Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

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

net/publication/327793216

Random finite element method for bearing capacity assessment of a shallow


foundation under varied uniaxial loadings

Article  in  International Journal of Reliability and Safety · January 2018


DOI: 10.1504/IJRS.2018.10016359

CITATIONS READS

0 376

4 authors:

Amal Hentati Mbarka Selmi


Ecole Polytechnique de Tunisie National Enginnering School of Gabes, Tunisia
6 PUBLICATIONS   16 CITATIONS    10 PUBLICATIONS   54 CITATIONS   

SEE PROFILE SEE PROFILE

Tarek Kormi Nizar Bel Hadj Ali


Ecole Polytechnique de Tunisie National Engineering School of Gabes
23 PUBLICATIONS   110 CITATIONS    59 PUBLICATIONS   732 CITATIONS   

SEE PROFILE SEE PROFILE

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

Fugitive methane estimation in Landfills View project

Modeling methane oxidation in the meditteranean basin View project

All content following this page was uploaded by Amal Hentati on 21 September 2018.

The user has requested enhancement of the downloaded file.


Int. J. Reliability and Safety, Vol. 12, No. 3, 2018 237

Random finite element method for bearing capacity


assessment of a shallow foundation under varied
uniaxial loadings

Amal Hentati*
Systems and Applied Mechanics Laboratory (LASMAP),
Polytechnic School of Tunisia,
El-Khawarizmi Street,
La Marsa, 2078, Tunisia
Email: amal.hentati@yahoo.fr
*Corresponding author

Mbarka Selmi
Civil Engineering Laboratory (L.G.C.),
National Engineering School of Tunis,
ENIT, University of Tunis EL-Manar,
LGC-ENIT, BP 37,
1002 Tunis Le Belvédère, Tunisia
Email: selmi.mbarka@yahoo.fr

Tarek Kormi and Nizar Bel Hadj Ali


Systems and Applied Mechanics Laboratory (LASMAP),
Polytechnic School of Tunisia,
El-Khawarizmi Street,
La Marsa, 2078, Tunisia
Email: tarekkormi@gmail.com
Email: nizar.belhadjali@enig.rnu.tn

Abstract: This paper focuses on the application of the random finite element
method (RFEM) for the assessment of the uniaxial bearing capacities of a
shallow foundation subjected to centred vertical, horizontal and rotational
loadings. The analysis combines finite element modelling, spatial variability
analysis and Monte Carlo simulation. For this, the soil undrained shear strength
is assumed to be variable in both horizontal and vertical directions with
spatial dependency expressed via a Markovian autocorrelation function. The
application of the proposed methodology to a shallow foundation permitted to
highlight the insufficiency of the deterministic approach to predict the uniaxial
foundation bearing capacities and led to different failure mechanisms.

Keywords: random finite element method; random fields; spatial variability;


shallow foundation reliability.

Reference to this paper should be made as follows: Hentati, A., Selmi, M.,
Kormi, T. and Ali, N.B.H. (2018) ‘Random finite element method for bearing
capacity assessment of a shallow foundation under varied uniaxial loadings’,
Int. J. Reliability and Safety, Vol. 12, No. 3, pp.237–260.

Copyright © 2018 Inderscience Enterprises Ltd.


238 A. Hentati et al.

Biographical notes: Amal Hentati is a PhD student in Systems and Applied


Mechanics Laboratory (LASMAP) of Polytechnic school of Tunisia. She
received her engineering diploma in 2012 and her master’s degree in 2013
from the National Engineering School of Tunis (ENIT).

Mbarka Selmi is an Assistant Professor of Civil Engineering at the National


Engineering School of Gabes (ENIG, University of Gabes) and a researcher at
the Civil Engineering Laboratory (L.G.C.-ENIT, University of Tunis EL-
Manar). She obtained her PhD degree from the National Engineering School of
Tunis in 2011. Her research interests include numerical soil modelling,
reliability analyses and geotechnical risk assessment, especially slopes stability
and foundations.

Tarek Kormi is an Assistant Professor in National Engineering School of


Gabes, University of Gabes since 2005. He received his PhD degree in Civil
Soil Mechanics, Ecole Nationale des Ponts et Chaussées, Paris in 2003, his MS
degree in Soil Mechanics, Ecole Centrale Paris, Paris in 1999 and his Graduate
Polytechnic Engineer from Ecole Polytechnique de Tunisie, Tunis in 1998.
His research interests include numerical modelling of landfills emissions,
unsaturated soil behaviour, pile and shallow foundations capacities

Nizar Bel Hadj Ali is an Associate Professor in Civil Engineering at the


National Engineering School of Gabes (ENIG, University of Gabes) and a
researcher in the Applied Mechanics and Systems Research Laboratory in the
Ecole Polytechnique de Tunisie (EPT, Carthage University). His research work
includes numerous projects in soft computing, structural health monitoring,
deployable and active structures. He is the author of many research articles and
attended dozen of international conferences.

1 Introduction

Predicting bearing capacity of shallow foundations is one of the most serious problems in
geotechnical engineering. Initially most works focused on the evaluation of the vertical
foundations capacity as this is the predominant loading due to the infrastructure self-
weight (e.g. Skempton, 1951). More recently, with the development of the offshore oil
and gas industry, a significant attention has been assigned to the capacity of foundations,
subjected to vertical (V), horizontal (H) and moment (M) loadings frequently due to
environmental factors likely to occur in deep water such as wind and waves. In this aim,
a deterministic bearing capacity assessment is, traditionally, derived using a global safety
factor based on constant soil properties values chosen by implementing local experience
and engineering judgment. However, it is well known in geotechnical field that soils are
highly variable in their properties and rarely homogeneous by nature because of different
depositional and post-depositional processes to which they were submitted during their
composition histories (Lacasse and Nadim, 1996). These variations, mainly attributed to
the inherent spatial soil variability, make the identification of the soils’ properties
incomplete and uncertain (Cherubini et al., 1993; Phoon and Kulhawy, 1999).
Consequently, the resulting safety factors are not consistent measures of risk evaluation
since foundations with the same safety factor may exhibit different risk levels depending
on the variability of the soil properties (Li and Lumb, 1987; Duncan, 2000). A statistical
description of these latter and a probability calculation constitute, hence, extremely
Random finite element method for bearing capacity assessment 239

precious tools to quantify the soil uncertainties and to incorporate them into foundations
designs and analyses (Lumb, 1974; Cherubini, 1997; Phoon and Kulhawy, 1999; Ang
and Tang, 2007; Uzielli et al., 2007; Fenton and Griffiths, 2008). In this context, a
probabilistic methodology focusing on soil variations is proposed in this paper to assess
the uncertainties effect on the uniaxial bearing capacities of shallow foundations resting on
deposits under undrained conditions. Since the bearing capacities, in such conditions,
depend primarily on the undrained shear strength of the soil, Su, only the spatial variability
of this latter is considered using 2D stationary random fields. A random finite element
method is established, in this study, using Matlab and the finite element software
Abaqus. The Monte Carlo simulation method is then used to characterise the stochastic
responses of the foundation under vertical, horizontal and rotational loadings separately.
The effect of soil uncertainties, mainly the undrained shear strength coefficient of
variation and scale of fluctuation, on the uniaxial bearing capacities and reliability
measures of the shallow foundation was investigated.
Based on the provided assessment findings, the importance of the probabilistic
foundation analyses with an appropriate statistical soil characterisation and the insufficiency
of the conventional approach to predict the foundation failure are highlighted.

2 Random fields modelling

The soil is a material resulting from a combination of geological, environmental, physical


and chemical processes. Most of these processes are continuous and cause local changes,
which result in spatial variability of soil properties. These spatial variations are not a
random process; rather they are controlled by location in space. To characterise spatial
soil properties pattern, geostatistical techniques are often used. Particularly, the random
field theory is widely used in geotechnical discipline to integrate the soil spatial
variations into geotechnical structures designs and assessments (Griffiths and Fenton,
2001; Griffiths et al., 2002; Popescu et al., 2005; Massih et al., 2008; Selmi, 2012).
Indeed, it assumes the soil parameter, at any location within a soil layer, to be an
uncertain quantity, or a random variable characterised by a probability distribution and
correlated with random variables at adjacent locations. The set of random variables at all
locations within the layer is referred to as random field which is characterised by the joint
probability distribution of all random variables and an autocorrelation distance (or its
associated scale of fluctuation, Vanmarcke, 1977) representing the distance under which
the spatially random values tend to be significantly correlated.
In the present study, the uniaxial undrained bearing capacities of a shallow
foundation are assessed. The variability of the undrained shear strength of the soil
beneath the foundation is assumed to be characterised by a lognormal distribution.
Random fields are generated using the Local Averaging Subdivision (LAS) algorithm
(Fenton and Griffiths, 2008) and completely defined by their mean, Su , variance,  Su 2
,
and autocorrelation function,  ( x, x ). To quantify the magnitude of uncertain fluctuations
of the considered random parameter around its mean value, the dimensionless coefficient
of variation ‘COV is used as:
S
COVSu  u
(1)
S u
240 A. Hentati et al.

Among autocorrelation functions commonly used in geotechnical engineering (Vanmarcke,


1983; Li and Lumb, 1987; Rackwitz, 2000; Griffiths and Fenton, 2009), a square
exponential autocorrelation (Markovian) function is used in the current study as follows:

  2 2

 2 x   2 z  
 ( x, y )  exp      (2)
  h   v  
 
where  x  | x2  x1 | and  z  | z2  z1 | are two components of the absolute distance
between two points in 2D space where the correlation function is calculated.  h and  v
are the horizontal and vertical scales of fluctuation (i.e. correlation lengths) respectively.
Since the foundation performance is generally controlled by the averaged soil
strength over spatial local domains, a Local Averaging Subdivision (LAS) was applied in
this study to discretise the continuous fields. Thus, each discrete local average given by a
realisation becomes the average property within each discrete element (Vanmarcke,
1983; Griffiths and Fenton, 2009). This approach was motivated first by the need to
represent engineering properties as local averages (since many properties are not well
defined at a point and show significant scale effects), and second to be easily able to
incorporate known data or change resolution within sub-regions.
Moreover, this study covers the heterogeneity anisotropy of the undrained shear
strength random field with making a distinction between vertical and horizontal
fluctuation scales. Available studies strongly show that a much higher horizontal degree
of correlation exists in comparison to the vertical direction, due to the process of
deposition. Indeed, the values of the vertical fluctuation scale,  v , are generally between
0.5 and 2 m, whereas the corresponding values for horizontal fluctuation scale,  h , are
generally of the order of 10–30 m (Cherubini, 2000). All these considerations were
implemented in a Matlab code to ensure the simulation of all random data.
Figure 1 shows three realisations of the undrained shear strength random field, Su,
associated to a mean value Su  10kPa, a coefficient of variation COVSu = 0.3 and a
vertical fluctuation scale  v  2m for different horizontal scales of fluctuation equal to
2 m, 10 m and 40 m respectively. Compared to Figure 1a, Figure 1b clearly shows that,
contrary to vertical variations, the soil property does not scatter considerably in the
horizontal direction due to the higher correlation length ( h  10m), albeit still random
from one location to another.

Figure 1 Realisations of the random field, Su, for different horizontal scales of fluctuation

(a)  v  2m;  h  2m (b)  v  2m;  h  10m


Random finite element method for bearing capacity assessment 241

Figure 1 Realisations of the random field, Su, for different horizontal scales of fluctuation
(continued)

(c)  v  2m;  h  40m

This result is more noticeable when considering higher horizontal fluctuation scale
 h  40m as illustrated in Figure 1c where the soil is near to be formed by horizontal layers
of stronger and weaker soil.

3 Finite element modelling

The FE analysis model used in the present paper is illustrated in Figure 2 where a two-
dimensional plane strain condition is assumed using the commercial Abaqus finite element
package (version 6.13, Dassault Systems). A shallow foundation with a width B = 1m,
subjected to a centred vertical loading (V), a horizontal loading (H) and a rotational
loading (M) is considered. The foundation is resting on the surface of the soil which is
assumed to be undrained and obeying a linear elastic-perfectly plastic constitutive law.
The foundation is modelled as a rigid body with a reference point at the centre. A fully
bonded foundation/soil interface is therefore appropriate. The elastic response is defined
by the Young’s modulus of the soil E = 500 Su. The Poisson’s ratio is set as 0.49 to model
the undrained conditions of no volume change as well as to ensure numerical stability.
The soil material failure is defined by the Tresca criterion, with the maximum shear
stress in any plane limited to the undrained shear strength Su. The soil domain is of width
5B and height 3B, large enough to ensure no obvious boundary effects. As shown in
Figure 2, displacement boundary conditions of roller supports were provided along the
vertical boundaries and along the base of the mesh while the top was left free in order to
simulate soil surface.
The soil domain was divided first into 1500 bidimensional elements in total. In each
element the soil properties were assumed to be constant and defined by the undrained
shear strength Su. However, these properties changed from element to element representing
the soil spatial variability in both horizontal and vertical directions. In other words, soil
domain under Abaqus environment is divided so as to agree well with the stochastic mesh.
In this analysis, the FE mesh consists of 8-node biquadratic plane strain elements. A
more refined mesh was used below and in the vicinity of the footing to improve the
numerical accuracy of the solution. Since the finite element size is an important
parameter which can affect the accuracy of reliability results when the spatial variability
242 A. Hentati et al.

is taken into account, it is carefully chosen so as to conserve the distribution and the
initial autocorrelation structure of stochastic field. As suggested by Griffiths and Fenton
(2004) and Ching and Phoon (2013), the ratio of the element size to the scale of
fluctuation has to be sufficiently small to ensure that the variance function approaches
1.0. The maximum width of one mesh is limited, hence, to the half of the autocorrelation
distance (Der Kiureghian and Ke, 1988). Owing to the large number of simulations that
must be completed, a balance must be found between calculation time and solution
accuracy with regards to mesh discretisation (Figure 2).

Figure 2 Finite element meshing into Abaqus

3.1 Sign convention and nomenclature


The sign convention for displacements and loads presented in this paper obeys a right-
handed axes and clockwise positive convention as proposed by Butterfield et al. (1997),
as shown in Figure 3.

Figure 3 Load and displacement convention adopted

The notation used in this paper is defined in Table 1. The V, H and M loads as well as the
corresponding footing movements w, u and  are defined in Figure 3. The foundation’s
ultimate uniaxial bearing capacities, corresponding to pure loading (for example, for
ultimate pure vertical load capacity H = M = 0), are denoted as V0, H0 and M0 for pure
Random finite element method for bearing capacity assessment 243

vertical, horizontal and moment loadings respectively (with further subscript det or ran
used to identify values derived from finite element analysis of a homogeneous soil or
from randomly generated and spatially varying soil, respectively).

4 Reliability analysis

There is a clear need for optimised foundation design tools that would enable more
detailed assessment of the risk of every installation. One such tool is likely to be the
systematic use of reliability measures instead of safety factors.
Among different definitions of the performance function, G, existing in the literature,
the definition adopted in this study is given as follows (Murff, 1994):
2 2 2
V  H  M  (3)
G   0    0    0  1
V   H   M 

where V0, H0 and M0 are respectively the vertical, horizontal and moment ultimate
bearing capacities.
V, H and M are respectively the vertical, horizontal and moment loadings.
To evaluate an estimate of the probability of failure Pf, the expected value and
standard deviation of the performance function, G, need to be determined. For this
purpose, the Monte-Carlo Simulation method is here used. The Cornell reliability index
is then calculated as follows (Cornell, 1969):
G
c  (4)
G

where G and  G are the expected value and the standard deviation of the performance
function G.
Using the cumulative distribution function of the standard normal distribution 
evaluated at –c, an estimate of the probability of failure, Pf, can be determined as follows:
Pf   (   c ) (5)

Conventionally, for geotechnical problems, the factor of safety typically ranges between
1.0 and 3.0. In a reliability context, the historical failure probabilities for engineering
conveniences are recognised to be between 10–2 and 10–4. Therefore, target failure
probabilities for new designs are typically within this range (Griffiths and Fenton, 2007)
which are associated to reliability index values lying in the range of 3–4 for a good
performance of the system (US Army Corps of Engineers, 1997; Baecher and Christian,
2003).
Table 1 Summary of notations adopted in this paper

Vertical Horizontal Moment


Load V H M
Displacement w u 
Uniaxial bearing capacity
V0,det H0,det M0,det
(deterministic)
Uniaxial bearing capacity
V0,ran H0,ran M0,ran
(random)
244 A. Hentati et al.

Table 1 Summary of notations adopted in this paper (continued)

Vertical Horizontal Moment


Bearing capacity factor N cV  V0 /BSu N cH  H 0 /BSu N cM  M 0 /B 2 Su
Normalised load v  Vran / V0,det h  H ran / H 0,det m  M ran /M 0,det
Normalised uniaxial bearing
vo  V0,ran / V0,det ho  H 0. ran /H 0,det mo  M 0,ran /M 0,det
capacity

5 Results and discussion

5.1 Uniaxial deterministic bearing capacities


A deterministic case was first modelled with a uniform undrained shear strength Su =10
kPa. The bearing capacity factors under pure uniaxial loading conditions (NcV, NcH, NcM)
are listed in Table 2 where they are also compared to available finite element and
analytical results. Figure 4, Figure 5 and Figure 6 make clear the convergence of
the bearing capacity factors for different loading cases. As coherent with other finite
element studies, the results with 2727 elements are slightly larger than the analytical or
exact solution.
The largest difference is noticed mainly for the uniaxial vertical bearing capacity
NcV = 5.32, which is over predicting the theoretical value of 5.14. This overestimation of
3.5% is assumed to be due to the relatively coarse mesh used in this study in order to
limit the computational time of each numerical simulation.
Although a finer mesh (FE with 3640 elements) can produce better estimation of
bearing capacity factors, the resulting computational time increases dramatically.
Consequently, since this study requires thousands of individual simulations within the
Monte Carlo framework and since exact numerical prediction of bearing capacity factors
is not the main goal, rather their relative variations due to soil heterogeneity, balance
must be found between calculation time and solution accuracy with regards to mesh
discretisation. To strike this balance, the coarser mesh with 2727 elements, refined below
and in the vicinity of the foundation, was viewed as acceptable (Figure 2) to determine
satisfying ultimate loads values.

Table 2 Summary of bearing capacities derived in deterministic soil conditions

FE FE (Gourvenec
(Cassidy et al.,
with 2727 with 3640 and Randolph, Analytical
2013)
elements elements 2003)
Vertical NcV 5.32 5.29 5.28 5.2 2 a
Horizontal NcH 1.03 1.003 1.1 1.02 1b
Rotational NcM 0.708 0.701 0.75 0.72 0.69c
a
Notes: Exact solution from Prandtl’s mechanism (Prandtl, 1921).
b
Based on the assumption that failure occurs as sliding along the foundation-
soil interface with a shear strength Su
c
Derived from upper bound theory.
Random finite element method for bearing capacity assessment 245

Figure 4 Typical load/deformation curve in the vertical bearing capacity analysis

Displacement w (m)

Figure 5 Typical load/deformation curve in the horizontal bearing capacity analysis

Displacement u (m)

Figure 6 Typical moment/rotation curve in the turning moment bearing capacity analysis

rotation  (rad)
246 A. Hentati et al.

5.2 Stochastic analysis


The present stochastic study is carried out to evaluate the impact of the undrained shear
strength variability on the resulting vertical, horizontal and rotational bearing capacities
of the considered footing. For this, we were mainly interested in the effects of the
coefficient of variation and the heterogeneity anisotropy of the soil undrained shear
strength on the foundation reliability.
In this aim, Su is statistically characterised by a lognormal distribution with a constant
mean value, Su , equal to 10 kPa and a coefficient of variation, COVSu , varying
systematically according to 0.1, 0.3 and 0.5 (resulting respectively in standard variation
values  Su  1, 3 and 5 kPa) which are typical values of the undrained shear strength
reported in the geotechnical literature (Lee et al., 1983; Duncan, 2000; Hicks and Samy,
2002).
A first case, assuming an isotropic random shear strength is considered i.e. the
horizontal and vertical fluctuation scales of Su were maintained constant and equal to
 v   h  2 m throughout all analyses. To show the random field anisotropy effect on the
footing reliability, a second case with different vertical and horizontal fluctuation scales
respectively equal to  v  2m and  h  10m, is then studied. For each of the three values
of COVSu considered in this study, 500 realisations of the Su random field were generated
to achieve a clear numerical convergence of bearing capacity values for both isotropic
and anisotropic cases.

5.2.1 Uniaxial stochastic bearing capacities


Figures 7 to 12 show the evolution of the mean value and COV of the vertical, horizontal
and rotational normalised uniaxial bearing capacities respectively denoted v0, h0, m0
throughout the Monte Carlo simulations respectively for both isotropic and
anisotropic cases. It can be seen that both statistics converge to an approximately
constant value so it is reasonable to assume that the response PDF has been well-captured
by the Monte Carlo simulations.

Figure 7 (a) Normalised mean value of vertical capacity vo and (b) COV of the normalised
vertical capacities vo (isotropic case)
Random finite element method for bearing capacity assessment 247

Figure 7 (a) Normalised mean value of vertical capacity vo and (b) COV of the normalised
vertical capacities vo (isotropic case) (continued)

Figure 8 (a) Normalised mean value of horizontal capacity ho and (b) COV of normalised
horizontal capacity ho (isotropic case)
248 A. Hentati et al.

Figure 9 (a) Normalised mean value of rotational capacity mo and (b) COV of normalised
rotational capacity mo (isotropic case)

Figure 10 (a) Normalised mean value of vertical capacity vo and (b) COV of normalised vertical
capacities vo (anisotropic case)

(a)
Random finite element method for bearing capacity assessment 249

Figure 10 (a) Normalised mean value of vertical capacity vo and (b) COV of normalised vertical
capacities vo (anisotropic case) (continued)

(b)

Figure 11 (a) Normalised mean value of horizontal capacity ho and (b) COV of normalised
horizontal capacity ho (anisotropic case)

(a)

(b)
250 A. Hentati et al.

Figure 12 (c) Normalised mean value of rotational capacity mo and (d) COV of normalised
rotational capacity mo (anisotropic case)

(a)

(b)

All these obtained results are summarised in Figure 13 in addition to the results of
(Griffiths et al., 2002) for the vertical bearing capacity. For different stochastic soil cases,
visual observations reveal that the mean value of v0, h0, m0 are less than 1, implying that
spatial variation reduces the mean uniaxial bearing capacity compared to the
deterministic case. This is consistent with the pure vertical bearing capacity results of
(Nobahar and Popescu, 2000; Griffiths et al., 2002). Also, a good agreement is found
here with the different uniaxial bearing capacity results of (Cassidy et al., 2013; Charlton
and Rouainia, 2016).
By observing the evolution of the normalised means and COV of the uniaxial bearing
capacities, a check can be made to make sure that both statistics are sensitive to the soil
properties variations. Obviously, any increase in spatial variability (COVSu from 0.1 to
0.5) results in a reduction of the mean value and an increase of the COV of each uniaxial
bearing capacity.
Another important conclusion is that the difference between deterministic and
stochastic capacities is greater under vertical loading.
This difference in terms of capacity may be due to the different failure mechanisms
for the three loading cases. Figure 14 shows the failure mechanisms for vertical,
horizontal and moment loadings for the deterministic case and two stochastic cases.
Random finite element method for bearing capacity assessment 251

It can be seen from Figure 14 that, for pure vertical loading, spatial variability near
the footing has triggered an asymmetric failure mechanism that is typically at a lower
bearing load than predicted by traditional homogeneous and symmetric failure analysis.
In contrast, under horizontal and moment loading, the failure mechanisms are slightly
different for both deterministic and stochastic cases with only minor variations in the
path taken by the shear plane.
The failure mechanism under vertical loading therefore appears to have a much
greater scope for variability compared to the other loading cases which justify the great
difference between deterministic and stochastic uniaxial capacities.
The most important conclusion derived from Figure 14 is that the inherent spatial
variability of the soil undrained shear strength is not only affecting the value of the
bearing capacity by making it a random variable that can be different from the
corresponding deterministic value, but it can also modify the basic form of the failure.
This change in the failure surface can lead to significant changes in the ultimate bearing
capacity. Otherwise, for the bearing capacity problem treated here, the consideration of
spatial variability of soil properties leads to totally different failure mechanism surfaces
from one realisation to another especially under vertical loading.
To show further the effect of the soil variability on the statistical characterisation of
the normalised uniaxial bearing capacities, histograms of these later are illustrated in
Figures 15, 16 and 17 for a COVSu of 0.3 for illustrative purposes.
Although it is stated that anisotropy does not have a significant effect on the mean
ultimate bearing capacity of shallow foundations laying on a spatially variable soil, it is
evident in Figures 15, 16 and 17 that it has an increasing effect on the mean bearing
capacities values and their corresponding standard deviations.
This is likely to be due to the fact that for the isotropic case with short horizontal
autocorrelation distance, capacity is dependent on a spatial average over the failure zone
due to the rapidly fluctuating areas of low and high strength soil. Conversely, for
anisotropic case with higher horizontal autocorrelation distance, the failure zone may
include large areas of strong or weak soil, as it can be clearly seen in Figure 1, resulting
in greater variability. This behaviour is expected when referring to Griffiths and Fenton
(2007) and Jamshidi Chenari and Mahigir (2014).

Figure 13 Mean value, and coefficient of variation, COV, of v0, h0, and m0 for isotropic and
anisotropic cases
252 A. Hentati et al.

Figure 13 Mean value, and coefficient of variation, COV, of v0, h0 and m0 for isotropic and
anisotropic cases (continued)

Figure 14 Displacement vectors at failure for (a) vertical, (b) horizontal and (c) moment loadings
Random finite element method for bearing capacity assessment 253

Figure15 Histograms of the normalised vertical capacities for the 500 simulations for isotropic
and anisotropic cases

Figure 16 Histograms of the normalised horizontal capacities for the 500 simulations for isotropic
and anisotropic cases
254 A. Hentati et al.

Figure 16 Histograms of the normalised horizontal capacities for the 500 simulations for isotropic
and anisotropic cases (continued)

Figure 17 Histograms of the normalised rotational capacities for the 500 simulations for isotropic
and anisotropic cases
Random finite element method for bearing capacity assessment 255

5.2.2 Reliability measures


This section proposes a reliability based analysis with a spatially variable shear strength
parameter Su. For this, equation (3) is used as a performance function.
All results are summarised in Tables 3 to 8 for a vertical scale of fluctuation
 v  2m . We notice first that for the same security level in a deterministic context
(i.e. the same safety factor, Fs), different levels of reliability may exist according to the
statistical characterisation of Su (the coefficients of variation and autocorrelation distance
values).
Another important conclusion is that, for suitable safety factors, reliability index βc
may take values below 3 and failure probability can reach values clearly greater than 10–3
which are not in the range of historical reliability measures for engineering conveniences
implying a bad performance of the foundation.
In other words, even if the foundation is considered safe in a deterministic context,
accounting for uncertainties could lead to an unsafe design with an important failure
probability.
It can be seen also that, unlike the failure probability, the reliability index decreases
with the increase of the applied loading V, H or M (i.e., with the decrease of the safety
factor FS) until it reaches its lowest values for the most comfortable factors of safety in
conventional analysis (3 for vertical loading and 2 for both horizontal and rotational
loadings).
Table 3 Reliability index, βc, for vertical loading

c
FS COVSu = 0.1 COVSu = 0.3 COVSu = 0.5
h = 2 m h = 10 m h = 2 m h = 10 m h = 2 m h = 10 m
1.5 5.40 3.45 1.88 1.16 1.21 0.71
2 8.10 4.65 2.82 1.53 1.82 0.9
2.5 9.72 5.20 3.39 1.71 2.19 0.997
3 10.80 5.50 3.76 1.8 2.43 1.04

Table 4 Probability of failure, Pf, for vertical loading

Pf(%)
FS COVSu = 0.1 COVSu = 0.3 COVSu = 0.5
h = 2 m h = 10 m h = 2 m h = 10 m h = 2 m h = 10 m
1.5 3.33E-06 2.80E-02 3.01 12.30 11.31 23.89
2 2.75E-14 1.66E-04 0.24 6.30 3.44 18.41
2.5 1.24E-20 9.96E-06 0.03 4.36 1.43 15.94
3 1.72E-25 1.90E-06 0.01 3.59 0.75 14.92
256 A. Hentati et al.

Table 5 Reliability index, c, for horizontal loading

c
FS COVSu = 0.1 COVSu = 0.3 COVSu = 0.5
h = 2 m h = 10 m h = 2 m h = 10 m h = 2 m h = 10 m
1.2 1.94 1.60 0.74 0.6 0.47 0.42
1.3 2.58 2.13 0.96 0.77 0.56 0.52
1.5 3.50 2.88 1.27 1.009 0.7 0.67
2 4.70 3.88 1.68 1.32 0.89 0.86

Table 6 Probability of failure, Pf, for horizontal loading

Pf(%)
FS COVSu = 0.1 COVSu = 0.3 COVSu = 0.5
h = 2 m h = 10 m h = 2 m h = 10 m h = 2 m h = 10 m
1.2 2.62 5.48 22.96 27.43 31.92 33.72
1.3 0.49 1.66 16.85 22.06 28.77 30.15
1.5 0.02 0.20 10.20 15.65 24.20 25.14
2 1.3E-04 0.01 4.65 9.34 18.67 19.49

Table 7 Reliability index, c, for rotational loading

c
FS COVSu = 0.1 COVSu = 0.3 COVSu = 0.5
h = 2 m h = 10 m h = 2 m h = 10 m h = 2 m h = 10 m
1.2 1.95 1.65 0.69 0.61 0.44 0.4
1.3 2.60 2.19 0.89 0.78 0.55 0.48
1.5 3.53 2.98 1.17 1.02 0.7 0.6
2 4.76 4.01 1.54 1.33 0.89 0.76

Table 8 Probability of failure, Pf, for rotational loading

Pf(%)
FS COVSu = 0.1 COVSu = 0.3 COVSu = 0.5
h = 2 m h = 10 m h = 2 m h = 10 m h = 2 m h = 10 m
1.2 2.56 4.95 24.51 27.09 33.00 34.46
1.3 0.47 1.43 18.67 21.77 29.12 31.56
1.5 0.02 0.14 12.10 15.39 24.20 27.43
2 9.68E-05 3.04E-03 6.18 9.18 18.67 22.36

In geotechnical engineering practice and for pure vertical loading, shallow foundations
with a factor of safety as high as FS = 3 rarely fail. However, for the considered
foundation, the most traditionally comfortable safety factor (FS = 3) would lead to a
Random finite element method for bearing capacity assessment 257

probability of failure as high as Pf= 3.59% and 14.92% associated with COVSu = 0.3 and
0.5 respectively for the anisotropic case ( h  10m,  v  2m). Moreover, usually for pure
horizontal and rotational loadings, dealing with safety factor of 1.5 normally leads to safe
design. Accounting for uncertainties shows that even with a safety factor as high as 2 the
foundation is expected to be insecure with probability of failure as high as 19.49% and
22.36% associated with COVSu = 0.5 for the anisotropic case respectively for pure
horizontal loading and rotational loading.
These results further support the conclusions drawn by Duncan (2000). He stated that
computing both factor of safety and probability of failure is better than computing either
one alone, both have value and each one enhances the value of the other.
Another interesting finding is that spatial variability of Su has a significant influence
on reliability measures. In other words, reliability index c decreases as COVSu increases
and the probability of failure changes several orders of magnitude when passing from the
lowest COVSu to the highest one (i.e., increasing from 1.72E-25 % to 0.75 % for the
isotropic case and for pure vertical loading). Therefore, reliability measures under
the assumption of ignoring spatial variability of Su may be severely biased since any
change in the coefficient of variation of Su highly affects the reliability index and its
corresponding failure probability.
Furthermore, according to the precedent tables, reliability measures highly depend on
fluctuation scales. Hence, taking into consideration soil anisotropy results in greater
probabilities of failure and smaller reliability index values compared to the isotropic case.
Physically, this implies that in a foundation where parameters are highly correlated in
space, if for example parameters at one location are found to have a value of βc greater
than 3, the other locations within the correlation scale will likely have a similar value.
However, if the correlation scale is short and the parameters at one location suggest
stability of the foundation, this information likely cannot be translated to other locations
at distances greater than the correlation scale. That is, the probability of failure thus
remains large. This is the physical explanation of effects of the fluctuation scales on the
reliability analysis of a foundation which illustrates the importance of choosing the
appropriate stochastic parameters to obtain design values in a probabilistic approach.

6 Conclusions

This paper constitutes a contribution to the development of probabilistic foundation


analyses by performing a mechanical-reliability coupling, using a code in Matlab and the
finite element software Abaqus, to evaluate the effect of uncertainties on the performance
of foundations under uniaxial vertical, horizontal and rotational loadings.
A probabilistic methodology focusing on soil variations is proposed. For this, the
undrained shear strength was assumed to be a random field which is log-normally
distributed through both isotropic and anisotropic configurations. The effect of soil
anisotropy on the uniaxial bearing capacities and reliability measures of the shallow
foundation was investigated for different values of coefficient of variation.
The paper findings highlight the insufficiency of the conventional approach and,
hence, the importance of the probabilistic foundation analyses to predict the foundation
failure.
258 A. Hentati et al.

In fact, results show that considering spatial variability of soil properties leads to
totally different failure mechanism surfaces from one realisation to another especially
under vertical loading. This change in the failure surface often leads to significant
changes in the ultimate bearing capacity. Moreover, for suitable safety factors, reliability
index βc may take values below 3 and failure probability can reach values clearly greater
than 10-3 which are not in the range of historical reliability measures for engineering
conveniences implying a bad performance of the foundation. It is not advocated here that
factor of safety analyses be abandoned in favour of reliability analyses. Instead, it is
suggested that factor of safety and reliability be used together, as complementary
measures of more acceptable design.
As far as anisotropy is concerned, the coefficient of variation of the undrained shear
strength was shown to induce uncertainty in the bearing capacity results while decreasing
the mean ultimate uniaxial capacity. Compared to the isotropic case with low horizontal
fluctuation scale, anisotropy with high horizontal fluctuation scale was found to increase
the bearing capacity of the shallow foundation. This means that neglecting the spatial
variability of the soil properties leads to an overestimation in the bearing capacity
prediction. Also, accurate determination of the uncertainties of stochastic parameters is
very important in obtaining reliable probabilistic results. In summary, for the considered
range of parameters, the results clearly show that the anisotropy and heterogeneity of
soils have a considerable influence on the bearing capacity and the reliability measures of
the shallow foundations. Among all, the coefficient of variation of the strength parameter
is more highlighted to influence the bearing capacity calculation.

References
Ang, A.H.S. and Tang W.H. (2007) Probability Concepts in Engineering: Emphasis on
Applications in Civil and Environmental Engineering.
Baecher, G.B. and Christian J.T. (2003) Reliability and Statistics in Geotechnical Engineering.
Butterfield, R., et al. (1997) ‘Standardized sign conventions and notation for generally loaded
foundations’, Geotechnique, Vol. 47, No. 5, pp.1051–1054.
Cassidy, M. J., et al. (2013) ‘Probabilistic combined loading failure envelopes of a strip footing on
spatially variable soil’, Computers and Geotechnics, Vol. 49 (Supplement C): pp.191–205.
Charlton, T.S. and Rouainia M. (2016) ‘Probabilistic capacity analysis of suction caissons in
spatially variable clay’. Computers and Geotechnics, Vol. 80 (Supplement C): pp.226–236.
Cherubini, C. (1997) ‘Data and considerations on the variability of geotechnical properties of
soils’, Proceedings of the international conference on safety and reliability, ESREL.
Cherubini, C. (2000) ‘Reliability evaluation of shallow foundation bearing capacity on c’  soils.’
Canadian Geotechnical Journal, Vol. 37, No. 1, pp.264–269.
Cherubini, C., et al. (1993) ‘The coefficients of variation of some geotechnical parameters’,
Probabilistic Methods in Geotechnical Engineering. Editado por KS Li y S.-CR Lo. AA
Balkema, Rotterdam, pp.179–184.
Ching, J. and Phoon K.-K. (2013) ‘Probability distribution for mobilised shear strengths of
spatially variable soils under uniform stress states’. Georisk: Assessment and Management of
Risk for Engineered Systems and Geohazards, Vol. 7, No. 3, pp.209–224.
Cornell, C.A. (1969) ‘A Probability Based Structural Code’, ACI-Journal, Vol. 66, pp.974–985.
Der Kiureghian, A. and Ke J.-B. (1988) ‘The stochastic finite element method in structural
reliability’, Probabilistic Engineering Mechanics, Vol. 3, No. 2, pp.83–91.
Random finite element method for bearing capacity assessment 259

Duncan, J. M. (2000) ‘Factors of Safety and Reliability in Geotechnical Engineering’, Journal of


Geotechnical and Geoenvironmental Engineering, Vol. 126, No. 4, pp.307–316.
Fenton, G. and Griffiths D. (2008) Risk Assessment in Geotechnical Engineering.
Gourvenec, S. and Randolph M. (2003) ‘Effect of strength non-homogeneity on the shape of failure
envelopes for combined loading of strip and circular foundations on clay’, Géotechnique, Vol.
53, No. 6, pp.575–586.
Griffiths, D. and Fenton G.A. (2009) ‘Probabilistic settlement analysis by stochastic and random
finite-element methods’, Journal of Geotechnical and Geoenvironmental Engineering,
Vol. 135, No. 11, pp.1629–1637.
Griffiths, D.V. and Fenton G.A. (2001) ‘Bearing capacity of spatially random soil: the undrained
clay Prandtl problem revisited’, Géotechnique, Vol. 51, No. 4, pp.351–359.
Griffiths, D.V. and Fenton G.A. (2004) ‘Probabilistic Slope Stability Analysis by Finite Elements’,
Journal of Geotechnical and Geoenvironmental Engineering, Vol. 130, No. 5, pp.507–518.
Griffiths, D.V. and Fenton G.A. (2007) Probabilistic Methods in Geotechnical Engineering,
Springer-Verlag Wien.
Griffiths, D.V., et al. (2002) ‘Bearing Capacity of Rough Rigid Strip Footing on Cohesive Soil:
Probabilistic Study’, Journal of Geotechnical and Geoenvironmental Engineering, Vol. 128,
No. 9, pp.743–755.
Hicks, M.A. and Samy K. (2002) ‘Influence of heterogeneity on undrained clay slope stability’,
Quarterly Journal of Engineering Geology and Hydrogeology, Vol. 35, No. 1, pp.41–49.
Jamshidi Chenari, R. and Mahigir A. (2014) ‘The effect of spatial variability and anisotropy of
soils on bearing capacity of shallow foundations’. Civil Engineering Infrastructures Journal,
Vol. 47, No. 2, pp.199–213.
Lacasse, S. and F. Nadim (1996) ‘Uncertainties in characterising soil properties’, Uncertainty in
the geologic environment: from theory to practice (Uncertainty ‘96), A Geotechnical Special
Publication, Madison, USA, pp.49–75.
Lee, I., et al. (1983) Geotechnical Engineering Pitman, London.
Li, K. and Lumb P. (1987) ‘Probabilistic design of slopes’, Canadian Geotechnical Journal,
Vol. 24, No. 4, pp.520–535.
Lumb, P. (1974) ‘Application of statistics in soil mechanics’. Soil Mechanics New Horizons. IK
Lee, ed.
Massih, D.S.Y.A., et al. (2008) ‘Reliability-Based Analysis and Design of Strip Footings against
Bearing Capacity Failure’, Journal of Geotechnical and Geoenvironmental Engineering,
Vol. 134, No. 7, pp.917–928.
Murff, J.D. (1994) ‘Limit analysis of multi-footing foundation systems’, 8th International
Conference on Computer Methods and Advances in Geomechanics, Rotterdam, Balkema.
Nobahar, A. and Popescu R. (2000) ‘Spatial variability of soil properties – effects on foundation
design’, the 53 rd Canadian Geotechnical Conference, Montreal, Quebec.
Phoon, K.-K. and Kulhawy F. H. (1999) ‘Characterization of geotechnical variability’, Canadian
Geotechnical Journal, Vol. 36, No. 4, pp.612–624.
Phoon, K.-K. and Kulhawy F.H. (1999) ‘Evaluation of geotechnical property variability’,
Canadian Geotechnical Journal, Vol. 36, No. 4, pp.625–639.
Popescu, R., et al. (2005) ‘Effects of random heterogeneity of soil properties on bearing capacity’,
Probabilistic Engineering Mechanics, Vol. 20, No. 4, pp.324–341.
Prandtl, L. (1921) ‘Hauptaufsätze: Über die Eindringungsfestigkeit (Härte) plastischer Baustoffe
und die Festigkeit von Schneiden’, ZAMM – Journal of Applied Mathematics and Mechanics /
Zeitschrift für Angewandte Mathematik und Mechanik Vol. 1, No. 1, pp.15–20.
Rackwitz, R. (2000) ‘Reviewing probabilistic soils modelling’, Computers and Geotechnics,
Vol. 26, Nos. 3-4, pp.199–223.
Selmi, M. (2012) Étude probabiliste de stabilité des pentes: Simulation par champs stochastiques,
Editions Universitaires Europeennes.
260 A. Hentati et al.

Skempton, A.W. (1951) ‘The Bearing Capacity of Clays’, Selected Papers on Soil Mechanics,
pp.50–59.
US Army Corps of Engineers (1997) ‘Engineering and design introduction to probability and
reliability methods for use in geotechnical engineering’, Engr. Tech. Letter Washington,
1110-2-547.
Uzielli, M., et al. (2007) ‘Soil Variability Analysis for Geotechnical Practice’, Characterization
and Engineering Properties of Natural Soils-Tan, Phoon, Hight& Leroueil, London: Taylor &
Francis group.
Vanmarcke, E. (1983) Random Fields Analysis and Synthesis.
Vanmarcke, E.H. (1977) ‘Probabilistic modeling of soil profiles’, Journal of the Geotechnical
Engineering Division, Vol. 103, No. 11, pp.1227–1246.

View publication stats

You might also like