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

net/publication/273023103

Load and Resistance Factor Design of Drilled Shafts in Sand

Article  in  Journal of Geotechnical and Geoenvironmental Engineering · December 2012

DOI: 10.1061/(ASCE)GT.1943-5606.0000714

CITATIONS READS
21 488

2 authors:

Dipanjan Basu Rodrigo Salgado

University of Waterloo Purdue University
123 PUBLICATIONS   648 CITATIONS    218 PUBLICATIONS   5,109 CITATIONS   

SEE PROFILE SEE PROFILE

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

Compressed Air Energy Storage System in Salt Deposits View project

Experimental Study of the Load Response of Large Diameter Closed-ended and Open-ended Pipe Piles Installed in Alluvial Soil View project

All content following this page was uploaded by Dipanjan Basu on 05 August 2015.

The user has requested enhancement of the downloaded file.

 Load and Resistance Factor Design of Drilled Shafts in Sand
D. Basu, M.ASCE1; and Rodrigo Salgado, F.ASCE2

Abstract: Resistance factors are developed for drilled shafts for a design method based on soil variables. The uncertainties associated with the
design variables and equations were systematically quantified, and Monte-Carlo simulations were performed to obtain the distributions of the
shaft and base capacities. Both the base and shaft capacities were found to follow normal distributions, and the applied dead and live loads were
assumed to follow normal and lognormal distributions, respectively. Reliability analysis was then performed to obtain the limit state and nom-
inal resistances and loads for a variety of soil profiles and pile dimensions. The optimal dead- and live-load factors were subsequently obtained
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

from the analysis. The optimal resistance factors were then adjusted for use with the load factors recommended by the Federal Highway Ad-
ministration. DOI: 10.1061/(ASCE)GT.1943-5606.0000714. © 2012 American Society of Civil Engineers.
CE Database subject headings: Load and resistance factor design; Drilled shafts; Axial loads; Design; Sand (soil type); Load factors; Limit
states; Reliability.
Author keywords: Load and resistance factor design; Drilled shaft; Axial load; Foundation design; Sand; Load factors; Resistance factors;
Limit states; Reliability analysis.

Introduction load tests to obtain target reliability indices or resistance factors

The Federal Highway Administration (FHWA) has mandated the
use of LRFD for foundations of all bridge structures. This mandate
may have started a trend in the United States of change from the
traditional working stress design (WSD) (or the allowable stress
design) to LRFD of foundations and geotechnical structures. This is
a natural trend, because LRFD is conceptually superior to WSD. In
LRFD, the uncertainties associated with the design variables and
methodologies can be systematically allocated to factors—the re-
sistance and load factors—that are associated separately with the
resistance and applied loads. In contrast, WSD relies on ad hoc use of
one factor—the factor of safety—that attempts to single-handedly
account for all the uncertainties. LRFD is appropriate for geo-
technical designs, because the variabilities and uncertainties asso-
ciated with natural systems, the ground in this case, are much greater
than those associated with well-controlled engineered systems
(Lacasse and Nadim 1996). Consequently, the development of LRFD
methods for various geotechnical problems (e.g., design of founda-
tions, slopes, and retaining structures) has been an active field of
research this past decade.
Early efforts toward establishing LRFD codes for geotechnical
designs in the United States were not sufficiently rigorous; the speci-
fications were calibrated based on a combination of simplistic re-
liability analysis, fitting to WSD and engineering judgment (Paikowsky
2004). A large number of research studies on reliability analysis or
LRFD of piles have relied on calibrations with respect to field pile
1
Assistant Professor, Dept. of Civil and Environmental Engineering,
Univ. of Waterloo, 200 University Ave. W., Waterloo, ON N2L 3G1,
Canada (corresponding author). E-mail: basu.dipan@gmail.com
2
Professor, School of Civil Engineering, Purdue Univ., 550 Stadium
Mall Dr., West Lafayette, IN 47907. E-mail: rodrigo@purdue.edu
Note. This manuscript was submitted on July 21, 2009; approved on
February 13, 2012; published online on February 15, 2012. Discussion
period open until May 1, 2013; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Geotechnical and
Geoenvironmental Engineering, Vol. 138, No. 12, December 1, 2012.
©ASCE, ISSN 1090-0241/2012/12-1455e1469/$25.00.
(Yoon and O’Neil 1996; Yoon et al. 2008). Most of these research
studies were done for driven piles (Titi et al. 2004; Kwak et al.
2007), although studies on drilled shafts (or bored piles) also exist.
Ochiai et al. (1994) proposed a reliability-based design method for
bored piles based on an in situ test-based design equation by
considering the spatial variation of standard penetration test (SPT)
results. Honjo et al. (2002) used a conditional first-order, second-
moment approach to obtain resistance factors of cast in situ concrete
piles in Japan. Misra and Roberts (2006, 2009) and Misra et al. (2007)
developed a reliability analysis for drilled shafts based on an elas-
toplastic pile shaft-soil interaction model by using Monte-Carlo
(M-C) analysis. Zhang et al. (2005) introduced a bias factor to
account for the differences in the failure criteria used for design of
bored piles. McVay et al. (2003), based on load tests of drilled
shafts in Florida limestone, assessed the reliability of drilled shafts
as a function of cost. Yang et al. (2008) calibrated, based on data
from O-cell tests, the side resistance factors for drilled shafts in
weak rock.
In this paper, the resistance factors for drilled shafts in sand for
a soil variable-based design method are developed. A systematic
probabilistic analysis is performed in which the uncertainties
(probability density functions [PDFs]) associated with each of the
soil variables and design equations are quantified, and the individual
uncertainties are then combined using M-C simulations to obtain the
PDFs of the shaft and base resistances. Subsequently, reliability
analysis is performed using the probability distributions of shaft and
base resistances and of dead loads (DLs) and live load (LLs) to
identify the most probable failure point for a target acceptable risk
(i.e., target probability of failure). This point gives the values of the
limit-state resistances and loads that are most likely to occur in
practical design. The load and resistance factors are obtained by
dividing the limit state values of loads and resistances by the corre-
sponding nominal values. The major difference between the current
study and prior studies on LRFD of drilled shafts is that, in this re-
search, the resistance factors are developed by quantifying in-
dividually the uncertainties of all the basic soil variables appearing in
the mechanistic design equations and not by calibrating the final
resistances obtained from the design equations against pile load test

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1455

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

 data. Calibrations with load tests cannot identify and discriminate
between the different sources of uncertainties associated with the
problem. The approach here was possible because the design
equations were developed from rigorous analyses of nondisplace-
ment piles (Lee and Salgado 1999; Loukidis and Salgado 2008;
Salgado and Prezzi 2007; Salgado 2008) that realistically simulate
the load-transfer mechanisms along the pile shaft and at the pile base.
Pile Resistance Equations
The ultimate total resistance Qult of a drilled shaft is the summation
of the limit shaft resistance QsL and ultimate base resistance Qb,ult
(Salgado 2008). The ultimate base and limit shaft resistances are
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.
expressed as
P somewhat wide range (Salgado 2008) and do not mechanistically
QsL ¼ qsL;i As;i ð1Þ
i Lee and Salgado (1999) performed nonlinear finite-element analysis
Qb;ult ¼ qb;ult Ab
where qsL,i 5 limit unit shaft resistance along the segment of the shaft
intersecting the ith sublayer; Hi 5 thickness of the ith sublayer; As,i
(5pBsHi) 5 corresponding shaft surface area (Bs is the pile shaft
diameter); qb,ult 5 ultimate unit base resistance; and Ab (5pBb2/4) 5
area of the pile base (Bb is the pile base diameter). The division into
soil sublayers along the pile shaft generally follows the natural
stratification, as established through the field investigation, with
individual soil layers being further subdivided into thinner sublayers
to reflect trends of penetration resistance (SPT blow counts or cone
resistance) within a layer and to increase the accuracy of the
computations.
The unit shaft resistance qsL in sand is often determined using the
b method, according to which
qsL ¼ ðK tan dÞs9v ¼ bs9v
where sv9 5 in situ vertical effective stress at the depth at which qsL is
calculated, K 5 coefficient, and d 5 friction angle mobilized along
the pile-soil interface. Several researchers have outlined different
procedures for estimating b (Reese et al. 1976; Stas and Kulhawy
1984; O’Neill and Resse 1999). However, these methods are mostly
empirical, do not always give reliable estimates of b, and are not
valid for all sand types (O’Neill and Hassan 1994; Loukidis et al.
2008). Recently, Loukidis and Salgado (2008) performed a finite-
element analysis coupled with an advanced constitutive model to
investigate the mechanics of load transfer at the interface of non-
displacement piles in sands. Based on their analysis, Loukidis and
Salgado (2008) proposed the following equation for K:
h  i
DR s9v
1:320:2ln
K0 100 pA
K ¼ pffiffiffiffiffiffiffiffiffiffiffi
ffi C1 e
e0:2 K0 20:4
where DR 5 relative density of sand expressed as a percentage, K0 5
coefficient of earth pressure at rest, and pA 5 reference stress
(5100 kPa). Eq. (4) mechanistically captures the dependence of K on
the relative density and overburden pressure (i.e., depth) and reliably
predicts K for angular and rounded sands. Loukidis and Salgado
(2008) found that the coefficient C1 is equal to 0.71 for angular sands
(based on results for Toyoura sand, which is angular) and is equal
to 0.63 for rounded sands (based on results for Ottawa sand) and
suggested a value of C1 5 0.7 to be used in calculations for clean
sands in general. Loukidis and Salgado (2008) also found that the
1456 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012
J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.
angle d was approximately equal to the triaxial-compression critical-
state friction angle fc; thus, d 5 fc can be assumed in calculations
without any significant error.
The ultimate unit base resistance qb,ult is related to the limit-
bearing capacity and is consequently affected by both friction and
dilatancy (Salgado 2008). Thus, qb,ult is a function of DR and fc
(Bolton 1986). Based on load tests, some researchers have expressed
qb,ult as ad hoc fractions of cone penetration resistance qc (Ghionna
et al. 1994) on the basis that the cone penetration test (CPT) can be
viewed as a scaled-down pile load test, and therefore qc is ap-
proximately equal to the limit unit base resistance qbL (which
corresponds to the limiting value of the unit base load at which the
soil mass surrounding the pile can no longer generate additional
resistance, leading to plunging of the pile). However, the qb,ult/qbL
(or qb,ult/qc) values proposed by these researchers vary over a
relate to the fundamental soil variables associated with the problem.
and used the results of plate load tests in calibration chambers to
ð2Þ find that qb,ult/qbL depends primarily on DR. Based on this analysis,
Salgado (2008) proposed an analytical expression for qb,ult for pile
head settlement equal to 10% of the pile diameter (i.e., for 10%
relative settlement)
qb;ult ¼ qb;10% ¼ 0:23 e20:0066DR qbL ð5Þ
where the limit unit base capacity qbL (5qc) can be expressed,
based on rigorous cavity expansion analysis (Salgado and Randolph
2001; Salgado and Prezzi 2007), in terms of the fundamental soil
variables as
 0:84120:0047DR
qbL s9h
¼ 1:64 e0:1041fc 1ð0:026420:0002fc ÞDR
pA pA
ð6Þ
ð3Þ
where sh9 (5K0sv9) 5 in situ effective horizontal stress. Thus, these
equations mechanistically relate the ultimate pile resistance to the
fundamental soil variables and are based on rigorous analyses.
For the calculations of qbL and qb,10%, appropriate fc, DR, and s9h
values must be used that are representatives of the zone below the
pile base within which the resistance to downward movement of the
pile develops. At the limit condition (i.e., when the pile is about to
plunge), a plastic zone forms immediately below the pile base. On
the other hand, the zone of influence for the condition corresponding
to 10% relative settlement is somewhat wider and deeper for sand,
because there has not been localization of strains as a result of
plasticity. To the author’s knowledge, there has been no study that
conclusively outlines the exact extent of the zone of influence below
the pile base under working and limit conditions. The authors
considered it appropriate to use the values of fc, DR, and s9h cor-
responding to a depth of Bb/2 below the pile base (Bb is the pile base
ð4Þ
diameter) for calculating qbL and qb,10%. For the cases studied, in
which the density was either constant or increased from the pile
base down, this was considered conservative, even in the event that
a large influence depth would be in effect.
Eqs. (1)e(6) have been used to calculate the capacity of drilled
shafts. The choice of Eqs. (4)e(6) for calculating shaft and base
capacities was based on the fact that these equations have been
obtained from rigorous analyses that mechanistically relate the pile
capacities to the fundamental soil variables; additionally, these
equations have been compared against specific high-quality ex-
perimental data. Such a fundamental approach automatically allows
a systematic consideration of the uncertainties in the soil variables
 and in the design equations that can be combined to calculate the The uncertainties considered are the uncertainties in (1) soil
uncertainties associated with the pile capacity. variables, (2) design equations (model uncertainties), (3) applied
loads, and (4) pile dimensions. For each variable, the PDF was as-
LRFD Framework sumed as either normal or lognormal distribution, and its mean
(expected value) and COV were estimated.
In the LRFD framework, the capacity (total resistance) and demand Biases often arise in the estimation of the design variables (ma-
(applied loads) are related as terial properties, loads, and resistances), making the nominal values
P calculated by the designers different from the corresponding mean
ðnÞ
ðRFÞ RðnÞ $ ðLFÞi Li ð7Þ values. In such cases, the mean and nominal values are related
i through bias factors. For a design variable Y, if y is the mean and yn
where (RF) 5 resistance factor, RðnÞ 5 nominal (or characteristic) is the nominal value, then the bias factor of Y is defined as the
resistance, (LF)i 5 load factor corresponding to the ith nominal (or ratio y=yn . If biases are known, it is appropriate in design to account
ðnÞ for them.
characteristic) load Li , and the superscript (n) represents nominal
loads and resistances. The deterministic loads and resistances es-
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

timated by design engineers based on procedures prescribed by Uncertainties in Soil Variables

codes (or other specific procedures), manuals and books, or on
experience are called the nominal loads and resistances, respec- The soil variables required for pile capacity calculations are fc,
tively. Thus, QsL and Qb;ult calculated deterministically using Eqs. DR, K0, and soil unit weight g (required to calculate the in situ
ðnÞ ðnÞ
(1)e(6) are the nominal shaft and base resistances QsL and Qb;ult . stresses). K0 is difficult to estimate in the field, and not much
For piles, Eq. (7) can be written in two different ways information (e.g., PDF or COV) is available regarding its vari-
h i ability. However, if a deposit is known to be normally consolidated,
ðnÞ ðnÞ ðnÞ P ðnÞ
ðRFÞ QsL þ Qb;ult ¼ ðRFÞ Qult $ ðLFÞi Li ð8Þ for example, then K0 falls within a relatively narrow range of
roughly 0.4e0.5, depending on the relative density of the deposit.
Accordingly, K0 was assumed to be deterministic with its value
ðnÞ ðnÞ P ðnÞ
ðRFÞs QsL þ ðRFÞb Qb;ult $ ðLFÞi Li ð9Þ in the 0.4e0.5 range and the variation of K0 was indirectly ac-
counted for by performing analyses with three fixed (deterministic)
The shaft and base resistances are calculated separately, and the values: K0 5 0.4, 0.45, and 0.5. The remaining variables were
mechanisms by which these resistances develop are quite different. The treated probabilistically.
loading that develops along the pile shaft closely approximates simple Based on experimental observations, Bolton (1986) reported
shear loading and, at the ultimate limit state, the stresses correspond to a 61° band encompassing all the measurements of critical-state
the critical-state values of shear stress. The loading around the base is friction angle fc. Thus, the maximum error in the estimation
much more complex, with the mean stress increasing at almost every of fc at a particular site is expected to be 61°. Assuming
point around the base, and shear stresses developing by different that fc follows normal distribution (Foye 2005), the spread of
degrees and at different rates depending on the point. Consequently, 2° results in an SD of 0.33° for fc. Because fc of different
the two capacity equations are subjected to different sets of uncer- types of sand lie typically within the 28e36° range (Salgado
tainties. This is reflected in Eq. (9), which is considered to be more 2008), the maximum and minimum values of the COV of fc
appropriate for this analysis. Eq. (9) is used as the LRFD equation, that can be expected at a particular site are 0.33°/28° 5 0.012 and
which, for the applied dead and live loads, can be rewritten as 0.33°/36° 5 0.009. Kim (2008) estimated the COV of fc to be
ðnÞ ðnÞ in the 0.0081e0.0172 range based on the experimental studies
ðRFÞs QsL þ ðRFÞb Qb;ult $ ðLFÞDL ðDLÞðnÞ þ ðLFÞLL ðLLÞðnÞ
by Verdugo and Ishihara (1996) and Negussey et al. (1988). In
ð10Þ this study, the COV of fc was conservatively assumed to be
equal to 0.02, which is slightly greater than the maximum value
where DL = dead load, LL = live load, (RF)s 5 shaft-resistance reported in the literature. fc was also assumed to follow normal
factor, (RF)b 5 base-resistance factor, (LF)DL 5 dead-load factor, distribution.
and (LF)LL 5 live-load factor. The variability of soil unit weight has been studied by many
researchers. Baecher and Christian (2003) reported, based on studies
Uncertainty Assessment of Design Variables by Lee et al. (1983), Lacasse and Nadim (1996), and Lumb (1974),
that the COV of unit weight does not exceed 0.1. Kim (2008) and
The uncertainties associated with the design variables are quantified Kim and Salgado (2012) corroborates the finding based on studies
by treating each of these variables as a random variable with an as- by Phoon and Kulhawy (1999), White et al. (2005), and Hammitt
sociated probability-density function (PDF). The PDFs used are ei- (1966). In this study, it was assumed that g follows normal distri-
ther normal or lognormal distributions. One measure of the variable bution with a COV 5 0.1.
uncertainty is the sample coefficient of variation (COV), defined as The estimation of relative density depends on the correct esti-
SX mation of the in situ void ratio e, and the maximum and minimum
COV ¼ ð11Þ void ratios emax and emin. Baecher and Christian (2003) reported,
x
based on a study by Phoon and Kulhawy (1999), that the COV of DR
where x 5 sample mean of a design variable X, and SX 5 sample SD varies between 0.1 and 0.4. However, such a wide variation seems
of X, defined as unlikely for a particular site and may have resulted from clubbing of
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi data applicable for a wide range of DR (30e70%). It is likely that the
u N
uP variability will be greater for lower values of DR and will decrease as
u ðxi 2 xÞ2
t DR increases. A more reliable and practical method of calculating DR
SX ¼ i ¼ 1 ð12Þ is from the CPT results, as proposed by Salgado (2008), which is
N 21
a result of rigorous cavity expansion analysis (Salgado and Prezzi
where xi 5 individual realizations of X, and N 5 sample size. 2007)

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1457

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

   
qc s9h EðYÞ ¼ M bias EðMÞEðf Þ ¼ M bias Eðf Þ ð16Þ
ln 2 0:4947 2 0:1041fc 2 0:841ln
pA p
DR (%) ¼   A Further, from Eq. (15) we get (sY/Mbias)2 5 {[E(M)]2 1 sM2}{[E(f)]2 1
s9h
0:0264 2 0:0002fc 2 0:0047 ln sf2} 2 [E(M)]2[E(f )]2 which, on simplification, produces
pA
  s2Y n o
¼ fDR qc ; fc ; s9h ð13Þ 2
¼ ½Eðf Þ2 þ s2f s2M þ s2f ð17Þ
ðM bias Þ

Instead of treating DR as a fundamental variable, cone resistance qc
was assumed for a soil profile as the starting point (similar to the
starting point in design if CPT results were known for a site) and
the PDF of DR from the PDF of qc and from the PDF of the re-
lationship between qc and DR (i.e., the qc→DR model uncertainty)
given by Eq. (13) was calculated. Although it is difficult to
separate the uncertainty in the measurement of qc (related to
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.
In these equations, E(Y) and sY are the expectation and SD of Y,
respectively, and E(f ) and sf are the expectation and SD of the
estimations of Y using y 5 f (x), respectively, without taking model
uncertainty into account. M and f are statistically independent.
Eq. (16) confirms that Mbias is the bias factor in the relationship y 5
f(x), relating the nominal value E(f) of Y to its mean value E(Y).
Eq. (17) shows that sM is related to sY and sf; and thus can be

hardware variability) from the uncertainty in the variables that qc determined if information about sY and sf are available.
depends on, certain attempts have been made to do that. According If during the estimation of sY, the uncertainty because of the input
to FHWA (2001), the COV of qc is 0.07. Foye (2005) found that variable X can be eliminated so that sf 5 0 (i.e., if X is assumed to be
the COV of qc is 0.08 and that qc follows normal distribution. deterministic), then sM can be expressed as
In this research, it is assumed that qc follows normal distribution s
with a COV 5 0.08. The model uncertainty associated with Eq. (13) sM ¼ biasY ð18Þ
M Eðf Þ
was estimated following a procedure in the next section.
Because only the information about a sample space is generally
known, the sample SD SY and SM [as defined in Eq. (12)] are usually
Model Uncertainties used in calculations instead of sY and sM. Thus, Eq. (18) can be
rewritten in terms of the sample SD as
Model uncertainties arise because the mathematical models (equa-
tions) are not perfect representation of a process but approximations SY
SM ¼ ð19Þ
of the actual behavior based on a set of assumptions. Model M bias Eðf Þ
uncertainties (also known as transformation uncertainties) exist in
the estimation of the unit base and shaft resistances and in the esti- To determine the model uncertainties, yi should be obtained from
mation of DR from qc. well-controlled test data, from which the error yi 2 f (xi) can be
If a function y 5 f (x) represents a mathematical relationship accurately calculated.
between two variables X and Y, then the model uncertainty as- The model uncertainty in the qc→DR relationship [in which Y 5
sociated with the function f appears as a difference between DR, X 5 qc, and f 5 fDR is given by Eq. (13)] was investigated by Foye
realizations yi and f (xi). The error wi 5 yi 2 f (xi) often has (2005) using results of 25 well-controlled calibration chamber tests
a random and a deterministic part. The random error is caused by (Salgado 1993). Based on the study by Foye (2005), the mean of the
random fluctuations of xi and yi. The deterministic error generally normalized error of DR [W* 5 (DR,measured 2 DR,calculated)/DR,calculated]
arises because of imperfections in f and results in a bias in the was estimated to be about 20.03 [i.e., E(W*) 5 20.03], which
estimations of y. This bias depends on f, and can be eliminated by implies that DR is overpredicted by 3% using Eq. (13). Thus, there is
using an appropriate bias factor. a bias present in Eq. (13) because of which the values of DR to be
A normalized error wpi can be defined to facilitate the calculations used in the M-C calculations should be obtained by multiplying the
as (Foye 2005; Foye et al. 2006a) nominal DR values calculated from Eq. (13) by 0.97 [i.e., the bias
factor 5 MDR bias
5 1 1 E(W*) 5 0.97]. Foye (2005) calculated the SD
yi 2 f ðxi Þ SDR (5 SY) of DR to be equal to 10%, (i.e., 0.1 if relative density is
wpi ¼ ð14Þ expressed as a fraction) with qc (5 X) as a deterministic variable
f ðxi Þ
(i.e., sf 5 Sf 5 0 was used in the calculations). Using Eq. (19), which
relates SY and SM when Sf 5 0, SM is equal to SDR =½MDR bias
EðfDR Þ 5
Eq. (14) can be used to redefine the relationship between Y and X as SDR =½0:97EðfDR Þ. Foye et al. (2006a) also observed that the normal-
  ized error follows normal distribution.
yi ¼ 1 þ wpi f ðxi Þ ¼ M bias mi f ðxi Þ ð15Þ The incorporation of the qc→DR model error in the M-C simu-
bias
lations was done by introducing a bias factor MDR and a new random
where M bias 5 deterministic bias in the model f, mi 5 realization of variable MDR representing the qc→DR model uncertainty and by
the random variable M representing the random part of the model rewriting Eq. (13) as
uncertainty, and wpi 5 realization of the random variable W* rep-  
resenting the normalized error. The bias factor Mbias is defined as DR (%) ¼ MDR bias
MDR fDR qc ; fc ; s9h
Mbias 5 E(1 1 W*) 5 1 1 E(W*). If the model has no bias, W* will    
qc s9
have an expected value of zero, which makes M bias 5 1; however, ln 2 0:4947 2 0:1041fc 2 0:841ln h
pA p
that is rarely the case with real engineering models. The realizations ¼ 0:97MDR  A
mi and wpi are related by mi 5 (1 1 wpi )/Mbias; thus, the random 0:0264 2 0:0002fc 2 0:0047 ln
s9h
variables M and W* are statistically related through their expected pA
(mean) values and SDs as E(M) 5 E(1 1 W*)/Mbias 5 [1 1 E(W*)]/ ð20Þ
Mbias 5 1, and sM 5 sW*/Mbias.
Assuming M and f are statistically independent and uncorrelated, In the M-C simulations, MDR is treated as a random variable that
the expectation of Y can be obtained from Eq. (15) follows normal distribution with a mean E(MDR) 5 1.0 and an SD

1458 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

 SMDR 5 SDR/[0.97E(fDR)] 5 10%/[0.97E(fDR)] {or SMDR 5 0.1/
[0.97E(fDR)] if relative density is expressed as a fraction}.
The model uncertainty associated with qsL arises in the estimation
of b through the use of Eq. (4). To estimate the uncertainty in Eq. (4),
results of eight centrifuge tests by Fioravante (2002) and Colombi
(2005) were used. The centrifuge tests were performed with a sand
with fc 5 32.3° for two relative densities (DR 5 66 and 90%)
and with K0 5 0.45. To compare the results from Eq. (4), which is
valid for field conditions, with the centrifuge test results, the scale
effects associated with the ratio of the pile diameter to the thickness
of the shear band that forms around the shaft should be incorporated.
Therefore, the finite-element analysis of Loukidis and Salgado
(2008) that was used to obtain Eq. (4) for field conditions was redone
to simulate the conditions prevailing in the centrifuge tests, and an
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.
equation of K for the centrifuge-test conditions was developed
following exactly the same procedure that was done for the case of
Eq. (4)
h  i
DR s9v
100
1:9520:48 ln pA
K ¼ 0:88K0 e
Eq. (21) was used to estimate b for the centrifuge tests, and the
normalized error [W* 5 (bmeasured 2 bcalculated)/bcalculated] between
the measured and estimated b values was calculated. For the tests
corresponding to DR 5 66%, the mean of the normalized error, E
(W*), was obtained as 20.08, while, for the tests corresponding to
DR 5 90%, the mean was 0.01. Thus, the model overpredicted b by
8% for DR 5 66% and underpredicted b by 1% for DR 5 90%.
Because the bias was opposite for the two relative densities and
because the data available from the centrifuge tests were limited, the
authors decided that no model bias for b would be considered in the
analysis (i.e., bias factor is Mbbias 5 1). The SD of the normalized
error SW* for DR 5 66% and DR 5 90% was found to be 0.215 and
0.216, respectively. Therefore, to incorporate the qsL model un-
certainty in the analysis, a new variable Mb that follows normal
distribution with expected value E(Mb) 5 1.0 and SD SMb 5 SW*/
Mbbias 5 0.2 was introduced. The equation used to calculate qsL in the
M-C simulations is
h  i
DR s9v
0:7K0 100
1:320:2ln pA
qsL ¼ Mb pffiffiffiffiffiffiffiffiffiffiffiffi e s9v tanfc
e0:2 K0 20:4
The model uncertainty associated with unit base resistance [Eqs.
(5) and (6)] was estimated from 21 well-controlled deep plate load
tests performed within a calibration chamber (Lee and Salgado
1999). The same procedure previously used for calculating the
normalized error was followed. However, before the estimation of
the model uncertainty, the authors investigated whether the tests
performed in the calibration chamber had any chamber boundary
effect. Lee and Salgado (1999) simulated, using their finite-element
model, the pile base response for real field conditions and for cal-
ibration chamber test conditions. They assumed identical values of
sand relative densities and stress states at the pile base for the field
and the calibration chamber. They considered two extreme boundary
conditions for the lateral walls of the calibration chamber: constant
horizontal stress and zero horizontal displacement. The simulated
field values of qb,10% were greater than the corresponding simulated
calibration chamber values by about 9% for the constant horizontal
stress boundary condition of the calibration chamber, while an
opposite result, with field values less than calibration chamber
values by about 4%, was observed for the zero horizontal dis-
placement boundary condition of the calibration chamber. The
JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1459
J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.
normalized error between the qb,10% values simulated for the field
(qb,10%,field) and the calibration chamber tests (qb,10%,plate) was
calculated, and it was found that the mean and SD of the normalized
error [(qb,10%,field 2 qb,10%,plate)/qb,10%,field], considering both the
constant horizontal stress and zero horizontal displacement con-
ditions, were 0.025 and 0.09. In the field, neither constant stress nor
zero displacement conditions exist; rather, a condition between these
two extremes prevails. Thus, the small overall bias of 2.5% will very
likely not exist. Consequently, the authors considered it appropriate
to neglect the boundary effects in the analysis.
The plate load tests in the calibration chamber were done for two
relative densities: DR 5 50% and DR 5 90%. For DR 5 50%, the mean
of the normalized error between measured and predicted unit base
resistance [W* 5 (qb,10%,measured 2 qb,10%,calculated)/qb,10%,calculated]
was 20.03 (i.e., E(W*) 5 20.03) while, for DR 5 90%, the mean
E(W*) was 0.16. Thus, the model overpredicts qb,10% by 3% for
DR 5 50% and underpredicts qb,10% by 16% for DR 5 90%. Con-
sequently, the calculated nominal qb,10% values from Eqs. (5) and (6)
must be multiplied by 0.97 for DR 5 50% [i.e., bias factor 5 Mqb bias
5
1 1 E(W*) 5 0.97] and by 1.16 for DR 5 90% [i.e., bias factor 5
ð21Þ Mqbbias
5 1 1 E(W*) 5 1.16] before they are used in the M-C
simulations. Because no reliable data were available for other val-
ues of DR, a linear interpolation of the bias factor for values of DR
intermediate between 50 and 90% was assumed. Also, it was as-
sumed that the bias factor is equal to 0.97 for DR # 50% and is equal to
1.16 for DR $ 90%. The SD of the normalized error SW* was found to
be equal to 0.15 and 0.08 for DR 5 50 and 90%, respectively. A large
SD of 0.15 occurred for DR 5 50% because, in the calibration
chamber test, relative densities at these levels are not as reproducible
or uniform as those for denser samples. Additionally, in the specific
case of these tests, samples with DR between 49.1 and 60.2% were
grouped for the comparisons. For real field conditions, the SD was
expected to be less. This was corroborated by the SD value 0.08 for
the tests corresponding to DR 5 90%, which were known to be more
uniform and more reproducible (witnessed by test samples having
relative densities that fall in the much narrower range of 90.6e92.8%).
Therefore, an SD of 0.1 was assumed in the calculations.
To incorporate the model uncertainty of qb,10% in the calcu-
bias
lations, a bias factor Mqb was introduced in the equation of qb,10%
bias
with Mqb equal to 0.97 for DR # 50%, equal to 1.16 for DR $ 90%,
and equal to a linearly interpolated value between 0.97 and 1.16 for
50% , DR , 90%. A new random variable Mqb was also introduced
ð22Þ
with E(Mqb) 5 1 and SMqb 5 0.1. Mqb was assumed to follow normal
distribution. The modified equation of qb,10% used in the M-C
simulation is given by
qb;10% ¼ Mqb
bias
Mqb 0:38pA e20:0066DR e0:1041fc1ð0:026420:0002fc ÞDR
 0:84120:0047DR
s9h
 ð23Þ
pA
Uncertainties in Applied Loads
According to Ellingwood and Tekie (1999), the DL can be de-
scribed with normal distribution with a bias factor of 1.05 and
a COV equal to 0.1. According to Nowak (1994) and FHWA
(2001), the bias factor and COV for the DL are in the range
1.03e1.05 and 0.08e0.10, respectively, depending on the type
of structural components. In this analysis, it was assumed that the
DL follows normal distribution with a bias factor of 1.05 and a
COV of 0.1.
The LL is generally described using lognormal distribution (Foye
et al. 2006a). According to FHWA (2001), the LL has a bias factor
 of 1.1e1.2 and a COV of 0.18. According to Ellingwood and Tekie
(1999), however, LL is represented by a Type 1 distribution of the
largest values with a bias factor of 1.0 and a COV of 0.25. The PDFs
of the Type 1 extreme value distribution and the lognormal distri-
bution were plotted for identical values of the bias factor and COV,
and it was found that the PDFs of both distributions are nearly
identical for the range of the LL values encountered in practice for
piles. This finding was indirectly corroborated by Ghosn and Moses
(1998), who separately calculated reliability indices for highway
bridge structures with LLs following lognormal distribution and
a Type 1 extreme value distribution, and found that the results of
both calculations were sufficiently close. Consequently, the widely
used lognormal distribution was used to describe LLs. For most of
the problems analyzed, the authors conservatively chose a COV 5
0.25 and the corresponding bias factor 5 1.0, as recommended
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

by Ellingwood and Tekie (1999). For a few selected problems,

however, analyses were performed for LL COV 5 0.18 and bias
factor 5 1.15, as recommended by FHWA (2001), and the results
were compared.

Uncertainties in Pile Dimensions

Drilled shafts are constructed by removing soil from the ground by
drilling and filling the resulting cylindrical void with concrete and
reinforcement. The construction process is controlled; therefore,
the as-built diameter of the drilled shaft varies little. To the author’s
knowledge, there is no systematic study available regarding the
variability of the drilled shaft diameter. Based on experience and
typical construction tolerances, it was assumed that the drilled shaft
diameter Bp (5 Bs 5 Bb) follows normal distribution with a COV 5
0.02.
Typically, drilling is done to refusal or to a minimal embedment
into the bearing layer, noticeable by a substantial change in drilling
resistance. Thus, the lengths of drilled shafts do not vary much,
and any variation that may occur usually results in lengths slightly
greater than the design lengths. Therefore, it was assumed that
the pile length Lp is deterministic (equal to the design length),
because any additional length (because of variability in the depth to
the bearing layer, for example) will work toward a conservative
design.
Practical Cases Considered
The primary goal of this paper is to present resistance factors that
can be used in the design of drilled shafts in normally consolidated
sand deposits. Therefore, a range of soil profiles, pile dimensions,
and (LL)/(DL) ratios were considered that may occur in real field
conditions.
The soil profiles (Fig. 1) considered consist of: (1) a homoge-
neous, completely dry deposit of sandy soil with a mean relative
density DR,mean 5 70%; (2) the same homogeneous sand deposit
described in (1) (i.e., with DR,mean 5 70%) with a water table located
at the ground surface; (3) a completely dry sand deposit with a loose
layer (DR,mean 5 50%) overlying a strong bearing layer (DR,mean 5
80%) that extends to great depth; (4) a two-layer system, as in (3)—
with DR,mean 5 50% for the top layer and DR,mean 5 80% for the
underlying layer—with a water table located at a depth of 2 m below
the ground surface (the soil above the water table was assumed to be
completely dry); (5) a two-layer system with the top layer consisting
of extremely loose sand having DR,mean 5 20% and the bearing layer
consisting of dense sand having DR,mean 5 80% and with a water
table located at the ground surface; and (6) a four-layer, completely
dry deposit consisting of a loose top layer with DR,mean 5 30%
Fig. 1. Details of soil profiles: (a) Profiles 1 and 2; (b) Profiles 3e5;
(c) Profile 6
spanning 0e5 m down from the ground surface, a second layer with
DR,mean 5 45% spanning 5e10 m below the ground surface, a third
layer with DR,mean 5 60% spanning 10e15 m below the ground
surface, and a bearing layer with DR,mean 5 75% that lies below the
third layer and extends down to great depth. The thicknesses of the
soil layers in Profiles (3)e(6) were assumed to be deterministic
variables. In deciding the thicknesses of the top layers of the two-
layer profiles [(3)e(5)], it was assumed that the depth of pile em-
bedment Hbearing in the bearing layer is two times the mean pile
diameter Bp,mean (Fig. 1).
For these deposits, three different sand types were considered,
with mean critical-state friction angle fc,mean 5 30, 33, and 36°,
and, for each type, three different values of K0—0.4, 0.45 and
0.5—were assumed. The mean values of sand unit weight
g mean ½5ðGs 1 Semean Þ=ð1 1 emean Þ (Gs is the specific gravity of
sand solid particles, asssumed to be equal to 2.65, S is the degree
of saturation) was calculated using the mean values of void
ratio emean. The mean void ratio emean was obtained from the
mean relative density DR,mean using the equation emean 5
emax 2 ðDR;mean =100Þðemax 2 emin Þ, where emax (50.9) and emin
(50.45) are the maximum and minimum void ratios, respectively.
In the M-C simulations, mean qc profiles qc,mean(z) (i.e., qc,mean
as a function of depth z) were given as inputs. These qc,mean(z)
curves were backcalculated initially (i.e., before the M-C sim-
ulations were performed) from preassumed values of DR,mean,
g mean, fc,mean, and K0, using the inverse of Eq. (20). As a conse-
quence, when given as an input to the M-C simulation code, they
produced the desired mean relative density for each layer.
The slenderness ratio (Lp/Bp) and the diameter and length of
drilled shafts considered in this study ranged from values that would

1460 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

 be considered very low to values considered high for real field
conditions. The responses of four drilled shafts were studied with (A)
mean diameter Bp,mean 5 0.3 m and length Lp 5 10 m, (B) Bp,mean 5
1.5 m and Lp 5 10 m, (C) Bp,mean 5 0.3 m and Lp 5 30 m, and (D) Bp,
mean 5 1.5 m and Lp 5 30 m for Profiles 1e5. For Profile 6, a fifth
drilled shaft (E) was considered with Bp,mean 5 1.0 m and Lp 5 20 m.
The proportion of LLs and DLs acting on a bridge structure
depends on the span length (Hansell and Viest 1971). Titi et al.
(2004) tabulated the (LL)/(DL) ratios recommended by AASHTO
(2007) and FHWA (2001) for design of bridge structures; the rec-
ommended values vary over a wide range of 0.2821.92. Accord-
ingly, (LL)/(DL) 5 0.25, 1.0, and 2.0 was considered in this study.

Distributions of Shaft and Base Resistances

Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

M-C simulations were performed to obtain the probability dis-

tributions of QsL and Qb;ult , with the assumption that all the random
variables describing the soil properties and model uncertainties are
uncorrelated. The analysis started with the selection of a soil profile
with known (or assumed) mean values of fc, g, and K0 (K0 is
deterministic, and therefore the mean value was the constant de-
terministic value) and with an assumed mean trend of CPT profile
qc,mean(z). A drilled shaft with an assumed length Lp and an as-
sumed mean diameter Bp,mean embedded in the soil profile was
considered. With the mean soil profile and pile dimensions es-
tablished, the M-C simulations were started. The program started
with shaft capacity calculations at depth z 5 0 m and moved down
along the pile shaft. First, a random value of qc was generated for
a particular depth along with random values of fc, g, and MDR to
calculate a random value of DR using Eq. (20). Subsequently,
random numbers of the remaining variables, Mb and Bp, were
generated to calculate a random value of shaft resistance for that
depth. This shaft capacity calculation was repeated for different
depths along the pile shaft (typically, soil sublayers of 1- or 2-m
thickness were assumed) with new random values of qc, DR, and the Fig. 2. Histograms and best fit normal plots of (a) limit shaft capacity;
other variables. The calculated shaft capacities at the different (b) ultimate base capacity of a 30-m-long drilled shaft with a diameter of
depths were summed over the entire pile length to obtain the 1.5 m in a homogeneous profile consisting of sand with mean DR 5 70%,
random value of the total shaft resistance QsL. Then, as the pile mean fc 5 33°, and K0 5 0.45 and the with water table at the ground
base was reached, a random value of the base resistance Qb,ult surface
was calculated using the random values of soil variables at
z 5 Lp 1 Bp,mean/2, a random value of Bp, a random value of Mqb,
bias
and the appropriate value of the bias factor Mqb . Eqs. (1), (2), (22),
and (23) were used to obtain these random values of QsL and Qb;ult . Reliability Analysis and LRFD
The set of calculations (corresponding to one M-C run) was re-
peated 5,000 times to obtain the probability distributions (histo-
grams) of QsL and Qb;ult . The random values, mean, and SD of QsL Reliability Index and Most Probable Failure Point
and Qb;ult were recorded for Drilled Shafts AeE in Profiles 1e6.
Figs. 2(a and b) show the histograms of QsL and Qb;ult for At the limit state, the total capacity (or resistance) equals the total
Drilled Shaft D in Soil Profile 2 with fc 5 33° and K0 5 0.45. demand (or load). Thus, the limit-state (performance) function is
The shaft and base capacities follow normal distributions—the given by
trend was typical of all the shaft and base capacities calculated. ðLSÞ ðLSÞ
Thus, QsL and Qb;ult were assumed to be normally distributed QsL þ Qb;ult ¼ ðDLÞðLSÞ þ ðLLÞðLSÞ ð24Þ
random variables. Also, the COV of Qb;ult lies within a range of
0.129e0.135 for all the drilled shafts, while the COV of QsL lies in where the superscript (LS) denotes the limit state. The equation was
the range of 0.094e0.115 and 0.075e0.086 for the 10- and 30-m- rewritten for the convenience of mathematical operations as
long drilled shafts, respectively. The COV of QsL depends on the
variability of layers crossed by the pile and their thicknesses: the gðXÞ ¼ X1 þ X2 2 X3 2 X4 ¼ 0 ð25Þ
variability in QsL decreases as the pile length increases. Biases
were observed between the mean and deterministic (nominal) where X 5 {X1, X2, X3, X4} is the vector of the random variables
capacities [the deterministic values of QsL and Qb;ult were cal- representing QsL (5X1), Qb;ult (5X2), DL (5X3), and LL (5X4),
culated using Eqs. (1)e(6)]. The bias factor for QsL lies in the respectively. X1, X2, and X3 follow normal distribution, while X4
narrow range of 0.976e0.990, while the bias factor for Qb;ult lies follows lognormal distribution. The probability of failure pf is
in the range of 1.065e1.109. given by

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1461

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

 ð
pf ¼ fx ðxÞdx ð26Þ bHL s21 b s2 b s2
gðYp Þ ¼ 2 þ m1 2 HL 2 þ m2 2 HL 3
L L L
gðXÞ , 0  

bHL 224 exp 24 Y4p þ l4

2 m3 2 exp þ l4 ¼ 0
L
where fX(x) 5 joint probability distribution of X1, X2, X3, and X4, and
x 5 domain of X. ð30Þ
Eq. (25) is expressed in terms of the standard normal variables
by performing the Rosenblatt transformation (Ang and Tang 1984) where
on X as
1=2
ðY4 24 1l4 Þ
gðYÞ ¼ Y1 s1 þ m1 þ Y2 s2 þ m2 2 Y3 s3 2 m3 2 e
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.
where Yi 5 (Xi 2 mi)/si for i 5 1, 2, and 3, mi and si are the mean
and SDs of Xi (for ffii 5 1, 2, 3, and 4), Y4 5 ðln X4 2 l4 Þ=24 ,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
24 5 lnð1 1 s24 =m24 Þ, and l4 5 ln m4 2 224 =2. In the standard
normal space, pf is given by
ð The most probable failure point Y* (i.e., the required limit state
pf ¼ fY ðyÞdy
gðYÞ , 0
where f(×) 5 PDF of the standard normal variate, and y 5 domain
of Y. Thus, the magnitude of pf depends on the relative position of
the limit-state hyper surface g(Y) 5 0 with respect to the origin of
the four-dimensional standard normal hyperspace of Y 5 {Y1, Y2,
Y3, Y4}. The position of the limit-state surface is often described
by the minimum distance bHL from the limit state surface to the
origin—bHL is commonly known as the Hasofer-Lind reliability
index. The greater the value of bHL, the further the limit state
surface is from the origin and the smaller the probability of
failure. This minimum distance bHL was calculated from the
origin to a particular point on g(Y) 5 0, known as the most
probable failure point Y*—this point corresponds to the maxi-
mum value of the integrand fY(y). It is important to locate Y*
because the value of bHL depends on it and because the values
of Xi corresponding to the most probable failure point gives
the most probable values of Xi at the limit state. In other words,
ðLSÞ ðLSÞ
QsL , Qb;ult , ðDLÞðLSÞ , and ðLLÞðLSÞ in Eq. (24) correspond to the
most probable failure point.
For uncorrelated and independent standard normal variables Yi
with a linear limit-state function, bHL gives an accurate estimate of
the probability of failure as
pf ¼ 1 2 FðbHL Þ
where F(×) 5 cumulative density function of the standard normal
variate. For example, bHL 5 3.719, 3.090, 2.326, and 1.282
for pf 5 1024, 1023, 1022, and 1021, respectively, for standard
normal variables with linear limit state functions (note that pf
decreases as bHL increases). For the nonlinear limit state func-
tion given by Eq. (27), the estimates of pf from Eq. (29) are
approximate.
The minimum distance bHL can be obtained by using Lagrange’s
multiplier technique in which the function Lg 5 ðYT YÞ1=2 1
lgðYÞ—where the distance (YTY)1/2 from the origin is the ob-
jective function, g(Y) is the constraint, and l is Lagrange’s
multiplier—is minimized (Ang and Tang 1984). Thus, setting
∂Lg =∂Yi 5 Yi =ðYT YÞ1=2 1l∂gðYÞ=∂Yi 5 0 and ∂Lg =∂l 5 gðYÞ 5 0,
g(Y) at the most probable failure point can be expressed in terms
of bHL as
1462 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012
J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.
   
L ¼ s21 þ s22 þ s23 þ 224 exp 2 Y4p 24 þ l4 ð31Þ
¼ 0
and Y4* is the value of Y4 at the most probable failure point. The
ð27Þ
reliability index bHL is given by
1  
bHL ¼ 2 s1 Y1p þ s2 Y2p 2 s3 Y3p 2 24 Y4p exp 24 Y4p þ l4
L
ð32Þ
point) in the reduced variable space is related to bHL as
ð28Þ
8
>
> bHL si
>
> 2 i ¼ 1; 2
>
>
>
<
L
p bHL s3
Yi ¼ i ¼ 3 ð33Þ
>
> L 
>
>
>
> bHL 24 exp 24 Y4p þ l4
>
: i ¼ 4
L
Target Probability of Failure and Calculation of Mean
Dead and Live Loads
For most structural designs, a target pf of approximately 1023 is often
assumed (Foye 2005; Ellingwood et al. 1980). However, depending
on the importance of a structure and the relative cost of safety
measures, different target pf may be set (Joint Committee on
Structural Safety 2000). Moreover, particular components of a
structure, for example steel connections, may require a target pf
greater than 1023 (Fisher et al. 1978). A target probability of failure
of 1023 for drilled shafts implies that one in every 1,000 piles may
fail. Under typical conditions, this incidence of failure may be too
high. A target pf 5 1024, corresponding to a failure rate of one in
every 10,000 drilled shafts, is perhaps more realistic. Accordingly,
two values of target pf , 1023 and 1024 (i.e., target bHL of 3.090 and
ð29Þ
3.719) were considered and resistance factors for both these target
probabilities of failure were developed.
A target value of bHL (typically, at 3.090 or 3.719) was set, and
Eq. (30) was used to calculate the mean values of DL and LL (i.e., m3
and m4) from the known mean (m1 and m2) and SD (s1 and s2) values
of QsL and Qb,ult (obtained by M-C simulations previously de-
scribed) for Profiles 1e6 and Drilled Shafts AeE. The calculations
were performed for three values of (LL)/(DL), 0.25, 1.0, and 2.0,
after expressing s3 and s4 as products of the corresponding COV
and mean. Thus, in the actual calculations, the following rela-
tionships were used: m4 5 rm3 (r 5 0.25, 1.0, and 2.0), s3 5
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
m3(COV)DL, s4 5 rm3(COV)LL, 24 5 lnf1 1 ½rm3 ðCOVÞLL 2 =ðrm3 Þ2 g,
and l4 5 lnðrm3 Þ 2 224 =2. Consequently, the only two unknowns
in Eq. (30) are m3 and Y 4*. Initial guesses of m3 and Y *4 were given
as inputs to the calculations and iterations performed on m3 and
Y 4* until Eq. (30) was satisfied, such that the values of Y 4* be-
tween two consecutive iterations were within 0.01%. The new
 value of Y 4* after each iteration was calculated using Eq. (33). After safety (FS)mean can be defined as the ratio of the mean total resistance
Eq. (30) was satisfied, the mean values of DL (m3) and LL (5 m4 5 and mean total load. Thus, the mean factor of safety is given by
rm3) and the corresponding SDs s3 and s4 were recorded. It was
assumed in the calculations that QsL, Qb,ult, DL, and LL were sta- ðmeanÞ ðmeanÞ
QsL þ Qb;ult m1 þ m2
tistically independent and uncorrelated. ðFSÞmean ¼ ¼ ð40Þ
ðDLÞ ðmeanÞ
þ ðLLÞ ðmeanÞ m3 þ m4

ðmeanÞ ðmeanÞ
Location of Limit States and Calculation of Load and where QsL , Qb;ult , ðDLÞðmeanÞ , and ðLLÞðmeanÞ 5 mean values of
Resistance Factors the shaft and base resistances and DLs and LLs, respectively.
ðLSÞ ðLSÞ
The limit state values Qb;ult , QsL , ðDLÞðLSÞ , and ðLLÞðLSÞ were
obtained after Eq. (30) was satisfied for a target bHL by first cal-
culating Y* at the most probable failure point from Eq. (33) and then Verification of Target Probability of Failure Using
transforming Y* back to X* using the relationships: Xi* 5 si Y i*1 Monte-Carlo Simulations
ðLSÞ
mi for i 5 1, 2, and 3, and X4p 5 expð24 Y4p 1 l4 Þ [X1p 5 QsL , Eq. (29) gives an approximate estimate of the probability of failure pf
ðLSÞ ðLSÞ ðLSÞ
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

X2p 5 Qb;ult ,X3p 5 ðDLÞ , and X4p 5 ðLLÞ ]. At the limit state, from bHL for the nonlinear limit state function [Eq. (30)] used in the
Eq. (24) is valid, which can be rewritten as reliability analysis. Because target bHL was used in the calculations
in place of target pf to estimate the resistance and load factors, it was
ðLSÞ ðLSÞ
QsL ðnÞ Qb;ult ðnÞ ðDLÞðLSÞ ðnÞ ðLLÞðLSÞ necessary to determine whether the target pf assumed in the cal-
ðnÞ
QsL þ ðnÞ
Qb;ult ¼ ðnÞ
ðDLÞ þ ðnÞ
ðLLÞðnÞ culations based on the target bHL was sufficiently accurate. This was
QsL Qb;ult ðDLÞ ðLLÞ done by performing M-C simulations, which were performed in
ð34Þ addition to the M-C simulations performed previously to determine
the distributions of shaft and base capacities for Soil Profiles 1e6
Defining and Drilled Shafts AeE. The earlier set of M-C simulations was
necessary to perform the reliability analysis, while this new set acts
ðLSÞ ðnÞ as a means to check the accuracy of the reliability analysis.
ðRFÞs ¼ QsL =QsL ð35Þ
Using the means and COVs of shaft and base resistances (as
obtained from the earlier set of M-C simulations) and the means and
ðLSÞ ðnÞ
ðRFÞb ¼ Qb;ult =Qb;ult ð36Þ COVs of DLs and LLs (the mean m3 and m4 of the DLs and LLs were
obtained from the reliability analysis), the new set of M-C simu-
lations was performed with ntotal number of runs to generate random
ðLFÞDL ¼ ðDLÞðLSÞ =ðDLÞðnÞ ð37Þ values of QsL, Qb,ult, DL, and LL, and the number of runs nf for
which ðQsL 1 Qb;ult Þ 2 ðDL 1 LLÞ was less than zero (i.e., the num-
ðLFÞLL ¼ ðLLÞðLSÞ =ðLLÞðnÞ ð38Þ ber of times for which demand exceeded capacity) was recorded. The
ratio nf /ntotal gives an estimate of the probability of failure pf.
Eq. (34) can be rewritten as The minimum value of ntotal required to obtain a reliable pre-
diction of pf depends on the target pf, and can be determined from the
ðnÞ ðnÞ following equation (Fenton and Griffiths 2008):
ðRFÞs QsL þ ðRFÞb Qb;ult ¼ ðLFÞDL ðDLÞðnÞ þ ðLFÞLL ðLLÞðnÞ
ð39Þ  za 2
ntotal ¼ pf 1 2 pf ð41Þ
ðnÞ ðnÞ e
The nominal resistances QsL and Qb;ult
were calculated using the
deterministic Eqs. (1)e(6). The nominal loads, ðDLÞðnÞ and where e 5 maximum error in pf estimated with a confidence interval
ðLLÞðnÞ , were calculated from the corresponding mean loads of a (expressed either as a percentage or a fraction), and za 5 number
ðDLÞðmeanÞ ð5m3 Þ and ðLLÞmean ð5m4 Þ, by dividing the mean loads of SDs of the standard normal variate Z (with mean z 5 0 and
by the corresponding bias factors (DL bias factor 5 1.05; LL bias SD sz 5 1), such that the probability P(z 2 zasz , Z , z 1 zasz) 5
factor 5 1.0 or 1.15). Using the limit state and nominal values, P(2za , Z , za) (i.e., the area under the standard normal curve
the resistance and load factors were calculated using Eqs. extending symmetrically about the zero mean to a distance of 6za)
(35)e(38). equals a. The values of za are calculated from an inverse error
In going from Eq. (24) to Eq. (39), the point of focus shifts from function and are readily available. For example, for 90% confidence,
the limit (failure) state values of loads and resistances to values a 5 0.9, for which za can be calculated as 1.645. This equation is
corresponding to a safe state by a distance bHL that has an acceptable obtained with the assumption that each M-C run is a Bernoulli trial.
quantified risk (or probability of failure). In design, the equality sign Using Eq. (41), the required number of M-C runs ntotal was
in Eq. (39) is replaced by $ [Eq. (10)], indicating that the sum of the estimated to be equal to 270,332 for a target pf 5 1023 with 90%
factored resistances must be no less than the sum of the factored confidence (a 5 0.9; za 5 1.645) that the maximum relative error
loads. in the estimation of pf is 10% (i.e., e 5 0.0001). The authors chose
ntotal 5 271,000; thus, nf 5 271 corresponds to pf 5 1023. For the
target pf 5 1024, ntotal required to maintain the error in pf within the
Mean Factor of Safety
stringent bound of 10%, with 90% confidence, was prohibitively
To feel comfortable with the LRFD framework, some designers like large. Consequently, the authors chose ntotal 5 1,000,000, which
to estimate the factor of safety of their designs so that they can link ensured, with 90% confidence, that the maximum relative error in the
their designs performed using the LRFD method with the WSD estimation of pf is 16.5%. For practical purposes, the relative error
framework. Although the traditional factor of safety of the WSD of 16.5% is perfectly acceptable for the low probability-of-failure
method cannot be defined in the LRFD framework, a mean factor of case where only one out of 10,000 piles is expected to reach the limit

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1463

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

 state. For ntotal 5 1,000,000, nf 5 1,000 corresponds to the target
pf 5 1024.
It was observed from the M-C simulations that pf was estimated
quite accurately from the reliability analysis using target bHL for the
cases with (LL)/(DL) 5 1.0 and 2.0—the calculated values of pf
from the M-C simulations fall within a narrow range around 1 3 1023
for the cases for which target bHL 5 3.090 was used and around
1 3 1024 for which target bHL 5 3.719 was used. However, for the
case with (LL)/(DL) 5 0.25, conservative results were obtained using
the reliability analysis with target bHL—the values of pf calculated
from the M-C simulations was around 0.4 3 1023 for the cases for
which target bHL 5 3.090 was used and around 0.35 3 1024 for the
cases for which target bHL 5 3.719 was used. Because the estimates
from bHL were conservative for the cases with (LL)/(DL) 5 0.25 and
accurate for the cases with (LL)/(DL) 5 1.0 and 2.0, the results
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

of the reliability analysis using bHL were accepted as final.

Adjusted Load and Resistance Factors

Fig. 3. Variation of optimal resistance and load factors with critical state
The resistance and load factors obtained using the method described friction angle fc
in the previous section are optimal factors that satisfy the limit state
condition [Eq. (39)]. Ideally, the optimal factors should be used in
design. However, most designs of drilled shafts are done with
factored loads estimated by structural engineers who use load factors
prescribed in codes. The optimal resistance factors are not com-
patible with the code-prescribed load factors because their combi-
nation does not satisfy the limit state [Eq. (39)]. Consequently, the
optimal resistance factors were adjusted to make them compatible
with the load factors prescribed in the codes. This was done using
adjusted resistance factors (RF)code, to be used with design codes,
given by (Foye et al. 2006b)
( )
code opt ðLFÞcode ðLFÞ code
ðRFÞ ¼ ðRFÞ min DL
;  LL
ð42Þ
ðLFÞopt
DL ðLFÞopt
LL

where (RF)opt 5 optimal resistance factor (for either shaft or base

resistance) as obtained from the analysis previously described;
ðLFÞcode code
DL , ðLFÞLL 5 dead and live-load factors prescribed in code,
respectively; ðLFÞopt opt
DL , ðLFÞLL 5 optimal dead and live-load factors
(obtained from analysis previously described), respectively; and min
{×, ×} 5 minimum of the two arguments within {}. In design,
(RF)code should be used with code-specified load factors. However,
if designs can be done with optimal load factors, then (RF)opt should
be used. This paper uses the AASHTO (2007) recommended load
factors, ðLFÞcode code
DL 5 1:25 and ðLFÞLL 5 1:75, to obtain the adjusted
resistance factors ðRFÞb and ðRFÞcode
code
. and load factors of Drilled Shaft B were plotted for the Soil Profiles
s
Results
Although the primary objective of this paper is to present resistance
factors for use in design, the authors also investigated how the op-
timal resistance and load factors change with different field con-
ditions and variables. The optimal load and resistance factors were
found to be independent of both fc and K0 for all practical purposes.
Figs. 3 and 4 show the plots of the resistance and load factors as
functions of fc,mean and K0, respectively, for Drilled Shaft D in-
stalled in Soil Profile 1 with (LL)/(DL) 5 1.0 and target pf 5 1023.
The invariance of the resistance and load factors observed in Figs. 3
and 4 were typical of all the other drilled shafts and soil profiles; the
minor fluctuations observed were rather random without any par-
ticular trend. The effect of the water table on the resistance and load
factors was also minimal. This was evident from a comparison of the
Fig. 4. Variation of optimal resistance and load factors with coefficient
of earth pressure K0 at rest
results of Soil Profiles 1e4 in Fig. 5, in which the optimal resistance
1e5 and for (LL)/(DL) 5 0.25 and target pf 5 1024. Fig. 5 shows
that the fluctuations in the resistance and load factors were not
significant from one soil profile to the other for all practical purposes.
This trend was also observed for the other drilled shafts and for other
values of (LL)/(DL) and pf. The fluctuations in the resistance and
load factors from one pile to another were also not significant, as
evident from Fig. 6, which shows the optimal resistance and load
factors for Drilled Shafts AeD installed in Soil Profile 3 and for
(LL)/(DL) 5 2.0 and pf 5 1024. The trends observed in Fig. 6 are
typical of all the other cases.
Although soil properties and profiles and pile dimensions have
little effect on the resistance and load factors, the (LL)/(DL) ratio has
a nonnegligible effect on the live-load factor. As Fig. 7 shows, the
optimal resistance factors slightly increase and the optimal DL factor
slightly decreases with an increase in the (LL)/(DL) ratio; however,
the increase in the optimal live-load factor with increase in the
(LL)/(DL) ratio is significant. Fig. 7 was plotted for Drilled Shaft D

1464 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Optimal resistance and load factors for different soil profiles Fig. 7. Variation of optimal resistance and load factors with live load to
dead load ratio

Fig. 6. Optimal resistance and load factors for different drilled shaft
dimensions

installed in Soil Profile 1 for different values of (LL)/(DL) and pf 5

1024; the trend was consistent for all the other cases.
Because (LL)/(DL) affects the resistance and load factors, it must
have a nonnegligible effect on the mean factor of safety, (FS)mean.
(FS)mean was plotted against pf in Fig. 8(a) for different values of
(LL)/(DL) for Drilled Shaft B installed in Soil Profile 3. Fig. 8(a)
shows that for a particular pf, (FS)mean increases with an increase in
the (LL)/(DL) ratio. Thus, (FS)mean is not a true indicator of the
reliability of a design: for the same mean factor of safety, the
probabilities of failure are different for different proportions of
the applied LL. From another viewpoint, for the same probability of
failure, a higher factor of safety is required to account for the higher
uncertainty levels of LL if (LL)/(DL) increases. The (FS)mean plots
were almost identical for all the different cases studied.
The effect of target pf on the optimal resistance and load factors
are shown in Fig. 8(b) for the same drilled shaft and soil profile of Fig. 8. Variation of (a) mean factor of safety; (b) optimal resistance and
Fig. 8(a) and for (LL)/(DL) 5 1.0. The change in the values of the load factors with target probability of failure
resistance and load factors because of the change in pf is most

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1465

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

 Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

live load to dead load ratio

optimal factors; (b) optimal and code-adjusted resistance factors with
Fig. 9. Variation of (a) Ellingwood and Tekie (1999) and FHWA (2001)
code
Table 1. Resistance Factors Averaged across Different Drilled Shafts and Adjusted with ðLFÞcode
DL 5 1:25 and ðLFÞLL 5 1:75

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

Soil Profile

(1) (2) (3) (4) (5) (6)

Probability
of failure Statistics ðRFÞcode
s ðRFÞcode
b ðRFÞcode
s ðRFÞcode
b ðRFÞcode
s ðRFÞcode
b ðRFÞcode
s ðRFÞcode
b ðRFÞcode
s ðRFÞcode
b ðRFÞcode
s ðRFÞcode
b
23
10 Mean 0.805 0.916 0.801 0.970 0.823 0.959 0.821 0.955 0.815 0.956 0.835 0.920

1466 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012

SD 0.027 0.062 0.023 0.050 0.024 0.069 0.022 0.069 0.026 0.069 0.031 0.031
Maximum 0.851 0.984 0.852 1.067 0.861 1.091 0.856 1.088 0.851 1.088 0.871 0.955
Minimum 0.745 0.789 0.774 0.890 0.789 0.867 0.789 0.862 0.766 0.856 0.811 0.896
1024 Mean 0.704 0.809 0.715 0.831 0.723 0.851 0.721 0.848 0.713 0.847 0.740 0.813
SD 0.029 0.077 0.052 0.103 0.053 0.101 0.051 0.101 0.048 0.098 0.076 0.079
Maximum 0.760 0.947 0.808 1.036 0.814 1.062 0.807 1.059 0.799 1.057 0.827 0.904
Minimum 0.663 0.708 0.657 0.706 0.667 0.734 0.667 0.729 0.654 0.730 0.691 0.762
 Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.

J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.

Fig. 10. Summary of design equations, variable and model uncertainties, load and resistance factors

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1467

 pronounced for the live-load factor. This also corroborates that the
change in (FS)mean because of the change in pf is strongly influenced
by the LL. Similar trends were observed for the other drilled shafts
and soil profiles. To study the effect of pf, additional results were
generated with target pf 5 1022, 1021, and 1.0.
Because the LL has the most effect on the reliability study, it was
considered necessary to investigate how much change would occur
in the values of resistance factors if, instead of the choice of LL
COV 5 0.25 and bias factor 5 1.0 (Ellingwood and Tekie 1999), the
FHWA (2001) recommended LL COV 5 0.18 and bias factor 5
1.15 was used. For this purpose, additional analyses for the different
drilled shafts installed in Soil Profile 1 were performed. Fig. 9(a)
shows the optimal resistance and load factors for both sets of live-
load COV and bias factor plotted as a function of the (LL)/(DL) ratio
for Drilled Shaft D and for pf 5 1024. Fig. 9(b) shows the optimal
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.
and code-adjusted resistance factors for the same case [code-
adjusted resistance factors were obtained from the optimal re-
sistance factors by using Eq. (42)]. The figures, typical of all the
drilled shafts, show that the resistance factors obtained by using both
sets of live-load COV and bias factor are nearly identical. The set by
Ellingwood and Tekie (1999) produced slightly conservative code-
adjusted resistance factors on average; hence, this set was used in the
calculations. Fig. 9(b) also shows that the adjusted resistance factors
do not vary significantly with the (LL)/(DL) ratio. At the same time,
the adjusted factors, similar to the optimal factors, do not change
much across soil profiles or across drilled shafts.
Because the code-adjusted resistance factors do not vary signif-
icantly for the different drilled shafts, the results of the adjusted
resistance factors over all the drilled shafts for a particular soil profile
were consolidated. The mean, SD, and maximum and minimum of
the resistance factors obtained for the different drilled shafts in-
stalled in a particular soil profile were calculated. These statistics for
the different profiles are reported in Table 1. The data for pf 5 1023
and 1024 is presented separately. Based on Table 1, the following
values are recommended for use in design: ðRFÞcode s 5 0:70 and
23
ðRFÞcode
b 5 0:75 for p f 5 10 , and ðRFÞ code
s 5 0:65 and ðRFÞcode
b 5
24
0:70 for pf 5 10 . Fig. 10 summarizes the design methodology; it
contains the design equations, the different soil variables with their
uncertainties, and the resistance and load factors.
Conclusions
A systematic probabilistic analysis was performed to develop the
resistance factors for drilled shafts in normally consolidated sand for
a soil variable-based design method. The analysis involved identifi-
cation of a robust design method, quantification of the uncertainties
(probability distributions) associated with the design variables and the
design equations, generation of the probability distributions of shaft
and base capacities using M-C simulations, performance of a re-
liability analysis based on target reliability index to obtain the limit
state and nominal resistances and loads, and performing M-C simu-
lations using the results of the reliability analysis to ensure that the
target reliability index produced the desired target probability of
failure. From the calculated limit state and nominal values of shaft and
base capacities and DLs and LLs, the optimal resistance and load
factors were obtained. The optimal resistance factors were then ad-
justed to make them compatible with the dead and live-load factors
recommended by AASHTO (2007).
In the course of the study, performed for six different soil profiles
with different soil properties and for five different dimensions of
drilled shafts, the resistance and load factors did not vary to any
significant extent between the different soil profiles and drilled
shafts considered. The ratio of (LL)/(DL) was identified as the only
1468 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012
J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.
variable that affected the results; however, it affected mostly the LL
factor, with minimal effect on the resistance and DL factors. The
mean factor of safety was found, reasonably, to increase with a de-
crease in the target probability of failure. The change in the mean
factor of safety with the target probability of failure was a function of
the (LL)/(DL) ratio.
Based on the study, the authors recommend base and shaft re-
sistance factors that can be used in design with the AASHTO (2007)
recommended DL and LL factors of 1.25 and 1.75, respectively. The
recommended shaft and base resistance values are 0.70 and 0.75 for
a probability of failure of 1023 and 0.65 and 0.70 for a probability of
failure of 1024.
Acknowledgments
This work was supported by the Joint Transportation Research Pro-
gram administered by the Indiana Department of Transportation and
Purdue University. The contents of this paper reflect the views of the
authors, who are responsible for the facts and the accuracy of the
data presented herein, and do not necessarily reflect the official
views or policies of the Federal Highway Administration and the
Indiana Department of Transportation, nor do the contents constitute
a standard, specification, or regulation.
References
AASHTO. (2007). AASHTO LRFD bridge design specifications, Wash-
ington, DC.
Ang, A. H.-S., and Tang, W. H. (1984). Probability concepts in engineering
planning and design: Decision, risk and reliability, Vol. 2, Wiley, New York.
Baecher, G. B., and Christian, J. T. (2003). Reliability and statistics in
geotechnical engineering, Wiley, New York.
Bolton, M. D. (1986). “The strength and dilatancy of sands.” Geotechnique,
36(1), 65e78.
Colombi, A. (2005). “Physical modeling of an isolated pile in coarse grained
soils.” Ph.D. thesis, Univ. of Ferrara, Ferrara, India.
Ellingwood, B., MacGregor, J. G., Galambos, T. V., and Cornell, C. A.
(1980). “Probability based load criteria: Load factors and load combi-
nations.” J. Struct. Eng. Div., 108(5), 978e997.
Ellingwood, B. R., and Tekie, P. B. (1999). “Wind load statistics for
probability-based structural design.” J. Struct. Eng., 125(4), 453e463.
Federal Highway Administration (FHWA). (2001). “Load and resistance
factor design (LRFD) for highway bridge substructures.” FHWA HI-98-
032, FHWA, Washington, DC.
Fenton, G. A., and Griffiths, D. V. (2008). Risk assessment in geotechnical
engineering, Wiley, New York.
Fioravante, V. (2002). “On the shaft friction modelling of non-displacement
piles in sand.” Soils Found., 42(2), 23e33.
Fisher, J. W., Galambos, T. V., Kulak, G. L., and Ravindra, M. K. (1978).
“Load and resistance factor design criteria for connectors.” J. Struct.
Eng. Div., 104(9), 1427e1441.
Foye, K. C. (2005). “A rational probabilistic method for the development of
geotechnical load and resistance factor design.” Ph.D. thesis, Purdue
Univ., West Layfayette, IN.
Foye, K. C., Salgado, R., and Scott, B. (2006a). “Assessment of variable
uncertainties for reliability-based design of foundations.” J. Geotech.
Geoenviron. Eng., 132(9), 1197e1207.
Foye, K. C., Salgado, R., and Scott, B. (2006b). “Resistance factors for use in
shallow foundation LRFD.” J. Geotech. Geoenviron. Eng., 132(9),
1208e1218.
Ghionna, V. N., Jamiolkowski, M., Pedroni, D. and Salgado, R. (1994). “The tip
resistance of drilled shafts in sands.” Vertical Horizontal Deform. Found.
Embank.: Proc., Settlement 1994, Vol. 2, ASCE, Reston, VA, 1039e1057.
Ghosn, M., and Moses, F. (1998). “Redundancy in highway bridge
superstructures.” NCHRP Rep. 406, Transportation Research Board,
Washington, DC.
 Hammitt, G. M. (1966). Statistical analysis of data from comparative
laboratory test program sponsored by ACIL, U.S. Army Waterways
Experiment Station, Vicksburg, MS.
Hansell, W. C., and Viest, I. M. (1971). “Load factor design for steel
highway bridges.” AISC Eng. J., 8(4), 113e123.
Honjo, Y., Suzuki, M., Shirato, M., and And Fukui, J. (2002). “De-
termination of partial factors for a vertically loaded pile based on re-
liability analysis.” Soils Found., 42(5), 91e109.
Joint Committee on Structural Safety. (2000). “Probabilistic model code.”
JCSS-OSTL/DIA/VROU-10-11-2000, Joint Committee on Structural
Safety, Washington, DC.
Kim, D. W. (2008). “Load and resistance factor design of slopes and MSE
walls.” Ph.D. thesis, Purdue Univ., West Lafayette, IN.
Kim, D. W., and Salgado, R. (2012). “Load and resistance factors for ex-
ternal stability checks of mechanically stabilized earth walls.” J. Geo-
tech. Geoenviron. Eng., 137(3), 241e251.
Downloaded from ascelibrary.org by Univ Of Waterloo on 12/10/12. Copyright ASCE. For personal use only; all rights reserved.
Kwak, K., Park, J. H., Kim, K. J., Choi, Y. K., and Huh, J. (2007).
“Evaluation of resistance bias factors for load and resistance factor
design of driven pipe piles.” Contemporary Issues in Deep Foundations,
GSP 158, ASCE, Reston, VA.
Lacasse, S., and Nadim, F. (1996). “Uncertainties in characterising soil
properties.” Proc., Uncertain. Geol. Environ., 96, 49e73.
Lee, J. H., and Salgado, R. (1999). “Determination of pile base resistance in
sands.” J. Geotech. Geoenviron. Eng., 125(8), 673e683.
Lee, I. K., White, W., and Ingles, O. G. (1983). Geotechnical engineering,
Pitman Publishing, Pitman, NJ.
Loukidis, D., and Salgado, R. (2008). “Analysis of shaft resistance of non-
displacement piles in sand.” Geotechnique, 58(4), 283e296.
Loukidis, D., Salgado, R., and Abou-Jaoude, G. (2008). “Assessment of
axially-loaded pile dynamic design methods and Review of INDOT
axially-loaded pile design procedure.” FHWA/IN/JTRP-2008/6, Purdue Univ.
Lumb, P. (1974). “Application of statistics in soil mechanics.” Soil mechanics—
New horizons, I. K. Lee, ed., Elsevier.
McVay, M., et al. (2003). “Use of LRFD, cost and risk to design a drilled
shaft load test program in Florida limestone.” TRB 2003 Annual Meeting
(CD-ROM), Transportation Research Board, Washington, DC.
Misra, A., and Roberts, L. A. (2006). “Probabilistic analysis of drilled shaft
service limit state using the “tez” method.” Can. Geotech. J., 43(12),
1324e1332.
Misra, A., and Roberts, L. A. (2009). “Service limit state resistance factors
for drilled shafts.” Geotechnique, 59(1), 53e61.
Misra, A., Roberts, L. A., and Levorson, S. M. (2007). “Reliability of drilled
shaft behavior using finite difference method and Monte Carlo simu-
lation.” Geotech. Geol. Eng., 25(1), 65e77.
Negussey, D., Wijewickreme, W. K. D., and Vaid, Y. P. (1988). “Constant-
volume friction angle of granular materials.” Can. Geotech. J., 25(1),
50e55.
Nowak, A. S. (1994). “Load model for bridge design code.” Can. Geotech. J.,
21, 36e49.
JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / DECEMBER 2012 / 1469
View publication stats J. Geotech. Geoenviron. Eng. 2012.138:1455-1469.
O’Neill, M. W., and Hassan, K. M. (1994). “Drilled shafts: Effects of
construction on performance and design criteria.” Proc., Int. Conf.
Design Constr. Deep Found., FHWA, Washington, DC, 137e187.
O’Neill, M. W., and Resse, L. C. (1999). “Drilled shafts: Construction
procedures and design methods.” Rep. No. FHWA-IF-99-025, Federal
Highway Administration, Washington, DC.
Ochiai, H., Otani, J., and Matsui, K. (1994). “Performance factor for bearing
resistance of bored friction piles.” Struct. Saf., 14(1e2), 103e130.
Paikowsky, S. G. (2004). “Load and resistance factor design (LRFD) for
deep foundations.” NCHRP Rep. 507, Transportation Research Board,
Washington, DC.
Phoon, K.-K., and Kulhawy, F. H. (1999). “Characterization of geotechnical
variability.” Can. Geotech. J., 36(4), 612e624.
Reese, L. C., Touma, F. J., and O’Neill, M. W. (1976). “Behavior of drilled
piers under axial loading.” J. Geotech. Eng. Div., 102(5), 493e510.
Salgado, R. (1993). “Analysis of penetration resistance in sands.” Ph.D.
thesis, Univ. of CaliforniaeBerkeley, Berkeley, CA.
Salgado, R. (2008). The engineering of foundations, McGraw Hill, New
York.
Salgado, R., and Prezzi, M. (2007). “Computation of cavity expansion
pressure and penetration resistance in sands.” Int. J. Geomech., 7(4),
251e265.
Salgado, R., and Randolph, M. F. (2001). “Analysis of cavity expansion in
sand.” Int. J. Geomech., 1(2), 175e192.
Stas, C. V., and Kulhawy, F. H. (1984). “Critical evaluation of design
methods for foundations under axial uplift and compression loading.”
EPRI Rep. EL-3771, Research Project 1493-1, Electric Power Research
Institute, Palo Alto, CA.
Titi, H. H., Mahamid, M., Abu-Farsakh, M. Y., and Elias, M. (2004).
“Evaluation of CPT methods for load and resistance factor design of
driven piles.” GeoTrans 2004, ASCE, Reston, VA.
Verdugo, R., and Ishihara, K. (1996). “The steady state of sandy soils.” Soils
Found., 36(2), 81e91.
White, D. J., Schaefer, V. R., Yang, H., and Thompson, M. J. (2005).
“Innovative solutions for slope stability reinforcement and character-
ization: Vol. I.” Final Rep. CTRE Project 03-127, Center for Trans-
portation Research and Education, Iowa State Univ., IA.
Yang, X., Han, J., Parsons, R. L., and Henthorne, R. W. (2008). “Resistance
factors for drilled shafts in weak rock based on O-cell test data.” J. TRB,
2045, 62e67.
Yoon, G. L., and O’Neil, M. W. (1996). “Design model bias factors for
driven piles from experiments at NGES-UH.” Proc., Uncertain. Geol.
Environ., 58(2), 759e773.
Yoon, S., Abu-Farsakh, M. Y., Tsai, C., and Zhang, Z. (2008). “Calibration
of resistance factors for axially loaded concrete piles driven into soft
soils.” J. TRB, 2045, 39e50.
Zhang, L. M., Li, D. Q., and Tang, W. H. (2005). “Reliability of bored pile
foundations considering bias in failure criteria.” Can. Geotech. J., 42(4),
1086e1093.