Beruflich Dokumente
Kultur Dokumente
Reliability of API, NGI, ICP and Fugro Axial Pile Capacity Calculation Methods
S. Lacasse, F. Nadim, K. H. Andersen, Siren Knudsen, Unni K. Eidsvig, Norwegian Geotechnical Institute (NGI),
Gülin Yetginer, Tom R. Guttormsen and Asle Eide, Statoil.
This paper was prepared for presentation at the Offshore Technology Conference held in Houston, Texas, USA, 6–9 May 2013.
This paper was selected for presentation by an OTC program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been
reviewed by the Offshore Technology Conference and are subject to correction by the author(s). The material does not necessarily reflect any position of the Offshore Technology Conference, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Offshore Technology Conference is prohibited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of OTC copyright.
Abstract
The API RP2A (RP2GEO) and ISO 19902 guidelines include four CPT-methods for calculating the axial capacity of piles in
sands. The guidelines require that if newer methods are to be implemented in design, the same level of safety shall be docu-
mented for these methods as for existing methods. The designer is required to select an appropriate safety factor when using
the newer design methods. The challenge lies in deciding which safety factor will ensure a consistent safety level for different
soil conditions and pile dimensions. To evaluate the required material factor, the probability of failure was quantified in two
case studies for piles designed with the API method and with the newer NGI, ICP and Fugro methods. A calibration of the
required material factor for a target probability of failure of 10-4/yr was also performed. The results show that the annual relia-
bility index and probability of failure vary with the axial pile capacity calculation method. The study provides a contribution to
the discussion on the reliability of the API, the NGI, the ICP and the Fugro methods. The material factor needs to be associated
with the characteristic soil parameters selected for design. A large number of case studies should be added to quantify the reli-
ability and the required material factor for each pile capacity method. The findings on margin of safety and the definition of the
characteristic shear strength have important implications for the design of piles offshore and can result in significant savings.
Introduction
Ensuring adequate reliability under severe loading conditions is a necessary consideration for offshore platforms, and the safe-
ty margin depends on the uncertainty in the parameters entering the analyses, in addition to the model uncertainty. The design
engineer attempts to compensate for the uncertainties by introducing an appropriate "factor of safety" in design. There will
always be a finite probability that the forces of the environment can cause damage, or the total collapse, of an offshore struc-
ture. Defining the level of finite probability that is tolerable is the challenge.
The API RP2A (RP2GEO) (1) and ISO 19902 (2) guidelines included four CPT-methods for calculating the axial capacity
of piles in sands in 2007. The designer is required to select an appropriate safety factor when using the newer design methods.
The difficulty lies in deciding which safety factor will ensure a consistent safety level for different soil conditions and pile
dimensions. The axial capacity of tubular steel piles for offshore installations is frequently based on the Recommended Prac-
tice of the American Petroleum Institute (API) (1). To evaluate the required material factor, the probability of failure was
quantified for piles designed with the API method and with the NGI, ICP and Fugro methods. A calibration of the required
material factors for a target probability of failure of 10-4/yr was also carried out. The paper presents the results of the reliability
analyses and the calibration of the material factor for each axial pile capacity method. The approach is illustrated with two case
studies of jacket on piles, the first on a mainly clay profile and the second on a very dense sand profile.
Scope of Study
The study was undertaken to document that the pile foundations were designed according to governing regulations. The goal
was to make a recommendation on the appropriate material factor and minimum pile length to use for the design of the piles on
an offshore jacket. The reliability study of the axial capacity of the piles considered two jackets on two very different soil pro-
files. For each of the case studies, the assessment included:
Statistical evaluations of the soil and load parameters,
Statistical analyses of the model uncertainty associated with each of the pile capacity methods,
Deterministic and probabilistic analyses of the axial pile capacity with different pile capacity methods to determine
the annual reliability index and the annual probability of failure for each axial pile capacity calculation method, and
2 OTC 24063
Calibration of the required material factor for a target annual probability of failure.
Table 1 lists the axial pile capacity methods considered: the API method, the NGI-05 method, the ICP-05 method and the
Fugro-96/05 method. For the ICP method in sand, the simplified method relevant for large diameter piles was used as
described in API and ISO (1; 2). Table 2 lists the soil parameters required for each method. The unit weight of the soil is also
required to calculate the total or effective stress versus depth.
Table 2. Required soil parameters for each pile capacity method considered in the study.
Soil type Sand Clay
Cone Relative Effective internal Effective interface Undrained Cone Plasticity Overconsoli- Sensi-
resistance density friction angle friction angle shear strength resistance index dation ratio tivity
Method qc Dr ' suUU qc Ip OCR(or pc') St
API (✓) ✓ (✓) ✓ ✓
NGI-05 ✓ (✓) ✓ ✓
ICP-05 ✓ (✓) ✓ (or d50) (✓) ✓ ✓ ✓
Fugro-96/05 ✓ ✓
Parenthesis indicates that the parameter is derived from another parameter.
d50 = Diameter of particles at 50% passing on grain size distribution curve.
suUU = Undrained shear strength from unconsolidated undrained (UU) tests.
for design, because the effect of cyclic loading is normally more severe in tension than in compression. Screening analyses
showed that for the two case studies in this paper, the compression piles were critical for design. The effect of cyclic loading
was small for Case Study A and negligible for Case Study B.
The effect of time should also be considered, and a decision should be made on whether to include this effect or not. This
effect can be important, especially for piles in sands on the short term. The effect of time would be the same for each of the
calculation methods used.
Figure 1. Selected characteristic undrained shear strength profile Figure 2. Mean and standard deviation of undrained shear strength
for the deterministic design of piles, Case Study A. compared with characteristic untrained shear strength.
PDF of Soil Parameters. The probability density functions (PDF) are listed at the top of the columns for each soil parame-
ter in Table 3. The PDFs were selected as either normal (N), lognormal (LN) or beta (), based on a study of the best fit of the
data. To ensure that the parameters would not become negative in the probabilistic analyses, parameters like undrained shear
strength and sensitivity were given a lognormal PDF. The best fit of the data was done with the commercial program EasyFit,
produced by Mathwave (13). In many layers however, there were too few data to establish a good fit of the data.
Key Soil Parameters. Figures 2 to 4 illustrate key results of the statistical analyses. Figure 2 compares the mean and stand-
ard deviation of the undrained shear strength with the characteristic shear strength (p'o is the in situ effective vertical stress).
Figure 3 compares the mean and standard deviation of the undrained shear strength with the results of piezocone tests, where
the qc measurements were converted to the triaxial compression undrained shear strength (suC), and not the UU-strength (suUU).
Figure 4 presents the mean and standard deviation of the plasticity index.
Table 3. Input parameters for clay and sand used in probabilistic analyses of axial pile capacity, Case Study A.
Depth Submerged unit Undrained shear CPTU Presonsolidation
Plasticity index, Ip
to weight, strength, Top-bottom Top-bottom (Lower–Upper) Sensitivity, St
Soil Soil (%)
bottom (kN/m3) suUU ( kPa) qc (MPa) p (kPa)
Unit type
of soil Mean; CoV (%) Mean; CoV (%) Mean; CoV Mean (range) Mean; CoV (%) Mean; CoV (%)
unit, z PDF=N PDF=LN (%) PDF=N PDF= PDF=LN PDF=LN
I 1.0 SAND 10.0; 6% - 1 – 5; 20 % - - -
II 3.6 CLAY 11.3; 4% 38 – 108; 30 – 40 % 0.7 – 2 300; (200 - 400) 1.7-1.9; 20-40% 13.4; 25 %
IIIa 18.6 CLAY 11.2; 4% 108 – 172; 25 – 30 % 2–3 300; (200 - 400) 1.5-1.6; 20-30% 18.7 – 22.8; 10 %
IIIb 24.1 SAND 10.7; 7% - 70 – 70: 20 % - - -
IIIc 32.3 CLAY 11.4; 4% 192; 10 % 3–3 300; (200 - 400) 2.1; 15 – 20 % 19.6; 10 %
IVa 38.6 CLAY 10.5; 3% 342; 25 – 30 % 6–6 550; (400 - 700) 1.5; 20 – 30 % 33.2; 20 %
IVb 50.0 CLAY 10.7; 5% 342; 25 – 30 % 6–6 550; (400 - 700) 1.5; 20 – 30 % 17.0; 40 %
IVc 56.3 CLAY 11.2; 3% 492; 35 % 40 – 15 550; (400 - 700) 1.5; 20 – 30 % 13.3; 40 %
IVd 69.5 CLAY 9.7; 6% 288 – 240; 25 % 7.5 – 7.5 550; (400 - 700) 1.5: 20 – 30 % 30.7; 10 %
IVe 91.9 CLAY 11.2; 5% 337; 25 – 30 % 8–8 550; (400 - 700) 1.5: 20 – 30 % 19.3; 20 %
IVf 96.0 SAND 11.5; 4% - 50 – 50 - - -
V 102.0 CLAY 11.7; 2% - 8 – 11 550; (400 - 700) - 20
PDF = Probability density function: N=normal; LN= lognormal; = beta
Depth below seafloor, m
Figure 3. Mean and standard deviation of the undrained shear Figure 4. Mean and standard deviation of the plasticity index
strength compared with suC from CPTU tests. compared with values used for preliminary design.
OTC 24063 5
Other parameters are also required for the calculation of axial pile capacity. These include specific parameters for the sand
layers, such as the coefficient of horizontal stress (K), the friction angle and the API limiting skin friction. Bias and coefficient
of variation were obtained from the survey done of 40 offshore pile specialists summarized by Lacasse and GOULOIS (14). A
summary of these additional parameters is given in Table 4. The values of the friction angle in the sand are based on the rec-
ommendations by the current API-guidelines. For very dense sand, the friction angle would be expected to be at least 45 °, but
the API guidelines were followed in the reliability study. The values of relative density were obtained from the procedure in
PACER (15) to obtain relative density from CPTU tests. Values of Dr greater than 100% were derived. More discussion of this
aspect can be found under Case Study B.
Scour and gapping. Gapping at the top of the pile may occur under horizontal loading. In the deterministic analysis, the
axial pile capacity in the top 1.7 m was taken as zero. A study of the gapping was not included in the reliability analyses. In
principle, one should analyze the possibility of loss in capacity due to gapping or scour at the top of the pile.
Dashed curves are the
shifted exponential distri‐
bution fitted to the data
on the full curves. The
fitting was governed by a
best fit for return periods
greater than 100 years.
a) Fitted Gumbel distributions for environmental loads (ref 16). b) Storm-induced loads with shifted exponential distribution.
Figure 5. Statistical representation of environmental loads for Case Study A
method, where the original data were not available and the uncertainty was estimated from the results published by the authors
(7). The axial pile capacity methods were used as described by the references in Table 1. No bias to account for the effects of
plasticity (for the API methods) or increasing pile diameter (ICP and Fugro methods) was included in this paper.
Most effort for Case Study A was spent on the model uncertainty for the prediction of skin friction, since the friction capac-
ity was the most important component of the pile capacity. The uncertainty in the end bearing was taken as a large value and
the same for all methods for the study of the reliability of the methods. For Case Study A, introducing a relatively large uncer-
tainty in the end bearing model did not have a significant effect on the annual probability of failure.
Table 5. Model uncertainty Qm/Qc used for each pile capacity method in reliability analyses, Case Study A
Model uncertainty, Qm/Qc, in clay layers Model uncertainty, Qm/Qc, in sand layers
Axial pile capacity method
Skin friction End bearing Skin friction
Bias = 1.09; SD = 0.26 Bias = 1.0; SD = 0.30 Bias = 1.58; SD = 0.66
API
CoV = 0.24; PDF = LN or N CoV = 0.30; PDF = LN or N CoV = 0.42; PDF = LN or N
NGI-05 Bias = 1.06; SD = 0.17 Bias = 1.0: SD = 0.30 Bias = 1.05; SD = 0.20
CoV = 0.16; PDF = LN or N CoV = 0.30; PDF = LN or N CoV = 0.19; PDF = LN or N
ICP- 05 Bias = 1.05; SD = 0.31 Bias = 1.0; SD = 0.30 Bias = 1.13; SD = 0.21
CoV = 0.30; PDF = LN or N CoV = 0.30; PDF = LN or N CoV = 0.18; PDF = LN or N
Fugro-96/05 method Bias = 1.01; SD = 0.17 Bias = 1.0; SD = 0.30 Bias = 1.00; SD = 0.28
CoV = 0.17; PDF = LN or N CoV = 0.30; PDF = LN or N CoV = 0.28; PDF = LN or N
Bias Qm/Qc > 1.0 means that method under-predicts measured capacity in pile load tests
Bias Qm/Qc < 1.0 means that method over-predicts measured capacity in pile load tests
SD = Standard deviation CoV = Coefficient of variation LN or N = Lognormal or normal PDF
Probabilistic Analyses
The annual probability of pile foundation failure was estimated using a two-stage approach. In the first stage, the uncertain-
ty in the ultimate axial pile capacity was evaluated and quantified by a probability distribution function. The NGI software
RELPAX (20) was used. In the second stage, the results of the first stage were combined with the probabilistic description of
the maximum annual axial load on the pile to calculate the annual probability of failure. The commercial software COMREL
(21) was used. Appendix C describes the approach in more detail. Table 7 presents a selection of the probabilistic RELPAX
and COMREL analyses done.
Table 7.Results of probabilistic analyses of axial capacity for 90 m pile, Case Study A.
Ultimate pile capacity Deterministic Scaling factor Calculated Relative contribution to Pf
Annual
Method Mean ± SD (CoV) capacity * on capacity load factor Soil resistance
No Pf and Loads
PDF = LN (MN) for Pf=10-4/yr for Pf=10-4/yr (Model uncertainty)
Pf = 2.010-7 49 %
I API 151.7 ± 31.1 (21 %) 138.6 0.67 1.116 51 %
= 5.0 (40 %)
Pf = 1.310-6 33 %
II NGI-05 120.5 ± 20.0 (17 %) 118.6 0.76 1.259 67 %
= 4.7 (24 %)
Pf = 3.910-7 37 %
III ICP-05 136.8 ± 25.4 (19 %) 136.8 0.705 1.186 63 %
= 4.9 (30 %)
Fugro- Pf = 2.410-4 32 %
IV 114.6 ± 30.8 (27 %) 136.1 1.07 0.926 68 %
96/05 = 3.5 (24 %)
* With mean soil parameters
OTC 24063 7
Figure 6. Ultimate axial pile capacity in compression with the Figure 7. Ultimate axial pile capacity in compression with the
characteristic undrained shear strength, Case Study A. mean undrained shear strength at Case Study A.
Axial Pile Capacity and Annual Probability of Failure. The ultimate axial pile capacity for a 90-m long pile, with a
Lognormal distribution, had a coefficient of variation between 16 and 20 % for the API, the NGI-05 and the ICP-05 methods.
The annual probability of failure was between 10-6 and10-7. The coefficient of variations for the Fugro methods was closer to
30 %, and the probability of failure was significantly higher.
Table 8 lists the axial pile capacity and annual probability of failure as a function of pile length between 75 and 90 m. The
numbers are for the NGI-05 method. The calculations show that a 75-m long pile had an annual probability of failure of about
10-5. The table shows only small changes in the annual probability of failure when the second-order reliability method (SORM)
was used instead of the first-order reliability method (FORM) (Appendix C) and when a normal PDF for the axial pile capacity
was used instead of the lognormal PDF.
Table 8. Axial pile capacity and annual probability of failure as a function of pile length, NGI-05 method, Case Study A.
Depth (m) Qult char (MN) Qult mean (MN) Annual Pf - FORM, QLN Annual Pf - SORM, QLN Annual Pf - FORM, QN
75 77.8 96.6 2.110-5 2.110-5 5.210-5
80 84.1 103.9 2.310-5 2.510-5 2.310-5
90 96.8 118.6 1.210-6 1.310-5 1.310-5
In the COMREL analyses (21), a scaling factor was established to calculate the probabilistic foundation capacity required
for an annual failure probability of 10-4. The scaling factors shown in Table 7 represent the factor by which one needs to multi-
ply the axial pile capacity (the mean or the characteristic capacity) to bring the probability of failure to 10-4/yr. The environ-
mental load in the probabilistic analyses giving an annual failure probability of 10-4 related to the 100-year deterministic envi-
ronmental load was also found through a scaling factor. These scaling factors are used for the calibration of the material factor.
Influence factors. The FORM analyses with RELPAX and COMREL provided sensitivity factors for each random varia-
ble. The sensitivity factors describe the relative contribution of the random variables to the total uncertainty. The influence
factors of the uncertainties in the loads and the uncertainties in the soil resistance on the probability of failure, Pf, are given in
Table 7. For the NGI-05 method, the uncertainties in the loads influence the probability of failure with 67 % weight, while the
uncertainties in the soil parameters and pile capacity calculation model influence the probability of failure with 33% weight,
whereof 24 % was due to the uncertainty the capacity calculation method. Model uncertainty in the capacity calculation meth-
ods, mainly the skin friction, was important for all methods, contributing more than 50 % of the uncertainty in the capacity,
thereby contributing to increasing the probability of failure.
Observations. The results of the probabilistic analyses indicate that (1) the deterministic capacity calculated by RELPAX
with the mean values of the soil parameters were in general quite close to the probabilistic mean for the NGI-05, ICP-05 and
Fugro-96/05 methods; (2) the model uncertainty for each of the calculation methods was by far the most important contributor
to the uncertainty in the capacity; and (3) the probabilistic distribution of the axial pile capacity was best modeled with a
lognormal distribution.
8 OTC 24063
The characteristic shear strength in the sand, based on the measured cone resistance from CPTU tests, was selected as a
carefully assessed value as required by NORSOK (11). Figure 9 presents the soil layering used and characteristic cone re-
sistance profile. For piles loaded in compression, a penetration depth of 51 m was originally required to ensure the required
bearing capacity. The NGI-05 method, giving in this case the lowest capacity among the CPT-methods, was used for this eval-
uation of pile length.
Table 10. Input parameters for sand and clay in the probabilistic analyses of axial pile capacity, Case Study B.
Depth Total unit weight, Undrained shear strength qc (top-bottom) Relative density
Plasticity index, Ip (%)
Soil to top Soil (kN/m3) (top-bottom), suUU ( kPa) (MPa) Dr (%)
Unit of unit type Mean ± SD Mean; CoV (%) Mean; CoV (%) Mean; CoV(%) Mean; CoV (%)
(m) (CoV=2-7 %); PDF=N PDF=LN PDF = LN PDF = N PDF=LN
I 0 SAND 18.6 ± 0.47 - 0 – 4.5; 17 % 0.94; 7 % -
II 0.3 SAND 18.6 ± 0.47 - 4.5 – 12.5; 17 % 1.15; 6 % -
II 0.8 SAND 18.6 ± 0.47 - 12.6 – 52.9; 13 % 1.25; 4 % -
IIIb 8.7 SAND 18.8 ± 0.51 - 74.2; 25 % 1.36; 7 % -
IIIb 21.7 CLAY 21.2 ± 0.47 415-478; 6.5 % 10; 68 % - 16.3; 15 %
IVb 22.2 SAND 19.0 ± 0.64 - 60.6 – 94.8; 25 % 1.28; 8 % -
IVb 25.0 SAND 19.0 ± 0.64 - 112.9 - 83.3; 32 % 1.37; 9 % -
IVb 33.2 SAND 19.0 ± 0.64 - 67.2; 43 %- 1.15; 15% -
IVb 34.9 CLAY 20.2 ± 0.24 277-294; 34 % 4 – 6.6; 33 % - 29.5; 3 %
IVb 36.1 SAND 19.0 ± 0.64 - 20.6 – 51.0; 42 % 0.88; 19 % -
IVb 38.5 SAND 19.0 ± 0.64 - 65.3; 27 % 1.09; 10 % -
Iva 46.2 CLAY 22.2 ± 1.6 324-1471; 38 % 7.4 – 30.4; 49 % - 14; 19 %
IVb 49.4 SAND 19.0 ± 0.64 - 112.9 - 90.5; 34 % 1.27; 11 % -
Preconsolidation stress (p’) is 1800 kPa in clay layer IIIb, 1000 kPa in clay layer IVb and 3300 kPa in clay layer IVa.
Sensitivity in clay layers: mean =1.5, CoV = 25 %; su in clay from qc trend using = 20.0 kN/m3 and N = 20. su ~ (qc – v0)/N.
PDF = Probability density function: N = normal; LN = lognormal
OTC 24063 9
Figure 8. Cone resistance versus depth, Case Study B. Figure 9. Mean and characteristic cone resistance, Case Study B
Relative density, Dr , %
0 25 50 75 100 125 150
0 0.0
0.2
8.7
10
15
20
21.4
22.5
Depth below seafloor, m
25
30
35 34.8
36.1
40
45
46.2
49.3
50
55
Characteristic Dr
Mean Dr
60 60.0
Figure 10. Undrained shear strength of clay layers, Case Study B Figure 11. Relative density implied by the NGI 05 method for the
deterministic and statistical qc profiles, Case Study B.
Figure 10 gives the average shear strength in the "clay" layers. The clay did not contribute significant resistance to the ul-
timate pile capacity. The plasticity index was assumed to be constant with depth within each clay layer. Other parameters re-
quired for the calculation of axial pile capacity in the clay layers were the difference in effective stress between the preconsoli-
dation stress and the overburden stress, Δp', and the sensitivity, St. Standard deviation and coefficient of variation were esti-
mated on the basis of experience and engineering judgment, and earlier published results (22; 23).
10 OTC 24063
Relative Density Dr.The relative density of the sand, Dr, was estimated from the cone resistance (8):
0.4 ln .
22 ′ ∙
where qc is the cone resistance, σ'vo the vertical effective stress and σatm the atmospheric reference pressure (100 kPa). Using
the First Order Second Moment (FOSM), and neglecting the uncertainty in the vertical effective stress, the relationship be-
tween the standard deviation (SD) of Dr and the coefficient of variation (CoV) of qc was SD (Dr) = 0.4·CoV(qc). The choice of
a lognormal distribution for qc implies a normal distribution for Dr, due to the logarithmic relationship between Dr and qc.
The resulting relative density of the sand, used by the NGI-05 and API methods, was in some layers as high as 135 %. Fig-
ure 11 presents the inferred relative density with depth, for both the characteristic and the mean profiles, and the measured
cone resistance. The very high relative densities are directly related to the very high values of cone resistance measured. It is
not unusual that the relative density in situ is higher than what can be achieved in the laboratory, especially considering the
ASTM and British Standard laboratory techniques that were used to establish the correlation among qc-Dr and effective stress.
The above expression is a compromise between the two diagrams recommended in Lunne et al (24), which were based on
work done by Baldi et al (25). Relative densities greater than 100 % are not unrealistic (8). Laboratory testing techniques used
at NGI can give relative densities that are 20 percentage points higher than the standard ASTM and British standards methods.
Relative densities measured in a recovered tube from the Frigg field have also shown relative densities that were higher than
what could be achieved in the laboratory. Although the calculated Dr may be correct with respect to the reliability of the empir-
ical NGI method to predict axial pile capacity, sensitivity analyses were made and it was decided to limit the maximum rela-
tive density to 120 %.
Evaluation of other parameters for sand. As for Case Study A, other parameters are required for the calculation of axial
pile capacity. The type of required parameters depend on the method of axial pile capacity calculation used. These additional
parameters include specific parameters for the sand layers, such as the coefficient of horizontal stress (K), the friction angle
('), and the API specified limit on the skin friction (flim). The mean and coefficient of variation were obtained from the survey
done of 40 offshore pile specialists summarized by (14). The values from this survey for very dense sand, in terms of mean and
coefficient of variation (CoV), are presented in Table 11. The total unit weight was assumed to be constant with depth within
each layer. The values of the friction angle in the sand were based on the recommendations by the API-guidelines (9). For very
dense sand, the friction angle would be expected to be 45° or even more, but the API guidelines were followed in the reliability
study of the API methods.
Table 11. Mean and CoV for additional parameters in sand in pile capacity analyses.
Unit ' (°) Flim (kPa) K δwall (°)
Mean CoV Mean CoV Mean CoV Mean CoV
I 40 15% 115 15% 0.8 25% 28 15%
II 40 15% 115 15% 0.8 25% 28 15%
IIIb 40 15% 115 15% 0.8 25% 28.8 15%
IVb_i 40 15% 115 15% 0.8 25% 28.8 15%
IVb_ii 40 15% 115 15% 0.8 25% 28.8 15%
IVb_iii 40 15% 115 15% 0.8 25% 28.8 15%
' = friction angle Flim = limiting skin friction δwall = effective stress shaft friction angle
K = coefficient of horizontal stress Dr = relative density
Figure 12. Fitted extreme value Gumbel distributions for annual maximum load, Case Study B.
OTC 24063 11
Table 12. Model uncertainty Qm/Qc used for each pile capacity method in reliability analyses, Case Study B.
Model uncertainty, Qm/Qc, in sand layers Model uncertainty, Qm/Qc, in clay layers
Axial pile capacity method
Skin friction End bearing Skin friction
Bias = 1.27; SD = 0.81 Bias = 1.60; SD = 0.91 Bias = 1.09; SD = 0.26
API
CoV = 0.64; PDF = LN or N CoV = 0.57; PDF = LN or N CoV = 0.24; PDF = LN or N
NGI 05 Bias = 1.09; SD = 0.26 Bias = 1.02: SD = 0.17 Bias = 1.06; SD = 0.17
CoV = 0.23; PDF = LN or N CoV = 0.17; PDF = LN or N CoV = 0.16; PDF = LN or N
ICP 05 Bias = 1.08; SD = 0.26 Bias = 1.16; SD = 0.25 Bias = 0.99; SD = 0.21
CoV = 0.25; PDF = LN or N CoV = 0.24; PDF = LN or N CoV = 0.21; PDF = LN or N
Fugro-96/05 method Bias = 1.25; SD = 0.375 Bias = 0.90; SD = 0.225 Bias = 1.01; SD = 0.21
CoV = 0.30; PDF = LN or N CoV = 0.25; PDF = LN or N CoV = 0.21; PDF = LN or N
Bias Qm/Qc > 1.0 means that method under-predicts measured capacity in pile load tests
Bias Qm/Qc < 1.0 means that method over-predicts measured capacity in pile load tests
SD = Standard deviation CoV = Coefficient of variation LN or N = Lognormal or normal PDF
With the selected characteristic parameters and the NGI-05 axial pile capacity calculation method, a pile penetration depth
of 51 m was required for a material factor of 1.5, whereas a pile penetration depth of 39 m was required for a material factor of
1.3. With the ICP-05 method, a pile penetration depth of 26 m was required for material factors of 1.5 and 1.3. This unusual
effect is due to the particular profile and the application of API’s transition rules between layers. Large decreases in capacities
were associated with the presence of very thin layers of clay. The required pile penetration depths can be from 26 m to 51 m,
depending on the method and material factor used. For the API method, the limit end bearing, qlim, was not reached until the
pile penetrated 54.2 m depth.
Effect of pile tip embedment and punch-through. The plugged pile tip resistance values computed by the axial pile capaci-
ty calculation methods include reduction factors to account for the embedment and punch-through, as recommended by (1; 2).
Figure 15 shows the interpretation of the API recommendations for the reduction in pile tip bearing in a sand layer when
the pile tip is near the top or the bottom of a sand layer. With large diameter piles and thin weak layers, the effect of punch-
through shown on Figure 15 is believed to be conservative, as the punching hypothesis assumes that a failure surface can be
mobilized in the weaker layer. For Case Study B, the layers at 22 and 35-m depths are too thin to allow such surface to mobi-
lize. The pile will have already found bearing in the next competent sand layer before the soil can fail in the very thin layer.
To study the effect of thin clay layers on the axial pile capacity, the entire soil profile was modeled as continuous sand, us-
ing qc as input. The results of the deterministic analyses with the NGI-05 method are presented in Figure 16 for the characteris-
tic and the mean qc profile. The analyses were done with two assumptions of qc in the clay layers: the bounds represent includ-
ing and neglecting the clay layers at 22, 35 and 48 m. It is believed that with the thin weaker layers at 22 and 35 m, the actual
capacity will be closer to that of the full sand profile (upper bound) than that of including the thin clay layers (lower bound).
The capacity will be in between the curves presented on Figure 16. Considering the above and using Figures 13, 14, 15 and 16,
the required pile lengths for different material factors for the API, NGI-05 and ICP-05 methods are given in Table 14.
12 OTC 24063
Scour and gapping. To partly account for scour, the cone resistance was reduced to a minimum of 0.1 MPa in the upper
1.5 m. Gapping caused by the horizontal load should be considered, especially for piles in clays. For Case Study B with sand
in the top 20 m, the sand will probably fill in the gap if gapping should occur. The very dense sand will also have a tendency to
dilate under loading, which should tend to reduce any gapping. Gapping is therefore not considered to be an issue.
Observations. There was a large difference in required pile length depending on the method used. The API method re-
quired much longer piles than the other methods. It is not a reliable method for very dense sands. The nature and strength char-
acteristics with depth in Case Study B (very dense sand with considerably weaker thin clay layers) made for an unusual in-
crease in capacity from 26 m to 51 m. The assumptions in the analyses to account for the effect of transition from one layer to
the other and the possible effect of punch-through tend to give too low ultimate axial pile capacity, because the thin weaker
layers at 22 and 35 m are too thin to allow full mobilization of a failure surface beneath the 2.4-m diameter pile.
Figure 13. Ultimate axial pile capacity in compression with the Figure 14. Ultimate axial pile capacity in compression with the
characteristic qc for Case Study B. mean qc (right diagram) for Case Study B.
Table 14. Required pile length, deterministic analyses with the API, NGI and ICP methods.
Method Strength profile Required pile length form = 1.5 Required pile length form = 1.3
API Statistical mean > 51 m 51 m
API Statistical mean, all sand* > 51 m 55 m
NGI-05 Characteristic 51 m 39 m
NGI-05 Characteristic all sand* 39 m 27 m
NGI-05 Mean 39 m 25 m
NGI-05 Mean, all sand* 26 m 23 m
ICP-05 Characteristic 26 m 26 m
ICP-05 Characteristic all sand* ≈23 m ≈18 m
ICP-05 Mean 23 m 13 m
ICP-05 Mean, all sand* 17 m 13 m
* Intermediate value between "low" qc and "high" qc
Probabilistic Analyses
Results. The annual probability of pile foundation failure was estimated with the same two-stage approach as for Case
Study A (Appendix C). Table 15 presents a selection of the probabilistic RELPAX and COMREL analyses done. The table
also shows the scaling factors for capacity and load established from the COMREL results to calculate the probabilistic foun-
dation capacity for an annual failure probability target of 10-4.
OTC 24063 13
q q
Pile tip depth below seafloor (m )
Plug 2 Pile
ge d
diameters
Co
San d
ri n
g
3 Pile
diameters
C la y
c:\cjfc\text\figures\pile-tip-stress.grf
Figure 15. Reduction in end bearing for pile tip embedded in sand Figure 16. Ultimate axial pile capacity in compression with charac-
layer, as recommended by API (1) and used in PACER (15). teristic and mean qc "all sand" profile, Case Study B.
Table 15. Results of probabilistic analyses of axial capacity for Case Study B
Ultimate pile capaci- Determinis- Scaling factor Calculated Relative contribution to Pf
Annual
Method ty Mean ± SD ticcapacity * on capacity for load factor Resistance
No Pf and Loads
(CoV); PDF = LN (MN) Pf = 10-4/yr for Pf=10-4/yr (Model uncertainty)
I API 106.9 ± 37.6 Pf = 1.310-2
83.3 1.676 0.700 -- ---
(51-m pile) (35%) = 2.2
III NGI-05 83.5 ± 12.0 Pf = 5.810-6 55 %
80.7 0.976 0.923 45 %
(26-m pile) (14 %) = 4.4 (45 %)
IV ICP-05 162.0 ± 29.0 Pf = 2.710-11 72 %
148.4 0.569 0.831 28 %
(26-m pile) (18 %) = 6.6 (58 %)
V Fugro-96/05 160.9 ± 34.0 Pf = 5.610-9 85 %
155.3 0.644 0.783 15 %
(26-m pile) (21 %) = 5.7 (70 %)
* With mean soil parameters
Observations. The probabilistic analyses of ultimate axial pile capacity, including all uncertainties in the soil parameters
gave coefficients of variation of 15 to 20 %, except for the API method. Model uncertainty in the pile capacity calculation
methods, either on the skin friction or the end bearing, was overwhelmingly the most significant random parameter contri-
buting most to increasing the probability of failure. The model uncertainties used are believed to be on the conservative side
(with a CoV larger than in reality) (17).
The results confirm the importance of quantifying this uncertainty parameter as realistically as possible. A parametric study
was also done with a 20-30 % increase in the model uncertainty. This increased the standard deviation and coefficient of varia-
tion of the ultimate axial pile capacity by about 5%. The model uncertainty was modeled with a lognormal PDF. NGI believes
this assumption to be more realistic than a normal PDF, because the normal distribution allows the calculations to select values
of biases (Qm/Qc) which are unrealistically high (and unrealistically low), which led to unrealistically low probabilities of fail-
ure.
ultimate pile capacity (Qult) and probabilistic environmental load (Penv), may help explain Steps 1 to 4 below. The PDF for the
Penv was taken as the same for Pf1 and Pf2 in the calculations. The calibration of the material factor should be consistent with
the definition of characteristic design load and the characteristic soil strength profile used for the calculation of axial pile ca-
pacity. Steps 5 to 9 below explain how the calibration is done for different ultimate capacities.
1) Obtain the scaling factor required to shift the PDF from the calculated annual Pf to the target Pf of 10-4 /yr;
2) Find the ultimate axial pile capacity, Qult mean, for the target Pf with the scaling factor;
3) Find the load on the pile (static, Pstat, + environmental, Penv⃰) at the design point for the target Pf;
4) Find the ultimate axial pile capacity at the design point, Qult⃰, for the target Pf;
5) Calculate the required material factor for Qult mean for Penv⃰ (design point);
6) Calculate the required material factor for Qult char for Penv⃰ (design point);
7) Calculate the load factor on Penv at the design point, l env⃰ (relative to the 100-yr characteristic load);
8) Calculate the required material factor for Qult mean for a load factor, l env, set to 1.3 (l stat is 1.0);
9) Calculate the required material factor for Qult char for a load factor, l env, set to 1.3 (l stat is 1.0).
Figure 17. 2D simplification of the PDFs of the environmental load Penv and ultimate pile capacity Qult.
Results
Tables 16 and 17 present the results of the calibration of the material factor for Case Studies A and B. The first column gives
the pile capacity method used; the second column, the characteristic and mean ultimate capacity from the RELPAX analyses.
The next six columns present the results for the capacities and the derived material factor and load factor for the target annual
probability of failure of 10-4/yr. Two material factors were obtained: one for the axial pile capacity calculated with the mean
undrained shear strength (Qult mean), and one for the axial pile capacity calculated with the characteristic undrained shear
strength (Qult char). The last two columns in the tables give the required material factor for a load factor on the environmental
100-yr load set to 1.3. The load factor at the design point was less than 1.3. In design however, the material factor would be
expected to be associated with a load factor of 1.3 on the 100-yr environmental load.
Table 16. Results of calibration of material factor for Case Study A (pile length 90 m).
RELPAX Pf = 10-4/yr l env = 1.3
Method Qult char Qult mean Scaling Qult mean Qult ⃰ (design point) m m l env ⃰ m m
(MN) (MN) factor (MN) (MN) (Qult mean) (Qult char) Design point (Qult mean) (Qult char)
API 124.7 151.7 0.671 101.8 57.5 1.77 2.46 1.12 1.64 1.35
NGI-05 100.1 120.5 0.762 91.8 61.2 1.50 1.25 1.26 1.48 1.23
ICP-05 133.5 136.8 0.706 96.6 59.2 1.63 1.59 1.19 1.55 1.52
Fugro-96/05 117.5 114.6 1.077 123.4 52.6 2.34 1.34 1.25 1.49 1.31
Table 17. Results of calibration of material factor for Case Study B (pile length 26 m, except for API, 51 m).
RELPAX Pf = 10-4/yr l env = 1.3
Method Qult char Qult mean Scaling Qult mean Qult ⃰ (design point) m m l env ⃰ m m
(MN) (MN) factor (MN) (MN) (Qult mean) (Qult char) Design point (Qult mean) (Qult char)
API 83.3 106.9 1.676 179.2 48.2 3.72 2.90 0.70 3.09 2.41
NGI-05 87.9 83.5 0.976 81.5 50.4 1.62 1.56 0.92 1.44 1.40
ICP-05 148.4 162.0 0.569 92.2 48.9 1.89 1.73 0.83 1.63 1.50
Fugro-96/05 155.3 160.9 0.644 103.6 48.1 2.15 2.08 0.78 1.84 1.77
OTC 24063 15
For both case studies, the calibrated material factors apply to those sites only, and cannot be transferred to other sites with-
out site-specific reliability studies. The calibrated material factor varies with the method of axial pile capacity used. The fac-
tors reflect the varying influence of the uncertainty in the soil parameters and of the model uncertainties for the different meth-
ods. The results present generally consistent trends, where the axial pile capacity methods predicting higher axial pile capacity
require a higher material factor to ensure that the probability of failure does not exceed 10-4/yr. The calibrated material factor
depends on the strength parameters used in the equilibrium equation to do the deterministic analyses, and should be used only
with the strength parameters it was derived from.
Case Study A. Using the NGI-05 pile capacity method, the required material factor is 1.50 if the mean undrained shear
strength is used, and the required material factor is 1.25 if the characteristic undrained shear strength (Fig. 1) is used. The load
factor,l env, was then 1.26. If the load factor is increased to 1.3, the material factors reduce to 1.48 and 1.23. With the ICP-05
method, which gave significantly higher capacity than the NGI-05 method, the calibrated material factor was 1.5 with a load
factor of 1.3. With the Fugro-96/05 method, a material factor of 1.3 seems appropriate with a load factor of 1.3. These numbers
apply to Case Study A only.
The calculations made the following assumptions: the case applies to a 90 m long pile. The calibrated factors represent an
approximation for the methods other than the NGI-05 method which determined the pile length of 90 m. A more complete
analysis would need to calibrate the material factor for the pile lengths required by each of the calculation methods.
Case Study B. Similar trends are seen for the pile capacity methods investigated for Case Study B. Because of the larger
model uncertainties in the axial pile capacity calculations, the calibrated material factors are higher than in Case Study A. Case
Study B is difficult to interpret, in terms of pile length and probability of failure, because of the relative little gain in capacity
as the pile length increases and the effect of the transition to and from the weaker layers.
For Case Study B and using the NGI-05 pile capacity method, the calibrated material factor is 1.62 if the mean soil parame-
ters are used, and the calibrated material factor reduces to 1.56 if the characteristic parameters selected for design are used. The
load factor, l env, was then 0.92. If the load factor is set to 1.3, the respective material factors reduce to 1.44 and 1.40. With the
ICP-05 method, which gave higher capacity than the NGI-05 method, the calibrated material factor was 1.6 with a load factor
of 1.3. With the Fugro-96/05 method, a material factor of 1.8 seems appropriate with a load factor of 1.3. Again, these num-
bers apply to Case Study B only.
The calibrated material factor of 1.40 with the NGI-05 method applies to the characteristic capacity in Figure 16. The actu-
al capacity in situ is believed to be closer the upper bound shown in Figure 16. One should do a reliability analysis with this
profile to complete the reliability study and to conclude on the required material factor.
It could be worthwhile to also consider using the mean of the soil parameters in the deterministic analyses and rather ad-
justing the required material factor. For example, in Case Study A, with the NGI-05 method, the material factor would be 1.5
rather than 1.3 if the mean undrained shear strength was used. For Case Study B, the material factor would be 1.45 rather than
1.4 if the mean strength parameters were used. Using mean values would have the advantage of removing the subjectivity in
the selection of the characteristic strength parameters. In addition, the mean is simpler to quantify and it is more objective.
Acknowledgments
The authors with to acknowledge Statoil for the permission to present this work. The authors thank Gunnar Lian and Bjørn Ivo
Krokeide from Statoil, Knut O. Ronold from DNV, Marc Lefranc, Rolf Hansson and Hans Jørgen Mikkelsen from FORCE
Technology, Patrick Joyce, Ping Lu, Neil Glover and David MacLaren from SNC-Lavalin, and Pasquale Carocanuto, Birger
Hansen, Morten Saue, Thomas Langford, Maarten Vanneste and Amir Rahim from NGI, for their assistance with the work.
References
(1) American Petroleum Institute (2011). Geotechnical and Foundation Design Considerations. ANSI/API Recommended practice 2GEO.
ISO 19901-4:2003 (Modified), Petroleum and natural gas industries–Specific requirements for offshore structures, Part 4–Geotechnical
and foundation design considerations. 1st edition, Washington, April 2011.
(2) ISO 19902 (2007). Petroleum and natural gas industries - Fixed Steel Offshore Structures. 1st ed., Switzerland, Dec. 2007.
(3) American Petroleum Institute (1993). Recommended Practice for Planning, Designing and Constructing Fixed Offshore Platforms -
Working Stress Design. API RP 2A-WSD, 20th Edition, Washington, 1 July 1993.
(4) Karlsrud K., Clausen, C.J.F. and Aas, P.M. (2005). Bearing Capacity of Driven Piles in Clay, the NGI Approach. Proc., International
Symposium on Frontiers in Offshore Geotechnics, Perth Sept. 2005, A.A. Balkema Publishers, ISBN 0 415 39063 X.
(5) Jardine, R.J., and F.C. Chow (1996). New design procedures for offshore piles. Marine Technology Directorate. Publ. 96/103, London,
UK.
(6) Jardine R.J., Chow F.C., Overy R.F. and Standing J.R. (2005). ICP design methods for driven piles in sands and clays. Imperial Col-
lege, Thomas Telford Publishing, London.
OTC 24063 17
(7) Kolk H.J. & E.v.d. Velde (1996). A Reliable Method to Determine Friction Capacity of Piles Driven into Clays.OTC paper no. 7993,
Houston, May 1996.
(8) Clausen, C., Aas, P.M. and Karlsrud K. (2005). Bearing Capacity of Driven Piles in Sand, the NGI Approach. Proc. 1st International
Symp. on Frontiers in Offshore Geotechnics, ISFOG, Univ. of Western Australia, Perth, Taylor & Francis, London, pp. 677-681.
(9) American Petroleum Institute (2007). Recommended Practice for Planning, Designing and Construction Fixed Offshore Platforms –
Working Stress Design. API RP 2A-WSD, 21st Edition, Washington, August 2007.
(10) Kolk H.J., Baaijens A.E. and Senders M. (2005). Design criteria for pipe piles in silica sands. Proc. 1st International Symp. on Frontiers
in Offshore Geotechnics, ISFOG, Univ. of Western Australia, Perth, Taylor & Francis, Lon-don, pp. 711-716.
(11) NORSOK (2004). NORSOK N-004, Design of steel structures. Rev.2, October 2004.
(12) DNV (2012). Recommended Practice DNV-RP-C207: Statistical representation of soil data, January 2012.
(13) Mathwave. www.mathwave.com/products/easyfit.htm.
(14) Lacasse, S. and A. Goulois (1989). Uncertainty in API parameters for prediction of axial capacity of driven piles in sand. Proc., 21st
Annual OTC, Houston, Texas, pp. 353-358.
(15) NGI (2011). PACER – A Computer Program for Calculation of Axial Capacity of Driven Piles. Program Documentation. NGI Report
No. 20061125-1, Rev.2 dated 1 November 2011.
(16) Haver. S. (2012). Personal communication. April 2012.
(17) Lacasse, S. F. Nadim, T. Langford, S. Kudsen, G. Yetginer. T.R. Guttormsen and A. Eide (2013). Model Uncertainty in Axial Pile
Capacity Design Methods. Paper OTC-24066-MS. Houston, Texas. May 2013.
(18) NGI (2000). Bearing Capacity of Driven Piles in Clay. Report 525211-1. 23 March 2000. Norwegian Geotechnical Institute. Oslo.
(19) NGI (2001). Bearing Capacity of Driven Piles in Sand. Report 525211-2. 21 Jan. 2001. Norwegian Geotechnical Institute. Oslo.
(20) NGI (1999). Deterministic and probabilistic analysis of an offshore pile. NGI report 514161-1, 23 March 1999.
(21) RCP GmbH (1999). STRUREL – A Structural Reliability Analysis Program System. RCP GmbH, Munich, Germany.
(22) Keaveny, J.M., F. Nadim and S. Lacasse (1989). Auto-correlation functions for offshore geotechnical data. Proc. 5th ICOSSAR, San
Francisco, USA, pp. 263–270.
(23) Lacasse, S. and F. Nadim (1996). Uncertainties in characterizing soil properties. Uncertainty in the Geologic Environment: From Theo-
ry to Practice. Proc. Uncertainty '96. Madison, WI. American Society of Civil Engineers. Geotechnical Special Publ., 58. pp. 49-75.
(24) Lunne, T. Robertson, P.K. and Powell, J.J.M. (1997). Cone Penetration Testing in Geotechnical Practice. Blackie A&P. LondonUK. 312p.
(25) Baldi, G. Bellotti, R. Ghionna, V., Jamiolkowski, M. and Pasqualini, E. (1986). Interpretation of CPTs and CPTUs; 2nd Part: drained
penetration of sands. Proc. 4th International Geotechnical Seminar, Singapore. pp.143-156.
(26) NGI (2008). PACER – A Computer Program for Calculation of Axial Capacity of Driven Piles. Program Documentation. NGI Report
No. 20061125-1, Rev. 1 dated 23 March 2008.
(27) Lehane BM, Schneider JA and Xu X. (2005). A Review of Design Methods for Offshore Driven Piles in Siliceous Sand. UWA Report
No. GEO 05358, The University of Western Australia, Perth, Australia, Sept. 2005.
(28) Schneider JA, Xu X and Lehane BM. (2008). Database Assessment of CPT-Based Design Methods for Axial Capacity of Driven Piles
in Siliceous Sands. ASCE Journal of Geotechnical and Geoenv. Eng. 134(9): 1227-1244, Sept.
(29) Simpson, B. (2012). Eurocode 7. Fundamental issues and some implications for users. Keynote Lecture, NGM 2012. Proc. DGF Bulle-
tin 27. V 1. pp. 29-52.
(30) Lacasse, S., Nadim, F., Rahim, A., Guttormsen, T.R. (2007a). Statistical description of characteristic soil properties. Offshore Tech-
nology Conference, 37. Houston 2007. Proceedings paper 19117.
(31) Lacasse, S., Guttormsen, T., Nadim, F., Rahim, A. and Lunne, T. (2007b. Use of statistical methods for selecting design soil parame-
ters. International Offshore Site Investigation and Geotechnics Conference, 6. London 2007. Proceedings, pp. 449-460.
(32) Lacasse, S. (1988). Uncertainty in offshore geotechnical engineering - Deterministic and probabilistic analysis of axial capacity of
single piles. NGI report 525285-8, 1 May 1988.
(33) Gollwitzer, S., T. Abdo, and R. Rackwitz (1988). FORM (First-Order Reliability Method) Manual. RCP GmbH, Munich, Germany.
List of Appendices
Appendix A. Statistical Analyses of Soil Data - Recommended Practice by DNV-RP-C207
Appendix B. Statistical Representation of Loads for Foundation Reliability Analyses
Appendix C. Reliability Analysis of Axial Pile Capacity
Appendix D. Procedure for the Statistical Treatment of CPT Data
18 OTC 24063
Dependent Soil Variables. Let Y be a dependent soil variable, with linear variation with depth. It is assumed that n obser-
vations of pairs (zi, yi) where i = 1, 2, 3, ...n are available from the soil investigations. The variation of Y with depth can be
expressed as , where the term represents the linear mean variation with depth, the term ε the natu-
ral variability of Y about the mean, and z denotes the depth below the soil surface. The coefficients and represent the
surface intercept and the depth gradient of the mean of Y. The standard deviation of Y is assumed to be constant, i.e. to be in-
dependent of the depth z. Estimators for the coefficients a0 and a1 (denoted â0 and â1 respectively) for the line with the best fit
to the data can be found from:
∑ ̅
∑ ̅
and ̅
in which ∑ and ̅ ∑ . The standard deviation of Y is estimated by the sample standard deviation:
1
2
For comparison of uncertainties for each soil parameter in the different layers, the coefficient of variation, CoV, is useful.
The CoV expresses the standard deviation as a fraction of the mean, often in percent. An estimator for CoV is not specified
(12) for dependent soil variables. The coefficient of variation could in principle vary with depth for dependent soil variables.
For the present study, the CoV-value (denoted ), was estimated for each layer as a value representative for the entire depth
of each layer with the following equation: CoV s/y. For simplicity, CoV was denoted CoV in the paper.
The extreme value, xq, corresponding to an annual exceedence probability q, is given by:
1
If the q-probability extreme values, xq1 and xq2, for respectively two exceedance probability levels q1 and q2 are given, then α
and β can be estimated from:
and 1
The extreme loads corresponding to 0.63, 0.1, 0.01, 0.001 and 0.0001 exceedence probability per year were provided by the
structural engineer. The parameters of the extreme value distribution can be estimated from any two of these extreme loads.
Figures in the main text provide the parameters corresponding to the various two-by-two combinations and the corresponding
Gumbel distributions. Table B1 provides the parameters corresponding to the two-by-two combinations for Case Study A.
Table B2 provides the parameters corresponding to the two-by-two combinations for Case Study B.
Uncertainty in environmental loads, Case Study A. At a given return period, there is some uncertainty in the calculated loads
due to the uncertainties in wave height and period, as well as the mathematical idealization. It was beyond the scope of the study
to do a detailed assessment of these uncertainties. Their effect were however included in the analyses. According to (16), the
model uncertainty for the storm-induced loads varies for different return periods. This uncertainty could be as high as 30% for the
10,000-yr return period, 25% for the 1,000-yr, 20 to25% for the 100-yr, and 10 to 15% for the 10-yr and 1-yr return periods.
These values reflect the range (upper and lower bounds), not the standard deviation of the model uncertainty, and they comprise
both the statistical uncertainty in wave height and period, and the modeling uncertainty of the mathematical idealization. For the
platform analyzed, the estimated uncertainty bands were within a range of ±10 to 16% for extreme loads with return periods of
100 to 10,000 years (16). To account for this, a modeling uncertainty parameter with mean of 1, standard deviation of 0.1, and
normal distribution was assumed for the storm-induced loads in the foundation reliability analyses. Sensitivity analyses were done
where this uncertainty was modeled by a Beta-distribution with a variable standard deviation as function of the load level. The
results of the analyses were not sensitive to how the uncertainty in the calculated loads was modeled.
Uncertainty in environmental loads, Case Study B. The uncertainty in the calculated environmental loads for Case Study
B was not assessed in detail. Based on discussions with the structural engineers, a modeling uncertainty parameter with mean
of 1, standard deviation of 0.1, and normal distribution was assumed for the storm-induced loads in the foundation reliability
OTC 24063 19
Table B1. Gumbel parameters, annual maximum storm-induced load on most heavily-loaded pile group, Case Study A.
Case q1/q2
xq1 xq2 β Mean Standard deviation
(MN) (MN) (MN) (MN) (MN) (MN)
0.1/0.01 75 800 101 900 50 804 11 107 57 215 14 245
0.1/0.0001 75 800 167 600 46 118 13 190 53 731 16 917
0.01/0.001 101 900 132 300 41 285 13 177 48 891 16 900
0.01/0.0001 101 900 167 600 36 342 14 251 44 658 18 278
0.001/0.0001 132 300 167 600 26 428 15 327 35 275 19 658
Table B2. Gumbel parameters, annual maximum storm-induced load on most heavily-loaded pile group, Case Study B.
xq1 xq2 α β Mean Standard deviation
Case q1/q2
(MN) (MN) (MN) (MN) (MN) (MN)
0.1/0.01 78 307 97 174 8 029 60 238 64 873 10 298
0.1/0.001 78 307 122 444 9 478 56 978 62 449 12 155
0.1/0.0001 78 307 159 172 11 619 52 161 58 867 14 901
0.01/0.001 97 174 122 444 10 953 46 788 53 110 14 047
0.01/0.0001 97 174 159 172 13 448 35 310 43 073 17 247
0.001/0.0001 122 444 159 172 15 948 12 290 21 494 20 453
where L is the domain of X where g(X) < 0. In general the above integral cannot be solved analytically, and an approximation
is obtained by the FORM approach. In this approach, the general case is approximated to an ideal situation where X is a vector
of independent Gaussian variables with zero mean and unit standard deviation, and g(X) is a linear function. The probability of
failure Pf at the "design point"(where the probability of failure is highest) is:
20 OTC 24063
n
Pf = P [ g(X) < 0 ] = P [
i 1
iUi – < 0 ] = (-)
where P[…] reads “the probability that”, i is the direction cosine of random variable Xi, Ui is the transformation of variable Xi
in the standard normal space, is the distance between the origin and the hyperplane g(U) = 0 in the standard normal space, n
is the number of basic random variables Xi (and its transformation in the standard normal space U), and is the standard nor-
mal distribution function. The vector of the direction cosines of the random variables (i) is called the vector of sensitivity
factors, and the distance is called the reliability index. The relationship between the reliability index, , and the probability
of failure, Pf, is shown in Figure C2.
The direction cosines or sensitivity factors are an important by-product of the FORM analysis. The square of the sensi-
tivity factors (i2), which sum is equal to unity, quantifies the relative contribution of the uncertainty in each random variable
Xi to the total uncertainty. The statistical subroutine packages FORM and SORM developed by (33) are implemented in
RELPAX software. Spot verifications were done with the second-order reliability method (SORM) and the changes in the
results were found to be small.
Evaluation of the Annual Probability of Pile Foundation Failure. The assessment of the annual probability of foundation
failure used the FORM approximation in the COMREL-Symbolic software package (21). The following limit state function g
was used in the calculation of the annual foundation failure for a group of four piles in compression:
g = scale*capacity – pile_weight – (storm_load*errload + perm_load) /4
where capacity = Axial pile capacity from RELPAX probabilistic analyses
storm_load = Sum of annual maximum storm-induced loads on pile group
perm_load = Sum of permanent static loads on pile group (deterministic)
errload = Random variable describing the epistemic uncertainty in load calculation model
pile_weight = Net sum of submerged weight of a 90-m steel pile and soil plug, and their buoyancy (deterministic)
scale = Scaling factor on capacity (set to unity in first analysis)
The limit state function g is based on the following assumptions: (1) failure of the most heavily-loaded pile in the foundation
system does not mean that the foundation system has failed. The jacket and the pile group are capable of redistributing the load
to other piles in the group once the axial capacity of a pile is fully mobilized. This assumption may be non-conservative; (2)
the uncertainty in gravity-induced loads is negligible. The variables used for the base case analyses of Case Studies A and B
are summarized in Table C1.
Figure C1. Ultimate axial capacity of a pile. Figure C2. Relationship between probability of failure, Pf, and reliability index .
Pult + W’ = Qs + Qtip
Qs = Ultimate skin friction
Qtip = Ultimate tip capacity
Pult = Axial pile capacity
OTC 24063 21
Table C1. Parameters used of the evaluation of the annual probability of failure for Case Studies A and B
Case Study Case Study A (90- m long pile) Case Study B (26-m long pile)
Variable PDF Mean SD PDF Mean SD
Capacity Lognormal or Normal See main text See main text Lognormal or Normal See main text See main text
Gumbel Gumbel
storm_load 44 658 kN 18 278 kN 62 449 kN 12 155 kN
= 36 342, = 14 251 = 9 478, = 56 978
storm_load Gumbel
Shifted exponential 50 467 kN 14 267 kN 43 073 kN 17 247 kN
(Sensitivity analysis) = 13 448, = 35 310
errload Beta with = 3, = 3 1.0 Variable * Normal 1.0 0.1
pile_weight ** -- 3 253 kN -- -- 1 490 kN --
perm_load * -- 103 300 kN -- -- 203 561 kN --
* Function of the load level, see main text PDF Probability density function SD Standard deviation
** Deterministic parameter
Figure D1. Filtered, re-sampled and merged cone resistance qc with mean and ±one standard deviation.