Sunteți pe pagina 1din 12

Proceedings of the 6th International Offshore Site Investigation and Geotechnics Conference:

Confronting New Challenges and Sharing Knowledge, 11–13 September 2007, London, UK

USE OF STATISTICAL METHODS FOR SELECTING


DESIGN SOIL PARAMETERS
Suzanne Lacasse
International Centre for Geohazards, Norwegian Geotechnical Institute, Norway
Tom Guttormsen
Norsk Hydro Production, Norway
Farrokh Nadim
International Centre for Geohazards, Norwegian Geotechnical Institute, Norway
Amir Rahim
NGI Inc, Houston, USA
Tom Lunne
International Centre for Geohazards, Norwegian Geotechnical Institute, Norway

Abstract
The mechanical properties of soils can never be established with complete certainty. The uncertainty
is due to natural variability of soils, imperfect interpretation models, measurement errors, insufficient
data, etc. The selection of soil properties for geotechnical design is often based on subjective judgment
and experience, and the uncertainties in soil properties are only indirectly accounted for when the
characteristic design values are selected. Although statistical methods can quantify uncertainties and
account for them in a rational manner, they are rarely used in establishing the design soil parameters.
The profession also uses imprecise definitions of ‘characteristic’, ‘best estimate’, ‘upper bound’ and
‘lower bound’ values for design. The profession needs to make a recommendation of which values to
use in design. The paper demonstrates how characteristic values of soil parameters can be extracted
from available data with statistical methods. Examples of characteristic (design) soil parameters over
the past 25 years were re-analysed using statistical methods. Recommendations for choosing charac-
teristic values and expressing variability are given.

1. Introduction Perhaps it has become a habit that no one questions, or the


Soils are naturally variable because of the way they are restricted use of statistical methods may be a reflection of
formed and the continuous processes of the environment the fact that often not enough data are available to actually
that alter them. The uncertainty in the mechanical proper- implement statistical methods with confidence. Det Norske
ties of offshore soils is due to the natural variability from Veritas (DNV) and Norsk Hydro1 prepared a guidance note
point to point within a soil volume, insufficient data and on the statistical representation of soil data.
imperfect interpretation models, measurement errors and When describing the design soil profile, expressions such
other sources. as ‘characteristic’, ‘best estimate’, ‘upper bound’ and ‘lower
The selection of soil parameters for geotechnical assessment bound’ values are used in practice. A common understand-
is often based on subjective judgment and accumulated ex- ing or a standard criterion does not exist for the selection
of ‘best estimate’, ‘upper bound’ and ‘lower bound’ values,
perience. The uncertainties in the soil properties are only
and there is an uncertainty on what is covered by these
indirectly accounted for when the characteristic (design)
expressions. The profession now needs to make a recom-
value(s) are chosen.
mendation of which values to use in design. The paper
Statistics and probability are useful tools for the quantifica- attempts to quantify some of these concepts through the
tion of the mean (most probable, expected) value and the reanalysis of case studies in the North Sea and elsewhere,
possible range of values of a parameter. Statistical and prob- where characteristic (design) values have been selected by
abilistic methods can quantify the uncertainties and make experienced engineers. Examples of design soil parameters
it possible to account for them in a rational and consistent recommended by the Norwegian Geotechnical Institute
manner. They are, however, rarely used in practice to estab- (NGI) for offshore sites over the past two decades were
lish the design soil parameters. The reason for this is unclear. re-evaluated using statistical methods, and comparisons of

449

OSIG final.indb 449 09/08/2007 00:08:00


Lacasse, Guttormsen, Nadim, Rahim and Lunne. Use of Statistical Methods for Selecting Design Soil Parameters

characteristic value with best estimates and variance were ment and accumulated experience. The uncertainties which
made. To ensure neutrality for the statistical analysis, the are involved with the soil properties are only, to a limited
cases were analysed by engineers not having had any prior extent, brought into the picture when the characteristic val-
involvement in the original selection of the characteristic ues are chosen and reported to the client. The main sources
parameters. The characteristic values were then compared of uncertainty in estimated shear of strength soil within a
with the results of unbiased statistical analyses. nominally uniform layer are
The paper discusses ‘characteristic value’, ‘best estimate’, • Limited geo-exploration
‘upper bound’ and ‘lower bound’, as they are used to de- • Measurement errors
scribe a design soil profile. Characteristic (design) values of • Limited parameter evaluation
undrained shear strength were compared with the results
• Spatial variability of mechanical properties
of unbiased statistical analyses. The paper suggests that the
profession adopts a univocal definition for the term ‘best • Sample disturbance
estimate’ and for parameter variability. This suggestion is • Bias in test method.
proposed as a subject for debate in the coming months so These sources, except for the inherent spatial variability, are
that agreement can be established for use in practice. epistemic uncertainties. This type of uncertainty can be re-
duced by, for example, increasing the number of tests or
2. Uncertainties in Soil Shear Strength improving the measurement methods.
Soil design parameters for offshore geotechnical design are
usually obtained from laboratory tests on obtained samples 2.2 Uncertainties due to sample disturbance
and from in situ tests, mainly the piezocone penetration When evaluating the uncertainty of a soil property, it is im-
test (CPTU). The paper focuses on the undrained shear portant to consider the factors that can influence the qual-
strength of the soil, which is the most important parameter ity of the soil properties derived from the laboratory or in
needed for the evaluation of the capacity of offshore foun- situ tests, and to take the uncertainties associated with the
dations (both skirt penetration and load resistance). The assessed soil parameters into account. The main cause of
characteristic (design) undrained shear strength in the case uncertainty in results of laboratory testing is sample distur-
studies was based on laboratory tests under different stress bance. Sample disturbance can be due to
conditions, the interpretation of in situ CPTU or correla- 1. The drilling process for borehole preparation for down-
tions for normalised undrained shear strength as a function hole mode sampling
of over-consolidation ratio (OCR). Lacasse et al.2 presented 2. Sampling by pressing the sample tube into the soil, ei-
six case studies, three of which were for foundation design ther at the bottom of a borehole for downhole mode
for offshore installations, and three related to geohazards as- sampling, or into the seabed for sea bed mode sampling
sessment at the site. The paper presents the statistical analy-
3. Stress relief when bringing the sample up to deck; this
ses for three recent case studies of geohazards assessment
is particularly important if gas is present in situ, in dis-
studies in the North Sea and combines these results with
solved or free form
the case studies reported earlier.
4. Handling, storing and transportation of sample to on-
In the engineering literature, the terms variability and un- shore laboratory
certainty are often employed interchangeably. While the two
5. Storage, testing, equipment and procedures in the labo-
terms refer to concepts that are closely related, there are some
ratory.
important differences. Variability, which represents the natu-
ral randomness of the property, is also referred to as aleatory It is outside the scope of this paper to discuss all these fac-
(or inherent) uncertainty. It is defined as the observable man- tors in detail. In most cases the tube sampling disturbance
ifestation of one or more physical processes. In principle, a is the most severe factor. A comprehensive study by Lunne
variable property could be described if a sufficient number of et al.5 confirmed the effects of tube sampling disturbance on
measurements are available and if the quality of the measure- shear strength parameters, as measured in constant rate of
ments themselves is sufficient to ensure a confident evalua- strain consolidation (CRSC), anisotropically consolidated
tion of the observations. Hence, the observation of variability undrained triaxial compression (CAUC), anisotropically
implicitly provides a more or less detailed assessment of the consolidated undrained triaxial extension (CAUE) and di-
level of knowledge on a phenomenon of interest and the ca- rect simple shear (DSS) tests. Parallel laboratory tests were
pability to measure and model the phenomenon itself3. carried out on high quality block samples and on different
types of piston tube samples from 12 soft Norwegian marine
Uncertainty can also be epistemic. Epistemic uncertainty clays.
refers to the lack of knowledge for a given soil property4, 2.
The quantification of epistemic uncertainty always requires With reference to offshore applications, block samples rep-
some degree of expert judgement. resent the best possible samples that can be recovered today.
The measured soil parameters are believed to be as close
2.1 Sources of uncertainty to the in situ conditions as feasible with today’s technol-
The selection of characteristic values of soil properties for ogy. The 75-mm piston thin wall samples can be considered
use in geotechnical design is often based on subjective judg- to be the best piston samples obtainable offshore today. The

450

OSIG final.indb 450 09/08/2007 00:08:01


Proceedings of the 6th International Offshore Site Investigation and Geotechnics Conference:
Confronting New Challenges and Sharing Knowledge, 11–13 September 2007, London, UK

Figure 1: Normalised stress-strain curves and stress paths from CAUC tests on Lierstranda clay5

composite 54-mm piston tube samples are considered to have samples are owing to the fact that the clay, when disturbed,
characteristics (in terms of geometry of cutting shoe and area changed from a contractant to a dilatant behaviour, as a
ratio) similar to those of a gravity or Kullenberg sampler used large volume change, ∆e/e0 (see Table 1), occurred when the
for seabed sampling offshore. Figure 1 compares the results clay was re-consolidated to the in situ conditions.
of CAUC tests on specimens from three types of samples
The study by Lunne et al.5 indicated that the undrained
of Lierstranda clay. Large differences in the undrained shear
shear strength measured on gravity core samples may, in
strength are measured on the three types of samples.
the worst case, underestimate the in situ value by a factor of
Table 1 summarises the ratios of undrained shear strength almost 2. It is possible to classify the degree of sample dis-
as measured on 54-mm tube sample to that measured on a turbance by considering the volume change, ∆e/e0. For the
block sample on the 12 types of clay included in the NGI evaluation of the most reliable undrained shear strength,
study5. In most cases, the undrained shear strength ratios one places more emphasis on the samples that have the
(ratio block sampler to 54-mm sampler) vary between 1.2 lowest degree of disturbance. The results of laboratory tests
and 1.9. Two clays, however, came out with ratios less than may also not be representative of the in situ conditions if
unity: these unusually lower shear strength of the block erroneous consolidation stresses are used in the laboratory.

Table 1: Effects of sample disturbance for 12 clays in NGI’s block sample database5
Clay Depth Water Plasticity Ave. Volume Change, Sample Shear Strength
(m) Content, w Index, Ip ∆e/e0 Quality Ratio, CAUC
(%) (%) Block 54mm Index* su,block/ su,54-mm

Lierstranda 6 40–42 20 0.025 0.053 1/2 1.35


12.3–16.4 34–38 16–18 0.032 0.080 1/3 1.22
Daneviksgt., 5–10 51–54 27–29 0.021 0.080 1/3 1.19
Drammen 14–19 21–36 10–20 0.042 0.118 1–2/3–4 0.79
Onsøy 7–15 61–65 34–41 0.026 0.071 1/2–3 1.28
Elveplassen 10.1–10.4 34–36 14–16 0.027 0.103 1/3 1.19
Leirsund 7–16 31–39 12–18 0.036 0.061 1/2 1.25***
Eidsvold 6–15 25–33 13–19 0.038 0.090 1/3 0.96
Glava** 5.7–6.1 33 17–18 0.013 0.026 1/2 1.47
Emmerstad 4.1–5.9 37–44 6–11 0.003 0.017 1/1 1.23***
Ellingsrud 7–13 35–39 6–7 0.022 0.079 1/2–3 1.43***
Nykirke 6–10 25–36 4–9 0.018 0.129 1/3–4 1.91
Kvenild 7–8.5 31–37 10–14 0.032 0.071 1/2–3 1.43
5,
* According to sample disturbance criteria in Lunne et al. block sample compared to 54mm sample
** OCR = 4.5–6
*** su,block sample/su,95-mm sample

451

OSIG final.indb 451 09/08/2007 00:08:03


Lacasse, Guttormsen, Nadim, Rahim and Lunne. Use of Statistical Methods for Selecting Design Soil Parameters

Cases have been seen where the undrained shear strength


results are unreliable for excellent samples, because the
wrong in situ stresses have been used for the consolidation
in the laboratory.
2.3 Uncertainties in shear strength estimates from
in situ tests
The most used offshore in situ test nowadays is the piezocone
test (CPTU). The sources of error for soil shear strength pa-
rameters derived from CPTU tests include measurement er-
ror, assumptions used in interpreting the results and choice
of the so-called N-factors for estimating the undrained shear
strength on the basis of the CPTU results. Just as for the
laboratory data, Lunne et al.5, NGI6, Karlsrud et al.7 and
Lunne et al.8 showed that the derived soil parameters may Figure 2: Variability about mean of depth-dependent soil
vary by a factor of as much as 2, depending on the effects of parameter
sample disturbance, measurement error and interpretation.
In summary, there can be significant sources of error in the 3.2 Parameter estimation for dependent soil variables
results of both laboratory and in situ tests. Before doing a Let Y be a soil variable, e.g. the shear strength, whose varia-
statistical study on soil data from laboratory and/or in situ tion with depth, z, can be reasonably well modelled as a linear
tests, the quality of the data should be carefully assessed. function: Y is then a so-called dependent variable, whereas z
When differences in quality are documented, the relative is independent. Assuming that n observations of pairs, zi, yi,
importance to the data of variable quality should be con- where i = 1, 2, 3, ... n are available from a soil investigation,
sidered given. Only data from the same geological layer the variation of Y with depth can be expressed as
should be used in the statistical analysis. Y = a0 + a1z + ε (2)

3. Statistical Description of Soil Properties where the term a0 + a1z represents the linear mean variation
with depth, the term ε is the natural variability of Y about
A quantitative geotechnical variability investigation relies on
the mean and z is the depth below the soil surface. The vari-
sets (samples in statistical terms) of measured data which are
able Y versus z is shown in Figure 2. The coefficients a0 and
limited in size and quality. Hence, it is necessary to refer to
a1 represent the surface intercept and the depth gradient,
sample statistics. Sample statistics are imperfect estimators of
respectively, of the mean of Y. The variability term, ε, has
the ‘real’ population parameters and may be biased. ‘Sample
zero mean and standard deviation, σ. The standard devia-
statistics’ refers to any mathematical function for a sample.
tion, σ, is assumed to be constant, i.e. it is assumed to be
An infinite number of sample statistics may be calculated
independent of depth, z. The coefficients, a0 and a1, in the
from any given data set. For most geotechnical engineering
linear expression can be estimated from the n observed data
applications, however, it is usually sufficient to calculate the
pairs, z and y, as
first two statistical moments of a sample, i.e. the mean and
standard deviation. Higher moments can be unreliable when
estimated from the usual sample sizes available for offshore (3)
sites.
3.1 Parameter estimation for independent soil variables and
Let X be a soil parameter, for example the unit weight, for (4)
which n observations, X1 ... Xn, exist. The mean value of X in which
is µ and the standard deviation is σ. Any dependency on
depth or any other conditioning quantity is disregarded in (5)
this example, i.e. x is a so-called independent variable.
and
The mean value, µ, is estimated by the sample mean denoted (6)
, and the standard deviation, σ, is estimated by the sample
standard deviation, where n is the number of data points: The standard deviation, σ, of the natural variability term is
estimated by the sample standard deviation

(7)
(1)
When the standard deviation or variance of ε is not con-
The sample estimates and s are central estimates of µ and stant with depth, the derivation above is no longer valid
σ. The coefficient of variation (CoV) is the ratio of the (Figure 3). The guidance note by DNV and Norsk Hydro1
standard deviation to the mean, CoV = σ/µ. provided the equations for a0 and a1 for the situation where

452

OSIG final.indb 452 09/08/2007 00:08:05


Proceedings of the 6th International Offshore Site Investigation and Geotechnics Conference:
Confronting New Challenges and Sharing Knowledge, 11–13 September 2007, London, UK

drained shear strength is often selected as


a conservative estimate representative of
the available data. For the statistical study,
the mean and standard deviations were
obtained from the parameter estimation
method for dependent variables, as de-
scribed previously. Only the triaxial com-
pression data were used, as they represent
the most reliable test results.
4.2 Undrained shear strength interpreted
Figure 3: Depth-dependent soil parameter with constant and non-constant variance from CPTU results
with depth When the characteristic undrained shear
strength profile is established on the basis
the standard deviation of ε increases with depth, and the of CPTUs, the results can be interpreted from theoretical
increase is proportional with depth, z. or empirical correlations. The empirical approach was used
in this study. To interpret undrained shear strength from
4. Undrained Shear Strength CPTU test results, three main categories can be used
The three new sets of data analysed are from an offshore site 1. ‘Total’ cone resistance
in the North Sea, where a geohazards assessment was carried
2. ‘Effective’ cone resistance
out in 2006. Shear strength profiles were needed for the cal-
culation of both skirt penetration resistance (Characteristic, 3. Excess pore pressure.
High) and the load capacity (Characteristic, Low) of the off- In most of the cases presented in this paper, the estimation
shore installation. Eleven CPTUs were analysed. As the pro- of undrained shear stregth, su, from CPTU tests used the
filing varied at the different locations at this site (denoted G), corrected total cone resistance. The interpretation was done
the CPTUs were clustered in groups of three and four at the with the following equation
three locations. The locations are denoted G1, G2 and G3. (8)
The earlier sites reported in Lacasse et al. were three geo-
2

hazards assessment sites, designated Sites A, B and C, and where qt is the corrected cone resistance (measured cone re-
three soil profiles established for stability analysis of offshore sistance plus measured pore pressure behind the cone), σvo
installations, denoted Sites D, E and F. The characteristic un- is the total overburden stress and Nkt is the cone factor.
drained shear strength was based on CAUE, DSS laboratory There are two main sources of uncertainty in the evaluation
tests and/or CPTUs. At times, alternative methods of inter- of su from equation 8: the corrected cone resistance, qt, and
pretation were used to derive the undrained shear strength. the cone factor, Nkt. The uncertainty in the total overbur-
Although the geotechnical specialists who established the den stress is considered as small, as σvo is calculated from
characteristic (design) undrained shear strength profiles the unit weight of the clay times the depth. Data in the
were aware of the variability of the measurements at these literature show little variability in the unit weight of clay
three sites, as well as the uncertainty in the estimation of the within a layer. Therefore, in the following derivation, the
undrained shear strength, no attempt was made at the time uncertainty in the overburden stress is ignored.
to quantify the variability or uncertainties.
The mean and standard deviation of the undrained shear
The characteristic low estimate of undrained shear strength strength, su, were evaluated using the first order, second
for G2 recommended for the soil layer down to 6m depth moment (FOSM) approach9. The FOSM approach pro-
(see Figure 4a in Section 5.1) appears to be conservative. vides an effective means of investigating the propagation of
However, at this location, the recommendation of the de- second moment uncertainties. It provides an approximate
sign engineer was influenced by the large variability in the estimate of the central tendency parameter (e.g. mean) and
interpreted undrained shear strength below 6m. Figure 4b the dispersion parameter (e.g. standard deviation) of a ran-
(see Section 5.1) shows the results of the four CPTUs at dom variable which is a function of other random variable.
this location down to 20m depth, and the interpreted un- The statistical parameters for the undrained shear strength
drained shear strength down to 12m depth (a separate soil were evaluated as follows
layer was considered below 6m depth). It could be argued
(9)
that the design engineer should not have let the large vari-
ability in the CPTU results below 6m influence his recom- with mean at any depth as
mendations for the top 6m the soil profile.
(10)
4.1 Undrained shear strength from laboratory measurements
When undrained shear strength is measured in the labo- Ignoring the uncertainty which may be present in the es-
ratory, the shear strength profile versus depth consists of timation of the overburden stress, the variance of the un-
discrete measurements, and the characteristic (design) un- drained shear strength, su, accounting for the uncertainties

453

OSIG final.indb 453 09/08/2007 00:08:07


Lacasse, Guttormsen, Nadim, Rahim and Lunne. Use of Statistical Methods for Selecting Design Soil Parameters

for both qt and Nkt, becomes resented by the µsu ± σsu. The dispersion, σsu, represents
both the spatial variability of the undrained shear strength
(11) in the soil layer in question (aleatory uncertainty) and the
uncertainty in the estimation of the in situ values (epistemic
where the mean and standard deviation of the cone resist- uncertainty).
ance are given by 5.1 Three locations at Site G
Figure 4 presents the results of the CPTU and of the statis-
tical analysis done for three locations at Site G. At Location
G1, the three CPTU results are rather homogeneous; at
(12) Site G2, the four CPTUs give results not too different from
those at Location G1, whereas at Site G3, the four cone
The soil design parameter reports that were used in this resistance values show two separate trends.
study often provided upper bound and lower bound esti- The statistical interpretation done is shown on the right-
mates of the cone factor. If the average cone factor was not hand side graphs in Figure 4a. Two characteristic undrained
given, the statistical analyses assumed that the cone factor shear strength profiles were established by the designing en-
had a normal probability distribution and that the mean gineer: a ‘low’ characteristic undrained shear strength for
and standard deviation of Nkt were the calculation of bearing capacity and stability, and a ‘high’
characteristic undrained shear strength for the calculation of
the soil resistance to skirt penetration. The graphs present
the ‘low and high’ characteristic (design) values, and the
(13)
mean ± 1 standard deviation based on the statistical analysis
The above equations for the mean and standard deviation of the corrected cone resistance, qt, from the CPTU. The
of Nkt are based on judgement, not statistics. The equation CoV (ratio of standard deviation to mean) at Locations G1,
for the mean value is valid if one assumes a symmetric prob- G2 and G3 were about 50–60% near the seabed, reducing
ability distribution for Nkt. If one assumes a uniform distri- to about 15–25% at a depth of 6m. The high CoV near the
bution for Nkt between its upper bound and lower bound seabed is due to the low mean value of the undrained shear
values, then the standard deviation would be 0.289 times strength.
the difference between the bounds. On the other hand, if 5.1.1 su for bearing capacity calculations
one assumes a symmetric triangular distribution, then the
If one considers the clay below a depth of 1m, the statistical
standard deviation would be 0.204 times the difference be-
analyses suggest the following:
tween the bounds. The real distribution is believed to be
somewhere between uniform and triangular, so the stand- • At Location G1, the characteristic su for stability calcula-
ard deviation was assumed to be 0.25 times the difference tions was 1 to 2 times the standard deviation below the
between the upper bound and lower bound values. mean value.
• At Location G2, the characteristic su for stability calcula-
If one should interpret the undrained shear strength from
tions was between 1 and 3 times the standard deviation
CPTU with either the effective cone resistance or excess
below the mean value. If one considers the su to follow
pore pressure, the derivation of the mean and standard devi-
a normal distribution, it is difficult to imagine an su less
ation of su using either of those two interpretation methods
than three times the standard deviation from the mean.
can be done in a manner similar to the procedure outlined
previously. For Site E, the estimation of su from CPTU tests • At Location G3, the characteristic su for stability calculations
used two alternative approaches: (1) correlation between was typically 1 standard deviation below the mean value.
corrected cone resistance, Nkt, and the pore pressure fac- 5.1.2 su for skirt penetration calculations
tor, Bq10; and (2) site-specific correlation between the pore Again excepting the top 1m of the clay, the following can
pressure factor, Bq, and the OCR, followed by calculation be observed at all three locations, the characteristic su for
of the su from the OCR using the SHANSEP approach11. skirt penetration calculations was close to approximately
The pore pressure factor, Bq, is defined as the mean plus 1 standard deviation.
(14) 5.1.3 Comment
With respect to the results of unbiased statistical analy-
where qt is the corrected cone resistance, σvo is the total over- ses, the characteristic values selected seemed to fit well the
burden stress and ∆u the measured excess pore pressure. range of ±1 standard deviation, which is close to the range
of uncertainty wanted to account for when choosing ‘con-
5. Characteristic Soil Profiles versus Statistical servative’ values for bearing capacity design. The exception
Soil Parameters was Location G2, where the ‘low’ characteristic value was
The statistical results are shown as the mean of the und- much lower than the mean –1 standard deviation at 6m.
rained shear strength, µsu, based on the laboratory and/or This was due to the fact that the interpretation was being
in situ test results used, and the standard deviation rep- carried out at shallow depths and that the undrained shear

454

OSIG final.indb 454 09/08/2007 00:08:08


Proceedings of the 6th International Offshore Site Investigation and Geotechnics Conference:
Confronting New Challenges and Sharing Knowledge, 11–13 September 2007, London, UK

Figure 4a: Results at three


locations at Site G (2006)

Left: CPTU tests


(G1, three CTPUs; G2, four
CPTUs; G3, four CPTUs)

Right: Characteristic
undrained shear strength com-
pared with statistical results

Figure 4b: Results below


6m depth at
location G2 at Site G (2006)

Left: Four CPTUs


(down to 20m)

Right: Undrained shear


strength
(down to 12m)

455

OSIG final.indb 455 09/08/2007 00:08:13


Lacasse, Guttormsen, Nadim, Rahim and Lunne. Use of Statistical Methods for Selecting Design Soil Parameters

Figure 5: Characteristic undrained shear strength in triaxial Figure 7: Characteristic triaxial compression undrained
compression at Site A (1984) compared with result of shear strength compared with result of statistical
statistical analysis analysis at Site B

strength values were low (between 3 and 25kPa in the top measured CAUC values. The characteristic undrained shear
6m). In addition, especially at Location G2, the selection of strength in triaxial compression, suC, selected was even low-
the characteristic undrained shear strength was influenced er than the mean –1 standard deviation below depth 65m.
by the results of the CPTUs below 6m. Statistical analyses
5.2.2 Site B
are presently underway for the clay layers below the top 6m
of the deposit, and will help establish whether the trends At Site B, six fairly uniform CPTU profiles were analysed
observed at Site G and the six other sites reported in Lacasse statistically. Figures 6 and 7 present the available CPTU
et al.2 (Section 5.2) are consistent. data and the results of the statistical analysis at Site B com-
pared with the characteristic values. The undrained shear
5.2 Other sites strength was interpreted from the CPTUs to correspond
5.2.1 Site A to loading under triaxial compression conditions, suC. The
At Site A, CAUC tests were used to establish the charac- statistical mean and the recommended characteristic value
teristic undrained shear strength under triaxial conditions. compare quite well and are in agreement at 1 to 2m depth
Twenty-two tests were available between depths of 45 and and at about 8m depth. Otherwise, the statistical mean is
95m. Figure 5 presents the results for Site A. The statis- only a few percentage points (up to 5%) below the charac-
tical analysis is shown as the means and ±1 standard de- teristic value. The CoV was higher, approximately 16%.
viation (±1 std. dev.). The characteristic (design) undrained 5.2.3 Site C
shear strength profile was taken at the lower bound of the At Site C (Figure 8), strength index tests (torvane, pocket
penetrometer, lab vane), in situ vane tests and unconsoli-
Figure 6: Results of six CPTU tests at Site B (2003) dated undrained (UU) tests were available. More recent
anisotropically and isotropically undrained triaxial com-
pression tests (CIUC and CAUC) were run at depths of

Figure 8: Undrained shear strength tests at Site C (2003)

456

OSIG final.indb 456 09/08/2007 00:08:16


Proceedings of the 6th International Offshore Site Investigation and Geotechnics Conference:
Confronting New Challenges and Sharing Knowledge, 11–13 September 2007, London, UK

at a depth of 70m may also have been affected by sample


disturbance, though probably to a smaller degree than the
strength index tests.
Figure 9 compares the characteristic undrained shear
strength with the statistical analysis results. The mean is
10–15% higher than the characteristic undrained shear
strength. The characteristic shear strength is close to the
mean –1 standard deviation. The characteristic shear
strength below a depth of 50m tends to be higher than the
mean –1 standard deviation. The relative position of the
characteristic profile and triaxial data at 69m can confirm
that the triaxial data may have also been affected by sample
disturbance at 70m, if there is no change in the layering at
that depth.
5.2.4 Site D
Figure 9: Characteristic triaxial compression undrained shear At Site D, four highly variable CPTU profiles were avail-
strength at Site C compared with result of statistical analysis able. Figure 10 reproduces the corrected cone resistance,
and Figure 11 presents the characteristic lower and upper
28 and 69m. These are the most reliable results. The char- bound interpreted suC (corresponding to triaxial compres-
acteristic undrained shear strength in triaxial compression, sion) in the design report. The strength recommendations
suC, was obtained from a correlation with in situ overburden were made for calculations associated with skirt penetra-
stress, p’o, and OCR, where the undrained shear strength tion. Figure 11 also shows the statistical mean ±1 standard
was taken as 0.31*p’o. Figure 9 presents the statistical analy- deviation based on the four CPTU results. The character-
sis. Using the method described above, the CoV about the istic upper bound agrees well the mean +1 standard devia-
means was between 10 and 15%. Since there were only four tion in the top 3m, but is much higher than that below
data points available, the standard deviation was verified us- 3m. On the other hand, the characteristic lower bound is
ing the Snedecor and Cochran12 approximation, where the about 2 standard deviations lower than the mean in the
range of values at a given depth is multiplied by a weighting top 3m, and 1 standard deviation below the mean between
factor4. This led to a CoV between 6 and 14%. For a more 3 and 6m.
complete study, one should also include the statistics for 5.2.5 Site E
the field vane tests. The field vane showed a shear strength
At Site E, two CPTU profiles were available in two loca-
profile close to the triaxial compression results, and is con-
tions that had very different soil profiles. Figures 12 and
sidered to be reliable.
13 present the results of the analyses (sites E1 and E2).
The strength index and UU tests show linearly increasing For this site, the effect of using two different methods of
shear strength with depth down to a depth of 70m, and a interpretation for the CPTU results was studied, and the
decrease in strength down to 90m. The data below 70m are undrained shear strength profiles from the two methods
probably affected by sampling disturbance. The triaxial data was compared. The two interpretations were used to obtain

Figure 10: Results of four CPTUs at Site D (2001) Figure 11: Characteristic upper and lower bound undrained
shear strength from CPTUs at Site D compared with result of
statistical analysis

457

OSIG final.indb 457 09/08/2007 00:08:18


Lacasse, Guttormsen, Nadim, Rahim and Lunne. Use of Statistical Methods for Selecting Design Soil Parameters

Figure 12: Characteristic DSS undrained shear strength from Figure 13: Characteristic DSS undrained shear strength from
CPTUs interpreted with two methods at Site E1 (2003) CPTUs interpreted with two methods at Site E2 (2003)

the undrained shear strength for a direct simple shear state acteristic undrained shear strength was selected close to
of stress, suDSS. The results illustrate the uncertainty in the the mean value –1 standard deviation. Characteristic un-
interpretation method. For Site E1, the two interpretations drained shear strengths from more recent case studies lie
gave very different results. The characteristic profile was se- close to the mean value –1/2 standard deviation, although
lected as close to the lower of the two interpretations. For there are exceptions.
Site E2, the two interpretation methods gave similar results.
When the engineer is faced with two sets of data for the
The characteristic strength was again selected close to the
shear strength profile, the engineer tends to choose the
lower of the two interpretations.
lowest value of the two as the characteristic profile. It may
be more logical to combine the two sets of data through a
6. Observed Trends
Bayesian model, for example.
In a deterministic framework, the safety margin of a geo-
technical design, e.g foundation bearing capacity or the 7. Recommendations
stability situation for a slope, is expressed by the factor of
safety (FoS). The FoS is defined as the ratio of the char-
7.1 Summary
acteristic resisting force to the characteristic driving force. The study aimed at setting the stage for a discussion in the
The approach does not address the uncertainty in load and profession on the consistency in the selection of design pa-
resistance in a consistent manner. The ambiguous defini- rameters. The profession should aim at providing consistent
tion of ‘characteristic’ value requires that the engineer safety margins in design, as was the intention with load and
implicitly account for uncertainties by choosing conserva- resistance factored design (LRFD) codes. The study shows
tive values of load (high) and resistance parameters (low). that there can be large and highly variable uncertainties in
The choice does not reflect a consistent safety margin and, the soil parameters. Probability approaches or some simpli-
because of this, appears to be somewhat arbitrary, while fied version of them are probably the only rigorous ways
remaining conservative. Foundations with nominally the to approach a consistent safety margin. Ideally, character-
same FoS could have significantly different safety margins istic parameters should be estimated to be above or below
a threshold within a distribution in which the user has a
because of the uncertainties and the way in which they
reasonable level of confidence (threshold recommendations
are dealt.
are made in the following section). Such an approach would
The characteristic shear strength profile for seafloor stabil- also give recognition to the necessity and cost-effectiveness
ity and bearing capacity evaluations recommended by the of high quality and reliable soil sampling and testing.
experienced engineer seems to fit well with the statistically
From the case studies, it would seem that when one relies
determined mean value –1/2 to –1 standard deviation. At
on laboratory test results to establish the characteristic (de-
shallow depths, however, the engineer seems to be more
sign) strength, the experienced engineers tend to lie much
conservative, with values equal to the mean –1 standard de-
lower than the mean, perhaps at about 1 standard deviation
viation or more being used. This is associated with the low
below the mean. On the other hand, when one relies on the
values of undrained shear strength close to the seafloor, and
results of in situ piezocone tests, the characteristic strength
the possible larger uncertainties associated with the field
can be much closer to the mean of the interpreted measure-
operation of in situ testing devices and sampling at very
ments in situ, except at shallow depths. This may be due
shallow depths.
to the fact that one feels that for CPTU tests in clays, the
The case studies suggest that in the early 1980s, the char- selection of the cone resistance values is already based on

458

OSIG final.indb 458 09/08/2007 00:08:19


Proceedings of the 6th International Offshore Site Investigation and Geotechnics Conference:
Confronting New Challenges and Sharing Knowledge, 11–13 September 2007, London, UK

the data that lie towards the lower bound of the measured 7.3 Recommendations
data. A study of the case study databank suggests that the Which definition of the characteristic value applies usually
experienced engineer was more conservative in the early to depends on the design code and on the actual application.
mid-1980s than in the past 10–15 years. For problems that are governed by a local soil strength value,
a low quantile in the strength distribution is a logical choice
7.2 Confidence in statistical estimates of soil properties
for the characteristic value. For problems that involve large
An important issue is the confidence level in the statisti-
soil volumes where local strength variations from point to
cal estimates of soil properties. The equations in Sections
point can be assumed to average out (such as in stability
3.1 and 3.2 give the sample mean and standard deviation,
calculations for large gravity-based foundations), the mean
which are only estimates of the true mean and standard de- value of the soil strength would be a more logical choice for
viation of a given property. The uncertainty in the estimates the characteristic value. In principle, the characteristic value
of the true mean and standard deviation can be quantified should be chosen in the same way in which it was chosen
based on the theory for one-sided confidence intervals13. for the reliability analyses used to calibrate the design codes
For example, if X is a soil parameter for which n observa- against a given probability of failure.
tions exist, then the mean value of X, can be estimated with
a confidence of 1 – α from the following equation The profession needs to clarify and unify the vocabulary
used in the description of soil parameters used for design.
(15) Characteristic strength is probably fairly well understood
by most practitioners to be the recommended value for de-
where and s are, respectively, the sample mean sample and sign. The authors recommend that best estimate be used to
the standard deviation of X based on n observations, and designate the statistical mean obtained by methods similar
tn-1(α) is a factor tabulated in statistical textbooks. DNV13 to those described herein. The use of the expressions lower
gives an example where n = 22 observations of the undrained bound and upper bound should be discouraged, but one
shear strength, su, in a soil deposit are available. The mean should express the uncertainties (or the variability) as the
value is estimated to be µsu = 60.2kPa, and the standard de- standard deviation or the CoV.
viation is estimated to be s = 10.2kPa. In the DNV example,
the mean value of the undrained shear strength is to be esti- There should be a series of characteristic profiles for differ-
mated with 90% confidence. For n = 22 and (1 – α) = 90%, ent design cases, e.g. skirt penetration, stability of skirted
the value of tn-1(α) is 1.33. Thus one obtains gravity foundations, seismic response, etc. The characteris-
tic profile could include a range or a standard deviation to
(16) account for uncertainty and specific modifications to ac-
count for shape or depth effects. The standard deviation
considered should represent the total uncertainty, and not
If the characteristic estimate of a soil parameter, X, is de-
only the statistical uncertainty represented by the standard
fined as some quantile in the distribution of X, then the
deviation of the data about the mean trend. This means that
following expression for the estimate of the characteristic
the uncertainty both in the measurements and due to the
value with confidence of 1 – α applies13
limited number of measurements should also be evaluated
(17)
and represented in the standard deviation.
where and s are, respectively, the sample mean sample and When determining the characteristic undrained shear
standard deviation of X based on n observations, and c1-α(n) strength, using ±0.5 or ±1 standard deviations in the analy-
is a factor that is obtained from the theory for tolerance ses will depend on what the parameters will be used for
bounds. DNV13 provides the values of c1-α(n) for XC defined (e.g. stability analysis versus skirt penetration analysis). A
as the 5% quantile of X, as well as the equations for param- wider bound would be logical if one is mainly concerned
eter estimation with confidence for dependent variables. with the punctual values of the undrained shear strength. If
A problem often encountered in practice in the offshore the undrained shear strength is averaged over a larger depth
environment is scarcity of data coverage in the lateral and/or volume of soil, the wide range is, however, probably
direction(s). In most situations, the number of geotechnical too conservative because of spatial averaging effects.
boreholes or CPTUs within a soil layer is too few to perform When the parameters are based on high quality laboratory
any sort of meaningful analysis of confidence intervals. The or reliable in situ tests, the following are suggested
high confidence of the geotechnical engineer in the recom-
• Design over large volumes of soil: the characteristic un-
mended characteristic values, despite the insufficient spatial
drained shear strength should be set equal to the mean
data coverage, is often justified by judgement, experience
–0.5 standard deviation
and understanding of the geological setting. This is a classical
problem in the Bayesian approach to estimation, which is an • Design of installations over smaller areas or volumes
alternative to estimation by the classical statistical approach. (e.g. penetration of short skirts): one should use a char-
The Bayesian approach allows for inclusion of information acteristic value equal to the mean –1 standard deviation
other than sample data and for updating of prior informa- • Design based on less reliable test data: the range should
tion when new additional information becomes available. be 1 to 2 standard deviations, depending on the volume
Further details are provided in e.g. Ang and Tang9. of interest

459

OSIG final.indb 459 09/08/2007 00:08:20


Lacasse, Guttormsen, Nadim, Rahim and Lunne. Use of Statistical Methods for Selecting Design Soil Parameters

• The stability assessment of natural submarine slopes: the 3. Uzielli M, Lacasse S, Nadim F and Phoon KK. (2006). Soil
mean undrained shear strength should be used. variability analysis for geotechnical practice. In Tan TS, Phoon
KK, Hight DW and Leroueil S. (eds.), Characterization and
These definitions and suggestions are set forth as a subject Engineering Properties of Natural Soils, Vol. 3. London: Taylor and
for debate in the coming months so that the issue can be Francis, 1653–1752.
discussed and agreement can be reached for use in practice. 4. Lacasse S and Nadim F. (1996). Uncertainty in characterizing
soil properties, In Shackleford CD, Nelson PP and Roth MJS
The authors believe that the geotechnical profession should (eds.), Uncertainty in the Geologic Environment: From Theory and
use statistical analysis of the parameters more extensively Practice, SSP 58. New York: American Society of Civil Engineers,
and more routinely. Patterns emerge on the relationship be- 49–75.
tween characteristic soil parameters and statistically derived 5. Lunne T, Berre T, Andersen KH, Strandvik S and Sjursen M.
ones. The authors recommend, however, that the profession (2006). Effects of sample disturbance and consolidation proce-
should look beyond simple statistical analyses and address dures on measured shear strength of soft marine Norwegian clays.
the variability and uncertainty in soil parameters explicitly Canadian Geotech. J. 43, 726–750.
by implementing probability theory and reliability analy- 6. Norwegian Geotechnical Institute (NGI). (2002). Quality of
ses. These analyses provide a rational framework for dealing CPTU. Statistical analyses of CPTU data from Onsøy. NGI
Report No. 20011099-2, 5 September.
with uncertainties and decision-making under uncertainty.
Depending on the level of sophistication, the analyses pro- 7. Karlsrud K, Lunne T, Kort DA and Strandvik S. (2005). CPTU
correlations for clays. Proc. Int. Conf. Soil Mech. and Geotech.
vide the following output: probability of failure or unsat- Engng, Osaka, Vol. 2, 693–702.
isfactory performance, reliability index, the most probable
8. Lunne T, Randolph MF, Chung SF, Andersen KH and Sjursen
combination of parameters leading to failure and sensitivity MA. (2005). Comparison of cone and T-bar factors in two onshore
of result to any change in parameters. and one offshore clay sediments. In Gourvenec S and Cassidy M
(eds.), Proc. Int. Symp. on Frontiers in Offshore Geotechnics: ISFOG
Acknowledgment 2005. London: Taylor and Francis, 981–989.
The authors wish to acknowledge the help from Katherine 9. Ang AHS and Tang WH. (1975). Probability Concepts in Engineering
Feeley, on leave at NGI from Northeastern University, with Planning and Design. New York: Wiley and Sons, 409p.
the statistical analyses. 10. Lunne T, Robertson PK and Powell JJM. (1997). Cone Penetration
Testing in Geotechnical Practice. London: Spon Press, 312p.
References 11. Ladd CC, Foott R, Ishihara K, Schlosser F and Poulos HG.
1. Det Norske Veritas (DNV) and Norsk Hydro. (2006). Technical (1977). Stress deformation and strength characteristics, SOA
Report No. 2006-1437: Guidance Note Statistical Representation Report. Proc. ICSMFE Tokyo, Vol. 2, 421–494.
of Soil Data, 30 August. 12. Snedecor GW and Cochran WG. (1964). Statistical methods.
2. Lacasse S, Nadim F, Rahim A and Guttormsen TR. (2007). Ames, IA: Iowa State University Press, 140 p.
Statistical description of characteristic soil properties. Offshore 13. DNV. (2007). Recommended Practice DNV-RP-C207: Statistical
Tech. Conf., Houston, USA, OTC 19117. Representation of Soil Data, April.

460

OSIG final.indb 460 09/08/2007 00:08:20

S-ar putea să vă placă și