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

net/publication/260228707

Reliability Analysis of CPT Measurements for Calculating Undrained Shear

Strength

Article in Geotechnical Testing Journal · November 2011

DOI: 10.1520/GTJ103771

CITATIONS READS

3 1,913

4 authors, including:

Khalid A. Alshibli Ayman M Okeil

University of Tennessee Louisiana State University
147 PUBLICATIONS 4,072 CITATIONS 98 PUBLICATIONS 1,292 CITATIONS

SEE PROFILE SEE PROFILE

Bashar Alramahi
ExxonMobil
22 PUBLICATIONS 208 CITATIONS

SEE PROFILE

All content following this page was uploaded by Khalid A. Alshibli on 19 August 2015.

The user has requested enhancement of the downloaded file.

 Geotechnical Testing Journal, Vol. 34 , No. 6
Paper ID GTJ103771
Available online at: www.astm.org

Khalid A. Alshibli,1 Ayman M. Okeil,2 Bashar Alramahi,3 and Zhongjie Zhang4

Reliability Analysis of CPT Measurements for

Calculating Undrained Shear Strength

ABSTRACT: The cone penetration test (CPT) has been widely used in Louisiana to classify soils, measure undrained shear strength (Su), and
identify bearing stratum for driven piles. This paper compares the values of Su based on CPT measurement with Su of the unconfined compression
test. A total of 752 CPT soundings were collected and archived using ArcGIS software in which 503 were matched with adjacent boreholes and
249 did not have adjacent borehole data available. The dataset was analyzed for general as well as specific trends in order to identify appropriate
parameters to be included in the investigation. The calibration of the CPT expression for Su was conducted using the first order reliability method
(FORM) and accounting for all sources of uncertainty. Optimum CPT coefficient (Nkt ) values to calculate Su were computed for various target
reliability values. It was determined that the soil classification is the only parameter showing clear trends that affect CPT estimates of the undrained
shear strength. Values of Nkt for each soil type based on the Robertson (1990) classification and the Zhang and Tumay (1999) classification were
determined for three target reliability levels. It is obvious that the Nkt coefficient for soils with higher clay content is lower than those with less clay
content. A single Nkt value that is valid for all soil types is unwarranted as will lead to acceptable results for some soil conditions and unacceptable
results for others, which can be unconservative.
KEYWORDS: undrained shear strength, cone penetration, soft soils, reliability, LRFD

Introduction 1983b), estimate the in situ stresses, compressibility, permeability,

and other soil properties.
Shear strength of soils is a key parameter in many applications. The CPT involves pushing a series of metal rods with a
The standard methods to determine shear strength of soils are cone-shaped tip into the ground at a constant penetration rate.
broadly classified as laboratory (e.g., triaxial, unconfined com- During this process, load sensors measure the amount of force
pression, direct shear, and California Bearing Ratio) and in situ required to penetrate different soil layers. Mainly, two force
procedures (e.g., dynamic cone penetration (DCP), vane shear components are measured during a CPT: the force acting on the
test, plate load test, cone penetration test (CPT), and standard pen- cone (Qc) and the total combined force acting on the cone and
etration test (SPT)). In situ procedures generally provide robust cylindrical friction sleeve located behind the tip (Qt). The cone
and convenient measurements of soil shear strength as compared resistance (qc) is calculated as the force (Qc) divided by the pro-
to laboratory procedures. However, they are less accurate and in jected area of the cone (Ac), while the sleeve friction (fs) is cal-
some cases are based on empirical models. In the last three deca- culated as the net force acting on the friction sleeve (Qt-Qc)
des, the CPT has gained popularity as a fast, inexpensive, and divided by the surface of the sleeve (As). Another parameter
fairly accurate method for in situ characterization of sub-surface commonly calculated from CPT data is the ratio of the sleeve
soil layers. This is attributed to its ability to obtain nearly continu- friction to the tip resistance, known as the friction ratio. This pa-
ous measurements providing a much higher data resolution than rameter is particularly useful in soil classification because differ-
standard soil sampling procedures. The data obtained from a CPT ent soil types exhibit different relative amounts of tip resistance
sounding can be used to classify soils (e.g., Robertson 1990; and sleeve friction.
Zhang and Tumay 1999), determine the soil shear strength (e.g., The Louisiana Department of Transportation and Development
Villet and Mitchell 1981; Robertson and Campanella 1983a, (LA-DOTD) has been one of the leading States in developing and
implementing the CPT technology. In most cases, the CPT sound-
ings are complemented with drilling boreholes and carrying out
Manuscript received January 19, 2011; accepted for publication April 11, the corresponding laboratory experiments (classification, physical
2011; published online July 2011.
1
Professor, Department of Civil and Environmental Engineering, University
properties, consolidation, and shear strength). The wealth of CPT
of Tennessee, Knoxville, TN 37996, e-mail: [email protected] measurements in Louisiana warrant a thorough analysis and eval-
2
Associate Professor, Department of Civil and Environmental Engineering, uation of such measurements and compare them with results
Louisiana State University, Baton Rouge, LA 70803, e-mail: [email protected] acquired using laboratory testing of specimens retrieved using
3
ExxonMobil, Upstream Research Company, Houston, TX 77098, e-mail: soil boring techniques. The objective of this paper is to assess the
[email protected]
4
Pavement and Geotechnical Research Administrator, Louisiana
reliability of CPT in predicting the undrained shear strength (Su)
Transportation Research Center, 4101 Gourrier Avenue, Baton Rouge, LA of soils and comparing it with Su of the unconfined compression
70808, e-mail: [email protected] test.

Copyright
Copyright
C 2011 by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959.
V by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012 1
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
2 GEOTECHNICAL TESTING JOURNAL

Background cone tip resistance and sleeve friction. This paper focuses on the
undrained shear strength which is applicable to clayey soils.
Soil Classification Using CPT Therefore, the predicted probability of clay percentage according
to the Zhang and Tumay (1999) method is chosen as a classifica-
Data obtained from a CPT can be used to classify the soils along tion parameter. Also, the soil classification using the Robertson
the path of the cone. It was observed that different types of soils (1990) method that provides soil classifications as one of six pos-
exhibit distinctive responses during cone penetration making it sible types is investigated. The Robertson method classifications
possible to classify the soils based on their response. For example, are two for organic soils and peats, three for clays (silty clay to
while sandy soils are characterized by high cone resistance and clays), four for silt mixtures (clayey silt to silty clay), five for sand
low friction ratio; soft clays produce low cone resistance and high mixtures (silty sand to sandy silty), six for sands (clean sand to
friction ratio (Lunne et al. 1997). Begemann (1965) noted that the silty sand), and seven for (gravelly sand to dense sand). The clas-
soil type is not a strict function of the tip resistance or the sleeve sifications of interest to this paper are two through four where the
friction but rather a combination of these values. Several efforts undrained shear strength is relevant.
were made to present a dependable classification chart using cone
penetrometer data (e.g., Begemann 1965; Sanglerat et al. 1974; CPT Models to Calculate Soils Undrained Shear
Schmertmann 1970; Douglas and Olsen 1981; Robertson et al.
Strength
1986). In order to perform a more accurate classification scheme,
Robertson et al. (1986) suggested classification charts based on qt, Since cone penetration is a complex phenomenon, all theoretical
fs and the pore water pressure parameter. These charts classify the solutions make several simplifying assumptions regarding soil
soils based on their response to 12 distinct soil types and provide behavior, failure mechanism, and boundary conditions. Such solu-
information about the relative density and over-consolidation ratio tions need to be verified using actual field and=or laboratory test
(OCR) of the soils. data. They have limitations in modeling the real soil behavior
Eslami and Fellenius (1997) presented a classification chart under conditions of varying stress history, anisotropy, sensitivity,
based on data obtained from 20 sites in 5 countries. They used an aging, and micro fabric. Hence, empirical correlations are gener-
“effective” tip resistance parameter (q0 ¼ qt  u, where u is pore ally preferred although the theoretical solutions have provided a
water pressure measured at the joint of the cone tip and the friction useful framework of understanding. Many researchers attempted
sleeve) to provide a more consistent delineation of envelopes than to develop a dependable correlation between the parameters
a plot of only the cone resistance. After comparing several soil obtained from cone penetration tests and the undrained shear
classification methods using CPT data, Fellenius and Eslami strength of cohesive soils. Some of the presented correlations
(2000) concluded that the classification methods that do not cor- were based on theoretical solutions such as the bearing capacity
rect for the pore pressure on the cone shoulder may not be relevant theory, cavity expansion method, strain path method, and numeri-
outside the areas where they were developed, and the error due to cal methods using linear and non-linear soil models (e.g., Hill
omitting the pore water pressure is highest in fine-grained soils. 1950; Vesic 1972; Baligh 1985; Teh and Houlsby 1991; Abu-
Zhang and Tumay (1999) argued that due to complicated envi- Farsakh et al. 1998). Other correlations, however, were empiri-
ronmental conditions, the correlation between soil and mechanical cally developed by comparing the CPT results with laboratory
properties would never be a simple one-to-one correspondence. shear strength experiments. For example, the undrained shear
They also indicated that the CPT classification charts do not pres- strength (Su) is expressed as a function of the cone tip resistance
ent an accurate prediction of soil types based on compositional (qc) using the following relationship
properties but rather a guide to soil behavior type. They suggested qc  r
Su ¼ (1)
an alternative classification method using statistical and fuzzy sub- Nc
set approaches to calculate a soil classification index (U) from
where:
CPT data that are used to determine probable unified soil classifi-
Nc is a theoretical cone factor and
cation system (USCS) (compositional) soil types. This alternative
r is the in situ total overburden pressure.
classification method utilizes the fuzzy subset approach which
Depending on the theory used to calculate Nc, r can be the
aims to release the constraint of soil composition and put empha-
total vertical overburden (rvo), horizontal (rho), or mean (rmean)
sis on the soil behavior instead (Kaufmann 1975; Brown et al.
stress. Lunne et al. (1997) presented a similar model to predict the
1985). Three empirically defined density functions that corre-
undrained shear strength using the total cone resistance
spond to three soil groups are presented: highly probable clay
qc  rvo
(HPC), highly probable mixed (HPM), and highly probable sand Su ¼ (2)
(HPS). The three functions as a whole will reflect the overall per- Nk
spective of soil properties. This results in a classification that does where Nk denotes the cone factor that includes the influence of
not yield sharp boundaries between layers but rather a smooth the cone shape and depth factor.
transition from one soil type to another. This statistical approach Its value typically ranges from 11 to 19 for normally consoli-
has been adopted by LA-DOTD as a standard procedure for the dated clays according to Lunne et al. (1997). Kjekstad et al.
interpretation of CPT data. (1978) reported a Nk value of 17 for over-consolidated clays.
The CPT data used in this paper do not have pore water pres- Equation 2 was later modified where the corrected cone resistance
sure measurements; therefore, soil classifications are based on the (qt) was used as follows:

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
 ALSHIBLI ET AL. ON RELIABILITY ANALYSIS OF CPT MEASUREMENTS 3

qt  rvo For illustration purposes, a scenario of two boreholes at distan-

Su ¼ (3)
Nkt ces D1 ¼ 10.6 m and D2 ¼ 33.5 m would yield the following
weighted average
Many experiments were performed to estimate the value (or
range of values) for Nkt (e.g., Aas et al. 1986; La Rouchelle et al. q ¼ 0:7586q1 þ 0:2414q2 (5)
1988; Rad and Lunne 1988). A wide range of values were
reported depending on the type of cone, field conditions, penetra-
tion rate, and laboratory testing method. However, most of the Averaging CPT Readings
reported values of Nkt ranged between 10 and 30. CPT readings are known to have spikes due to many factors. One
of these factors is the fact that continuous readings are recorded at
very small intervals which is in many cases smaller than 25 mm
Data Collection and Processing of depth. Relying on pinpoint data readings would result in
The CPT data reported in this paper were collected from unnecessary scatter, especially that soil properties obtained from
LA-DOTD project archives in the form of paper documents con- borehole tests are usually based on specimens of tangible dimen-
taining CPT soundings, borehole logs, and project maps. The next sions (200 mm in the case of Shelby Tube sample). As a result, it
step was to match the CPT data with adjacent borehole data. was deemed necessary to average CPT data readings over a depth
Therefore, for every CPT point, one or more boreholes were iden- similar to what are used in borehole soil property testing. In this
tified based on the distance from the CPT to allow for the compar- paper, 11 consecutive CPT data readings were averaged around
ison of parameters obtained from boreholes to CPT. A total of 752 every desired depth where borehole data were available. These
CPT soundings were documented, of which 503 were matched readings correspond to a depth of about 198 mm (7.8 in.). Figure 1
with adjacent boreholes. All the CPT measurements were con- illustrates the averaging of the device readings for the cone tip re-
ducted using a friction cone with a cross-sectional area of 10 cm2. sistance. Figure 1(a) depicts the raw data obtained from the CPT
All undrained shear strength measurements reported in the bore- device, while Fig. 1(b) shows the same results after averaging the,
hole logs were calculated from the unconfined compression tests over a depth of 198 mm (7.8 in.). It can be seen that localized
(ASTM-D2166) that were conducted by LA-DOTD laboratories spikes are almost eliminated without losing the general trend of
and consulting agencies. The test sites represent 52 projects that the readings. This is the purpose of averaging and it was done for
cover many parts of the state of Louisiana with different geologi- all CPT readings in the database.
cal and geomorphological features (see Alshibli et al. 2008 for
more details about projects and their locations). Filtering Data Results
As expected, the CPT and borehole locations were not exact
matches with respect to the location. This was due to many site As is the case with any natural and hence uncontrollable material,
factors and is now considered historic data that are not changea- a wide range of readings is always expected. Soils fall under this
ble at the time of conducting this research. Thus, it was impera- category of materials. Nevertheless, soil properties are known to
tive that acceptable criteria be set to justify the hypothesis of fall within a confined physical range albeit it may be a wide range.
equal soil properties at both the CPT and borehole locations. Values outside this range represent soil properties that represent
These criteria address the fact that the distance between the soil conditions that may lead to outcast data points. For example,
CPTs and boreholes ranged from 0.3 m (1 ft) to 562 m (1844 ft). extremely low undrained shear strength results indicate very soft
At the same time, more than one borehole existed in the vicinity soils that are difficult to test under normal laboratory conditions,
of some CPTs in the database, and a weighing function was thus results within this low range are in many cases unrepresenta-
needed for these locations. There is no published guidelines that tive of actual soil conditions. At the other extreme, high undrained
set a distance limit to associate a borehole measurements with shear strength values are usually attributed to very stiff clays,
CPT. As a result, the following criteria were set in this project to where fluctuations in the cone readings may be excessive leading
address these issues: to unreliable CPT estimates of the undrained shear strength. It was
therefore deemed necessary to set limits on the range of undrained
• A maximum distance between the CPT and borehole loca- shear strength, Su, collected in the database. Values that exceed a
tion was set to be 45.7 m (150 ft). maximum threshold or fall below a minimum threshold were
• In the case of having more than one borehole in the vicinity excluded from the database. Based on the literature, Su values fall-
of a CPT location, a 1=D weight function was used to ing below 7.18 kPa (150 psf) and above 119.7 kPa (2500 psf)
account for the proximity of borehole locations to CPT were excluded from the database in subsequent analyses.
locations. The weighted average function for any quantity, Figure 2(a) shows a histogram of the undrained shear strength
q, based on the results from n boreholes at distances D1, complied in the database. The plot shows that the database
D2, …, Di…, Dn can be expressed as includes points of extremely low and high Su values. Su values
2 0n 13 falling below 150 psf and above 2500 psf were excluded from the
P
n 6 B
Dj  Di C7 database in subsequent analyses. Figures 2(b) and 2(c) shows the
X 6 Bj¼1 C7
q¼ 6q i B P n C7 (4) impact the chosen threshold has on the scatter of the transforma-
4 @ A5
i¼1 Dj tion model used in this study. The scatter is measured
 by the coef-
j¼1 ficient of variation (COV), of the ratio SuCPT SuUC . It can be seen

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
4 GEOTECHNICAL TESTING JOURNAL

FIG. 1—Illustrative example showing averaging CPT values (a) raw tip resistance and (b) averaged tip resistance.

that as more data points with high or low undrained shear strength (2009) investigated the variability of CPT measurements by con-
values are excluded, the smaller the COV of the transformation ducting 16 CPT penetrations in proximity to each other to a depth
model becomes. Filtering the data using these two threshold of 24.4 m (80 ft). A reliability analysis was conducted to deter-
results in a reduction in the number of data points used in the sub- mine the coefficient of variation of qc as well as soil unit weight
sequent analyses. The database included 862 data points after (c) and overburden pressure (rvo). The results showed that COV
applying the aforementioned thresholds. (qc) is higher when compared to COV(c) and COV(rvo) with val-
ues close to 40% in some cases. On average, the COV for qc, c
and rvo are 19.6, 1.46, and 0.51%, respectively. These parameters
are incorporated in the reliability analysis as will be explained in
Uncertainty in Soil Properties
“Reliability Analysis of CPT”.
There are two main approaches for addressing uncertainties in
geotechnical applications. In the first approach, uncertainties are
dissected to their main sources and each source is investigated
Reliability Analysis of CPT
methodically in a rational way. The main sources of uncertainty
for geotechnical applications include inherent soil variability, Most civil engineering design codes have adopted the load and re-
measurement errors, and expression (transformation model) sistance factor design (LRFD) philosophy and moving away from
uncertainty. Figure 3 shows a schematic of these sources. Several the allowable (working) stress design (ASD) approach. In the
researchers used this approach to study geotechnical design within USA, Building Code Requirements for Concrete Structures (ACI
a reliability-based framework (e.g., Phoon and Kulhawy, 1999a, 318), steel construction (AISC-LRFD), and highway bridges
1999b). Conversely, reliability studies for geotechnical applica- (AASHTO-LRFD (2004)) are all based on the LRFD design phi-
tions are often conducted at the foundation capacity level rather losophy. Despite the extensive efforts in this field, the geotechni-
than the soil property level. In this second approach, the resistance cal profession has not fully accepted the new load and resistance
of the component is investigated by comparing analytical esti- factors developed for AASHTO-LRFD. This is one of the most
mates with experimental results for which a database of available difficult hindrances of LRFD to implement in geotechnical appli-
results is assembled. This approach helps in overcoming the com- cations. As a result, more research studies have been conducted to
plexity and lack of information that may exist at the soil property alleviate doubts and overcome the reluctance in the geotechnical
level. community. For example, Allen (2005) developed geotechnical
One of the identified sources of uncertainty is the inherent vari- resistance factors and downdrag load factors for LRFD foundation
ation in the device and measurement equipment. This uncertainty strength limit state design based on the information used in
needs to be accounted for in the reliability analysis forming the NCHRP Projects 12-04 and 12-17. These efforts cover a wide
basis for the correlation of the CPT measurements. Alshibli et al. range of geotechnical applications including shallow foundations,

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
 ALSHIBLI ET AL. ON RELIABILITY ANALYSIS OF CPT MEASUREMENTS 5

of design coefficients that help account for excessive overloading

conditions (load factors) and unforeseen strength deficiencies (re-
sistance factors). In general, the main design equation is usually
given as
X
/Rn  ci Qi (6)
i

where Rn is the nominal capacity of the designed member, / is

a resistance factor, Qi is the demand due to applied loads (dead,
live, etc.), and ci is the load factor. The development of LRFD
codes was initially started for structural applications. The resist-
ance of structural components has variabilities associated with it;
however, more uncertainty is inherent in loads than in resistance.
Geotechnical applications are unique in the sense that uncertain-
ties are high on the resistance side as well. Nevertheless, LRFD
research ensued in the geotechnical engineering community early
on to develop a rational design methodology based on available
statistical information of geotechnical applications.
Several researchers studied the reliability of CPT as a feasible
tool for identifying the probability of soil liquefaction for a given
factor of safety. For example, Liu and Chen (2006) used data
from 49 sites (71 cone penetrations) to develop statistical parame-
ters for cone tip resistance, qc, sleeve friction, fs, groundwater ta-
ble and soil unit weight. This information was then used to treat
each parameter as a random variable in analyzing both demand
(cyclic resistance ratio – CRR) and resistance (cyclic stress ratio –
CSR). Based on this information, Monte Carlo simulations were
conducted to determine soil properties in unsampled sites, which
were then used to map the liquefaction potential.
Juang et al. (2006) also studied the liquefaction potential using
CPT results after accounting for parameter and=or model uncer-
tainties. The first-order reliability method (FORM) was used to
FIG. 2—(a) Histogram of Su frequency of compiled database, (b) and (c) Effect estimate the reliability index, b, based on a limit state function
of thresholds on uncertainty of transformation model. involving CRR, CSR, and the model uncertainty. Each of the six
input parameters (qc, fs, rv, r0v , amax, and Mw) was considered to
deep foundations, and retaining structures (Allen 2005; Phoon be a random variable and correlation between these parameters
et al. 2003; Phoon and Kulhawy 2005; Abd Alghaffar and was accounted for. The study used a database of 96 liquefaction
Dymiotis-Wellington 2005; Low 2005; Goh and Kulhawy 2005; case histories to determine these parameters, where they consid-
Fenton et al. 2005). ered spatial variations using Bayesian mapping functions. The
LRFD design codes appear to be deterministic to their user; authors concluded that traditional liquefaction boundaries are of-
while in fact they are probabilistically calibrated to ensure that a ten based on the conservative side, which may lead to erroneous
target risk level is not exceeded. This is achieved using a number post-earthquake investigations.

FIG. 3—Sources of uncertainty in estimating soil properties (Kulhawy 1992).

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
6 GEOTECHNICAL TESTING JOURNAL

Calibration of CPT Coefficient

LA-DOTD uses Eq 3 (referred to as the transformation model
hereafter) to calculate Su from CPT measurements. The compiled
database of CPT results was used to calibrate the transformation
model of Su for Louisiana soils. This database was obtained from
projects all over Louisiana. The goal of the paper is to calibrate
the transformation model utilized by geotechnical engineers by
comparing CPT results to corresponding results from borehole
tests. All reported Su values in the borehole logs are based on the
unconfined compression test (UC) (denoted as SuUC ). Each CPT
test provided several data points to the database. The soil proper-
ties were computed at the depth intervals that match the reported
borehole results. In all, 2630 data points for SuUC values from bore-
holes were compiled. It was only possible to compute Su from
CPT readings in 1886 locations and depths. However, the number
of data points where Su results were available from matching CPT
and borehole data were 1814. By applying the maximum distance
limit discussed earlier, the final number of data points included in
subsequent analyses dropped to 1263 from 251 CPT locations.
The calibration of the CPT coefficient, Nkt (Eq 3) to achieve ac-
ceptable reliability in Su estimates was conducted using the first
order reliability method based on a limit state function as
described next. An acceptable reliability can be defined in terms
of a probability of exceedance (Pe) which is described in detail in FIG. 4—Illustration of the limit state function (LSF).
“Limit State Function”. The statistical characteristics of the data
in the compiled database of CPT results first had to be analyzed. the origin, which can be represented in multiples of the standard
These characteristics are essential for the reliability-based calibra- deviation, rZ. This multiple is defined as the reliability index (b).
tion. Statistical characteristics of random variables not analyzed in The higher the reliability index, the higher the probability of
this paper were obtained from the literature. The calibration was exceedance (Pe).
conducted for the sole coefficient in the transformation model The reliability calibration conducted in this paper is based on
(Eq 3), Nkt, which was performed for different soil classifications, the limit state function (LSF) described above. However, the CPT
different target reliability levels, and rvo=qc ratio. estimate, SuCPT , will be replaced by the transformation model used
to obtain its values (Eq 3). This allows for including the uncertain-
ties inherent in the original CPT readings, namely qc and rvo.
Limit State Function Equation (7) shows the details the LSF starting with the random
The limit state function is simply the boundary between accepta- variables SUCu and SCPT
u . The expanded version is shown in
ble and unacceptable performance. In this paper, the undrained Eq (7b), where SUC UC
u is replaced by cSu in which c is random vari-
shear strength estimates from the unconfined compression (bore- able that accounts for uncertainty inherent in the soil property,
hole data), SuUC , and CPT data, SuCPT , should be identical in ideal SuUC , which is taken as a deterministic value representing soils in
circumstances. In real life, the difference between both is inevita- location under investigation. The other two newly introduced ran-
ble since both estimates have inherent randomness in their values. dom variables, f and g, represent the uncertainties in the transfor-
An acceptable performance is achieved if CPT estimates for SuCPT mation model and tip resistance readings, respectively
is less than the borehole result, SuUC , and an unacceptable perform-
Z ¼ SUC
u  Su
CPT
(7a)
ance is to be expected if the opposite is true. If the formula used in  
the determining SuCPT , transformation model in Eq 3 is conserva- UC gqc  rvo
Z ¼ cSu  n (7b)
tive, then the mean value of CPT estimates, lSuCPT , would be less Nkt
than the mean value of borehole results, lSuUC (Fig. 4(a)). Alterna- Table 1 lists the statistical characteristics of the random variables
tively, this can be described using a limit state function (LSF) that used in the analysis. The characteristics of the random variable f
represents the difference between both estimates (Z ¼ SuUC are given in Tables 2 and 3 for both classification techniques;
 SuCPT ), which will also be a random variable that can be plotted namely Robertson (1990), Zhang and Tumay (1999). The bias and
as can be seen in Fig. 4(b). The probability of estimating the standard deviation values were computed by determining the statis-
undrained shear strength from borehole unconfined compression tical descriptors (mean and standard deviation) for the ratio between
tests that will be higher than those obtain from the CPT readings the CPT estimates and borehole results, SuCPT =SuUC , in the database
is graphically represented by the shaded area in Fig. 4(b). It can  CPT 
be seen that the larger this area, the safer the CPT estimates. An S
bias; kn ¼ mean uUC (8)
indirect measure of the area is the distance from the mean, lZ, to Su

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
 ALSHIBLI ET AL. ON RELIABILITY ANALYSIS OF CPT MEASUREMENTS 7

TABLE 1—Random variables used in reliability calibration. TABLE 3—Statistical characteristics of transformation model (Zhang and
Tumay (1999) classification).
Variable Mean COV (%) Distribution Source
ALL Cases Clay > 75% Clay ¼ 50%–75% Clay ¼ 25%–50%
Soil uncertainty, c 1.0 33 Lognormal Phoon and
Kulhawy (1996b) Nkt Bias STDEV Bias STDEV Bias STDEV Bias STDEV
Transformation Varies Lognormal Current investigation
12 2.401 1.693 2.172 1.387 2.426 1.981 2.662 1.906
model, f
15 2.009 1.427 1.822 1.124 1.954 1.476 2.273 1.774
Tip resistance, g Varies Normal Alshibli et al. (2009)
18 1.793 1.418 1.614 1.038 1.776 1.732 1.946 1.521
Overburden pressure Deterministic Alshibli et al. (2009)
21 1.562 1.218 1.414 0.891 1.521 1.441 1.675 1.297
  24 1.414 1.154 1.258 0.825 1.357 1.255 1.542 1.251
SuCPT 27 1.285 1.126 1.129 0.737 1.216 1.120 1.426 1.419
standard deviation SuUC
coefficient of variation, COVðnÞ ¼  CPT  30 1.201 1.125 1.053 0.728 1.130 1.017 1.368 1.545
S
mean SuUC 33 1.100 1.026 0.958 0.662 1.045 0.926 1.244 1.402
u
rn 36 1.023 0.971 0.892 0.614 0.959 0.851 1.143 1.281
¼ (9)
kn 39 0.961 0.924 0.828 0.569 0.886 0.788 1.075 1.210
42 0.906 0.914 0.769 0.529 0.823 0.732 0.998 1.124
Several CPT coefficient values were considered in determining
45 0.860 0.873 0.744 0.581 0.771 0.684 0.937 1.049
these statistical descriptors. For each value, the SuCPT would
48 0.817 0.825 0.701 0.545 0.726 0.643 0.892 0.990
change and cause the bias and coefficient of variation given in
51 0.773 0.777 0.660 0.513 0.694 0.609 0.842 0.933
Tables 2 and 3. The various values are needed for the optimization
step that follows for determining an appropriate CPT coefficient,
Nkt, that meets a target reliability level. Finally, the tip resistance
uncertainty was obtained from the repeatability study reported by how well the chosen mathematical distribution fits the collected
Alshibli et al. (2009) while the overburden pressure was taken as data (observations). Many statistical tests can be used to evaluate
a deterministic quantity due to its low COV (see Alshibli et al. how well a certain distribution fits the collected data. In this paper,
2009 for more details). the Chi-Square Test was used. The random variables were tested
for two possible distribution types; namely Normal and Lognor-
mal. The results from conducting these tests led to the choices
Chi-Square Statistical Test “Goodness of Fit” shown in Table 1 for the two random variables that were deter-
mined in this paper, f and g.
In addition to the statistical descriptors that were determined in
“Limit State Function” (bias, k, and coefficient of variation, Reliability-Based Calibration
COV), a distribution type was needed for the reliability study
used in the calibration process. Statistical distributions are mathe- The dataset was further analyzed by grouping data points in sub-
matical expressions that represent the frequency within a dataset groups based on different parameters associated with each point.
continuously over the possible range for the entire population. The parameters considered in this paper for grouping the data are
Any random variables can be described using different distribu- (1) depth, (2) soil classification (two different methods), and (3)
tion types. The choice of one distribution over another is based on cone tip resistance. It was determined that the soil classification is
the only parameter showing clear trends that affect CPT estimates
TABLE 2—Statistical characteristics of transformation model (Robertson
(1990) classification).
of the undrained shear strength.
Calibration of the CPT coefficient, Nkt, was conducted using
ALL Cases 2 3 4 the LSF described earlier. The goal is to identify Nktvalues that
Nkt Bias STDEV Bias STDEV Bias STDEV Bias STDEV
result in desired reliability levels, which will be measured in terms
of a reliability index, b, defined as the ratio between the mean
12 2.401 1.693 1.541 1.196 2.284 1.366 2.484 1.652
value of the LSF and its standard deviation, which was previously
15 2.009 1.427 1.358 0.962 1.899 1.268 2.067 1.299
illustrated in Fig. 4(b)
18 1.793 1.418 1.178 0.806 1.604 1.048 1.881 1.252
21 1.562 1.218 1.041 0.680 1.390 0.898 1.612 1.055
24 1.414 1.154 0.964 0.580 1.229 0.792 1.482 1.108 TABLE 4—Range of parameters covered in reliability investigation.
27 1.285 1.126 0.879 0.512 1.093 0.705 1.329 0.982
30 1.201 1.125 0.791 0.461 0.991 0.633 1.220 0.916 Parameter Range
33 1.100 1.026 0.719 0.419 0.904 0.576 1.115 0.829 CPT coefficient, Nkt 12, 15, 18, 21, …, 39,
36 1.023 0.971 0.700 0.368 0.835 0.536 1.027 0.759 42, 45, 48
39 0.961 0.924 0.678 0.350 0.772 0.496 0.971 0.748 Ratio, rvo=qc 0.05, 0.10, 0.20, 0.50,
42 0.906 0.914 0.630 0.325 0.717 0.461 0.941 0.874 0.80, 0.90
45 0.860 0.873 0.609 0.294 0.673 0.431 0.900 0.866 Soil classification Robertson (1990) 2, 3, 4, 5
48 0.817 0.825 0.571 0.275 0.635 0.405 0.854 0.816 Zhang and Tumay (1999) >75%, 50%–75%,
51 0.773 0.777 0.554 0.254 0.600 0.381 0.806 0.767 25%–50%, <25%

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
8 GEOTECHNICAL TESTING JOURNAL

TABLE 6—Recommended Nkt values for Zhang and Tumay (1999)

classification.

Nkt Values for Different Soil Classifications

(Clay Probability)

Probability of Exceedance ALL > 75% 50%–75% 25%–50% <25%

50% (bT ¼ 0.0) 27.5 26.9 25.3 28.8 31.5
55% (bT ¼ 0.1257) 31.1 30.0 29.3 32.5 35.3
66.7% (bT ¼ 0.4308) 42.0 38.4 39.3 45.2 50.1

ability level. The Nkt results presented in Tables 5 and 6 are based
on the entire dataset compiled for this investigation. Values of Nkt
FIG. 5—Determining optimum Nkt values (bT ¼ 0.1257, rvo =qc ¼ 0.05, Rob- for each soil type based on the Robertson (1990) classification and
ertson Method).
the Zhang and Tumay (1999) classification were determined for
lz three target reliability levels. It is obvious that the Nkt coefficient
b¼ (10)
rz for soils with higher clay content is lower than those with less clay
b is related to the probability of exceedance, Pe, using the fol- content. Now designers can choose optimum Nkt values for differ-
lowing expression ent types of soils and satisfy a certain target reliability values.

Pe ¼ UðbÞ (11)
Conclusions
where U(b) is a cumulative distribution function (CDF) for a
limit state function, Z. The reliability index was evaluated for each The first order reliability method was used to conduct reliability anal-
considered case. In all, a total of 672 cases were considered as ysis for CPT measurements. It accounts for all sources of uncertainty
summarized in Table 4 (14 Nkt values  4 soil classifications (soil properties, device measurement, and transformation model) to
 6 rvo=qc ratios  2 classification methods). The ratio of over- calibrate the cone coefficient, Nkt, for achieving certain target reliabil-
burden pressure to the tip resistance, rvo=qc, in the reliability anal- ity levels. A single Nkt value that is valid for all soil types is unwar-
yses was considered in the reliability study since the results ranted as it will lead to acceptable results for some soil conditions
should cover a wide range of possibilities (design space). The and unacceptable results for others, which can be unconservative.
FORM was used to compute b for all cases. The FORM is based The reliability analysis yielded Nkt values equal to 27.5, 31.0, and
on a first order Taylor series expansion of the limit state function, 42.0 for target bT values equal to 0, (0%), 0.1257 (55%), and 0.4308
which approximates the failure surface by a tangent plane at the (66.7%), respectively, for the entire dataset; i.e., if soil classification
point of interest. Three target probability of exceedance (Pe) levels is not considered. Tables 5 and 6 list the recommended values Nkt
were investigated in this paper; they are Pe ¼ 50% (bT ¼ 0.0), based on soil type and probability of exceedance.
Pe ¼ 55% (bT ¼ 0.1257), and Pe ¼ 66.7% (bT ¼ 0.4308). An opti-
mization process was conducted to determine the optimum Nkt for Acknowledgments
the three levels of Pe as illustrated in Fig. 5 for the case of consid- The authors gratefully acknowledged the financial support pro-
ered cases. Summaries of the results of the analyses are listed in vided by the Louisiana Transportation Research Center (LTRC)
Tables 5 and 6. and Louisiana Department of Transportation and Development
The current practice by many State and Federal agencies is to (LADOTD).
assume a single Nkt value (e.g., 15 in Louisiana) to estimate SuCPT
using Eq 3 without consideration for soil type and risk level. As a References
matter of fact, many agencies plot the values of undrained shear
strength from different measurement sources (e.g., conventional Aas, G., Lacasse, S., Lunne, T., and Hoeg, K., 1986, “Use of In
unconfined compression, unconsolidated undrained triaxial com- Situ Tests for Foundation Design on Clay,” Proceedings of the
pression, and CPT) with depth. Such measurements exhibit a wide ASCE Specialty conference In Situ ‘86: Use of In Situ Tests in
scatter; then a trend line that splits the measurements into 2=3 ver- Geotechnical Engineering, Blacksburg, American Society of
sus 1=3 is used as the design parameter. Such approach does not Civil Engineers (ASCE), pp. 1–30.
account for the sources of uncertainty and a certain target of reli- AASHTO, 2004, LRFD Bridge Design Specifications. American
Association of State Highway and Transportation Officials,
Washington, D.C.
TABLE 5—Recommended Nkt values for Robertson (1990) classification.
Abd Alghaffar, M. A. and Dymiotis-Wellington, C., 2005,
Nkt Values for Different Soil Classifications “Reliability Analysis of Retaining Walls Designed to British
and European Standards,” Struct. Infrastruct. Eng., Vol. 1,
Probability of Exceedance ALL 2 3 4 No. 4, pp. 271–284.
50% (bT ¼ 0.0) 27.5 18.6 26.2 30.9 Abu-Farsakh, M. Y., Voyiadjis, G. Z., and Tumay, M. T., 1998,
55% (bT ¼ 0.1257) 31.1 21.0 28.6 34.3 “Numerical Analysis of the Miniature Piezocone Penetration
66.7% (bT ¼ 0.4308) 42.0 28.5 35.7 45.0
Tests (PCPT) in Cohesive Soils,” Int. J. Numer. Analyt. Meth.
Geomech., Vol. 22, No. 10, pp. 791–818.

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
 ALSHIBLI ET AL. ON RELIABILITY ANALYSIS OF CPT MEASUREMENTS 9

Allen, T. M., 2005, “Development of Geotechnical Resistance Liu, C. N. and Chen, C. H., 2006, “Mapping Liquefaction Poten-
Factors and Downdrag Load Factors for LRFD Foundation tial Considering Spatial Correlations of CPT Measurements,”
Strength Limit State Design,” Report No. FHWA-NHI–05– J. Geotech. Geoenviron. Eng., Vol. 132, No. 9, pp. 1178–
052, Federal Highway Administration, Washington, D.C. 1187.
Alshibli, K. A., Okeil, A., and Alramahi, B. 2008, “ Update of Low, B. K., 2005, “Reliability-Based Design Applied to Retaining
Correlations between Cone Penetration and Boring Log Data”, Walls,” Geotechnique, Vol. 55, No. 4, pp. 343.
Louisiana Transportation Research Center, December 2008, Lunne, T., Robertson, P. K., and Powell, J. J. M., 1997, Cone
pp. 147. Penetration Testing in Geotechnical Practice, Blackie Aca-
Alshibli, K. A., Okeil A. M., Alramahi, B., and Zhang, Z., 2009, demic, EF Spon=Routledge, NY.
“Statistical Assessment of Repeatability of CPT Meas- Phoon, K. K. and Kulhawy, F. H., 1999a, “Characterization of
urements,” Contemporary Topics in In Situ Testing, Analysis, Geotechnical Variability,” Can. Geotech. J., Vol. 36, No. 4,
and Reliability of Foundations, M. Iskander, D. Laefer, and M. pp. 612–624.
Hussein, Eds., ASCE, GSP # 186, pp. 87–94. Phoon, K. K. and Kulhawy, F. H., 1999b, “Evaluation of Geo-
Baligh, M. M., 1985, “Strain Path Method,” J. Geotech. Engrg. technical Property Variability,” Can. Geotech. J., Vol. 36, No.
Div., Vol. 111, No. 9), pp. 1108–1136. 4, pp. 625–639.
Begemann, H. K. S. Ph., 1965, “The Friction Jacket Cone an Aid Phoon, K. K. and , Kulhawy, F. H., 2005, “Characterisation of
in Determining the Soil Profile,” Proceedings of the 6th Inter- Model Uncertainties for Laterally Loaded Rigid Drilled
national Conference on Soil Mechanics and Foundation Engi- Shafts,” Geotechnique, Vol. 55, No. 1, pp. 45–54.
neering, University of Toronto Press, Montréal, Vol. 1, pp. Phoon, K. K., Kulhawy, F. H., and Grigoriu, M. D., 2003,
17–20. “Multiple Resistance Factor Design for Shallow Transmission
Douglas, B. J. and Olsen, R. S., 1981, “Soil Classification Using Line Structure Foundations,” J. Geotech. Geoenviron. Eng.,
Electric Cone Penetrometer. Cone Penetration Testing and Vol. 129, No. 9, pp. 807.
Experience,” Proceedings of the ASCE National Convention, Rad, N. S., and Lunne, T., 1988, “Direct Correlations Between
American Society of Civil Engineers (ASCE), St. Louis, pp. Piezocone Test Results and Undrained Shear Strength of
209–227. Clay,” Proceedings if The International Symposium on Pene-
Eslami, A. and Fellenius, B. H., 1997, “Pile Capacity by Direct tration Testing ISOPT–1, Orlando, Balkema, Rotterdam, Vol.
CPT and CPTu Methods Applied to 102 Case Histories.” Can. 2, pp. 911–917.
Geotech. J., Vol. 34, No. 6, pp. 880–898. Robertson, P. K., 1990, “Soil Classification Using the Cone Pene-
Fellenius, B. H. and Eslami, A., 2000, “Soil Profile Interpreted tration Test,” Can. Geotech. J., Vol. 27, No. 1. pp. 151–158.
from CPTu data,” Year 2000 Geotechnics, Geotechnical Engi- Robertson, P.K. and Campanella, R. G., 1983a, “Interpretation of
neering Conference, Asian Institute of Technology, Bangkok, Cone Penetrometer Test: Part I: Sand,” Can. Geotech. J.,
Thailand, November 27–30, 2000, pp. 18. Vol. 20 No. 4, pp.718–733.
Fenton, G. A., Griffiths, D. V., and Williams, M. B., 2005, Robertson, P. K. and Campanella, R. G., 1983b, “Interpretation of
“Reliability of Traditional Retaining Wall Design,” Geotechni- Cone Penetrometer Test: Part II: Clay,” Can. Geotech. J.,
que, Vol. 55, No. 1, pp. 55–62. Vol. 20, No. 4, pp. 734–745.
Goh, A. T. C. and Kulhawy, F. H., 2005, “Reliability Assessment Robertson, P. K., Campanella, R. G., Gillespie, D., and Greig, J.,
of Serviceability Performance of Braced Retaining Walls 1986, “Use of Piezometer Cone Data,” Proceedings of the
Using a Neural Network Approach,” Int. J. Numer. Analyt. ASCE Specialty Conference In Situ ‘86: Use of In Situ Tests in
Meth. Geomech., Vol. 29, No. 6, pp. 627–642. Geotechnical Engineering, American Society of Civil Engi-
Hill, R., 1950, The Mathematical Theory of Plasticity, Oxford neers (ASCE), Blacksburg. pp. 1263–1280.
University Press, Oxford, UK. Sanglerat, G., Nhim, T. V., Sejourne, M., and Andina, R., 1974,
Juang, C. H., Fang, S. Y., and Khor, E. H., 2006, “First-Order “Direct Soil Classification by Static Penetrometer with Special
Reliability Method for Probabilistic Liquefaction Triggering Friction Sleeve,” Proceedings of the First European Sympo-
Analysis Using CPT,” J. Geotech. Geoenviron. Eng., Vol. sium on Penetration Testing, ESOPT–1, June 5–7, Stockholm,
132, No. 3, pp. 337–350. Vol. 2.2, pp. 337–344.
Kulhawy, F. H. 1992, “On Evaluation of the Static Soil Proper- Schmertmann, J. H., 1970, “Static Cone to Compute Static Settle-
ties,” In Stability and Performance of Slopes and Embank- ment Over Sand,” J. Soil Mech. and Found. Div., Vol. 96, No.
ments II (GSP31), R. B. Seed and R. W. Boulanger, Eds., SM3, pp. 1011–1043.
ASCE, NY, pp. 95–115. Teh, C. I. and Houlsby, G. T., 1991, “An Analytical Study of the
Kaufmann, A., 1975, Introduction to the Theory of Fuzzy Cone Penetration Test in Clay,” Geotechnique, Vol. 41, No.1,
Subsets, I: Fundamental Theoretical Elements, Academic pp. 17–34.
Press, NY. Vesic, A. S., 1972, “Expansion of Cavities on the Infinite Soil
Kjekstad, O., Lunne, T., and Clausen, C. J. F., 1978, “Comparison Masses,” J. Soil Mech. and Found. Div., Vol. 98, No. SM3,
Between In Situ Cone Resistance and Laboratory Strength for pp. 265–290.
Overconsolidated North Sea Clays,” Mar. Georesources Geo- Villet, W. C. B. and Mitchell, J. K., 1981, “Cone Resistance, Rela-
technology, Vol. 3, No.1, pp. 23–36. tive Density and Friction Angle,” Cone Penetration Testing and
La Rouchelle, P., Zebdi, P. M., Leroueil, S., Tavenas, F., and Experience; Session at the ASCE National Convention, American
Virely, D., 1988, “Piezocone Tests in Sensitive Clays of East- Society of Engineers (ASCE), St. Louis, pp. 178–207.
ern Canada,” Proceedings of the International Symposium on Zhang, Z. and Tumay, M.T., 1999, “Statistical to Fuzzy Approach
Penetration Testing, ISOPT–1, Orlando, Balkema, Rotterdam, Toward CPT Soil Classification,” J. Geotech. Geoenviron.
Vol. 2 pp. 831–841. Eng., Vol. 125, No. 3, pp. 179–186.

Copyright by ASTM Int'l (all rights reserved); Thu Jan 12 13:01:08 EST 2012
Downloaded/printed by
Tennessee Univ pursuant to License Agreement. No further reproductions authorized.
View publication stats