Model Uncertainty of Eurocode 7 Approach for Bearing

Capacity of Circular Footings on Dense Sand

Chong Tang1 and Kok-Kwang Phoon, F.ASCE2
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

Abstract: This paper presents a critical evaluation of the model factor M = qu,m/qu,c for Eurocode 7 calculating the bearing capacity of
circular footings on dense sand, where qu,m = measured capacity and qu,c = Eurocode 7 calculated capacity. Regression analysis is
required to remove the dependency of M on the input parameters. Because the input parameters cannot be varied systematically in load
tests, previous studies showed that finite-element limit analysis (FELA) can be used as an alternative to load tests for regression. This is
further verified from the model factor MFELA = qu,m/qu,FELA with a mean of 1 and a coefficient of variation (cov) of 0.1, where qu,FELA = FELA
predicted capacity. A correction factor (Ms = qu,FELA/qu,c) is next defined, which can be decomposed as a product of a systematic part f and a
residual part h (i.e., Ms = f h ), which is modeled as a lognormal random variable with mean = 1 and cov = 0.11. Finally, a new model factor
(M0 = qu,m/q0 u,c = qu,m/fqu,c) is defined. The model statistics of M0 = h MFELA can be obtained from those of h and MFELA, where mean = 1 and
cov = 0.11. This is consistent with the results (mean = 1.02 and cov = 0.15) characterized using the load-test database directly, because M0 no
longer depends on the input parameters. DOI: 10.1061/(ASCE)GM.1943-5622.0000737. © 2016 American Society of Civil Engineers.

Introduction (2006) characterized model uncertainty in the normalized load-

displacement curves for augered cast-in-place piles under axial
compression using a pair of correlated curve-fitting parameters.
Characterization of Model Uncertainty Based on four load-test databases from South Africa for driven
Because of the simplifications of many geotechnical calculation piles in noncohesive soils (53 tests), bored piles in noncohesive
models developed for practical design, model uncertainty always soils (33 tests), driven piles in cohesive soils (59 tests), and bored
exists and can be significant (e.g., Ang and Tang 1984; Phoon and piles in cohesive soils (53 tests), Dithinde et al. (2011) character-
Kulhawy 2005). As mentioned by Tang and Gilbert (1993) and ized the model uncertainty for some static pile design formulae.
Lacasse and Nadim (1994), the calculated failure probabilities with- Recently, Burlon et al. (2014) proposed a new, improved calcula-
out considering model uncertainty are not actual failure probabil- tion model (i.e., pressuremeter test 2012 model) with the corre-
ities of geotechnical systems. For reliability-based design of geo- sponding model factor to meet the requirements of Eurocode 7
technical structures, it is essential to characterize the model (CEN 2004), where a database of 174 full-scale static pile load
uncertainty of a predictive model. In practice, the model uncertainty tests carried out by the Laboratoire Central des Ponts et Chausees
is usually represented in terms of the probability model (including (French Institute of Science and Technology for Transport,
the probability distribution function, the mean, and the standard Spatial Planning, Development and Networks, Paris) is used.
deviation) of a model factor, which is defined as the ratio of the For shallow foundations, Muganga (2008) quantified the uncer-
measured performance to the predicted value. tainty in the prediction of the displacement and capacity of shallow
Some studies have been carried out to establish model uncer- foundations on rock. Paikowsky (2010) estimated the model statis-
tainty over the past decades. For axial-loaded piles in various soil tics under various loading conditions using an assembled database,
conditions, Briaud and Tucker (1988) evaluated 13 methods which includes 549 load tests of foundations mostly in or on granu-
designed to predict the ultimate load of a pile and five methods lar soils. Phoon and Tang (2015a, b) investigated the model uncer-
designed to predict the settlement of a pile based on a 98-pile load- tainty of Eurocode 7 (CEN 2004) for bearing capacity of strip foot-
test database. Paikowsky (2002) investigated the model statistics ings on sand under general combined loading. Phoon et al. (2007)
for deep foundation resistance factors. In the case of laterally loaded characterized the uncertainty in the uplift load-displacement curves
rigid drilled shafts (bored piles), Phoon and Kulhawy (2005) pre- associated with spread footings, drilled shafts, and pressure-injected
sented a critical evaluation of model factors both for lateral and footings. Uzielli & Mayne (2012) presented the statistical character-
moment limit capacity using an extensive database. Phoon et al. ization of the load-displacement behavior of shallow footings on
cohesionless soils and the probabilistic estimation of settlement for
serviceability limit state design. Sivakugan and Johnson (2004) car-
1
Research Fellow, Dept. of Civil and Environmental Engineering, ried out a statistical analysis on settlements of shallow foundations
National Univ. of Singapore, Block E1A, #07-03, 1 Engineering Dr. 2, on sand predicted by some empirical approaches. It was found that
Singapore 117576 (corresponding author). E-mail: ceetc@[Link] the settlement ratio, defined as the ratio of predicted to actual settle-
2
Professor, Dept. of Civil and Environmental Engineering, National ments, follows a beta distribution. These aforementioned research
Univ. of Singapore, Block E1A, #07-03, 1 Engineering Dr. 2, Singapore works were based on a load-test database.
117576. E-mail: kkphoon@[Link]
In contrast, Zhang et al. (2009) proposed a framework for charac-
Note. This manuscript was submitted on December 2, 2015; approved
on May 26, 2016; published online on August 11, 2016. Discussion period terizing geotechnical model uncertainty, which is based on the con-
open until January 11, 2017; separate discussions must be submitted for cept of multivariate Bayesian updating. The proposed framework
individual papers. This paper is part of the International Journal of was applied to establish the model uncertainty of four limit equilib-
Geomechanics, © ASCE, ISSN 1532-3641. rium methods for slope-stability analysis using centrifuge test data.

© ASCE 04016069-1 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

 Bearing Capacity of Circular Footings on Sand combined loading (e.g., Phoon and Tang 2015a, b). The key postu-
late of this framework is that the finite-element limit analysis
According to Eurocode 7 (CEN 2004), the ultimate bearing capacity
(FELA) is good enough to replace load tests, which will be exam-
of a surface circular foundation with diameter D on sand is usually
ined again in the present work. Note that Zhang et al. (2015) has
calculated as
also applied the same framework successfully to characterize the
qu ¼ 0:5 g DN g s g (1a) model uncertainty for cantilever-wall deflections in undrained clay.
The numerical method adopted in their paper is the FEM rather than
FELA.
where g = unit weight of sand; the design values of dimensionless
It would appear that sufficiently sophisticated numerical meth-
bearing-capacity factors Nq and Ng are determined as follows:
ods that model the physics of the problem correctly can produce
predictions that are close to field measurements. This is a key ele-
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

Nq ¼ ep tan f tan 2 ðp =4 þ f =2Þ

N g ¼ 2ðNq  1Þtan f (1b) ment of the model uncertainty approach discussed later.

and the shape factor sg = 0.7, where f = the internal friction angle FELA with Stress Level Effect
of sand. Some theoretical and numerical studies associated with the
bearing capacity of circular foundations on sand can be found in
Lee and Salgado (2005), Lyamin et al. (2007), Kumar and Summary of FELA
Chakraborty (2013), Azzam and ElWakil (2015), and Chakraborty Limit analysis is based on the lower- and upper-bound theorems
and Kumar (2015). developed by Drucker et al. (1951, 1952), where the soil is
According to Eqs. (1a) and (1b), the model uncertainty of assumed to be a perfectly plastic material and obeys an associated
Eurocode 7 (CEN 2004) mainly arises from the bearing-capacity flow rule. On this basis, this method can give a lower and upper
factor (Ng ) due to the scale effect. This can be partly explained as bound of the actual collapse load (Chen 1975). Over the past dec-
the dependency of friction angle ( f ) on the stress level (e.g., Bolton ades, finite-element (FE) formulations of limit analysis have
1986; Bolton and Lau 1989). A large foundation diameter (D) leads become increasingly popular (Sloan 2013) because FELA can
to increased qu, reduced friction angle ( f ) mobilized along the fail- easily model complex problem geometries, layered soil profiles,
ure mechanism, and reduced Ng (e.g., Okahara et al. 1988; and complicated loadings.
Kusakabe et al. 1991; Ueno et al. 1998; Perkins and Madson 2000; In this paper, only the lower-bound method is used, because it
Zhu et al. 2001; Cerato and Lutenegger 2007; Loukidis and Salgado can provide a conservative estimate of the actual collapse load.
2011). White et al. (2008) stated that “there is not yet an accepted The lower-bound theorem states that when the stress field within
consensus on: (a) appropriate factors for use in the bearing capacity the soil mass is stable (i.e., the stress field does not violate the
equation; (b) methods for selecting an appropriate friction angle for yield criterion at any point) and statically admissible (i.e., the
calculating bearing capacity; or (c) a quantitative method to capture stress field is in equilibrium with the surface tractions and body
the observed scale effects that lead to differences in the behavior forces), collapse does not occur. The stress field is discretized
observed in small-scale tests at unit gravity compared with behavior into a set of nodal stresses at each node using the FE concept. As a
at full scale or within a geotechnical centrifuge.” For the first issue, result, the determination of the lower bound to the collapse load is
Diaz-Segura (2013) evaluated the range of variation of the bearing- formulated as a second-order cone programming problem, which
capacity factor (Ng ) using 60 estimation methods. This shows a can be efficiently solved using MOSEK. Full details about this
marked dependence on the methods used to determine Ng , where the methodology can be found in Tang et al. (2014).
differences for the same ( f ) values are up to 267% between esti- Because of the high stress gradient around the footing edges that
mated values. For the second aspect, Diaz-Segura (2013) also cannot be captured using the linear elements and the change in the
pointed out that uncertainty in the estimation of f , due to the use of direction of the principal stress, a fine fan of elements at the edges
correlations with in situ tests, leads to a range of variation for Ng of the footing is desirable. An illustration of the FE mesh together
higher than that observed using the 60 estimation methods. with the stress boundary conditions are presented in Fig. 1. The do-
main is discretized into a number of three-node triangular elements.
Objective and Scope of This Paper Unlike the usual form of mesh used by FE analysis, each node is
unique to a particular element, and more than one node may share
The main objective of this paper is to establish the probability the same coordinate. The foundation–soil interface (i.e., Edge AB
model of a model factor for the Eurocode 7 (CEN 2004) approach in Fig. 1) is assumed to be fully rough. In this case, the shear stress
to calculate the ultimate bearing capacity of a circular foundation on (t xy) along Edge AB is unconstrained (Makrodimopoulos and
dense sand. The model factor is defined as the ratio of the measured Martin 2006).
capacity to the predicted capacity according to Eurocode 7 (CEN
2004). Note that the model factor may be systematically affected by
input parameters, particularly if the prediction model is overly sim-
plified to reduce the solution to a simple analytical form as adopted
by Eurocode 7 (CEN 2004) [i.e., Eqs. (1a) and (1b)].
It is not easy to remove the dependence of the model factor on
these input parameters by regressing the model factor against each
input parameter, because the values of these input parameters can-
not be varied systematically in a load-test database for regression
analysis. Therefore, the framework proposed by Phoon and Tang
(2015a) is used here, which has been applied to characterize the
model uncertainty of the Eurocode 7 (CEN 2004) approach for esti- Fig. 1. Illustration of FE mesh used in numerical simulation
mating the bearing capacity of a strip footing on sand under general

© ASCE 04016069-2 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

 Bolton’s Strength-Dilatancy Relation 10 kN/m3 is chosen for the unit weight of sand. For Toyoura sand,
the critical-state friction angle ( f cv) can be 32° according to
It has been recognized that the friction angle ( f ) of sand strongly
Jamiolkowski et al. (2003). In addition, the following values for the
depends on the stress level, which has a significant effect on the ulti-
foundation diameter (D) and the relative density (DR) of sand are
mate vertical bearing capacity of a foundation on sand (e.g.,
selected, which are consistent with the parameter ranges of Loukidis
Loukidis and Salgado 2011). To consider the stress-level effect, the
and Salgado (2011):
strength–dilatancy relation [Eqs. (2a)–(2d)] suggested by Bolton
1. Foundation diameter D = 1, 2, 3, 4, 5, and 6 m. As a result, g D/
(1986) is incorporated into FELA in this paper
pa = 0.10.6, where pa is the reference atmospheric pressure
IR ¼ DR ðQ  ln p0 Þ  R (2a) equal to 100 kPa.
2. Relative density of sand DR = 0.7, 0.75, 0.8, 0.85, and 0.9 for
dense sand.
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

f ¼ f cv þ mIR (2b)
Consequently, 30 cases are investigated. The Ng values from
FELA and from Eqs. (3a) and (3b) are compared in Fig. 2(a) for
0:8 c ¼ f  f cv (2c) strip footings and Fig. 2(b) for circular footings. The accuracy can
be evaluated by the ratio of Ng ,FELA/Ng ,FEM, where Ng ,FELA = Ng
sin f cos c value from FELA, and Ng ,FEM = Ng value from FEM. A good
tan f  ¼ (2d) agreement is observed in Fig. 2(a) for strip footings. The mean and
1  sin f sin c
standard deviation of Ng ,FELA/Ng ,FEM are 0.97 and 0.1, respec-
where IR = dilation indicator; DR = relative density index of sand; Q tively. The difference is because FELA and FEM are two different
and R = empirical material constants with values of 10 and 1; f = numerical methods. In addition, Eq. (3a) is an approximation of
peak friction angle (in degrees); f cv = critical state friction angle FEM results only. In Fig. 2(b), it seems that the difference between
(in degrees); c = dilation angle (in degrees); f * = reduced friction FEM and FELA results becomes larger for circular footings. The
angle; p0 = mean stress level; and m = 3 for triaxial or general corresponding mean and standard deviation of Ng ,FELA/Ng ,FEM
stresses and 5 for plane-strain conditions. Because the friction angle are 0.84 and 0.11, respectively. This can be explained according to
( f ) is a function of the stress level, an iterative procedure is the shape factor (sg ). The FELA results for the shape factor (sg ) are
illustrated in Fig. 2(c). It shows sg is between 0.56 and 0.8, whereas
required to determine the bearing capacity (qu).
To incorporate Bolton’s strength–dilatancy relation in the foun- De Beer (1965) suggested s g = 0.7–0.9 based on small-scale model
tests. The shape factor (s g ) decreases with increasing DR as
dation problem, it is more convenient to replace the mean effective
observed by Loukidis and Salgado (2011) [i.e., Eq. (3b)] and
stress at failure (p0 ) in Eq. (2a) with the bearing capacity (qu). The
increases as g D/pa increases. The latter effect is not considered in
Meyerhof (1950) recommendation of p0 = 0.1qu was used in the pres-
Eq. (3b), which contributes to the difference between Ng values
ent study, which is typically found to be appropriate (e.g., Perkins
from FEM and FELA presented in Figure 2(b).
and Madson 2000; Randolph et al. 2004; White et al. 2008). This is
Interestingly, the mean of s g from FELA is approximately 0.71
also suggested by ISO 19905-1 (see Section E.2 in ISO 2012).
(the corresponding value of FEM is 0.82), which is almost equal to
the value (s g = 0.7) suggested by Eurocode 7 (CEN 2004). The pres-
Validation with FEM Results ent results (sg < 1) are inconsistent with those calculated from limit
Loukidis and Salgado (2011) investigated the effect of relative den- analysis without stress-level effect (Lyamin et al. 2007), where
sity and stress level on the bearing capacity of footings on Toyoura s g > 1. This is because the same friction angle is used by Lyamin et
and Ottawa sands using FEM. In their work, a two-surface plasticity al. (2007) irrespective of triaxial or plane-strain conditions. This
constitutive model was incorporated into FEM to capture the will indeed produce s g > 1. However, in reality, the friction angle
aspects of the mechanical response of sands that are relevant to the under triaxial conditions is smaller than that under plane-strain con-
bearing-capacity problem. Consequently, Loukidis and Salgado ditions, which leads to sg < 1. This has also been discussed by
(2011) proposed the following equation to estimate the bearing- Loukidis and Salgado (2011).
capacity factor (Ng ) of strip footings:
 0:4 Model Uncertainty of FELA
gD
N g ¼ 2:82 exp ð3:64DR Þ (3a) It should be pointed out that the numerical model may only be an
pa approximation of the real behavior, and therefore, it may also be
subjected to model uncertainty. This can be seen in Zhang et al.
For circular footings, Loukidis and Salgado (2011) also pro- (2015) for FEM and Phoon and Tang (2015a, b) for FELA. In this
posed the following shape factor (sg ): case, the model uncertainty of FELA is mainly attributed to the fol-
lowing factors:
s g ¼ 1  0:23DR (3b) 1. The soil is idealized as a perfectly plastic material. This is a
simplification of the real soil behavior, which is quite complex.
Accordingly, the bearing capacity depends on (1) the foundation 2. Eqs. (2a)–(2d) proposed by Bolton (1986) is empirical in
diameter (D), (2) unit weight ( g ) of sand, (3) relative density (DR) nature.
of sand, and (4) the critical-state friction angle ( f cv) of sand. These The model factor MFELA for FELA can be defined as follows:
parameters can be transformed to three dimensionless parameters:
(1) g D/pa (where pa = 100 kPa), which has been widely used to MFELA ¼ qu;m =qu;FELA (4)
investigate the scale effect on the bearing capacity (e.g., De Beer
1965); (2) tan f cv; and (3) DR. where qu,m = measured capacity; and qu,FELA = FELA-predicted
Eqs. (3a) and (3b) is then used to verify the accuracy of FELA. capacity.
Loukidis and Salgado (2011) pointed out that the choice of the unit To characterize the model uncertainty of FELA, a database of 27
weight ( g ) is immaterial. Therefore, throughout this paper, g = centrifuge tests (see Table 1) for strip footings and a database of 26

© ASCE 04016069-3 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. (a) Comparison between bearing-capacity factor Ng from FELA and FEM for strip footings; (b) comparison between bearing-capacity factor
Ng from FELA and FEM for circular footings; (c) shape factor sg

Table 1. Centrifuge Test Database for Strip Foundations on Dense Sand Table 2. Centrifuge Test Database for Circular Foundations on Dense
Sand
Reference g (kN/m )3
f cv (°) DR D (m) qu,m (kPa)
Reference g (kN/m3) f cv (°) DR D (m) qu,m (kPa)
Okahara et al. (1988) 15.7 32 0.88 0.3 362
15.7 32 0.88 0.6 648 Kusakabe et al. (1991) 16.5 34 0.82 0.9 780
15.7 32 0.88 1 1,153 16.5 34 0.82 1.8 1,195
15.7 32 0.88 1.4 1,208 16.5 34 0.82 2.7 1,559
15.7 32 0.88 2 1,740 Ueno et al. (1994) 15.7 33 0.7 3 1,154
15.7 32 0.88 3 2,314 15.7 33 0.7 1.5 619
15.7 32 0.88 4 2,742 15.7 33 0.7 0.75 371
15.2 32 0.74 0.3 289 15.7 33 0.7 0.3 165
15.2 32 0.74 0.6 475 Zhu et al. (2001) 15.4 36 0.9 0.44 578
15.2 32 0.74 1 633 15.4 36 0.9 1.75 1,556
15.2 32 0.74 1.4 893 15.4 36 0.9 4.4 3,036
15.2 32 0.74 2 1,118 15.4 36 0.9 7 4,226
15.2 32 0.74 3 1,995 Clark (1998) 15.04 39.3 0.95 0.5 642
15.2 32 0.74 4 1,673 15.04 38.5 0.95 1 1,026
Ueno et al. (1998) 15.7 33 0.88 0.9 1,255 15.04 36.8 0.95 5 3,053
15.7 33 0.88 2.1 2,167 15.04 36 0.95 10 4,843
15.7 33 0.88 3 2,953 Kutter et al. (1988) 20.6 36.5 0.94 1.91 2,134
15.7 33 0.88 4 2,808 20.6 36.5 0.94 0.96 1,440
15.2 33 0.74 0.9 721 Okamura et al. (1997) 9.7 33 0.88 3 1,293
15.2 33 0.74 2.1 1,374 9.7 33 0.88 1.5 703
15.2 33 0.74 3 2,044 9.7 33 0.88 2 937
Okamura et al. (1997) 9.7 34 0.88 1 1,020 9.7 33 0.88 3 1,273
9.7 34 0.88 2 1,831 White et al. (2008) 17.2 31 0.78 0.6 314
Zhu et al. (2001) 15.4 36 0.9 0.6 2,070 17.2 31 0.78 2.4 1,170
15.4 36 0.9 1.2 2,920 Bolton and Lau (1989) 16.5 37.5 0.84 5 3,430
15.4 36 0.9 1.8 3,830 Toyosawa et al. (2013) 15.7 36 0.95 0.5 900
15.4 36 0.9 2.4 5,060 Cerato and Lutenegger (2007) 16.2 38 0.7 0.61 640

© ASCE 04016069-4 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

 Table 3. Spearman Rank Correlation Diagnostics for h , MFELA, and M'
with Input Parameters

Parameter g D/pa DR tan f cv

M 0 0.9 0.03
h 0.82 0.9 0.44
M0 0.43 0.61 0.12
MFELA 0.91 0.8 0.59
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. Comparison between measured capacity and FELA-calculated

capacity: (a) strip footing; (b) circular footing

centrifuge tests (see Table 2) for circular footings are developed in

this paper. An evaluation of the model factor usually involved two
aspects:
1. Verify the randomness of the model factor. This can be accom-
plished using the MATLAB function corr for the Spearman
rank correlation diagnostics analysis with the input parameters.
2. Identify a reasonable probability model of the model factor.
This is accomplished using the MATLAB function kstest for the Fig. 4. Empirical probability distribution of MFELA
Kolmogorov-Smirnov goodness-of-fit test (KS test).
The comparison between the measured capacity (qu,m) and the
FELA-predicted capacity (qu,FELA) is illustrated in Figs. 3(a and b),
respectively, for strip and circular footings on dense sand, where 53
cases are compared. The differences between the FELA results and
the test results tend to be less than 15%, except for five outlying
cases at approximately 20%.
For circular footings, the p values of the Spearman rank correla-
tion diagnostics with respect to the input parameters (i.e., g D/pa,
tan f cv, and DR) are given in Table 3, which are higher than 0.05. It
indicates that MFELA can be modeled as a random variable directly.
The mean and standard deviation of MFELA are 1 and 0.1, respec-
tively. The p value of the KS test for the normality of lnMFELA is
0.12. This indicates that the lognormal distribution is a reasonable
model to describe MFELA. The empirical distribution of MFELA is
illustrated in Fig. 4. Based on the aforementioned results, it is rea- Fig. 5. Comparison between measured capacity and calculated
sonable to conclude that FELA can provide a relatively accurate capacity from Eurocode 7 and modified Eurocode 7
prediction of the bearing capacity.

Table 4. Ranges of Input Parameters Used for Numerical Simulations

Comparison of Load Tests and Eurocode 7 (CEN 2004)
Parameter Range of values Multicollinearity analysis (VIF)
To statistically characterize the model uncertainty, the following g (D) /pa
a
0.1–2 (0.5–20 m) 1.23
model factor M for Eurocode 7 (CEN 2004) is suggested: DR 0.7–0.95 1.12
tan( f cv)b 0.58–0.67 (30°–40°) 1.05
M ¼ qu;m =qu;c (5)
a
Range of foundation diameter shown in parentheses.
b
where qu,m = measured capacity; and qu,c = Eurocode 7–calculated Range of critical-state friction angle shown in parentheses.
capacity. From Eqs. (1a) and (1b), the calculation of qu,c depends
on the f value. However, the f value was not given in these tests. 7 (CEN 2004) approach underestimates the bearing capacity of cir-
More importantly, the uncertainty in the estimation of f will lead cular foundations on dense sand. This is because the effect of stress
to a significantly higher range of variation for Ng (Diaz-Segura level on the ultimate bearing capacity is not considered in Eurocode
2013). Therefore, the critical-state friction angle ( f cv) is used in 7 (CEN 2004). The p values of the Spearman rank correlation diag-
Eurocode 7 (CEN 2004) [i.e., Eqs. (1a) and (1b)] to calculate qu,c. nostics with respect to g D/pa, DR, and tan f cv given in Table 3 are
Some reported values of f cv for different sands can be found in almost equal to zero. This suggests a significant correlation of M
Table 1 of Bolton (1986), which can be used as a guideline for with the input parameters due to idealizations made to retain analyti-
selecting f cv. cal simplicity. Therefore, the model factor M for Eurocode 7 (CEN
For 26 cases in Table 2, the comparison between the measured 2004) cannot be modeled as a random variable directly.
capacity (qu,m) and the Eurocode 7 (CEN 2004)–calculated bearing Regression analysis is required to remove the dependency of the
capacity (qu,c) is given in Fig. 5, where q0u;c = fqu,c, and f is the regres- model factor M on the input parameters. However, the input param-
sion equation obtained next. It is clearly observed that the Eurocode eters (i.e., g D/pa, DR, and tan f cv) in Tables 1 and 2 are not varied

© ASCE 04016069-5 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

 systematically over their full practical ranges (see Table 4).
Because it has been proven in the previous section that FELA can
give a reasonable prediction of the bearing capacity of circular foun-
dations on dense sand, FELA is used to replace load tests in this crit-
ical regression step in the following section.

Comparison of FELA and Eurocode 7 (CEN 2004)

Regression Analysis
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

In this section, a set of orthogonal experiments are designed to

remove the dependency of the model factor of Eurocode 7 (CEN
2004) with respect to the input parameters. For this purpose, the
model factor Ms is defined as follows:

Ms ¼ qu;FELA =qu;c (6)

For onshore footings, the foundation diameter (D) is typically

between 0.5 and 3 m. For offshore foundations (e.g., spudcan foun-
dations), the foundation diameter D usually varies from 10 to 20 m.
Therefore, the foundation diameter D = 0.5–20 m in this paper. For
dense sand, the relative density is usually larger than 70%, and then Fig. 6. (a) Comparison of calculated capacity from Eurocode 7 with
DR = 0.7–0.95. Kulhawy and Mayne (1990) pointed out that the crit- FELA results; (b) comparison of calculated capacity from modified
ical-state friction angle ( f cv) for most natural sand deposits will Eurocode 7 approach with FELA results; (c) plot of residual part h
rarely be much above 30°–34°. However, to be consistent with the against FELA results; (d) empirical probability distribution of h
load-test database, f cv = 30°–40° is used here. In fact, the f cv
value may be up to 40° for feldspathic sand. Consequently, the rea-
sonable ranges for the dimensionless parameters g D/pa, DR, and Probability Model of Residual Part (g)
tan f cv are summarized in Table 4. For practical convenience, the model factor Ms can be expressed in
For regression analysis, 450 sets of the input parameters are the following multiplicative form:
designed. A multicollinearity diagnostic is carried out for the 450
input parameter sets according to the variance inflation factor Ms ¼ f h (8)
(VIF). This is performed using the MATLAB function colldiag. All
of the calculated values of VIF for the three dimensionless parame- where f = regression Eq. (7); and h = residual part of the model fac-
ters are slightly higher than 1, as shown in Table 4. This indicates tor Ms.
the orthogonality of these parameter sets. The residual part h in Eq. (8) is plotted against the FELA results
The bearing capacities calculated from Eurocode 7 (CEN 2004) in Fig. 6(c). It visually shows that h is randomly distributed with
for 450 cases are plotted against the FELA-calculated capacities in respect to qu,FELA. This is verified by the p values of the Spearman
Fig. 6(a). It can also be observed that Eurocode 7 (CEN 2004) rank correlation diagnostics given in Table 2, which are higher than
underestimates the capacity. Based on the results, the systematic 0.05. This indicates that h is independent of the input parameters
variation (f) of the model factor Ms can be expressed as (i.e., g D/pa, tan f cv, and DR) and can be modeled as a random vari-
able. The mean and standard deviation of h are 1.02 and 0.11,
f ¼ eb0 eb1 tan f cv eb2 DR eb3 g D=pa (7) respectively. In this case, the p value of the KS test for the normality
of ln h is 0.14. This indicates that the lognormal distribution is a rea-
The model coefficients {bi}i = 0,···,3 are determined using the sonable model to describe h . The empirical distribution of h is
MATLAB function regress to carry out multiple linear regression illustrated in Fig. 6(d).
analysis. The results are given in Table 5. The coefficient determina-
tion (R2) is 0.9. Fig. 6(b) plots the capacity q0u;c = fqu,c from the modi-
Model Uncertainty of Modified Eurocode 7 (CEN 2004)
fied Eurocode 7 (CEN 2004) against the FELA capacity qu,FELA. A
good agreement is observed., which indicates that the performance of
For the 26 cases in Table 2, the bearing capacity q0u;c from the modi-
Eurocode 7 (CEN 2004) can be improved by multiplying the regres- fied Eurocode 7 (CEN 2004) (i.e., q0u;c = fqu,c) is plotted against the
sion equation f. measured value qu,m in Fig. 5. Good agreement is shown between
The critical-state friction angle ( f cv) is an input parameter in q0u;c and qu,m. This implies the performance of Eurocode 7 can be
Eq. (7), which is usually determined by conventional triaxial tests. improved by multiplying the regression equation f, which reason-
For comparison with centrifuge tests, the f cv value in Tables 1 and ably accounts for the scale effect.
2 is adopted. In the following regression analysis, f cv = 30°–40° Similarly, the model factor M0 for the modified Eurocode 7
for various granular materials, which is consistent with the values in (CEN 2004) can be expressed as
Table 1 of Bolton (1986). In future applications, the selection of the
f cv value can also refer to Table 1 in Bolton (1986). These values M 0 ¼ qu;m =q0u;c ¼ qu;m =fqu;c (9)
are also subjected to measurement error, which is outside the scope
of this paper. However, such error will be reflected in the model where f is given in Eq. (7) with the coefficient values summarized in
uncertainty of the calculation model. Table 5.

© ASCE 04016069-6 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

 Table 5. Summary of Model Uncertainty of Some Capacity Calculation Models in Geotechnical Engineering

M0
Problem Variable lnf = b0 þ Rbixi Mean Coefficient of variation No. of tests
Strip footings on sand b0 0.28
under positive combined x1 g D/pa b1 –5.05 1.04 0.1 120
loading (Phoon and x2 j b2 11.4
Tang 2015a)a x3 tan f a b3 –0.26
x4 d/B b4 –0.09
x5 a/ f a b5 0.21
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

x6 e/B b6 –1.12
x7 (e/B)(a/ f a) b7 –0.98
Strip footings on sand b0 0.1 1.07 0.1 72
under negative combined x1 g D/pa b1 –4.5
loading (Phoon and x2 j b2 10.4
Tang 2015b)a x3 tan f a b3 –0.25
x4 d/B b4 –0.12
x5 a/ f a b5 –1.03
x6 e/B b6 –0.45
x7 (e/B)(a/ f a) b7 –1.81
Strip footings on sand b0 0.1 1.06 0.13 192
under general combined x1 g D/pa b1 –4.5
loading (Phoon and x2 j b2 10.25
Tang 2015b)a x3 tan f a b3 –0.15
x4 d/B b4 0.05
x5 a/ f a b5 –0.93
x6 e/B b6 –0.05
x7 (e/B)(a/ f a) b7 –2.53
Helical anchors in clay b0 0.75 0.95 0.11 78
under tension loading x1 n b1 –0.05
(Tang and Phoon 2016)b x2 S/D b2 –0.11
x3 H/D b3 –0.03
x4 g H/cu b4 –0.11
Circular footings on b0 1.97 1.02 0.15 26
dense sandc x1 tan f cv b1 –3.12
x2 DR b2 2.23
x3 g D/pa b3 –0.68
a
D = foundation diameter; d = embedment depth; e = load eccentricity; pa = atmospheric pressure, 100 kPa; a = load inclination; g = unit weight of sand;
j = empirical parameter, 0.02–0.12; f a = repose angle of sand.
b
cu = shear strength of clay; D = diameter of helix plate; H = embedment depth of top helix; n = number of helix plates; S = plate spacing; g = unit weight of
clay.
c
D = foundation diameter; DR = relative density of sand; g = unit weight of sand; f cv = critical-state friction angle of sand.

From a statistical viewpoint, Eq. (9) is preferred over Eq. (5). On

this basis, there are two ways to statistically characterize the model
factor M0 :
1. If the model factor M0 is independent of the input parameters
(i.e., g D/pa, tan f cv, and DR), the probability model, mean, and
standard deviation of M0 can be characterized directly using the
load-test database with size = 26, given in Table 2. The p values
of the Spearman rank correlation diagnostics with the input pa-
rameters shown in Table 3 are higher than 0.05. This suggests
the model factor M0 is no longer a function of g D/pa, tan f cv,
and DR and can be modeled as a random variable directly. The
KS test for the normality of lnM0 is 0.1, which indicates that M0
can be modeled as a lognormal random variable. The empirical
distribution of lnM0 is illustrated in Fig. 7. The mean and stand-
ard deviation of M0 are 1 and 0.11, respectively.
2. In contrast, the model factor M0 in Eq. (9) can also be written as
Fig. 7. Empirical probability distribution of M0 the product of two independent lognormal random variables ( h
and MFELA) according to Eqs. (4), (6), and (8), namely

© ASCE 04016069-7 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

 M0 ¼ h MFELA (10) CEN (European Committee for Standardization). (2004). EN 1997-1:
Eurocode 7: Geotechnical design—Part 1: General rules, Brussels,
Belgium.
which is also a lognormal random variable. Under this assumption,
Cerato, A. B., and Lutenegger, A. J. (2007). “Scale effect of shallow foun-
the mean and standard deviation of M0 are 1.02 and 0.15, respec- dation bearing capacity on granular material.” J. Geotech. Geoenviron.
tively, which are slightly larger than that directly characterized Eng., 10.1061/(ASCE)1090-0241(2007)133:10(1192), 1192–1202.
from the load-test database as shown in Table 2, where mean = 1 Chakraborty, M., and Kumar, J. (2015). “Lower-bound axisymmetric for-
and standard deviation = 0.11. mulation for geomechanics problems using nonlinear optimization.” Int.
Table 5 summarizes the results obtained recently using this J. Geomech., 10.1061/(ASCE)GM.1943-5622.0000454, 06014024.
framework. The standard deviation of M0 for footings on a homoge- Chen, W. F. (1975). Limit analysis and soil plasticity, Elsevier, Amsterdam,
neous soil layer is approximately 0.1. This can be explained by the The Netherlands.
Clark, J. I. (1998). “The settlement and bearing capacity of very large foun-
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

fact that these simple calculation models are based on a physically

dations on strong soils: 1996 R.M. Hardy keynote address.” Can.
correct model. Geotech. J., 35(1), 131–145.
In addition, Paikowsky (2010) thoroughly evaluated the model De Beer, E. E. (1965). “Bearing capacity and settlement of shallow founda-
factor for the current design practice estimating the bearing capacity tions on sand.” Proc., Symp. on Bearing Capacity and Settlement of
of shallow foundations under various loading conditions based on Foundations, Durham, NC, pp. 15–33.
an extensive database. In their work, the model factor was modeled Diaz-Segura, E. G. (2013). “Assessment of the range of variation of Ng
as a random variable directly without examining its dependence on from 60 estimation methods for footings on sand.” Can. Geotech. J.,
the input parameters. Under vertical loading, the standard deviation 50(7), 793–800.
of the model factor is approximately 0.46, which is much larger Dithinde, M., Phoon, K. K., De Wet, D., and Retief, J. V. (2011).
“Characterization of model uncertainty in the static pile design for-
than the 0.1 obtained in the current work. This is because regression mula.” J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)GT.1943-5606
analysis is performed to express the model factor as a function of .0000401, 70–85.
the input parameters. Consequently, the standard deviation of the Drucker, D. C., Greenberg, W., and Prager, W. (1951). “The safety factor of
model factor reduces from 0.33 to 0.1. an elastic plastic body in plane strain.” Trans. ASME J. Appl. Mech., 73,
371–378.
Drucker, D. C., Prager, W., and Greenberg, H. J. (1952). “Extended limit
Conclusions design theorems for continuous media.” Q. J. Appl. Math., 9(4), 381–389.
ISO (International Organization for Standardization). (2012). ISO 19905-1:
In this paper, the model uncertainty of Eurocode 7 (CEN 2004) for Petroleum and natural gas industries-site specific assessment of mobile
calculating the bearing capacity of circular foundations on dense offshore units—Part 1: Jack-ups, Geneva, Switzerland.
sand is investigated using the framework proposed by Phoon and Jamiolkowski, M. B., Lo Presti, D. C. F., and Manassero, M. (2003).
Tang (2015a). Consequently, the following conclusions can be made: “Evaluation of relative density and shear strength of sand from cone
penetration test (CPT) and flat dilatometer (DMT).” Soil behavior and
1. The foundation shape has an insignificant effect on the model
soft ground construction, J. T. Germaine, T. C. Sheahan, and R. V.
uncertainty, because the FELA calculated value of s g is very Whitman, ed., ASCE, Reston, VA, 201–238.
close to the suggested value by Eurocode 7 (CEN 2004). Kulhawy, F. H., and Mayne, P. W. (1990). “Manual on estimating the soil
2. The systematic variation f of the model factor M = qu,m/qu,c for properties on foundation design.” EL-6800, Research project 1493-6,
Eurocode 7 can be expressed as an exponential function of Final Rep., Electric Power Research Institute, Palo Alto, CA.
three dimensionless parameters: tan f cv, g D/pa, and DR. Kumar, J., and Chakraborty, D. (2013). “Linearization of Drucker-Prager
3. In contrast to M, a new model factor defined as M0 = qu,m/fqu,c yield criterion for axisymmetric problems: Implementation in lower-
is found to be independent of tan f cv, g D/pa, and DR. Using bound limit analysis.” Int. J. Geomech., 10.1061/(ASCE)GM.1943
the load-test database of size = 26, M0 can be characterized as -5622.0000200, 153–161.
Kusakabe, O., Yamaguchi, H., and Morikage, A. (1991). “Experiment and
a lognormal random variable with mean = 1 and standard analysis on the scale effect of Ng for circular and rectangular footings.”
deviation = 0.11. Proc., Int. Conf., Centrifuge 91, A. A. Balkema, Rotterdam, The
Netherlands, 179–186.
Kutter, B. L., Abghari, A., and Cheney, J. A. (1988). “Strength parameters
References for bearing capacity of sand.” J. Geotech. Engrg., 10.1061/(ASCE)0733
-9410(1988)114:4(491), 491–498.
Ang, A. H. S., and Tang, W. H. (1984). Probability concepts in engineering Lacasse, S., and Nadim, F. (1994). “Reliability issues and future challenges
planning and design. II: Decision, risk and reliability, John Wiley & in geotechnical engineering for offshore structures.” Proc., 7th Int. Conf.
Sons, New York. on Behavior of Offshore Structures, MIT Press, Cambridge, MA, 9–38.
Azzam, W. R., and ElWakil, A. Z. (2015). “Experimental and numerical Lee, J. H., and Salgado, R. (2005). “Estimation of bearing capacity of
studies of circular footing resting on confined granular subgrade adja- circular footings on sands based on cone penetration test.” J.
cent to slope.” Int. J. Geomech., 10.1061/(ASCE)GM.1943-5622 Geotech. Geoenviron. Eng., 10.1061/(ASCE)1090-0241(2005)131:
.0000500, 04015028. 4(442), 442–452.
Bolton, M. D. (1986). “The strength and dilatancy of sands.” Geotechnique, Loukidis, D., and Salgado, R. (2011). “Effect of relative density and stress
36(1), 65–78. level on the bearing capacity of footings on sand.” Geotechnique, 61(2),
Bolton, M. D., and Lau, C. K. (1989). “Scale effect in the bearing capacity 107–119.
of granular soils.” Proc., 12th Int. Conf. on Soil Mechanics and Lyamin, A., Salgado, R., Sloan, S. W., and Prezzi, M. (2007). “Two- and
Foundation Engineering, Vol. 2, A. A. Balkema, Rotterdam, The three-dimensional bearing capacity of footings in sand.” Geotechnique,
Netherlands, 895–898. 57(8), 647–662.
Briaud, J. L., and Tucker, L. M. (1988). “Measured and predicted axial Makrodimopoulos, A., and Martin, C. M. (2006). “Lower bound limit anal-
response of 98 piles.” J. Geotech. Engrg, 10.1061/(ASCE)0733 ysis of cohesive-frictional materials using second-order cone program-
-9410(1988)114:9(984), 984–1001. ming.” Int. J. Numer. Methods Eng., 66(4), 604–634.
Burlon, S., Frank, R., Baguelin, F., Habert, J., and Legrand, S. (2014). MATLAB 7.10.0 [Computer software]. MathWorks, Natick, MA.
“Model factor for the bearing capacity of piles from pressuremeter test Meyerhof, G. G. (1950). “The bearing capacity of sand.” PhD thesis, Univ.
results-Eurocode 7 approach.” Geotechnique, 64(7), 513–525. of London, London.

© ASCE 04016069-8 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069

 MOSEK [Computer software]. MOSEK ApS, Copenhagen, Denmark. Sivakugan, N., and Johnson, K. (2004). “Settlement predictions in granular
Muganga, R. T. (2008). “Uncertainty evaluation of displacement and soils: A probabilistic approach.” Geotechnique, 54(7), 499–502.
capacity of shallow foundations on rock.” Master's thesis, Univ. of Sloan, S. W. (2013). “Geotechnical stability analysis.” Geotechnique,
Massachusetts, Lowell, MA. 63(7), 531–572.
Okahara, M., Takagi, S., Obata, H., Mori, K., and Tatsuta, M. (1988). Tang, C., and Phoon, K. K. (2016). “Model uncertainty of cylindrical shear
“Centrifuge tests on scale effect of bearing capacity.” Proc., 42nd Japan method for calculating the uplift capacity of helical anchors in clay.”
Annual Conference on Civil Engineering, Vol. III, 250–251. Eng. Geol., 207, 14–23.
Okamura, M., Takemura, J., and Kimura, T. (1997). “Centrifuge model tests Tang, C., Toh, K. C., and Phoon, K. K. (2014). “Axisymmetric lower-
on bearing capacity and deformation of sand layer overlying clay.” Soils bound limit analysis using finite elements and second-order cone
Found., 37(1), 73–88. programming.” J. Eng. Mech., 10.1061/(ASCE)EM.1943-7889
Paikowsky, S. G. (2002). “Load and resistance factor design (LRFD) for .0000669, 268–278.
deep foundations.” NCHRP Rep. 24-17, Transportation Research Tang, W. H., and Gilbert, R. B. (1993). “Case study of offshore pile system
Downloaded from [Link] by Indian Institute of Technology Bhubaneswar on 06/15/20. Copyright ASCE. For personal use only; all rights reserved.

Board, Washington, DC. reliability.” Proc., 25th Offshore Technology Conf., Society of
Paikowsky, S. G. (2010). “LRFD design and construction of shallow founda- Petroleum Engineers, Houston, TX, 677–686.
tions for highway bridge structures.” NCHRP Rep. 651, Transportation Toyosawa, Y., Itoh, K., Kikkawa, N., Yang, J. J., and Liu, F. (2013).
Research Board, Washington, DC. “Influence of model footing diameter and embedded depth on particle
Perkins, S. W., and Madson, C. R. (2000). “Bearing capacity of shallow foun- size effect in centrifugal bearing capacity tests.” Soils Found., 53(2),
dations on sand: A relative density approach.” J. Geotech. Geoenviron. 349–356.
Eng., 10.1061/(ASCE)1090-0241(2000)126:6(521), 521–530. Ueno, K., Miura, K., and Maeda, Y. (1998). “Prediction of ultimate bearing
Phoon, K. K., Chen, J. R., and Kulhawy, F. H. (2006). “Characterization of capacity of surface footings with regard to size effect.” Soils Found.,
model uncertainties for augured cast-in-place (ACIP) piles under axial 38(3), 165–178.
compression.” Foundation analysis and design: Innovative methods Ueno, K., Nakatomi, T., Mito, K., and Kusakabe, O. (1994). “Initial condi-
(GSP 153), R. L. Parsons, L. M. Zhang, W. D. Guo, K. K. Phoon, and tions and their influences on bearing characteristics of sand.” Proc. Int.
M. Yang, eds., ASCE, Reston, VA, 82–89. Conf. Centrifuge, 94, 541–546.
Phoon, K. K., Chen, J. R., and Kulhawy, F. H. (2007). “Probabilistic hyper- Uzielli, M., and Mayne, P. W. (2012). “Load-displacement uncertainty of
bolic models for foundation uplift movement.” Probabilistic applica- vertically loaded shallow footings on sands and effects on probabilistic
tions in geotechnical engineering (GSP 170) (CD-ROM), K. K. Phoon, settlement estimation.” Georisk: Assess. Manage. Risk Eng. Syst.
et al., eds., ASCE, Reston, VA. Geohazards, 6(1), 50–69.
Phoon, K. K., and Kulhawy, F. H. (2005). “Characterisation of model White, D. J., Teh, K. L., Leung, C. F., and Chow, Y. K. (2008). “A compari-
uncertainties for laterally loaded rigid drilled shafts.” Geotechnique, son of the bearing capacity of flat and conical circular foundations on
55(1), 45–54. sand.” Geotechnique, 58(10), 781–792.
Phoon, K. K., and Tang, C. (2015a). “Model uncertainty for the capacity of Zhang, D. M., Phoon, K. K., Huang, H. W., and Hu, Q. F. (2015).
strip footings under combined loading.” Geotechnical Special “Characterization of model uncertainty for cantilever deflections in
Publication in honor of Wilson H. Tang, ASCE, Reston, VA. undrained clay.” J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)GT
Phoon, K. K., and Tang, C. (2015b). “Model uncertainty for the capacity of .1943-5606.0001205, 04014088.
strip footings under negative and general combined loading.” 12th Int. Zhang, J., Zhang, L. M., and Tang, W. H. (2009). “Bayesian framework for
Conf. on Applications of Statistics and Probability in Civil Engineering, characterizing geotechnical model uncertainty.” J. Geotech.
ICASP 12, Vancouver, Canada, July 12-15. Geoenviron. Eng., 10.1061/(ASCE)GT.1943-5606.0000018, 932–940.
Randolph, M. F., Jamiolkowski, M. B., and Zdravkovic, L. (2004). “Load Zhu, F., Clark, J. I., and Phillips, R. (2001). “Scale effect of strip and circu-
carrying capacity of foundations.” Proc., Skempton Memorial Conf., lar footings resting on dense sand.” J. Geotech. Geoenviron. Eng., 10
Vol. 1, Telford, London, 207–240. .1061/(ASCE)1090-0241(2001)127:7(613), 613–621.

© ASCE 04016069-9 Int. J. Geomech.

Int. J. Geomech., 2017, 17(3): 04016069