p-y approach
while retaining a comparable accuracy to the underlying 3D finite element model. The
application of this design approach indicates potential for a significant reduction in design
conservatism and substantial savings for selected design scenarios.
 A medium scale field testing campaign was carried out in which a total of 28 piles were
tested with varied diameter, length and wall thickness with both monotonic and cyclic loads
and with up to 130 simultaneous instrument measurements of pile and soil response. These
field tests, and the supporting site characterisation, delivers a new industry standard
database against which design models in clay and sand may be compared, developed and
validated.
 Adoption of the PISA design approach is likely to result in reduced conservatism in monopile
design. It is important that as these methods are adopted, consideration is given to the
realistic design drivers of the foundations, including effects such as cyclic degradation of
strength and stiffness and accumulated deformation.
There are a number of areas where the PISA project has exceeded the original expectations:
 The project has established itself with significantly more project partners than was originally
anticipated. As a project, it has been supported by the vast majority of developers building
offshore windfarms in the UK and in Europe;
 The existence of the PISA project, and the developments that are already in the public
domain, have been recognised internationally, with companies and academic institutions
worldwide (including USA, China, Australia) expressing an interest in becoming associated
with the project;
 The project has established a world leading and successful academic collaboration;
 The project has seen the completion of extremely high quality reference pile loading tests,
the data from which are already providing technical input to a number of academic research
projects within the AWG, and are likely to continue to do so in years to come;
 Within the timescale of the project, the development of the new design approaches has
impacted on design practices being adopted by some of the project partners;
Page 62/272
PISA Project – Final Report Section 6 - Achievements of the Project and Conclusions
 The project has brought foundation designers from across the offshore wind industry
together and provided an effective technical forum for them to discuss and generate ideas
for reducing the cost of offshore wind energy;
 The project has interfaced with certification agencies and is indirectly responsible for
changes to the DNV design code for offshore wind DNV-OS-J101;
 The project originally set out to concentrate on monotonic loading. Due to additional funding
raised during the course of the project, it was possible to complete a significant quantity of
cyclic testing, which will provide input into the development of the modelling of cyclic
loading; and
 Field testing of piles at varied loading rates has identified rate effects, particularly for stiff low
plasticity clay soils, which indicate that the capacity under wave or impulse loads may be
higher than under relatively steady wind loads.
At the outset of the project, the supporting partners agreed that the scope of PISA should be
focused to maximise the quality of and confidence in the improved design methods. In particular, the
decision was made to:
 Focus on two sites, one with a homogeneous clay profile (i.e. Cowden clay) and one with a
homogeneous sand profile (i.e. Dunkirk sand). Layered soils have not been considered.
 Focus on understanding the monotonic behaviour.
 Focus on piles with limited vertical loading component.
 Neglect the effect of pile installation on the subsequent pile response.
 Ignore effects of scour and scour protection.
Despite the limitations of the work of the PISA project represents a significant improvement in design
methods for offshore wind turbine monopiles and will lead to a lowering of the cost of wind energy.
Page 63/272
PISA Project – Final Report Appendix A - Site investigation, laboratory testing and derivation of soil
parameters for numerical modelling
Appendix A Site investigation, laboratory testing and derivation of

soil parameters for numerical modelling
Finite element analyses of the PISA test piles at the Cowden and Dunkirk test sites, reported in the
draft design methodologies for clays and sands (AWG 2014a, AWG 2014b respectively), were
based on soil parameters derived from historic site investigation data reported in the literature. The
main reference for characterising the glacial clay tills at Cowden was that of Powell & Butcher
(2003), which summarises extensive field and laboratory site investigations performed on this
deposit. However, the reported experimental results, and in particular those obtained through
laboratory element testing, pre-date contemporary experimental techniques and lack accurate
measurements. In a similar manner, the key references for Dunkirk sand (Chow, 1997; Kuwano,
1999) lacked information for the derivation of Dunkirk sand parameters within the critical state
framework of the sand constitutive model used in the FE analyses. Consequently, a limited data-set
of new laboratory experiments on Dunkirk sand, which formed part of Aghakouchak’s (2015) PhD
thesis on offshore driven piles, which was not related to the PISA project, were used for derivation of
the Dunkirk sand parameters for FE modelling. This appendix summarises the new field and
laboratory testing campaigns conducted in conjunction with the large-scale pile testing at both sites,
with the purpose of enabling more accurate derivation of soil parameters for FE modelling and the
formulation of the final design methodologies. Derivation of final soil parameters is presented for
each site.
A1 Cowden till characterisation

A1.1 New experimental programmes
The sampling of Cowden till was conducted in June 2014 by Concept Ltd, using the Geobor-S rotary
sampling technique. The four rotary boreholes attempted extended down to a maximum depth of
around 13 m, of which only about 5 m of the retrieved 100 mm diameter core was considered of
sufficient quality for high-quality mechanical testing in the laboratory. These samples formed the
basis of new laboratory testing programmes at ICL and GEOLABS Ltd., conducted for the purpose
of deriving revised Cowden clay till input parameters for FE modelling. At the same time, new shear
wave velocity measurements were made on site using in-situ Seismic Cone Penetration (SCPT)
techniques, which were also utilised in the recalibration of the Cowden till constitutive model.
A number of in-situ Cone Penetration tests (CPT) were performed during the installation of the piles
in October/November 2014, which indicated that the soil’s undrained strength in compression down
to 2-3 m depth may not have been as high as originally derived from the historic information. An
additional sampling campaign was organised in January 2015, this time with large block samples (18
in total) taken from trial pits excavated by ESG at depths down to about 3 m. A number of Hand
Shear Vane (HSV) tests to about 1 m depth complemented this sampling.
The interpreted new soil test data became available for recalibration of the Cowden till soil model in
May 2015 (Ushev et al., 2015), comprising the completed laboratory testing programme on the
Geobor-S rotary samples, as well as in-situ SCPT, CPT and HSV data. The recalibration of the
model and the reanalyses of test piles were completed by June 2015, before the field testing of the
large diameter (2.0 m) test piles commenced. Laboratory testing of the block samples commenced
in summer 2015. Consequently, this data was not available for the revision of the Cowden soil
parameters for the FE modelling.
Page 64/272
A1.2 Testing by GEOLABS

GEOLABS were initially commissioned to establish whether the best quality Geobor-S samples
taken at Cowden were suitable for advanced testing. A limited set of triaxial stress path tests was
performed between July and September 2014 on 100 mm diameter samples. The data showed
credible values of water content, stiffness and shear strength of the glacial clay till that were broadly
consistent with earlier work. It was therefore concluded that the quality of samples extracted in June
2014 was suitable for research testing, but recognising that the volume of available core would be
insufficient for the whole PhD programme. A summary of GEOLABS triaxial stress path tests is
given in Table 10, together with those performed at ICL by the PhD student Emil Ushev.
GEOLABS were further commissioned to undertake additional index and oedometer testing to
complement the advanced triaxial stress path testing at ICL, necessary for the revision of soil
parameters as input to FE modelling. Table 10 and Table 11 summarise the two groups of tests.
Table 10: Summary of index testing carried out to May 2015
Chemical tests
Depth (m) LL (%) PL (%) PI Organic (%) Carbonate (%) Resource
0.74 60 23 37 0.7 2.7 GEOLABS

3.50 35 17 18 1.3 6.6 GEOLABS
5.93 34 16 18 1.3 12 GEOLABS
7.30 34 16 18 1.5 11 GEOLABS
10.15 35 15 20 1.6 12 GEOLABS
1.53 44 18 26 GEOLABS
3.61 35 16 19 GEOLABS
5.99 34 15 19 GEOLABS
8.16 35 15 20 GEOLABS
10.04 34 14 20 GEOLABS
1.5 35.4 18.09 17.31 ICL
3.5 33.4 ICL
3.7 37.3 16.72 20.58 ICL
5.7 32.3 14.20 18.10 ICL
8.2 32.7 16.52 16.18 ICL
10.1 34.6 16.62 17.98 ICL
Table 11: Summary of oedometer tests and compression indices, by GEOLABS
Depth (m) Intact Cc Reconstituted Cc*

0.74 0.279 0.387
3.50 0.187 0.221
5.93 0.143 0.221
7.30 0.145 0.237
10.15 0.163 0.315
Page 65/272
A1.3 Testing at Imperial College London (ICL)

The new programme of Cowden till testing started at the ICL Geotechnics laboratory in mid-2014,
with a total of just 3.1 m good quality core being available for testing. The work forms part of the
three-year PhD study of Emil Ushev, which commenced in January 2014 as part of the PISA project.
The triaxial tests involved 38 and 100 mm diameter samples, with height to diameter ratio of 2, in
ICL designed automated stress path apparatus. All stress path cells have been instrumented to
measure local strains and, for the 100 mm diameter samples, mid-height pore pressures and shear
wave velocities. The latest equipment capabilities are broadly as described in the recent PhD
studies by Hosseini-Kamal (2012) and Al-Haj (2014) on clay testing. Large diameter cells offer better
instrumentation resolution and more representative element volumes of the stony Cowden till.
However, the 38 mm samples consume less core and are quicker to consolidate, therefore enabling
a larger number of tests to be performed. In addition, the 38 mm cells were more readily available
for the PISA project in 2014, with two 100 mm cells being used until May 2015. Consequently, a mix
of both sample sizes was involved in the testing and this strategy has continued in subsequent
experiments, with a difference that four 100 mm cells are now being utilised.
The experimental procedure involved consolidation of all samples to estimated K 0 initial stresses. A
rest period was then allowed until creep strains diminished to very low residual strain rates.
Monotonic undrained (U) shearing followed in both compression (C) and extension (E) modes. The
use of mid-height pore pressure probes ensured that consolidation processes were completed
before shearing. The axial strain rate of 5% per day during shearing was chosen to ensure that high
quality effective stress and stress-strain data can be recorded with the available equipment.
A summary of all ICL and GEOLABS triaxial stress path testing to May 2015 is given in Table 12.
The code for each test, in order of appearance of each symbol, identifies the following:
C Cowden site;
R or B rotary or block samples (only rotary utilised in testing programme to May 2015);
38 or 100 sample diameter in millimetres;
K or I K0 or isotropic consolidation;
U or D undrained or drained shearing (only undrained in testing programme to May 2015)’
C or E compression or extension shearing mode;
X.Y sample depth below ground surface in metres.
The letter ‘G’ at the beginning of the code signifies a test performed by GEOLABS. In total, the
results of 20 triaxial compression (of which 4 were by GEOLABS) and 7 triaxial extension tests were
available for the recalibration of the Cowden soil model. The samples covered the initial stresses in
the deposit from 0.5 m to 11.5 m depth.
A1.4 Cowden geotechnical profile

The new site investigation broadly confirmed the Cowden geotechnical profile derived from the
historic information and reported by AWG (2014a). The summary of index properties in Table 10,
from tests regarding water contents, Atterberg limits and chemical testing, confirm that the soil in the
profile is mainly a low plasticity glacial clay till down to around 12 m depth, below which a sand layer
is encountered. This sand has not been tested as the maximum depth of the test piles is 10.5 m.
The upper part of the till is weathered, penetrating to around 4.5 m.
Sample preparation for triaxial testing confirmed the stony nature of the deposit, as well as slight
sub-vertical fissuring, particularly in superficial layers. Figure 51 shows two examples of clay till
Page 66/272
samples from about 2.0 m depth which fell apart during trimming, with sample (b) being an
exceptional case of a particularly large stone inclusion. Generally, the intensity of stone inclusion
was particularly high at this elevation and not representative of other depths of the deposit.
Table 12: Summary of Cowden triaxial tests to May 2015
Test code Pre-shear Pre-shear Pre-shear

p’0 (kPa) v’ (kPa) K0
CR38KUC0.5 19.3 14.5 1.5

CR38KUC1.0 27.7 20.8 1.5
CR38KUC1.5 35.5 26.6 1.5
GCR100KUC1.6 35 26.2 1.5
CR100KUC2.0 46.4 34.8 1.5
CR100KUC2.5 54.8 41.1 1.5
CR38KUC3.0T 62.4 46.8 1.5
GCR100KUC3.7 70 52.5 1.5
CR38KUC3.7 70.2 52.6 1.5
CR100KUC5.0 92.7 69.5 1.5
CR38KUC5.0 76.2 57 1.5
CR38KUC5.4 96.6 72.5 1.5
GCR100KUC6.1 99.6 74.7 1.5
CR38IUC8.0 113.5 113.5 1.0
CR38IUC8.0 113.8 113.8 1.0
CR100IUC8.2 121 121 1.0
CR38IUC10.0 156.2 156.2 1.0
CR38IUC10.0 153.1 153.1 1.0
GCR100IUC10.0 161 161 1.0
CR38IUC11.5 198 198 1.0
CR38KUE0.5 19.7 14.7 1.5
CR38KUE1.0 27.5 20.6 1.5
CR38KUE3.4 65.9 49.4 1.5
CR100KUE4.5 80.2 60.2 1.5
CR38KUE5.4 97.5 73.1 1.5
CR38IUE7.5 113.5 113.5 1.0
CR38IUE10.0 153.6 153.6 1.0
Page 67/272
200 mm
200 mm
Exposed stones in samples
(a) (b)
Figure 51: Examples of Cowden till samples
A2 Soil constitutive model for Cowden till

The constitutive model for simulating the behaviour of the glacial clay tills at Cowden has remained
the same throughout the PISA project. The principal requirements for the model are the ability to
accurately reproduce the soil’s nonlinear small strain stiffness and its ultimate strength. The adopted
model is an expanded generalised version of the Modified Cam Clay (MCC) model of Roscoe &
Burland (1968).
The model adopts the original formulation in the isotropic v – ln p' plane, which assumes a linear
shape of the normal (virgin) compression line, defined as:
𝑣 = 𝑣1 − 𝜆 ln 𝑝′ Equation 7
where 𝑝′ is the mean effective stress, 𝑣 = 1 + 𝑒 is the specific volume and 𝑣1 and 𝜆 are model
parameters. The plasticity develops under shearing, when stress state reaches the yield surface, 𝐹,
which in the present version of the model combines the elliptical shape adopted by the original MCC
model with the nonlinear Hvorslev surface proposed by Tsiampousi et al. (2013):
𝑛
𝐽 𝛼 𝛼 𝑝0′ 𝑝0′
− − (1 − ) ( ) = 0, 𝑝′ <
𝑝′ 𝑔(𝜃) 𝑔(𝜃) 𝑔(𝜃) 2𝑝′ 2
𝐹= 2 Equation 8
𝐽 𝑝0′ 𝑝0′
( ′
) − ( ′ − 1) = 0, 𝑝′ ≥
{ 𝑝 𝑔(𝜃) 𝑝 2
Page 68/272
where 𝐽 (= 𝑞/√3 in triaxial conditions) is the second invariant of the deviatoric stress tensor, 𝑝’0 is
the hardening parameter defining the size of the yield surface, 𝛼 and 𝑛 are parameters of the
Hvorslev surface and 𝑔(𝜃) is the stress ratio at Critical State for a given value of the Lode angle, 𝜃.
The complete shape of the yield surface, in the p' - J plane, is illustrated in Figure 52, where it can
be seen that the use of the Hvorslev surface aims at limiting the maximum strength simulated for
over-consolidated materials, which is known to be greatly overestimated by the original MCC ellipse.
Figure 52: Yield surface in J-p' plane adopted by the expanded MCC model
The variation of soil strength in the deviatoric plane, defined by the stress ratio at Critical State, 𝑔(𝜃),
adopts the Van Eekelen (1980) general expression:
𝑋
𝑔(𝜃) = Equation 9
(1 + 𝑌 sin 3𝜃) 𝑍
where 𝑋, 𝑌 and 𝑍 are model constants which need to be determined. Based on the values of 𝑔(𝜃)
and 𝜃, it is possible to establish the corresponding angle of shearing resistance using:
√3𝑔(𝜃) cos 𝜃
𝜑𝐶𝑆 = sin−1 ( ) Equation 10
√3 − 𝑔(𝜃) sin 𝜃
The plastic potential at a given stress state on the yield surface (p'c, Jc) is defined by:
𝑛 𝑚
𝐽 𝛼 𝑝𝑐′ 𝛼 𝑝𝑐′ 𝑝0′ 𝑝′ − 𝑝𝑐 𝛽 𝑝0′ − 2𝑝𝑐′ 𝑝0′
− − (1 − ) ( ) − ( ) = 0, 𝑝′ <
𝑝′ 𝑔(𝜃) 𝑔(𝜃) 𝑝′ 𝑔(𝜃) 𝑝′ 2𝑝𝑐′ 𝑝′ 𝑔(𝜃) 𝑝0′ 2
𝑃= 2
𝐽 𝑝0′ 𝑝0′
( ′ ) − ( ′ − 1) = 0, 𝑝′ ≥
{ 𝑝 𝑔(𝜃) 𝑝 2
where 𝛽 is a parameter determining the gradient of the flow vector for 𝑝′ of 0.0 kPa. The expressions
above imply that, in general, plastic behaviour is only associated in the 𝑝′ − J plane if the stress point
encounters the yield surface on the elliptical part. In the deviatoric plane, it is assumed that the
plastic potential has the same shape as the yield surface, being given by the expression proposed
by Van Eekelen (1980) with the same set of parameters 𝑋, 𝑌 and 𝑍.
Page 69/272
In terms of elastic behaviour, the formulation for the tangent bulk stiffness, 𝐾𝑡𝑎𝑛 , established for the
original MCC model, is adopted:
𝑣𝑝′
𝐾𝑡𝑎𝑛 = Equation 11
𝜅
where 𝜅 is the slope of the isotropic swelling line in the 𝑣 − ln 𝑝′ space. Conversely, a modified
hyperbolic expression is adopted for the tangent shear modulus, 𝐺𝑡𝑎𝑛 :
𝑝′ 1 − 𝑅𝑚𝑖𝑛
𝐺𝑡𝑎𝑛 = 𝐺0 ( ′ ) (𝑅𝑚𝑖𝑛 + (1 ) Equation 12
𝑝𝑟𝑒𝑓 + 𝑎𝐸𝑑 )𝑏
′ ′
where 𝑝𝑟𝑒𝑓 is a reference pressure, 𝐺0 is the small-strain shear modulus at 𝑝′ = 𝑝𝑟𝑒𝑓 , 𝐸𝑑 is the
generalised deviatoric strain [= (2/√6)√(𝜀1 − 𝜀2 )2 + (𝜀1 − 𝜀3 )2 + (𝜀2 − 𝜀3 )2 ], and 𝑎, 𝑏 and 𝑅𝑚𝑖𝑛 are
model parameters.
A3 Recalibration of the Cowden constitutive model
A3.1 Strength
The first parameter to be assessed is that defining the drained strength of the Cowden glacial till.
Triaxial tests in compression (TXC) and extension (TXE) have been analysed to identify those which
reached failure conditions that could be interpreted as corresponding to the occurrence of Critical
State. From such analyses, Figure 53 summarises stress conditions at Critical States in q - p' space
from individual tests. ‘Historic’ data represent tests carried out by the BRE (Powell, 2003) which
were used in the original calibration of the constitutive model. ‘New’ data result from the new PISA
experimental campaigns at ICL and GEOLABS, summarised in Section A1. Both sets of critical state
data plot closely together, rendering the conclusion that the originally derived strength parameters
for this soil are appropriate.
Consequently, average ultimate stress ratios q/p' remain at 1.07 and 0.90 in triaxial compression
and triaxial extension, respectively. These values correspond to angles of shearing resistance at
Critical State of 𝜑TXC = 27° and 𝜑TXE = 32°, suggesting the existence of a considerable degree of
Lode’s angle dependency and/or strength anisotropy in this material.
In order to reproduce the different strengths corresponding to triaxial compression and triaxial
extension conditions, the generalised Van Eekelen (1980) shape in the deviatoric plane was used
for the yield surface. The parameters of this function (Equation 9) in the deviatoric plane remain the
same as originally derived, with 𝑋 = 0.548, 𝑌 = 0.698 and 𝑍 = 0.100. Figure 54 shows the resulting
variation of the angle of shearing resistance with Lode’s angle, demonstrating the ability of the
model to capture the experimentally observed strength variation. Note that the chosen parameters
ensure that the yield surface is convex, a condition necessary to guarantee numerical stability.
The need to employ the Hvorslev surface was established by the original model calibration. The new
data from isotropically consolidated (K0 = 1.0) undrained TXC tests, summarised in Table 12, are
used to examine the validity of the originally derived parameters of the Hvorslev surface. Figure 55
summarises normalised effective stress paths from these tests, in terms of p'/p0' ratio versus J/p'
ratio. The originally calibrated shape of the Hvorslev surface with parameters α = 0.25 and n = 0.40
Page 70/272
(Equation 8), also plotted in the figure, shows an excellent fit to the new data. Consequently, the
Hvorslev surface parameters remain unchanged.
Figure 53: Cowden clay till strengths in triaxial compression and extension
Figure 54: Adopted variation of angle of shearing resistance at critical state in the deviatoric
plane
The plastic potential of the Hvorslev surface requires parameters  and m to be evaluated. The
former controls the gradient of the flow vector at p' = 0 kPa, while the latter controls the nonlinearity
of its variation to the Critical State, at which the gradient of this vector must be zero (i.e. no
volumetric plastic strains are simulated). For this calibration it is necessary to use drained TXC tests,
which were available in the original model calibration. As new drained TXC tests were not available
Page 71/272
at the time of model recalibration, the values of  = 0.20 and m = 1.0 originally established were
adopted.
Figure 55: Fitting of the Hvorslev surface to new experimental data
A3.2 1D compressibility
The historic data from oedometer tests on intact samples, described in Robson (1988), are
reproduced in v – lnp' space in Figure 56. Their interpretation suggested that the virgin compression
line (VCL, Equation 7) could be approximated by setting v1 = 1.757 and  = 0.062, the former being
the magnitude of the specific volume at unit mean effective stress and the latter being the gradient of
the line in v – lnp' space. From unload-reload loops carried out in these experiments it was not
possible to establish with accuracy the gradient of the swelling lines, , which is also a model
parameter. Consequently, a value of 5.0 was assumed for the ratio /, leading to  = 0.0124. The
suitability of this estimate was verified by the agreement between the measured undrained strength
and that predicted by the constitutive model in the original model calibration.
Figure 56: Interpretation of oedometer data
Page 72/272
New oedometer tests on intact Cowden till specimens were conducted by GEOLABS, applying
constant rate of strain (CRS) consolidation to samples 65 mm in diameter and 18.6 mm high. These
tests are also interpreted in v – lnp' space and added to Figure 56, showing larger compressibility of
the soil. The newly interpreted parameters of the virgin compression line are v 1 = 2.2,  = 0.1145
and  = 0.021, with ratio /= 5.45 being close to the previously assumed value of 5.0.
A3.3 Shear stiffness

The original model calibration for shear stiffness relied on historic data from in situ crosshole,
downhole and surface seismic wave measurements, and laboratory resonant column and triaxial
testing. The set of in-situ data showed significant scatter and it was therefore decided to fit the
maximum shear modulus G0 to the resonant column and triaxial data, as these were performed
under more controlled laboratory conditions. Assuming a reference effective pressure p'ref = 100 kPa
in Equation 12 and setting G0 = 80 MPa, the measured stiffness variation with depth was adequately
presented (i.e. G0 = 800p’). This G0 profile is reproduced in Figure 58. For clarity, previous
experimental data are not included in this figure, but were reported in AWG (2014a).
In terms of stiffness degradation with increasing strain, the historic data, reproduced as a range of
measurements shaded in grey in Figure 57, show the inability of the instrumentation to resolve the
readings at very small strains. The originally adopted stiffness degradation, in terms of the
normalised secant shear modulus, Gsec/p', in Figure 57, is fitted to 800p’ at the beginning of shearing
and forced to pass close to the upper boundary of the shaded stiffness range between 0.01% and
1.0% deviatoric strain, as recommended by Lehane (1992) for the interpretation of pile testing in
-5
axial loading at Cowden at that time. The fitting parameters in Equation 12 are a = 9.87x10 , b =
0.987 and Rmin = 0.05.
The new triaxial tests performed in the ICL laboratory were equipped with high resolution amplified
local strain LVDT devices. The 38 mm diameter samples were equipped with two diametrically
opposite axial local strain devices, whereas 100 mm diameter samples employed three axial gauges
o
set at 120 apart circumferentially. The latter were also equipped with an advanced system for radial
strain measurements and bender element (BE) systems. The laboratory data, in particular from 100
mm samples, showed clear initial linear portions of the stress-strain curves, which were interpreted
as equivalent G0 profiles in triaxial compression (TXC) and extension (TXE) in Figure 58. These two
profiles are in reasonable agreement with the originally derived G0 profile of 800p' in the deeper
deposit, although higher G0 is indicated in the superficial soil.
However, the interpretation of G0 from dynamic shear wave propagation using the new BE results
revealed higher stiffness profiles from both vertically (vh) and horizontally (hh) travelling waves
oscillating in the horizontal direction. Additionally, new in-situ stiffness data interpreted from two
Seismic Cone Penetration Tests (SCPT), show G0 profiles similar to those from the BE tests (Figure
58). Consequently, there appears to be some discrepancy in the stiffness profiles between static and
dynamic measurement techniques, which could not be explained from available information.
In terms of stiffness degradation with strain, reproduced in Figure 57 from TXC tests, the linear
normalisation of the secant shear modulus, Gsec, with p’ shows a wide range of curves which plot at
the upper boundary or above the previously measured range of stiffness degradation. A nonlinear
normalisation of shear stiffness would perhaps be more appropriate, however as the MCC model
assumes a linear relationship between the bulk stiffness, K, and p', the same is maintained for G and
Page 73/272
p'. The newly adopted G0 profile of 1100p' is approximately an upper boundary for G0 interpretation
from triaxial local strain measurements, and a lower boundary for G0 interpretation from dynamic
measurement techniques. The stiffness degradation, however, is fitted with the same a, b and Rmin
parameters as previously established.
Figure 57: Shear stiffness degradation at small strains
Figure 58: G0 profile with depth
Page 74/272
A summary of the recalibrated parameters of the generalised MCC model, following the procedures
summarised in this section, is given in Table 13. Comparisons between experimental results and
numerical simulations using these model parameters for three undrained TXC tests, on samples
from three different depths, are shown in Figure 59 and Figure 60, in terms of stress-strain and
excess pore pressure-strain curves.
Table 13: Summary of recalibrated MCC parameters for Cowden till

Component Parameters
Strength (Van Eekelen, 1980) 𝑋 = 0.548; 𝑌 = 0.698; 𝑍 = 0.100
Hvorslev surface – shape (Tsiampousi et al., 2013) 𝛼 = 0.25; 𝑛 = 0.40
Hvorslev surface – plastic potential (Tsiampousi et al., 2013) 𝛽 = 0.20; 𝑚 = 1.00
Virgin consolidation line 𝑣1 = 2.20; 𝜆 = 0.1145
Nonlinear elasticity – bulk stiffness 𝜅 = 0.021
Nonlinear elasticity – small strain shear stiffness ′
𝐺0 = 110MPa; 𝑝𝑟𝑒𝑓 = 100.0kPa
Nonlinear elasticity – shear stiffness degradation 𝑎 = 9.78 × 10−5 ; 𝑏 = 0.987; 𝑅𝑚𝑖𝑛 = 0.05
Figure 59: Measured and simulated stress-strain curves for selected undrained TXC tests
Figure 60: Measured and simulated excess pore water pressures for selected undrained TXC
tests
Page 75/272
A3.4 Initial ground conditions

The new site investigation (SI) at Cowden was aimed at profiling the depth of the deposit down to
approximately the maximum pile length of 10.5 m. Early CPT tests confirmed the existence of a
sand layer at about 12 m depth, the presence of which was known also from historic site
investigations. No changes in ground water conditions were reported from the new SI, consequently
the same ground water level (at 1.0 m depth) and the under-drained pore water pressure profile,
shown in Figure 61, were adopted in the reanalysis of the test piles.
Figure 61: Ground profile and ground water profile at Cowden

In a similar manner, no new data was available to inform possible changes in the in situ K0 profile,
consequently the profile previously established was adopted for the reanalyses of the test piles and
is reproduced in Figure 62(a). The high superficial K0 values implied by simple interpretation of the
oedometer tests shown in the figure were considered unrealistic, as it was believed that the one-
dimensional nature of an oedometer test is not representative of processes induced by glaciation. A
value of K0 = 1.5 was therefore adopted down to 4.0 m depth, with the remaining profile averaging
the measurements to about 25.0 m depth.
The calculation of the vertical effective stress also remains the same, utilising a saturated bulk unit
3
weight  = 21.19 kN/m and the pore water pressure profile depicted in Figure 61.
The profiles of undrained strength, su, in triaxial compression (TXC) with depth are shown in Figure
63. The original evaluation of this profile was fitted to historic undrained strength measurements in
triaxial compression on 100 mm diameter pushed samples, indicating a strong surface crust to about
2.0 m depth.
The new SI, in particular early CPT probing, indicated significantly lower strength in surface layers
(Figure 64). It is noted that these test were performed following a wet period with significant rainfall,
Page 76/272
which also required a piling mat to be constructed at the Cowden site, due to very soft ground
surface conditions, to enable installation of the test piles. Laboratory triaxial compression
experiments and shallow hand shear vane (HSV) tests, the latter carried out during block sampling,
confirmed the lower su TXC profile in the top 2.0 m. The TXC tests also showed very good
agreement with originally derived strength profile below 2.0 m depth.
(a) K0 (b) OCR
Figure 62: Profiles with depth adopted for FE reanalysis
It is interesting to note from Figure 63 that sample size (38 or 100 mm diameter) in TXC tests does
not appear to have a significant effect on measured undrained strengths, which may be counter-
intuitive considering the stony nature of the soil and fissuring. All test results reported by Ushev et al.
(2015) are very consistent in this respect, with samples showing ductile behaviour to failure similar
to the examples given in Figure 59. It is also worth noting that the strengths measured from 100 mm
TXC tests by GEOLABS are consistent with values obtained from ICL experiments.
A comparison of the new laboratory TXC undrained strength data with the undrained strength
profiles interpreted from the new CPT traces is shown separately in Figure 64. The erratic nature of
the CPT traces is perhaps amplified by the stone content in the deposit, but if obvious spikes are
ignored then generally good agreement is seen from the two experimental sets. A higher CPT
strength at about 2.0 m depth is perhaps explained by the reported higher stone content at this
horizon (shown in Figure 51), where higher resistance is easily picked up by the cone tip. On the
other hand, a triaxial sample tests the whole soil matrix, not just stones, hence resulting in smaller
undrained strength. The CPT traces consistently detect the presence of a sand layer at about 12.0
m depth, which is about 1.0 m thick.
For the generalised MCC model to reproduce the newly adopted su TXC profile in FE simulations, a
new profile of overconsolidation ratio, OCR, with depth, consistent with the same initial effective
Page 77/272
stresses in the ground applied originally (i.e. 'v and K0), is derived as shown in Figure 62(b). The
new profile differs from the original only in the top 2.0 m.
Figure 63: Undrained strength profile in triaxial compression
Due to K0 > 1.0 in the top 8.0 m of the clay till deposit, the initial soil state to a depth of 8.0m is in
triaxial extension (TXE). This implies that the available undrained strength, su, in the ground down to
8.0m depth is the TXE strength, followed by the TXC strength in the deeper deposit, as shown in
Figure 64.
Whereas the simulated TXC strength profile was derived and calibrated from the new experimental
evidence, as explained above, the TXE strength profile plotted in Figure 65 is a result of the adopted
shape of the yield surface in the deviatoric plane. When compared to the measured TXE undrained
strength from 38 mm samples, the profile fits reasonably well through the measured data. The
observed larger scatter in the measured TXE data could be attributed to the necking failure of
samples which is typical for triaxial testing in tension.
In the subsequent finite element analyses the whole Cowden deposit was simulated as a single
layer of the clay till soil, extending to 40 m depth (top of Chalk), with constitutive model parameters
as summarised in Table 13. Variations in the deposit were ensured through the imposed initial
ground conditions. The presence of sand layers was ignored, as they are positioned well below the
longest tested pile.
Page 78/272
Figure 64: Comparison of laboratory and field interpretations of undrained compression

strength
Figure 65: Initial undrained strength profile in the ground
Page 79/272
A4 Dunkirk sand characterisation
A4.1 New experimental programmes
The sampling of Dunkirk sand was conducted in summer 2014, immediately after the Cowden
sampling campaign, using the same Geobor-S sampling technique. Unfortunately, this campaign
was unsuccessful and no testable sand samples were obtained. Three boreholes were drilled, with a
total core recovery rate of 45%. Attempts were made to preserve samples in plastic containers,
bags, cling film and wax, and in plastic liners. None of the methods proved to be successful and in
all cases the samples deformed readily during handling. Due to the high permeability of the sand,
substantial water content changes occurred. The samples were also contaminated with the drilling
polymer mud, so none was brought to the laboratory. Examples of the recovered sand samples are
shown in Figure 66.
Figure 66: Examples of Dunkirk sand samples
No further sampling, to attempt the retrieval of intact specimens, was organised at the Dunkirk site.
A new laboratory testing programme at ICL, to facilitate the recalibration of the Dunkirk sand
constitutive model, was carried out on reconstituted samples of hand-dug Dunkirk sand from the
superficial layers.
A number of in-situ Cone Penetration Tests (CPTu) were performed during the installation of the
piles, at the locations of all medium (D=0.762m) and large (D=2.0m) diameter piles, and their
interpretation was used in the assessment of the initial conditions at the site. Additionally, new shear
wave velocity measurements, from in-situ Seismic Cone Penetration Tests (SCPT), were also
employed in the recalibration of the Dunkirk sand constitutive model.
The new soil test data became available for recalibration of the sand model between mid-November
2015 and mid-January 2016.
A4.2 Laboratory testing at ICL
The new programme of Dunkirk sand laboratory testing was conducted at the ICL Geotechnics
laboratory from August to November 2015. The work forms part of the PhD study of Tingfa Liu,
which commenced in October 2014.
A suite of triaxial tests on 38mm diameter samples, summarised in Table 14, was conducted in the
same ICL designed automated stress path cells used previously for the new Cowden till testing
Page 80/272
programme. The stress path cells were instrumented to measure local strains and the experimental
set up for testing sands is described in the recent PhD thesis of Aghakouchak (2015).
All samples were reconstituted to initial relative densities (void ratios) that are broadly representative
of those in-situ. They were then isotropically compressed to a desired stress level, p0’, followed by a
rest period to allow creep strains to diminish to very low strain rates. Subsequent monotonic drained
shearing in compression (C) and extension (E) was applied at an axial strain rate of 5% per day,
which was sufficiently slow to prevent the build-up of excess pore water pressures.
Table 14: Summary of Dunkirk triaxial tests, November 2015
Initial void Pre-shear Pre-shear Pre-shear

Test
ratio, e0 p’0 (kPa) OCR K0
1 (C) 0.640 50 1.0 1.0
2 (C, E) 0.640 100 1.0 1.0
3 (C) 0.640 150 1.0 1.0
4 (C) 0.640 400 1.0 1.0
5 (C) 0.587 100 1.0 1.0
6 (C) 0.740 100 1.0 1.0
7 (C, E) 0.740 400 1.0 1.0
8 (C) 0.640 50 4.0 1.0
9 (C) 0.640 100 4.0 1.0
10 (C) 0.640 150 4.0 1.0
11 (C) 0.640 50 12.0 1.0
12 (C) 0.587 100 4.0 1.0
1 (C)3 0.740 100 4.0 1.0
A4.3 Dunkirk geotechnical profile
The new site investigation broadly confirmed the Dunkirk geotechnical profile derived from the
historic information and reported by AWG (2014b). Table 15 summarises the stratigraphy of the site,
as reported by Chow (1997), where the hydraulic fill and the natural sand underneath were
described as uniform, fine to medium sands of the same geological origin, with a mean particle size
D50 = 0.25mm.
A summary of the new CPT data is presented in Figure 67, showing the average (mean), maximum
and minimum recorded values for all pile locations. From the new pore water pressure
measurements the ground water level is estimated at around 5.4m depth, which is about 1.4m
deeper compared to the historic data. Due to the lack of further site investigation, the uncertainty of
the hydraulic conditions above the water table has remained, as this part of the deposit is likely to be
unsaturated. The observed creation of stable gaps to sizeable depths, with a vertical soil face
around the laterally loaded test piles, clearly confirmed the existence of considerable effective
stresses (potentially suctions) above the water level, or a certain degree of cementation which could
not be confirmed or quantified.
Page 81/272
Table 15: Stratigraphy at Dunkirk

0-3m Hydraulic Fill Sand fill that was dredged from the offshore Flandrian deposits, and
placed to raise the ground level. No compaction or surcharging has
taken place.
3-30m Flandrian Sand Marine sand deposited during three local marine transgressions.
These sands are often separated by organic layers which
accumulated between transgressions. A 600mm thick organic layer
is found at around 8m depth, separating the Flandrain sand into
upper and middle units.
30m + Yprésienne Clay An Eocene marine clay (also known as London Clay and Argile de
Flandres) which extends beneath the southern North Sea.
Figure 67: CPT profiles from PISA site investigation at Dunkirk

The interpretation of the relative density (Dr) profiles from the new CPT data showed significant
dependency on the adopted correlation with the cone resistance, qc. The correlation of Baldi et al.
(1986) indicated the relative density in excess of 100% throughout the deposit. Adopting the
correlation of Mayne & Kulhawy (1990) resulted in a Dr profile that is closer to that estimated in
Chow (1997), which used the CPT correlation of Lunne & Christoffersen (1983). Consequently, the
adopted Dr profile for the final 3D FE analyses remains unchanged, with Dr=100% in the hydraulic
fill (top 3m) and Dr=75% in the rest of the deposit.
A5 Soil constitutive model for Dunkirk sand

The constitutive model for simulating the behaviour of the dense marine sand at Dunkirk has
remained the same throughout the PISA Project. It is a Critical State model based on the state
parameter framework for sands (Been & Jefferies, 1985), with the current model version, described
in detail in Taborda et al. (2014), being an evolution of the bounding surface plasticity model
proposed initially by Manzari & Dafalias (1997) and subsequently improved by Papadimitriou &
Bouckovalas (2002).
Page 82/272
In addition to a yield surface, employed to determine the onset of plasticity, the model uses three
additional surfaces to characterise the response of the material: the critical state surface, which
defines the stress conditions at failure, the dilatancy surface, which defines the transition from plastic
contraction to plastic dilation, and the bounding surface, which controls the material’s peak strength
and the plastic strain rate. While the critical state surface is defined by the stress ratio at failure
under triaxial compression conditions – a model parameter denoted as 𝑀𝑐𝑐 –, the other surfaces
depend on the value of the state parameter, which represents the difference between the current
void ratio, 𝑒, and that at critical state for the same mean effective stress, 𝑒𝐶𝑆 :
𝜉
𝑝′
𝜓 = 𝑒 − 𝑒𝐶𝑆 = 𝑒 − (𝑒𝐶𝑆,𝑟𝑒𝑓 − 𝜆 ∙ ( ′ ) ) Equation 13
𝑝𝑟𝑒𝑓
′
where 𝑒𝐶𝑆,𝑟𝑒𝑓 is the void ratio at critical state for 𝑝′ = 0 kPa, 𝑝𝑟𝑒𝑓 is a reference pressure and 𝜆 and
𝜉 are parameters defining the shape of the critical state line (CSL) in the 𝑒 - 𝑝′ plane. Based on the
current value of 𝜓, the positions of the dilatancy and bounding surfaces, defined by the respective
stress ratios in triaxial compression 𝑀𝑐𝑑 and 𝑀𝑐𝑏 , can be established using:
𝑀𝑐𝑑 = 𝑀𝑐𝑐 + 𝑘𝑐𝑑 ∙ 𝜓

Equation 14
𝑀𝑐𝑏 = 𝑀𝑐𝑐 + 𝑘𝑐𝑏 ∙ 〈−𝜓〉
where 𝑘𝑐𝑑 and 𝑘𝑐𝑏 are model parameters and 〈 〉 are the Macauley brackets, according to which
〈𝑥〉 = 𝑥 if 𝑥 > 0 and 〈𝑥〉 = 0 if 𝑥 < 0.
A6 Recalibration of the Dunkirk sand constitutive model
A6.1 Critical state strength
The initial triaxial tests carried out by Aghakouchak (2015) were complemented by a series of triaxial
tests performed to assist the recalibration of the sand model for the Dunkirk site (Table 14). To
determine the strength at critical state, the variations of stress ratio 𝑞/𝑝′ with strain level were
analysed, as shown in Figure 68. Clearly, in initially dense samples, critical state can only be
reached for very high deformation levels, rendering the identification of their occurrence a difficult
process. However, as shown in Figure 68, in the new triaxial tests most samples were sheared until
strains in excess of 25% and 10% were reached in compression and extension, respectively,
enabling stress ratios of 𝑀𝑐𝑐 = 1.28 and 𝑀𝑒𝑐 = 0.92 to be estimated. These values are identical to
those determined in the original model calibration reported by AWG (2014b), with the former
corresponding to an angle of shearing resistance of 𝜑 𝑇𝑋𝐶 = 32° .
Page 83/272
Figure 68: Determination of stress ratios at critical state under triaxial loading conditions
A6.2 Critical state line (CSL)
To assess the position of the critical state line in 𝑒 − 𝑝′ space, new triaxial compression tests on
samples with an initial void ratio 𝑒 of 0.64 are compared with those performed by Aghakouchak
(2015) and used in the original model calibration. The paths in this plane are shown in Figure 69,
together with the CSL adopted in the original calibration. The CSL is defined by Equation 15, in
which it was assumed that 𝑒𝐶𝑆 = 𝑒𝑚𝑎𝑥 = 0.91 (Sim et al., 2013) for 𝑝′ = 0 𝑘𝑃𝑎, as suggested by
Klotz & Coop (2002). Note that, for clarity, the tests by Aghakouchak (2015) are represented in
Figure 69 only by the points corresponding to the estimated occurrence of critical state conditions.
Clearly, the response obtained in the new tests is in accordance with the previously established
CSL, meaning that no adjustment of this component of the model is required.
𝜉 0.179
𝑝′ 𝑝′
𝑒𝐶𝑆 = 𝑒𝐶𝑆,𝑟𝑒𝑓 − 𝜆 ∙ ( ′ ) = 0.91 − 0.135 ∙ ( ) Equation 15
𝑝𝑟𝑒𝑓 101.3
A6.3 Elastic stiffness
The maximum shear stiffness simulated by the constitutive model is dependent on the current
values of mean effective stress, 𝑝′, and void ratio, 𝑒, through the relationship proposed by Hardin &
Richart (1963):
′
𝐵 ∙ 𝑝𝑟𝑒𝑓 𝑝′
𝐺0 = ∙ √ ′ Equation 16
0.3 + 0.7 ∙ 𝑒 2 𝑝𝑟𝑒𝑓
where 𝐵 is a model parameter requiring calibration. The two sets of available triaxial tests were
analysed in order to establish the vertical Young’s modulus, 𝐸𝑣 , at very small strains, the values of
which were subsequently converted to shear modulus by assuming a constant Poisson’s ratio
𝑣 = 0.17 (Kuwano, 1999) using:
Page 84/272
𝐸𝑣
𝐺= Equation 17
2 ∙ (1 + 𝜈)
Given the shape of Equation 16, the value of 𝐵 can be calculated through linear regression by
plotting the values of 𝐺0 against the modified mean effective stress, 𝑝∗ :
′
𝑝𝑟𝑒𝑓 𝑝′
𝑝∗ = ∙ √ ′ Equation 18
0.3 + 0.7 ∙ 𝑒 2 𝑝𝑟𝑒𝑓
Clearly, Figure 70 demonstrates that the two sets of tests yield distinct values of 𝐵, probably due to
the different consolidation states. Tests by Aghakouchak (2015) were anisotropically consolidated
with 𝐾0 = 0.5, while the samples in the new set of tests were isotropically consolidated (𝐾0 = 1.0)
prior to shearing. Also depicted in Figure 70 is the value of 𝐵 determined by considering only the
available new SCPT results, which are shown in Figure 71, together with the approximation provided
by Equation 16. Given that 𝐵 = 875 reproduces the measured field data and is within the range
measured in the laboratory, this value was adopted in the recalibration of the model for Dunkirk
sand.
Figure 69: Assessment of the position of the critical state line in e – p’ plane
Page 85/272
Figure 70: Calibration of the expression for the shear modulus at very small strains
Figure 71: G0 profile with depth

The elastic stiffness in the adopted model is assumed to change with the mobilised stress ratio,
according to a Ramberg-Osgood-type relation. This aspect of the material’s behaviour is controlled
by parameters 𝑎1 , 𝛾1 and 𝜅. The latter was assumed to have the value of 2.0, in accordance with the
suggestion by Papadimitriou & Bouckovalas (2002), while a value of 0.40 was prescribed for 𝑎1 . The
final parameter, 𝛾1 was calculated using the expression proposed by Loukidis & Salgado (2009),
-3
which relates this quantity to 𝑎1 , returning a value of 1.031×10 .
Page 86/272
A6.4 Peak strength and phase transformation
According to the adopted constitutive framework, the peak angle of shearing resistance and the
stress ratio at phase transformation are linear functions of the state parameter (Equation 14).
Therefore, to determine the parameter 𝑘𝑐𝑏 , which controls the first of these two aspects of soil
response, the peak mobilised stress ratio needs to be plotted against the corresponding value of
state parameter at that stage of the test. A linear regression analysis can then be carried out,
specifying that, for 𝜓 = 0, the peak stress ratio must coincide with that at critical state conditions,
𝑀𝑐𝑐 = 1.28. This process is shown in Figure 72, yielding a preliminary estimate for 𝑘𝑐𝑏 of 3.30. It is
important to note that this value is only an initial calculation as it depends greatly on the assessed
magnitude of 𝜓.
Figure 72: Calculation of the parameter controlling the peak strength

To avoid the simulation of excessively high values of angle of shearing resistance, a limit is
introduced in the calculation on the opening of the bounding surface using Equation 14. A maximum
of 50° was imposed for 𝜑′ under plane strain conditions (approximately 𝜃 = 0°), meaning that,
𝑏
under triaxial compression loading, the maximum stress ratio that can be mobilised is 𝑀𝑐,𝑚𝑎𝑥 =
1.631, which corresponds to 𝜑𝑝𝑒𝑎𝑘,𝑇𝑋𝐶 = 40° (see Taborda et al. (2014) for full details on the
assumed deviatoric shape for the various model surfaces). This limit is only slightly below the
maximum mobilised strength in the available laboratory tests (Test 5, 𝜑𝑝𝑒𝑎𝑘,𝑇𝑋𝐶 = 42°). Additional
tests carried out on dense samples under low stress levels and non-triaxial loading conditions would
be required to validate the assumed limiting value.
A similar process is carried out for phase transformation, which in drained tests is characterised by
the transition from plastic contraction to plastic dilation. Therefore, the occurrence of this distinctive
state can be established by determining the stress ratio at which the plastic dilatancy ratio changes
sign:
Δ𝑝′
𝑝𝑙𝑎𝑠𝑡𝑖𝑐
Δ𝜀𝑣𝑜𝑙 𝑒𝑙𝑎𝑠𝑡𝑖𝑐
Δ𝜀𝑣𝑜𝑙 − Δ𝜀𝑣𝑜𝑙 Δ𝜀𝑣𝑜𝑙 −
𝐾𝑡𝑎𝑛
𝐷 𝑝𝑙𝑎𝑠𝑡𝑖𝑐 = = = Equation 19
Δ𝜀𝑑
𝑝𝑙𝑎𝑠𝑡𝑖𝑐
Δ𝜀𝑑 − Δ𝜀𝑑𝑒𝑙𝑎𝑠𝑡𝑖𝑐 Δ𝑞
Δ𝜀𝑑 − 3𝐺
𝑡𝑎𝑛
Page 87/272
Clearly, this process has a greater degree of uncertainty since it depends on the assumed elastic
behaviour, which itself is an idealisation. The identified stress ratio is plotted against the
corresponding state parameter and a linear regression is carried out, ensuring that, at 𝜓 = 0, the
phase transformation needs to take place at the stress ratio corresponding to the critical state
strength. This process is illustrated in Figure 73, where a value of 𝑘𝑐𝑑 of 0.88 is determined.
Figure 73: Calculation of the parameter controlling the onset of plastic dilation
A6.5 Plastic response and model adjustments
The plastic behaviour is governed by the flow rule and the hardening modulus. In the chosen model,
the flow rule is characterised by a single parameter 𝐴0 , which can be estimated by performing the
adjustment of a linear relationship between the stress ratio 𝑞/𝑝′ and the plastic dilatancy rate
𝐷 𝑝𝑙𝑎𝑠𝑡𝑖𝑐 . This procedure, which, when applied to the available tests, returns a value of 𝐴0 = 1.3, is
explained in detail in Loukidis & Salgado (2009) and Taborda (2011).
The hardening modulus is calculated using:
𝐴 = 𝑝′ ∙ ℎ0 ∙ ℎ𝑒 ∙ ℎ𝑔 ∙ ℎ𝑏 ∙ ℎ𝑓 ∙ 𝑑𝑏 Equation 20
Where ℎ0 is a model parameter, ℎ𝑒 denotes the influence of the void ratio, ℎ𝑔 incorporates the
nonlinear elastic shear modulus, ℎ𝑏 is a function of the distance to the bounding surface and its
opening, ℎ𝑓 introduces the impact of generated soil fabric and 𝑑 𝑏 is the current distance to the
bounding surface. For the details of each of these components, see Taborda et al. (2014). Given the
limited available data, it was assumed that ℎ𝑒 = 1.0 (i.e. model parameter 𝛾 = 0.0) and that
ℎ𝑓 = 1.0 (i.e. model parameters 𝐻0 = 0.0 and 𝜁 = 0.0). The current value of the nonlinear elastic
shear modulus without any modification was introduced in the calculation of 𝐴 by setting 𝛼 = 1.0, as
suggested in Taborda et al. (2014). Parameters 𝜇 and 𝛽 of the component related to the distance
and opening of the bounding surface, ℎ𝑏 , were set to their recommended values of 1.0 and 0.0,
Page 88/272
respectively (Papadimitriou & Bouckovalas, 2002). The last parameter to be determined is ℎ0 , which
controls the overall magnitude of the hardening modulus A and therefore controls the plastic strain
rate. A trial-and-error procedure is typically required to determine its value, as outlined below.
The calculation of ℎ0 was performed in two steps: the first was to determine the value required for
the best possible approximation to the available laboratory tests, while the second attempted at
reconciling the field stiffness data with those from triaxial tests. The first of these steps, despite
generating a very good fitting to the laboratory data, essentially ignores the fact that the stiffness at
very small strains, which is based on field measurements, was found to be higher than that
determined from local strain measurements in the triaxial tests. Therefore, this initial estimate leads
to a very steep reduction in stiffness from 𝐺0 to the values of stiffness observed in the laboratory
data. To force this transition to take place over a wider strain range, rather than forcing it to take
place at very small strains, a second adjustment of ℎ0 was carried out by increasing its value until a
more satisfactory relationship between shear modulus and deviatoric strain was obtained. This
adjustment, however, results in higher peak strengths being mobilised, meaning that this step needs
to be accompanied by a reduction in 𝑘𝑐𝑏 until the correct value of peak strength is simulated for the
new value of ℎ0 . In the present case, this has resulted in ℎ0 = 0.4 and 𝑘𝑐𝑏 = 2.70. Although a
similar process could be carried out for the position of the dilatancy surface, it was found that the
increase in plastic stiffness resulting from the use of higher values of ℎ0 has little impact on this
parameter, since phase transformation tends to happen very early in the test.
A6.6 Final set of parameters and model performance
The parameters used in the analysis are summarised in Table 16, while the reproductions of the
triaxial tests are shown in Figure 74 in terms of axial strain-deviatoric stress and in terms of axial
strain-volumetric strain. Clearly, the higher stiffness adopted based on the new field data has
resulted in the simulation of a slightly stiffer response than that observed in triaxial tests. Despite
this, it can be seen that the overall reproduction of the measured behaviour is very satisfactory. It
should be noted that if the field data for the very small stiffness are ignored, a better fit to laboratory
tests is obtained.
Table 16: Adopted model parameters
Component Parameters
′
𝑝𝑟𝑒𝑓 = 101.3 𝑘𝑃𝑎; 𝑒𝐶𝑆,𝑟𝑒𝑓 = 0.910;
Critical State Line
𝜆 = 0.135; 𝜉 = 0.179
Strength 𝑀𝑐𝑐 = 1.28; 𝑀𝑒𝑐 = 0.92
𝑘𝑐𝑏 = 2.70; 𝑘𝑐𝑑 = 0.88; 𝑚 = 0.065;
Model surfaces ′
𝑝𝑌𝑆 = 1.0 𝑘𝑃𝑎; 𝐴0 = 1.30
ℎ0 = 0.045; 𝛼 = 1.0; 𝛾 = 0.0;
Hardening modulus
𝛽 = 0.0; 𝜇 = 1.0
Nonlinear elasticity – small strain stiffness 𝐵 = 875.0; 𝜈 = 0.17
Nonlinear elasticity – shear stiffness degradation 𝑎1 = 0.40; 𝛾1 = 1.031 × 10−3 ; 𝜅 = 2.0
Fabric tensor 𝐻0 = 0.0; 𝜁 = 0.0
Page 89/272
Figure 74: Comparison of measured and reproduced soil behaviour in triaxial tests
Page 90/272
A6.7 Initial ground conditions
The only change in the initial ground conditions at the Dunkirk site is the pore water pressure (pwp)
profile, derived from the new CPT tests performed at all pile locations and summarised in Figure 75.
The new data indicate the ground water table at 5.4 m below the ground surface, compared to 4m
from the historic record. The adopted hydrostatic pwp distribution extends to about 1.4m above the
water table, with a maximum suction of 13.5kPa at 4m depth. This pwp distribution is corroborated
by the in-situ measurements, as shown in Figure 75. At depths shallower than 4m the piezocone
readings are unreliable as the soil is likely to be unsaturated. The adopted profile extends the
13.5kPa suction to 2m depth, then falls to 10kPa at the ground surface. This is the approximate
value of suction originally assumed in Chow (1997). As discussed earlier in Section A4, the
formation of stable gaps around the laterally loaded test piles justifies the assumption of suction
above the water table, although not its exact magnitude.
Figure 75: Pore water pressure profile at Dunkirk

As also discussed in Section A4, the adopted initial relative density profile has remained the same,
with Dr=100% in the top 3m of the hydraulic fill, followed by Dr=75% in the natural sand deposit. The
corresponding initial void ratios, e0, are 0.57 and 0.628, respectively. These values have stayed the
same as originally determined, as no new data was available for reinterpretation. The calculation of
3
the vertical stress adopts the same bulk unit weights of 19.9 kN/m below the water table and 17.1
3
kN/m above it. The K0 remains at 0.4.
In the subsequent finite element analyses the whole Dunkirk deposit was simulated as a single layer
of sand, with constitutive model parameters as summarised in Table 16. Variations in the deposit
were ensured through the imposed initial ground conditions.
Page 91/272
PISA Project – Final Report Appendix B - Analysis of monotonic field test measurements
Appendix B Analysis of monotonic field test measurements

As described in detail in the Field Test Factual Report (AWG 2015), large scale field tests were
carried out at a dense sand site and a stiff clay site, on laterally loaded piles of a range of geometry
and loading patterns, in order to validate new design methods over a realistic range of parameter
space. The scope of the PISA project focuses on the prediction of the response of piles to
monotonic loading and therefore the comparison with existing ‘p-y’ and new design methods will be
carried out predominantly with monotonic loading tests. However, the field tests also explored some
additional effects such as varied rate and cyclic loading, which are explored in Appendix G.
This appendix focusses on the processing of the monotonic field test data, in order to confirm the
consistency of the pile tests and to provide concise data on pile response, which can be compared
with predictions from existing and new design models.
B1 Monotonic field test pile geometries

Field tests were carried out on piles covering a range of diameters, lengths and thicknesses. Figure
76 shows a matrix of the monotonic tests carried out at Cowden and Dunkirk. Where repeat tests of
a given geometry were made, the test with the most complete dataset in terms of monotonic loading
has been selected for further analysis, as signified by a bold typeface.
L/D
CS4
10 CM3
DS4
CS3
8 DM3
DS3
CS2 CM6 & CM9 CL1 & CL2

5.25
DS1 DM4 & DM9 DL1 & DL2
CM2 & CM8

3
DM5 & DM7
0.273 0.762 2.0 D (m)
Figure 76: Matrix of monotonic field tests where the pile name consists of the test site (C –
Cowden, D – Dunkirk) pile diameter (S – small, M – mid, L – large) and location number. Bold
type face signifies repeat geometry selected for further analysis.
A full list of all test pile geometries and test description may be found in the Field Test Factual
Report (AWG 2015). The geometries of the monotonic pile tests are given in Table 17.
Page 92/272
Table 17: Monotonic field test geometries
Test Pile D L/D D/t Material Description

5 CS2 0.273 5.25 27 Clay Small diameter, medium length
7 CS3 0.273 8 27 Clay Small diameter, long
8 CS4 0.273 10 27 Clay Small diameter, very long
10 CM9 0.762 5.25 69 Clay Mid-diameter, medium length
11 CM8 0.762 3 76 Clay Mid-diameter, short
12 CM3 0.762 10 30 Clay Mid-diameter, long
15 CM6 0.762 5.25 69 Clay Mid-diameter, medium length
21 CM2 0.762 3 76 Clay Mid-diameter, short
22 DS1 0.273 5.25 27 Sand Small diameter, medium length
23 DS3 0.273 8 27 Sand Small diameter, long
24 DS4 0.273 10 27 Sand Small diameter, very long
26 DM9 0.762 5.25 54 Sand Mid-diameter, medium length
27 DM5 0.762 3 76 Sand Mid-diameter, short
28 DM4 0.762 5.25 54 Sand Mid-diameter, medium length
29 DM7 0.762 3 76 Sand Mid-diameter, short
33 DM3 0.762 8 30 Sand Mid-diameter, long
39 DL1 & DL2 2 5.25 53 Sand Large diameter, medium length
42 CL1 & CL2 2 5.25 80 Clay Large diameter, medium length
B2 Test repeatability and consistency

This section demonstrates that where tests were carried out on piles of similar geometry, results
were consistent and repeatable, which shows that the obtained data are representative of the field
response, within experimental tolerances.
Figure 77(a) shows a comparison of the load displacement response for the mid-diameter short piles
at Cowden (D = 0.762m, L/D = 3, t = 10mm). This plot shows that the fidelity of the response
between the two tests is good, with secant stiffness at small displacement (after load step 3) with
less than 2.1% error and loads at an interpolated ultimate displacement of vG = 0.1D within 13%.
The moderate difference in response at large displacements is assumed to be predominantly
caused by potential differences in soil strength, as shown by the CPT measurements taken before
pile installation in Figure 77(b). However, the test of CM8 was one of the first to be carried out, when
the testing apparatus and procedure was still undergoing optimisation and the small unload-reload
loops during loading are anticipated to result in a marginally reduced load for a given displacement.
Page 93/272
(a) ground load-displacement (b) CPT qt profile
Figure 77: Comparison of short (L/D = 3) medium-diameter (D = 0.762m) pile tests at Cowden
(CM2 and CM8)
Figure 78 shows a comparison of the load displacement response for the mid-diameter mid-length
piles at Cowden (D = 0.762m, L/D = 5.25, t = 11mm). Following load step 4, pile CM6 was subject to
cyclic loading and a comparison of the ultimate response is therefore not possible. However, the
values of small displacement secant stiffness (after load step 3) are within 23%, which is indicative
of the variance that can be expected between similar tests.
Figure 78: Comparison of first three load steps of mid-length (L/D = 5.25) medium-diameter (D
= 0.762m) pile tests at Cowden (CM9 and CM6)
Figure 79 shows a comparison of the load displacement response for the large diameter piles at
Cowden (D = 2.0m, L/D = 5.25, t = 25mm). The small displacement secant stiffness (after load step
Page 94/272
3) and large displacement loads interpolated at vG = 0.1D are within 1.2% and 5.3% respectively.
This indicates good repeatability and consistency of the test data, despite a difference in embedded
pile depths with 10.35m and 10.6m for CL1 and CL2 respectively.
Figure 79: Comparison of large-diameter (D = 2.0m) pile tests at Cowden (CL1 and CL2)
Figure 80 shows a comparison of the mid-diameter, short piles tested at Dunkirk. Note that these
piles reached their maximum load prior to the target displacement of vG = 0.1D, and experienced a
subsequent softening. During the first test of this geometry (DM5) this resulted in an acceleration
and oscillation of the test control system, which was trying to maintain the target load. During the
second test (DM7) the test was transferred to displacement control once the pile displayed signs of
yielding. The small displacement secant stiffness (at load step 3) and ultimate loads are within 2.9%
and 5.4% respectively, indicating a consistent and repeatable dataset.
(a) ground load-displacement (b) CPT qc profile
Figure 80: Comparison of short (L/D = 3) mid-diameter (D = 0.762m) pile tests at Dunkirk (DM5
and DM7)
Page 95/272
Figure 81 shows a comparison of the mid-diameter, medium length piles at Dunkirk. The small
displacement secant stiffness (after load step 3) and large displacement loads interpolated at vG =
0.1D are within 18.5% and 8.6% respectively, which represent the greatest observed variability of
the tests in sand.
Figure 81: Comparison of mid-length (L/D = 5.25) medium-diameter (D = 0.762m) pile tests at
Dunkirk (DM4 and DM9)
Figure 82 shows a comparison of the large-diameter pile tests at Dunkirk. The two piles were loaded
against each other and were therefore tested simultaneously. The resulting load-displacement
responses show good fidelity, with small displacement secant stiffness (after load step 3) and large
displacement loads interpolated at vG = 0.1D within 2.3% and 1.8% respectively.
Figure 82: Comparison of large-diameter (D = 2.0m) pile tests at Dunkirk (DL1 and DL2)
Page 96/272
B3 Comparison of varied L/D ratios

Figure 83 shows a comparison of the load-displacement response for the small diameter piles
(D=0.273m) with variations of length to diameter ratio in clay. Note that for clarity the plot of the
small displacement response in Figure 83(b) shows the envelope response of the monotonic curve,
which is calculated as described in Appendix F3. Figure 83(c) features a plot of; the initial stiffness,
calculated from the envelope monotonic curve up to vG=D/1,000; the ultimate response, determined
from the envelope monotonic curve at vG=0.1D; and the reload stiffness. During each monotonic pile
test an unload-reload operation was carried out after the third load step (vG=D/200 in clay and
vG=3D/200 in sand). The reload stiffness is determined over the initial D/1,000 displacement of the
reload curve, after smoothing with a 20 sample wide Gaussian window function. This results in a
stiffness that is comparable to the initial stiffness and indicates what change in pile stiffness occurs
rd
as a result of previous loading. Note that the initial stiffness defined here differs from the ‘3 load
step’ definition provided in AWG (2015). This change represents the result of additional processing
and a measure of initial stiffness that is more representative of small displacements.
As expected, the ultimate response increases as the pile length is increased. However, the initial
stiffness appears to reach a threshold above which the initial stiffness remains unchanged. This is
likely due to the fact that for longer piles the initial stiffness is entirely dependent on the flexure at
shallow depths and does not depend on the deeper pile sections, which are not mobilised. For
shorter piles the initial stiffness is a function of the flexure and rotation of the entire pile.
The proportion of the initial stiffness that is retained after previous loading is greater for the longer
piles (average of 80% for CS3 and CS4) compared to the medium length pile (59% for CS2). This
result is not surprising as the flexure of the longer piles will result in reduced soil remoulding towards
the toe of the pile at a given displacement and greater recovery of both displacement and stiffness
on unloading.
(a) Overall response (b) small displacement (c) Performance metrics

(envelope)
Figure 83: Variation of load-displacement response in clay with L/D at a pile diameter of 0.273m
(CS2, CS3 and CS4)
Page 97/272
Figure 84 shows a comparison of the load-displacement response for the mid-diameter piles
(D=0.762m) with variations of length to diameter ratio in clay. A similar behaviour is observed to the
small diameter piles, however, full recovery of the initial stiffness is observed for the long geometry.

(envelope)
Figure 84: Variation of load-displacement response in clay with L/D at a pile diameter of 0.762m
(CM2, CM9 and CM3)
Figure 85 shows a comparison of the load-displacement response for the mid-diameter piles
(D=0.762m) with variations of length to diameter ratio in sand. Whilst similar patterns of ultimate
response and initial stiffness are observed to the tests in clay, there is a significant drop in stiffness
during the reload step for the medium length and longer pile. This may occur due to the larger
preceding displacement, when compared to the clay tests. However, it is unclear why the initial
stiffness of the short pile is so well retained.
Page 98/272

(envelope)
Figure 85: Variation of load-displacement response in sand with L/D at a pile diameter of 0.762
(DM7, DM4 and DM3)
B4 Comparison between field tests and existing API/DNV ‘p-y method’

The calculation of the pile response using the API/DNV approach has been conducted as an
independent analysis by DONG Energy, as described in Appendix F4.
B4.1 Comparison of measured and predicted response in Cowden Till clay
As shown in Figure 86, modelling of the pile response using the API/DNV approach significantly
under-predicts the loads at ultimate displacement for all geometries.
Page 99/272
(a) CM2 (L/D = 3) (b) CM9 (L/D = 5.25
(c) CM3 (L/D = 10) (d) CL2 (L/D = 5.25)
Figure 86: Representative examples of Cowden field test pile response against prediction
using existing API/DNV ‘p-y method’
The correlation between the measured and predicted load curves is quantified using the ‘quality of
fit’ metric defined in Section 4.4, and plotted in Figure 87. Separate metrics have been determined
for the performance up to ultimate response (𝜂0.1𝐷 ) and the small displacement performance (𝜂𝑠𝑑 ).
Note that the accuracy metric for the small displacements is calculated up to a displacement of
D/1,000, which is the smallest measured displacement that is common between all field tests. An
accuracy of 100% indicates that both curves are identical.
At large displacements the API/DNV approach predicts an area under the load-displacement curve
up to ultimate response that is on average 50% of that observed in the field tests. The API/DNV
approach performs particularly poorly when predicting the small displacement stiffness, predicting an
Page 100/272
average of only 24% of the measured stiffness. At these displacements there is a clear relationship
between accuracy and normalised pile length, with the stiffness of shorter piles being particularly
under-predicted.
Pile Pile
D D
L/D L/D
(a) Ultimate response (up to vG = 0.1D), (b) Small displacement (up to vG = D/1,000),
average η0.1D of 50% average ηsd of 24%
Figure 87: Accuracy metrics of fit for the API/DNV ‘p-y’ method to the Cowden field tests
Page 101/272
B4.2 Comparison of measured and predicted response in Dunkirk sand
Figure 88 shows a comparison of the measured field test response against the predicted response
when using the API/DNV approach.
(a) DM7 (L/D = 3) (b) DM4 (L/D = 5.25
(c) DM3 (L/D = 8) (d) DL2 (L/D = 5.25)
Figure 88: Representative examples of Dunkirk field test pile response against prediction
using existing API/DNV ‘p-y method’
The resulting quality of fit metrics are shown in Figure 89 and demonstrate that the accuracy with
which the API/DNV approach predicts the ultimate pile response is greater for piles of greater
diameter D or length to diameter ratio L/D. It is unclear whether this trend indicates an increase in
accuracy with diameter or a trend towards over-prediction at larger diameters. A similar pattern is
observed for the small displacement predictions, but with lower average accuracies.
Page 102/272
Pile Pile
D D
L/D L/D
(a) Ultimate response (up to vG = 0.1D), (b) Small displacement (up to vG = D/1,000),
average η0.1D of 61% average ηsd of 39%
Figure 89: Accuracy metrics of fit for the API/DNV ‘p-y’ method to the Dunkirk field tests
Page 103/272
B5 Interpretation of embedded pile response

During the field tests some piles featured below-ground instrumentation, which monitored the pile
strains and rotations, from which the bending moment, displacement, section shear and equivalent
lateral distributed load can be calculated, as described in Appendix F2. Figure 90(b) and (c) shows
an example of how a piecewise expression has been fitted to the measurements of bending moment
and rotation for the long pile CM3 (D = 0.762m, L/D = 10) at a large displacement.
(a) (b) (c)
(d) (e)
Figure 90: Embedded pile response for CM3 (D = 0.762m, L/D = 10) at load step 10
(HG =425kN, vG =99.3mm)
Page 104/272
The displacement profile in Figure 90(a) is obtained by integrating the piecewise expressions, whilst
the section shear force and equivalent distributed lateral load in Figure 90(d) and (e) are obtained by
differentiation. It should be noted that the shear force and equivalent distributed loads are those that
would be required to generate the measured bending moment and rotation profile. This means that
they not only include the effect of the true distributed load, but also the distributed moment and base
components. The error in this approximation should be small for such a slender pile, but will become
significant for shorter piles, as the remaining components have a more significant influence on the
pile behaviour. This exercise has therefore only been performed for the long CM3 pile.
It is worth noting that the piecewise expression of bending moment is cubic, which when double
differentiated to obtain the equivalent distributed load curves results in linear expressions with depth.
Unfortunately, there is insufficient data to estimate the variation of equivalent distributed load with
any greater resolution.
A comparison of the interpreted field test pile response with the existing and new design methods is
carried out in Appendix C.
Figure 91 shows how the displacement bending moment profiles develop during the monotonic pile
test of CM3. As shown in Figure 91(a), at a small load only the pile at shallow depth flexes, which
results in a shallow pivot point, defined as the location of zero lateral displacement. As the load is
increased the flexure of the pile increases and eventually evolves into a rotational failure
mechanism, with the pivot location moving gradually to greater depths. Figure 91(b) demonstrates
how the point of the peak bending moment gradually moves to greater depth as the applied load is
increased. This occurs due to a greater mobilisation of bending moment with depth as the pile
deflection increases.
(a) Displacement (b) Section moment
Figure 91: Evolution of embedded pile response for CM3 (D = 0.762m & L/D = 10)
Figure 92 shows a comparison of the metrics of embedded response for the three monotonic pile
tests in clay that were monitored below ground. Note that for the medium length and short piles, the
bending moment at the pile toe was not constrained to 0kNm.
Page 105/272
(a) Peak to ground moment ratio (b) normalised depth of peak moment
(c) normalised depth of pivot point (d) normalised deflection curvature
Figure 92: Embedded pile response metrics for clay field tests (CM2 – D = 0.762m & L/D = 3,
CM3 – D = 0.762m & L/D = 10, CL2 – D = 2.0m & L/D = 5.25)
Figure 92(a) shows the ratio of the peak bending moment to the ground level bending moment. It
can be seen that for the medium and long piles (CL2 and CM3) the ratio of this bending moment
increases as the pile is further displaced. However, the short pile (CM2) features a more constant
bending moment close to the surface at both large and small displacements, which does not
increase significantly above the ground level moment, as indicated by Figure 93. This is likely due to
the fact that the entire pile length is mobilised with relatively little flexure from very small
displacements, resulting in a relatively slow decrease in bending moment towards the pile toe.
Page 106/272
Figure 93: Evolution of embedded bending moment profile for pile CM2 (D = 0.762m & L/D = 3)
Figure 92(b) shows the depth of the peak in bending moment, normalised by the pile length. As
expected, the normalised depth of the peak moment increases with load for the medium length and
long piles (CL2 and CM3), as the proportion of pile length that is mobilised in flexure increases. For
the short pile the depth of maximum bending moment is more difficult to accurately locate, due to a
less pronounced peak in bending moment and fewer datapoints with which to fit the profile.
However, as the pile load is increased the bending moment profile tends towards a near surface
peak and a quicker drop off at depth.
Figure 92(c) shows the normalised depth of the pivot location. As expected, the pivot point of the
longer piles is shallow at small loads and progresses to greater depth. However, the relative
stiffness of the small pile (CM2) results in a rotation of the pile about a pivot point which remains
relatively constant throughout loading. Interestingly, the depth of the pivot point at ultimate response
is similar for each geometry.
Figure 92(d) shows the variation of the mean normalised displacement curvature, which represents
the flexure of the pile relative to its length and is calculated in five steps:
 At each load step the displacement is normalised by the difference between the
displacement at the ground level and the pile toe;
 The depth is normalised by the pile length;
th
 A 5 order polynomial is fitted to the variation of the normalised displacement profile;
 The curvature of the profile is calculated from the second derivative of the resulting
polynomial; and
 The mean value of curvature is obtained.
As expected, the normalised curvature of the longer pile is large at small displacements and tends
towards a reduced value as the pile transitions towards a rotational mechanism. Conversely, the
short pile retains a low normalised curvature throughout loading, due to a small proportion of flexure
in the displacement mechanism.
Figure 94 shows a similar set of embedded response metrics for monotonic test piles in sand. The
general trends are similar to those described for clays.
Page 107/272
(a) Peak to ground moment ratio (b) normalised depth of peak moment
(c) normalised depth of pivot point (d) normalised deflection curvature
Figure 94: Embedded pile response metrics for sand field tests (DM7 – D = 0.762m & L/D = 3,
DM3 – D = 0.762m & L/D = 10, DM4 – D = 0.762m & L/D = 5.25, DL2 – D = 2.0m & L/D = 5.25)
Page 108/272
B6 Further processing of field data
B6.1 Temperature Stability
This section provides further analysis of the field test data provided in the factual report and
accompanying data files. As part of the PISA field test campaign at Dunkirk and Cowden, several
piles (namely DM1, DM6, DL1, DL2, CL1 and CL2) were monitored for a prolonged period prior to
testing with the full instrumentation setup to allow the assessment of the effects of temperature on
instrument performance. Temperature effects are of most importance when considering the small-
strain behaviour. This section is considered to be pertinent to piles subjected to small displacements
over prolonged periods, such as during low amplitude cyclic and damping tests. For standard
monotonic testing the effects of temperature are small relative to the overall pile and pile movements
achieved over the course of a test. Nevertheless, the small strain behaviour during monotonic
testing may be notably influenced by temperature effects. These effects are more significant for
above ground instruments (LVDTs / SUIs / Load Cells etc.) where the temperature fluctuations are
largest. The temperature was recorded for each pile using a thermocouple attached to the outside of
the pile above ground level. Additional temperature measurements from the instrumented test piles
were provided by the Fibre Optic temperature sensors, however, during load testing, these were
seen to be sensitive to strain and are therefore considered unreliable for assessing temperature
during loading.
B6.1.1 LVDT Temperature Effects

The effect of temperature on the LVDT measurements during continuous monitoring is plotted
against time for pile DM1 in Figure 95.
Figure 95: LVDT and temperature against time during overnight monitoring for pile DM1
Page 109/272
All of the LVDTs show an approximately inverse proportional linear relationship with temperature. A
linear least squares fit was used to determine a simple temperature correction factor for each
instrument. From analysing all the available monitoring data from the various pile tests, and due to
the variability in correction factor across all instrument types and the dearth of temperature
monitoring data available, it was concluded that an average correction factor of -0.11 mm/°C would
provide the most robust fit to all the data. The correction factor was then applied to estimate the
LVDT error due to temperature for each test (Error = Temperature difference from start of test * -
0.11). A list of the max, min and range of temperatures during each monotonic test is provided in
Table 18. The absolute temperature error normalised by the passive low (PL) displacement (which is
assumed to be approximately equal to the ground line displacement, vG) at small strain (vG/D =
0.1%) and ultimate (vG/D = 10%) values are also provided in Table 18. It is evident that aside from
some of the small pile tests (CS4, DS1, DS3, DS4) temperature effects are typically less than 10%
of the LVDT value at small strain (vG/D = 0.1%) and less than 1% at the ultimate value (vG/D = 10%).
It is therefore deemed unnecessary to correct for temperature for all mid and large-diameter
monotonic tests. Temperature correction may be applied to the LVDTs for the small pile monotonic
tests if the small strain response is to be accurately calculated. Further study has shown that the
LVDT temperature effects are of greater importance for the cyclic and damping load tests,
particularly for small amplitude cycles, but this data is considered outside the scope of this report.
Table 18: Monotonic test temperature range and normalised LVDT temperature error
Temp Norm Error at Norm Error at

Pilename Max Temp Min Temp Range vG/D = 0.1% vG/D = 10%
(°C) (°C) (°C) (%) (%)
Test5-CS2 5.64 2.48 3.16 2.78 1.14
Test7-CS3 4.70 1.51 3.19 3.87 1.05
Test8-CS4 4.30 -0.18 4.48 43.42 1.03
Test10-CM9 3.96 3.12 0.84 4.11 0.05
Test11-CM8 5.91 5.21 0.69 4.19 0.01
Test12-CM3 7.41 2.06 5.35 6.49 0.30
Test21-CM2 23.26 10.36 12.90 13.39 1.49
Test22-DS1 11.63 9.48 2.15 23.72 0.19
Test23-DS3 13.72 9.79 3.93 34.73 1.41
Test24-DS4 20.00 12.54 7.46 44.63 3.60
Test25-DS2 13.14 12.75 0.39 7.37 0.03
Test26-DM9 16.09 10.23 5.86 6.77 0.39
Test27-DM5 15.62 14.17 1.45 0.66 0.08
Test28-DM4 17.94 11.40 6.53 11.46 0.85
Test29-DM7 12.50 11.55 0.96 1.67 0.04
Test31-DM2 16.28 16.00 0.28 0.03 0.04
Test33-DM3 17.31 15.22 2.09 2.22 0.17
Test39-DL1 20.27 13.49 6.78 12.75 0.04
Test42-CL2 22.26 12.15 10.11 11.60 0.09
Page 110/272
B6.1.2 Load Cell (LCL) Temperature Effects

In a similar manner to the assessment of the LVDT temperature effects, the load cell measurements
were compared with temperature data recorded prior to load testing. Only data where the load cells
were fully free at one end (allowed to thermally expand / contract) was used. It was evident that
there is an approximate linear correlation for the change in load with temperature, although the
magnitude of the temperature effect is relatively small. The magnitude of the correction factor was
found to be related to the load cell capacity, such that the average correction factor for the 175t load
cells = +0.95 kN/°C, compared with +0.0167 kN/°C for the 12t load cells. Considering the largest
temperature change recorded during a test was 12.9 degrees, this relates to a maximum
temperature error of 0.7% of fullscale for the 175t load cell or 0.2% of fullscale for the 12t load cell. A
further check was undertaken, calculating the error in load due to temperature at small (vG/D = 0.1%)
and ultimate (vG/D = 10%) ground-line displacements for each pile, which found an average error of
≈2% at small strain and ≈0.5% at ultimate. It is evident that the errors caused by temperature effects
on the load cells are very small and therefore temperature correction of the load cells is deemed
unnecessary.
B6.1.3 Fibre Optic (FO) Temperature Effects

The effect of temperature on the fibre optic (FO) strain and temperature gauges was also examined.
When assessing the effect of temperature, the location of the strain and FO temperature gauges
must be considered. For gauges above ground level (S1, S1B, T1, T1B) the effect of temperature
will be significantly larger (due to changes in air temperature and exposure to direct sunlight) than
below ground level where the temperature will remain relatively constant. It has previously been
noted however that the fibre optic temperature sensors (T1 – T4) vary with the strain and bending
applied to the pile and are therefore not suitable for temperature compensation at high levels of
strain. Analysis of the two sensors above ground level show they are affected by temperature and
match the thermocouple measurements (temp) well. All other gauge levels remain relatively
constant due to the relatively constant temperature below ground level. While there was a
reasonably good correlation between the above ground FO strain sensor and the above ground FO
temperature sensor, the FO temperature sensor could not be used (due to the effect of strain on the
temperature sensor during loading). Therefore, it is recommended that no temperature
compensation be applied to the strain gauges, and the top level of strain gauge measurements
(above ground) have been ignored due to temperature effects which cannot be accurately corrected.
B6.1.4 Inclinometer (SUI & IPI) Temperature Effects

The effect of temperature on the stick-up inclinometers (SUI) and in-place inclinometers (IPI) was
also investigated. Due to the high level of noise present on the inclinometer signal, the raw data was
filtered using a gaussian window function with a period of 1 minute. Using the filtered data there
appears to be a slight correlation with temperature such that there is a ≈0.003° change in angle per
degree Celsius change in temperature. It should also be noted that the change in temperature below
ground level was seen to be <0.2°C (from the below ground FO temperature gauge) and so the
change in IPI values (below ground level) is most likely due to cable or other temperature effects
and not directly related to the change in temperature of the instrument itself. The change in angle
noted during the test is of a similar magnitude to the accuracy and noise band of the instruments
and so it is recommended that the inclinometers do not need to be corrected for temperature effects.
B6.1.5 Extensometer (IPE) Temperature Effects

A similar procedure was applied to the extensometer data for pile DM1. It is found that there is
negligible change in strain (<1με / degree change in temperature) with temperature. It is therefore
deemed unnecessary to correct for temperature effects on the in-place extensometers.
Page 111/272
B6.2 Piezometer response
A number of Geosense VWP-3400 type drive-in piezometer pressure sensors were installed in the
vicinity of piles CM5, CM6, CL1 and CL2 at Cowden. The installation procedures and locations of
the piezometers relative to each pile along with the factual output data from the piezometers is
provided in the Field Testing Factual Report (AWG, 2015, pages 38 – 39 and 252 – 262). The layout
of the piezometers is shown in Figure 96. The piezometers are named according to the pile, depth
and location relative to their pile. For example, piezometer L1 – 3.1A refers to the piezometer beside
pile L1, at 3.1m depth, where A refers to the nearest set of piezometers to the pile (at 0.5m from the
nd
pile wall) and B refers to the 2 set of piezometers (at 2m from the pile wall). For the mid-size piles,
which have piezometers on both the front and back sides of the piles, C and D refers to the nearest
and furthest set of piezometers on the active side of the pile. It should also be noted that for the
large piles, the piezometers beside L1 are on the front side of the pile and the piezometers beside
L2 are on the back side of the wall relative to the line of loading. For the mid-size piles with
piezometers, pile CM6 was subjected to monotonic loading while pile CM5 was subjected to cyclic
loading. Due to the complication of interpreting the pore pressure response during cyclic loading, the
pore pressure response of CM5 is outside the scope of this report.
Figure 96: Layout of piezometer for mid-diameter and large-diameter piles
B6.2.1 Large-diameter pile installation

The time series of the piezometer data is presented in the Field Testing Factual Report (AWG,
2015). The depth profiles of the equalised piezometer measurements and those after pile installation
for the large piles are shown in Figure 97. The pre-installation equalised pore pressures are seen to
increase with depth between 30 – 50 kPa and match reasonably closely with the adopted in-situ
pore pressure profile used in the 3D FE analysis (Figure 97a). The piezometer measurements at
9.5m depth after pile installation show large increases in pore pressure of ~220kPa on the
piezometers 0.5m radial distance from the pile wall, with a smaller increase of ~80kPa at 2m from
Page 112/272
the edge of the pile. At 5.25m depth, smaller increases in pore pressure of ~70kPa and ~20 kPa
were noted at 0.5m and 2m from the edge of the pile respectively. At the shallowest depth of 3.1m,
very small increases in pore pressure of 5 – 10 kPa were noted.
(a) pile pre-installation equalisation (b) post-installation
Figure 97: Pore pressure measurements for CL1 and CL2 (D = 2.0m)
Randolph and Wroth (1979) proposed a simplified closed form analytical solution for the radial
consolidation of the soil around a driven pile based on the cylindrical cavity expansion method
(CEM). The initial pore pressure after driving was suggested to vary radially away from the pile as:
𝑅
𝑢0 = 2 𝑠𝑢 ln ( ) Equation 21
𝑟
𝑅 = 𝑟0 (𝛽𝐺/𝑠𝑢 )0.5 Equation 22
Where R is the width of the plastic zone (u0 = 0 for r > R), r0 is the pile radius and β is a modification
factor for open-ended piles (=ratio of net volume to gross volume of the pile). For the large piles in
Cowden β = 0.05. Estimating the operational value of G after pile installation is much more difficult
and will depend on the stress and strain levels imposed during installation. For this simplified
analysis, the range of G/su ≈ 80 - 400 would be approximately representative of Cowden till in at
strains of ≈0.05% - 1% (assuming no increase in mean stress during pile installation). Figure 98
shows the CEM solutions predict the pore pressure distribution well for G/su = 300 for the
piezometers at 5.25m and G/su = 80 for the piezometers at 9.5m.
Page 113/272
Figure 98: Radial distribution of pore pressure compared with CEM solution
B6.2.2 Large-diameter load tests

The piezometer response for Cowden L1 and L2 piles during static loading are shown in Figure 99
and are compared with predictions from the FE analysis at the piezometer locations. Analysis of
piezometer response during instrument installation and pile installation indicates that the response
time response of the different instruments can vary significantly, possibly depending on the type and
porosity of the material in or adjacent to the piezometer port. Instruments that feature a slow
response tend towards the appropriate steady state pore pressure measurement, but are not
informative for short term response. For the case of the large pile tests, several piezometers appear
to exhibit a slow response behaviour (L1-3.1A, L2-5.25A and L1-9.5A).
For the piezometers under ‘passive’ loading (Figure 99a) there is a very good match between the
FE and measured results with the exception of the piezometer at 3.1m depth (L1-3.1A). For the
piezometers under ‘active’ loading (Figure 99a) the FE predicts the pore pressure reduction very
well at 3.1m. In general, the FE predicts the pore pressure at the ultimate displacement well, but
tends to overpredict at low displacements, which may be attributed to the slow rate of displacement
(including unload cycles) during the early phase of the field testing, which is not considered in the FE
analysis. Profiles of the change in pore pressure (at ultimate displacement) with depth are shown in
Figure 100.
Page 114/272
(a) ‘passive’ sensor locations
(b) ‘active’ sensor locations
Figure 99: Comparison of FE predicted and measured pore pressures during loading
Page 115/272
(a) 0.5m from pile wall (b) 2m from pile wall
Figure 100: Profile of change in pore pressure during monotonic loading for piezometers
B6.3 Camera gap imaging
For each monotonic pile test the response of the gap forming behind the pile was monitored using a
series of images, captured periodically from a camera mounted on the pile. A ruler was placed
underneath the camera at ground level to allow estimation of the size of the gap from the image. The
size of the gap at ground level was estimated using functions available in the software package
Matlab, with the procedure outlined below.
B6.3.1 Procedure
The camera images were initially corrected for rotation of the camera during load testing using a 3D
affine transformation, where the angle of rotation was based on the estimated initial camera angle
combined with the measurement of pile rotation at ground level, matched with the timestamp from
the camera image. The gap edge was identified, using edge detection algorithms and colour
gradient thresholds in the ‘hsv’ (hue, saturation, value) colour spaces. As the gap is generally
considerably darker than the surrounding soil, the ‘value’ parameter in the hsv colour space was
used with a threshold value, as a means of identifying the gap edge. The images were then
converted to a binary gap image, which allowed simple measurements of the number of pixels in the
gap from the image matrix.
Page 116/272
Figure 101: Raw image (left) and binary gap image (right)
One of the main sources for error in estimating the size of the gap from the image was due to image
distortion due to the characteristics of the lens and the position of the camera in relation to the gap.
When estimating the size of the gap using the ruler placed at ground level, the length of a pixel at
ground level was initially estimated by selecting two points along the ruler of known length (typically
5cm) and counting the number of pixels in a straight line between the points. This method assumes
that the length per pixel across the whole image is constant and there is no image distortion
(uncorrected). However due to the position of the camera close to the gap it is evident from the
images that there is significant distortion around the outer edges of the image (as light enters the
outer edges of the lens). Calibration of the camera using a checker board can allow removal of the
radial lens distortion in an image, however, in the absence of the required information an alternative
method was used. Individual points at 1cm lengths along the ruler were selected for a number of
images throughout a load test. The distances between the points (in terms of pixels) was then
plotted against absolute radial distance from the centre of the image, allowing the variation in pixel
length across the image to be established. The integrated effect of the change in pixel length with
distance from the centre of the image was used to provide more accurate estimates of the gap size.
As the lens correction and edge detection algorithm is less reliable near the edges of the image, the
value of the gap measurement used in the following sections typically corresponds to an average
value taken over the centre 10% of the image (from 0.45 – 0.55 of the image width).
B6.3.2 Gap measurements

The uncorrected gap measurements (using 2 points at 5cm distance on ruler) and those corrected
using the method described above are compared to the ground-line displacement for the CM3 pile
test in Figure 102. It is evident that the uncorrected method tends to underestimate the displacement
at ground level, while in general the corrected method provides a good match with the measured
ground-line displacement from the factual report data. The same procedure was applied to all the
pile tests where good image data was available and all show similar results. The difference between
the corrected and uncorrected gap measurements show the potential errors due to lens distortion
and highlight the need for accurate camera calibration. In the absence of this information, it is
difficult to draw definitive conclusions from the gap images. However, it appears, from the
information available and using the unrefined correction techniques, that the soil movement on the
Page 117/272
active (back) side of the pile at ground level is minimal and the size of the gap which develops is
directly comparable to the ground line movement of the pile.
Figure 102: Gap measurement and ground line displacement for pile CM3
B6.4 Total station measurements
Throughout the course of the load testing campaign, survey measurements were taken with a total
station of the critical points on the pile, including all above ground instrument locations. The survey
data was used primary to accurately identify the instrument positions and heights for use in the
interpretation of the test data. The survey data also provides an additional check on the pile
displacements during load testing. The data from the total station is output in terms of Eastings,
Northings and Height above a reference datum. This data was then processed through a custom
Matlab program where direction of loading for each test was identified from survey points along the
line of loading and the Easting and Northing survey data was then processed using a 2D
transformation matrix to calculate the pile movements parallel and perpendicular to the line of
loading. A table of the pile movements measured by the LVDTs and calculated from the survey data
at the passive low (PL) and active low (AL) displacement gauge locations following the final load
step is shown in Table 19. It should be noted that there is some inaccuracy in the survey
measurements of the order of 1 – 2mm, but in general the survey data shows good agreement with
the measured LVDT data.
Table 19: Comparison of total station survey and instrument measurements at Dunkirk
Pile DM3 DM4 DM5 DM6 DM7 DM9

LVDT PL - Measured 87.55 93.55 95.72 83.15 67.74 74.20
LVDT AL - Measured 91.79 97.68 98.73 86.55 70.67 74.81
LVDT PL - Surveyed 90.69 95.66 99.74 84.05 65.29 -
LVDT AL - Surveyed 94.08 97.81 99.72 84.47 67.96 74.51
Error 2.94% 1.16% 2.51% -0.70% -3.87% -0.40%
Page 118/272
PISA Project – Final Report Appendix C - Development and validation of 3D finite element models for pile behaviour prediction
Appendix C Development and validation of 3D finite element

models for pile behaviour prediction
This appendix discusses the robustness of the developed finite element (FE) model in predicting the
behaviour of monopiles tested in the field, with the objective of establishing whether the FE
modelling can serve as a basis for developing a design methodology. In general, an FE model is
required to represent realistic ground conditions and soil behaviour, together with a realistic problem
geometry and boundary conditions, to be able to produce accurate predictions for the problem under
investigation.
Appendix A describes interpreted and modelled initial ground conditions at both the Cowden and
Dunkirk sites, in terms of soil layering, initial effective stresses, pore water pressures, relative
densities, undrained strength and maximum shear modulus profiles. It also presents the
interpretation of the most relevant aspects of the hydro-mechanical behaviour of Cowden clay till
and Dunkirk sand from recent experimental investigations, and how this information is used to
calibrate the chosen constitutive models for each material. Both models, a generalised extended
Modified Cam Clay (MCC) model and a bounding surface plasticity model, for clay till and sand
respectively, are capable of reproducing nonlinear small strain behaviour and failure conditions in
the soil. Derived parameters for both models are summarised in Appendix A.
This chapter discusses the discretisation of the problem geometry and application of appropriate
boundary conditions for simulating lateral loading of piles, followed by a comparison of the predicted
and measured response of the test piles.
C1 3D FE model and boundary conditions
C1.1 Geometry and FE mesh

Four test pile geometries are reanalysed here, following recalibration of Cowden till and Dunkirk
sand constitutive models based on the results from the new site investigation (SI) and laboratory
testing interpreted in Appendix A. The main geometric characteristics of the piles are summarised in
Table 20. Principal geometric parameters are pile diameter, D, embedded pile length, L, pile height
above ground surface (stickup height), h, and thickness of the pile wall, t.
Table 20: Geometry of reanalysed test piles

Analysis D h h/D L L/D t D/t Equivalent test Pile
(m) (m) (m) (mm) Cowden Dunkirk
T2 0.76 10 13.2 2.3 3 10 76 CM2 DM7
T4 0.76 10 13.2 4.0 5.25 13 / 14 58 / 54 CM9 DM4
T6 0.76 10 13.2 7.6 / 6.1 10 / 8 25 30 CM3 DM3
T7 2.0 10 5.0 10.5 5.25 25 / 38 80 / 52 CL2 DL2
*Dual numbers represent Cowden / Dunkirk geometric value
The problem geometry for simulating the lateral loading of a circular pile contains one plane of
symmetry, hence only half of the problem is discretised into a finite element mesh as shown in
Figure 103. This is a mesh for pile T6, with the size of the discretised domain being the same for all
medium diameter piles (D=0.76 m): 40 m depth to the bottom boundary of the mesh and 30 m radial
distance to the far vertical boundary. For a 2 m diameter pile the radial distance is 80 m, with the
same 40 m depth of the mesh.
Page 119/272
The soil domain in the mesh is discretised with 6464 20-noded hexahedral displacement-based
isoparametric solid elements. The interface between the pile and the soil on the outside of the pile is
simulated with 180 16-noded zero-thickness interface element (Day & Potts, 1994), whereas the pile
itself is discretised with 280 8-noded shell elements (Schroeder et al., 2007). In the Z-coordinate
direction the embedded part of each pile is discretised with 18 rows of elements, while 10 rows of
elements are in the stickup.
Consistent with the original pile test analyses, reported by the AWG (2014a and 2014b), the tubular
open-ended pile is modelled as an elastic material, with properties representative of steel: Young’s
modulus E = 200 GPa and a Poisson’s ratio of 0.3. The constitutive model also requires the
thickness of the pile wall as an input parameter, as shell elements have no physical thickness in a
FE mesh.
The interface between the pile and the soil is represented with an elasto-plastic Tresca model with
limited tensile capacity (i.e. zero) in the case of Cowden test piles, and with an elasto-plastic Mohr-
Coulomb model with zero cohesion in the case of Dunkirk test piles. In both cases interface
elements enable opening of a gap around the pile during lateral loading. The input parameters for
the two interface models are the elastic shear and normal stiffness, KS and KN, which are both set to
5 3
1.0x10 kN/m . The plastic part of the Cowden interface model is defined by the initial undrained
strength in the soil, whereas the Dunkirk interface model adopts the angle of shearing resistance of
o
the soil of 32 .
Figure 103: Typical FE mesh employed in the analyses of test piles
C1.2 Boundary conditions

To prevent rigid body movements of the mesh all three displacement components in the three
coordinate directions (X, Y and Z) are prescribed as zero over the bottom boundary of the mesh (at
Z = -40 m). In addition, the displacements normal to the vertical cylindrical boundary are also set to
zero. To ensure that the X-Z plane at Y = 0 is a plane of symmetry, the displacements in the Y-
Page 120/272
direction over this plane are set to zero, as are the rotational degrees of freedom with respect to the
X- and Z-axes along the edges of pile shell elements at the plane of symmetry.
The horizontal load at the pile top (i.e. Z= h) is applied in a displacement controlled manner, in that
the increments of displacement in the X-direction are applied uniformly around the pile perimeter at
Z = +10 m. The horizontal load, H, on the pile is then obtained as the reaction to the applied
displacements. A sufficient number of displacement increments are applied in each analysis to
provide a complete load-displacement curve for each pile, thus enabling the interpretation of the
shape of the parametrised p-y curves.
All analyses for the Cowden test piles are performed under undrained conditions, which are
achieved by prescribing a high value for the bulk modulus of the pore fluid (Potts & Zdravkovic,
1999). For the Dunkirk test pile analyses the soil is assumed to behave in a drained manner.
C2 3D FE predictions of pile response
The following sections show comparisons of measured and predicted response of test piles from
both the 3D FE analyses and the existing API/DNV method. The load-displacement comparison
includes both preliminary and final 3D FE predictions, in order to demonstrate the effect of some
aspects of soil behaviour emphasised by the new site investigation. The response of the embedded
part of test piles utilises only the final 3D FE predictions.
C2.1 Load-displacement response at Cowden

The new SI and laboratory testing at Cowden, interpreted in Appendix A, showed primary
differences between the historic and new soil data to be in the small strain stiffness (Figure 57 and
Figure 58 in Appendix A) and the undrained strength profile in triaxial compression (Figure 63 in
Appendix A). All constitutive model parameters and initial ground conditions have been derived
strictly from the available soil data obtained from the field and laboratory investigations. In particular,
care has been taken to ensure the model parameters are within the experimental scatter of this data.
Predictions of pile response have been obtained from a single set of soil parameters derived from a
combination of historic and new experimental data on Cowden till.
The geometries of the four test piles are sketched to scale in Figure 63 in Appendix A, indicating that
the difference in the interpreted strength in the surface crust is likely to most significantly influence
the predicted capacity in lateral loading for the very short CM2 and the medium length CM9
equivalent piles. Figure 104 and Figure 105 show the applied horizontal load on the pile versus the
horizontal pile displacement at mudline. Comparisons are made between the measured curve for
each pile, the predictions for that pile from the preliminary and final 3D finite element analyses and
the existing API/DNV method. Figure 104 shows comparisons for mudline displacements up to 10%
of pile diameter (which is taken as nominal pile failure), and Figure 105 shows the same
comparisons at very small displacements at the beginning of loading. The objective of Figure 105 is
to examine the success of the numerical model in reproducing the initial gradient of the load-
displacement curve.
The final 3D FE predictions in Figure 104 demonstrate very good agreement of the overall load-
displacement response with field test data, despite significant variability of the Cowden till deposit
(e.g. varied stone content with depth and varied degree of weathering with depth). The effect of the
originally assessed higher undrained strength in the crust is clearly seen in the preliminary 3D FE
predictions for piles CM2 and CM9.
Page 121/272
The recalibration of the constitutive model is further supported by the accuracy of the final 3D FE
predictions at small displacements. The derived higher stiffness from the new data in Figure 57 and
Figure 58 in Appendix A leads to good agreement of the initial gradients of the load-displacement
curves with filed test responses.
Overall, the developed 3D FE numerical model is sufficiently robust to predict the response of
different pile geometries with a high degree of accuracy. The API/DNV predictions are consistently
and substantially lower over the whole range of piles’ lateral displacements.
(a) CM2 (D = 0.762, L/D = 3) (b) CM9 (D = 0.762, L/D = 5.25)
(c) CM3 (D = 0.762, L/D = 10) (d) CL2 (D = 2.0, L/D = 5.25)
Figure 104: Comparison up to ultimate displacement of field measured response, 3D finite

element modelling and existing API/DNV methods
Page 122/272
(a) CM2 (D = 0.762, L/D = 3) (b) CM9 (D = 0.762, L/D = 5.25)
(c) CM3 (D = 0.762, L/D = 10) (d) CL2 (D = 2.0, L/D = 5.25)
Figure 105: Comparison at small displacements of field measured response, 3D finite

Figure 106 plots the accuracy of the modelled curve shape, using the quality of fit metrics defined in
section 4.4, when comparing to the simplified field test response curves, defined in Appendix F3.
These plots indicate that the 3D finite element modelling predicts the ultimate and small
displacement response of the field tests with average accuracies of 88% and 86% respectively. This
represents a high accuracy when compared to other numerical modelling campaigns and is
significantly higher than the accuracy of the existing API/DNV design methods of 49% and 21%
respectively.
Page 123/272
Equivalent pile Equivalent pile

D D
L/D L/D
(a) Ultimate response (η0.1D) up to vG = 0.1D (b) Small displacement response (ηsd) up to
vG = D/1,000
Figure 106: Accuracy metrics of field test response prediction for 3D finite element modelling
and the existing API/DNV approach
C2.2 Embedded pile response at Cowden

The figures presented in this section show the embedded pile response for the instrumented piles
CM2, CM3 and CL2, corresponding to the short, long and mid-length piles (within the parameter
space of this study). The embedded response has been computed following the methodology
described in Appendix F2, adopting an optimisation process that fits a piecewise spline to the
measurements of bending moment and inclination. The spline can be integrated to assess the
displacement and differentiated to determine the applied load. Also plotted on the figures are the
embedded response predictions derived from the 3D finite element and the existing API/DNV
modelling approaches.
It should be noted that the shear force and equivalent distributed loads are those that would be
required to generate the measured bending moment and rotation profile. This means that they not
only include the effect of the true distributed load, but also the distributed moment and base
components. The error in this approximation should be small for slender piles, but will become
significant for shorter piles, as the remaining components have a more significant influence on the
pile behaviour. The calculation of the applied load in this way is therefore only carried out for the
longer CM3 pile, where there is greater confidence that the equivalent distributed load is
representative of the loads applied to the test pile.
C2.2.1 CL2 pile response (D = 2.0m, L/D = 5.25)

Figure 107 - Figure 110 show the embedded response of the large diameter CL2 pile (D = 2.0m, L/D
= 5.25). Figure 107 and Figure 108 show the response following an early load step at small
moments and displacements, where the predicted ground level displacement is matched in Figure
107 and the moment at ground level is matched to the measured value in Figure 108. Figure 107(a)
shows that at a matched small displacement, the API/DNV approach predicts a significantly more
rigid pile response than that by the 3D finite element analysis. This indicates an under-estimate of
the soil stiffness, which is reflected in the under-prediction of bending moment shown in Figure
107(b). The significant under-prediction of the soil reaction and bending moment by the API/DNV
Page 124/272
approach is a direct cause of the under-prediction of pile load-displacement response seen in Figure
105.
(a) (b)
Figure 107: Comparison of embedded response for CL2 (D = 2.0m, L/D = 5.25) after load step 3
at a matching ground displacement vG = 9.5mm
Figure 108(b) shows that at a matched applied moment, the bending profile predicted by both the
API/DNV and 3D FE models has a good fidelity with the measured response, despite a marginal
over-prediction by both of the peak bending moment. The disparity in the load-displacement
response prediction is again evident in Figure 108(a), where the displacement required by the
API/DNV approach to reach the target load and moment is an order of magnitude greater than that
measured, and has been omitted to provide a detailed view of the fidelity in the displacement
response predicted by the 3D finite element approach.
(a) (b)
at a matching ground level moment MG = 7.3MNm
Page 125/272
Figure 109 and Figure 110 show the response at a later load step, with high load and displacement,
again where the predicted ground level displacement are matched in Figure 109 and the moment at
ground level is matched to the measured value in Figure 110. At this higher load, it is evident that
the 3D finite element approach achieves a good prediction of the shape and magnitude of both the
displacement and bending moment profile. However, Figure 109(b) again demonstrates the under-
prediction of the soil reaction and bending moment profile by the API/DNV approach. Due to the fact
that the API/DNV approach predicts complete pile failure prior to the ultimate response, only the 3D
finite element response is shown in Figure 110.
(a) (b)
at a matching ground displacement vG = 208mm
(a) (b)
at a matching ground level moment MG = 22.2MNm
Page 126/272
Based on Figure 107 - Figure 110, it is evident that where the load-displacement response is poorly
predicted, there is a great disparity between the fidelity of the displacement response and the
bending moment response. This poor prediction of load-displacement response can be more easily
observed in the earlier Figure 104 and Figure 105. The following plotted embedded responses are
therefore conducted only for matching values at ground level of the appropriate quantity, to provide a
fair measure of the embedded response fidelity.
C2.2.2 CM2 pile response (D = 0.762m, L/D = 3)

Figure 111 and Figure 112 show the embedded response of the short CM2 pile (D = 0.762m, L/D =
3) at a small and large displacements respectively. Note that during the spline fitting, the toe
moment is not constrained to 0kNm. Figure 112 demonstrates that at large displacements the 3D
finite element simulation predicts both the bending moment and displacement profiles with high
accuracy, indicating that the physics of the problem are well modelled. However, at both small and
large displacement the API/DNV approach significantly under-predicts the reaction at the pile toe,
which contributes to a poor prediction of the response shape.
(a) vG = 4.1mm (b) MG = 105kNm
Figure 111: Comparison of embedded response for CM2 (D = 0.762m, L/D = 3) after load step 3
Note that the spline fit shown in Figure 111(b) indicates a significant value of bending moment at the
pile toe. This behaviour is regarded as being strong conditioned by the likely errors in the strain
gauges associated with the small magnitudes of the measurements.
Page 127/272
(a) vG = 71.9mm (b) MG = 312kNm
Figure 112: Comparison of embedded response for CM2 (D = 0.762m, L/D = 3) after load step 8
C2.2.3 CM3 pile response (D = 0.762m, L/D = 10)

Figure 113 and Figure 114 show the embedded response of the long CM3 pile (D = 0.762m, L/D =
10). This pile represents a geometry that would be most appropriately modelled by the API/DNV
approach, as it is a relatively slender pile with a response dominated by lateral soil reactions. In
addition to the plots of displacement and bending moment profile, the equivalent lateral distributed
load peq at a matching matched ground level moment MG is plotted. In order to achieve a consistent
comparison, the equivalent distributed loads plotted for the 3D finite element is the p-v component
only.
As shown in Figure 113(a) and (b), at a small displacement and small load respectively, the 3D finite
element simulation predicts the shape and magnitude of the displacement and bending moment
response with good accuracy. However, at small displacements the API/DNV approach under-
predicts the curvature of the displacement profile, again indicating that the soil reactions are under-
predicted, which is reflected in the poor prediction of the load-displacement response in Figure 104
and Figure 105. Even at the matching load, shown in Figure 113(b), an underestimate of the soil-
reactions results in an over-estimate of the loads and bending moments transferred to greater
depths, which is reflected in the prediction of peq, in Figure 113(c).
Page 128/272
(a) vG = 3.9mm (b) MG = 536kNm (c) MG = 536kNm
Figure 113: Comparison of embedded response for CM3 (D = 0.762m, L/D = 10) at load step 3
At the larger loads and displacements shown in Figure 114, the API/DNV approach continues to
under-predict the soil reactions and pile curvature at matching ground level displacements, but at a
matching load, both the API/DNV and 3D finite element analysis achieve excellent predictions of
both the bending moment distribution and the soil reaction distribution, peq.
Page 129/272
(a) (b) (c)
Figure 114: Comparison of embedded response for CM3 (D = 0.762m, L/D = 10) at load step 9
(vG = 99.3mm, MG = 4247kNm)
By selecting displacements and equivalent distributed loads at regular load steps, it is possible to
build soil response curves at specified depths. Figure 115(a) shows a comparison of normalised
equivalent p-v curves extracted from the field test data with curves generated using the API/DNV
approach plotted as dashed lines (normalised using the expressions described in Table 5 of section
2 and using the strength and stiffness profile used for the 3D finite element calculations). Whilst the
predicted response of the soil at small depths shows a moderate correlation with the measured field
data, the increase in response with depth is not simulated. This is likely due to the combined factors
of an under-prediction of the ultimate value of soil reaction and the over-prediction of the strain at
which this response is mobilised. These findings indicate that the relatively good prediction of the
load displacement response for the long pile (CM3) by the API/DNV approach is largely due to the
reasonable prediction of soil response most influential on the resulting pile response, at shallow
depths. However, the physics of the model in general is poor and not able to adapt to the demands
of varied pile geometry.
Figure 115(b) shows a comparison of normalised equivalent p-v curves extracted from the field test
data with the lateral distributed load component extracted from the 3D finite element model. This
figure shows a good correlation at shallow and intermediate depths, but an under-prediction towards
the pile toe. This under-prediction is accommodated in the 3D finite element model by the additional
components of base shear and moment that will achieve a more accurate overall reaction to the pile.
Page 130/272
(a) API/DNV shown as dashed lines (b) 3D FE shown as dashed lines
Figure 115: Equivalent distributed lateral loads extracted from CM3 (D = 0.762m, L/D = 10)
field test data, with corresponding predicted soil reactions shown as dashed lines
C2.3 Load-displacement at Dunkirk

Figure 116 and Figure 117 show comparisons of measured and predicted response, from
preliminary and final 3D FE analyses as well as the existing API/DNV approach, of test piles at
Dunkirk.
Two immediate observations can be drawn from these comparisons: the accuracy of the final 3D FE
predictions is lower than that observed in the clay test simulations and the accuracy of the API/DNV
approach is signfiicantly greater than for the clay field tests. However, the fidelity of the 3DFE
predicted ultimate response over the range of geometries shows a greater consistency than that of
the API/DNV approach. Note that neither method was capable of predicting the premature failure of
the short pile (DM7) prior to the ultimate displacement of vG = 0.1D.
Page 131/272
(a) DM7 (D = 0.762, L/D = 3) (b) DM4 (D = 0.762, L/D = 5.25)
(c) DM3 (D = 0.762, L/D = 8) (d) DL2 (D = 2.0, L/D = 5.25)
Figure 116: Comparison up to ultimate displacement of field measured response, 3D finite

Very similar conclusions can be drawn from the small displacement predictions shown in Figure 117,
where the 3D finite element modelling shows a moderate improvement in response prediction over
the API/DNV approach.
Page 132/272
(a) DM7 (D = 0.762, L/D = 3) (b) DM4 (D = 0.762, L/D = 5.25)
(c) DM3 (D = 0.762, L/D = 8) (d) DL2 (D = 2.0, L/D = 5.25)
Figure 117: Comparison at small displacement of field measured response, 3D finite element
modelling and existing API/DNV methods
These observations are reflected in the quality of fit metrics, shown in Figure 118, where the 3D FE
achieves an average accuracy on the ultimate response of 81% compared to 70% achieved by the
API/DNV approach. The difference in accuracy at small displacements is marginally greater, with
72% and 42% average accuracy for the 3D FE and API/DNV approaches respectively.
Page 133/272
Equivalent pile
D
L/D
(a) Ultimate response (η0.1D) up to vG = 0.1D (b) Small displacement response (ηsd) up to
vG = D/1,000
Figure 118: Accuracy metrics of field test response prediction for 3D finite element modelling
and the existing API/DNV approach
The predictions with the new set of model parameters, as derived in Appendix A, are significantly
closer to the test data than those obtained using the preliminary calibration. This reflects two
important aspects of modelling pile behaviour in sands: the sensitivity of the overall response to
variations in model parameters and initial conditions, and the need to have a good characterisation
of the material using a combination of field and laboratory testing.
The first of these issues highlights the difficulty of predicting with confidence the response of sands,
where the mobilisable strength is directly related to the plastic volumetric deformations experienced
by the soil. Both these aspects of soil behaviour are controlled by the state parameter (i.e. the
difference in void ratio between the current state and critical state for the same mean effective
stress). Therefore, uncertainties in the position of the critical state line and, more crucially, in the
initial stress state and void ratio have the potential to have a considerable effect on the simulated
response. Indeed, the reassessment of the initial pore water pressure profile, which resulted in the
lowering of the water table (Figure 75 in Appendix A) and the consequent increase in mean effective
stress throughout the deposit, partly justifies the higher stiffness obtained with the new set of
parameters. Unfortunately, the substantial variation in the values returned by the different
correlations between the results of CPTs and relative density (as discussed in Appendix A), as well
as lack of sampling at depth, meant that a reinterpretation of the initial void ratio in the soil deposit
could not be carried out with higher confidence than that associated with the values reported by
Chow (1997) and adopted in the preliminary model calibration (AWG, 2014b).
The second of the aspects mentioned above – the need to combine field data with laboratory data –
arises mostly from the lack of high quality samples at a range of depths. Indeed, the observation that
the initial stiffness of the samples tested was lower than that measured in the SCPTs, suggested
that, rather than trying to reproduce directly the results of laboratory testing, the choice of
parameters should aim at simulating a more gradual transition between field stiffness at low strains
and measurements from triaxial tests at larger deformations, in order to compensate for sample
disturbance or other aspects of in-situ behaviour. This criterion led to the accurate reproduction of
dilatancy rates and mobilised peak strength (see Figure 74 in Appendix A), though the latter state
Page 134/272
took place at smaller strains in the simulations, reflecting the effect of higher stiffness expected in
the field. More importantly, the increase in the number of triaxial tests available, compared to the
original calibration, allowed a much more complete characterisation of the behaviour of the soil at
Dunkirk.
In particular, the execution of tests at a range of initial relative densities and mean effective stress
levels (see Table 14 in Appendix A) allowed the calibration of the component of the model
controlling the peak strength to be carried out with greater confidence than what was possible for the
preliminary set of parameters, where only one density was available and the mean effective stress
levels to which the samples were subjected were all considerably higher.
Lastly, unlike in the case of the Cowden till model calibration where lateral pile loading tests from
previous research programmes were available for validation of the original model calibratoin (AWG,
2014a), the only available tests for the validation of the original Dunkirk model calibration were those
for axially loaded piles, the behaviour of which is controlled by a significantly different mechanism.
Although a very good prediciton of the axial pile response was achieved with the originally derived
set of model parameters (see AWG, 2014b), the axially loaded pile test had a limited value as a
validation tool.
C2.4 Embedded pile response at Dunkirk

The following section shows the embedded pile response for the instrumented DM7 (D = 0.762m,
L/D = 3), DM4 (D = 0.762m, L/D = 5.25), DM3 (D = 0.762m, L/D = 8) and DL2 (D = 2.0m, L/D =
5.25). In a similar fashion to the response for clay, shown in section C2.2, the embedded response
has been calculated using the methodology described in Appendix F2. This adopts an optimisation
process for fitting a piecewise spline to the measurements of bending moment and inclination,
enabling integration for displacement and differentiation for applied load. The calculation of the
applied distributed load is only carried out for the longer DM3 pile, where there is greater confidence
that the equivalent distributed load is representative of the loads applied to the test pile.
C2.4.1 DL2 pile response (D = 2.0m, L/D = 5.25)

Figure 119 - Figure 122 show the embedded response of the large diameter DL2 pile (D = 2.0m, L/D
= 5.25). Figure 119 and Figure 120 show the response following an early load step with small load
and displacement, where the predicted ground level displacements are matched in Figure 119 and
the moments at ground level are matched to the measured value in Figure 120. Figure 119 shows
that at a matched small displacement, both API/DNV and 3D finite element approaches predict very
similar response, with an under-estimate of the soil reactions, resulting in an under-prediction of both
the displacement curvature, as well as the bending moment profile.
Page 135/272
(a) (b)
Figure 119: Comparison of embedded response for DL2 (D = 2.0m, L/D = 5.25) after load step 2
at a matching ground displacement vG = 10.1mm
At a matched moment, as shown in Figure 120, both the 3D finite element and API/DNV approaches
over-estimate the pile displacements, which is a result of the under-estimated pile stiffness and load-
displacement response seen in Figure 116 and Figure 117. Figure 120(b) shows that whilst the
API/DNV approach over-predicts the peak bending moment, both approaches under-predict the
bending moment at the pile toe.
(a) (b)
at a matching matched ground level moment MG = 10.8MNm
The embedded response at a larger matched displacement is shown in Figure 121, which shows
that whilst the shape of the displacement profile (and therefore the corresponding pile-soil stiffness
distribution) is well predicted, the inaccuracy in the load-displacement response results in over and
Page 136/272
under-predictions of the bending moment profile for the API/DNV and 3D finite element methods
respectively.
(a) (b)
at a matching ground displacement vG = 110mm
At the matched moment, as shown in Figure 122, the 3D finite element approach achieves a good
fidelity on the shape of the bending moment profile. Whereas the distribution of the soil reactions
predicted by the API/DNV approach result in a peak in bending moment at a greater depth.
(a) (b)
at a matching matched ground level moment MG = 34.1MNm
In a similar fashion to the clay analysis in section C2.2 the following plotted embedded responses
are conducted only for matching values at ground level of the appropriate quantity, to provide a fair
measure of the embedded response fidelity.
Page 137/272
C2.4.2 DM7 pile response (D = 0.762m, L/D = 3)

Figure 123 and Figure 124 show the embedded response of the short DM7 pile (D = 0.762m, L/D =
3) at an early load step and a later load step respectively. Note that during the spline fitting, the toe
moment is not constrained to 0kNm. Figure 123(a) demonstrates that at a matched displacement,
both the API/DNV approach and 3D finite element approach under-predict the pile stiffness.
However, at a matched moment, the 3D finite element approach achieves a good correlation with
the shape of the measured bending moment profile towards the soil surface. Both API/DNV and 3D
finite element analysis under-predict the soil reaction at the pile toe.
(a) vG = 3.6mm (b) MG = 211kNm
Figure 123: Comparison of embedded response for DM7 (D = 0.762m, L/D = 3) after load step 2
At the later load step shown in Figure 124, the API/DNV approach predicts a pile failure before
reaching the measured load and so the response of the 3D finite element analysis is shown only.
The 3D finite element approach achieves an excellent correlation in terms of both magnitude and
shape of the displacement and bending moment profiles.
Displacement (mm) Moment (kNm)

-200 0 200 400 600
-40 -20 0 20 40 60
0
0
0.5 0.5
1 1
Depth (m)
Depth (m)
1.5 1.5
Surface M
2 Surface v 2 Strain gauges
Spline fit Spline fit
3D FE 3D FE
2.5 2.5
(a) vG = 45.6mm (b) MG = 497kNm
Figure 124: Comparison of embedded response for DM7 (D = 0.762m, L/D = 3) after load step 6
Page 138/272
C2.4.3 DM3 pile response (D = 0.762m, L/D = 8)

Figure 125 and Figure 126 show the embedded response of the long DM3 pile (D = 0.762m, L/D =
8). In a similar fashion to pile CM3 in the clay tests, DM3 represents a geometry that should be
relatively well conditioned to modelling with the API/DNV approach, as it is a relatively slender pile
with a response dominated by lateral soil reactions. In addition to the plots of displacement, rotation
and bending moment, the equivalent lateral distributed load peq required to achieve the measured
embedded response is plotted. In order to achieve a consistent comparison, the equivalent
distributed loads plotted for the 3D finite element is the p-v component only.
As shown in Figure 125(a), at a matched small displacement both the 3D finite element and
API/DNV approaches under-predict the soil stiffness and pile curvature. This is also observed in the
predictions at a matched ground level moment, where both prediction approaches under-estimate
the soil stiffness at shallow depths in Figure 125(c), resulting in a greater transfer of load and
bending moment towards the pile to shown in Figure 125(b).
(a) vG = 3.6mm (b) MG = 777kNm (c) MG = 777kNm
Figure 125: Comparison of embedded response for DM3 (D = 0.762m, L/D = 8) at load step 2
Figure 126 shows that at a higher matched moment, the 3D finite element approach achieves a
good prediction of the bending moment profile, whilst the API/DNV approach significantly under-
predicts the soil reactions towards the soil surface, transferring a very large proportion of the load
and bending moment to significantly greater depth. Figure 126(b) demonstrates the difference in the
predicted bending moments at high load, which would result in significantly different design of pile
wall thickness distribution between the API/DNV and 3D FE approaches.
Page 139/272
(a) vG = 49.1mm (b) MG = 4.4MNm (c) HG = 4.4MNm
Figure 126: Comparison of embedded response for DM3 (D = 0.762m, L/D = 8) at load step 7
Page 140/272
PISA Project – Final Report Appendix D - Detailed design method application to a stiff over-consolidated clay site
Appendix D Detailed design method application to a stiff over-

consolidated clay site
D1 Introduction
The following section sets out an example application of the numerical-based method, for a soil
profile that resembles those seen at some North Sea clay sites. The approach is described in some
detail to provide a template for future applications of the method. The resulting soil reaction curves
can be used to conduct rule-based analyses at clay sites with comparable undrained strength and
stiffness profiles.
Initially, the ground conditions at the example site are established. These conditions are based on
the profiles of strength, stiffness, pore water pressure, OCR and 𝐾𝑜 that have been measured at the
Cowden low plasticity glacial till test site. Minor modifications are applied to these profiles (principally
on the initial distribution of pore water pressures) to provide a closer representation of the
comparable conditions in the deeper quaternary layers at North Sea sites.
Next, a set of 3D finite element models of an embedded monopile are developed to cover a range of
expected values of pile dimension and load eccentricity. Lateral loading computations are conducted
and the results are interrogated to extract the various soil reaction components that act on the pile.
These 3D finite element analyses were conducted using the finite element software ICFEP (Potts
and Zdravkovic, 1999, 2001), which employs a modified Newton-Raphson nonlinear solver with an
error-controlled sub-stepping stress-point algorithm.
Finally, the computed soil reaction curves are represented by a set of equations. These equations
are then used, in conjunction with a 1D model of an embedded monopile, to compute the
performance of a monopile for any geometric configurations that fall within the parameters space
explored using the 3D model.
D2 Assumed ground conditions

The initial ground conditions in this design example are based on those estimated at the Cowden
site and described in Appendix A. They are idealised, however, for closer representation of the more
likely conditions at a similar deep clay till in the North Sea. The adopted profiles are labelled as
‘idealised clay till’ in Figure 127. For comparison, the adopted Cowden test site profiles are also
shown in the figure. The principal change is that of the pore water pressure distribution which is
assumed hydrostatic from the ground surface (Figure 127(a)). The undrained strength profile in
triaxial compression in Figure 127 (b) is also assumed to have a constant gradient below 5 m depth,
which is the same as the gradient of the Cowden profile in a deeper deposit. As a consequence of
higher pore pressures, the effective stresses in the idealised deposit are reduced, resulting in a
slightly different K0 (Figure 127 (c)) and OCR (Figure 127 (d)) profiles (compared to the Cowden
site) necessary to reproduce the idealised su profile in triaxial compression in Figure 127 (b).
The resulting G0 profile (Figure 128(a)) remains a function of the mean effective stress and is also
slightly altered by different initial effective stresses in the design example. The assumed normalised
stiffness degradation curve in Figure 128(b) retains the same shape as that for the Cowden site.
Page 141/272
(a) (b)
(c) (d)
Figure 127: Soil profiles for clay design example
Page 142/272
(a) (b)
Figure 128: Assumed maximum stiffness profile and stiffness degradation curve for the clay
D3 Constitutive models
An extended generalised version of the nonlinear elasto-plastic Modified Cam Clay model is adopted
for the soil. The model utilises a Hvorlsev surface on the dry side (Potts & Zdravkovic, 1999;
Tsiampousi et al., 2013) and the general expression of Van Eekelen (1980) for the soil strength
variation in the deviatoric plane. It also simulates the nonlinear degradation of soil stiffness with
strain and its dependence on stress level. The model is calibrated against the test data from the
Cowden site, as explained in Appendix A.
The interface between the pile and the soil is represented with an elasto-plastic Tresca model with
zero tensile capacity to enable opening of a gap around the pile during lateral loading. The shear
strength in compression is set equal to the local undrained soil shear strength, su. The elastic part of
5 3
the interface model is defined by the shear and normal stiffness which are both set to 1.0x10 kN/m ,
ensuring that the pile-soil interface is rough (shear stiffness) and that there is no deformation in
compression across the element (normal stiffness). The monopile is modelled as an elastic material,
with properties representative of steel: Young’s modulus E = 200 GPa and Poisson’s ratio ν = 0.3.
The pile wall thickness is specified as an additional constitutive parameter.
D4 Specification of 3D calibration analyses

Table 21 summarises the geometries selected for the 3D calibration analyses. The chosen pile
diameters are in the range of 5 to 10 m, covering current monopile geometries (D = 6 to 7 m) as well
as likely future developments with monopile diameters of up to 10 m. Figure 129 shows that the
selected normalised pile length and thickness are focussed on shorter and relatively thinner pile
geometries, which are anticipated to occur more frequently as pile and turbine geometries increase.
Load eccentricities h have been selected to cover a range of wave and wind loading, where it is
Page 143/272
understood that realistic load cases will feature a combination of the two and will fall within the
specified range.
The use of this relatively large parameter space means that the resulting soil reaction curves will
have a wide applicability when used in rule-based mode. In any practical application of the
numerical-based method, it is advisable to adopt an appropriate reduced parameter space, to bound
more closely the expected pile diameter and length and the load eccentricity. In the current set of
calibration analyses, separate calculations on similar piles (C1/C3, C6/C7) are conducted with
different values of pile wall thickness, t. In any practical application of the method it is considered
that variations in pile wall thickness do not need to be included within the calibration study (on the
basis that (a) pile wall thickness has only a minor influence on the behaviour of the pile and (b)
variations of pile wall thickness are accounted for directly within the 1D model).
Table 21: Geometry and loading adopted in the 3D finite element calibration
Analysis D h L t
h/D h/L L/D D/t
C1 10 50 5 20 2.5 2 91 110
C2 10 150 15 20 7.5 2 91 110
C3 10 50 5 20 2.5 2 125 80
C4 10 50 5 60 0.83 6 91 110
C5 10 150 15 60 2.5 6 91 110
C6 5 25 5 10 2.5 2 45 110
C7 5 25 5 10 2.5 2 83 60
C8 5 25 5 30 0.83 6 45 110
C9 5 75 15 30 2.5 6 45 110
C10 7.5 37.5 5 15 2.5 2 68 110
C11 7.5 37.5 5 45 0.83 6 68 110
140
Future
120
Jacket
100 Current
Parametric study
80
D/t
60
40
20
0
0 2 4 6 8 10 12 14
L/D
Figure 129: Normalised parameter space of pile length vs thickness, showing existing, future
and jacket pile geometries, as well as the parametric study geometry range. Data provided by
DONG Energy in 2013
Page 144/272
D5 Finite element mesh and boundary conditions

Lateral loading of a circular pile exhibits one plane of symmetry and it is therefore sufficient to
discretise only half of the problem. The FE mesh used for the C4 pile is shown in Figure 130 as an
example. The soil domain is discretised with 10530 20-noded hexahedral displacement-based
isoparametric solid elements. The interface between the pile and the surrounding soil is simulated
with 360 16-noded zero-thickness interface elements (Day & Potts, 1994), whereas the pile itself is
discretised with 600 8-noded shell elements (Schroeder et al., 2007). In the Z-coordinate direction
each pile is discretised with 30 rows of elements below the mudline and 20 rows of elements above.
Similar meshes are employed for the other piles in the calibration analyses.
Figure 130: Finite element mesh for monopiles in Table 21 (dimensions in brackets for pile
C4, D = 10, L/D = 6). Note that the coordinate system used to define the finite element mesh
(capital Z positive in the upwards direction) differs from the coordinate system used to define
the 1D model (in which low case z is positive downwards).
To prevent rigid body movements of the mesh all three displacement components in the three
Z = -100 m). In addition, the displacements normal to the vertical cylindrical boundary are also
prescribed to be zero. To ensure that the X-Z plane at Y = 0 is a plane of symmetry, the
displacements in the Y-direction over this plane are set to zero, as are the rotational degrees of
freedom with respect to X- and Z-axes along the edges of pile shell elements in the plane of
symmetry. The horizontal load at the pile top (i.e. at 𝑍 = ℎ) is applied in a displacement controlled
manner, such that the increments of displacement in the X-direction are applied uniformly around
the pile perimeter. The horizontal load, H, on the pile is then obtained as the reaction to the applied
displacements. An appropriate number of displacement increments are applied to provide a
complete load-displacement curve for each pile, thus enabling the interpretation of the shape of the
extracted soil reaction curves. All analyses are performed under undrained conditions, achieved by
prescribing a high value for the bulk modulus of the pore fluid (Potts & Zdravkovic, 1999).
D5.1 Consideration of gap opening
The interface elements employed in the finite element model at the pile/soil boundary allow the
possibility of gap formation on the active side of the pile, near to the ground surface. Since the
Page 145/272
detailed performance of the pile is strongly influenced by near surface soil/structure interaction
effects, inclusion of the possibility of gapping was a key element of the modelling process.
In the development of finite element models used for assisting the design of the test piles, any gap
formed around the pile was assumed to be filled with air (on the basis that the tests were conducted
onshore with the water table at some depth the surface). This same assumption was also employed
for the calibration analyses outlined in this Appendix, and used as a basis for the soil reaction curve
parameters. The process of deriving these parameters and the validation of the soil reaction curves
when interpolating within the calibration parameter space is demonstrated.
However, in an offshore environment, any gap on the active side of the pile will fill with water and, as
a consequence, additional lateral loading will be applied to the monopile. Additional analyses have
demonstrated that, as expected, the inclusion of water within the gap slightly reduces the ultimate
capacity. These analyses have also shown that the behaviour of a pile with a water-filled gap can be
approximated using the air-filled analyses with hydrostatic pressure applied to the pile at locations
where gaps have formed. This approach is adopted at the end of this Appendix, to provide soil
reaction curve parameters modified for a water filled gap, for offshore design.
In any future application of the numerical-based approach, for the offshore application, it is
recommended that interface element formulations are used in which any gap formed on the active
side of the pile is assumed to fill with water.
D6 3D Finite element results

Figure 131(a) shows the relationship between the lateral load and the deflection at the mud-line, for
the 10 m diameter piles specified in Table 21. Note that these displacements are significantly larger
than would normally be considered in design, where typically 10% of the pile diameter (i.e. 1m for
these piles) is defined as ULS failure. These relatively large displacements have been employed in
the current example to enable the various soil reaction curves to be fully defined. In addition, Figure
131(b) and (c) show the deflected shapes of the D=10 m piles, with depth z normalised by pile
length L, for the cases when the horizontal mud-line displacement reaches, respectively, 0.001D
(early loading stage) and 0.1D (nominal failure). Principal observations from Figure 131 include: (a)
the load capacity increases with increasing pile length and reduces with increasing stick-up; (b) short
piles deform in an essentially rigid manner whereas the longer piles are seen to deform in a more
flexible manner; (c) the pile wall thickness has a negligible influence on behaviour.
Page 146/272
(a)
(b) (c)
Figure 131: D = 10 m monopiles: (a) load-displacement curves; pile deflections for mudline
horizontal displacement at (b) 0.0001D and (c) 0.1D
D7 Comparison of 3D finite element and existing ‘p-y’ method predictions

An exercise has been conducted to compare the computed pile response with results obtained using
the traditional p-y approach based on the API/DNV p-y curves for clay. The calculation of the pile
response using the API/DNV approach has been conducted through an independent analysis by
DONG Energy, as described in Appendix F4. Comparisons of this sort are not required for routine
application of the numerical-based method. They are, however, provided here to illustrate the
benefits of the numerical-based method compared with current design methods.
Page 147/272
Comparisons between the API/DNV approach and the finite element analyses are principally on the
following calibration analyses:
 C1 – large diameter (D = 10), short (L/D = 2)

 C4 – large diameter (D = 10), long (L/D = 6)
 C6 – small diameter (D = 5), short (L/D = 2)
As shown in Figure 132 and Figure 133, the API/DNV approach significantly under-predicts the
ultimate load and the initial stiffness for both short and relatively long monopiles, as well as
monopiles of small and large diameter.
(a) C1 (D = 10, L/D = 2) (b) C4 (D = 10, L/D = 6) (c) C6 (D = 5, L/D = 2)
Figure 132: Comparison between 3D finite element and API/DNV predictions
Figure 133: Comparison between 3D finite element and API/DNV predictions at small
displacements
Page 148/272
To quantify the difference in area below the computed load-displacement responses a ‘quality of fit’
metric has been adopted, as described in section 4.4 of the main report. Note that the range over
which the small displacement is calculated for comparison with the 3D FE results is D/10,000, which
th
corresponds to approximately 1/200 of the ultimate load.
Figure 134(a) demonstrates that the shape of the complete response curve (determined from the 3D
finite element models) is predicted by the API/DNV ‘p-y’ method with an average accuracy of 37%.
The accuracy is lower for the shorter piles (L/D = 2) and greater for the longer piles (L/D = 6). The
API/DNV approach performs less well at small displacements, predicting with an average accuracy
metric of 16%. The accuracy metric is as low as 5% for some short piles. It is observed that
variations in load eccentricity and pile wall thickness have relatively little effect on prediction
accuracy.
It is commonly observed that the fundamental natural frequency of wind turbines founded on
monopile foundations is greater than would be expected on the basis of predictions made using the
API p-y method, suggesting that the API approach systematically under-predicts the stiffness of the
foundation. This observation is consistent with the data in Figure 134(b).
Page 149/272
Pile
D
L/D
M/HD
D/t
(a) ultimate response (vG = 0.1D) average η0.1D of 41%
Pile
D
L/D
M/HD
D/t
(b) small displacement (vG = D/10,000) average ηsd of 15%
Figure 134: Metrics for the accuracy of prediction of the parametric study using the API/DNV
‘p-y’ approach
D8 Extraction of soil reaction curve

A soil reaction curve for a given depth may be determined from forces and displacements at element
nodes or Gauss points, but must be summed over each element before being used to calculate the
corresponding depth-wise soil reaction. In this study, the soil reaction curves are determined by
systematically extracting stresses from the interface elements around the perimeter of the pile and
also on the basis of computed nodal forces acting on the shell elements and the soil elements at the
base of the pile. Local displacements and rotations of the pile are determined from the
corresponding Gauss point and node displacements. The components of the 3D model that are
relevant to this soil reaction curve extraction process are highlighted in red in Figure 135(b).
Page 150/272
H PILE
SHELLS
INTERFACE
ELEMENTS
TOE SOIL
ELEMENTS
(b) Section view of mesh with

(a) Typical FE mesh
element types
Figure 135: Finite element simulation mesh and element types
D8.1 Boundary equilibrium check on model convergence
The first step in analysing the finite element results is to check that the computed nodal forces
applied to the pile are in equilibrium. This is achieved by summing the nodal forces on a boundary
around the pile (as shown by the red highlighted nodes in Figure 135(b)), and confirming that the
sum is equal to the applied load (to within an acceptable tolerance). The maximum proportional
difference between the sum of the boundary forces and the applied load for the finite element
calibration simulations up to a ground level displacement of vG = 0.1D are shown in Figure 136. It
should be noted that these maximum differences tend to occur towards a ground displacement of vG
= 0.1D and are generally significantly lower for small and intermediate displacements. It should also
be noted that this difference is only local, applying to the pile section of the mesh (Figure 135 (b))
from which the FE data is extracted. The error is worsened by the presence of a gap in interface
elements, as in this locality any residual stresses at integration points of “gapped” elements have to
reduce to zero during the solution process. This would require exceptionally tight tolerances which
would increase analyses run-times. This was deemed unnecessary as the overall convergence for
the whole FE mesh is much smaller than 2%.
Page 151/272
Pile
D
L/D
M/HD
D/t
Figure 136: Maximum percentage difference between sum of boundary forces and applied
load
D8.2 Component extraction
Four separate soil reaction components are assumed to act on the pile (e.g. see Figure 35 Section 5
in the main report).
The distributed lateral load p (units of force / length) is calculated by integrating the y-component of
the computed normal and shear stresses in the interface elements at a given depth. The
corresponding displacement v is calculated from the average of the gauss point displacements
around the pile perimeter at the corresponding depth.
The distributed moment m (units of force x length / length) is calculated by integrating the product of
the vertical components of computed shear stress and the lateral distance of the gauss points from
the neutral axis of the pile, using Equation 23. The corresponding rotation of the pile cross section 
is calculated by computing a least squares fit of an inclined plane about the x-axis to the computed
interface element displacements at the appropriate depth.
𝐷2 2𝜋
d𝑀 = ( ∫ 𝜏 𝑐𝑜𝑠𝜙 𝑑𝜑) 𝑑𝑧 = 𝑚 𝑑𝑧
4 0 Equation 23
where 𝜏 is the local vertically-acting shear stress and 𝜙 is the azimuthal angle, defined with respect
to the loading direction (see Figure 36(a) in section 5 of the main report).
The base horizontal force HB (units of force) is calculated from the sum of the y-component of nodal
force for the remaining nodes across the base of the pile boundary (shown in Figure 135(b)). This
includes contributions from the nodes on the soil elements across the toe of the pile, the nodes at
the base of the pile shell elements and any remaining interface nodes on the boundary that were not
used for the distributed load calculations. The corresponding displacement v is calculated by taking
the average of the displacements in the y-direction of the pile nodes at the base of the pile.
Page 152/272
The base moment MB (units of force x length) is calculated with reference to the same nodes as
those for the base horizontal force, but summing the product of the vertical component of force and
the lateral distance from the pile neutral axis. The corresponding rotation of the toe cross section 
is calculated by minimising the root mean square error for an inclined plane about the x-axis that is
fit to the pile base nodal displacements.
Examples of the distributed soil reactions extracted at selected depths of the C4 simulation (D =
10m, L/D = 6) are shown in Figure 137(a) and (b) and the base horizontal force and moment
components are shown in Figure 137(c) and (d).
(a) distributed load-displacement (p-v) (b) distributed moment-rotation (m-)
(c) base horizontal force-displacement (HB-v) (d) base moment-rotation (MB-)
Figure 137: Example soil reaction curves extracted from C4 (D = 10, L/D = 6) 3D FE simulation
Page 153/272
D8.3 Correction for reversing displacements
When implementing soil reaction curves in a numerical scheme it is necessary that the response is
monotonic, so that a given displacement corresponds to a single load. However, by simulating
realistic behaviour of a pile under monotonic lateral loads, the response at some depths is not
monotonic. Figure 138 plots the interpreted deflected shape from field measurements of the long pile
in clay, CM3 (D = 0.762m, L/D = 10), at two different load levels. Figure 138(a) shows that as the
pile flexes during early load steps some points on the pile move in the opposite direction to the
applied load (e.g. z = 3.5m). Once the load has overcome the shallow soil reactions and the pile
adopts a more rotational deflected shape, as shown in Figure 138(b), these points often deflect in
the direction of the applied load.
(a) Load step 1 (HG = 17.5kN) (b) Load step 8 (HG = 319.4kN)
Figure 138: Pile deflection profiles interpreted from field measurements of CM3 (D = 0.762m,
L/D = 10)
This change in the sign of the displacement means that the soil reaction curves at a given depth will
not be monotonic, as shown in Figure 139, and so it is necessary to extract the realistic
characteristics of the response to obtain a monotonic curve. Whilst the initial response may be in a
different direction to the final displacement, it is assumed to occur at the correct stiffness. If the
reversed curve regains the correct sign of response at a negative strain, as shown in Figure 139(b)
then it is assumed that the reaction responds at the initial stiffness until the raw curve is met.
However, if the reversed curve regains the opposite sign of response at a positive strain, as shown
in Figure 139(a), a different approach must be adopted. Several options are available, but for this
analysis it has been decided to reverse the sign of the displacement and reaction for the initial
response, up to the point at which the pile reverses its direction of displacement. A linear response
then links this reversed section to the point of the next highest reaction on the raw curve. Whilst this
approach can yield a curve that looks disjointed, it is worth noting that the magnitude of the reversed
displacements is generally very small and the effect on the overall response is almost
unperceivable. The result of this process is a set of soil reaction curves that exhibit an appropriate
initial and ultimate response and can be applied within a numerical scheme.
Whilst the majority of soil reaction which feature reversals can be corrected, the reaction at the final
pile pivot depth (e.g. z ≈ 4.8m in Figure 138(b)) is unlikely to obtain sufficient displacement to
provide a monotonic response curve. These soil reactions are omitted from further processing.
Page 154/272
(a) opposite sign zero strain (b) correct sign zero strain
Figure 139: Reversing soil reaction components extracted from FE simulation of CM3
(D = 0.762m, L/D = 10)
D8.4 Normalisation of soil reactions
The soil reaction curves extracted from the 3D analysis are normalised using the dimensionless
groups listed in Table 5 of Section 5. The local soil strength and stiffness data used in this
normalization process are calculated from the average value over the height of each element, based
on the idealised profile shown in Figure 127(b) and Figure 128(a). An example set of curves
normalised in this way (for the C4 simulation) is shown in Figure 140.
(a) Distributed load curves (b) Normalised distributed load curves
Figure 140: Distributed load curves from analysis C4 (D = 10, L/D = 6)
Page 155/272
D9 Application of normalised soil reaction to the 1D model
D9.1 1D Model framework
Key features of the 1D finite element model are illustrated in Figure 141. Timoshenko beam theory is
used to represent the pile. The model is formulated within a variational framework; this provides a
consistent procedure to incorporate the soil reaction curves within the analysis. This modelling
approach is, essentially, a generalisation of the conventional ‘p-y’ method. To reflect the general
nature of the analysis framework, however, the nomenclature ‘p-y curve’ is avoided (on the basis
that it implies a specific coordinate system and is not readily generalized to other components of soil
reaction). Furthermore, consistent with normal conventions in applied mechanics, the lateral
displacement of the pile is denoted v (rather than y).
MG
HG
z v
Lateral soil
reaction
p(z,v)
Distributed
moment
m(z,)
Base shear
force HB(vB) Base moment
MB(B)
Figure 141: Key features of the 1D finite element model

In the conventional p-y approach, the soil/pile interaction is represented by a lateral distributed load
applied along the length of the pile. A lateral load soil reaction curve, which has a similar function to
a conventional p-y curve, is adopted in the current model. The model is extended to include a
distributed moment acting along the embedded length of the pile, a base horizontal force, HB, and a
base moment, MB. These extensions to the conventional p-y approach follow previous work by
Davidson (1982), Lam and Martin (1986) and Lam (2013) for the design of caissons, principally for
onshore applications.
Page 156/272
The pile is represented in the 1D model as a line of beam finite elements. The soil deformations
(rotation and displacement) are prescribed to conform to the pile displacements (and cross-section
rotations) along the embedded length. The soil is represented in the formulation by a separate set of
‘soil finite elements’ that are connected to the embedded length of the pile. The soil response is
incorporated within the analysis, on the basis of the Winkler assumption, using specified functions
for the various soil reaction curves. The soil reaction curves are incorporated within the model in a
way that is analogous to the incorporation of constitutive models within a conventional finite element
framework.
Further details on the formulation and implementation of the 1D model can be found in Appendix F1.
D9.2 Validation of applying extracted soil reaction curves to the 1D model
Figure 142 shows a comparison for the C1, C4 and C6 geometries between the load-displacement
responses directly computed by the 3D finite element model and those predicted when applying the
extracted and normalised soil reactions to the 1D model; this form of model is referred to below as
‘1D (numerical)’. These figures confirm that, as expected, when the soil reactions determined from
the 3D finite element model are incorporated within the 1D model, the overall computed response of
the pile is similar to that obtained using the 3D model. The small differences between the predicted
responses of the 3D finite element and 1D models are attributed to a combination of factors
including residual stresses in the 3D finite element modelling, the correction of non-monotonic
interpreted soil reaction curves and any inaccuracies in the assumed physics of the Timoshenko
beam elements of the 1D model.
Figure 142: Load-displacement response comparison of 3D finite element and 1D model using
3D FE derived and normalised soil reactions
The 1D model is found to reproduce the shape of the computed 3D finite element response, for large
and small displacements, with average qualities of fit of 95% and 97% respectively, over the range
of calibration simulations (see Figure 143).
Page 157/272
Pile
D
L/D
M/HD
D/t
Pile
D
L/D
M/HD
D/t
Figure 143: Metrics for the accuracy of prediction of the parametric study when applying 3D
finite element derived soil reactions to the 1D model
Page 158/272
(a) Ultimate response (b) Small displacement

calibration analyses and the corresponding results obtained using the 1D model. Analysis C4,
(D= 10m, L = 60m)
For Analysis C4 (L/D = 6), the 1D model provides a good fit with the 3D finite element results.
Separate analyses have been conducted in which the distributed moment and the base horizontal
force and moment are removed from the analysis. The results plotted in Figure 144(a) and (b) show
that the two sets of results are similar indicating that, in this case, the distributed moment and base
horizontal force and moment have an insignificant influence on the behaviour of the pile.
Figure 145 shows a comparison of the bending moment profiles obtained from the 1D model when
using soil reactions extracted from the 3D calibration analyses. This demonstrates that little error is
introduced by assuming that longer piles are subject only to laterally distributed loads.
(a) Ultimate response (vG = 0.1D) (b) Small displacement (vG = D/10,000)
Figure 145: Comparison of bending moment profiles obtained using the 1D model and soil
reactions determined from the 3D calibration analyses. Analysis C4, (D= 10m, L = 60m)
Page 159/272
For Analysis C1 (L/D = 2), the 1D model is also seen to provide a good fit with the 3D calibration
data (Figure 146(a)). However, the response computed using the 1D model including only the lateral
soil reactions falls significantly below the 3D calibration data. In this case, in which the pile is
relatively short, it is clear that the distributed moment and the base horizontal force/moment soil
reaction curves need to be included in the 1D model to obtain reliable results. Results of a more
detailed exploration of the contribution of each component in the 1D models for Analysis C1 is
shown in Figure 147. In this case all four of the soil reaction components employed in the model are
seen to contribute significantly to the overall pile behaviour.

calibration analyses and the corresponding results obtained using the 1D model. Analysis C1
(D = 10m, L = 20m)
Figure 147: Influence of the various components of soil reaction on the computed response
for Analysis C1 (D = 10m, L = 20m)
Page 160/272
Figure 148 demonstrates that when the additional soil reaction components are excluded for the
shorter geometry, the under-prediction of load-displacement response is accompanied by a
corresponding under-prediction of the distribution of bending moment along the pile length.
Figure 148: Comparison of bending moment profiles obtained using the 1D model and soil
reactions determined from the 3D calibration analyses. Analysis C1 (D = 10m, L = 20m)
These results suggest that it is possible with the proposed approach to model the full range of pile
geometries from large diameter short monopiles, where all four soil reaction curves appear to be
significant, to smaller diameter jacket piles where the distributed lateral load component is dominant.
D9.3 Influence of soil stiffness and strength parameter variability
Figure 149 shows a comparison of the head load-displacement response for pile C1 (D = 10m, L/D =
2), predicted using the 1D model and adopting the numerically derived soil reactions curves, when
the soil strength and stiffness are scaled. At large displacements, shown in Figure 149(a), variations
in the soil shear stiffness have a negligible effect on the predicted response. However, a 10%
variation in the soil strength profile provides an approximately proportionate variation in response
prediction. At small displacements, shown in Figure 149(b), variations in shear stiffness and strength
appear to have similar influence on the predicted response.
Page 161/272
(a) Comparison up to ultimate displacement (vG (b) Comparison at small displacement

= 0.1D)
Figure 149: Head load-displacement response for analysis C1 (D = 10m, L/D = 2), comparing the
1D model predicted response with scaled soil strength and stiffness
These results demonstrate the dependency of the predicted response on the accuracy of the
measured soil characteristics. Errors in measured soil strength would be likely to result in
proportional errors in the predicted ultimate capacity, whilst the response small displacements is
likely to be highly dependent on accurate measurements of both strength and stiffness
characteristics.
D10 Parameterised curve fitting

The computed normalised soil reaction curves are represented by a parametric equation in a form
that is suitable for implementation in a 1D model of monopile foundations. A four-parameter conic
form is adopted and defined in Section 5.2. This parametric equation is fitted to the normalised soil
reaction curves extracted from the 3D FE analysis using a numerical fitting process.
D10.1 Distributed lateral load soil reaction curve
The distributed load curves (for Pile C4) are represented using the conic form shown in Figure
150(b). Comparisons between the computed soil reaction curves and the conic equations
determined from the fitting process are illustrated in Figure 150(a).
Page 162/272
p
suD
np=0
pu
np
kp
np=1
vpu v
I
D R
(a) Normalised distributed load curve (fitted (b) Normalised distributed load curve shape
parameterised curves shown with dashes)
Figure 150: Distributed normalised components from analysis C4 (D = 10, L/D = 6)
To fit the conic form to the computed data, four normalised parameters are required at each depth;
the initial stiffness kp, the ultimate response 𝑝̅𝑢 , the ultimate displacement 𝑣̅𝑝𝑢 and the curvature np.
These parameters are determined as follows:
1. A single value of ultimate displacement 𝑣̅𝑢 is selected manually at which the reactions for all
depths have reached their approximate ultimate values (𝑣̅𝑝𝑢 = 200 in Figure 150(a)).
2. The initial stiffness kp is selected by fitting the linear expression 𝑝̅ = 𝑘𝑝 𝑣̅ using a least square fit
𝑣̅
over the small displacement region of 0 < 𝑣̅ < 𝑝𝑢⁄20.
3. The ultimate response 𝑝̅𝑢 , and curvature np, are fitted by minimising the proportional least
square error of Equation 3 in section 5.2.2 of the main report using a suitable optimisation
function (e.g. fmincon in MATLAB).
The variation of the soil reaction curve parameters with normalised depth (z/D) is characterised
using simple ‘depth variation’ functions (shown as dashed lines in Figure 151). These depth variation
functions allow the soil reaction curves to be assembled for piles of an arbitrary length (within the
original parameter space). Procedures used to establish the depth variation functions are set out
below.
The depth variation function for initial stiffness 𝑘𝑝 is represented with a linear function, as shown in
Figure 151(a). Note that the initial stiffness parameter 𝑘𝑝 , can also be expressed as the dimensioned
stiffness, divided by the soil shear stiffness, as shown in Equation 24. This means that if the
normalising value of G0 approaches zero towards the soil surface (as shown in Figure 128) then the
initial stiffness tends towards infinity. This will result in large values that may lead to unrealistic pile
stiffness if a different profile of G0 were used to during a design exercise. This potential risk has
been avoided in this dataset by omitting the near surface values from the fit of the linear function,
and fitting only to the region 𝑧⁄𝐷 > 0.2.
Page 163/272
𝑝
𝑝̅ 𝑠𝑢 𝐷 𝑝 1
𝑘𝑝 = = ⁄𝑣𝐼 = Equation 24
𝑣̅ 𝑅 𝑣 𝐺0
𝐷
The depth variation function of the curvature parameter np is represented with a single best fit line
(shown as a dashed line in the plots) to represent the variation of the parameter with z/D.
The variation of 𝑝̅𝑢 with depth, for shallow depths, is related to the wedge-type mechanism illustrated
in Figure 36(b) of section 5 of the main report. Upper-bound calculations described by Murff and
Hamilton (1993) suggest that 𝑝̅𝑢 is likely to vary with depth according to an exponential function. For
large depths, the soil is expected to fail in a flow-around mechanism, implying 𝑝̅𝑢 = 9.14 for a
smooth pile and 𝑝̅𝑢 = 11.92 for a rough pile (Randolph and Houlsby 1984, Martin and Randolph
2006). The depth variation function adopted for 𝑝̅𝑢 is based on an exponential form (consistent with
the calculations given by Murff and Hamilton results):
−𝜉𝑧
𝑝̅𝑢 = 𝑁1 − 𝑁2 𝑒𝑥𝑝 ( )
𝐷 Equation 25
where 𝑁1 , 𝑁2 and 𝜉 are parameters determined by non-linear regression, minimising the proportional
least squares error using an unconstrained optimisation function (fminsearch in MATLAB). In
developing this expression, it is assumed (based on visual observation of the extracted load
displacement curves) that the value of 𝑝̅𝑢 at 𝐿⁄𝐷 = 6 corresponds to the flow-around mechanism.
The regions of poor fit around normalised depths of z/D = 1.4 and z/D = 4.4 seen in Figure 151(c)
occur near the pile rotation point (for values of L/D of 2 and 6 respectively). Due to the relatively
small mobilised soil reactions in these regions, the soil reactions have an almost insignificant effect
on the overall pile response. Their effect on the process of determining the depth variation function
for 𝑝̅𝑢 is mitigated by finding the depth at which the minimum final lateral pile displacement occurs
(corresponding to the rotation point) and excluding any soil reaction data within L/10D of this point.
Page 164/272
(a) Initial stiffness parameter kp (b) Curvature parameter np
̅u
(c) Ultimate response paramter 𝒑
Figure 151: Depth variation functions for the soil reaction curve parameters. Analyses C1 to
C11
Page 165/272
(a) Ultimate resistance (b) Ultimate displacement
Figure 152: Comparison between parameter derived response, API/DNV prediction and Murff
and Hamilton (1993) smooth pile prediction
Figure 152(a) shows a comparison between the depth variation function for 𝑝̅𝑢 (Equation 25 with the
parameters in Table 22), and corresponding data from Murff and Hamilton (1993) for the translation
of smooth and rough piles in a rigid plastic soil with a constant strength with depth. Note that the
Murff and Hamilton (1993) data is for weightless soil whereas the “Parametric” and “API/DNV”
curves account for soil weight. Also shown is the equivalent variation of 𝑝̅𝑢 with depth given by the
API/DNV approach (Appendix F4 Equation 75). Our results are consistent with the Murff and
Hamilton (1993) upper-bound calculations, although the underlying methodology adopted by Murff
and Hamilton is very different from that used here. Values of 𝑝̅𝑢 computed using the API/DNV
approach fall well below the other curves.
Further comparisons can be made between the API p-y formulation and the current parameterised
soil reaction curve. The API formulation (Appendix F4 Equation 76) indicates that the ultimate
resistance 𝑝𝑢 is not mobilised until a displacement of approximately 𝑣𝑢 = 0.23𝐷 (for all depths). This
displacement is generally significantly higher than the ultimate displacement observed in the
parametric fit to the 3D finite element results, as shown in Figure 152(b). Figure 153 compares
predictions of how the lateral distributed load curves, normalised by the ultimate resistance, evolve
with normalised displacement and shows that the API/DNV curves not only under-predict the
displacement of ultimate resistance, but also the intermediate curve shape. When the under-
prediction of the ultimate resistance is also considered, it is straightforward to understand why the
API/DNV approach provides a conservative estimate of the response in the stiff over-consolidated
clay material.
Page 166/272
Figure 153: Comparison between PISA parametric (solid lines) and API/DNV (dashed lines)
predicted lateral distributed load, normalised by the ultimate resistance, at various depths of a
10m diameter pile
D10.2 Distributed moment curves
m
suD2
mu
km
ψmu ψIR
(a) 3D finite element extracted and (b) Parameter curve shape

parameterised fit for C4
Figure 154: Distributed moment curve shape for normalised curves

Figure 154(a) shows an example set of normalised distributed moment reaction curves for one of the
piles analysed in the calibration process (C4). In contrast to the lateral distributed load results, these
data indicate an initial peak in resistance reducing to a significantly lower value as the rotation is
further increased. A bi-linear form of the parametric curve shown in Figure 150(b), has been adopted
by selecting nm = 0 and ensuring that 𝜓̅𝑚𝑢 > 𝑚 ̅ 𝑢 /𝑘𝑚 , resulting in the curve shape shown in Figure
154(b). By selecting the strength and stiffness parameters for the bi-linear curve, the peaks in the
Page 167/272
computed response are ignored. In adopting this approach, it is assumed that the peak response
occurs over a sufficiently small range of rotation that it will not significantly affect the overall pile
behaviour. The ultimate moment parameter 𝑚 ̅ 𝑢 is determined at each depth by taking the mean of
the values that satisfy 𝑚 ̅ > 0.9𝑚̅ final . The initial stiffness parameter 𝑘𝑚 is then determined using a
least squares fit to the values in the region 𝑚 ̅ <𝑚 ̅ 𝑢 /10.
In a similar fashion to the lateral distributed loads, the initial stiffness parameter 𝑘𝑚 is inversely
proportional to the soil shear stiffness G0, which can lead to unrealistically large values when the
shear stiffness tends to zero, as shown in Figure 155(a). The fitting of the linear depth variation
function for 𝑘𝑚 is therefore carried out over the region z/D > 0.2. A linear best-fit line (determined by
least squares regression) is also adopted for the depth variation of 𝑚 ̅ 𝑢 . The resulting best-fit lines
are specified in Table 22 and plotted in Figure 155.
Unlike the distributed lateral loads, distributed moments close to the rotation point of the pile are
significant to the overall pile response and are therefore included in the fit of the soil reaction
function for the ultimate distributed moment.
(a) Initial stiffness parameter km ̅u

(b) Ultimate response paramter 𝒎
Figure 155: Variation of normalised distributed moment parameters against normalised depth
for the field test 3D finite element simulations
It is noted that the piles included in the calibration analyses appear to be insufficiently long to
mobilise the full-flow mechanism (e.g. Figure 36(b) of Section 5 of the main report). The distributed
moments extracted from the calibration analyses are therefore regarded as being associated
principally with the vertical movements developed within surface wedge-type mechanisms.
Page 168/272
D10.3 Base horizontal force curves
Figure 156: Base horizontal force extracted, normalised and parameterised for analysis C4
(D = 10, L/D = 6)
The base horizontal force soil reactions extracted from the calibration analyses are shown in Figure
156. The conic form is used to represent these data. The ultimate displacement parameter 𝑣̅ Bu is
assumed to be independent of depth and a single parameter value is selected in a similar fashion to
̅ Bu (determined at values of L/D
the distributed load curves. The remaining three parameters 𝑘𝐻 , 𝑛𝐻 , 𝐻
= 2 and L/D = 6) are fitted to a linear depth variation function as shown in Figure 157. The resulting
regression lines are specified in Table 22.
(a) kH (b) nH ̅ Bu
(c) 𝑯
Initial stiffness parameter Curvature parameter Ultimate response paramter
Figure 157: Variation of normalised base horizontal force parameters against normalised depth
for the field test 3D finite element simulations
Page 169/272
D10.4 Base moment curves
Figure 158: Base moment extracted, normalised and parameterised for analysis C4
(D = 10, L/D = 6)
Conic-form equations are fit to the computed base moment reaction curves. The determined linear
depth variation functions are shown in Figure 159.
(a) kM (b) nM ̅ Bu
(c) 𝑴
Figure 159: Variation of normalised base moment parameters against normalised depth for the
field test 3D finite element simulations
Page 170/272
D10.5 Soil reaction curves and depth variation functions
Parameters for the example clay till site soil reaction curves and the depth variation functions are
listed in Table 22.
Table 22: Parameters for normalised soil reaction curves for a pile with an air-filled gap in an
idealised clay till as defined in section 5.3.2, calibrated within the parameter space defined in
Table 21 (𝟐 < 𝑳⁄𝑫 < 𝟔, 𝟓 < 𝐃 < 𝟏𝟎, 𝟓 < 𝒉⁄𝑫 < 𝟏𝟓, 𝟔𝟎 < 𝑫⁄𝒕 < 𝟏𝟏𝟎)

Distributed lateral load, 𝒑 Ultimate strain, 𝑣̅ pu 200
𝑧
𝐷
𝑧
Curvature, np −0.07 + 0.92
𝐷
𝑧
Ultimate reaction, 𝑝̅u 11.66 − 8.64𝑒 (−0.37𝐷)
𝑧
Initial stiffness, km −0.12 + 0.98
𝐷
Curvature, nm 0
𝑧
̅u −0.05 + 0.38
𝐷
Base horizontal force, 𝑯𝑩 Ultimate strain, 𝑣̅ Hu 300
𝐿
𝐷
𝐿
𝐷
𝐿
̅ Bu
Ultimate reaction, 𝐻 0.07 + 0.59
𝐷
𝐿
Initial stiffness, kM −0.002 + 0.19
𝐷
𝐿
Curvature, nM −0.15 + 0.99
𝐷
𝐿
̅ Bu
𝐷
Page 171/272
D11 Design calculations conducted using the 1D model
D11.1 Accuracy of the parametric expressions when applied to the 1D model
A study has been conducted to assess the performance of the 1D model based on the soil reaction
curves given in Table 22. In the results presented below, this form of the model is referred to as ‘1D
(parametric)’. Prediction made using the 1D (parametric) approach are compared with results
obtained using the current API p-y method (described in Appendix F4).
Comparisons between the predictions obtained using the 1D (parametric) model, the results of 3D
finite element model and the existing API/DNV approach for the short C1 pile and the long C4 pile
are shown in Figure 160. Whilst observable error occurs in the predictions of the short C1 pile
response made using the 1D model, the overall behaviour is well represented with accurate
predictions of the initial stiffness, ultimate response and intermediate curvature. This pile provides a
stern test of the parameterisation process, because the ultimate soil reaction is reached over a
significant length of the pile and therefore a good approximation of the reaction curvature is required
to obtain an accurate prediction of the shape of the overall load/displacement response.
Page 172/272

As shown in Figure 161 the shape of the load-displacement response up to both large and small
displacement is well predicted when employing the 1D model with parameterised soil reaction
curves. The average accuracy of prediction at large displacements of η0.1D = 94%, is significantly
higher than that achieved by the API/DNV approach of 41%. The latter is especially poor in the
simulation of shorter piles. However, the greatest improvement in performance is seen in the
accuracy of prediction of the small displacement response ηsd = 87%. The reduced accuracy of the
small displacement prediction for the shorter piles (L/D = 2) occurs as a result of omitting the peak
during the fitting of the depth variation functions of initial stiffness (kp and km) close to the soil
surface, but will result in a more robust calculation of small displacement behaviour when applied to
varied soil profiles.
Page 173/272
D
L/D
M/HD
D/t
D
L/D
M/HD
D/t
D11.2 Interpolated design scenario
Page 174/272
To test the predictive capability of the 1D model two additional test analyses (D1 and D2) have been
conducted using the 3D finite element model. The geometric configurations adopted for these tests
cases are provided in Table 23 and were selected to fall within the parameter space adopted for the
initial calibration analyses, as shown in Figure 162.
Analysis D h L t
h/D L/D D/t
D1 7.5 37.5 5 22.5 3 68 110
D2 8.75 87.5 10 35 4 91 97
Figure 162: Parametric geometry space for the field test, full scale calibration analyses and
design scenarios
parametric 1D model are shown in Figure 163 and Figure 164. The results indicate a close fit
between the 1D and 3D finite element data over the full range of applied loading (Figure 163(a) and
Figure 164(a)). The small displacement response computed using the 1D model is seen to be
almost identical to the response computed using the 3D finite element model (Figure 163(b) and
Figure 164(b)). Also shown for comparison are the corresponding results obtained using the API p-y
method.
Page 175/272
Figure 163: Comparison of load-displacement response from 3D FE, 1D model simulations

using parameterised soil reactions and the existing API/DNV approach for the D1 design case
(D = 7.5m, L/D = 3)
Figure 164: Comparison of load-displacement response from 3D FE, 1D model simulations

using parameterised soil reactions and the existing API/DNV approach for the D2 design case
(D = 8.75m, L/D = 4)
Page 176/272
D12 Explorative analysis using calibrated soil reactions in the 1D model

This section outlines additional studies conducted using the 1D model with the calibrated soil
reaction curves given in Table 22.
In the current study the pile geometries are:
 pile diameters 5m < D < 10m,

 normalised lengths 2 < L/D < 6
 load eccentricities 5 < M/HD < 15.
D12.1 Contribution of reaction components
The relative importance of the four soil reaction components (distributed lateral load p, distributed
moment m, base horizontal force HB and base moment MB) is investigated as described below.
D12.1.1 Varied pile diameter
Figure 165 shows the percentage contribution of each soil reaction component to the resistance of
the lateral load, as the pile diameter is varied. The contribution has been calculated by simulating
the response of piles with a fixed L/D = 4, D/t = 110 and M/HD = 5. Components are sequentially
added in to the simulation, in the order p-v, m-ψ, HB-v and finally MB-ψ. The difference in the pile
capacity between successive simulations is attributed to the additional reaction component. Using
the accuracy metric approach (described in section 4.4 of the main report), the additional
contribution of that component up to the target displacement is quantified. Note that the lateral
distributed load p-v, dominates the pile response and has therefore been plotted on the primary (left
hand side) y-axis. The smaller contributions of the remaining components are plotted against the
secondary (right hand side) y-axis.
(a) Ultimate response (up to vG = 0.1D) (b) Small displacement (up to vG = D/10,000)
Figure 165: Soil reaction component contribution of pile lateral capacity with varied diameter
for L/D = 4, D/t = 110, M/HD = 5. The contribution of the lateral load is indicated on the left axis.
The contributions of the other components (dashed lines) are indicated in the right axis.
Page 177/272
Figure 165(a) shows that at the ultimate response the contribution of the different components does
not vary significantly, with the base horizontal force and distributed moment components contributing
a similar proportion of the pile capacity. The base moment component is demonstrated to contribute
the least to the pile capacity.
Figure 165(b) shows that at a relatively common pile aspect ratio L/D = 4, the distributed moment
contributes a significant proportion of the initial pile stiffness. Unsurprisingly, the base horizontal
force and base moment do not contribute significantly and have probably not been mobilised at such
small displacements.
D12.1.2 Varied pile slenderness L/D
Figure 166 shows the contribution of the different soil reaction components when a pile of a fixed D
= 7m, D/t = 110 and M/HD = 5 is varied in slenderness 2 < L/D < 6.
Figure 166: Soil reaction component contribution of pile lateral capacity with varied L/D for D =
7m, D/t = 110, M/HD = 5. The contribution of the lateral load is indicated on the left axis. The
contributions of the other components (dashed lines) are indicated in the right axis.
Figure 166(a) shows that at very low L/D, all three additional components of soil reaction have a
significant influence on the pile capacity at ultimate response. However, as the slenderness of the
pile is increased, the relative influence of these components decreases, to the point that the
response is dominated by the lateral distributed load.
However, Figure 166(b) demonstrates that the distributed moment has a significant contribution to
the initial stiffness at high slenderness ratios and increases as L/D is reduced. For very short piles
the additional soil reaction components contribute approximately 38% of the total initial stiffness,
with the most significant being the distributed moment.
Page 178/272
D13 Influence of shear strains on the 1D model calculations

As shown in Figure 167 the influence of the shear compliance within the Timoshenko beam element
of the 1D model has a relatively small effect on the obtained response prediction. On average over
the calibration geometries, omitting the shear compliance from the model results in an increase of
the predicted load at small displacement (vG = D/10,000) and ultimate response (vG = 0.1D) by
0.79% and 1.97% respectively.
Figure 167: Comparison of load-displacement 1D (parametric) prediction for calibration C4 (D =

10m, L/D = 6) with and without shear strain compliance
D14 Consideration of the gap on the active side of the pile

The analysis described so far in this Appendix has assumed that the gap developing on the active
side is air-filled. However, in an offshore environment, any gap on the active side will fill with water
and, as a consequence, additional lateral loading will be applied to the monopile. Figure 168 shows
comparisons of predicted load-displacement response when using the 1D model. The green line
shows the response predicted when adopting soil reactions extracted from a 3D FE analysis with an
air-filled gap. The blue line shows the response predicted when adopting soil reactions extracted
from a 3D FE analysis with a water-filled gap. While the difference in the response at small
displacements is negligible (due to little to no gap development) the ultimate capacity at vG=0.1D is
reduced due to the additional hydrostatic loading by approximately 10%.
The red line in Figure 168 shows the response predicted when the results of the air-filled 3D FE
analysis are corrected by applying a hydrostatic pressure to the pile at any location that a gap has
developed. The good correlation between the water-filled simulation and the air-corrected approach
validates this process for predicting the performance with a water-filled gap, using air-filled gap 3D
FE simulations.
By applying this approach to the entire calibration study set of simulations a set of soil reaction curve
parameters has been derived, as shown in Table 24, which are appropriate for the design of an
offshore foundation with a water-filled gap. Note that there is little to no difference in the majority of
the parameter expressions for the air-filled and submerged cases, except for the ultimate lateral
Page 179/272
distributed load parameter, 𝑝̅𝑢 , which remains largely unchanged at the soil surface, but is reduced
at depth.
(a) C1 (D = 10m, L/D = 2) (b) C4 (D = 10m, L/D = 2)
Figure 168: Comparison of load-displacement responses predicted using the 1D (numerical)

model that feature an air-filled gap, a water-filled gap and also an air-filled gap corrected to
account for the hydrostatic pressure of a water-filled gap.
Figure 169 shows a more detailed analysis of the size of the gap developed in the two simulations.
Not unsurprisingly the water filled gap is marginally larger than the air-sized gap, although the
overall effect on the results appears to be marginal (as shown in Figure 168).
(a) C1 (D = 10m, L/D = 2) (b) C4 (D = 10m, L/D = 2)
Figure 169: Comparison of area of pile exposed by gap opening during loading
Page 180/272
section 5.3.2, calibrated within the parameter space defined in Table 21 and with an air gap
corrected to be water-filled (𝟐 < 𝑳⁄𝑫 < 𝟔, 𝟓 < 𝐃 < 𝟏𝟎, 𝟓 < 𝒉⁄𝑫 < 𝟏𝟓, 𝟔𝟎 < 𝑫⁄𝒕 < 𝟏𝟏𝟎)

Distributed lateral load, 𝒑 Ultimate strain, 𝑣̅ pu 200
𝑧
𝐷
𝑧
Curvature, np −0.06 + 0.93
𝐷
𝑧
Ultimate reaction, 𝑝̅u 9.78 − 6.73𝑒 (−0.36𝐷)
𝑧
Initial stiffness, km −0.12 + 0.98
𝐷
Curvature, nm 0
𝑧
̅u −0.05 + 0.38
𝐷
Base horizontal force, 𝑯𝑩 Ultimate strain, 𝑣̅ Hu 300
𝐿
𝐷
𝐿
𝐷
𝐿
̅ Bu
Ultimate reaction, 𝐻 0.07 + 0.59
𝐷
𝐿
Initial stiffness, kM −0.002 + 0.19
𝐷
𝐿
Curvature, nM −0.15 + 0.99
𝐷
𝐿
̅ Bu
𝐷
D14.1 Equivalent API/DNV design
A simplified design exercise has been carried out in which the pile lengths used for the design
cases, D1 and D2, have been varied, so as to achieve similar pile capacity at ultimate response (vG
= 0.1D) with the API/DNV approach. This exercise has been carried out by comparing to the
simplified 1D model using the submerged parametric soil reaction curves (which are appropriate for
a submerged design), as described in section 0. As shown in Figure 170, to achieve similar
response using the API/DNV approach for the D1 design case (D = 7.5m, L/D = 3, h/D = 5), it is
necessary to extend the pile length by 55%, to an L/D = 4.65. Assuming that the ultimate response
was the design driver in this case, the use of the PISA parametric curves would therefore allow a
significant saving on both material and installation costs, when compared to an equivalent design
using the previous API/DNV methods. Whilst other design drivers, such as the accumulated fatigue
Page 181/272
from dynamic loading, are also likely to be potential design drivers, it should be noted that even
when adopting a significantly longer pile design, the API/DNV approach is incapable of matching the
predicted stiffness of the PISA approach.
Very similar results were observed for D2, with a pile length extension of 55% (L/D = 6.2) required to
allow the ultimate capacity predicted by the API/DNV approach to match that achieved by the PISA
parameterised approach (L/D = 4).
Figure 170: Comparison of load-displacement response from 3D FE and the existing API/DNV
approach for the D1 design case (D = 7.5m, L/D = 3)
D14.2 Contribution of reaction components
In a similar fashion to the exercise shown in section D12.1, the contribution of the different soil
reaction components has been calculated, based on the water-filled gap soil reaction parametric
curves shown in Table 24.
Figure 171 and Figure 172 show the variation of water-filled gap soil reaction component
contributions with a varied pile diameter and slenderness ratio, respectively. A marginal reduction in
the influence of the distributed lateral load p, is observed when compared to Figure 165 and Figure
166, which adopt soil reaction curves with an air-filled gap, but do not substantively change any
assessment of the influence or trends.
Page 182/272
D14.2.1 Varied pile diameter D
for L/D = 4, D/t = 110, M/HD = 5. The contribution of the lateral load is indicated on the left axis.
The contributions of the other components (dashed lines) are indicated in the right axis.
D14.2.2 Varied pile slenderness L/D
7m, D/t = 110, M/HD = 5. The contribution of the lateral load is indicated on the left axis. The
contributions of the other components (dashed lines) are indicated in the right axis.
Page 183/272
D15 Results of water-filled gap full 3D FE simulations

Subsequent to the analysis described in section D14, where a correction to the air-filled gap 3D FE
simulations was used to predict pile response with a water-filled gap, 3D FE simulations have been
conducted of all calibration pile geometries with a water-filled gap and hydrostatic pressure on any
opening face. Using the procedure outlined in sections D8 and D10, expressions for the parameters
of soil reaction have been derived and are shown in Table 25 (column ‘Water-filled gap
expressions’) and compared to the previously derived expressions using the corrected air-filled gap
simulations (column ‘Air-corrected expressions’).
section 5.3.2, calibrated within the parameter space defined in Table 21 and with a water-filled
gap calculated by a correction on the air-filled gap and with full 3D FE simulation of the
water-filled gap (𝟐 < 𝑳⁄𝑫 < 𝟔, 𝟓 < 𝐃 < 𝟏𝟎, 𝟓 < 𝒉⁄𝑫 < 𝟏𝟓, 𝟔𝟎 < 𝑫⁄𝒕 < 𝟏𝟏𝟎)
Soil reaction Parameter Air-corrected Water-filled gap

component expression expressions
(Table24)
Distributed Ultimate strain, 𝑣̅ pu 200 200
lateral load, 𝒑 𝑧 𝑧
Initial stiffness, kp −1.11 + 8.13 −1.10 + 8.12
𝐷 𝐷
𝑧 𝑧
Curvature, np −0.06 + 0.93 −0.05 + 0.92
𝐷 𝐷
𝑧 𝑧
Ultimate reaction, 𝑝̅u 9.78 − 6.73𝑒 (−0.36𝐷) 10.21 − 7.22𝑒 (−0.33𝐷)
Distributed Ultimate rotation, 𝜓̅mu 10 10
moment, 𝒎 𝑧 𝑧
Initial stiffness, km −0.12 + 0.98 −0.11 + 0.97
𝐷 𝐷
Curvature, nm 0 0
𝑧 𝑧
̅u −0.05 + 0.38 −0.04 + 0.38
𝐷 𝐷
Base Ultimate strain, 𝑣̅ Hu 300 300
horizontal
𝐿 𝐿
force, 𝑯𝑩 Initial stiffness, kH −0.32 + 2.58 −0.32 + 2.56
𝐷 𝐷
𝐿 𝐿
Curvature, nH −0.04 + 0.76 −0.03 + 0.74
𝐷 𝐷
𝐿 𝐿
̅ Bu
Ultimate reaction, 𝐻 0.07 + 0.59 0.07 + 0.60
𝐷 𝐷
Base Ultimate rotation, 𝜓̅Mu 200 200
moment, 𝑴𝑩
𝐿 𝐿
Initial stiffness, kM −0.002 + 0.19 −0.003 + 0.20
𝐷 𝐷
𝐿 𝐿
Curvature, nM −0.15 + 0.99 −0.16 + 1.01
𝐷 𝐷
𝐿 𝐿
̅ Bu
Ultimate reaction, 𝑀 −0.07 + 0.65 −0.08 + 0.65
𝐷 𝐷
Page 184/272
As shown in Table 25, soil reaction expressions calculated for a water-filled gap, using either the air-
corrected approach or a full 3D FE simulation of the water-filled gap show good correlation. Figure
173, shows the difference in the lateral distributed ultimate soil reaction, which demonstrates that the
difference between two expressions is less than 3% over the entire length of all pile geometries of
the calibration study.
Figure 173: Comparison of distributed lateral ultimate soil reactions with a water-filled gap,
using an air-corrected approach and full 3D FE water-filled gap simulations
As shown in Figure 174, the accuracy of predictions using the 1D model and parameters determined
from the 3D FE simulations with water in the gap is similar to the accuracy of the dry gap calibration
described in D11, with average accuracy metrics up to ultimate response and at small
displacements of 95% and 87% respectively.
Page 185/272
D
L/D
M/HD
D/t
D
L/D
M/HD
D/t
Figure 174: Accuracy metrics comparing load-displacement shape accuracy of the 1D

model using water-filled gap parameterised soil reaction curves to 3D finite element
simulations
Page 186/272
PISA Project – Final Report Appendix E - Detailed design method application to a dense sand site
Appendix E Detailed design method application to a dense sand

site
E1 Introduction
This appendix sets out an example application of the numerical-based method, for an idealised soil
profile that resembles the conditions at some North Sea sand sites. The approach is described in
some detail to provide a template for future applications. The resulting soil reaction curves can be
used to conduct rule-based analyses at dense sand sites with comparable relative density, stiffness
and friction angle profiles.
Initially, the ground conditions at the example site are established. These conditions are based on
the profiles of initial density, stiffness, pore water pressure and 𝐾𝑜 that were measured at the
Dunkirk dense sand test site. Minor modifications are applied to these profiles (principally on the
distribution of the pore pressures) to provide a closer representation of the likely conditions at typical
North Sea sites.
Next, a set of 3D finite element models of an embedded monopile (referred to as ‘calibration

analyses’) are developed to cover a range of expected values of pile dimensions and load
eccentricity. These analyses are conducted using the finite element software ICFEP (Potts and
Zdravkovic, 1999, 2001). The calibration analysis results are interrogated to extract the various soil
reaction components that act on the pile.
Finally, the computed soil reaction curves are represented by a set of equations, based on the conic
functions specified in Section 5.2.2 of the main report. It is demonstrated that these equations can
be used, in conjunction with a 1D model of an embedded monopile, to compute the performance of
a monopile for geometric and loading configurations that fall within the parameters space explored in
the calibration process.
The procedures described in this appendix follow closely those adopted for the clay design study
that is described in Appendix D.
E2 Assumed ground conditions

The initial ground conditions at the example site are based on those estimated at the Dunkirk test
site and specified in Appendix A. The conditions are idealised, however, for closer representation of
the likely conditions at a typical North Sea sand site. The pore pressure and stiffness profiles
adopted in this design example are shown in Figure 175, labelled as ‘Idealised dense sand’. For
comparison, the equivalent profiles adopted for the analyses of the field test piles at the Dunkirk site
are also shown in the figure, labelled ‘Dunkirk site’. In the idealised dense sand profile, the pore
pressure is specified to be hydrostatic, with zero pore pressure at the ground surface (Figure
175(a)). This is in contrast to the Dunkirk site profile in which the water table is located beneath the
soil surface. In the idealised dense sand profile, the initial density of the sand is uniform throughout
the deposit, at Dr = 75%. This is the same as the initial relative density of the natural Dunkirk sand
and corresponds to an initial void ratio e0 of 0.628. The initial vertical total stress is calculated by
3
adopting the bulk unit weight of the sand as 𝛾𝑠𝑎𝑡 = 19.9 kN/m (i.e. the same value adopted for the
Dunkirk sand below the water table). The coefficient of earth pressure at rest is 𝐾𝑜 = 0.4, which is
the same as the value adopted for the Dunkirk site ground model.
Page 187/272
The small strain shear modulus adopted in the idealised dense sand profile is determined from the
local values of mean effective stress 𝑝′ and the initial void ratio 𝑒0 using Equation 16 of Appendix A.
As a consequence of higher pore water pressures in the design example (compared with the
Dunkirk site ground model) the effective stresses are reduced, resulting in reduced values of G0, as
shown in Figure 175(b).
(a) (b)
Figure 175: Soil profiles for the sand design example
E3 Constitutive models
A Critical State model based on the state parameter framework for sands (Been & Jefferies, 1985) is
adopted to simulate the behaviour of the sand in the 3D finite element calibration analyses. The
model is described in detail in Taborda et al. (2014), and is an evolution of the bounding surface
plasticity model proposed initially by Manzari & Dafalias (1997) and subsequently improved by
Papadimitriou & Bouckovalas (2002).
The model is calibrated against the in situ and laboratory data from the Dunkirk site, as explained in
Appendix A. The model parameters are summarised in Table 16 of Appendix A.
The interface between the pile and the soil is represented with an elasto-plastic Mohr-Coulomb
model. The elastic part of the interface model is defined by a shear and a normal stiffness which, in
5 3
this case, are both set to 1.0x10 kN/m . These values ensure a rough pile-soil interface (shear
stiffness) and no deformation in compression across the element (normal stiffness). The interface is
o
specified to have zero cohesion and an angle of friction of 32 ; these values are consistent with the
strength parameters adopted for the sand. The monopile is modelled as an elastic material, with
properties representative of steel: Young’s modulus E = 200 GPa and Poisson’s ratio ν = 0.3. The
pile wall thickness is specified as an additional model parameter.
Page 188/272
E4 Specification of the 3D calibration analyses

The pile geometries selected for the 3D calibration analyses are the same as those adopted for the
clay design example in Appendix D and are listed in Table 26.
Table 26: Geometry and loading adopted in the 3D finite element calibration analyses
Analysis D h L t
h/D h/L L/D D/t
C1 10 50 5 20 2.5 2 91 110
C2 10 150 15 20 7.5 2 91 110
C3 10 50 5 20 2.5 2 125 80
C4 10 50 5 60 0.83 6 91 110
C5 10 150 15 60 2.5 6 91 110
C6 5 25 5 10 2.5 2 45 110
C7 5 25 5 10 2.5 2 83 60
C8 5 25 5 30 0.83 6 45 110
C9 5 75 15 30 2.5 6 45 110
C10 7.5 37.5 5 15 2.5 2 68 110
C11 7.5 37.5 5 45 0.83 6 68 110
E5 Finite element mesh and boundary conditions

Lateral loading of a circular pile exhibits one plane of symmetry and it is therefore sufficient to
discretise only half of the problem. The FE mesh used for the C4 pile is shown in Figure 176 as an
example. The soil domain is discretised with 10530 20-noded hexahedral displacement-based
isoparametric solid elements. The interface between the pile and the surrounding soil is simulated
with 360 16-noded zero-thickness interface elements (Day & Potts, 1994), whereas the pile itself is
discretised with 600 8-noded shell elements (Schroeder et al., 2007). In the Z-coordinate direction
each pile is discretised with 30 rows of elements below the mudline and 20 rows of elements above.
Similar meshes are employed for the other piles in the calibration analyses.
Page 189/272
Figure 176: Finite element mesh for the monopiles in Table 26 (dimensions in brackets for
pile C4, D = 10, L/D = 6). Note that the coordinate system used to define the finite element
mesh (upper case Z positive in the upwards direction) differs from the coordinate system
used to define the 1D model (in which lower case z is positive downwards).
To prevent rigid body movements of the mesh, all three displacement components in the three
Z = -100 m). In addition, the displacements normal to the vertical cylindrical boundary are also
prescribed to be zero. To ensure that the X-Z plane at Y = 0 is a plane of symmetry, the
displacements in the Y-direction over this plane are set to zero, as are the rotational degrees of
freedom with respect to X- and Z-axes along the edges of pile shell elements in the plane of
symmetry. The horizontal load at the pile top (i.e. 𝑍 = ℎ) is applied in a displacement controlled
manner, such that the increments of displacement in the X-direction are applied uniformly around
the pile perimeter. The horizontal load, H, on the pile is then obtained as the reaction to the applied
displacements. An appropriate number of displacement increments are applied to provide a
complete load-displacement curve for each pile, thus enabling the interpretation of the shape of the
extracted soil reaction curves. All analyses are performed under drained conditions.
E6 3D finite element results

Figure 177(a) shows the relationship between the lateral load and the mudline deflection for the 10
m diameter piles specified in Table 26. Relatively large pile displacements have been applied in the
calibration examples to enable the various soil reaction curves to be fully defined. In addition, Figure
177 (b) and (c) show the deflected shapes of the D=10 m piles, with the depth z normalised by pile
length L, for cases when the horizontal mudline displacement reaches, respectively, 0.0001D (initial
loading stage) and 0.1D (nominal failure). Principal observations from Figure 177 include: (a) the
load capacity increases with increasing pile length and reduces with increasing stick-up height, ℎ; (b)
short piles deform in an essentially rigid manner whereas the longer piles deform more flexibly; (c)
variations in the pile wall thickness has a negligible influence on pile response.
Page 190/272
(a)
(b) (c)
Figure 177: D = 10 m monopiles: (a) load-displacement curves; pile deflections for mudline
horizontal displacement at (b) 0.0001D and (c) 0.1D
Page 191/272
E7 Comparison between the 3D finite element results and the API/DNV ‘p-y’
method
An exercise has been conducted to compare the computed pile response with results obtained using
the conventional p-y approach based on the API/DNV p-y curves for sand. The calculation of the pile
response using the API/DNV approach has been conducted as an independent analysis by DONG
Energy, as described in Appendix F4.
The comparisons between the API/DNV approach and the calibration analyses, presented below,
are principally concerned with the following configurations:
 C1 – large diameter (D = 10), short (L/D = 2)

 C4 – large diameter (D = 10), long (L/D = 6)
 C6 – small diameter (D = 5), short (L/D = 2)
As shown in Figure 178 and Figure 179, the API/DNV approach appears to provide a reasonably
good match with the 3D finite element results. At large displacements the response of the shorter
piles computed using the API/DNV approach is rather softer than the equivalent finite element data,
especially for the smaller diameter (D = 5m) piles. However, the API/DNV approach provides a
closer comparison with the 3D results than is the case for the clay design study described in
Appendix D.
Figure 178: Comparison between the 3D finite element results and the API/DNV ’p-y’ method
Page 192/272
Figure 179: Comparison between the 3D finite element results and the API/DNV method at small
displacements
To quantify the difference in area below the computed load-displacement responses the ‘quality of
fit’ metric described in section 4.4 is adopted. Separate comparisons have been undertaken for the
ultimate response (𝑣𝐺 = 0.1𝐷) and the small displacement response (𝑣𝐺 = 𝐷/10,000).
Figure 180(a) demonstrates that the API/DNV approach provides an overall shape of response that
matches well with the calibration analyses, with an average accuracy (with respect to the finite
element data) of 88%. A marginally lower accuracy is observed for the smaller diameter simulations
(D = 5m), but, in general, the prediction accuracy for long and short piles is similar. As shown in
Figure 180 (b), the small displacement prediction accuracy achieved by the API/DNV approach is
rather lower (average ηsd = 43%), with an observable difference in prediction accuracy between the
long piles (average ηsd = 62%) and the short (average ηsd = 28%) piles.
Page 193/272
Pile
D
L/D
M/HD
D/t
(a) Quality of fit metric for ultimate response (vG = 0.1D). Average η0.1D is 88%.
Pile
D
L/D
M/HD
D/t
(b) Quality of fit metric for small displacement response (vG = D/10,000). Average ηsd is 43%.
Figure 180: Metrics for the accuracy of prediction of the calibration analyses using the
API/DNV ‘p-y’ approach
E8 Extraction of soil reaction curves

A soil reaction curve for a given depth may be determined from forces and displacements computed
using the 3D finite element model at element nodes or Gauss points. However, the forces must be
summed over each element before being used to calculate the corresponding depth-wise soil
reaction. In this study, the soil reaction curves are determined by systematically extracting stresses
from the interface elements around the perimeter of the pile and also on the basis of the computed
nodal forces acting on the shell elements and the soil elements at the base of the pile. Local
displacements and rotations of the pile are determined from the corresponding Gauss point and
node displacements. The components of the 3D model that are relevant to this soil reaction curve
extraction process are highlighted in red in Figure 181(b).
Page 194/272
H PILE
SHELLS
INTERFACE
ELEMENTS
TOE SOIL
ELEMENTS
(b) Section view of mesh with

(a) Typical FE mesh
element types
Figure 181: Finite element mesh and element types
E8.1 Boundary equilibrium check on model convergence
A check that the computed nodal forces applied to the pile are in equilibrium is achieved by
summing the nodal forces on a boundary around the pile (as shown by the red highlighted nodes in
Figure 181(b)), and confirming that the sum is equal to the applied load, (to within an acceptable
tolerance). The maximum proportional difference between the sum of the boundary forces and the
applied load for the finite element calibration simulations up to a ground level displacement of vG =
0.1D are shown in Figure 182. These errors are small (compared to those that develop in the clay
analyses), largely due to the lack of any gap formation and the resulting constraints that this
imposes on the FE solution.
Pile
D
L/D
M/HD
D/t
Figure 182: Maximum proportional difference between sum of the boundary forces and the
applied load
Page 195/272
E8.2 Component extraction
Four separate soil reaction components are assumed to act on the pile (e.g. see Figure 35 of
Section 5 of the main report).
The distributed load p (units of force / length) is determined by integrating the y-component of the
computed normal stresses in the interface elements at a given depth. The corresponding
displacement v is calculated from the average of the Gauss point displacements around the pile
perimeter at the corresponding depth.
The distributed moment m (units of force x length / length) is calculated by integrating the product of
the vertical components of computed shear stress and the lateral distance of the Gauss point from
the neutral axis of the pile, using Equation 26. The corresponding rotation of the pile cross section 
is calculated by computing a least squares fit of an inclined plane about the x-axis to the computed
interface element displacements at the appropriate depth.
𝐷2 2𝜋
d𝑀 = ( ∫ 𝜏 𝑐𝑜𝑠𝜙 𝑑𝜑) 𝑑𝑧 = 𝑚 𝑑𝑧
4 0 Equation 26
where 𝜏 is the local vertically-acting shear traction and 𝜙 is the azimuthal angle, defined with respect
to the loading direction (see Figure 36(a) of Section 5 of the main report).
The base shear force HB (units of force) is calculated from the sum of the y-component of nodal
force for the remaining nodes across the base of the pile boundary (shown in Figure 181(b)). This
includes contributions from the nodes on the soil elements across the toe of the pile, the nodes at
the base of the pile shell elements and any remaining interface nodes on the boundary that were not
used for the distributed load calculations. The corresponding displacement v is calculated by taking
the average of the displacements of the pile nodes at the base of the pile.
The base moment MB (units of force x length) is calculated with reference to the same nodes as
those for the base shear force, but summing the product of the vertical component of force and the
lateral distance from the pile neutral axis. The corresponding rotation of the pile base  is
calculated by minimising the root mean square error for an inclined plane about the x-axis that is fit
to the pile base nodal displacements.
Examples of the distributed soil reactions extracted at selected depths of the C4 simulation (D =
10m, L/D = 6) are shown in Figure 183(a)-(b). The base shear and base moment components are
shown in Figure 183(c)-(d).
Page 196/272
(a) distributed load-displacement (p-v) (b) distributed moment-rotation (m-)
(c) base shear-displacement (HB-vB) (d) base moment-rotation (MB-)
Figure 183: Example soil reaction curves extracted from C4 (D = 10, L/D = 6) 3D FE calibration
analyses
Page 197/272
E8.3 Normalisation of soil reactions
The soil reaction curves extracted from the 3D analysis are normalised using the dimensionless
groups listed in Table 27.
Table 27: Normalisation of pile reaction components. Note that 𝝈′𝒗𝒊 is the local value of the
initial vertical effective stress.
Component Clay normalisation Sand Normalisation
𝑝 𝑝
̅
Distributed load, 𝒑 ′
𝑠𝑢 𝐷 𝜎𝑣𝑖 𝐷
𝑣 𝑣 𝑝𝑎 𝑣 𝐺
̅
Lateral displacement, 𝒗 𝐼 𝐼𝑠 √ ′ [= ′ ]
𝐷 𝑅 𝐷 𝜎𝑣𝑖 𝐷 𝜎𝑣𝑖
𝑚 𝑚
̅
Distributed moment, 𝒎
𝑠𝑢 𝐷 2 𝑝𝐷
𝑝𝑎 𝐺
̅
Pile rotation, 𝝍 𝜓𝐼𝑅 𝜓𝐼𝑠 √ ′ [= 𝜓 ′ ]
𝜎𝑣𝑖 𝜎𝑣𝑖
𝐻𝐵 𝐻𝐵
̅𝑩
Base shear load, 𝑯 ′ 2
𝑀𝐵 𝑀𝐵
̅𝑩
Base moment, 𝑴 ′ 3
The local values of initial vertical effective stress used in the normalisation process are calculated
from the average values determined over the height of each element, based on the idealised dense
sand profile described in Appendix A4.
It should be carefully noted that the process that is employed for the distributed moment component
for the sand involves an approach in which the local distributed moment, 𝑚, is normalised with
respect to the local distributed load, 𝑝. This approach has been adopted on the basis that, for a
frictional soil/pile interface, the magnitude of the pile/soil shear tractions (and, as a consequence, the
magnitude of the distributed moment) is likely to be strongly conditioned by the magnitude of the
local lateral stresses. The distributed moment is therefore thought more likely to be correlated with
the distributed load 𝑝 than, for example, with the local vertical effective stress.
To illustrate the use of the normalisation approach for distributed moment given in Table 27, data
are provided in Figure 184 for distributed load and distributed moment in which both sets of data are
′
normalised using the local vertical effective stress, 𝜎𝑣𝑖 . At particular values of 𝑧⁄𝐷 the general shape
of the normalised distributed load and moment curves appear similar. This seems to indicate that the
data are correlated.
Page 198/272
(a) Normalised distributed load curves (b) Normalised distributed moment curves
Figure 184: Distributed load and moment soil reactions from analysis C4 (D = 10, L/D = 6)
Data on distributed moment normalised with respect to the local distributed load are indicated in
Figure 185. These data conform to a consistent pattern that can be described, straightforwardly,
using a bilinear form of the conic function used to represent the soil reaction curves (Section 5.2.2 of
the main report).
Figure 185: Normalised distributed moment curves from analysis C4 (D = 10, L/D = 6)
E9 Application of normalised soil reaction to the 1D model
E9.1 Validation of applying extracted soil reaction curves to the 1D model
Figure 186 shows a comparison for the C1, C4 and C6 geometries between the load-displacement
responses directly computed by the 3D finite element model and those computed when applying the
extracted and normalised soil reactions to the 1D model; this form of the 1D model is referred to
Page 199/272
below as ‘1D (numerical)’. The data in Figure 186 confirm that, as expected, when the soil reactions
determined from the 3D finite element model are incorporated directly within the 1D model, the
overall computed responses of the 1D and 3D models are similar. The small differences between
the results obtained from the 3D and 1D models are attributed to a combination of factors including
the correction of non-monotonic interpreted soil reaction curves (as described in Appendix D8.3),
any inaccuracies in the assumed kinematics of the Timoshenko beam elements employed in the 1D
model, errors associated with the Winkler assumption and residual convergence errors in the 3D
finite element modelling.
Figure 186: Load-displacement response comparison of the 3D finite element results and the
1D (numerical) model.
The 1D (numerical) model is found to reproduce the shape of the 3D finite element calibration
analysis results for both large and small displacements, with average accuracies of 95% and 99%
respectively, over the range of calibration simulations (see Figure 187).
Page 200/272
Pile
D
L/D
M/HD
D/t
Pile
D
L/D
M/HD
D/t
Figure 187: Metrics for the accuracy of the 1D (numerical) model
Page 201/272

calibration analyses and the corresponding results obtained using the 1D (numerical) model.
Analysis C4, (D = 10m, L = 60m)
For Analysis C4 (L/D = 6), the 1D (numerical) model provides a good fit with the 3D calibration data.
Separate analyses have been conducted in which the distributed moment, and the base shear and
moment are removed from the analysis. The results, plotted in Figure 188(a) and (b), show that the
two sets of data are similar indicating that, in this case, the distributed moment and the base shear
and moment have a minimal influence on the behaviour of the pile.
Figure 189 shows the bending moment profiles obtained from the 1D (numerical) model for the case
(i) when all four soil reaction terms are included and (ii) when only the distributed load term is
included. It is clear that, for his particular analysis, the distributed moment and the base
shear/moment components of soil reaction have a relatively minor influence on the computed
bending moment, compared with the lateral load component, especially at large displacements.
Page 202/272
Figure 189: Comparison of bending moment profiles obtained using the 1D (numerical) model
and soil reactions determined from the 3D calibration analyses. Analysis C4, (D= 10m, L =
60m)
For Analysis C1 (L/D = 2), the 1D model is also seen to provide a good fit with the 3D calibration
data (Figure 190(a)). However, the lateral load response computed using the 1D model including
only the distributed load soil reactions falls significantly below the 3D calibration data. In this case, in
which the pile is relatively short, it is clear that the other three soil reaction components need to be
included in the 1D model to obtain reliable results.

calibration analyses and the corresponding results obtained using the 1D (numerical) model.
Analysis C1 (D = 10m, L = 20m)
Results of a more detailed exploration of the contribution of each soil reaction component in the 1D
(numerical) model for analysis C1 are shown in Figure 191. In this case, the distributed moment is
Page 203/272
seen to provide a significant contribution to the overall response, with the two base components
having a relatively small influence.
Figure 191: Influence of the various components of soil reaction on the computed response
for Analysis C1 (D = 10m, L = 20m)
Figure 192 demonstrates that when selected soil reaction components are excluded for the shorter
geometry, the under-prediction of the load-displacement response (compared with the 3D finite
element data) is accompanied by a corresponding under-prediction of the bending moments. Whilst
the base reaction components have some influence, especially at small displacements, the majority
of the response appears to be captured by the distributed load and distributed moment components.
Figure 192: Bending moment profiles obtained using the 1D (numerical) model. Separate plots
are provided to demonstrate the influence of individual soil reaction components. Analysis C1
(D = 10m, L = 20m)
Page 204/272
E10 Parameterised curve fitting

As shown in Figure 184, the normalised distributed load exhibits behaviour, at relatively shallow
depths, in which a peak occurs in the response, followed by softening behaviour. It is presumed that
similar behaviour would be observed at greater depths along the pile provided that sufficiently large
lateral displacement are imposed. It seems plausible that the existence of a peak in the distributed
load response is attributable to the tendency of the (dense) sand to dilate; in cases where the sand
is loose, peak behaviour of the form shown in Figure 184 may not occur. In any practical application
of the 1D design method, the tendency of the distributed load to exhibit peak behaviour (which will
depend, for example, on the local relative density and stress level) is difficult to model in detail. The
process adopted here to parameterise the soil reaction curve for the distributed load provides a
relatively simple way of representing this behaviour. The development of more detailed approaches
is likely to be the subject of future work.
The computed normalised soil reaction curves are represented by the four-parameter conic form that
is defined in Section 5.2.2 of the main report. This form cannot represent the peak behaviour, and
post-peak softening, that is apparent in Figure 184. A more complex curve shape featuring a post-
peak softening could be developed, with additional parameters, but preliminary experimentation has
demonstrated that this approach does not offer any improvement in robustness or accuracy, and
risks over-fitting of the soil response.
E10.1 Distributed lateral load soil reaction curve
The distributed load curves (for Pile C4) are represented using the conic form shown in Figure 193.
p
σ'viD
np=0
pu
np
kp
np=1
vpu v pa
I
D S σ'vi
Figure 193: Soil reaction curve for distributed load

To fit the conic form to the computed data, four parameters are required at each depth; the initial
stiffness kp, the ultimate response 𝑝̅𝑢 , the ultimate displacement 𝑣̅𝑝𝑢 and the curvature np.
The first stage of fitting to the calibration data is to calculate the initial stiffness, kp, for the distributed
load, which has a significant influence on the small displacement pile response. The initial stiffness
kp is selected by fitting the linear expression 𝑝̅ = 𝑘𝑝 𝑣̅ using a least squares fit over the small
displacement region 0 < 𝑝̅ < 0.1. The variation of this initial stiffness parameter with depth, for each
of the calibration analyses, is shown in Figure 194. To provide a simple expression that can be
Page 205/272
applied within the 1D model, a linear depth variation function (showed by a dashed line) has been
determined using a least squares regression across the dataset. Note that by fitting a linear
expression the initial stiffness at small depths will be underestimated.
Figure 194: Initial stiffness parameter kp variation for analyses C1 to C11
Developing an approach to define the depth variation functions for the ultimate response, 𝑝̅𝑢 , is
rather less straightforward, as a consequence of the post-peak softening behaviour that has been
found to occur. One possible approach is to determine 𝑝̅𝑢 from the peak of the soil reaction curve
and to ignore the post-peak softening behaviour; soil reaction curves obtained using this approach
are indicated in Figure 195(a) as dashed lines. However, this approach (referred to below as Method
A) has been found to cause the stiffness of the monopile, computed using the 1D model, to be
unrealistically large, especially for shorter monopiles. An alternative approach, referred to below as
Method B, is to relate 𝑝̅𝑢 to the ultimate response of the pile determined at large displacements (i.e.
at the end of the calibration analysis); soil reactions determined using this approach are indicated on
Figure 195(b) as dashed lines. A key difficulty here, however, is that the ultimate value of the lateral
load is somewhat arbitrary as it depends on the magnitude of the lateral displacements actually
applied during the 3D calibration analyses
Page 206/272
̅𝒖 at peak value (Method A)

(a) 𝒑 ̅𝒖 at ultimate value (Method B)
(b) 𝒑
Figure 195: Distributed load and parametric curve fits for analysis C4 (D = 10, L/D = 6).
Numerical soil reaction curves are shown as solid lines. Parametric equations are shown as
dashed lines.
The proposed soil reaction curve for the distributed load, for use in the 1D (parametric) model, is
based on a representation that is intermediate between Methods A and B. Initially, separate soil
reaction curves are determined for each of Method A and B. For each method, the ultimate
displacement 𝑣̅𝑝𝑢 is determined from the displacement at which the ultimate response 𝑝̅𝑢 first occurs
or is first exceeded. Then, the curvature parameter np is determined by minimising the proportional
least squares error of Equation 3 of section 5.2.2 of the main report over the region 0 < 𝑣̅ <𝑣̅𝑝𝑢 using
a suitable optimisation function (e.g. fmincon in MATLAB).
Figure 196 shows the variation of the parameter values for each approach, averaged over binns of
width z/20L. It should be carefully noted that two different forms of normalised depth are employed in
this figure. For np, the data are found to correlate well with normalised depth expressed as z/D. It
was discovered, however, that for 𝒑 ̅𝒖 and 𝒗 ̅𝒑𝒖 , an improved correlation could be achieved by
adopting an alternative approach in which the normalised depth is taken to be z/L. To obtain a soil
reaction curve model for practical applications, linear depth variation functions have been selected
that fall between the Method A and Method B lines in Figure 196. These functions are specified in
Table 28.
.
Page 207/272
(a) Curvature parameter np ̅u

(b) Ultimate response paramter 𝒑
̅pu
(c) Ultimate displacement paramter 𝒗
Figure 196: Soil reaction parameter variations with depth (averaged over bin widths of z/20L)
and selected depth variation functions from analyses C1 to C11
E10.2 Distributed moment curves
A bi-linear form of the conic form Figure 197(b), has been adopted by specifying nm = 0 and ensuring
that 𝜓̅𝑚𝑢 > 𝑚̅ 𝑢 /𝑘𝑚 ; this results in the curve shape shown in Figure 197(a). The response tends to a
limiting value 𝑚 ̅ 𝑢 after a very sharp initial rise. To determine the model parameters, the following
approach has been adopted:
 A suitably high value of initial stiffness is chosen (km = 20).
 The ultimate moment parameter 𝑚 ̅ 𝑢 is selected by taking the mean of the values that satisfy
𝑚
̅ > 0.9𝑚̅ 𝑓𝑖𝑛𝑎𝑙 at each soil reaction depth, where 𝑚 ̅𝑓𝑖𝑛𝑎𝑙 is the value of distributed moment at a
given depth at the last simulated increment
Page 208/272
m
pD
mu
km
ψmu ψIs(pa/σ’vi)½
(a) 3D finite element extracted and (b) Parametric curve shape

parameterised fit for C4
Figure 197: Distributed moment soil reaction curve shape for normalised curves
The variation of the ultimate moment with normalised depth (𝑧⁄𝐿) for each calibration analysis is
shown in Figure 198. Also shown is a linear depth variation function, which has been determined
using a least squares fit. The depth variation function is specified in Table 28.
̅u
Figure 198: Variation of normalised distributed moment ultimate response paramter 𝒎
Page 209/272
E10.3 Base horizontal force curves
Figure 199: Normalised base horizontal force for analysis C4 (D = 10, L/D = 6)
Figure 199 shows an example of a base horizontal force soil reaction curve (for the C4 analysis).
Similar to the lateral distributed load, this base reaction indicates behaviour in which the response
reaches a peak and then softens. To represent this response using the conic function, the following
approach has been adopted:
1. The initial stiffness parameter kH, is calculated from a least squares regression for 0 < 𝑣̅𝐻 <
0.01.
2. The ultimate displacement 𝑣̅𝐻𝑢 is selected as the first point at which the ultimate response
parameter is met or exceeded.
3. The ultimate response parameter 𝐻 ̅𝐵𝑢 is calculated from the average of the normalised base
shear values in the region 𝑣̅𝐻 > 𝑣̅𝐻𝑢 .
4. The curvature parameter nH is fitted by minimising the proportional least square error of
Equation 3 in section 5.2.2 of the main report over the region 0 < 𝑣̅ <𝑣̅𝐻𝑢 using a suitable
optimisation function (e.g. fmincon in MATLAB).
Linear depth variation functions for each soil reaction parameter have been determined using a least
squares regression. The resulting functions are shown in Figure 200. The corresponding equations
are given in Table 28.
Page 210/272
(a) kH Initial stiffness parameter (b) nH Curvature parameter
̅ Bu Ultimate response parameter

(c) 𝑯 ̅Bu Ultimate displacement parameter
(d) 𝒗
Figure 200: Depth variation functions for the base shear force soil reaction curve parameters
Page 211/272
E10.4 Base moment curves
Figure 201: Normalised base moment reaction for analysis C4 (D = 10, L/D = 6)
Figure 201 shows the normalised base moment reaction curve for the C4 analysis. It is noted that
the base moment reaction does not exhibit a peak. The following process is adopted for fitting the
parameterised curve to the computed data:
1. The initial stiffness parameter kM, is calculated using a least squares regression over the
region 0 < 𝑣̅𝑀 < 0.05.
2. Due to the lack of an obvious peak in the response, a value of ultimate rotation parameter
𝜓̅𝑀𝑢 is manually selected (𝜓̅𝑀𝑢 = 50) that exceeds the observed rotations and allows a
reasonable value of curvature parameter to be selected.
3. The curvature parameter nH and ultimate response parameter 𝑀 ̅𝐵𝑢 are selected by
minimising the proportional least square error of Equation 3 in section 5.2.2 of the main
report using a suitable optimisation function (e.g. fmincon in MATLAB).
The fitted parameter values and depth variation functions are shown in Figure 202 are specified in
Table 28.
(a) kM (b) nM ̅ Bu
(c) 𝑴
Figure 202: Depth variation functions for the base moment soil reaction curve parameters
Page 212/272
E10.5 Soil reaction curves and depth variation functions
Parameters for the example dense sand site soil reaction curves and the depth variation functions,
for the idealised dense sand site, are listed in Table 28)
Table 28: Parameters for normalised soil reaction curves in Dunkirk sand, calibrated within
the parameter space defined in Table 26 (𝟐 < 𝑳⁄𝑫 < 𝟔, 𝟓 < 𝑫 < 𝟏𝟎, 𝟓 < 𝒉⁄𝑫 < 𝟏𝟓, 𝟔𝟎 < 𝑫⁄𝒕 <
𝟏𝟏𝟎)

Distributed lateral load, 𝒑 Ultimate strain, 𝑣̅ pu 53.1
𝑧
𝐷
Curvature, np 0.944
𝑧
Ultimate reaction, 𝑝̅u −10.18 + 21.61
𝐿
Initial stiffness, km 20
Curvature, nm 0
𝑧
̅u −0.05 + 0.21
𝐿
Base shear, 𝑯𝑩 𝐿
Ultimate strain, 𝑣̅ Hu −0.29 + 2.31
𝐷
𝐿
𝐷
𝐿
𝐷
𝐿
̅ Bu
Ultimate reaction, 𝐻 −0.07 + 0.62
𝐷
Initial stiffness, kM 0.29
Curvature, nM 0.89
𝐿
̅ Bu
𝐷
Page 213/272
E11 Design calculations conducted using the 1D model
E11.1 Accuracy of the parametric expressions when applied to the 1D model
A study has been conducted to assess the performance of the 1D (parametric) model (i.e. the form
of the model that is based on the soil reaction curves specified in Table 28 ). Results obtained using
the 1D (parametric) approach are compared with the API/DNV p-y method (described in Appendix
F4).

calibration analyses and the 1D (parametric) model.
Comparisons between the results obtained using the 1D (parametric) model, the 3D finite element
calibration analyses and the current API/DNV approach for the (short) C1 pile and the (long) C4 pile
are shown in Figure 203. For both piles, the 1D (parametric) model is seen to provide a good match
Page 214/272
with the 3D calibration data up to the ultimate response. The small displacement behaviour
determined using the 1D (numerical) model agrees well with the 3D calibration data for the C4 pile.
However, the 1D (numerical) results are marginally softer than the calibration data for C1. This
discrepancy is expected to occur, due to the use of a linear expression for the depth variation
function for kp; this linear function underestimates the stiffness at shallow depths leading to errors
that are likely to be more significant for short piles..
In general the API/DNV approach performs relatively well, but is inconsistent in whether it over-
predicts or under-predicts the ultimate response computed using the 3D model for different
geometries. The API/DNV approach consistently under-predicts the small displacement stiffness,
especially for the shorter geometries.
As shown in Figure 204(a), the 1D (parametric) model consistently predicts the shape and
magnitude of the ultimate response (compared to the calibration analyses) with an average accuracy
of η0.1D = 96%. This is despite a marginal under-prediction of the response for the shorter D = 5m
piles (C6 and C7). Despite its simplicity, the API/DNV approach also achieves a generally good
average accuracy of η0.1D = 88%.
Figure 204(b) demonstrates that there is a greater disparity in the quality of the prediction for the
small displacement response, with accuracy metrics of ηsd = 92% and ηsd = 43% for the 1D
(parametric) and API/DNV approaches respectively. Whilst the 1D (parametric) approach indicates a
marginal reduction in small displacement accuracy for the shorter pile geometries, the API/DNV
approach exhibits a significant reduction, suggesting that the existing design methods are poorly
conditioned for predicting the small displacement response for shorter geometries.
Page 215/272
D
L/D
M/HD
D/t
D
L/D
M/HD
D/t
Figure 204: Accuracy metrics comparing load-displacement shape accuracy of the 1D

(parametric) model and the API/DNV ’p-y’ method, to the 3D finite element calibration
analyses
E11.2 Interpolated design scenario
The 1D (parametric) model is intended for use as a predictive tool. Once the model has been trained
(using the 3D calibration data), it can be used to determine the performance of a monopile for
arbitrary values of the geometric and loading parameters that lie within the parameter space
employed in the 3D finite element calibration analyses.
Page 216/272
To test the predictive capability of the 1D (parametric) model, two additional test analyses (D1 and
D2) have been conducted using the 3D finite element modelling approach. The geometric
configurations adopted for these tests cases are provided in Table 29 and were selected to fall
within the parameter space adopted for the initial calibration analyses, as shown in Figure 205.
Analysis D h L t
h/D L/D D/t
D1 7.5 37.5 5 22.5 3 68 110
D2 8.75 87.5 10 35 4 91 97
Figure 205: Parametric geometry space for the field tests, 3D finite element calibration
analyses and the two design cases
The load-displacement responses computed using the 3D finite element model and the 1D
(parametric) model are shown in Figure 206 and Figure 207. The results indicate a close fit between
the two sets of data over the full range of applied loading (Figure 206 (a) and Figure 207 (a)). The
small displacement response computed using the 1D (parametric) model is seen to be almost
identical to the response computed using the 3D finite element model (Figure 206 (b) and Figure
207 (b)). Also shown for comparison are the corresponding results obtained using the API/DNV p-y
method.
Page 217/272
Figure 206: Comparison of load-displacement response from the 3D finite element model, the 1D
(parametric) model and the existing API/DNV approach for the D1 design case (D = 7.5, L/D = 3)
Figure 207: Comparison of load-displacement response from the 3D finite element model, the 1D
(parametric) model and the existing API/DNV approach for the D2 design case (D = 8.75, L/D = 4)
The process of parameterising the soil reactions involves a simplification of the post-peak softening
behaviour that is apparent in some of the extracted soil reaction curves. Despite this simplification,
the shape and magnitude of the bending moment distribution along the pile computed using the 1D
(parametric) model compares well with results obtained using the 1D (numerical) model (i.e. the
form of the 1D model that is based directly on the numerical values of the soil reaction curves
determined from the 3D finite element analysis), see Figure 208 and Figure 209.
Page 218/272
Figure 208: Comparison of bending moment distributions from the 1D (numerical), 1D

(parametric) and the current API/DNV approach for the D1 design case (D = 7.5, L/D = 3)
Figure 209: Comparison of bending moment distributions from the 1D (numerical), 1D

(parametric) and the current API/DNV approach for the D2 design case (D = 8.75, L/D = 4)
Page 219/272
E12 Explorative analysis using calibrated soil reactions in the 1D model

This section outlines additional studies conducted using the 1D (parametric) model. In the current
study the pile geometries are:
 pile diameters 5m < D < 10m,

 normalised lengths 2 < L/D < 6
 load eccentricities 5 < M/HD < 15.
E12.1 Contribution of reaction components
The relative importance of the four soil reaction components (distributed load p, distributed moment
m, base shear force HB and base moment MB) is investigated as described below.
E12.1.1 Varied pile diameter
Figure 210 shows the percentage contribution of each soil reaction component to the resistance of
the lateral load, as the pile diameter is varied. The contribution has been calculated by simulating
the response of piles with a fixed L/D = 4, D/t = 110 and M/HD = 5. Components are sequentially
added to the simulation in the order 𝑝 − 𝑣, 𝑚 − 𝜓, 𝐻𝐵 − 𝑣𝐵 and finally 𝑀𝐵 − 𝜓𝐵 . The difference in
computed pile lateral load between successive simulations is attributed to the additional reaction
component. Using the accuracy metric approach (described in Section 4.4 of the main report), the
additional contribution of that component up to the target displacement is quantified. Note that since
the lateral distributed load p-v, dominates the pile response, it has been plotted on the primary (left
hand side) y-axis. The smaller contributions of the remaining components are plotted against the
secondary (right hand side) y-axis.
for L/D = 4, D/t = 110, M/HD = 5. The contribution of the lateral load (solid line) is indicated on
the left axis. The contributions of the other components (dashed lines) are indicated in the
right axis.
Page 220/272
Figure 210 shows that for a pile with an aspect ratio L/D = 4, at both small displacements and the
ultimate response, the contribution of the different components does not vary significantly and the
distributed load and distributed moment are the most significant components. The base soil reaction
components are shown to have negligible impact on the pile capacity.
E12.1.2 Varied pile slenderness L/D
Figure 211 shows the contribution of the different soil reaction components when a pile of a fixed D
= 7m, D/t = 110 and M/HD = 5 is varied in slenderness 2 < L/D < 6.
7m, D/t = 110, M/HD = 5. The contribution of the distributed load (solid line) is indicated on the
left axis. The contributions of the other components (dashed lines) are indicated in the right
axis.
Figure 211 shows that the contribution of the distributed moment is particularly significant at small
L/D ratios and should also be considered for longer pile lengths. At small displacements for shorter
pile geometries, the base shear and base moment components can also contribute a significant
proportion of the small displacement stiffness (16% for L/D = 2), but are relatively insignificant for
most other configurations.
Page 221/272
PISA Project – Final Report Appendix F - Analysis tools
Appendix F Analysis tools
F1 Formulation of the 1D model
F1.1 Specification of the modelling procedure
The modelling procedure adopted for the analysis of an embedded monopile is based on extensions
to the conventional p-y approach to include additional soil reaction components that have been
found to be significant for monopiles with relatively small length-to-diameter ratios. These extensions
follow previous work by Davidson (1982), Lam and Martin (1986) and Lam (2013) for the design of
caissons, principally for onshore applications.
A 1D representation is adopted for the monopile and tower (see Figure 212). In the formulation
presented below, a horizontal force 𝐻 and and a moment 𝑀 is assumed to be applied at the top of
the tower. (In all of the analyses presented in this report, however, the applied moment at the top of
the tower is set to zero). The coordinate directions and sign conventions employed in the formulation
are as indicated in Figure 212. To reflect the general nature of the analysis framework, the use of
the nomenclature ‘p-y curve’ is avoided in the discussion given below (on the basis that it implies a
specific coordinate system and is not readily generalized to other components of soil reaction).
M H
Mudline
z v
Lateral soil
reaction
p(z,v)
L
Distributed
moment
m(z,)
Base shear
force HB(vB)
Base
moment
MB(B)
Figure 212: Key features of the 1D finite element model
Four separate components of soil reaction are assumed to be applied to the pile – a lateral load 𝑝, a
distributed moment 𝑚, a base horizontal force 𝐻𝐵 and a base moment 𝑀𝐵 . The model is formulated
within a variational (virtual work) framework; this provides a consistent way of incorporating the soil
Page 222/272
reaction curves within the model (e.g. it avoids the need to lump the various soil reactions at the
nodes in an arbitrary way). The soil reaction curves employed in the analysis are based on the
mathematical forms given in section 5.2.2 of the main report.
The pile is represented in the model as a line of beam finite elements. In the current formulation, 2-
noded, 5 degree-of-freedom beam elements (based on the formulation given in Astley 1992) are
employed. The soil is represented in the formulation by a separate set of 2-noded ‘soil finite
elements’ that are connected to the beam elements (at the nodes) used to represent the pile. The
soil deformations (rotation and displacement) are therefore prescribed to conform to the pile
displacements and rotations along the embedded length. The soil response is incorporated within
the analysis on the basis of the Winkler assumption (i.e. local values of load and moment are
assumed to depend only on local values of displacement and rotation). The soil reaction curves are
incorporated within the model in a way that is analogous to the incorporation of constitutive models
within a conventional finite element framework.
F1.2 Formulation and implementation
F1.2.1 Virtual work equations

The displacements in the pile are represented using Timoshenko beam theory as indicated in Figure
213.
y, v Pile cross-section

Pile/tower
z, w
Figure 213: Timoshenko beam deformations
The assumed displacement field (for the case where axial strains on the neutral axis are zero) is:
𝑤(𝑦, 𝑧) = 𝑤0 + 𝑦𝜓(𝑧) 𝑣(𝑦, 𝑧) = 𝑣𝑜 (𝑧) Equation 27
where 𝑤0 is a (constant) axial displacement and 𝑣𝑜 (𝑧) is the lateral displacement of the centroid of
the beam. The corresponding strains are:
𝜕𝑤 𝑑𝜓 𝜕𝑣 𝜕𝑤 𝜕𝑣 𝑑𝑣
𝜀𝑧𝑧 = =𝑦 ; 𝜀𝑦𝑦 = =0 ; 𝛾𝑦𝑧 = + = 𝜓+ Equation 28
𝜕𝑧 𝑑𝑧 𝜕𝑦 𝜕𝑦 𝜕𝑧 𝑑𝑧
The bending moment in the pile (a positive moment causes tension on the left hand side of the
pile/tower as indicated in Figure 212) is:
Page 223/272
𝑑𝜓
𝑀𝑖𝑛𝑡 = ∫ −𝐸𝜀𝑧𝑧 𝑦 𝑑𝑥 = −𝐸𝐼 Equation 29
𝑑𝑧
The shear force in the pile (sign convention; shear force is positive in the tower for a horizontal load
applied as shown in the Figure 212) is:
𝑑𝑣
𝑉𝑖𝑛𝑡 = −𝐺𝐴𝜅 (𝜓 + ) Equation 30
𝑑𝑧
where 𝐺 is the shear modulus, 𝐴 is the area and 𝜅 is a shear factor (in all of the 1D analyses
presented in this report κ = 0.5, E = 210GPa. A Poisson’s ratio ν = 0.3 is adopted to calculate G).
Using the terminology of Reddy (2002), at equilibrium the sum of the internal and external virtual
work is zero. This gives:
0 = 𝛿𝑊𝐸 + 𝛿𝑊𝐼 Equation 31
where 𝛿𝑊𝐸 is the external virtual work, given by:
𝛿𝑊𝐸 = −𝐻𝛿𝑣𝑇 − 𝑀𝛿𝜓 𝑇 Equation 32
where 𝛿𝑣𝑇 is the virtual displacement at the top of the tower and 𝛿𝜓 𝑇 is the virtual rotation at the top
of the tower. The internal virtual work, 𝛿𝑊𝐼 , is:
𝑑𝛿𝜓
𝛿𝑊𝐼 = ∫ [−𝑀𝑖𝑛𝑡 − 𝑉𝑖𝑛𝑡 𝛿𝛾 + 𝑝(𝑧, 𝑣)𝛿𝑣 + 𝑚(𝑧, 𝜓)𝛿𝜓] 𝑑𝑧 + 𝐻𝐵 𝛿𝑣𝐵 + 𝑀𝐵 𝛿𝜓𝐵
𝑑𝑧
𝑑𝜓 𝑑𝛿𝜓 𝑑𝑣 𝑑𝛿𝑣 Equation 33
= ∫ [ 𝐸𝐼 + ( + 𝜓) 𝐺𝐴𝜅 ( + 𝛿𝜓) + 𝑝(𝑧, 𝑣)𝛿𝑣 + 𝑚(𝑧, 𝜓)𝛿𝜓] 𝑑𝑧 + 𝐻𝐵 𝛿𝑣𝐵
𝑑𝑧 𝑑𝑧 𝑑𝑧 𝑑𝑧
+ 𝑀𝐵 𝛿𝜓𝐵 𝑑𝑧
where the integration is taken over the entire length of the pile and tower.
F1.2.2 Displacement interpolation employed for the beam and soil elements
The formulation employed for the finite elements used to represent the pile is based on the 2-noded
5 degree-of-freedom element given in Astley (1992). The approach used to interpolate the
displacement and rotation within these beam elements is specified below. The same interpolation
approach is employed for the 2-noded elements that are used to model the soil.
The lateral displacement within each beam (and soil) element is given by the interpolation equation:
𝑣 = 𝑁1 𝑉1 + 𝑁2 θ1 + 𝑁3 𝑉2 + 𝑁4 θ2 Equation 34
Page 224/272
where 𝑁𝑖 are the conventional set of Hermite cubic interpolation functions and 𝑉𝑗 , θ𝑗 are the nodal
values of lateral displacement and overall rotation . Following Astley (1992), it is assumed that within
each element the shear strain is constant, 𝛾0 . On this basis, the overall rotation within each element
is 𝜃 = (𝑑𝑣⁄𝑑𝑧) = 𝛾0 − 𝜓. The interpolated displacements may therefore be written:
𝑣 = 𝑁1 𝑉1 − 𝑁2 Ψ1 + (𝑁2 + 𝑁4 )𝛾0 + 𝑁3 𝑉2 − 𝑁4 ψ2 Equation 35
where Ψi are the nodal values of the cross-section rotation 𝜓 (assumed to be continuous at the
element nodes) and 𝛾0 is treated as an additional element degree-of-freedom.
The detailed formulation of the finite element equations for the pile and the soil may be derived from
this displacement interpolation, on the basis of standard procedures.
In the analyses presented in this report, the various integrals that arise from the discretized form of
Equation 33 are evaluated using Gauss integration with four Gauss points per element. Solutions to
the resulting finite element equations are obtained using a Newton-Raphson scheme.
Page 225/272
F2 Embedded pile response calculation

During the Cowden and Dunkirk tests, most of the medium diameter piles and the large diameter
piles were instrumented with inclinometers, extensometers and fibre optic strain gauges, as
illustrated in Figure 214. The data from this instrumentation, in combination with the calculated
ground level displacement and rotation allows for the calculation of the pile behaviour below ground
level.
Figure 214: Illustration of the instrumentation below ground level

A procedure to calculate the pile displacements and an equivalent distributed lateral load is
described below.
F2.1 Optimisation process for interpretation of pile behaviour
Various options are available to determine the below-ground pile behaviour on the basis of data
measured during the field tests. In the current approach, the pile behaviour is determined using a
least squares optimisation process, where an optimal fit is found to the inclinometer and the fibre
optic data. The extensometer data is found to be generally consistent with the fibre optic data.
However, as the fibre optic system appears to provide more accurate and reliable data, the
extensometer data is not employed in the optimisation process. The measured data on strain and
inclination are used in combination with a Timoshenko beam model for the monopile to extract an
equivalent set of lateral load soil reaction curves (p-v). It is noted that there are insufficient data to
determine the distributed moment applied along the pile and the force and moment applied at the
base of the pile.
Each monopile has a uniform flexural rigidity. In the following analysis, the flexural rigidity adopted
for the monopile includes an allowance for the stiffness of the tubes for the extensometers and the
inclinometers. For the shear stiffness, only the circular section of the foundation pile is considered;
the tubes are assumed to have only a minor influence on the shear stiffness.
F2.2 Governing equations for the foundation model
The governing equations for the pile, based on Timoshenko beam theory, are:
Page 226/272
𝑑𝜓
𝑀𝑖𝑛𝑡 = −𝐸𝐼 Equation 36
𝑑𝑧
𝑑𝑣 𝑉𝑖𝑛𝑡
𝜓 = −( + ) Equation 37
𝑑𝑧 𝜅𝐴𝐺
where 𝜓 is the (clockwise) rotation of the beam cross-section, 𝑀𝑖𝑛𝑡 is the bending moment (a
positive moment causes tension on the left hand side of the stick-up), 𝑉𝑖𝑛𝑡 is the shear force, 𝐸𝐼 is
the flexural stiffness, 𝜅𝐴𝐺 is the shear stiffness and 𝑣 is the lateral displacement in the 𝑦-direction. In
the following analysis it is assumed that: 𝜅 = 0.3, 𝐸 = 210 GPa and 𝐺 = 80.77 GPa.
𝑀𝐺 𝑀𝐺
𝐻𝐺 𝐻𝐺
𝑦
𝑧 𝑝
𝐿 𝑝
Figure 215: Illustration of the discretisation of the pile
The pile is divided into 𝑛 equal sections along its depth, where the soil reaction is defined by:
𝑝 = 𝑎𝑧 + 𝑏 Equation 38
The shear force and moment along the foundation pile are found by integrating over the soil
reaction:
𝑎𝑧 2
𝑉𝑖𝑛𝑡 = ∫ 𝑝 𝑑𝑧 = + 𝑏𝑧 + 𝑐 Equation 39
2
𝑎𝑧 3 𝑏𝑧 2
𝑀𝑖𝑛𝑡 = ∫ 𝑉 𝑑𝑧 = + + 𝑐𝑧 + 𝑑 Equation 40
6 2
Combining this with the governing equations of the Timoshenko beam gives:
𝑀𝑖𝑛𝑡 𝑑2𝑣 1 𝑑𝑉𝑖𝑛𝑡

=( 2+ ) Equation 41
𝐸𝐼 𝑑𝑧 𝜅𝐴𝐺 𝑑𝑧
𝑑2𝑣 𝑑𝑉𝑖𝑛𝑡 𝑎𝑧 3 𝑏𝑧 2
𝐸𝐼 = 𝑀𝑖𝑛𝑡 − 𝜆 = + + 𝑐𝑧 + 𝑑 − 𝜆(𝑎𝑧 + 𝑏) Equation 42
𝑑𝑧 2 𝑑𝑧 6 2
Page 227/272
with 𝜆 = 𝐸𝐼/𝜅𝐴𝐺. The cross-section rotation, overall rotation and lateral displacement are then given
as:
𝑎𝑧 4 𝑏𝑧 3 𝑐𝑧 2
−𝐸𝐼 𝜓 = + + + 𝑑𝑧 + 𝑒 Equation 43
24 6 2
𝑎𝑧 4 𝑏𝑧 3 𝑐𝑧 2 𝑎𝑧 2
−𝐸𝐼 𝜃 = + + + 𝑑𝑧 + 𝑒 − 𝜆 ( + 𝑏𝑧 + 𝑐) Equation 44
24 6 2 2
𝑎𝑧 5 𝑏𝑧 4 𝑐𝑧 3 𝑑𝑧 2 𝑎𝑧 3 𝑏𝑧 2
𝐸𝐼 𝑣 = + + + + 𝑒𝑧 + 𝑓 − 𝜆 ( + + 𝑐𝑧) Equation 45
120 24 6 2 6 2
In the above equations variables 𝑎 to 𝑓 are unknowns which have to be determined for each of the 𝑛
sections of the pile.
F2.3 Continuity between the sections
At the interface of two sections, continuity is enforced for the displacement, cross-section rotation,
moment, shear force, and lateral load. This results in the following equations:
𝑎𝑖 𝑧𝑖5 𝑏𝑖 𝑧𝑖4 𝑐𝑖 𝑧𝑖3 𝑑𝑖 𝑧𝑖2 𝑎𝑖 𝑧𝑖3 𝑏𝑖 𝑧𝑖2

𝑣𝑖 = 𝑣𝑖+1 → + + + + 𝑒𝑖 𝑧𝑖 + 𝑓𝑖 − 𝜆 ( + + 𝑐𝑖 𝑧𝑖 )
120 24 6 2 6 2
𝑎𝑖+1 𝑧𝑖5 𝑏𝑖+1 𝑧𝑖4 𝑐𝑖+1 𝑧𝑖3 𝑑𝑖+1 𝑧𝑖2
= + + + + 𝑒𝑖+1 𝑧𝑖 + 𝑓𝑖+1 Equation 46
120 24 6 2
𝑎𝑖+1 𝑧𝑖3 𝑏𝑖+1 𝑧𝑖2
−𝜆( + + 𝑐𝑖+1 𝑧𝑖 )
6 2
𝑎𝑖 𝑧𝑖4 𝑏𝑖 𝑧𝑖3 𝑐𝑖 𝑧𝑖2 𝑎𝑖+1 𝑧𝑖4 𝑏𝑖+1 𝑧𝑖3 𝑐𝑖+1 𝑧𝑖2

𝜓𝑖 = 𝜓𝑖+1 → + + + 𝑑𝑖 𝑧 + 𝑒𝑖 = + + + 𝑑𝑖+1 𝑧𝑖 + 𝑒𝑖+1 Equation 47
24 6 2 24 6 2
𝑎𝑖 𝑧𝑖3 𝑏𝑖 𝑧𝑖2 𝑎𝑖+1 𝑧𝑖3 𝑏𝑖+1 𝑧𝑖2

(𝑀𝑖𝑛𝑡 )𝑖 = (𝑀𝑖𝑛𝑡 )𝑖+1 → + + 𝑐𝑖 𝑧𝑖 + 𝑑𝑖 = + + 𝑐𝑖+1 𝑧𝑖 + 𝑑𝑖+1 Equation 48
6 2 6 2
𝑎𝑖 𝑧𝑖2 𝑎𝑖+1 𝑧𝑖2

(𝑉𝑖𝑛𝑡 )𝑖 = (𝑉𝑖𝑛𝑡 )𝑖+1 → + 𝑏𝑖 𝑧𝑖 + 𝑐𝑖 = + 𝑏𝑖+1 𝑧𝑖 + 𝑐𝑖+1 Equation 49
2 2
𝑝𝑖 = 𝑝𝑖+1 → 𝑎𝑖 𝑧𝑖 + 𝑏𝑖 = 𝑎𝑖+1 𝑧𝑖 + 𝑏𝑖+1 Equation 50
where 𝑧𝑖 is the coordinate of the interface between section 𝑖 and section 𝑖 + 1. For each additional
section, the five additional equations provide six unknowns. The degree of non-determination of the
foundation model is therefore equal to DoND = 5 + 𝑛.
Page 228/272
F2.4 Physical constraints
In the approach of finding a solution where the soil reaction is governed by the lateral soil reaction
along the pile, the soil reactions at the base of the foundation can be set to zero.
𝑎𝑛 𝐿2
𝐻𝐵 = 0 = + 𝑏𝑛 𝐿 + 𝑐𝑛 Equation 51
2
𝑎𝑛 𝐿3 𝑏𝑛 𝐿2
𝑀𝐵 = 0 = + + 𝑐𝑛 𝐿 + 𝑑𝑛 Equation 52
6 2
Due to errors in the measured data, the optimised solution for the soil reaction may give a reaction in
the same direction as the displacement at ground level which is physically unrealistic. The lateral
reaction at ground level is therefore also constrained to zero:
𝑝𝐺 = 0 = 𝑏1 Equation 53
Using these additional constraints, the degree of non-determination reduces to: DoND = 5 + 𝑛 − 3.
F2.5 Boundary conditions
The displacement, overall rotation, moment and shear force at ground level, obtained from the
factual report, provide boundary conditions which are used in combination with the continuity
conditions to solve for the unknowns:
𝑓1
𝑣𝐺 = Equation 54
𝐸𝐼
𝑒1 − 𝜆𝑐1
𝜃𝐺 = − Equation 55
𝐸𝐼
𝑀𝐺 = 𝑑1 Equation 56
𝑉𝐺 = 𝑐1 Equation 57
Along the foundation pile, the inclinometers and strain gauges also provide sets of boundary
conditions:
4 3 2 2
1 𝑎𝑖 𝑧𝐼𝑃𝐼 𝑏𝑖 𝑧𝐼𝑃𝐼 𝑐𝑖 𝑧𝐼𝑃𝐼 𝑎𝑖 𝑧𝐼𝑃𝐼
𝜃𝐼𝑃𝐼 = − ( + + + 𝑑𝑖 𝑧𝐼𝑃𝐼 + 𝑒𝑖 − 𝜆 ( + 𝑏𝑖 𝑧𝐼𝑃𝐼 + 𝑐𝑖 )) Equation 58
𝐸𝐼 24 6 2 2
Page 229/272
𝑎𝑖 𝑧𝑆3 𝑏𝑖 𝑧𝑆2
𝑀𝑆 = + + 𝑐𝑖 𝑧𝑆 + 𝑑𝑖 Equation 59
6 2
where 𝜃𝐼𝑃𝐼 is the rotation measured by the inclinometers and 𝑀𝑆 is the bending moment determined
from strains measured by the fibre optic system. Subscript 𝑖 indicates which pile section the
instrument is located on.
In finding an optimal fit to the measured data, an appropriate ratio needs to be used of the number of
specified boundary conditions over the degree of non-determination. It is found that if this ratio is
bigger than 2, only counting the data points on one side of the foundation, then the optimised fit
gives good results. This constrains the number elements that can be employed.
F2.6 Least-squares solution to the problem
Values of the undetermined parameters 𝑎𝑖 to 𝑓𝑖 are found by obtaining a least-squares solution to

the boundary condition equations subject to the continuity equations. The process is described
below.
F2.6.1 Instrument weighting

It is desirable to apply suitable weighting factors to the LVDT and inclinometer measurements to
ensure that each data set has an appropriate influence on the computed least-squares solution for
the undetermined parameters. In the current approach, it is desired that the weighting given to the
ground level data, inclinometer data and fibre optic data should be broadly comparable. Moreover,
the weightings employed for the instruments are required to normalise the data, to reflect the
anticipated magnitudes of the measurements in each set of data. The weighting factors using this
approach are specified below:
1
𝑤𝑣 = Equation 60
0.1𝐷
1
𝑤𝜃 = Equation 61
2∘
1
𝑤𝑀 = Equation 62
𝑀0.1𝐷
1
𝑤𝑉 = Equation 63
𝑉0.1𝐷
F2.7 Boundary condition and continuity equations
For the least squares solution to the problem, the equations are expressed as 𝐀𝐱 = 𝐛 for the
boundary conditions and 𝐂𝐱 = 𝐝 for the continuity equations and physical constraints. In these
equations, 𝐱 = [𝑎1 𝑏1 𝑐1 𝑑1 𝑒1 𝑓1 … 𝑎𝑛 𝑏𝑛 𝑐𝑛 𝑑𝑛 𝑒𝑛 𝑓𝑛 ]𝑇 , is the vector of unknowns to be determined.
The matrix for the continuity equations and physical constraints 𝐂 is a 5(𝑛 − 1) + 3 by 6𝑛 matrix and
𝐝 is a vector with 5(𝑛 − 1) + 3 elements. These are constructed as:
Page 230/272
𝑧𝑖5 𝑧𝑖3 𝑧𝑖4 𝑧𝑖2 𝑧𝑖3 𝑧𝑖2

−𝜆 −𝜆 − 𝜆𝑧𝑖 𝑧𝑖 1
120 6 24 2 6 2
𝑧𝑖4 𝑧𝑖3 𝑧𝑖2 0
𝑧𝑖 1 0
24 6 2 0
𝐂𝐢 = 𝑧𝑖3 𝑧𝑖2 ; 𝐝𝐢 = 0 Equation 64
𝑧𝑖 1 0 0 0
6 2
[ 0]
𝑧𝑖2
𝑧𝑖 1 0 0 0
2
[ 𝑧𝑖 1 0 0 0 0]
𝐂𝐏𝐆 = [0 1 0 0 0 0]; 𝐝𝐆𝐏 = [0]; Equation 65
𝐿2
𝐿 1 0 0 0
0
𝐂𝐏𝐁 = 23 ; 𝐝𝐁𝐏 = [ ];
𝐿 𝐿2 0 Equation 66
[6 𝐿 1 0 0]
2
𝐂𝐏𝐆 𝟎 𝟎 ⋯ 𝟎 𝐝𝐆𝐏
𝐂𝟏 −𝐂𝟏 𝟎 ⋯ 𝟎 𝐝𝟏
𝟎 𝐂𝟐 −𝐂𝟐 ⋱ ⋮ 𝐝𝟐
𝐂= ;𝐝 =
⋮ ⋱ ⋱ ⋱ 𝟎 ⋮ Equation 67
𝟎 ⋯ 𝟎 𝐂𝐧−𝟏 −𝐂𝐧−𝟏 𝐝𝐧−𝟏
[𝟎 ⋯ 𝟎 𝟎 𝐂𝐏𝐁 ] [ 𝐝𝐁𝐏 ]
For the ground level boundary conditions, the rows in the 𝐀𝐆 and 𝐛𝐆 matrices are written as:
𝑤𝑣 0 0 0 0 0 1/𝐸𝐼 𝑤𝑣 𝑣𝐺
𝐆
𝑤𝜃 0 0 𝜆/𝐸𝐼 0 −1/𝐸𝐼 0 𝑤𝜃 𝜃𝐺
𝐀 = diag (𝑤 ) [ ] ; 𝐛𝐆 = diag (𝑤 ) [𝑀 ] Equation 68
𝑀 0 0 0 1 0 0 𝑀 𝐺
𝑤𝑉 0 0 1 0 0 0 𝑤𝑉 𝑉𝐺
where diag(𝐰) denotes a diagonal matrix with the diagonal elements of the vector 𝐰. For each data
point of the inclinometers and the fibre optics, a row in the boundary conditions matrices can be
written as:
𝑤𝜃 𝑧𝐼𝑃𝐼
4 2
𝑧𝐼𝑃𝐼 3
𝑧𝐼𝑃𝐼 2
𝑧𝐼𝑃𝐼
𝐀𝐈𝐏𝐈
𝐢 =− [ −𝜆 − 𝜆𝑧𝐼𝑃𝐼 −𝜆 𝑧𝐼𝑃𝐼 1 0] ;
𝐸𝐼 24 2 6 2 Equation 69
𝐈𝐏𝐈
𝐛𝐢 = 𝑤𝜃 [ 𝜃𝐼𝑃𝐼 ]
𝑧𝑆3 𝑧𝑆2
𝐀𝐒𝐢 = 𝑤𝑀 [ 𝑧𝑆 1 0
𝐒
0] ; 𝐛𝐢 = 𝑤𝑀 [𝑀𝑆 ] Equation 70
6 2
The total matrices are then assembled as:
𝐀𝐈𝐏𝐈
𝐢 𝐛𝐈𝐏𝐈
𝐢
𝐀𝐢 = [ 𝐒 ] ; 𝐛𝐢 = [ 𝐒 ] Equation 71
𝐀𝐢 𝐛𝐢
Page 231/272
𝐀𝐆 𝟎 ⋯ 𝟎 𝐛𝐆
𝐀𝟏 𝟎 ⋯ 𝟎 𝐛𝟏
𝐀= 𝟎 𝐀𝟐 ⋱ ⋮ ; 𝐛 = 𝐛𝟐 Equation 72
⋮ ⋱ ⋱ 𝟎 ⋮
[𝟎 ⋯ 𝟎 𝐀𝐧 ] [𝐛𝟐 ]
F2.8 Weighted least-squares solution to the undetermined parameters
The unknowns are determined by finding the least-squares solution to:
minimise ‖𝐀𝐱 − 𝐛‖22

subject to 𝐂𝐱 = 𝐝 Equation 73
where ‖⋅‖22 denotes the 2-norm of a vector. Least-squares solutions are obtained using the lsqlin
function available in Matlab.
Page 232/272
F3 Simplified backbone curve shape calculation

The field tests were carried out using a series of constant rate of displacement steps with maintained
static loads in between, which means that the load-displacement response is not a smooth curve. To
proceed with the analysis of the field data, it is necessary to define a single smooth response curve
for each test. This is achieved using the following process:
 A linear function is fitted to the initial loading curve (load step 0), where the intercept with the
displacement axis is taken as the offset in displacement at zero load, which is then
subtracted from all displacements for that test.
 Data for the load steps (constant rate of D/300 per minute) are isolated from the test data.
 In a similar approach to that described in Appendix E2, a piecewise cubic spline is fitted to
the isolated load steps, using a least-squares approach, with node locations at vG = 0,
D/1,000, D/500, D/200, D/50 and the maximum displacement in the series.
An example of the resulting field test response fit is shown in Figure 216 for pile CM3, which
demonstrates the accuracy of the fit at small and large displacements. It should be noted that this
fitted curve represents the response of the pile to a load which is applied at a constant displacement
rate of D/300 per minute. It is acknowledged that the displacement rate of full scale loads will vary
significantly and that this is likely to have an effect on the resulting pile response, as described in
(reference to appendix G1). However, this displacement rate is consistent between all field tests and
provides a fair basis for comparison.
(a) Overall response (b) Small displacement response
Figure 216: Fitting of a single response curve to field test data for CM3 (D = 0.762m, L/D = 10)
Page 233/272
F4 Existing API/DNV ‘p-y’ calculation

The API/DNV ‘p-y’ method is adopted to provide a benchmark for the performance of the analysis
procedures presented in this report. The particular form of the method that is adopted for this
purpose is specified below.
The API/DNV approach derives soil reactions based on the guidance specified in DNV-OS-J101
(DNV 2014).
F4.1 API/DNV in clay
For piles in cohesive soil the static ultimate lateral resistance is recommended to be calculated by:
1⁄
𝑝𝑢 𝑣 3
𝑝(𝑣) = { 2 (𝑣𝑢 ) for 𝑣 ≤ 8𝑣𝑢
Equation 74
𝑝𝑢 for 𝑣 > 8𝑣𝑢
where 𝑝𝑢 is the ultimate lateral resistance defined as:
𝑝𝑢 = min{(3𝑠𝑢 + 𝛾 ′ 𝑧)𝐷 + 𝐽𝑠𝑢 𝑧 | 9𝑠𝑢 𝐷}

Equation 75
where J is a dimensionless empirical coefficient. The value of 𝑣𝑢 is determined from:
𝑣𝑢 = 2.5𝜀50 𝐷
Equation 76
where ε50 is the strain that occurs at one-half of the maximum stress in laboratory undrained
compression tests of undisturbed soil samples. The initial slope of the p-y curve, k, is given by:
0.23𝑝𝑢
𝑘=
0.1𝑣𝑢 Equation 77
The model input parameters adopted in this report were selected by an independent analysis of the
site conditions by DONG Energy and are given in Table 30. A comparison of the CPT
measurements and the measured laboratory response indicates a best fit NKT factor of 20 (which
falls on the lower bound of the values reported by Powell and Butcher (2003)) and has been used to
derive the undrained strength profiles shown in Table 31, based on the net cone resistance of the
local CPT tests obtained during the PISA project (see AWG (2015)). Note that in Table 31, the
depth, z, defines the top of a layer of material of the given strength.
Table 30: Input parameters to the API/DNV approach for the Cowden test site
API/DNV parameter Value

3
Effective weight, γ’ (kN/m ) 11.19
J parameter 0.25
Half strain to failure, ε50 1.15%
Page 234/272
The lateral soil reaction curves from the API/DNV approach are applied to the 1D model, described
in Appendix E1, to obtain the overall pile load-displacement response, as well as the depth variation
of the lateral load applied to the pile.
Table 31: Profiles of undrained strength derived by DONG Energy and adopted for the
API/DNV comparison calculations
Calibration Field test

analyses CS1 & CS2 CS3 & CS4 CM2 CM3 CM9 CL2
z su z su z su z su z su z su z su
(m) (kPa) (m) (kPa) (m) (kPa) (m) (kPa) (m) (kPa) (m) (kPa) (m) (kPa)
0 200 0 10 0 5 0 25 0 10 0 10 0 8
2 200 0.3 45 0.3 45 0.5 70 0.2 35 0.3 45 0.2 40
3.5 125 0.8 60 0.8 50 0.75 120 0.8 70 0.9 70 1 107
5 105 1.2 80 1.2 110 1 175 1 120 1.1 100 1.3 152
10 120 1.4 110 1.6 150 2 105 1.3 158 1.5 150 1.6 180
20 160 1.6 170 1.8 190 3 88 1.7 205 1.6 170 2.4 149
40 225 2.5 150 2.1 155 3.5 75 1.9 180 1.8 180 2.8 105
2.9 135 2.4 165 4 80 2.8 130 2.2 170 3.3 85
3.1 110 2.9 140 4.8 65 3 108 2.4 150 4.4 95
3.4 95 3.4 120 5.3 85 3.9 85 2.6 140 5.3 83
3.9 80 3.6 100 6.5 63 4.5 95 2.8 130 6.9 110
4.7 70 4.1 80 7.2 80 5.1 78 3.2 88 7.2 95
5.1 130 5 125 7.6 70 5.5 110 4 80 7.8 83
5.4 100 5.5 110 8.5 85 6.6 77 5.3 100 11.6 111
5.8 125 5.8 105 8.9 60 7.4 100 5.8 120 12 230
6 150 6.2 90 9.3 75 8.7 90 6.1 140 13 95
6.3 93 8.5 85 10 95 10 98 6.3 85 13.7 121
7.6 78 10 100 11.5 125 11 110 7 80 14.2 105
8.3 70 11.6 115 11.7 160 11.6 135 9 75 15 130
9 78 12.1 250 12 900 11.8 175 10 45 16.5 115
9.5 85 12.8 88 12.8 600 12.5 900 10.3 70 17.1 150
10 100 13.1 105 13.5 130 13 500 10.8 57 17.8 130
11.6 115 13.6 145 14 100 13.2 800 11 74 18.2 145
12.1 250 14 90 15 125 13.7 500 11.9 100 19.3 120
12.8 88 14.9 115 16.3 80 13.8 350 12.1 125 20 120
13.1 105 15.3 135 17 80 14 135 12.4 400
13.6 145 15.9 115 14.8 115 13 400
14 90 16.3 85 15.4 140 13.3 85
14.9 115 16.6 65 15.7 120 13.6 110
15.3 135 16.9 65 16.3 80 14.5 95
15.9 115 17 80 15 115
16.3 85 15.4 135
16.6 65 15.9 120
16.9 65 16.3 80
16.6 70
17 70
Page 235/272
F4.2 API/DNV in sand
For piles in cohesionless soils, the static ultimate lateral resistance is recommended to be calculated
using:
(𝐶 𝑧 + 𝐶2 𝐷)𝛾′𝑧 0 < 𝑧 ≤ 𝑧𝑅
𝑝𝑢 = { 1 Equation 78
𝐶3 𝐷𝛾 ′ 𝑧 𝑧 > 𝑧𝑅
where C1, C2 and C3 are empirical coefficients that depend on the soil’s internal angle of friction f
and zR is the transition depth at which the value of (𝐶1 𝑧 + 𝐶2 𝐷)𝛾′𝑧 exceeds 𝐶3 𝐷𝛾 ′ 𝑧.
The lateral load per unit length p acting on the pile at a depth z is then calculated based on Equation
79, in which the initial modulus of subgrade reaction k is empirically prescribed as a function of the
soil’s friction angle f and A is a factor to account for static or cyclic loading conditions, as shown in
Equation 80.
𝑘𝑧
𝑝 = 𝐴𝑝𝑢 tanh ( 𝑣) Equation 79
𝐴𝑝𝑢
0.9 for cyclic loading

𝐴={ 𝑧 Equation 80
(3 − 0.8 ) ≥ 0.9 for static loading
𝐷
Note that for soils in which the effective unit weight varies significantly, such as at the transition
between fully dry and fully saturated material, the local effective unit weight no longer reflects the
true state of the effective stress at that location. For the calculations within this report an ‘effective
depth’ is determined as the depth required to achieve the local effective stress, assuming a uniform
profile of the local effective unit weight above. This effective depth is then used for the calculation of
the lateral load p.
The model input parameters adopted in this report were selected by an independent analysis of
historical site data (Jardine et al., 2000) by DONG Energy and are listed in Table 32. Note that in
Table 32, the depth, z, defines the top of a layer of material of the given unit weight and friction
angle.
The dry and saturated relative density, Dr, have been calculated using Equation 81 and Equation 82,
proposed by Jamiolkowski et al. (2003), and assuming a water table at 4m depth.
Equation 81
Page 236/272
Equation 82
0.72 sinφ'
where σp'=0.33(qt-σvo) (clean sand from Mayne, 2010), Ko=(1-sinφ')·OCR (after Mayne and
Kulhawy, 1982), φ' = 32 and OCR=σ'p/σ'v0.
Based on the calculated relative densities, layer intervals have been established, as shown by the
DL2 example in Figure 217.
Relative density, DR
0 40 80 120
0
Jamiolkowski
Depth, z (m)
6
BS Classification
Design Line
8
10
12
14
Figure 217: Example of relative density layer intervals for the interpreted DL2 CPT
measurements.
When deriving the soil friction angle, the method suggested by Mayne (2010) resulted in values
significantly greater than those presented by Jardine (2000). The approach of Bolton (1986) was
therefore adopted, using Equation 83 and Equation 84.
𝐼𝑅 = 𝐼𝐷 (10 − ln𝑝′ ) − 1 Equation 83
′ ′
𝜙max − 𝜙crit = 3𝐼𝑅 Equation 84
Page 237/272
A critical friction angle φ crit = 31° (recommended by Jardine et al., 2000) was adopted and a mean
’
effective stress at failure of p’ = 150 kPa. A comparison with Anderson (2014) indicates that the
likely consolidation stresses are well matched to the vertical effective stresses expected in-situ.
Mean friction angles were then determined over the layer intervals shown in Figure 217. A maximum
friction angle of 40° was adopted, in line with the API/DNV guidance.
The calibration 3D FE analyses were carried out assuming a profile entirely composed of the
Flandrian sand, seen at depths greater than 3-4m at the Dunkirk test site. The above analysis
indicates that peak friction angles of 40° or more would be expected and so a uniform peak friction
angle of 40° was adopted.
The water table depth of 4m was used as a boundary for the calculation of the subgrade modulus, k,
as a function of friction angle.
Table 32: Profiles of sand unit weight and friction angle adopted for API/DNV calculations
Field test
Calibration analyses
DS1 DM4, DM3 & DM7 DL2
z γ' φ z γ' φ z γ' φ z γ' φ
(m) (kN/m3) (deg) (m) (kN/m3) (deg) (m) (kN/m3) (deg) (m) (kN/m3) (deg)
0 10 40 0 17 38.82 0 17 37.7 0 17 38.28
5 10 40 0.36 17 43.75 0.35 17 43.2 0.31 17 43.32
10 10 40 2.51 17 38.52 2.86 17 39 2.28 17 39.62
15 10 40 3.12 17 36.12 3.46 17 43.4 2.65 17 36.76
20 10 40 3.56 17 42.08 4.62 10 40.1 3.25 17 31.03
25 10 40 4.07 10 40.05 8.36 10 36.5 3.63 17 42.2
30 10 40 8.69 10 42 4.23 10 39.41
35 10 40 10.34 10 39.5 4.92 10 42.18
40 10 40 10.85 10 41.2 5.63 10 40.34
45 10 40 6.35 10 41.11
50 10 40 8.48 10 38.54
55 10 40 8.96 10 42.07
60 10 40 10.79 10 40.13
65 10 40
Page 238/272
PISA Project – Final Report Appendix G - Additional areas of interest and future work
Appendix G Additional areas of interest and future work
G1 Rate effects
The effect of loading rate on the pile response was a serendipitous find from the PISA research
program. The effect was first identified in the Cowden test program, from the initial testing on the
small diameter piles. In Test 1 the loading was applied at a very fast rate, due to problems with the
load control system, with the loading phase taking about 10 seconds, as shown in Figure 218(a).
Test CS2 was conducted as planned with load hold phases, allowing for creep. By comparing the
first two tests, as shown in Figure 218(b), it is clear that substantial rate effects occurred during the
performance of Test CS1, with a factor of more than 2 on the loads obtained at small displacement
and about 2 towards the end of the test. The change is equivalent to a 24% increase in capacity per
log cycle of rate. The tests with the creep phases mimic the time periods over which wind loads
might be applied, and it is clear that considerable displacements occur at higher loads. The fast rate
test mimics a wave load or impulse load.
(a) (b)
Figure 218: Rate effect for small diameter, medium length piles (D = 0.273m, L/D = 5.25)
Given the insight obtained from the small diameter tests a decision was made to explore the effects
of rate for one of the medium size piles. Figure 219 shows the results for CM1, the mid-length mid-
size pile (L/D = 5.25). This pile is fully instrumented with fibre optic strain gauges, extensometers
and inclinometers. This test was originally specified to explore the effect of pile wall thickness,
although subsequently this was felt to be fairly unimportant for the particular wall thicknesses
explored in the testing program. This pile was therefore used to explore effects of rate on pile
response. The test procedure for the first five steps was not changed, so that a comparison could be
made with CM9. Figure 220 shows that CM1 is considerably stiffer than CM9, though the data
plotted represent a sub-sampled set of the total data collected. Following the completion of step 5
the pile was loaded at a fast rate for 23 mm, followed by a slow rate for 23 mm, a fast rate for 23
mm, finishing with a slow rate for the final phase. The transition between each phase of loading was
smooth, so that the test could explore so called “isotach” behaviour as shown in Figure 221. The
rates applied for the different phases are plotted on Figure 219 and Figure 220, and indicate a
difference of a factor of about 300, which was limited by the capability of the loading system. The
result indicates a 21% increase in capacity, equivalent to 8.7% increase in per log-cycle. This
compares with the general understanding of there being a 10-20% increase in clay undrained
strength per log cycle (Dayal and Allen 1975). Further, and more detailed, analysis of these effects
is required, but it is noted this rate effect is not currently provided for in current design calculations.
Page 239/272
However such behaviour can also describe the response during creep in the load-hold phases as
shown in Figure 222.
Figure 219: Rate test completed on CM1
Figure 220: Comparison of CM1 and CM9
Page 240/272
Figure 221: Qualitative understanding of rate effects
Figure 222: Qualitative understanding of creep periods
Page 241/272
It was felt that rate effects at Dunkirk, in the sand soil, would be less important. Nevertheless the
effect was explored both at small scale and at larger scale. A comparison of DS1 and DS2 in Figure
223 show a small increase in capacity, despite a difference in loading rate of a factor of
approximately 400.
Figure 223: Comparison of DS1 (rate ≈ 0.9mm/min) and DS2 (rate ≈ 400mm/min) at Dunkirk
Figure 224 to Figure 226 show the results for the thicker pile DM6, for which a variable rate of
loading was applied. The standard monotonic test sequence was applied for the first 12 mm; the
results are comparable to those obtained from DM4 as shown in Figure 225 and Figure 226. The
pile was then loaded for (i) 5 mm at a slow rate (approximately 0.45 mm/min), (ii) 20 mm at a fast
rate (330 mm/min), (iii) 10 mm at the slow rate (0.45 mm/min), (iv) 20 mm at the fast rate (330
mm/min) and (v) at the standard loading rate (2.5 mm/min) to reach 10% of the diameter. Following
this sequence, the pile was loaded in one-way cyclic loading but this part of the test is not described
in this section. It appears from the test results that there is a rate effect which is equivalent to 4.2%
change per log cycle, however further analysis will be required to properly disaggregate this effect
from the creep phases. It is clear, however, that the advantages of rate-dependency in the sand are
less significant than appears to be the case for clay. Figure 226 shows remarkable repeatability of
the two tests, DM6 and DM4, right up until the change in testing sequence.
Page 242/272
Figure 224: Test of rate effects in DM6
Figure 225: Comparison of DM6 and DM4
Page 243/272
Figure 226: Comparison of DM6 and DM4 at small displacements
Page 244/272
G2 Cyclic loading
In the testing program, a number of piles were subjected to 1 way cyclic loading. The relevant test
piles are listed in Table 33. As for the monotonic loading tests, a similar response is seen between
the two large piles in Dunkirk and in Cowden (Figure 227). Therefore only one of the piles at each
site, CL2 and DL2, are considered in the further analysis presented in this report.
The evaluation of the ground level rotation is based on Section 9.1 in the factual report (AWG 2015),
which involves an optimisation process to fit a model of the stickup pile to the displacement
transducers and stickup inclinometers. For the cyclic tests of CM5, DM2 the evaluation is based on
the results of the displacement transducers only as the stickup inclinometers had large instrument
errors due to the rapid loading. This method of analysis gives accurate results for the accumulation
of rotation throughout the load sets. However, the evaluation of damping and stiffness of individual
loops might be less accurate. Further analyses would be required to assess this and compare the
results with the damping report.
Table 33: Test piles subjected to cyclic loading
Test Pile D t L L/D M/H Cycles

15 CM6 0.762 11 4.00 5.25 10 55
18 CM5 0.762 11 4.00 5.25 10 27650
30 DM2 0.762 14 4.00 5.25 10 38110
DL1 2.00 38 10.50 5.25 10 588
40
DL2 2.00 38 10.50 5.25 10 588
CL1 2.00 25 10.50 5.25 10 1060
43
CL2 2.00 25 10.50 5.25 10 1060
(a) (b)
Figure 227: Comparison of the large cyclic load tests in (a) Cowden and (b) Dunkirk
The analysis of the cyclic tests first addresses the CM5 test pile. The results are then compared to
CM6, where cyclic loading has been applied at holds of the monotonic loading sequence. The
results are then compared to the larger CL2 test pile at the same test site. Following this, the piles in
Dunkirk are analysed.
To analyse the cyclic tests, first the load cycles are identified by finding the local maxima and
minima in the applied load (Figure 228a). For each load cycle, the stiffness 𝑘𝜃 is defined by the
Page 245/272
stiffness between the average of the minima and the maximum (Figure 228b). The accumulated
rotation is calculated as:
∗𝑖 −𝜃 0
𝜃𝑚𝑎𝑥 ∗𝑖 𝑖
AR = ∗1 −𝜃 0 with: 𝜃𝑚𝑎𝑥 = 𝜃𝑚𝑎𝑥 + (𝑀𝐴𝑖𝑚 − 𝑀𝑚𝑎𝑥 )𝑘𝜃 Equation 85
𝜃𝑚𝑎𝑥
(a) (b)
Figure 228: Analysis of the cyclic tests
where 𝜃 0 is the rotation of the pile before the load set. The dissipated energy in a cycle is defined
using the integral of the moment rotation curve where the last point is reconnected to the first point
of the loop: DE = ∫𝐿𝑜𝑜𝑝 𝑀(𝜃) 𝑑𝜃 . The equivalent elastic stored energy for dynamic calculations is
1 average average
calculated as: EE = (𝑀max − 𝑀min )(𝜃max − 𝜃min ). The damping ratio is then defined as:
8
1 DE
𝐷= Equation 86
4 𝜋 EE
Page 246/272
G2.1 CM5 Test analysis

The loading routine for CM5 is given in Table 34, where ζc is the cyclic characteristic parameter
defined by 𝜁𝑐 = 𝑀min ⁄𝑀max (Leblanc et al., 2010).
Table 34: Loading routine for CM5
Load set Max Load (kN) 𝜻𝒄 Cycles Comment

1 10 0 7000 Sinusoidal (+triangles)
2 20 0 7000 Sinusoidal
7 20 0 500 Varying loading frequency
8 40 0.5 1000 Sinusoidal
9 20-30-40 0 1500 Triangular
10 20-30-40-60 0 1000 Triangular
Examples of the loops for load sets 1, 2, 3 and 11 are given in Figure 229.
(a) (b)
(c) (d)
Figure 229: Accumulated rotation for load set (a) 1, (b) 2, (c) 3 and (d) 11 of CM5
Page 247/272
These figures show several effects: change of the hysteresis loop shape with loading amplitude and
with cycle number, and accumulated rotation due to the cyclic loading. The shape change of the
loop shape in load set 3 is illustrated in Figure 230. This figure shows stiffness degradation of the
foundation pile with the cycling.
Figure 230: Change of the loop shape in load set 3 of CM5
Page 248/272
Figure 231 illustrates the change stiffness, damping and accumulated rotation during load sets 1 to 6
and 11 using the approach given in section G2.
(a)
(b)
(c)
(d)
Figure 231: (a) Stiffness, (b) damping, (c) accumulated rotation and (d) zoom of the
accumulated rotation of the load sets for CM5
Page 249/272
Figure 231 shows an approximately constant stiffness and damping throughout each load set. In the
first loops, the accumulation of rotation results in an additional damping, which decreases
throughout the test. The accumulation of rotation increases with load amplitude, however, after a
higher amplitude cyclic loading has been applied, low amplitude cyclic loading does not result in
additional rotation of the foundation.
The average stiffness and damping for load sets 1 to 6 and 11 is plotted in function of the load
amplitude in Figure 232, with a dotted line indicating the sequence of the load sets. The stiffness is
compared with the stiffness of reload step 3 of the CM9 test pile.
(a) (b)
Figure 232: (a) Stiffness and (b) damping at varying loading amplitudes for CM5
Figure 232 shows that the stiffness decreases with load amplitude and that the stiffness degrades
after a higher load amplitude is reached. After load set 3, the stiffness during cycling does not return
to its original value. The stiffness of reload step 3 of CM9 can therefore be compared with the line
between load set 2 and 3. The stiffness of both test piles is in good agreement, confirming the
repeatability of the tests. The damping remains approximately constant across the loading
amplitudes, except at the highest loading amplitude, which could be attributed to the damping effect
of the rotation accumulation.
In load set 7, the loading period has been varied. The results for the stiffness and damping ratio are
given in Figure 233. The results show an approximately constant stiffness and damping across the
frequency range of applied loading.
Page 250/272
(a) (b)
Figure 233: (a) Stiffness and (b) damping at varying loading periods for load step 7 of CM5
In Load set 9 and 10, triangular loading has been applied with varying load amplitude (Figure 234).
The results confirm the earlier statement that the accumulation of rotation is governed by the largest
applied cyclic load amplitude at the foundation. During load stage 9 no accumulated rotation is seen.
In load stage 10, the load cycles of 60 kN result in additional accumulation of rotation.
(a) (b)
Figure 234: Moment-rotation response for load stage (a) 9 and (b) 10 of CM5
Page 251/272
G2.2 CM6 Test analysis
The CM6 test pile has been loaded similarly to the monotonic loading tests, but was subjected to 1-
way cyclic loading at 3 load holds. The loading routine is given in Table 35.
Table 35: Loading routine for CM6

1 43.5 0 9 Sinusoidal
2 70 0 9 Sinusoidal
The moment-rotation response of the pile is illustrated in Figure 235.
Figure 235: Moment-rotation response of CM6
Page 252/272
Figure 236 illustrates the stiffness, damping and accumulated rotation throughout the load sets.
(a)
(b)
(c)
(d)
accumulated rotation for CM6
Page 253/272
The qualitative results are in agreement with the results of CM5. The average stiffness and damping
of the load sets is given in Figure 237. This figure illustrates the stiffness decrease at higher load
amplitudes. The stiffness is also compared with the stiffness during reload step 3. Similarly to CM5,
the damping increases at the high load amplitudes.
(a) (b)
Figure 237: (a) Stiffness and (b) damping at varying loading amplitudes for CM6
G2.3 CL2 Test analysis

The cyclic loading for CL2 is applied after the monotonic loading routine where the test pile has been
loaded to the failure displacement of 0.1D. The loading routine for CL2 is given in Table 36.
Table 36: Loading routine for CL2

Page 254/272
(a)
(b)
(c)
(d)
accumulated rotation of the cyclic sets for CL2
Page 255/272
The average stiffness and damping throughout the load sets is given in Figure 239. The stiffness
can be compared to the stiffness of reload step 3. The comparison indicates the stiffness
degradation after a higher load amplitude has been reached. The qualitative damping results are in
agreement with CM5 and CM6.
(a) (b)
Figure 239: (a) Stiffness and (b) damping at varying loading amplitudes for CL2
G2.4 DM2 Test analysis

The loading routine for DM2 is given in Table 37.
Table 37: Loading routine for DM2

1 10 0 5100 Sinusoidal (+triangular)
8 10-20-40 0 1000 Triangular
9 10-20-40-80 0 1000 Triangular
12 10 0 100 Triangular
16 10-20-40-80 0 1000 Triangular
Page 256/272
Figure 240 illustrates the stiffness, damping and accumulated rotation throughout load sets 1 to 4,
11 and 13.
(a)
(b)
(c)
(d)
accumulated rotation of the cyclic tests for DM2
Page 257/272
In these tests, the results at the lower load levels appear to be affected by an instrumentation error.
The high stiffness and damping in load set 1 (and 2) could be attributed to this error. The qualitative
results on the rotation accumulation in this test in Dunkirk are similar to the results in Cowden:
accumulation of rotation increases with loading amplitude. However, the accumulation is governed
by the highest applied load in the load history. When low load cycling is applied after a higher load
cycle, there is no accumulation of rotation.
The average of the stiffness and damping at the load levels starting from 20kN are given in Figure
241. The stiffness is compared with the stiffness of reload step 3 of test pile DM4 and DM9. Again
there is reasonable agreement confirming the repeatability of the tests.
(a) (b)
Figure 241: (a) Stiffness and (b) damping at varying loading amplitudes for DM2
G2.5 DL2 Test analysis

The loading routine for DL2 is given in Table 38.
Table 38: Loading routine for DL2

Page 258/272
(a)
(b)
(c)
Figure 242: Stiffness, damping and accumulated rotation of the cyclic tests
The accumulation of rotation increases with load amplitude. At the lowest load amplitude a decrease
of the accumulated rotation is seen, which might be attributed to an instrumentation error, which is
less significant at the higher load levels. The average stiffness and damping is given in Figure 243.
Page 259/272
(a) (b)
Figure 243: (a) Stiffness and (b) damping at varying loading amplitudes
The results in Figure 243 show an approximately constant stiffness damping over the loading range.
It should be noted that these results are obtained after the pile has previously been loaded until
failure displacement of 0.1D.
Page 260/272
G3 Lessons learnt from pile testing program?

The on-site preparation and testing of the medium scale piles were carried out in three main stages.
Figure 244 shows the main structure of the project team throughout these stages. There was a clear
delineation between the responsibility of the different parties and a well-defined testing program. The
following records observations made following the completion of the testing, and which might be
useful for similar programs in the future.
Piezo Installation Pile Installation Testing
Site
Management
Contractor
Subcontractor
Academic
Advisors
Figure 244: Field test project team structure
G3.1 Test program design

The most important factor to note is that the quality of the testing campaign is strongly dependent on
the effort spent during the design phase, by both the client and the testing contractor. This is an
iterative process, which will not be successful without clear and frequent communication between
parties throughout. Loading rates or ranges which are not specified during the design phase are
unlikely to be possible or measureable during the actual testing. Once an understanding of the
general test arrangement has been made, an exhaustive list should be made of the types of test that
would be desirable. For each test the range of loading rate and magnitude should be specified,
along with the expected rate and range of measuring instrument response. The contractor should
use this specification to select loading apparatus and instruments that are well suited to the
specification of each test. Where a specification is difficult or expensive to achieve, this should be
highlighted to the client.
G3.2 Test program manual

Once the test program has been specified, it is vital to prepare, in advance, a test program manual
that covers the whole of the site works from (a) responsibilities / reporting, (b) test sequence, (c)
instrumentation, (d) loading rates, (e) data logging and so on. The test manual forms the basis for
work in the absence of further client instruction (although it may be updated as the testing
Page 261/272
proceeds). Dialogue between the parties during the testing ensures that lessons are learnt from the
earlier testing phases.
G3.3 Hydraulics
Large pile testing is likely to be carried out using hydraulic actuators. The specification of the
hydraulics will generally dictate what loading can, and cannot, be achieved. For a given flow rate
and oil pressure, a smaller diameter hydraulic ram will provide a greater displacement rate and lower
friction, but can only achieve a relatively low load. Conversely, a large diameter ram can apply high
loads, but will result in lower displacement rates and greater seal friction. High seal friction is
problematic when trying to achieve very small displacement rates, as the rams will have a tendency
to stick and slip, resulting in an unsteady load and displacement path.
G3.4 Test piles

Sacrificial piles are imperative for calibrating the load controllers prior to an actual test. High quality
testing campaigns are likely to be carried out using advanced load controllers. The selection of the
controller parameters will depend on the stiffness of the pile and the required loading rate. By
providing sacrificial piles of at least two different stiffnesses, the control parameters can be tuned
prior to testing. Note that faults will be encountered during these initial tests and it is recommended
that the time and equipment are available to return the pile to vertical and grout the pile in place for
additional testing.
G3.5 Long term testing

Where instrument response will be measured over a long period of time, such as during cyclic
testing, it is vital that temperature measurements are taken close to the instruments. At low load
levels it is possible that the temperature response of the instrument will be greater than the response
due to the loading, which will result in un-interpretable results without a means of excluding the
temperature effect during the test. It is therefore recommended that the instrument response and
temperature variation are measured prior to testing on a fully assembled test for a period of at least
24 hours. This measurement will allow the variation of instrument response due to temperature to be
excluded from the test results in post-processing. It is also worth noting that instruments will be
exposed to outdoor weather conditions with constant displacement for large periods of time. It is
therefore recommended that additional measures are taken to ensure that instruments remain
sealed. This is especially important on sites with sand or dust, which can result in instrument sticking
and poor measurement.
G3.6 Oversight
It is critical that an agent is available to the test contractor for feedback and validation of the test
procedure. Ideally this agent would operate on site during the setup and first small number of tests
as a minimum. This agent should have the ability to analyse the measured data and provide
feedback within a timescale that allows future testing to be optimised.
G3.7 System validation

Prior to the testing campaign, the test contractor should demonstrate the functionality of the entire
instrumentation and control system, using the maximum amount of anticipated instrumentation
during a test. This may be achieved during the test of a sacrificial pile or as a dummy test. The data
measured from this demonstration should be used by the client’s agent to develop or validate the
analysis tools, which will allow quick feedback of test performance during or closely following a test.
Page 262/272
G4 Ageing
During the project there was considerable discussion about the effect of pile ageing on the response
of the piles at testing. The following tables provide an estimate of the pile ages at testing.
Table 39: Cowden pile installation depths and dates
Pile Vibration Driving Plug height Driving Date Pile age at

Date tested
location depth (m) depth (m) AGL (m) data installed testing (days)
C-S-1 1.43 06/12/2014 16/01/2015 42
C-S-2 1.43 06/12/2014 22/01/2015 47
C-S-3 2.18 06/12/2014 28/01/2015 54
C-S-4 2.73 06/12/2014 20/01/2015 46
C-M-1 1.5 4 * 17/10/2014 05/03/2015 140
C-M-2 1.5 2.29 * 17/10/2014 07/04/2015 162
C-M-3 1 7.62 * 05/12/2014 02/03/2015 88
C-M-5 1.5 4 * 05/12/2014 18/03/2015 104
C-M-6 1.5 4 * 08/12/2014 10/03/2015 93
C-M-7 1.5 4 * 16/10/2014 26/03/2015 162
C-M-8 1 2.29 14/10/2014 16/02/2015 126
C-M-9 1.5 4 17/10/2014 12/02/2015 118
C-L-1 1.15 10.38 0.408 * 09/12/2014 21/07/2015 225
C-L-2 1.15 10.62 * 08/12/2014 21/07/2015 226
C-R-1 1.5 8.3 06/12/2014
Table 40: Dunkirk pile installation depths and dates
Pile Vibration Driving Plug height Driving Date Pile age at

Date tested
location depth (m) depth (m) AGL (m) data installed testing (days)
D-S-1 1.43 0.16 08/11/2014 22/04/2015 165
D-S-2 1.43 0.12 08/11/2014 25/04/2015 168
D-S-3 2.18 0.18 08/11/2014 23/04/2015 166
D-S-4 2.73 0.51 08/11/2014 24/04/2015 167
D-M-1 1.5 4 * 10/11/2014 08/06/2015 211
D-M-2 1.5 4 * 06/11/2014 16/05/2015 192
D-M-3 1.5 6.1 * 06/11/2014 03/06/2015 210
D-M-4 1.5 4 * 10/11/2014 12/05/2015 184
D-M-5 1.5 2.29 10/11/2014 08/05/2015 180
D-M-6 1.5 4 * 10/11/2014 05/06/2015 208
D-M-7 1.5 2.29 * 10/11/2014 14/05/2015 185
D-M-9 1.5 4 10/11/2014 30/04/2015 171
D-L-1 1.5 10.5 * 14/11/2014 20/06/2015 219
D-L-2 1.5 10.5 0.1991 * 13/11/2014 20/06/2015 220
D-R-1 2.4 8.5 14/11/2014
Page 263/272
For the tests at Cowden in the stiff clay, it was felt that the principal ageing mechanism would relate
to pore pressure effects. However, it became apparent, based on measurements of pore pressures
around the pile (these were discussed earlier in Appendix B), that the majority of the excess pore
pressures developed from driving had equilibrated prior to testing. The effect of changes in
undrained strength, due to consolidation, on the pile capacity was determined to be minimal.
For the piles in sand the effects of ageing were less well understood. Consequently, inspired by the
discussions of the AWG, a supplementary testing program was completed by University College
Dublin at the Blessington (Ireland) dense sand test site. The following is a report from UCD on these
tests to provide insight into pile ageing in sand. Note that G4.1 to G4.7 does not form a deliverable
for the PISA project.
G4.1 Overview
The majority of offshore wind turbines are founded on monopiles, which are large diameter, open-
ended steel tubes, usually driven into the sea bed. The piles support the vertical self-weight of the
turbine, and in addition are subjected to significant wind and wave forces which result in the
application of variable cyclic lateral loads and moments. While axially loaded piles typically develop
their resistance through end bearing beneath the pile base and shaft friction developed in a narrow
zone around the pile shaft, the failure mechanism developed by laterally loaded piles extend into the
soil mass and the pile-soil interface response is of less importance. The positive benefits of ageing
on the axial capacity of driven piles in sands have been demonstrated by recent research. Field
studies by Jardine and Standing (2012), Gavin et al. (2013) and Karlsrud et al. (2014) and others
have shown up to a 300% increase in axial capacity over 12 months following driving. Gavin et al.
(2015) note that the increase in capacity is derived from enhanced shaft resistance with capacity
gains depending on the set-up time, the loading history and the degree of damage caused during
pile installation. Whilst research on the mechanisms governing gains in axial shaft resistance is
ongoing dilation during loading of sand in the narrow shear zone developed at the pile-soil interface
and soil creep caused by the relaxation of stresses induced during pile installation appear to be
significant. Since lateral loading mobilises a much larger volume of soil than axial loading, dilation in
the narrow shear zone at the pile-soil interface is unlikely to have a significant effect on the response
of piles. However, changes in the distribution of radial and vertical stress around the pile as a result
of creep may be significant. While research is ongoing to improve our understanding of pile ageing
for axially loaded piles, the authors are not aware of any field tests investigating on the effects of
ageing on laterally loaded piles in sands.
G4.2 EXPERIMENTAL DESCRIPTION

The pile tests discussed in this section were conducted in dense sand at Blessington Quarry in
Ireland. Previous lateral load test programmes at the site were described by Li et al. (2014) and
Murphy et al. (2016). For the ageing tests described herein, a total of fourteen 510mm diameter
open-ended steel piles were driven in the arrangement shown in Figure 245. Ten of the piles were
statically loaded as part of the pile ageing study, see Table 41. The piles had a wall thickness, t of
10mm, with driven penetration depths of 1.5, 2.25 and 3m, which resulted in an L/D range of 3 – 6.
The spacing of the piles was such that there was a minimum spacing of 6 pile diameters between
the test piles, and at least 10 diameters spacing in the line of loading between the test piles and
reaction piles.
Page 264/272
Figure 245: Pile layout and CPT location for Blessington test site
Table 41: Pile test dates for ageing tests

Embedment
Pile Date driven Date tested Age
depth (m)
22/08/2014 7
L4 3 15/08/2014
02/12/2014 109
L5 3 14/08/2014 10/04/2015 239
L6 3 14/08/2014 16/12/2014 124
L7 1.5 14/08/2014 22/08/2014 8
L8 1.5 15/08/2014 15/12/2014 122
L9 1.5 15/08/2014 03/12/2014 110
L10 1.5 11/12/2014 12/12/2014 1
L11 2.25 11/12/2014 08/04/2015 118
L12 2.25 11/12/2014 16/12/2014 5
G4.3 Site Description

Blessington sand is heavily over-consolidated as a result of the glacial history of the region and
recent excavation at the site. Detailed descriptions of the geotechnical properties of the sand have
been reported by Tolooiyan & Gavin (2011), Igoe et al. (2011) and Doherty et al. (2012). Sand
replacement tests suggest the in-situ relative density is close to 100% and the unit weight of the
3
sand is 20kN/m . Triaxial compression and shear box tests on reconstituted Blessington sand
indicate the constant volume friction angle is 37° and the peak friction angle decreases from 54° at 1
m depth to 42° at about 5 m depth. The water table was measured at 2.2m below ground level.
Multiple Cone Penetration Tests were conducted in the vicinity of the test piles, and the averaged
cone tip resistance qc values was approximately 30 MPa at 0.5m depth, reducing to 15 to 20 MPa
Page 265/272
below 1 m depth (see Figure 246). The CPTs are quite uniform across the site for depths larger than
below ground level, however, significant variability is evident over the top 1m.
(a) Cone resistance (b) Sleeve friction
Figure 246: CPT profiles from Blessington test site
G4.4 Test Setup

Lateral loads were applied using the threaded bar system shown in Figure 247. The load was
applied at a height of 1m above ground level using a 100 tonne hydraulic jack connected to a
reaction pile (Pile R in Figure 245) using a specially fabricated load frame. The applied load was
measured with a hollow load cell placed in-line with the threaded bar system. The hydraulic pressure
was controlled manually using a hydraulic power pack. Pile displacements above ground were
measured using three Linear Variable Differential Transformers (LVDTs) placed at 100, 300 and 500
mm above ground level at the front edge of the pile. The pile rotation was measured using four
inclinometers placed at 100 and 450mm above ground level on either side of the pile, which were
riveted onto the pile wall. All measurements were logged on a single Campbell Scientific Cr9000x
data-logger at a logging frequency of 10 Hz. A list of dates for pile driving and load testing is
provided in Table 21.
Page 266/272
Load cell
LVDTs
Inclinometers
Figure 247: Pile load test setup for ageing tests at Blessington
G4.5 Test Results

Load tests performed on piles with an embedded length of 3 m (L/D = 6) are shown in Figure 248.
Pile L4 was first loaded 7 days after driving, however, due to a problem with the loading frame the
test pile was unloaded when the lateral displacement reached ~12mm. The pile was then reloaded
109 days after driving. Piles L5 and L6 were load tested 239 and 124 days after driving respectively.
The three piles appear to show a broadly similar load displacement response, with the initial
stiffness and maximum mobilized resistance being closely comparable. Whilst there is no evident
ageing trend, it should be noted that, due to limitations in the capacity of the loading frame, none of
the piles appeared to reach their ultimate lateral load capacity. At the higher applied loads, creep
during maintained load periods where the load was maintained became significant.
Page 267/272
Figure 248: Load – displacement curves for 3m embedded piles (L/D = 6)
The load-displacement responses for the piles embedded to 2.25 m below ground level (L11 & L12)
are shown in Figure 249. Pile L12, tested 5 days after driving, had an almost identical initial stiffness
to pile L11, tested 118 days after driving. However, pile L11 had an ultimate capacity approximately
15% higher, suggesting a possible increase in ultimate capacity with age.
Figure 249: Load – displacement curves for 2.25m embedded piles (L/D = 4.5)
Figure 250 shows the load-displacement response of the piles embedded 1.5m (L7, L8, L9 and
L10). While there is significantly more scatter in the observed data, pile L10, which was tested 1 day
after driving developed a maximum capacity that was ~20% higher than that measured for piles
which were allowed to age more than 100 days. It is likely that the variability of the upper 1m of soil,
as seen from the CPT profiles, had a more important influence on the results than any ageing
effects.
Page 268/272
Figure 250: Load – displacement curves for 1.5m embedded piles (L/D = 3)
G4.6 Conclusions
This section describes field tests conducted to investigate the effect of pile ageing on the lateral
resistance of driven piles. A total of ten piles, with varying slenderness (L/D) ratios varying from 3 to
6, were tested at different time intervals after driving. The test data suggest that ageing is much less
important for lateral loading than for axially loaded piles in sands. Three piles with a slenderness
ratio L/D = 6 that were tested at different time intervals, all showed broadly similar load-displacement
responses. For the piles with L/D = 4.5, a 15% increase in resistance was noted for a pile load
tested 118 days after driving when compared with a pile loaded 5 days after driving. For piles with
an L/D = 3, a pile tested 1 day after driving developed a significantly higher resistance than piles
load tested more than 100 days after driving. While some variation in pile capacity was noted, there
was no consistent trend for pile capacity to increase with time in the manner noted for axial load
tests performed at the same site. The observed difference in the lateral capacity with time were most
likely due to natural variability of the near surface sand deposit, with CPT q c values ranging by ±
20% of the mean over the first 1 m depth. Axial load tests conducted on 7 m long, 340 mm diameter
piles at the same site indicated a 250% increase in capacity over a time-period of 100 days after
driving. It was noted that the increase in capacity for these axially piles was a result of increased
shear resistance that developed within 3 m of the pile tip. It is noted that the test piles described in
this section had relatively small embedment depths. In order to confirm these trends for offshore
monopoles it is important that tests of this form be repeated with longer penetration lengths.
Page 269/272
PISA Project – Final Report
References
Aghakouchak, A. 2015. Advanced laboratory studies to explore the axial cyclic behaviour of driven
piles. PhD thesis, Imperial College London.
Al-Haj, K. 2014. Mechanical response of two plastic clay soils from Sudan PhD thesis, Imperial
College London.
Andersen, K. H., & Schjetne, K. (2014). Closure to “Database of Friction Angles of Sand and
Consolidation Characteristics of Sand, Silt, and Clay” by Knut H. Andersen and Knut
Schjetne. Journal of Geotechnical and Geoenvironmental Engineering, 140(6), 07014013.
API, 2010. RP 2A-WSD - Recommended Practice for Planning, Designing and Constructing Fixed
Offshore Platforms. Washington: American Petroleum Institute.
Astley, R.J. 1992. Finite Elements in Solids and Structures: An Introduction. Chapman Hall.
AWG, 2013. Detailed Project Plan, Report No. 1656626, PISA AWG.
AWG, 2014a. Draft Design Methodology, Report No. 1868186, PISA AWG.
AWG, 2014b. Draft Sand Design Methodology, Report No. 1868186, PISA AWG.
AWG, 2015. Field Test Factual Report, Report No. 2260925, PISA AWG.
Brinkgreve, R.B. 2013. Validating numerical modelling in Geotechnical Engineering. NAFEMS
Mar, A. 2002. How to undertake finite element based geotechnical analysis. NAFEMS
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., Kellahave, D., & Skov Gretlund, J. (2015). New design methods for large
diameter piles under lateral loading for offshore wind applications. Third International
Symposium on Frontiers in Offshore Geotechics, 1 (pp. 705-710).
Bolton, M.D. (1986). The shear strength and dilatancy of sands. Geotechnique, Vol. 36, No. 1, pp.
65-78.
Chow F.C. (1997), Investigations into the behaviour of displacement piles for offshore foundations.
PhD thesis, Imperial College, University of London.
Cox, W. R., Reese, L. & Grubbs, B. R. 1974. Field testing of laterally loaded piles in sand. Proc.
Offshore Technol. Conf., Houston, TX, paper no. 2080.
Day, R.A. and Potts, D.M. 1994. Zero thickness interface elements - numerical stability and
application. Int. Jnl. Num. Analy. Meth. Geomech. 18: 689-708.
Dayal, U. and Allen, J.H. 1975. The effect of penetration rate on the strength of remoulded clay and
sand samples. Canadian Geotechnical Journal, 12: 336-348
Doherty, P., Kirwan, L., Gavin, K., Igoe, D., Tyrrell, S., Ward, D. & O’Kelly, B. (2012) Soil properties
at the UCD geotechnical research site at Blessington, Proceedings of Bridge and
Concrete Research in Ireland Conference, Dublin, Ireland, 499-504.
DNV, 2014. DNV-OS-J101 - Design of Offshore Wind Turbine Structure. Oslo: Det Norske Veritas.
Davidson, H.L. 1982. Laterally loaded drilled pier research, Vol 1: Design methodology, Vol. 2:
Research documentation. Final Report by GAI Consultants Inc., to Electric Power Research
Institute (EPRI).
Gavin, K.. Igoe, D and Kirwan, L. (2013) The effect of ageing on piles in sand, ICE Journal of
Geotechnical Engineering on the Geotechnical challenges of renewable energy
projects, Vol. 166, Issue 2. April 2013.
Gavin, K., Jardine, R.J., Karlsrud, K. and Lehane, B.M. (2015). The Effects of Pile Ageing on the
Shaft Capacity of Offshore Piles in Sand. Keynote paper. Proc. international
Symposium Frontiers in Offshore Geotechnics (ISFOG), Oslo. June, p 25:
Gerolymos, N., & Gazetas, G. 2006. Development of Winkler model for static and dynamic response
of caisson foundations with soil and interface nonlinearities. Soil Dynamics and Earthquake
Engineering, 26 (5), 363-376.
Page 270/272
Hamre, L., Feizi Khankandi, S., Strøm, P.J. and Athanasiu, C. 2011 Lateral behaviour of large
diameter monopiles at Sherringham Shoal Wind Farm in: Gourvenec & White eds. Frontiers
in Offshore Geotechnics II London: Taylor Francis Group, pp 575-580
Hosseini-Kamal, R. 2012. Experimental study of the geotechnical properties of UK mudrocks PhD
thesis, Imperial College, University of London.
Hutchinson, J.R. 2000. Shear Coefficients for Timoshenko Beam Theory. Journal of Applied
Mechanics 68 1: 87-92.
Igoe, D., Gavin K. and O’Kelly, B. (2011), The shaft capacity of pipe piles in sand, ASCE Journal of
Geotechnical and Geoenvironmental Engineering,(2011) Vol 137, No.10, pp 903-912
Jamiolkowski, M., Lo Presti, D.C.F. and Manassero, M., 2003. Evaluation of relative density and
shear strength of sands from CPT and DMT. Soil behavior and soft ground
construction, 7(119), pp.201-238.
Jardine, R. and Standing, J. (2000). Pile load testing performed for HSE cyclic loading study at
Dunkirk, France. Two Volumes. Offshore Technology Report OTO 2000 007; Health
and Safety Executive, London.
Kallehave, D., LeBlanc Thilsted, C. & Liingaard, M. A. 2012. Modification of the API P-y formulation
of initial stiffness of sand. In: P. Allan, et al. eds. Offshore Site Investigation and
Geotechnics: Proceedings of the 7th International Conference. London: Society for
Underwater Technology, pp. 465-472.
Karlsrud, K., Jensen, T.G, Wensaas Lied, E.K, Nowacki, F, and Simonsen, A.S. (2014), Significant
ageing effects for axially loaded piles in sand and clay verified by new field load tests.
Proceedings of the Offshore Technology Conference, Houston, May 2014, Paper No.
OTC-25197-MS.
Kelly, R.B., Houlsby, G.T. and Byrne, B.W. 2006. A comparison of field and laboratory caisson tests
in sand and clay. Géotechnique 56 9: 617-626.
Kuwano R. (1999), The stiffness and yielding anisotropy of sand. PhD thesis, Imperial College,
University of London.
Lam, I.P.O. and Martin, G.R. 1986. Seismic Design of High-way Bridge Foundations. US
Department of Transportation Report No. FHWA/RD-86/102
Lam, I.P.O. 2013. Diameter effects on p-y curves. Deep Marine Foundations - A Perspective on the
Design and Con-strction of Deep Marine Foundations.
Leblanc, C., Byrne, B.W. and Houlsby, G.T. 2010. Response of stiff piles to random two-way lateral
loading.” Géotechnique 60 9: 715-721.
Lehane B.M. (1992), “Experimental investigations of pile behaviour using instrumented piles”. PhD
Thesis, Imperial College, University of London.
Li, W, Igoe, D and Gavin, K,, 2014. Evaluation of CPT-based P–y models for laterally loaded piles in
siliceous sand (2014), Geotechnique Letters, Vol 4, Issue April, pp 110-117
Martin, C., & Randolph, M. 2006. Upper-bound analysis of lateral pile capacity in cohesive soil., 56
Géotechnique (2), 141-146.
Manzari, M.T. and Dafalias, Y.F., 1997. A critical state two-surface plasticity model for
sands. Geotechnique, 47(2), pp.255-272.
Matlock, H. 1970. Correlations for Design of Laterally Loaded Piles in Soft Clay. Second Annual
Offshore Technology Conference, Houston, Taxes, Vol 1, pp 577-588.
Mayne, P.W. and Kulhawy, F.H., 1982. Ko- OCR Relationships in Soil.Journal of the Soil Mechanics
and Foundations Division, 108(6), pp.851-872.
Mayne, P.W., 2010. Regional report for North America. In Proceedings, 2nd Intl. Symposium on
Cone Penetration Testing (pp. 275-312).
Murff. J.D. and Hamilton J.M. 1993. P-Ultimate for Undrained Analysis of Laterally Loaded Piles.
ASCE Journal of Geotechnical Engineering 119 1: 91-107.
Page 271/272
Gerolymos, N. and Gazetas, G., 2006. Development of Winkler model for static and dynamic
response of caisson foundations with soil and interface nonlinearities. Soil Dynamics and
Earthquake Engineering, 26(5), pp.363-376.
Gerry Murphy, Paul Doherty, David Cadogan and Kenneth Gavin (2016), Field experiments on
instrumented Winged Monopiles, Accepted for Publication in the ICE Journal of
Geotechnical Engineering, December 2015
Papadimitriou, A. G. & Bouckovalas, G. D. 2002. Plasticity model for sand under small and large
cyclic strains: a multiaxial formulation. Soil Dynamics and Earthquake Engineering 22 (3):
191-204.
Potts, D. M. and Zdravkovic, L. 1999. Finite element analysis in geotechnical engineering: theory.
London, Thomas Telford.
Potts, D. M. and Zdravkovic, L. 2001. Finite element analysis in geotechnical engineering:
application. London, Thomas Telford.
Powell, J. J. M. and Butcher, A. P. 2003. Characterisation of a glacial clay till at Cowden,
Humberside. In: T. e. al., ed. Characterisation and Engineering Properties of Natural Soils.
Lisse: Swets & Zeitlinger, pp. 983-1020.
Randolph, M.F., & Houlsby, G. 1984. The limiting pressure on a circular pile loaded laterally in
cohesive soil. Geotechnique, 34 (4), 613-623.
Randolph, M.F., & Gourvenec, S. 2011. Offshore Geotechnical Engineering. (Spon).
Reddy, J.N. 2002. Energy Principles and Variational Methods in Applied Mechanics. Wiley.
Reese, L.C. and Matlock, H., 1956. Nondimensional solutions for laterally loaded piles with soil
modulus assumed proportional to depth. Proc. 8th Texas Conf. on Soil Mech. and Found.
Engng, Austin, TX.
Reese, L., Cox, W. R. & Koop, F. D., 1974. Analysis of laterally loaded piles in sand. Proc. 6th
Offshore Tech. Conf., Houston 2, paper no. 2079.
Robson R.J. 1988. Review of laboratory and field data from the Cowden test site, Holderness. MSc
Thesis, Imperial College, University of London.
Roscoe K.H. & Burland J.B. 1968. On the generalised stress-strain behaviour of ‘wet’ clay. Eng.
Plasticity, Cambridge Univ. Press: 535-609.
Schroeder, F.C., Day, R.A., Potts, D.M. & Addenbrooke, T. I. 2007. An 8-node isoparametric shear
deformable shell element. Int. Jnl of Geomechanics 7 1: 44-52.
Taborda, D. M. G. 2011. Development of constitutive models for application in soil dynamics. PhD
Thesis, Imperial College London.
Tolooiyan, A. & Gavin, K. (2011) Modelling the cone penetration test in sand using cavity expansion
and arbitrary lagrangian eulerian Finite Element Methods, Computers and
Geotechnics, 38(4), 482-490.
Tsiampousi A., Zdravkovic L. & Potts D.M. 2013. A new Hvorslev surface for critical state type
unsaturated and saturated constitutive models; Computers & Geotechnics 48: 156-166.
Ushev E., Jardine R.J. & Liu T. 2015. Interim report on soil testing for PISA project on samples taken
at Cowden site 2014-15, Imperial College London, UK.
Van Eekelen H.A.M. 1980. Isotropic yield surfaces in three dimensions for use in soil mechanics. Int.
Jnl. Num. Anal. Meth. Geomech., 4: 89-101
Winkler E.(1867). Die lehre von elasticitat und festigkeit, (H. Dominic us), Prague, 182-184.
Page 272/272

PISA I - Final Report

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PISA I - Final Report

Uploaded by

Copyright:

Available Formats

This report is Confidential to the PISA Project Industry Partners

PISA Final Report

This report is Confidential to the PISA Project Industry Partners

DONG Energy Document Information:

The PISA Project is a joint industry research project supported by:

(Lead Project Partner)

1 Introduction and project overview .......................................................................................... 7

2 Overview of site selection and characterisation .................................................................. 16

3 Numerical 3D finite element modelling ................................................................................. 24

4 Field testing ............................................................................................................................. 28

5 1D ‘soil reaction curve’ analysis procedure ......................................................................... 39

6 Achievements of the Project and Conclusions .................................................................... 62

Appendix B Analysis of monotonic field test measurements ...................................................... 92

Appendix F Analysis tools.............................................................................................................. 222

Appendix G Additional areas of interest and future work ........................................................... 239

References ....................................................................................................................................... 270

 Quality of fit (%)

‘ Effective quantity (after the subtraction of pore pressure)

1 Introduction and project overview

Figure 1: Overview of project approach and scope of work

1.1 Aim and objectives

1.2 Problem definition and current modelling approaches

H Key dimensionless groups

Embedded length 𝐿/𝐷

1.3 Shortcomings of current p-y approaches applied to wind turbine monopiles

1.4 Overview of project scope

Table 1: Lateral pile design approaches

3D FE Winkler-type Model Surface Spring Model

Finite element Site

Figure 5: Detailed schematic showing project execution

1.5 Project structure

Figure 6: Organogram for PISA project

Overall Project Co-ordination and

Element Testing / Field Testing

Field Testing, Instrumentation, Development of Modelling

Figure 7: Structure of contributions of the PISA Academic Work Group

1.6 Project timeline

The key highlights in the timeline include:

1.7 Meetings and workshops

2 Overview of site selection and characterisation

Figure 9: Location of Cowden test site (Google maps)

PISA test site

Previous BRE test site

Figure 10: Satellite image of Cowden test site (Google Earth)

(a) Ground and ground water (b) K0

Table 2: Stratigraphy at Cowden

Depth Material Description

(a) undrained strength su in triaxial (b) maximum shear modulus G0

Figure 13: Small strain stiffness of Cowden till

Figure 14: CPT profiles from PISA site investigation at Cowden

Figure 15: Location of Dunkirk test site (Google Maps)

PISA test site

Figure 16: Satellite image of Dunkirk test site (Google Earth)

The ground conditions at Dunkirk are summarised in Table 3:

Table 3: Stratigraphy at Dunkirk

Depth Material Description

Figure 17: CPT profiles from PISA site investigation at Dunkirk

3 Numerical 3D finite element modelling

3.1 Model development

Figure 20: Typical FE mesh for the analyses of test piles

Table 4: Geometry of test piles

3.1.2 Boundary conditions

The horizontal load at the pile top, at Z = h, is applied in a displacement-controlled manner, by

3.1.3 Ground conditions