Você está na página 1de 31

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281104304

Deterministic and probabilistic advances in the analysis of


spudcan behaviour

Conference Paper · June 2015


DOI: 10.1201/b18442-11

CITATIONS READS

5 314

5 authors, including:

Mark Cassidy Jinhui Lisa Li


University of Western Australia Harbin Institute of Technology
177 PUBLICATIONS   2,289 CITATIONS    49 PUBLICATIONS   421 CITATIONS   

SEE PROFILE SEE PROFILE

Pan Hu Marco Uzielli


University of Western Australia Georisk Engineering S.r.l.
18 PUBLICATIONS   108 CITATIONS    54 PUBLICATIONS   537 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Pile Reliability JIP View project

Feasibility study on manganese nodules recovery View project

All content following this page was uploaded by Jinhui Lisa Li on 20 August 2015.

The user has requested enhancement of the downloaded file.


Frontiers in Offshore Geotechnics III – Meyer (Ed.)
© 2015 Taylor & Francis Group, London, ISBN: 978-1-138-02848-7

Deterministic and probabilistic advances in the analysis


of spudcan behaviour

M. Cassidy, J. Li & P. Hu
Centre for Offshore Foundation Systems and ARC CoE for Geotechnical Science
and Engineering, University of Western Australia, Perth, Australia

M. Uzielli
Georisk Engineering S.r.l., Florence, Italy

S. Lacasse
Norwegian Geotechnical Institute, Oslo, Norway

ABSTRACT: This paper presents recent advances in the prediction of the installation of jack-up platforms.
New analytical models for spudcan installations in clays, sand-over-clay and stiff-over-soft clays are reviewed
and compared to the existing methods in the recently released ISO 19905-1 standard. The ability of each to
retrospectively simulate databases of field and centrifuge model tests are summarised and conclusions are
drawn on whether or not modifications to the ISO document are warranted. The paper also demonstrates how
deterministic approaches commonly used in jack-up assessment can be transformed into and completed by
probabilistic predictions. Prior predictions of jack-up installations can be improved with the use of model factors
obtained from existing field records or model tests. Curves of jack-up installation pressure versus depth can be
written as probability of exceedance curves, thus providing the engineer with information on how the installation
curve may develop. A Bayesian updating technique is introduced and provides a reliable and consistent framework
to update installation predictions with monitored information. This technique is then applied to predict punch-
through potential in layered soils. Hypothetical examples demonstrate how updated quantitative information can
benefit industry in guiding decisions during offshore jack-up installations. The authors hope that the introduction
of the probabilistic approaches, although less mature than the deterministic ones, will provoke discussion and
suggest a path forward for the jack-up industry.

1 INTRODUCTION

Mobile jack-up platforms remain a key contributor


to the development of offshore oil and gas reservoirs
through their ability to self-install and then drill in
moderate water depths. Modern jack-ups, such as the
Keppel FELS B class, consist of a buoyant triangular
hull, three independent truss-work legs of approxi-
mately 170 m in length and inverted conical spudcan
footings of approximately 20 m in diameter (Figure 1).
More recently, the same jack-up concept has been
extended to purpose-built vessels for the installation
of offshore wind farms.
A jack-up is self-installed by lifting the hull from
the water and pushing the large spudcans into the
seabed. On location, the legs are lowered to the seabed,
where they continue to be jacked until adequate bear-
ing capacity is achieved for the hull to be lifted clear
of the water.
Figure 2 illustrates the installation process schemat-
ically. The spudcan foundations are pre-loaded by Figure 1. Photograph of two jack-ups (courtesy of Keppel
pumping sea-water into ballast tanks in the hull. This Offshore and Marine).

183
exceed the load required for the spudcan to punch-
through. This has been difficult to achieve in practice
as a result of (i) scarce and possibly inadequate site-
investigation data (i.e., significant uncertainty in layer
location and thickness and in soil properties), (ii)
deficiencies in prediction methods (current guidelines
being based on incorrect failure mechanisms), and (iii)
the use of a deterministic model that does not account
for the different uncertainties. A reliability assessment
provides a framework that includes the uncertainties in
the analysis.
The paper has two main objectives: (1) present
recent advances in the prediction of the installation
of jack-up platforms, with focus on whether or not
new analytical models provide more reliable predic-
tions than the existing ISO 19905-1 standard (ISO
2012); and (2) show how the deterministic approaches
commonly used in jack-up assessment can be trans-
formed into and completed by probabilistic predic-
tions. The probabilistic approach is less mature than
Figure 2. Schematic of jack-up installation process (after the deterministic ones, but the authors hope to provoke
Dean 2010). discussion and suggest a way forward.
The paper concentrates on jack-up installations and
specifically where modifications to the ISO docu-
ment may be possible. The authors discuss how the
deterministic and probabilistic advances can benefit
industry, provide useful and quantitative information
and guide decisions made during offshore jack-up
installations.
After a short overview of the recent standards and
guidelines, the statistical analyses and Bayes’ theorem
used in the paper are briefly reviewed. The approaches
are introduced in order of increasingly complexity and
are progressively applied to clay, sand-over-clay and
then stiff clay-over-clay profiles in Sections 5 to 7
respectively. In each section, the current ISO guide-
Figure 3. Definition of punch-through failure. line and possible improvements from recent research
are presented, together with a summary of the predic-
tive capabilities of the guidelines for field or centrifuge
experiments. For the three soil profiles, a probabilistic
proof tests the foundations by exposing them to a
application is illustrated and discussed.
higher vertical load than would be expected during ser-
vice. The ballast tanks are emptied before operations
on the jack-up begin.
The installation and preloading process for jack- 2 STANDARDS AND GUIDELINES
ups is complex, with uncertainties in soil layering,
soil strength, structure, behaviour, the measured load Before a jack-up can operate at a location, a site-
applied and the current installation depth. Even in specific assessment of its installation, operation
relatively uniform soils, accurate predictions of the and subsequent extraction must be completed. Site-
installation and bearing capacity are essential. For specific assessments of jack-up installations typically
instance, in softer clay soils, reassurance that the spud- follow the recently released International Standards
cans will not penetrate past the available leg length Organisation document ISO 19905-1 (ISO 2012).
is required. Installation in layered soils is particularly The document SNAME TR5-5A Revision 2 (SNAME
hazardous, with the potential for a spudcan punching- 2008) was used as the starting point for the draft-
through a locally stronger soil into an underlying ing of the ISO standard. The geotechnical assessment
weaker material (as illustrated in Figure 3). Punch- approach was revised and updated with the results
through failures result in rig damage, such as buckling of industrial and academic research. The guidelines
of the legs and lost drilling time, and may even cause are deterministic and provide no guidance on using
temporary loss of serviceability of a rig with major probabilistic methods.
financial implications. Additional guidance for spudcan installation is
To reduce the occurrence of failure incidents, it is provided in the InSafe JIP (Joint-Industry Project)
critical to ensure that the applied preload does not guidelines (Osborne et al. 2009, 2010): 146 field

184
installations were collated and a subset used to cali-
brate predictive analysis methods. Preliminary InSafe
guidelines on the use of statistical methods were
provided, including:
• the statistical derivation of best-estimate strength
profiles from multiple data sources (see also Bienen
et al. 2010),
• the interpretation of single-layer design methods
with field data, with exceedance percentile curves
to illustrate variability in the predicted values,
• a Monte Carlo simulation method to do installation
predictions (see also Houlsby 2010), and
• guidance on interpreting monitored data against the
statistical results (see also Houlsby 2010).

Even though such probabilistic methods have been


outlined, there is little evidence of their use in practice,
and further development, verification and champi-
oning of their use is required.

3 INTRODUCTION TO PROBABILISTIC
METHODS

3.1 Current approach and InSafe


recommendations
Before a jack-up installation and preload, engineers
prepare a load-penetration curve to assess if the oper-
ations will be safe. It is well acknowledged that there
is considerable uncertainty in this prediction. These
include uncertainties in
• measuring and deriving the properties of the soils
(predominantly, but not limited to, strength);
• locations of layer boundaries;
• variation in properties across the site; and
• the application of the predictive model. Figure 4. Current methodology with soil strength estimates.

Furthermore, there are measurement uncertainties • No modelling (or transformational) uncertainty


(e.g. vertical load, depth) when comparing the mon- within the models is accounted for, and only soil
itored data from the installation against the prior strengths can incorporate variability;
prediction curve. • There is no use of empirical or historical records to
As highlighted in the InSafe JIP (Osborne et al. calibrate uncertainties.
2009, 2010) common practice in installation predic-
tion is to find a “best estimate” of the soil strength Often the best estimate is calculated as the statistical
properties and do a deterministic analysis (e.g. ISO, average of all the different soil characterization tests,
SNAME) to produce a “best estimate” of the load following methods advocated by Lacasse et al. (2007),
penetration curve. This methodology is schematically amongst others, and detailed in DNV (2012). These
shown in Figure 4. are discussed further in Section 4. Lower and upper
Knowing that there are uncertainties in this pre- estimates of strength often reflect judgment. The sig-
diction, engineers use a “lower estimate” and “upper nificant uncertainty in the site investigation data used
estimate” to complement their best prediction (Figure to predict the load-displacement installation of jack-
4b). This is most often produced in an ad-hoc fash- up units was discussed in Bienen et al. (2010), who
ion, usually with lower and upper bound soil strength provide a statistical interpretation of 14 cases of off-
properties (Figure 4a). For complex layering, differ- shore data collated as part of the InSafe Joint Industry
ent analysis methods may be used to produce the Project.
estimates. Problems with this methodology include: Using the same InSafe database, Houlsby (2010)
outlined a Monte-Carlo approach to account for the
• The lower and upper estimates of strength are not variability in soil conditions, geometry, calculation
statistically derived but rather judgment of outliers and observation uncertainties, and suggested proba-
of soil data; bilistic percentile curves of the installation load and

185
provide a qualitative assessment, and often only with
advice on what could improve the predictive models
retrospectively. The initial predictions are not updated
with the new information from the monitored data. To
do so would be a significant improvement, providing
more reliability for jack-up installation.

3.2 Use of Bayes’ theorem to update predictions


Bayes’ Theorem provides a theoretical framework to
allow updating of the load-penetration prediction with
monitored data. This can be done in real-time or in
retrospective to improve the predictive models. Quan-
titative probability of punch-through occurrence, for
instance, could be continuously updated. Potentially,
Figure 5. Probability of exceedance plots (method recom- the results of updating could suggest a riskier project
mended in InSafe JIP and discussed in Houlsby 2010).
than expected and warrant halting operations.
Table 1. Observed response and suggested action (a sub- Bayes’ theorem relates current probability to prior
group of the full table after Osborne et al. 2010 and Houlsby probability with the help of new (or additional) infor-
2010). mation. Bayes’ theorem is the essential means of
changing one’s opinion in the light of new evidence.
Mathematically it is expressed as:

It converts the prior assessment P(A) into the so-


called posterior assessment P(A|B), where A is the
prior (i.e. the predicted bearing pressure) and B is the
observation (i.e. the monitored data). P(A) and P(B)
are the probabilities of occurrence of A and B, and
P(A|B), called the conditional probability, is the prob-
ability of A given that B is true. It is natural that one
should be prepared and willing to alter one’s opinion
in the light of new facts, and Bayes’ theorem tells us
how to do it.
In the paper, Bayesian updating will be used to
investigate whether a peak punch-through pressure
will occur (qpeak , Figure 3) before the installation
preload pressure is reached (qpreload ). In this paper,
displacement. The engineer estimates the uncertain- a very loose definition of punch-through is used: it
ties and probabilistic distribution of all the parameters solely defines if a peak capacity will form, followed
and factors in the predictive model and does a Monte- by a reduction in bearing pressure, but it gives no
Carlo simulation of the installation curves. These can judgment of the consequence post-peak. However, a
be sorted into exceedance percentile curves relative to formulation to probabilistically define the likelihood
the mean, as illustrated in Figure 5. These percentile of qpeak occurring and the depth dpeak where peak
curves reflect the uncertainty in the predictions. bearing pressure occurs (see Figure 3) are derived.
This method is included in the InSafe guidelines One can illustrate the prediction of the punch-
(2010). No guidance on appropriate assumptions of through pressure at depth dpeak before installation,
uncertainty is given. Model uncertainty could be with its uncertainty, with a series of two-dimensional
included but determining the bias and standard devia- contours. The prior prediction, done before installa-
tion is difficult. tion, is illustrated in Figure 6. The probability that
The InSafe JIP and Houlsby (2010) do introduce a punch-through will occur before preload is reached
rationale for comparing monitored data recorded dur- can be evaluated by summing all the cases where
ing the installation process against the probabilistic qpeak ≤ qpreload (the portion left of qpreload in Figure 6).
curves. It is argued that this could guide decisions Figure 6 illustrates the updating of the prior
when a monitored event diverges from the original prediction with the help of monitoring data. To
deterministic and/or probabilistic predictions. Cate- achieve this, a probabilistic model describing the load-
gories of response to monitored behavior were pro- displacement paths leading to a punch-through event
vided, with a subgroup shown in Table 1. These only is needed.

186
bearing pressure vs depth curve for the stiff-over-soft
clay case; Bayesian updating of the bearing pres-
sure vs depth curve with monitored data for a series
of hypothetical examples; and the calculation of the
probability of exceeding a target preload.

4 INTERPRETING SOIL DATA


STATISTICALLY

Statistical methods used in the interpretation of site


investigation data are discussed in this section. Only a
few aspects are presented. A more complete treatment
Figure 6. Prior probability predictions of punch-through. can be found in e.g. Baecher and Christian (2003) and
Ang and Tang (1975, 1984, 2007).

4.1 Determining layer boundaries


Locations of layer boundaries are often identified by
eye, with engineers using their judgment and experi-
ence. This process, although appropriate, usually will
introduce additional uncertainties. Methods to identify
layer boundaries using statistical methods help reduce
the uncertainty. Examples include:
• Lacasse and Nadim (1996) proposed a geostatistical
analysis with three-dimensional graphical represen-
tation of cone resistance. This allows the engineer
to adjust the originally assumed position of the clay
layers.
• Houlsby and Houlsby (2013) developed an unbiased
fitting of strength profiles to undrained strength
data using Bayesian inference techniques. The nov-
elty of this approach is that, instead of determining
boundaries between soil layers a priori, the approach
automatically deduced both the number of layers
and the position of the boundaries. Further, in setting
the prior it is possible to incorporate expert knowl-
edge and information derived from other datasets
into the determination of soil layers and proper-
ties. An example is provided in Figure 8. For the
same undrained strength with depth data, a two-
layer and four-layer profile were fit to the data and
Figure 7. Bayesian updating of punch-through from moni- updated based on an assumed prior. The statistical
toring. method also allows the calculation of a posterior
probability of the most probable number of layers
• In Figure 7a, the monitored data lie above the a (Fig. 8c). This can be extremely useful to increase
priori predicted pressure-displacement curve and confidence in the layering. For the example shown
suggest higher bearing pressure and shallower depth in Figure 8, the most likely number of layers was
than initially predicted. The probability of punch- two, followed closely by three. Fewer and more than
through related to qpreload is lower than in Figure 6. two or three layers was unlikely. The difference in
• In Figure 7b, the monitored data show higher pen-
the soil profiles between two and four is shown in
etrations at lower bearing pressure, the updated Figure 8.
contours of punch-through probability will give
lower pressure and larger penetrations than initially 4.2 Fitting of the undrained shear strength data
predicted. The probability of punch-through related
Uncertainties arise when the geotechnical data (e.g.
to qpreload is significantly higher than in Figure 6.
undrained shear strength) show significant scatter.
In this paper, the probabilistic approach is used Although the distributions can be described to follow a
to study several aspects: probabilistic representa- probability distribution function, it is more usual to use
tion of the bearing pressure vs depth curve for the “best estimates” and “lower” and “upper estimates”
sand-over-clay case; probabilistic representation of the for jack-up assessments, as illustrated in Figure 4.

187
• the mean value minus two standard deviations;
• or the most probable value (modal).
Det Norske Veritas AS (DNV 2012) prepared a
detailed treatment of the statistical representation of
soil data. In particular, DNV (2012) proposed a method
to estimate parameters for dependent soil variables,
such as an undrained shear strength (su ) that linearly
increasing with soil depth (z):

with su0 being the strength intercept at the soil sur-


face and k the shear strength gradient with depth from
the soil surface. ε represents the natural variability
of su about the mean. The parameters su0 and k are
evaluated as:

where sui are the observations of undrained shear


strength at depth of zi ; z̄ and s̄u are the mean values of
depth and undrained shear strength, respectively.
When the characteristic value is estimated with a
confidence interval, DNV (2012) described the calcu-
lation of the lower and upper bound of a characteristic
value (i.e. the values at 5% quantile). If the character-
istic value is defined as the mean, the range of values
in which the true but unknown mean value lies with
confidence of 1-α is:

where x̄ and s are the estimated mean and standard


deviation, respectively. tn−1 (α/2) is the value of the t-
distributed variant at the probability of (1 − α) with
(n − 1) degrees of freedom.
For jack-up design, the InSafe JIP and Bienen
et al. (2010) looked into the statistical description of
the undrained shear strength as determined from UU
(unconsolidated undrained) tests, a number of strength
index tests (à la pocket penetrometer, miniature vane,
torvane and motor vane) and the piezocone tests. The
variation in the measured data derived for each test-
ing method (and expressed as CoVs) is provided in
Table 2. More generally, Lacasse et al. (2013) sug-
gested a method to do the statistical reduction of cone
Figure 8. Predicted soil layering using (a) Two-layer model; penetration data to help determine the soil layering.
(b) Four-layer model; and (c) Marginal posterior over the The approach includes data screening, data sorting and
number of layers (after Houlsby & Houlsby 2013). data filtering before obtaining the mean and standard
deviation of the cone resistance.
In geotechnical design, the characteristic values for
design are often assumed as either: 4.3 Spatial variability of a soil property
• the mean value; The estimations above are for one location without
• a low quantile in the probability distribution; considering the correlation between adjacent locations

188
Table 2. Coefficient of Variation (CoV) of undrained shear Li et al. (2015) present at this conference an application
strength derived from InSafe database (after Bienen et al. example with cone penetration tests at the Laminaria
2010). site, offshore Western Australia, with an interpolation
and reduction in variance from a limited number of
Mean of CoV Range of CoV
CPT and T-bar tests.
Testing method (%) (%)
Kriging is useful to estimate and interpolate soil
UU 20 3–37 properties (e.g. soil strength or depth to a weak layer)
Minature Vane 20 0–48 across an entire site. The kriging approach can help
Moter Vane 12 3–18 estimate the soil properties for the prediction of the
Torvane 13 2–31 installation prediction bearing curve at locations where
Pocket Penetrometer 19 4–44 there is little data, for example when a jack-up needs
CPT 23 2–41 to be relocated.
CoV of measured undrained shear strength data within a layer
assumed to be homogenous 4.4 Selection of soil parameters for design
Lacasse et al. (2007) and DNV (2012) recommended
a principle for the definition of characteristic value.
across the field. A critical statistical parameter that is For problems that are governed by a local soil
needed to describe inherent variability is the scale of strength value (and where a conservative prediction
fluctuation (ISO-2394 2014). is required), a value representative of the low quantile
Soils are by nature rarely homogeneous and the in the probability density function is often required.
properties may be strongly spatially correlated, in For problems that involve large soil volumes the mean
either or both the horizontal and vertical direction. One value of the soil strength is often specified as the
of the sources of uncertainties in soil properties is their characteristic value. This is because the local strength
inherent spatial variability. variation can be assumed to be averaged during shear-
The spatial averaging is a concept with which spa- ing. The confidence bands chosen (say as lower and
tial variability of the soil property is averaged to upper estimates, Figure 4) should also reflect the
approximate a random variable that represents a soil uncertainty and severity of consequences. The authors
parameter (Vanmarcke 1977). A variance reduction believe that it is better to use statistical methods to cal-
factor is derived in terms of a scale of fluctuation and culate these rather than to use ad hoc judgement based
an averaging distance. on an observation of the scatter in the data set.
The scale of fluctuation, describing the random Consistent with this logic, the InSafe JIP recom-
field, defines the distances over which there is a signif- mended using upper and lower estimates undrained
icant correlation of the geotechnical parameter, while shear strength at ±1 times the standard deviation about
the averaging distance is defined as the distance over the mean for jack-up installation predictions. This fol-
which a geotechnical property is averaged. lows the recommendation of Lacasse et al. (2007) to
The estimation of the scale of fluctuation can be use a ±0.5 standard deviation for designs that require
found in, among others, Vanmarcke (1977, 1983), resistance over large volumes of soil (as a spudcan
Phoon and Kulhawy (1999) and Fenton and Griffith does), and ±1 to 2 standard deviations for designs
(2008). Li et al. (2014) presented values for the scale based on less reliable test data.
of fluctuation at the Ballina national field test facility If probability distributions are desired, normal dis-
in Australia. tribution and lognormal distribution models are often
used. The normal distribution is typically used for vari-
4.3.1 Interpolation by Kriging ables derived as sums or averages of a large number of
An exact picture of the variation of a property within independent quantities (e.g. strength). The lognormal
a soil volume can not be obtained because there will distribution is used for variables which come about
never be enough subsurface data, but it is possible with as non-negative products of a large number of inde-
spatial variability approaches to obtain the best unbi- pendent quantities (DNV 2012). Lacasse and Nadim
ased estimate of the properties by interpolation (see (1996) evaluated the probability distribution (PDF)
also Lacasse and Nadim 1996). A stochastic interpo- and the coefficient of variation (CoV) for several soil
lation technique that is well suited for soils is Kriging, properties evaluated from different tests (see Table 3).
which is also recommended by DNV (2012).
Kriging is a method of interpolation for which the
interpolated values are modeled by a Gaussian pro-
5 JACK-UPS IN CLAYS
cess governed by covariances. Kriging gives the best
linear unbiased prediction of the interpolated values.
5.1 Current ISO recommendations
The basic idea of Kriging is to predict the value of
a function at a given point by computing a weighted The calculation of the ultimate vertical bearing capac-
average of the known values of the function in the ity of a spudcan foundation in uniform clay soil is influ-
neighborhood of the point. enced by (i) the geometry of the foundation; (ii) the
Some of the first geotechnical applications were strength parameters of the soil, including its undrained
presented by Nadim (1988) and Keaveny et al. (1990). shear strength and its sensitivity to remoulding; (iii) the

189
Table 3. Probability Density Function (PDF) and Coeffi- the Skempton Nc factors do not account for an increase
cient of Variation (CoV) of soil properties (after Lacasse & in strength with increasing depth). Further discussion
Nadim 1996). of the implementation of the Skempton method in the
ISO guidelines is provided by Wong et al. (2012).
Soil property Soil type PDF CoV
An alternative set of bearing capacity factors
that directly account for the axisymmetric geome-
Cone resistance Sand LN —
Clay N/LN try, embedment depth (d), cone apex angle (β), cone
Undrained shear Clay (triax) LN 5–20% roughness (α), and increasing the undrained shear
strength, su Clay (index su ) LN 10–35% strength with depth (k) inherent in the spudcan pene-
Clayey silt N 10–30% tration problem are also provided inAppendix E of ISO

Ratio su /σv0 Clay N/LN 5–15% (2012). These factors are tabulated in ISO (2012) and
Plastic limit Clay N 3–20% were calculated using the lower-bound method of char-
Liquid limit Clay N 3–20% acteristics (the axisymmetric slip-line method with the
Submerged unit All soils N 0–10% Tresca yield criterion), as reported by Martin (1994)
weight
and Houlsby & Martin (2003). The latter also provided
Friction angle Sand N 2–5%
Void ratio, porosity, All soils N 7–30% the following simple formula based on a curve fit of
initial void ratio four key parameters, the cone apex angle (solutions
OCR Clay N/LN 10–35% for 30◦ ≤ β ≤ 180◦ ), the cone roughness (α = 0 to 1),
the embedment ratio (d/D = 0 to 2.5) and the dimen-
N/LN: Normal and lognormal distribution sionless rate of strength increase with increasing depth
(kD/sum = 0 to 5, where sum is the mudline undrained
shear strength and k the increase of undrained shear
selected set of bearing capacity factors; and (iv) the strength with depth):
drainage conditions of the soil depending on the rela-
tive velocity of the footing penetration. For most clay
soils, spudcan preloading will be done fast enough
to ensure fully undrained conditions. For clays with
coefficient of consolidation (cv ) less than 100 m2 /yr,
the normalised installation velocities of vD/cv (where
v is the spudcan penetration velocity and D is spudcan
diameter) tend to be greater than 10, thereby ensuring with
undrained conditions. However, rate effects are also
possible (Hossain & Randolph 2009a , Randolph &
Gourvenec 2011, Cassidy 2012).
The primary method of calculating leg penetra-
tion in clay in ISO (2012) follows classical bearing
capacity theory, with the spudcan idealised as a flat
circular foundation of diameter D embedded at a depth
d below the mudline. A load-penetration curve is pro-
gressively constructed from independent calculations
of wished-in-place footings at increasing depth. The
recommended bearing capacity Nc factors of ISO are
based on solutions for a strip footing on homogeneous
clay, with shape and depth factors based on Skemp-
ton (1951). The gross vertical bearing capacity, QV , is
calculated as:
In this paper, Equations 7 and 8 (rather than the
tables) are used for the ISO Houlsby-Martin calcu-
lations. These ISO methods assume failure based on
where su is the design undrained shear strength; Nc sc dc wished-in-place footings, and adjustments must be
is the bearing capacity factor for a circular footing, made to account for the change in mechanisms with
accounting for shape and depth (with a recommended depth.
value of 6 (1.0 + 0.2d/D) ≤ 9.0); qo is the effective Perhaps the most significant advance in ISO (2012)
overburden stress at depth d; and A is the equiva- was the inclusion of a rational method for predicting
lent cross-sectional area of the spudcan. Engineers are the depth at which a flow-round failure mechanism
allowed considerable freedom to estimate the value becomes preferential, leading to backfilling of the cav-
and representative depth of an equivalent su . Scrutiny ity above a penetrating spudcan. This method is based
of field installations in Gulf of Mexico clays have on the thorough centrifuge and large-deformation
yielded recommendations to use a strength averaged finite element investigations of Hossain et al. (2003,
over the zone of soil from the spudcan tip to a depth of 2005, 2006, 2009a ), which yielded the observation
half a diameter (partly to compensate for the fact that (and, ultimately, the encapsulation in formulae) of

190
three distinct mechanisms of soil flow around an mechanism at depth. Equation 9 can be used to predict
advancing spudcan. Initially, there is an outward and the transition to back flow, although several different
upward flow, leading to surface heave and the forma- interpretations regarding the depth at which a full flow
tion of a cavity above the spudcan; then, there is a mechanism arises have been proposed. This is stated
gradual flow back into the cavity; and finally, a fully to occur at 0.7D to 1.0D by Hossain et al. (2006),
localised flow forms around the embedded spudcan, although the recent publication of Hossain et al. (2014)
leaving the cavity unchanged. This localised mecha- recommends that for simplicity, the fully localised
nism is the result of a preferential flow failure and not mechanism should be assumed immediately follow-
a wall failure caused by the instability of the open ing the onset of backflow. The latter is assumed in
cavity (as was previously assumed in the SNAME the calculations in this paper. When the full flow-
guidelines). In accordance to the findings of Hossain around mechanism is assumed, no overburden stress
et al. the ISO guidelines define the conditions for back is incorporated. The bearing capacity equation is then:
flow as

The formulation of the Nc factors when d ≤ hcavity


for smooth spudcans (α = 0) is
where hcavity is the limiting cavity depth (also the depth
at the onset of backflow), γ  is the submerged unit
weight and suh is the undrained shear strength at that
depth.
With a formula to predict the onset of backflow
and the limiting cavity depth, more accurate estimation
of the surcharge component of the capacity can be
achieved, as well as better prediction of the increase in and that for rough spudcans (α = 1) is:
the bearing capacity factor that occurs before the fully
localised flow-round mechanism is generated (as the
failure mechanism passes through the backfilled soil).
This transition has been found to occur at a penetration
of approximately 0.7D to 1.0D beyond the onset of
backflow (Hossain et al. 2006).
Therefore, for shallow penetration of less than the
limiting cavity depth (d ≤ hcavity ),
Once a full flow-around mechanism has mobilised,
the recommended Nc formulation, under the assump-
tion that this occurs immediately at d > hcavity , is:
where Vb is the (partial) volume of the spudcan
that is submerged in soil, whereas after back flow
(d > hcavity ),

where Ncd simply implies the bearing capacity of a


deep flow-around mechanism.
where V is now the full volume of the spudcan, as
it is entirely submerged. In Equations 10 and 11, the 5.2.1 Accounting for strain-rate dependency and
single Nc value represents the best estimate of the bear- strain-softening
ing capacity factor (and includes the sc dc factor of the Hossain & Randolph (2009a ) extended their method
Skempton formulation). to account for strain-rate dependency and strain-
softening as the soils are sheared and remoulded as a
5.2 Recent advances in prediction methods result of the large displacements during spudcan pene-
tration (which were shown to be important by Erbrich
Following their improved understanding of the evolu-
2005, Osborne 2007, Hossain & Randolph 2009a ).
tion of failure mechanisms for a penetrating spudcan,
Following a similar equation for reduction in shear
Hossain et al. (2006, 2014) and Hossain & Randolph
strength with shearing formulated by Einav & Ran-
(2009a ) suggested alternative bearing capacity formu-
dolph (2005), Hossain & Randolph (2009a ) suggested
lations that consider the evolution of the soil failure
that the Nc factors of Equations 12 to 15 be adjusted
mechanism at different stages of penetration. They
via multiplication by:
included the three distinct mechanisms of soil flow
around an advancing spudcan mentioned above: a gen-
eral shear before the onset of backflow, a transitional
failure mechanism and a fully localised flow-around

191
where µ is a rate parameter (assumed to be 0.1, repre-
senting a 10% increase in load per log cycle of rate),
Rb is a rate coefficient that is used to adjust for the
strain-rate dependency, St is the sensitivity of the soil,
and ξ95 is the absolute shear strain required for the soil
to undergo 95% of full remoulding.
Hossain & Randolph (2009a ) suggested Rb values
of 1.47 for rapid penetration (initial stages, in which
the spudcan velocity is assumed to be ∼2 m/s) and
0.77 for slow penetration (final stages, in which the
velocity is assumed to be ∼0.36 m/s). An Rb value of
zero implies no rate dependency, with the formulation
reverting to one in which only softening is considered.
The importance of the term Rb , which effectively acts
as a multiplier on the capacity (1.147 for fast and 1.077
for slow), is discussed in Sections 5.3 and 5.4, where Figure 9. Comparison of the performance of single-layer
comparisons with field data are presented. clay analysis methods for Site 1 of Menzies and Roper (2008)
In the Hossain & Randolph method, the foot- (a similar analysis is presented in Hossain et al. 2014)
ing roughness, clay sensitivity and ductility must be
specified. In the absence of detailed information, a ξ95 = 15, St = 2.85) and HRSslow (Rb = 0.77, ξ95 = 15,
roughness coefficient of 0.5 and a ξ95 value of 15 are St = 2.85) take rate effects into account.
recommended. Shown in Figure 9 are retrospective simulations
for Site 1 of Menzies & Roper (2008). A similar
plot was provided by Hossain et al. (2014). How-
5.3 Model performance against field data ever, additional simulations of the rate-dependency
Various databases of field installations in clay sites and softening parameters are provided in Figure 9.
have been collated and published, including 14 sites The HMA method yields the lowest load per pen-
from the InSafe JIP (∼100 monitored points) and 16 etration depth; the HRN method yields the highest
sites in the Gulf of Mexico, as published by Menzies & values, and the SKE method lies in between. This
Roper (2008) and Hossain et al. (2014). The latter behaviour was described by Menzies & Roper (2008)
database will be used to compare the two ISO methods and Hossain et al. (2014) as providing good “lower-
and the proposed Hossain & Randolph methodology bound”, “upper-bound” and “average” or “best” esti-
(an assessment against the InSafe database was also mates of spudcan penetration from the HMA, HRN
provided in Osborne et al. 2009 and will be discussed and SKE methods, respectively. As will be demon-
later in this paper in Section 5.5). A retrospective strated in Section 5.4, this loose terminology does not
prediction for one site using deterministic methods necessarily reflect the entire database. For this site, the
is initially discussed to allow for a qualitative com- HRS method reduces the HRN prediction (acknowl-
parison of the methods. This is followed in the next edged to be ∼20% too high by Menzies & Roper
section by a statistical and probabilistic interpretation, 2008 and Hossain et al. 2014), bringing it closer to
including a proposed method for forward probabilistic the lower-load predictions of the HMA analysis. The
assessments. rate-dependent HRSfast and HRSslow methods modify
Previous assessments of this database have been the reduction due to softening and appear well suited
conducted by Menzies & Roper (2008) and Hossain to this site (though there is an element of curve fit-
et al. (2014). The soils are predominantly nor- ting, with the magnitude of Rb being the best fit to
mally consolidated or overconsolidated clays with an large-deformation finite element data).
undrained shear strength that increases with depth
(Hossain et al. 2014). In this paper, the recommended 5.4 Use of probabilistic methods, jack-up in clay
profiles of Hossain et al. are used. This section details the procedure for and results
Field measurements are compared with the values of the statistical and probabilistic characterisation of
predicted by the four methods outlined in Sections the model factor for the vertical bearing capacity of
5.1 and 5.2: (a) the Skempton (1951) ISO method, spudcans in single-clay-layer stratigraphies. The study
hereinafter denoted by SKE; (b) the ISO method of relies on a total of 318 field measurements from 16
Houlsby & Martin (2003), denoted by HMA; (c) the sites in the Gulf of Mexico (Menzies & Roper 2008,
method of Hossain and Randolph, denoted by HRN; Hossain et al. 2014). The data are considered to be
and (d) the subsequent development of the HRN sufficiently homogeneous from a geotechnical point
method presented by Hossain & Randolph (2009a ), of view to be addressed as a single population for
denoted by HRS, which considers the effects of strain- statistical and probabilistic purposes. The model factor
rate on shear strength and gradual remoulding. Several was defined as:
variants of this last method are considered: HRS
addresses only strain-softening (no rate effects, Rb =
0, ξ95 = 15, St = 2.85), whereas HRSfast (Rb = 1.47,

192
Table 4. Sample statistics of the model factors.

Model Mean σ CoV (%)

SKE 0.9 0.07 8.0


HMA 1.01 0.08 8.0
HRN 0.83 0.07 9.0
HRS 1.08 0.09 8.0

Figure 10. Comparison of the performance of four sin-


gle-layer clay analysis methods with 318 field measurements Figure 11. Frequency histograms for the model factor for
(Menzies & Roper (2008) and Hossain et al. (2014). (a) all methods and (b) fitted lognormal distribution for the
HMA model factor.
where Qmeasured and Qcalculated are the measured (field)
and model-predicted values of the vertical bear- For reliability analysis using Monte Carlo simula-
ing capacity. The model factor included both epis- tion, the model factor should be characterised prob-
temic uncertainties (i.e., measurement uncertainty) abilistically. Visual inspection suggests a moderate
and uncertainties in the calculation model, namely, increase in scatter with increasing Q. The lognormal
parameter uncertainties (i.e., uncertainties in the input probability density function distribution was selected
parameters to a calculation model) and transforma- to ensure that no values could become negative. Fig-
tion uncertainties (e.g. converting the cone resistance ure 11 shows the frequency histograms of the model
to undrained shear strength). factors for the four models and the fitted lognormal
Figure 10 presents scatter plots of Qmeasured and distribution for the HMA model. The goodness of fit
Qcalculated for each method (note that only the non- was checked with the Anderson-Darling test. The sam-
rate dependent HRS variant was analysed). Samples ples for the SKE, HMA and HRN methods were found
of 318 model factors were obtained using Equation to have a lognormal (LN) distribution at the 95% con-
17. Figure 10a presents all data, whereas Figure 10b is fidence level. The HRS method showed slightly lower
only the data for Q < 100 MN. Table 4 gives the statis- confidence level for a LN distribution.
tics (mean, standard deviation (σ ) and coefficient of Based on the results of this analysis, the ISO
variation (CoV)) of the model factors. method of Houlsby and Martin (2003) was adopted to
The HMA method has the smallest bias, as its sam- probabilistically calculate the vertical bearing capaci-
ple mean is closest to unity. The relative scatter of ties of spudcans in single-layer clay stratigraphy. The
sample values around the mean is given by the CoV. recommendation of this method applies solely to soil-
The four models are comparable in terms of scatter, spudcan systems that are geotechnically similar to
with a moderate dispersion of 8 or 9% around the those in the dataset used for the study. Further updating
mean. and generalisation of this performance of the methods

193
Figure 13. Cumulative distribution of the ratio of measured
to predicted (HMA) resistances (after Osborne et al. 2009).
Figure 12. Probabilistic predictions using the ISO
Houlsby & Martin method and the random LN model factor.
Table 5. Measured/predicted load ratios for spudcans in
clay.

can be done with the 14 single-layer clay field cases Qfield /Qcalculated
collated in the InSafe JIP (Osborne et al. 2009, 2010).
If larger, more representative databases could be accu- Percentile InSafe1 This paper2
mulated by the industry after each installation, simple
probabilistic methods can be easily developed. 5% 0.64 0.884
25% 0.92 0.955
For the purpose of a probabilistic analysis, vertical 50% 1.07 1.007
bearing capacity Q can be expressed as: 75% 1.18 1.062
95% 1.42 1.147

1. Factors from InSafe database (Osborne et al. 2009)


2. Model factors calculated using the method in Section 5.4
where Qcalculated is the deterministic bearing pressure
calculated with the ISO model of Houlsby and Martin 5.5 Comparison with InSafe guidelines
and
The InSafe guidelines give a direct comparison of the
HMA and HRN methods. Figure 13 presents a cumu-
lative distribution of the ratio of the measured vertical
load to the HMA prediction for the ∼100 monitored
is a lognormal (LN) random variable for the model data points in the InSafe JIP database (14 sites around
factor with a mean of 1.01 and a standard deviation the world). The spread of the results is greater than
of 0.08. Equation 18 involves only one random vari- that found here for the more consistent 16 Gulf of
ate. Hence, a probabilistic analysis can be conducted Mexico sites of Menzies and Roper (2008) and Hos-
by retrieving the quantiles of the lognormal distribu- sain et al. (2014). The InSafe guidelines provide the
tion of η that correspond to relevant probabilities of model factors that can be directly read off Figure 13
exceedance or non-exceedance and multiplying these or Table 5.
by the deterministic value Qcalculated .
As an example, Site 1 of Menzies and Roper (anal- 6 PUNCH-THROUGH PREDICTION: SAND
ysed deterministically in Figure 9) was reanalysed with OVER CLAY
the ISO Houlsby-Martin method and the model fac-
tor distribution. Probabilities of exceedance of 0.995, A thin layer of sand (less than one spudcan diam-
0.95, 0.05 and 0.005 were considered (i.e., lines repre- eter) overlying a weaker stratum of clay is partic-
senting, for each depth, the probability that the load ularly hazardous during the installation of spudcan
will be exceeded by 99.5%, 95% etc). The corre- foundations.
sponding quantiles of the lognormal distribution of
η are 0.821, 0.884, 1.147 and 1.234, respectively.
6.1 Limitations of ISO recommendations
Capacity lines pertaining to the exceedance proba-
bility levels were obtained by applying Equation 18. The new ISO 19905-1 guidelines retain the SNAME
Figure 12 plots the measured capacities (black circles, recommendation to use either a load spread approach
squares and triangles), the deterministic capacity and or the punching shear mechanism to calculate the peak
the capacity lines pertaining to the 99.5, 95, 5 and 0.5% penetration resistance for a footing on sand overlaying
probabilities of exceedance. clay.

194
Figure 14. Load spread method of ISO (2012). Figure 15. Punching shear method of ISO (2012).

alternative design approach in the ISO guidelines. For


In the load spread approach, the load is projected a flat footing at a particular depth (i.e. wished into
through the upper sand layer at an angle of spread place), the failure mechanism is assumed to comprise
α to a fictitious footing of increased bearing area a truncated cone in the sand layer being depressed
at the sand-clay boundary, as shown in Figure 14. into the underlying clay. However, for calculation pur-
Some debate still remains regarding the applicable val- poses, a simpler vertical shear plane is used (SNAME
ues of the assumed angle of spread; ratios of 1:5 to 2008, ISO 2012), as illustrated in Figure 15. The peak
1:3 (horizontal:vertical) are typically recommended. resistance qpeak is calculated with the punching shear
In retrospective simulations of a database of 71 cen- method as follows:
trifuge tests that cover most practical punch-through
scenarios (and will be further described in Section 6.4),
the steeper spreading ratio of 1:5 yielded peak punch-
through loads that were, on average, 21% lower than
those in the flatter 1:3 case. where the punching shear coefficient Ks is related to
In the application of the load spread method in ISO the normalised shear strength of the underlying clay
19905-1, there seems to be ambiguity as to where the layer through the ratio of the bearing capacity of the
surcharge should be considered in the calculation of clay to that of the sand and through the effective angle
the bearing capacity of the fictitious larger footing. of internal friction of the sand; it is suggested that the
As the authors see it, it is consistent, when calculat- value of this coefficient be obtained from the chart pre-
ing the bearing capacity of the fictitious footing, that sented by Meyerhof & Hanna (1978). γs is the buoyant
the actual surcharge on the sand-clay layer should be unit weight of the sand.
assumed. That is, the surcharge at the spudcan level There is a belief in the industry that both meth-
(q0 ) should have the additional surcharge due to the ods of ISO under predict the peak capacity (Baglioni
sand layer between it and the fictitious footing added et al. 1982, Osborne et al. 2009). This has been indeed
to it (i.e. q0 + Hs γs ). However, close reading of the shown to be the case, with averages of 42% and 56%
ISO guidelines seems to indicate that the pressure under prediction with respect to centrifuge data for
should be at the spudcan and therefore remain as just the load spread and punching shear methods, respec-
q0 (i.e. engineers are referred to Figure A9.3-11 of tively (see Section 6.4 and also Lee et al. 2013b and Hu
ISO 19905-1 that shows the effective overpressure at et al. 2014a ). The methods were originally developed
the footing level to be used in conjunction with the (i) for shallow wished-in-place footings (rather than
bearing capacity Equation A9.3-7). Following the ISO continuously penetrating spudcans) and (ii) for stress
guidelines, the weight of the sand between the spud- magnitudes significantly lower than those experienced
can and the fictitious footing is subtracted from the by spudcans. There is an inconsistency between the
fictitious footing’s capacity in the qpeak reported here. parameters required to fit small-scale model test data
In the calculations of this paper, the surcharge at and those required to fit field observations (Young &
the fictitious footing level (i.e. the sand-clay inter- Focht 1981, Higham 1984).
face) is used, as it provides higher (less conservative) Based on observations from visualisation experi-
required capacities. The authors also believe that most ments using geotechnical centrifuges (Teh et al. 2008,
engineers, if not reading the guidelines, would do the Hu et al. 2014b ) and finite element analysis (Lee 2009,
calculation in this way. A discussion of the difference Hu et al. 2014c,d ), there is overwhelming evidence
in the two interpretations on the retrospective predic- that the mechanism of failure at the peak is signifi-
tion of the experimental database is provided in Hu cantly different from the mechanisms assumed in the
(2015). ISO guidelines. In the former study, digital images
The punching shear mechanism developed by were captured continuously during the installation of
Hanna and Meyerhof (1980) is retained as an a half-spudcan against a transparent window. Through

195
Figure 16. Normalised incremental deviatoric shear strain
contours measured at qpeak for a conical angle of 14◦ and
Hs /D = 0.68 (after Hu et al. 2014b ).

analysis using Particle Image Velocimetry (PIV) cou-


pled with close-range photogrammetry corrections, a
change in the spudcan failure mechanism with pene-
tration depth (and under stress conditions equivalent to
the offshore case) was observed. An example of such
a change is illustrated in Figure 16. In this visualisa-
tion experiment, the silica sand layer had a thickness
of Hs /D = 0.68, with a relative density Id of 74%, and
the underlying undrained shear strength of the clay
layer was ∼20 kPa at the interface and increased by
∼2 kPa/m with increasing prototype depth. The key
aspects of the observations were as follows:
Initial spudcan tip penetration: The soil movement
radiates from the spudcan tip, is very limited and is
concentrated in the upper sand layer.
Peak resistance: The peak resistance occurs at a
relative shallow embedment, consistently measured at
∼0.12Hs (consistent with the full-spudcan centrifuge
tests of Teh et al. (2010), Lee et al. (2013a ) and Hu
et al. (2014a ) and discussed further in a probabilistic
context in Section 6.5). The mechanism consists of
a truncated cone of sand being forced vertically into
the underlying clay layer; farther from the centre line,
the pure vertical movement (directly under the spud-
can) gradually changes to radial movement with small Figure 17. Normalised incremental deviatoric shear strain
amounts of heave. This behaviour is manifested in the distributions at depths d = 0.6D, 1.1D and 1.6D (with
normalised incremental deviatoric strain contours of Hs = 0.68D these represent depths just before the sand-clay
Figure 16. The outer angle of the sand frustum being interface and then 0.42D below and 0.92D below, respec-
tively).
forced into the clay reflects the dilation angle ψ of the
sand (a much narrower angle than that obtained using
the load spread method of ISO). 1.6D (just before to just less than 1D beyond the sand-
In the underlying clay layer: After punch-through, clay interface, which was initially located at 0.68D).
in the clay layer, the sand becomes trapped in a nearly As discussed in Hu et al. (2014b ), the sand plug is
vertical plug below the spudcan. The evolution of this not fully formed at the sand-clay interface; rather,
sand plug underneath a conical footing of 14◦ is illus- the sand is still undergoing a significant amount of
trated in Figure 17 for depths of d = 0.6D, 1.1D and shearing (Figure 17a). A conical wedge of clay is also

196
Figure 18. Normalised sand plug heights at various pene- Figure 20. Depth at which punch-through occurs (after Hu
tration depths below the interface (after Hu et al. 2014a,b ). et al. 2014a ).

equal to the dilation angle ψ is assumed to be pushed


into the underlying clay. The footing pressure and the
weight of the sand frustum are opposed by frictional
resistance along the sides of the sand block and by
the bearing capacity of the underlying clay (Lee et al.
2013b ; Hu et al. 2014a ).
The primary improvement in the equations of Hu
et al. is the incorporation in the mechanism of the pen-
etration depth at which punch-through occurs (dpeak ).
Punch-through has been demonstrated to consis-
tently occur at shallow embedments of approximately
0.12Hs , as shown for 66 centrifuge tests in Figure 20.
There is some debate regarding the optimal form of
this equation, as some degree of scatter is evident
Figure 19. Model of peak resistance of a spudcan in sand
overlying clay.
even in precise and carefully executed model exper-
iments. However, the scatter can be accounted for in
the probabilistic methodology detailed in Section 6.5.
trapped beneath the sand plug, and the combination A design equation for the peak resistance qpeak was
of the spudcan, the stiff sand and the wedged clay derived by treating the conceptual sand frustum as a
produces an “enlarged spudcan” and soil flow mech- series of infinitesimal horizontal discs and equating
anism. The bearing capacity factor required to predict vertical force equilibrium, as in a ‘silo’ analysis. The
the behaviour in this scenario, if a traditional undrained detailed derivation can be found in Lee et al. (2013b )
approach is used, will significantly increase. As pene- and Hu et al. (2014a ), and the final design equation,
tration continues, the size of the soil plug decreases which includes dpeak = 0.12Hs , is as follows:
and stabilises with a height Hplug of approximately
0.9Hs . This behaviour was consistently observed in
seven centrifuge tests, for conical angles from 0 to 21◦
(Figure 18).Accounting for this spudcan, plug and clay
wedge is essential to obtain accurate predictions of the
post-peak spudcan behaviour.

6.2 New approach to predict qpeak


A new calculation method that, for the first time,   
accounts for the stress level and dilatant response of where E = 2 1 + DF tan φ∗
tan ψ
− 1 and q0 is the sur-
the sand was first advocated by Lee et al. (2013b ) and charge. The definition of the model parameters are
was then improved on by Hu et al. (2014a,b,c,d ). The provided in Table 6. Although the equation appears
spudcan failure mechanism in this model is illustrated daunting, it looks more complex than it actually is.
in Figure 19. It is consistent with the observations from The first term is related to the bearing capacity of
the PIV observation experiments described above. In the underlying clay and the frictional resistance from
this mechanism, a sand frustum with a dispersion angle the upper sand layer, whereas the second term arises

197
Table 6. Parameters for the sand-over-clay model proposed
by Lee et al. (2013b ) and Hu et al. (2014a ).

Symbol Parameter Parameter determination

D Footing diameter Known parameter


Hs Sand thickness In situ tests
ID Relative density In situ or laboratory tests
γs Buoyant unit In situ or laboratory tests
weight
φcv Critical state Laboratory tests or based
friction angle on mineralogy
Q Natural log of grain Bolton’s (1986) equation
crushing strength
IR Relative dilatancy Iterative calculations
index between Equation 21 and
  
φ Friction angle IR = ID Q − ln qpeak − 1 Figure 21. DF measurements from half-footing PIV tests

φ − φcv = 2.65IR (solid) and centrifuge tests (non-solid) (Hu et al. 2014b ).
ψ Dilation angle 0.8ψ = φ  − φcv
(φ  , φcv and ψ in degrees) One of the primary difficulties encountered in the
sin φ  cos ψ derivation of an equation for sand over clay is the
φ∗ Reduced friction tan φ∗ = estimation of the shear stress that is acting on the sand’s
angle due to the 1 − sin φ  sin ψ
slip surface. In Equation 21, this difficulty is overcome
non-associated by introducing a distribution factor, DF , that relates the
flow rule normal effective stress at the slip surface to the mean
sum Undrained shear In-situ or laboratory tests vertical effective stress averaged across the sand frus-
strength at sand- trum. This approach allows Equation 21 to be easily
clay interface calculated, but the values of DF must be (pre)calibrated
κ Strength profile of k can be inferred from site to the experimental database (Lee et al. 2013b ; Hu et al.
the clay, where k is investigations, and 2014a,b ). The expression for DF provided in Table 6
the gradient with k (D + 1.76Hs tan ψ) accounts for various spudcan shapes (flat and coni-
depth and κ is a κ=
sum cal, 7◦ to 21◦ ) and sand densities (loose, medium and
non-dimensional
parameter dense) and was developed based on 56 centrifuge tests,
as shown in Figure 21.
Nc0 Bearing capacity Nc0 ≈ 6.34 + 0.56κ
factor (Houlsby and Martin 2003)
Following Equation 21 and Table 6, the stress level
and dilatant response of the sand are calculated using
DF Distribution factor: For spudcans with conical an iterative approach.
the ratio of the bases greater than 7◦ , use
 −0.576
vertical effective Hs
stress at the slip DF = 0.642 ,
D 6.3 Predicting behaviour in underlying clay
planes to the mean 0.16 < Hs /D < 1.0, but for
vertical effective flat spudcans, use As the spudcan penetrates through the sand layer
stress within the  −0.174 (Figure 17), a plug of sand becomes trapped under-
Hs
sand layer DF = 0.623 , neath the spudcan. This effectively enlarges the footing
D size and, as the undrained shear strength increases
0.21 < Hs /D < 1.12. with increasing depth, it mobilises soil with a higher
shear strength at the base of the plug. The ISO and
SNAME guidelines give no recommendations on how
to account for this plug whilst retaining a classical
from the self-weight of the sand frustum. The term bearing capacity solution. Lee (2009) derived new Nc -
qpeak , as calculated from Equation 21, must be less factors for buried cylinders via small-strain finite ele-
than the bearing capacity of the spudcan foundation ment analysis and recommended their use to approx-
in the sand alone. Examples of methods for calculat- imate a composite spudcan and soil plug (although
ing the bearing capacities of spudcans in sand alone the experimental data fit the model only for a range
include Brinch Hansen (1970), Cassidy & Houlsby of plug heights between 0.6 to 0.9 Hs ). Upon examin-
(1999, 2002), Houlsby & Cassidy (2002), Randolph ing 52 centrifuge and retrospective numerical Coupled
et al. (2004) and ISO (2012). Eulerian Lagrangian (CEL) large-deformation anal-
Recent advances reported by Hu et al. (2014a ) yses, Hu et al. (2014d ) recommended a simplified
have demonstrated the applicability of Equation 21 method in which the formula
in medium to loose sand conditions as well as in the
original dense sand. All investigated cases, however,
concerned silica sands.

198
can be used to obtain an estimate of the variabil-
ity in the prediction of the bearing pressure in the
underlying clay.

6.4 Full profiles: model performance compared to


centrifuge database
To the authors’ knowledge, 71 geotechnical centrifuge
tests of circular foundations in sand-over-clay scenar-
ios have been reported in the literature within the last
five years. These tests were assembled into a database
by Hu (2015) and used to compare the results of ISO
and three new predictive methods to the experimen-
tal results. In tests performed at both UWA and the
National University of Singapore, the conical angles
of the footings ranged from 0 to 21◦ , the Hs /D ratios
Figure 22. Bearing capacity factor in clay underlying sand ranged from 0.16 to 1.12, the sand ID values ranged
(classified in terms of the strength gradient in the clay; after from 43 to 99%, and the strength gradients kD/sum of
Hu et al. 2014d ). the clay ranged from 0.6 to 2.8.
Two tests, labelled D1F40a (Lee et al. 2013a ) and
L1SP3 (Hu et al. 2014a ), are presented in Figure 24.
D1F40a represents a dense silica sand case (ID = 0.92)
with Hs /D = 0.6 and a typical spudcan shape with a
conical angle of 13◦ . The same spudcan shape was
used in L1SP3, but the sand was loosely laid (ID =
0.43), with Hs /D = 0.6. In both cases, a clear qpeak
was observed, at 520 kPa and 365 kPa respectively,
which was followed by a sharp reduction in measured
capacity before the capacity again increased in the
underlying clay layer.
In this paper, it is assumed that the vertical distance
penetrated between qpeak and the depth at which the
same capacity is regained in the clay can be regarded
as a measure of the punch-through intensity; this quan-
tity is labelled dpunch . This is also shown on Figure 3.
Experimentally, this quantity took values of 9.3 m and
7.7 m in tests D1F40a and L1SP3, respectively.
Figure 24 shows that the two ISO-recommended
methods under predict the capacity both for qpeak and
Figure 23. Normal distribution fit of difference between in the underlying clay; in the latter case, the under
measured Nc and Equation 22. prediction is attributed to not accounting for the soil
plug. For these cases, the method of Hu et al., as
described in this paper, provides accurate predictions
is assumed to apply to the classical formulation of both the peak and the capacity in the underlying clay.
This performance is to be expected, as the method was
calibrated using these two tests. In the following, the
performance of the Hu et al. method is verified with
the entire database, including tests that were not used
where su0 is the soil strength at the lowest elevation to guide the derivation of the method.
of the spudcan’s widest cross-sectional area; hplug is Figure 25 summarizes the predictions by the method
the sand plug height, which is estimated to be 0.9Hs ; of Hu et al., represented by Equations 21 to 23 in this
and γc is the effective unit weight of the clay. Figure paper, assessed against the full 71-test database. The
22 illustrates the fit of Equation 22 to all experimen- predictions with the ISO recommendations are also
tal and numerical analysis considered in this study. shown. Comparisons of the predicted and experimen-
Also shown is the range ±one standard deviation tally measured values of qpeak , the bearing capacity
(σ ). The difference between the experimentally mea- at 1D below the sand-clay interface and dpunch are
sured Nc and the fit of the bearing capacity (i.e. presented in Figures (a), (b) and (c), respectively.
(Nc,measured − Nc,fit )/Nc,measured ) are all within 20%, The statistics of the performance of each method are
with the majority within 10%. The variation was a summarised in Table 7.
normal Gaussian fit with a mean of zero and a CoV of The ability of the Hu et al. method with regard
only 7.5%. This is shown in Figure 23. These bounds to qpeak was reported earlier (Lee et al. 2013b , Hu

199
Figure 24. Comparison of experimentally measured and
predicted punch-through using Equations 21 to 23 and the
ISO guidelines for two centrifuge tests.

et al. 2014a ), but the analysis herein is with a more


extensive database (Figure 25a). The results provide
evidence that the existing load spread and punch-
ing shear models recommended by current jack-up
industry guidelines (ISO 2012) significantly and con- Figure 25. Comparison of the performance of the Hu et al.
(2014a,b,c,d ) and ISO methods with the full database for (a)
sistently under predict qpeak . On average, the ISO qpeak , (b) Nc in the underlying clay, and (c) dpunch .
methods predict only 58% and 44% of the measured
pressure in the centrifuge tests. Furthermore, the ISO-
methods show significantly higher variability. The Hu
et al. method yields values that are 1% higher than the On average, only 46% of the capacity is predicted by
experimental values on average, with a relatively small the ISO methods. The Hu et al. method yields pre-
CoV of 8.2%. dictions for the bearing capacity in the clay layer (at
The same trend can be seen with the retrospective 1D below the sand-clay interface) that are on average
simulations with the ISO guidelines. The adoption 97% of the experimental values, with a relatively small
of Nc factors derived for single-layer clays and the CoV of 7.7%. Because of the accuracy of its predic-
assumption of no sand plug formation result in a sig- tions of qpeak and Nc , the Hu et al. method also predicts
nificant under prediction of the capacity in the clay. quite accurately the penetration dpunch (mean of 1.06),

200
Table 7. Statistics of the performance of the ISO and Hu
et al. methods from simulations of the 71 centrifuge tests in
database (also reported in Hu 2015).

ISO (2012)

Load Punching Hu et al.


Method spread∧ shear method

qpeak, calculated / No. of tests 71 71 71


qpeak, measured Mean 0.58 0.44 1.01
σ 0.09 0.07 0.08
CoV (%) 15.8 14.1 8.2
Nc, calculated / No. of tests# 54 54 54
Nc measured Mean 0.45 0.45 0.97
σ 0.07 0.07 0.07
CoV (%) 15.7 15.7 7.7
dpunch, calculated / No. of tests∗ 62 56 54 Figure 26. Variation of punch-through depth with sand layer
dpunch, measured Mean 2.50 1.64 1.06 thickness.
σ 3.90 1.56 0.39
CoV (%) 155.9 95.5 36.5 Figure 20 and the data that were used to fit the prob-
abilistic model of punch-through failure of sand over
# Only 54 tests are available for Nc comparison (as some did
not penetrate past 1D below the sand-clay interface)
clay). The punch-through depth varied in the determin-
*Number of punch-throughs predicted istic model, with its scatter increasing with a thicker
ˆLoad spread of 1:3 (horizontal:vertical) used in calcula- sand layer (Figure 26). A multiplicative random vari-
tions and the surcharge acting at the sand-clay interface was ate was used as a model factor (εd ) and applied to
assumed the deterministic model of the punch-through depth
(darker line in Figure 26):

although with a wider variation than for qpeak and Nc


(CoV of 36.5%).
A computational Bayesian Markov Chain Monte
Carlo (MCMC) approach was used to determine that
6.5 Probabilistic modelling of punch-through, the model factor, εd , follows a truncated lognormal
sand over clay distribution. The mean value and standard deviation of
the MCMC distribution were 0.98 and 0.14. The prob-
The previous section provided details of a new method ability density function (PDF) and the corresponding
for predicting spudcan punch-through from a layer of cumulative distribution function (CDF) of the model
strong sandy material into an underlying weak clay factor εd are shown in Figure 27.
layer (following Lee et al. 2013b and Hu et al. 2014a ). A similar analysis was performed on the punch-
The calculation of the peak capacity was based on a through pressure, qpeak . As the variation in qpeak
‘silo’ analysis and, for the first time, accounted for increased with higher bearing pressures (see Fig-
the stress level and dilatant response of the sand. ure 25a), a multiplicative model factor, εq , was also
New bearing capacity factors that account for the for- applied to the deterministic model for qpeak :
mation of a sand plug were suggested. The results
in the previous section show that the method can
be used to retrospectively simulate an experimental
database more accurately than the current ISO guide- where qpeak,calc is the deterministic value calculated by
lines, and that the scatter in the predictions provide the Hu et al. (2014a ) model (and provided in Equa-
initial guidance on the uncertainties in the model. tion 21 and Table 6). The model factor, εq , followed
The method is a deterministic model, yielding only a lognormal distribution with mean value of 1.00 and
one prediction for any given site. In reality, the soil standard deviation of 0.19.
properties and layering at a site are uncertain due to With these model factors, punch-through pressure
inherent variability and measurement errors. These and depth were estimated probabilistically. The prob-
uncertainties are ignored in the deterministic model. A ability of punch-through occurring at a set of pressure
probabilistic assessment of the likely punch-through values (qpeak,i ) was first investigated. This probability
pressure and depth is required to account for the is termed as the marginal probability for the punch-
uncertainties. In this section, an estimation on the through pressure (P(qpeak,i )). The marginal probability
probability of punch-through pressures and depths is for the punch-through depth (P(dpeak,j )) was also cal-
described. culated at a set of depth values (dpeak,j ). The marginal
The punch-through depths (dpeak ) in 30 centrifuge probability values were obtained independently by per-
tests are plotted as a function of sand layer thick- forming Monte Carlo simulation on the punch-through
ness (Hs ) in Figure 26 (these tests are a subgroup of pressure and depth models (Equations 24 and 25).

201
Figure 27. Probability density function of model factor εd .

Assuming the two random variables to be mutually


independent, a bivariate joint probability density in the
(qpeak , dpeak ) can be obtained by multiplying the two
marginal probabilities:

for i = 1 to nq and j = 1 to nd , and where P(qpeak,i ,


dpeak,j ) is the probability that punch-through occur-
ring at (qpeak,i , dpeak,j ) and nq and nd are the
number of Monte Carlo simulations. The probabil-
ity P(qpeak,i ,dpeak,j ) estimated before installation was
called the prior probability.

6.5.1 Example calculations


The probabilistic potential for punch-through was
evaluated for three cases: D1F40a, L1SP3, and NUS-
F5 (tests of Lee et al. 2013a , Hu et al. 2014a and Teh
et al. 2010 respectively). The deterministic penetra-
tion profiles for D1F40a and L1SP3 were discussed in
Section 6.4 and shown in Figure 24.
The probabilistic predictions of the spudcan pres- Figure 28. Prior probability contours of punch-through
compared to measured and deterministic values for three
sures and depth are shown as contours on Figure 28.
tests.
These predictions are based on 20 000 Monte Carlo
simulations. The contours are constructed with 50

202
discrete intervals subdividing the range from 0.9 of particularly helpful if the prediction model deviates
the minimum qpeak to 1.1 the maximum of qpeak from the measured punch-through value.
(and another 50 intervals between 0.9 min(dpeak ) to The prior probability can be further updated by
1.1 max(dpeak )). The contour give the probability of incorporating the monitored data using Bayes’ theo-
punch-through occurring within one of the increments rem. This will be presented in Section 7.3.5.
(qpeak,i , dpeak,j ). The extent of the contours “footprint”
along the qpeak and dpeak axes give the range of pos-
sible values for punch-through pressure and depth at 7 PUNCH-THROUGH PREDICTION: STIFF
punch-through. For higher probabilities, the contours OVER SOFT CLAYS
in Figure 28 are dark and for lower probabilities the
contours are light. The exact probability values are a Equally problematic for potential punch-through dur-
function of the number of discretization chosen (with ing spudcan installation is a layer of stiff stronger clay
the sum being one). Therefore, if more discretization overlying weaker clay. Such conditions are encoun-
had been used (say 100), the probability values of the tered increasingly in frontier regions of offshore
contours would be lower (around half for this case, but South-East Asia (Castleberry and Prebaharan 1985;
still summing to one). Osborne and Paisley 2002; Paisley and Chan 2006).
The marginal probability showing the probability
distribution of qpeak alone is shown in the upper sec- 7.1 Limitations of ISO recommendations
tion of each of the three diagrams in Figure 28 and
the marginal probability of dpeak alone is shown on The SNAME (2008) and ISO (2012) industry guide-
the right hand side of the diagrams. A description on lines both recommend the punching shear method of
how these marginal probabilities can be used to predict Brown and Meyerhof (1969) for predicting the peak
probability of exceeding a preload level is provided for load of a spudcan penetrating a seabed consisting of a
the stiff-over-soft clay example in Section 7.4. stiff clay crust overlying soft clay. The peak resistance
The punch-through depth and stress measured in qpeak is calculated, under the assumption that the spud-
the centrifuge are presented by the cross in the three can is positioned at the surface of the strong clay layer,
diagrams in Figure 28. The predicted values using as follows:
the deterministic Hu et al. (2014a) method are also
demonstrated in Figure 28 as a square.
Test D1F40a represents a dense silica sand case
(ID = 0.92) with Hs /D = 0.6 and a typical spudcan where H is the height of the stronger upper layer; sut
shape with a conical angle of 13◦ . The measured qpeak and sub are the undrained shear strengths of the over-
was observed at 520 kPa. The probability of punch- and underlying clays, respectively; and Nc is the appro-
through occurring at the measured values (qpeak = priate bearing capacity factor for the spudcan-shaped
520 kPa, dpeak = 0.67 m) is 0.0077, which is larger footing.There are two main components to the method:
than the probability of punch-through occurring at (a) the frictional resistance around the periphery of the
the predicted values, 0.0065 (qpeak = 514 kPa, dpeak = soil plug trapped beneath the advancing spudcan and
0.74 m). This prior probability indicates a higher (b) the end bearing at the base of the plug.
punch-through probability due to the numerous uncer- As highlighted by Hossain & Randolph (2009b ) and
tainties. demonstrated in Section 7.2 below, the method does
The same spudcan shape was used in L1SP3, but the not account for the movement of the plug into the
sand was loosely laid (ID = 0.43), with Hs /D = 0.6. In underlying soft layer (because the base of the plug is
this case, qpeak was measured at 365 kPa. The most assumed to be fixed at the upper-lower layer interface
probable contours were lying between the measured regardless of the spudcan penetration). This approxi-
punch-through values and the predicted values, as mation has been found to exert significant influence
shown in Figure 28b. Compared to a sole determinis- on the prediction results.
tic data point (the square), the contours and marginal
distributions provide more information and illustrate
7.2 New predictive methods
the effect of the uncertainties in qpeak and dpeak on the
possibility of a punch-through. Recently, Edwards & Potts (2004) and Hossain
Test NUS-F5 was conducted at the National and Randolph (2009b ) have proposed new design
University of Singapore (Teh et al. 2010) and was not approaches. Edwards & Potts (2004) proposed that the
used in the calibration of the deterministic Hu et al. peak resistance be determined as the sum of the bear-
model or the probabilistic extension of this paper. The ing capacity of the underlying clay and a fraction of the
deterministic punch-through pressure deviated signifi- capacity of the upper layer. Based on their finite ele-
cantly from the measured one (Figure 28c). In this case, ment analysis, they provided factors relating the depth
the prior probability calculated probabilistically indi- of the upper layer normalised with respect to the spud-
cated the range of punch-through qpeak and dpeak which can diameter to the fraction of the upper-layer strength
the engineer may consider likely. The predicted range that should be used.
captured even the measured value which was much By contrast, Hossain & Randolph (2009b ) based
lower than predicted deterministically. This process is their new design approach on the failure mechanisms

203
Figure 30. Parameters used for stiff-over-soft clay.

upper soil layers, subs /sut ; (b) the thickness of the upper
layer relative to the spudcan diameter, H/D; and (c) the
normalised clay strength, subs /γb D. Definitions of the
parameters used in the model are provided in Fig-
Figure 29. Failure mechanism (Hossain and Randolph
ure 30. Pressure and penetration data can be plotted
2009b ). and analysed efficiently with the following normalised
parameter (Hossain & Randolph 2009b ):
observed in large-deformation finite element and cen-
trifuge experiments performed using half-spudcan
models against a clear window via digital imaging and
PIV analysis. The Bayesian updating approach was used, with a
The following four types of flow mechanisms were framework to update the prior probabilistic estimate
observed by Hossain and Randolph in stiff-over-soft with new information gained through the monitoring
clay soils: (i) the initial vertical movement of the upper of load-displacement points during installation (see
layer into the lower layer and the consequent deforma- also geotechnical applications by Zhang et al. (2004,
tion of the interface, (ii) the creation of a soil plug 2009a,b ) and Uzielli et al. (2015).
consisting of clay from the strong upper layer beneath For the jack-up penetration analysis, Bayes’ theo-
the penetrating spudcan, (iii) the delayed (compared rem can be expressed as:
with the case of a single clay layer) backflow of soil
around the spudcan into the cavity above, and (iv) an
eventual localised flow around the spudcan in the soft
underlying layer.
Although the predictive method of Hossain &
Randolph (2009b ) can be used to estimate a full spud- or as “the probability of punch-through occurring at
can penetration profile (as it accounts for the evolution a loading pressure q∗peak and depth d∗peak conditioned
of the failure mechanism at each penetration depth), on the observed load-displacement behaviour (qmon ,
the discussions presented in this paper focus on the dmon ) observed during installation is proportional to
peak capacity prediction and the events leading to this the product of (1) the prior (pre-installation) probabil-
potential punch-through point. Hossain and Randolph ity of occurrence of punch-through for q∗peak and d∗peak
describe the critical punch-though mechanism of a and (2) the likelihood, i.e., the probability of observ-
stiff-over-soft clay as a punching shear model of a ing the actual qmon and dmon during installation should
truncated cone. This model is illustrated in Figure 29. punch-through occur for q∗peak and d∗peak ”.
The full details of the equations will not be provided A computational Bayesian approach relying on
here, as they will be recast in a probabilistic form in Markov Chain Monte Carlo (MCMC) methods was
Section 7.3. However, they are available in Hossain & implemented for uncertainty characterization by
Randolph (2009b ) and Uzielli et al. (2015). Uzielli et al. (2015). Only the final equations are
provided here, along with application examples to
illustrate the potential of such methods in the jack-
7.3 Probabilistic modelling of punch-through, up. Uzielli et al. present the details of the Hossain &
stiff clay over soft clay Randolph equations and the associated uncertainties
7.3.1 Deterministic and probabilistic models for predicting the punch-through pressure (qpeak ), the
The deterministic analytical model of Hossain and depth at which punch-through occurs (dpeak ) and the
Randolph (2009b ) was formulated into a probabilistic load-displacement response.
framework by Uzielli et al. (2015).
Hossain & Randolph (2009b ) showed the punch- 7.3.2 Approach to quantify model uncertainty
through pressure and depth in stiff-over-soft clay to To update the punch-through pressure based on moni-
be complex functions of the following dimensionless toring measurements requires a probabilistic formula-
factors: (a) the strength ratio between the lower and tion of the load penetration curve. Quantifying model

204
uncertainty in the calculation of punch-through pres- Table 8. Model uncertainty factors εq , εd , εb .
sure and depth is also required for the Bayesian
updating. Distribution Parameters# Note
The results of Large Deformation Finite Element
(LDFE) analyses were used to establish the uncertain- εq Normal µ = 0.02 σ = 0.29
ties in the analysis parameters. The equations derived εd Log-normal ξ = 0.05 λ = 0.23
εb Truncated µ=0 σ = 0.02 Truncated at
statistically to match the LDFE data points, differ
normal −0.05 and 0.05
slightly in format from the Hossain and Randolph
deterministic equations. #
µ: mean; σ : standard deviation; ξ : log-mean (mean of
The population of the LDFE results were compared the natural logarithm); λ: log-standard deviation (standard
with the predictions with the Hossain & Randolph deviation of the natural logarithm)
model was evaluated to derive a model of (i) the punch-
through pressure qpeak , (ii) the punch-through depth
dpeak , and (iii) the load-displacement curve leading to
the peak punch-through event (used for the Bayesian
updating).

7.3.3 Formulation of model uncertainty


Using the Hossain & Randolph parameter χ , the
uncertainty in qpeak was expressed as:

where η1q = 1.78, η2q = 12.38, η3q = 1.17 and η4q =


0.87 are model estimators and εq is an additive, nor-
mally distributed model uncertainty factor (Uzielli
et al. 2015). The latter accounts for scatter in the LDFE
results around the analytical deterministic model in
Equation 22, as well as for epistemic uncertainties in
the parameters appearing in the model (see Uzielli et al.
2015). The mean and standard deviation of the model
uncertainty factor were taken as the sample means of
the posterior distributions of the mean and standard
deviation, respectively, as obtained using the MCMC
approach. Its values are provided in Table 8.
To model the uncertainty of the punch-through
depth (dpeak ), a sigmoid model was found to offer the Figure 31. Uncertainty modelling of (a) punch-through
best fit to the LDFE data: pressure, (b) punch-through depth and (c) load-displacement
curve (dots = LDFE data; black curve = deterministic regres-
sion line; grey = probabilistic realisations using MCS) (after
Uzielli et al. 2015).

where the values of θb1 = 0.154 and θb2 = 0.515 were


found to fit the LDFE data well. Again, a ran-
dom variable εb was used to parameterize the model
uncertainty.
where η1d = 0.14, η2d = 3.77, η3d = 0.337, η4d = 7.0
and η5d = 0.047 are the model coefficients and εd is 7.3.4 Illustration of model uncertainties
the best-fit random model factor with a lognormal Table 8 provides the distributions of the model uncer-
distribution. tainty factors ε . Figure 31 (a, b, and c) compares the
The probabilistic description of the bearing pressure results of 20 000 probabilistic realisations of Equa-
(q) vs depth (d) curve was also based on the LDFE tions 30 and 31 (Monte-Carlo Simulations (MCS))
data of Hossain & Randolph (2009b ), with a single- with the LDFE results of Hossain and Randolph for
parameter power model: the normalized capacity.
The probabilistic model was calibrated for a broad
range of potential punch-through conditions, but its
applicability is still restricted to the bounds of the
LDFE database used. A discussion of its range of

205
Table 9. Range of applicability of model.

H/D χ min χ max

H/D ≤ 0.50 0.10 0.40


0.50 < H/D ≤ 1.00 0.20 0.80
1.00 < H/D ≤ 1.50 0.30 1.20
1.50 < H/D ≤ 2.00 0.40 1.60

subs /sut χ min χ max

subs /sut ≤ 0.20 0.10 0.70


0.20 < subs /sut ≤ 0.40 0.20 1.30
0.40 < subs /sut ≤ 0.60 0.60 1.30
0.60 < subs /sut ≤ 0.80 0.80 1.60
Figure 32. Soil and jack-up characteristics for example
case.

applicability is provided in Uzielli et al. (2015) and In the formulation of Bayes’theorem, the likelihood
in Table 9. term P(qmon , dmon |qpeak,i , dpeak,j ) (Term C) expresses
the probability of predicting spudcan penetration val-
7.3.5 Bayesian updating ues that may be compatible with the penetration values
Bayes’theorem was formulated for the present analysis observed during installation, given the occurrence of
with four terms, A, B, C and D, as follows: punch-through at the ith value of the peak pressure
qpeak,i and at the jth depth value dpeak,j . Equation 32
describes the model for estimating the likelihood of the
load-displacement behaviour, which can be assessed
for compatibility against a set of monitoring data.
The Bayesian likelihood is calculated for each can-
didate punch-through scenario (qpeak,i , dpeakj ) by doing
Monte Carlo simulations of Equations 30 and 31.
This allows for the calculation of a sample of load-
displacement model factors ηb that are compatible with
the monitoring data.
where qpi (i = 1, . . . , Nint ) is the ith discrete candidate
value of the punch-through pressure qpeak (of a set of 7.4 Example calculations
Nint ), dpj (j = 1, . . . , Nint ) is the jth discrete candidate The practical application of the Bayesian framework
value of the punch-through depth dpeak , qmon is the is illustrated through two examples. The details of the
set of Nm measured loading values, and dmon is the installation are provided in Figure 32.
set of Nm measured penetration-depth values for the In the two hypothetical examples, the spudcan is
loading pressures qmon . It is noted that in Equation 33 assumed to be 15 m in diameter (D). The engineer need
(in order to save space) qp is used to represent qpeak to predict whether a peak punch-through event (qpeak )
and dp to represent dpeak . Equation 33 states that the will occur before the preassigned preload pressure
posterior probability of a set of punch-through pres- (qpreload ) is reached. The seabed consists of a two-
sure and depth values given a set of monitored data layered clay with an upper layer of 8 m in thickness (H),
(Term A), is calculated by multiplying the prior prob- with an undrained shear strength of sut = 100 kPa, and
ability of their occurrence (Term B) with the likelihood an underlying clay layer with an increasing strength
of the measured points leading to that punch-through gradient starting at subs = 50 kPa and increasing at a
pair (Term C) and normalising by summing all possi- rate of k= 1.5 kPa/m. The buoyant unit weights of the
ble combinations of eligible punch-through pairs given clay in the top and bottom layers are γt = 5.3 kN/m3
the measured monitoring data (Term D). and γb = 6.0 kN/m3 , respectively. The probability of
Equations 30 and 31 are used to calculate the pre- punch-through is to be investigated for preloads of
installation expectation of punch-through occurrence qpreload = 530, 550 and 600 kPa.
(Term A). This joint probability density is calcu- Using the procedures described in the previous sec-
lated as the product of the two independent marginal tion, probabilistic predictions of the punch-through
probabilities, pressure and depth were made. These are illustrated
in Figure 33. These predictions were calculated based
on 20 000 Monte Carlo simulations for each model,
and refer to discrete intervals of stress and depth of
and is evaluated through Monte Carlo simulations of qu = 5 kPa and d = 0.1 m, respectively. The con-
the two models (as was performed for sand-over-clay tours represent iso-probabilistic punch-through sce-
scenarios in Section 6.5). narios, i.e. the combinations of punch-through stress

206
Table 10. Calculated “Factor of Safety” and probability of
punch-through for Example 1.

qpreload = 530 kPa qpreload = 550 kPa

Case FS Pf FS Pf

Prior 1.07 0.08 1.03 0.26


After mon1 2.26 0.28 2.26 0.54
After mon2 1.61 0.26 1.61 0.52
After mon3 1.41 0.21 1.41 0.52
Alternative mon3 points
mon3 : w = 2 m, q = 500 kPa 1.13 0.36
mon3 : w = 2 m, q = 525 kPa 1.08 0.07

Table 11. Monitoring data for Examples 1 and 2.

Figure 33. Prior probability of punch-through. Example 1 Example 2


Monitoring
and depth having equal joint discrete values of prior data dmon (m) qmon (kPa) dmon (m) qmon (kPa)
punch-through probability. The cross in Figure 33
represents the unique deterministic punch-through mon1 0.1 250 0.2 300
pressure of qpeak = 565 kPa and punch-through depth mon2 0.2 350 0.6 400
dpeak = 5.6 m. mon3 2.0 400 2.0 550
The deterministic punch-through point does not
necessarily coincide with the most probable punch-
through pressure and depth, which is found at the
centre of the concentric contours. This is due to the measured data. The data consist of sets of depth (dmon )
skewness in Monte Carlo simulation output samples and pressure (qmon ) measurements during the jack-up
of punch-through stress and depth. The most proba- installation. Three measurement sets are used (mon1 ,
bly failure corresponds to a pressure of approximately mon2 and mon3 ).
560 kPa and a depth of 5.4 m. Figure 34 presents the likelihood and updated pos-
terior probability for the punch-through after the
7.4.1 Example 1 acquisition of the first measured data set (mon1 for
The likelihood of reaching punch-through conditions Example 1). Figure 34a presents the probability con-
was calculated. With a deterministic analysis, a “factor tours for these curves. The single measured data point,
of safety” (FS) can be defined: dmon of 0.1 m and qmon of 250 kPa, limits the extent of
the possible load-displacement curves.
With Equation 33, the probability was updated. It
can be observed in Figure 34b that the probability con-
tours have moved to a punch-through scenario that
where values of FS ≤ 1 represent failure and values is deeper and at a lower pressure (the contours move
FS > 1 represent the margin of safety. If the jack- down and to the left in Figure 34b). For a preload target
up is to be preloaded to qpreload = 530 kPa, then the of 530 kPa, this modification increases the likelihood
FS calculated prior to installation is 565/530 = 1.07. of punch-through, with the prior Pf increasing from
As reported for many other geotechnical applications, 0.08 to 0.28 (Table 10). As the installation continues
one cannot convert a factor of safety to probability of and further data points are collected, the probability
failure. can be further updated. For Example 1, Figure 35 and
The probabilistic method allows for a direct calcu- Figure 36 present the results of the updating for two
lation of the probability that qpeak will be less than the and three sets of data. The contours in this example
preload qpreload . For a qpreload of 530 kPa, the proba- do not undergo significant changes with each addi-
bilistic calculation yields Pf = 0.08. This means that tional measurement (Pf decreases slightly to 0.26 and
there is an 8% probability that during preloading to then to 0.21). Because the contours are not changing
530 kPa, a peak punch-through load will be reached or this result implies that the monitoring points are fol-
exceeded. The first line in Table 10 presents the prior lowing typical load-displacement curves leading to a
“factor of safety” FS and the probability of failure Pf punch-through event, according to the formulation of
for preload pressures of 530 and 550 kPa. Equation 32. Table 10 presents the values of FS and Pf
With Bayes’ theorem formulated as in Equation 33, for the Example 1 cases analysed.
the probability of punch-through was updated with For a qpreload of 550 kPa (Table 10), a deterministic
the results of monitoring data. Two monitoring sce- FS of 1.03 was calculated, corresponding to a Pf of
narios are used as examples. Table 11 presents the 0.26 (or a 26% probability of a punch-through failure).

207
Figure 35. Likelihood and posterior probability of
Figure 34. Likelihood and posterior probability of punch-through updated with two sets of monitoring data
punch-through updated with one set of monitoring data (Ex. 1).
(Ex. 1).

7.4.2 Example 2
From Table 10, one can observe that Pf increases with Figure 38 and Table 12 present the results of Example
the updating with the first set of monitoring data. 2, where another set of monitored data (Example 2 in
Table 10 also shows the change FS values as more Table 11). In this example, the updating has the oppo-
data points become available. The FS values repre- site effect to that observed in Example 1. The higher
sent the ratio between the deterministic punch-through monitored pressure leads to the concentration of the
pressure predicted (565 kPa) and the most recently predicted punch-through pressure at shallower depths
measured data. and higher pressure, thus moving the contours up and
If a third set of monitoring data had been used for to the right in Figure 38. The likelihood that a punch-
Example 1, the results can be quite different. The data through pressure will be reached before the preloading
set mon3 is changed to reflect higher pressure recorded to 530 kPa has been achieved was reduced from the
at the depth of 2 m: qmon3 is increased from 400 kPa to prior 0.08 to 0.07 (for two monitored points). Interest-
500 kPa and then to 525 kPa (Table 10). ingly, Pf increases if a preload of 600 kPa is targeted,
In the deterministic calculation, the FS gradually as shown in Table 12 (increasing from a prior value
decreases as the measured data approaches qpeak = of 0.83 to 0.91 after the acquisition of three sets of
565 kPa (1.41 to 1.13 to 1.08 in Table 10). However, monitoring data). This occurs because the probability
the opposite occurs in the probabilistic calculation. In contours become more concentrated with the acqui-
Figure 37, the contours concentrate at pressure levels sition of the monitoring data (Figure 38) compared
higher than the preload of 550 kPa. This in turn reduces with the predicted pressure prior to installation (Fig-
the Pf values from 0.52 to 0.36 to 0.07. The updating ure 33). The more uncertain (less concentrated) prior
suggest that a punch-through failure is less likely to predictions, therefore, allow for a no-punch-through
occur than initially predicted. probability of 17% above 600 kPa.

208
Figure 37. Probability of punch-through at qpeak and dpeak
with alternative third monitored points (Ex. 1).
Figure 36. Likelihood and posterior probability of
punch-through updated with three sets of monitoring data
(Ex. 1).

8 CONCLUSIONS

This paper presented recent advances in both the


deterministic and probabilistic methods to analyse the
installation of the spudcan foundations of offshore
mobile jack-up platforms. The authors concentrated
on new methods that may lead to modifications to
the recently released ISO 19905-1 guidelines (ISO
2012). These include the deterministic analytical mod-
els predicting spudcan penetration in predominantly
clay sites, as well as the formulas to estimate punch-
through potential in sand-over-clay and stiff-over-soft
clay sites.
Compared to the ISO 19905-1 methods, the data
and methods in the paper show that the ISO recom-
mendations could be changed in two key areas: Figure 38. Updated probability of punch-through at qpeak
and dpeak (Ex. 2).

(1) In retrospectively simulating field installation


data from Gulf of Mexico clay sites, the primary and Martin (Appendix E of ISO), on average, fit the
“Skempton” method of ISO under predicts the field data better (mean of 101% of field data). The new
records (with a mean of 90% below the field data). methods of Hossain et al. fit the data better when the
The alternative bearing capacity factors of Houlsby more complex rate and softening effects are included

209
Table 12. Calculated “Factors of Safety” and probabilities for the probabilistic prediction of bearing pressure
of failure for Example 2. in predominantly clay layers and for punch-through
scenarios. The model factors findings could also ben-
qpreload = 530 kPa qpreload = 600 kPa efit from additional field data to help further refine
the probabilistic model factors.
Case FS Pf S Pf
A Bayesian framework to update the probabilis-
Prior 1.07 0.08 0.94 0.83 tic forecast with observations made during spudcan
After mon1 1.88 0.07 1.88 0.94 installation was introduced. The approach is poten-
After mon2 1.41 0.07 1.41 0.94 tially an extremely powerful tool for the jack-up
After mon3 N.A. N.A. 1.03 0.91 industry, as monitored installation data are commonly
available.
Methods for the probabilistic prediction of punch-
through in sand-over-clay and stiff-over-soft clay were
in the calculations. The scatter in the ratio of the
described. Example analyses showed how a prediction
predicted to recorded data was very small, with con-
prior to an installation could be updated with moni-
sistent CoV values of 8 or 9% for all four methods
tored data, and how the statistics such as the probability
(Table 4).
that a peak punch-through pressure would occur before
(2) In retrospective simulations of 71 geotechnical
preload is reached can be changed.
centrifuge tests of spudcans penetrating in a sand-
The probabilistic models are not as mature as the
over-clay profile, the ISO load spread and punching
deterministic models, and also need further develop-
shear mechanisms were shown to significantly under
ment and verification with industry monitored data.
predict the peak punch-through pressure. On average,
However, the authors believe that deterministic and
they predicted only 58% and 44% of the measured
probabilistic, when used as complements to each other,
centrifuge values respectively. These represent signifi-
provide an improved basis for assessing how reliable
cant variations and confirm the existing opinion within
one’s prediction is, and for making decisions.
industry that these ISO methods under predict cases
The deterministic models that are founded in cor-
considerably jack-up installation loads in sand-over-
rect science and realistic failure mechanisms will not
clay profiles. The new method of Lee et al. and Hu
be replaced by probabilistic models. However, the
et al. that calculates the peak capacity through a ‘silo’
deterministic models do not provide all the infor-
analysis and, for the first time, accounts for the stress
mation required by engineers. They give no guid-
level and dilatant response of the sand provides pre-
ance on the confidence the engineer should place in
dictions of the centrifuge database that are very close
his prediction or on what course of action should
(mean of 101%) to the measured values.
be taken if a jack-up installation diverges from the
The authors recommend that the profession con-
installation prediction done before the installation.
tinue to improve the deterministic models with
Taking into account uncertainties probabilistically
applications for buried strong layers, multi-layered
and updating a priori predictions as more informa-
approaches and for silty materials. Each topic requires
tion becomes available, will lead to safer jack-up
further research. Of utmost importance, remains the
installations.
verification of predictions against field measurements.
Industry is encouraged to collate these and make them
available to researchers and practitioners (such as in
van Dijk and Yetginer 2015 at this conference). ACKNOWLEDGMENT
There is so far little evidence that statistical and
probabilistic models are used in jack-up site assess- The authors thank colleagues at the Centre for Off-
ments. It is hoped that the advances described in this shore Foundation Systems (COFS) and the Australian
paper provide inspiration for the inclusion of such Research Council Centre of Excellence for Geotechni-
methods in practice and for further development and cal Science and Engineering, the Norwegian Geotech-
refinement. nical Institute, the University of Oxford and the
A significant product form the probabilistic National University of Singapore for their contribu-
approach is that engineers can calculate likelihood tion to the analysis of spudcan foundations from which
intervals in the load-penetration behaviour that have this paper builds on, particularly Mark Randolph,
quantitative meaning. The intervals were presented Shazzad Hossain, Sam Stanier, Dong Wang, Kok
as probability of exceedance plots in this paper. It is Kuen Lee (now at Advanced Geomechnics-Fugro),
well recognised in practice that uncertainties in spud- Britta Bienen, Youhu Zhang, Christophe Gaudin, Guy
can behaviour exist, and engineers do develop upper Houlsby, Colin Leung and Kar Lu Teh (now at Brae-
and lower estimates for the soil strength. However, mar Technical Services). COFS is supported by the
the consistent framework that allow the incorpora- Lloyd’s Register Foundation as a Centre of Excel-
tion of all significant uncertainties in a systematic lence and currently is one of the primary nodes of the
manner and that the reliability approach provides, Australian Research Council Centre of Excellence for
is not used. Geotechnical Science and Engineering. Lloyd’s Reg-
Applications of the reliability approach were out- ister Foundation invests in science, engineering, and
lined in this paper, including distributed model factors technology for public benefit, worldwide.

210
REFERENCES Hossain, M.S., Hu, Y. & Randolph, M.F. 2003. Spudcan
foundation penetration into uniform clay. Proc. Int. Symp.
Ang, A, H-S. & Tang, W.H. 1975. Probability concepts Offshore and Polar Engineering (ISOPE), Hawaii, USA,
in engineering planning and design. Volume I Basic 647–652.
principles. John Wiley & Sons. 409 p. Hossain, M.S., Hu, Y., Randolph, M.F. & White, D.J. 2005.
Ang, A, H-S. & Tang, W.H. 1984. Probability concepts in Limiting cavity depth for spudcan foundations penetrating
engineering planning and design. Volume II. Decision, clay. Géotechnique 55(9), 679–690.
risk and reliability. John Wiley & Sons. 562 p. Hossain, M.S. & Randolph, M.F. 2009a . New mechanism-
Ang, A, H-S. & Tang, W.H. 2007. Probability concepts in based design approach for spudcan foundations on single
engineering. Emphasis on applications to civil & envi- layer clay. Journal of Geotechnical and Geoenvironmental
ronmental engineering. 2nd ed., John Wiley & Sons. Engineering, ASCE, 135(9), 1264–1274.
406 p. Hossain, M.S. & Randolph, M.F. 2009b . New mechanism-
Baecher, G.B. & Christian, J.T. 2003. Reliability and statistics based design approach for spudcan foundations on stiff-
in geotechnical engineering. Wiley & Sons. 605 p. over-soft clay. Proc. Offshore Technology Conference,
Baglioni, V.P., Chow, G.S. & Endley, S.N. 1982. Jack-up Houston, Paper OTC19907.
rig foundation stability in stratified soil profiles. Proc. Hossain, M.S., Randolph, M.F., Hu, Y. & White, D.J. 2006.
Offshore Technology Conference, Houston, Paper OTC Cavity stability and bearing capacity of spudcan foun-
4408. dations on clay. Proc. Offshore Technology Conference,
Bienen, B., Cassidy, M.J., Randolph, M.F. & Teh, K.L. 2010. Houston, Paper OTC 17770.
Characterisation of undrained shear strengths using sta- Hossain, M.S., Zheng, J., Menzies, D., Meyer, L. & Randolph,
tistical methods. Proc. 2nd Int. Sym. on Frontiers of M.F. 2014. Spudcan penetration analysis for case histories
Offshore Geotechnics. (ISFOG-2010). Perth, Australia, in clay. Journal of Geotechnical and Geoenvironmental
pp. 661–666. Engineering, ASCE, 140(7), 04014034.
Bolton, M.D. 1986. The strength and dilatancy of sands. Houlsby, G.T. 2010. A probabilistic approach to the pre-
Géotechnique, 36(1), 65–78. diction of spudcan penetration of jack-up units. Proc.
Brinch Hansen, J. 1970. A revised and extended formula 2nd Int. Symp. on Frontiers in Offshore Geotechnics:
for bearing capacity. The Danish Geotechnical Institute, ISFOG-2010, Perth.
Copenhagen 98: 5–11. Houlsby, G.T. & Cassidy, M.J. 2002. A plasticity model for
Brown, J.D. & Meyerhof, G.G. 1969. Experimental study the behaviour of footings on sand under combined loading.
of bearing capacity in layered soils, Proc. 7th ICSMFE, Géotechnique, 52(2), 117–129.
Vol. 2. Houlsby, N.M.T. & Houlsby, G.T. 2013. Statistical fit-
Cassidy, M.J. 2012. Experimental observations of the pen- ting of undrained strength data. Géotechnique. 63(14),
etration of spudcan footings in silt. Géotechnique 62(8) 1253–1263.
727–732. Houlsby, G.T. & Martin, C.M. 2003. Undrained bearing
Cassidy, M.J. & Houlsby, G.T. 1999. On the modelling of capacity factors for conical footings on clay. Géotechnique
foundations for jack-up units on sand, Proc. 31st Offshore 53(5), 513–520.
Technology Conference, Houston, OTC 10995. Hu, P. 2015. Predicting punch-through failure of a spudcan
Cassidy, M.J. & Houlsby, G.T. 2002. Vertical bearing capacity on sand overlying clay. PhD Thesis, University of Western
factors for conical footings on sand. Géotechnique, 52(9), Australia, Australia.
687–692. Hu, P., Stanier, S., Cassidy, M.J. & Wang, D. 2014a .
Castleberry II, J.P. & Prebaharan, N. 1985. Clay crusts of the Predicting peak resistance of spudcan penetrating
Sunda Shelf - a hazard to jack-up operations. Proc. 8th sand overlying clay. Journal of Geotechnical and
SoutheastAsian Geotechnical Conference, Kuala Lumpur, Geoenvironmental Engineering, ASCE, 140(2), DOI:
40–48. 10.1061/(ASCE)GT.1943-5606.0001016.
Dean, E.T.R. 2010. Offshore geotechnical engineering. Hu, P., Stanier, S.A., Wang, D. & Cassidy, M.J. 2014b . The
Thomas Telford. effect of conical shape on a footing penetrating sand over-
Det Norske Veritas (DNV). 2012. Statistical representation lying clay. Journal of Geotechnical and Geoenvironmental
of soil data. Recommended Practice DNV-RP-C207. Jan. Engineering, ASCE, Under review.
2012. Hu, P., Wang, D., Cassidy, M.J. & Stanier, S.A. 2014c .
Edwards, D.H. & Potts, D.M. 2004. The bearing capacity of a Predicting the resistance profile of a spudcan on sand
circular footing under ‘punch through’ failure. Proc. Int. overlying clay, Canadian Geotechnical Journal, 51(10),
Symp. Num. Models in Geomech., (NUMOG), Ottawa, 1151–1164, DOI: 10.1139/cgj-2013-0374.
493–498. Hu, P., Wang, D., Stanier, S.A. & Cassidy, M.J. 2014d .
Einav, I. & Randolph, M.F. 2005. Combining upper bound Assessing the punch-through hazard of a spudcan on sand
and strain path methods for evaluating penetration resis- overlying clay, Géotechnique, Under review.
tance. International Journal of Numerical Methods in ISO. 2012. Petroleum and natural gas industries: Site-
Engineering, 63(14), 1991–2016. specific assessment of mobile offshore unit. 1: jack-ups.
Erbrich, C.T. 2005. Australian frontiers – spudcans on the ISO 19905-1, ISO/FDIS, Geneva.
edge. Proc. Int. Symp. on Frontiers in Offshore Geotech- ISO DIS 2394. 2015. General principles on reliability for
nics (ISFOG), Perth, pp. 49–74. structures. International Standard. Draft.
Fenton, G.A. & Griffiths, D.V. 2008. Risk assessment in Keaveny, J.M., Nadim, F. & Lacasse, S. 1990.Autocorrelation
geotechnical engineering. Wiley and Sons. 461p. functions for offshore geotechnical data. Proc. ICOSSAR
Hanna, A.M. & Meyerhof, G.G. 1980. Design charts for ulti- 1990. International Conference on Structural Safety and
mate bearing capacity of foundations on sand overlaying Reliability. Perth, Australia, 263–270.
soft clay. Canadian Geotechnical Journal. 1: 300–303. Lacasse, S., Guttormsen, T., Nadim, F., Rahim, A. & Lunne,
Higham M.D. 1984. Models of jack-up rig foundations. MSc T. 2007. Use of statistical methods for selecting design soil
Thesis. Manchester University. parameters. Proc. 6th Int. Offshore Site Investigation and

211
Geotechnics Conference: Confronting new challenges and Paisley, J.M. & Chan, N. 2006. SE Asia jack-up punch-
sharing knowledge, London, UK, 449–460. throughs: technical guidance note on site assessment.
Lacasse, S. & Nadim, F. 1996. Uncertainties in characterizing Proc. Jack-up Asia Conference and Exhibition, Singapore.
soil properties. Uncertainty in the Geologic Environment: Phoon, K.K. & Kulhawy, F.H. 1999. Characterization of
From Theory to Practice, ASCE Geotechnical Special geotechnical variability. Canadian Geotechnical Journal,
Publication, 58: 49–75. 36: 612–624.
Lacasse, S., Nadim, F., Andersen, K.H., Knudsen, S., Eidsvig, Randolph, M.F. & Gourvenec, S. 2011. Offshore Geotech-
U.K., Yetginer, G., Guttormsen, T.R. & Eide, A. 2013. nical Engineering. CRC Press/ Taylor & Francis. ISBN:
Reliability of API, NGI, ICP and Fugro axial pile capac- 978-0-415-47744-4.
ity calculation methods. Offshore Technology Conference, Randolph, M.F., Jamiolkowski, M.B. & Zdravkovi, L. 2004.
OTC-24063-MS. Houston, Texas. Load carrying capacity of foundations. Proc. Skempton
Lee, K.K. 2009. Investigation of potential spudcan punch- Memorial Conf., London, 1: 207–240.
through failure on sand overlying clay soils. PhD thesis, Skempton, A.W. 1951. The bearing capacity of clays. Build-
University of Western Australia, Perth, Australia. ing Research Congress, London, 1, 180–189.
Lee, K.K., Cassidy, M.J. & Randolph, M.F. 2013a . Bearing SNAME 2008. Guidelines for site specific assessment of
capacity on sand overlying clay soils: Experimental and mobile jack-up units, Society of Naval Architects and
finite element investigation of potential punch-through Marine Engineers.Technical and Research Bulletin 5–5A,
failure, Géotechnique, 63(15), 1271–1284. New Jersey.
Lee, K.K., Randolph, M.F. & Cassidy, M.J. 2013b . Bear- Teh, K.L., Cassidy, M.J., Leung, C.F., Chow, Y.K., Randolph,
ing capacity on sand overlying clay soils: A simplified M.F. & Quah, C.K. 2008. Revealing the bearing fail-
conceptual model, Géotechnique, 63(15), 1285–1297. ure mechanisms of a penetrating spudcan through sand
Li, J.H., Huang, J., Cassidy, M.J. & Kelly, R. 2014. Spa- overlaying clay. Géotechnique. 58(10), 793–804.
tial variability of the soil at the Ballina national field test Teh, K.L., Leung, C.F., Chow, Y.K. & Cassidy, M.J. 2010.
facility. Australian Geomechanics, 49(4), 41–48. Centrifuge model study of spudcan penetration in sand
Li, J.H., Uzielli, M. & Cassidy, M.J. 2015. Uncertainty-based overlying clay. Géotechnique, 60(11), 825–842.
characterization of Piezocone and T-bar data for the Lam- Uzielli, M., Cassidy, M.J. & Hossain, M.S. 2015. Bayesian
inaria offshore site. Proc. 3rd Int. Sym. on Frontiers of prediction of punch-through probability for spudcans in
Offshore Geotechnics. (ISFOG-2015). Oslo, Norway. stiff-over-soft clay. To appear in the Wilson Tang Special
Martin, C.M. 1994. Physical and numerical modelling of off- Edition, GeoInstitute, ASCE.
shore foundations under combined loads. D.Phil. Thesis, van Dijk, B.F.J. & Yetginer, A.G. 2015. Findings of
University of Oxford. the ISSMGE jack-up leg penetration prediction event.
Menzies, D. & Roper, R. 2008. Comparison of jackup rig Proc. 3rd Int. Sym. on Frontiers of Offshore Geotechnics.
spudcan penetration methods in clay. Proc. 40th Offshore (ISFOG-2015). Oslo, Norway.
Technology Conference, Houston. Vanmarcke, E.H. 1977. Probabilistic modeling of soil
Meyerhof, G.G. & Hanna, A.M. 1978. Ultimate bearing profiles. J. Geotechnical Engineering Division ASCE
capacity of foundations of layered soils under inclined 103(11): 1227–1246.
loads. Canadian Geotechnical Journal, 15, 565–572. Vanmarcke, E.H. 1983. Random Fields: analysis and synthe-
Nadim, F. 1988. Geotechnical site description using Stochas- sis. Cambridge: MIT Press. 382p.
tic Interpolation. NGM.1988. 10th NGM (“Nordiske Wong, P.C., Templeton, J.S., Purwana, O.A., Hofstede, H.,
Geoteknikermøte”). Oslo. 158–161. Cassidy, M.J., Hossain, M.S. & Martin, C.M. 2012. Foun-
Osborne, J.J. 2007. Unpredicted jack-up foundation perfor- dation modeling and assessment in the new ISO standard
mance, PetroMin, Singapore, pp. 42–51. 19905-1. Proc. 44th Offshore Technology Conference,
Osborne, J.J., Houlsby, G.T., Teh, K.L., Bienen, B., Cas- Houston, USA, OTC23521.
sidy, M.J., Randolph, M.F. & Leung, C.F. 2009. Improved Young, A.G. & Focht, J.A. 1981. Subsurface hazards affect
guidelines for the prediction of geotechnical performance mobile jack-up rig operations. Sounding, McClelland
of spudcan foundations during installation and removal of Engineers Inc., Houston, 3(2): 4–9.
jack-up units. Proc. 41st Offshore Technology Conference, Zhang, L.M., Tang, W.H., Zhang, L.L. & Zheng, J.G. 2004.
Houston, OTC-20291. Reducing uncertainty of prediction from empirical corre-
Osborne, J.J. & Paisley, J.M. 2002. SE Asia jack-up punch- lations. Journal of Geotechnical and Geoenvironmental
throughs: The way forward? Proc. Int. Conf. on Offshore Engineering, ASCE, 130(5), 526–534.
Site Investigation and Geotechnics – Sustainability and Zhang, L.L., Tang, W.H. & Zhang, L.M. 2009a . Bayesian
Diversity. London, 301–306. model calibration using geotechnical centrifuge tests.
Osborne, J.J., Teh, K.L., Houlsby, G.T., Cassidy, M.J., Journal of Geotechnical and Geoenvironmental Engineer-
Bienen, B. & Leung, C.F. 2010. Improved guidelines ing, ASCE, 135(2), 291–299.
for the prediction of geotechnical performance of spud- Zhang, J., Zhang, L.M. & Tang, W.H. 2009b . Bayesian
can foundations during installation and removal of framework for characterizing geotechnical model uncer-
jack-up units. RPS Energy Report Number EOG0574- tainty. Journal of Geotechnical and Geoenvironmental
Rev1. Final Guidelines of the InSafe Joint Indus- Engineering, ASCE, 135(7), 932–940.
try Project 124p. Now available to all industry at
http://insafe.woking.rpsplc.co.uk/Default.asp.
Osborne, J.J., Teh, K.L., Houlsby, G.T., Cassidy, M.J., Ran-
dolph, M.F., Bienen, B., Hossain, M.S. & Leung, C.F.
2009. InSafe Joint Industry Project. FirstYear Report. RPS
Energy Report.

212

View publication stats

Você também pode gostar