3rd International ASRANet Colloquium

10 12th July 2006, Glasgow, UK.

RELIABILITY ANALYSIS OF EARTH DAMS:

CASE OF EL HOUAREB DAM KAIROUAN - TUNISIA
Z. Mrabet, Geo-Risk Consulting, USA
Mohamed Ridha El ouni, Institut National Agronomique de Tunis, Tunisia
Khaled Kheder, Institut Superieur des Etudes Technologiques de Nabeul, Tunisia
ABSTRACT
The reliability index of an earth dam in commonly taken as the value corresponding to the failure surface
associated with minimum reliability index. However, embankment dams are considered as systems
composed of several infinite number of possible failure surfaces associated with different reliability
indices. In this paper, the reliability analysis has been performed on El Houareb embankment dam
(Tunisia). Here, basic assumption, which considers soil properties of the embankment dam are
statistically homogeneous, has been followed. Special attention has been paid to the global probability of
failure. The calculated global probability of failure value is found to be close to the value associated with
the critical ellipsoid failure mechanism. Hence, the concept of global probability of failure is coherent;
should be considered, later, as the probability of failure of the project.

1.

INTRODUCTION

There is growing need within geotechnical

engineering for rational ways of handling
uncertainty and taking it into account for decisionmaking. Many problems still facing geotechnical
engineers regarding the reliability analysis and its
applicability for practical problems. The following
relevant questions should be addressed: What
factors most strongly influence the level of
uncertainties in earth structures evaluations?
Where are improvements most needed to reduce
uncertainty in these evaluations? And what level
of reliability of earth structures would be achieved
under average conditions? Unfortunately, general
responses are very difficult to these questions.
The demand for risk analyses is growing in all
scientific and technical fields. Increasing attention
is being paid to risk and uncertainty in
geotechnical and civil engineering, because of the
drive for improved reliability and safety. The
engineer is confronted with uncertainty associated
with the random nature, spatial variability of
geotechnical properties, and in the complexity of
engineering projects. The engineer is expected to
make dependable and clear decisions. To do so
requires an understanding of both the nature of
uncertainty and appropriate techniques to manage
it.
Unfortunately, many geotechnical engineers are
still skeptical of the outcomes of reliability
methods in geotechnical engineering. They prefer

to use conventional methods that are more

straightforward. The engineer has often resorts to
simple criteria to evaluate the safety of an earth
structure (i.e. earth dam). In general, these criteria
are used to test if the function of the structure is
fulfilled. Engineer, for example, use a criteria for
existing structures based on consideration of
lifetime. For a given highway, for instance, a life
cycle is a criterion associated with traffic, which
expressed in cycle numbers and verify that the
fatigue strength withstands the traffic. For an earth
dam, the approach is more complex, constraints
and criteria are multiple. Lifetime is not a
criterion. However, the risk of failure could be
one.
Thus, the philosophy regarding to the safety of
these structures should reconciled two points of
view:
A partial evaluation issued from the
engineering practice and based on either the
factor of safety or the uncertainty.
A global theoritecal evaluation strongly
anchored in the theory of probabilities and
notably the optimization theory of the global
cost of the structure.
Reliability analyses can be used in routine
geotechnical engineering practice. How should
probabilistic methods be introduced to practicing
geotechnical engineers who have no background
in the probabilistic theory? These simple
reliability analyses require a little effort more that
involved in conventional geotechnical analyses.
They provide a means of evaluating the combined

effects of uncertainties in the parameters involved

in the calculations, and they offer a useful
supplement to conventional analyses. The
additional parameters needed for the reliability
analyses standard deviations of the parameters can
be evaluated using the same amount of data and
types of correlations that are widely used in
geotechnical engineering practice.
During last two decades a significant body of
literature has been published including several
methodologies and applications (Magnan, 1982,
Benjamin and Cornell, 1970, Li et al., 1987a,
Vanmarcke, 1983). Probabilistic methods have
been developed to solve geotechnical design
problems (Harr, 1987). The geotechnical
engineering designer has to provide a way to
systematically incorporate uncertainty into the
design process in a rational manner and to must
take it into account the soil variability and
optimize design (Cherubini, 1987, Fenton, 1996,
Mrabet, 1999, Mrabet and Giles, 2002, Mrabet,
2004).
Within the literature, there is a multitude of
examples illustrating probabilistic techniques for a
wide variety of problems in areas such as slope
stability, embankments, seepage, offshore
structures, foundation settlement, pillar stability
and bearing capacity.
On the other hand, reliability assessment
techniques should be recognized as an additional
tool to existing deterministic methods in the
evaluation of, par example, risk of failure of an
earth dam.
2.

RELIABILITY ANALYSIS

Geotechnical engineering reliability analysis is

concerned with finding the reliability or
probability of failure (or reliability index) of a
structure or a system. The benefit of reliability
analysis in geotechnical engineering can be
summarized in the following points:
to highlight the uncertainties in design of these
structures; reliability analysis plays a major
role in considering the uncertainties
influencing the design of earth structures. For
example, an optimum procedure for design of
an embankment can be discussed where there
are uncertainties with regard to a stability
problem.
allow the geotechnical engineer to quantify the
effect of various failure preventive measures

on these structures in order to develop an

inspection and maintenance program.
The reliability evaluation of most geotechnical
structures, in particular existing earth dams, the
capacity-demand model is the most simplest
utilized, as the question of interest is the
probability of failure related to a load event rather
than the probability of failure within a time
interval.
Reliability assessment methods are being adopted
for use to develop rigorous risk-management
programs. Implementing the programs will ensure
that safety is maintained to a robust and acceptable
level.
Any simple reliability analysis should include the
following steps:
Establishing limit states
Identifying failure modes
Formulating limit state functions
Analyzing uncertainty
Evaluating reliability
Assessment results
The reliability index of an earth dam is commonly
taken as the value corresponding to the failure
surface associated with minimum reliability index.
Thus, the conventional factor of safety is defined
as the ratio of limit capacity of soil to a demand in
terms of loads:

F=

R
S

(1)

in which R= capacity (resisting force or resisting

moment); and S= demand (driving force or driving
moment).
In probabilistic modeling of safety, R and S are
assumed to be random variables. Let fR(r) and fS(s)
be the probability densities functions of variables
R and S. The probabilistic measure of safety is the
probability of failure, Pf in which should be
smaller than certain reference values set a priori.
The probability of failure is defined as (failure
occurs if R<S):

Pf = P 1
S

(2)

Assuming statistical independence between the

variable R and S the probability of failure can be
expressed as:

 r =s

(
)
f
s
S f R (r )dr ds
+

Pf =

(3)

The use of later formulation of probability of

failure makes the simplification possible only for
certain types of distribution of R and S such a
normal distribution. In such case the notion of
safety margin, MS=R-S (cornell 1970) can be
introduced. It is Possible to derive the density
function fMS(MS) of the random variable MS and
the risk of failure is given as:
0

Pf =

f (MS )d (ms)
MS

(4)

In general, the calculus of the integrals in the

preceding equations is particularly cumbersome.
In this case, safety is defined by the reliability
index, , as (Cornell 1970):

E{MS }

MS

(5)

in which E{MS} = expected value of MS; and MS

= standard deviation of MS, provides a simple
quantitative basis for assessing risk i.e. probability
of failure.
An advantage of reliability index is that it can be
determined from two first statistic moments (mean
value and variance) of probability density
functions of R and S without any assumption on
the specific shape of these functions.

3.
PROBABILISTIC MODELING AND
RELIABILITY OF EARTH STRUCTURES

Probabilistic approaches have considered the

uncertainty of natural and compacted material
properties as random variables (Mrabet, 1993,
1997). Early, random variable model in
geotechnical concept has been introduced and
discussed in Benjamin and Cornell (Benjamin and
Cornell, 1970).
Where random variable of geotechnical property is
not available due to, for example, a lack of sitespecific data, uncertainty can be characterized by
assuming that the coefficient of variation of a
geotechnical parameter is similar to that observed

at other sites. Harr (1987) reported typical values

of coefficients of variation for soil properties.
However, caution should be taken when using
typical values, as coefficients of variation solely
does not reveal much in accordance with the
correlation structure of soil properties (Mrabet,
2004).
Li and White (1987a) pointed out that the
probability of failure may be reduce by three
orders of magnitude if the auto-correlation
function is taken into account. Cherubini (1997)
pointed out that integration of the fluctuation scale
(i.e. auto-correlation distance) in probabilistic
models in the geotechnical context generates
failure probabilities consistent with frequencies of
failure observed in practice.
In geotechnical engineering, the reliability
analysis of earth fills (earth dams, for example)
can be performed using Bayesian Updating
technique in conjunction with conditional random
field to evaluate the uncertainty related to spatial
variation of the materials properties within a dam
based on quality control results during
construction (Mrabet, 1997, 1999, Mrabet and
Bouayed, 2000). Thus, the mathematical
expectation and variance of the average shear
strength along failure surface have been estimated
from these results. Standard quality control
programs has incorporated the results of control
tests in reliability analysis as soon as they were
available in order to take decision based on
actualized evaluation of the reliability of earth
structures (Mrabet, 1999, 2004).
Stochastic finite element method (SFEM) is a
good alternative for solving the geotechnical
problem associated with material variability. The
formulation of this method is described in Ghanem
& Spanos (1989). The following points are
relevant when performing stochastic finite element
modeling in a geotechnical engineering context:
SFEM is useful in evaluating a range of
variation of finite element modeling in a
geotechnical engineering context.
SFEM is a useful tool in providing a quick
insight into the relative importance of different
parameters of soil constitutive laws i.e.
sensitivity analysis.
SFEM can be used as part of reliability
analysis leading to more robust designs for
geotechnical structures.
Recently, Random Finite Element Method
(RFEM) has been introduced to solve a variety of
practical problems in geotechnical engineering

design. RFEM combines random field theory

(Fenton, 1990) with the finite element method
(Smith and Griffiths, 1998) and Monte Carlo
simulation to produce probabilistic results. They
estimated the slope failure probability and
investigated resulting probability as a function of
the soils statistical parameters.
Mrabet (2002) and Mrabet and Bouayed (2003)
have used a methodology (figure 1) based on
random field theory in conjunction with stochastic
finite element method (SFEM) to describe the
uncertainty in both the input material properties of
a geotechnical system and the result of the
analysis of an embankment dam.
External Parameters
Geometry
Loading
Boundary conditions

Internal parameters
Soil parameters
(Material properties)

Parameter uncertainty

Uncertainty modelling
Random variable
Random field
(Simple, conditional)

Finite Element Algorithm

Output response
Displacements
Strains
Stresses
Pore pressure

Uncertainty modeling

Reliability analysis

Wadi Floods. El Houareb reservoir is a man-made

water-body built on the Oued Merguellil, 35 km
east of Kairouan, for flood-control and watersupply purposes. Where the river emerges from
the Dorsale the reservoir is bordered by higher
land, but also has extensive flat shores. It retains
the waters, which once flowed into Sebkha Kelbia
(Kairouan city, Tunisia). It has an average depth
of nearly 20 m, but in periods of poor rainfall, it
can remain completely dry for several years on
end. The water plant grows commonly in the
reservoir and provides the main food-source for
wildfowl. The hydraulic characteristics of El
Houareb dam are shown in Table 1
Table 1. Hydraulic characteristics of El
Houareb dam
Area of basin pouring
1140 km2
Total capacity
110 .106 m3
Yearly average contribution
70,00 106 m3

El Houareb dam is founded on a sedimentary

basin. It is filled with a Triassic, Cretaceous and
Tertiary aged marine and fluvial sedimentary
fractured rock masses on its right side.
The dam has a height of 32 m and crest width of
8.5 m. A berm of 40 m long located at the
downstream shoulder contributes into the stability
of the structure. The clayey core, which provides
impermeable barrier within the body of the dam,
has a sloppy upstream 1V: 0.45H, 6 m wide at the
top and 21 m wide at the foundation level. The
outer slopes of the dam are made of 1V: 4H
upstream shoulder and 1V: 3H and 1V: 2.65H
below and above the berm platform located at the
downstream shoulder respectively.
Engineering analyses for the proposed dam were
performed to evaluate suitable dam sections for
the site conditions and available on-site
construction materials (figure 2).
Filter

Figure 1 A flow chart of the methodology (Mrabet and Bouayed,

2003)

Drain
Downstream
shoulder
Berm

Core

4.

CASE STUDY

4.1
DESCRIPTION AND PRESENTATION
OF EL-HOUAREB DAM
El Houareb dam (Central Tunisia) is selected as a
case study to perform the reliability analysis. The
main objective of this dam is to contain Merguellil

Upstream
shoulder
Foundation

Figure 2. General cross section schematic of El-Houareb

dam.

PROPERTIES OF THE DAM AND

FOUNDATION MATERIALS

4.2

The compacted materials were evaluated

according to their maximum dry unit weight
(dmax), optimum water content (wopt), specific
gravity (s), liquid limit (LL) and plasticity index
(Ip), all the parameters except specific gravity
indicated the desirable characteristic as an
impervious fill material.
The construction material for embankment and
foundation were:
Material A- Filter zone: granular material
Material B- downstream shoulder zone:
granular material
Material C- Core zone: very plastic clay has
been selected in order to withstand any
eventual settlements and to resist into water
seepage in case of occurrence of fissures.
Material D- soil foundation and upstream
shoulder zones: clay with medium plasticity
The main physical and shear strength parameters
are reported in Table 2.

The former method has been selected to evaluate

the safety factor.
Due to the foundation layers and the structures of
the embankment dam, two mechanisms of failure
have been considered; ellipsoid failure and wedge
failure mechanisms. Minimum safety factors of
2.01 and 1.94 for ellipsoid and wedge failure
mechanisms have been found respectively. An
example of Clara 3D analyses of El Houareb dam
is shown in figure 4.

Table 2. Physical properties and shear strength

parameters.
Material
A
B
C
D
Properties
WL (%)
IP (%)
W (%)
d/w
33,30 30
(Degree)
0
0
c (kPa)
c= cohesive strength,=friction

53
31
13,5
1,91
23
25

37
21
15
1,89
25,30
10

4.3
PERFORMANCE AND ANALYSIS OF
EL-HOUAREB DAM
The stability analysis of El Houareb embankment
dam and its foundation is carried out using a
deterministic approach. The three-dimensional
limit equilibrium CLARA program (Hungr 1987)
is used here to evaluate the factor of safety against
different mechanisms of failure. Different modes
of failures are implemented in CLARA program
which provides a choice of methods of analysis
including the following: Fellenius's method,
Bishop's method, Spencer's method, Janbu's
method and 3D extensions of Bishop's Simplified.

Figure 3: Analysis of upstream slope of El Houareb dam

using limit equilibrium to find the ellipsoid failure surface
(3D extension of Bishops simplified method, Clara 3D)

Subsequently, for the reliability analysis of El

Houareb dam, only the ellipsoid failure
mechanism will be considered.

5.

RELIABILITY ANALYSIS

Spatial correlation has long been ignored in

modeling variability of soil properties. However,
the spatial dependency within the medium should
be considered, particularly in a strongly
compacted soil i.e earth dams (Mrabet and
Bouayed 2003). To take into account spatial
correlation, it is possible to model the spatial
variability of soil properties with a spatial
stochastic process also known as random field
(Mrabet, 1999) in which the variable exhibits
autocorrelation, the tendency for values of the
variable at one point to be correlated to values at
nearby points.

Recently, special attention has given to the role of

spatial correlation. Some recent papers dealing
with the concept include those by Mrabet and
Giles (2002)
Many studies stressed out the effect of existing
auto-correlation on the results of probabilistic
models of compacted earth slopes analysis.
Ignoring auto-correlation is conservative and
considerably more than desired (Mrabet 1998).
The analysis that considers typical auto-correlation
distances results in reduction of probability of
failure (Cherubini 1997).
Extensive measurement program performed on
earth dams such Mirgenbach and Vieux-Pre dams
(Rossa & Fry 1988) and others led to the
conclusion that, for such structures, a significant
spatial correlation exists. The influence distance at
which auto-covariance becomes negligible is
sensitive to the construction procedure as well as
the material nature, but is practically identical for
all properties. A pronounced anisotropy of the
auto-covariance exists with a vertical distance of
influence of the order of meters and horizontal
distance of influence of the order of tenth of
meters [Mrabet, 1999, Mrabet and Giles 2002].
Similar pattern have been found for mechanical
properties and the exponential auto-correlation
function between two different points within the
compacted soil of the El Houareb dam has been
retained.
Due to the lack of data concerning the horizontal
auto-correlation function, we were reported to
Andersons work (1981) to establish one. We
therefore, obtained the following function:

hor 9 x ) = exp(0,065 x)

(6)

The reliability index of earth dam in commonly

taken as the value corresponding to the failure
surface associated with minimum reliability index.
However, embankment dams are considered as
systems composed of several infinite number of
possible failure surfaces associated with different
reliability indices. Therefore, the global
probability of failure of embankment dam is
however, at least for the moment, a complicated
problem to handle since correlation exists between
different failure surfaces. In practice, the
reliability of whole system may be governed by a
few subsystems or components. The calculated
probability associated to the critical failure surface

constitutes a lower limit of the global probability

of failure of El Houareb dam.
Subsequently, we calculate the global probability
of failure in respect to the following conditions:
-Cross-section of El Houareb dam as considered in
the above analysis (figure 2).
-Horizontal auto-correlation distance =60 m
-Vertical auto-correlation distance =3 m
-Coefficient of variation of the cohesion of the El
Houareb dam= 0.35
Hence, the dam is divided into different ellipsoids.
They are considered independents and vertically
separated by 3 m.
Table 3. Failure Probability
Hi (m)
Fi
Pfi
i
0m
2.01
3.45 0.025
Contact
DamFoundati
on
2.63
4.00 0.00013
3
2.95
4.23 0.0003
6
3.50
4.55 0.00002
9
3.78
4.80 0.000003
12
5.14
5.23 0.0000023
15
7.02
5.40 0.0000001
18
9.04
5.63
0.000000025
21
15.02
5.90 0.000000004
24
18.3
6.13 0.0000000032
27
25
6.23 0.0000000005
30
Hi = depth of the ellipsoid regarding the interface
dam-foundation (or the bottom of the dam)
Fi = Factor of Safety corresponding to the ellipsoid
failure Surface number i.
i = Minimum Reliability index corresponding to
the ellipsoid failure surface number i.
Pfi = Failure probability corresponding to the
ellipsoid failure surface i,
The global failure probability is calculated using
the following equation:

Pglobal = 1 (1 Pfi )
n

(7)

i =1

The global probability of failure is P = 0.02. This

value is close to the value associated with the
critical ellipsoid failure surface. This calculation
shows that the concept of global probability is
coherent; should be considered, later, as the global
probability of the project.

In reality, these results show that other sources of

uncertainty that should be taken into account
including, but not limited to:
ignorance of mechanisms of failure
ignorance of the entire history of dam behavior
ignorance of the horizontal auto-correlation
length.

5.

CONCLUSIONS

The probabilistic approach applied to clayey

embankment dams seems, from now on, an
interesting tool for the engineer who has to deal
with the safety and reliability of these structures.
However, the reliability analysis should be
considered as an efficient tool that complemented
a conventional deterministic analysis such as the
equilibrium limit analysis.
A complete reliability analysis should include all
sources of uncertainty. Particularly, correlations
between different properties that characterize
compacted materials and their corresponding
horizontal auto-correlation lengths generate main
uncertainties in the probabilistic model.
Similar reliability analysis could be performed
using conditional random field to evaluate the
uncertainty related to spatial variation of the
material properties within the dam based on
quality control.

REFERENCES

1. Benjamin, J., and Cornell, C.A. 1970.

Probability, Statistics and Decision for
Civil Engineers, McGraw-Hill, New York.
2. Cherubini, C.1997. Data and consideration
on the variability of geotechnical
properties of soils. Proceedings of the
Conference on advances in Safety and
Reliability: ESREL 1997. pp. 1583-1591.
3. El ouni M. R, 2004. Calcul en deformation
dun barrage. Colloque Materiaux, Sols et
Structures MS2 2004.
4. Hungr, O. 1987. An extension of Bishops
Simplified method of slope stability to
three dimensions. Geotechnique 37 (1):
113-117.
5. Magnan, J. P. 1982. Les methodes
statistiques et probabilistes en mecanique

des sols. Presse de lEcole Nationale des

Ponts et Chaussees, France.
6. Mrabet, Z., 1998. Geostatistics and shortterm reliability analysis of homogeneous
compacted
earth
fills.
Journees
Gestatistique, ENSMP, Fontainebleau,
France, Cahiers de Geostatistique, (6): pp
91-104.
7. Mrabet, Z., 1999. Reliability analysis of
homogeneous earth fills, A new approach.
Proceeding of the eight International
Conference on Applications of Statistics
and Probability in Civil Engineering,
ICASP8, 12-15 December 1999, Sydney,
New South Wales, Australia pp 499 - 507
8. Mrabet, Z. and Bouayed, A.
2000.
Reducing Uncertainty on the Results of
Reliability Analysis of Earth Fills Using
Stochastic
Estimations.
Second
International Conference on Computer
simulation in Risk Analysis and Hazard
Mitigation, Risk Analysis 2000, Bologna,
Italy, 13 October 2000.pp 203-214
9. Mrabet, Z. 2002. Reliability analysis of
earth fills using stochastic estimation
methods. Third International Conference
on Mathematical Methods in Reliability,
MMR 2002, Trondheim, Norway, 17 - 20
June 2002.
10. Mrabet, Z. & A. Bouayed. 2003.
Probabilistic
Risk
Assessment
of
Homogeneous Earth Dams. Proceeding of
the Ninth International Conference on
Applications of Statistics and Probability
in Civil Engineering, ICASP9, July 6-9,
2003, San Francisco, California pp. 367 372
11. Mrabet, Z. & D. Giles. 2002. Probabilistic
risk assessment: a key tool for reducing
uncertainty in geotechnical engineering.
Invited
paper.
Third
International
Conference on Computer Simulation In
Risk Analysis and Hazard Mitigation,
RISK 2002, Sintra, Portugal, 19 - 21 June
2002, pp. 3-14.
12. Mrabet, Z. 2004. Invited paper. Some
aspect on reliability in geotechnical
engineering.
Fourth
International
Conference on Computer Simulation In
Risk Analysis and Hazard Mitigation,

RISK 2004, Rhodes, Greece, 19 - 21 June

2004, pp. 75-84
13. Mrabet, Z., El Ouni, Kheder, K. 2006.
Probabilistic modeling and reliability
analysis of earth structures in geotechnical
engineering. Paper accepted to be
presented in the First Euro Mediterranean
in Advances on Geomaterials and
Structures, Hammamet 3-5 May Tunisia
14. Rossa, O. and Fry, J.J. 1988. Exploitation
des donnees recueillies sur la digue Aube.
Rapport interne.
15. Vanmarcke. E.H. 1983. Random fields:
Analysis and synthesis. The MIT Press,
Cambridge, Mass., 1983.