2008:14

RESEARCH REP O RT

ON INTERNAL EROSION IN
EMBANKMENT DAMS
A literature survey of the phenomenon and
the prospect to model it numerically

H. Mattsson, J. G. I. Hellström & T. S. Lundström

Luleå University of Technology
Department of Civil, Mining and Environmental Engineering
Division of Mining and Geotechnical Engineering

Luleå University of Technology

Department of Applied Physics and Mechanical Engineering
Division of Fluid Mechanics

2008:14|:฀1402-1528|:฀-f฀฀--฀08฀⁄14฀--฀฀
 ON INTERNAL EROSION IN
EMBANKMENT DAMS

A literature survey of the phenomenon and the

prospect to model it numerically

H. Mattsson*, J. G. I. Hellström† & T. S. Lundström†

* Luleå University of Technology

Department of Civil, Mining and Environmental Engineering
Division of Mining and Geotechnical Engineering

†
Luleå University of Technology
Department of Applied Physics and Mechanical Engineering
Division of Fluid Mechanics
Cover illustration: sinkhole on the tailings dam at Zinkgruvan in Sweden
(picture courtesy of SWECO).
PREFACE
The work presented in this research report has been carried out as a collaboration project
at Luleå University of Technology between the research divisions:

x Mining and Geotechnical Engineering, Department of Civil, Mining and

Environmental Engineering
x Fluid Mechanics, Department of Applied Physics and Mechanical Engineering.

This research project has been financially supported by SVC (Swedish Hydropower
Centre).

Many people have assisted us during the work with this literature survey. We would like
to especially thank Prof. Sven Knutsson and Prof. Håkan Gustavsson for helpful
suggestions during this project.

Luleå, October 2008

Hans Mattsson, J. Gunnar I. Hellström and T. Staffan Lundström

i
ABSTRACT
The main objective with this literature survey is to elucidate the state of the art of internal
erosion in embankment dams in order to be able to formulate a research program for
numerical modelling of internal erosion in a physically sound manner. Since these
processes normally are localised to specific zones in a dam, the ordinary continuum
approach frequently utilised in soil modelling will not, by itself, be successful. The plan of
the research group is therefore to treat internal erosion numerically as a type of localisation
and describe the constitutive behaviour with micromechanical models in localised zones.
In the next step, the internal erosion model developed will be implemented in a
mathematical consistent fashion in a continuum model, based on e.g. the finite element
method. In such software, ordinary computations of stresses, strains, deformations and pore
pressures in an embankment dam can be performed; results which possible lead to
conclusions about the initiation of internal erosion processes. When internal erosion is
initiated, the micromechanical models will describe these processes in localised zones.

It was decided to restrict the literature survey to areas important for the research direction
defined above. The chosen areas are thus: numerical modelling of embankment dams,
internal erosion processes, embankment dam failures and accidents, filter design,
micromechanical models of hydro dynamical loads on single particles and system of
particles as well as micromechanical models for friction between single grains and system
of grains. The most essential findings for each area, in the context of the research, will be
described below.

There exists much experience in the field numerical modelling of embankment dams
which is exemplified by the large amount of published papers on the subject. However,
the processes of internal erosion have not been modelled in a general manner, as outlined
in the research idea above, in any of the papers cited in this literature survey.

There are three main internal erosion processes that can initiate piping: backward erosion,
concentrated leak and suffusion. Piping can occur in the embankment, through the
foundation and from the embankment into the foundation. Embankment dams are
normally constructed in zones of different materials. Different compressibilities of the
various zones might lead to internal redistribution of stresses and uneven settlements,
which might cause cracks or ”soft zones” where internal erosion can be initiated. Cracks
can also occur later on by hydraulic fracturing. All of that described above are examples of
mechanical processes that are theoretically possible to model numerically. To receive
information about potential internal erosion problems in a dam at an early stage,

iii
geophysical methods are a promising alternative or complement to numerical methods.
However, much research and development remain before the results from geophysical
methods utilised for permanent monitoring and surveillance of dam structures are
completely trustful.

It is often very difficult to determine the exact reason for a dam accident or a failure since
the processes involved have a tendency to destroy evidence which might have existed.
Statistical data, however, shows that failure by overtopping and piping are the most
common modes of failure while failure by slides is less common. Many piping failures
occur very fast, leaving a short time for proper actions.

The safety of large embankment dams is strongly dependent on the reliability of the
performance of their critical filters. Existing filter design criteria are in most cases
empirically derived, implying that a lot of knowledge can be gained by taking a
mechanical approach to the problem. More research would also be desirable in the field of
ageing effects of dams and uncertainties of core/filter appearance during and after
accidents, incidents etc. For this, numerical modelling also seems to be a promising
approach.

Continuum formulations of flow through porous media often results in equations

consisting of a few unknown parameters such as the permeability. The physical
background of such parameters can often be traced to the detailed flow in the pores and it
is therefore in place to study the flow on this level, as well. The forces on individual
particles have been exploited for certain geometries and for a number of flow conditions.
We however need to investigate further higher Reynolds number flows, more complex
geometries and instationary conditions.

The forces from the micromechanical models must be balanced by gravity and forces that
emanates from particle interactions for the dam to be stable. As a first, and most simple,
criteria the size distribution of the particles being subjected to hydrodynamic pressure is
compared to the pore size distribution of the porous medium. Hence the particles will
move if they are small enough independent on magnitude of the force on it. In reality,
however, the hydraulic forces must exceed particle interaction forces and or gravitational
forces keeping the particles in place at normal conditions. The forces are strongly
dependent on the size of the particles and therefore dependent on different phenomena
such as cohesion, adhesion and static friction.

In the final part of the literature survey a concept for numerical modelling of internal
erosion is presented based on ideas that have emerged from this work. The concept

iv
involves as well mathematical developments in order to formulate a micromechanical
model for internal erosion as laboratory tests. The theoretical work and the practical
laboratory work should be performed simultaneously and in an interactive manner. A
software containing a model that simulates internal erosion could be useful for: an increase
of the knowledge about internal erosion processes, evaluating the risk for dam incidents
caused by internal erosion, estimating the time for progression of internal erosion and
piping, studying self-healing of leaks, changes in filter behaviour subsequent to particle
accumulation and ageing effects in dams, analysing the amount of instrumentation needed
in a dam and the proper location for monitoring and surveillance as well as designing dams
to mention a few examples. It is therefore apparent that the route suggested has a high
potential to become a tool for future improvements of dam safety.

v
TABLE OF CONTENTS
Preface ............................................................................................ i
Abstract..........................................................................................iii
1. Introduction ...........................................................................1
2. Numerical modelling of embankment dams ..............................3
2.1. Continuum models................................................................... 5
3. Internal erosion processes ........................................................9
3.1. Uneven settlements and hydraulic fracturing..............................11
3.2. Methods to examine and measure internal erosion .....................19
4. Embankment dam failures and accidents ................................. 23
4.1. Internal erosion susceptibility of glacial till ................................27
5. Filter design ......................................................................... 29
5.1. Segregation of filter materials...................................................33
6. Micromechanical models of hydro dynamical loads on single
particles and system of particles.............................................. 37
7. Micromechanical models for friction between single grains and
system of grains for non-cohesive soils .................................... 41
8. Concept for numerical modelling of internal erosion ................ 43
Acknowledgement .......................................................................... 47
References ..................................................................................... 49

vii
1. INTRODUCTION
Internal erosion in embankment dams is not a completely understood phenomenon. Dam
failures, accidents and deterioration of dams because of ageing effects do take place as a
result of internal erosion. Even if dams generally should be considered as safe
constructions, a large part of the present knowledge about problems associated with
internal erosion is a result of studies of former dam incidents. This circumstance is almost
inevitable since internal erosion processes are very complex and take place inside the
embankment or foundation making the erosion transparency limited until it has progressed
enough to be visible or detected by measurements. Instrumentation inside a dam could
perhaps detect erosion processes at an early stage but at the same time internal erosion has
a tendency to concentrate adjacent to different kinds of installations, so it is not desirable
to have more installations than necessary.

There has been much progress in the field of numerical modelling of advanced
geotechnical problems during the last decades. To the authors´ knowledge, the processes
of internal erosion in embankment dams have, however, still not been incorporated
successfully in such numerical models. A possible explanation for this could be that soil
normally is interpreted as a continuum in commercial numerical software, while the
analysis of formation of cracks, softened zones and pipes as a result of the erosion process,
almost certainly need some additional approach based on e.g. localisation. It is, of course,
also important to have sufficient knowledge of the fundamental mechanical behaviour of
internal erosion to be able to model it in a proper way.

As a part of the initiative from the Swedish Hydropower Centre (SVC: Svenskt
Vattenkraftscentrum) to secure and develop knowledge and competence in Sweden within
the hydropower engineering field, our research group is focusing on dam safety issues. The
overall objective for future research in the group is to develop a numerical model that can
simulate the processes of internal erosion in a physically sound fashion. In addition, the
model should be implemented in a software code in order to be applicable in research and
for practical purposes. As a first step, the software should be validated by comparing the
current knowledge of the mechanical responses from the internal erosion processes with
results from simulations in order to determine how successful the tool is; and if the result is
positive the software can be used in more general cases. Beyond the obvious fact that it is
of considerable value to increase the understanding about the internal erosion processes,
there are many potential advantages with a software containing a model that simulates
internal erosion, it can for instant be utilized for: design purposes, evaluating the risk for
accidents and failures involving internal erosion, analysing the time for development of

1
internal erosion and piping, studying the mechanisms associated with ageing of
embankment dams, investigating the self-healing phenomenon and the potential remaining
damage to the structure, examining the quantity of instrumentation needed as well as
suitable locations for them. Naturally, such software should also be able to model ordinary
variables dealt with in numerical geotechnical modelling software, like: stresses, strains,
pore pressures, deformations etc. To comprehend the physical processes involved in
internal erosion, laboratory tests must be performed and the result should be a base for the
formulation of mathematical criteria.

This literature survey is a result of the research group’s ambition to understand the state of
the art of internal erosion and numerical modelling of the phenomenon before the final
research programme is formulated. In Section 8, some ideas are presented, that have
emerged from this literature survey, about how to actually carry out the numerical
modelling project of internal erosion.

2
2. NUMERICAL MODELLING OF EMBANKMENT
DAMS
Numerical modelling can be a useful tool in geotechnical engineering since a wide variety
of problems is theoretically possible to model. Embankment dams, that geometrically are
quite simple structures, are especially attractive to model numerically in order to get an
early notion of potential hazardous processes, not visible on the outside but that can be in
progress inside the dam. The main problem with numerical modelling (e.g. noticed by Fell
et al., 2005) is that those performing such analyses normally need specialist knowledge in a
wide range of technical subjects. At the same time, numerical software is within reach for
almost anyone with a computer. Not seldom, commercial software are extremely user-
friendly which makes it easy to produce results, even wrong ones, often presented in the
form of colourful plots. Potts and Zdravkovi´c (1999, 2001) list the knowledge desirable to
be able to perform useful geotechnical finite element analyses. ”Firstly, a sound
understanding of soil mechanics and finite element theory is required. Secondly, an in-
depth understanding and appreciation of the limitations of the various constitutive models
that are currently available is needed. Lastly, users must be fully conversant with the
manner in which the software they are using works. Unfortunately, it is not easy for a
geotechnical engineer to gain all these skills, as it is very rare for all of them to be part of a
single undergraduate or postgraduate degree course. It is perhaps, therefore, not surprising
that many engineers, who carry out such analyses and/or use the results from such
analyses, are not aware of the potential restrictions and pitfalls involved.” The books by
Potts and Zdravkovi´c (1999, 2001) and Muir Wood (2004) are examples of very good
texts on the subject ”numerical modelling in geotechnical engineering”; the books contain
many valuable general advises about how numerical modelling should be performed.

In practical modelling of dams, there will probably always exist uncertainties whether the
dam will have the same characteristic and behaviour after construction that was intended at
the design stage or not. This is certainly a challenge for a user of numerical software which
obviously besides detailed understanding of the computer program and its mathematical
basis must have engineering judgement and practical experience; since if the model is not
representative for the actual dam the outcome of the computation must be questioned. It
is not realistic to believe, that a precise prediction of the performance of the dam is always
possible and with that in mind a user should avoid to draw to “high-flying” conclusions.
Examples of insecurities, to deal with in embankment dam engineering are discussed in
e.g. Milligan (2003). Especially, the high probability of segregation at handling and
placement of some common types of core and filter materials should be paid attention in
studies of internal erosion. It is likely that erosion processes could be concentrated to or
3
initiated at zones with segregated material. The problem of segregation is described further
in Section 5.1 in connection to filter design.

The Finite Element Method (FEM) has been used to estimate stresses and deformations in
embankments for about 30 years, so there exists much experience in this field. In the
literature many papers can be found that contain result from finite element analyses on
embankment dams. Such papers, of special interest in the context of internal erosion, are
e.g. Sherard (1986), Ng and Small (1999), Zhang and Du (1997), Day et al. (1998), Sharif
et al. (2001) and Huang (1996). However, the processes of internal erosion have not been
modelled in a general manner in any of the papers the authors´ studied in this literature
survey.

When modelling internal erosion it can be convenient to start from a fluid mechanic point
of view, which creates possibilities regarding the solution of the fluid flow through porous
media, and the modelling can be performed in numerous ways as outlined by Bear (1972).
It can, for instance, be convenient to apply Hodographic methods in which the boundaries
of the fluid including the free surface within the dam are transformed to form simpler
geometries. Although the flow in it-self can be treated with explicit expressions such as
Darcy’s law non-homogeneous and anisotropic permeabilities and the naturally formed
free surface often makes realistic modelling with analytical tools complex. One possible
way to model the flow through embankment dams is instead to solve for the momentum
equations for two phases and then add an advection equation to deal with the motion of
the surface formed between the phases, the air and the water, for instance. To deal with
the sharp transition in viscosity and density a modified Heaviside-step function can be
introduced, see Sharif et al. (2001). Adaptive FEM analysis of free surface flow is also
applicable, Sheng-Hong (1996). Another method is to combine spectral analysis with a
Laplace transform in order to model the transport of material through for instance the
riprap, Wörman and Xu (2001). A combination of the finite element method with a
stochastic approach and the FEM part is used to solve the mechanical behaviour of the
earth structure while the stochastic method is used for the probabilistic part of the earth
modelling, Mellah et al. (2000). Numerical procedures based solely on the finite element
method are also common. Such a technique can, for instance, be applied to hydraulic
fracturing in the core of earth and rock-fill dams. The finite element procedure makes
then use of special joint elements that allow fluid flow and fracture to be modelled
allowing the progress of pore pressure in the core of a dam to be simulated, Ng and Small
(1999). The Finite Volume Method can be used to solve the flow through the porous
material. With such approaches the main flow features are resolved in a better way Elforsk
(1999) and Elforsk (2005).

4
2.1. Continuum models
Flow through porous media, which embankment dams can be considered as, is generally
modelled by a continuum approach implying that the flow in the pores is averaged and
that averaged quantities such as the permeability are introduced.

The flow through porous materials is often described by Darcy’s law that gives a linear
relationship between the flow rate and the pressure. The law was originally derived from
experiment Bear (1972), Scheidegger (1972) and Dullien (1992) but Tucker III and
Dessenberger (1994) have shown that the law can be derived from the Navier-Stokes
equations. The conditions are that inertia and long-range viscous effects are negligible of
which the former is generally a too crude assumption for at least the final stages of internal
erosion. Let us, however, for the moment stick to the stated assumptions and write Darcy’s
law in its general form:

K ij
vi  p, j (2.1)
P

where vi is the superficial velocity, Kij the permeability tensor, P the viscosity and p the
pressure. An obvious interpretation of (2.1) is that the permeability can vary as a function
of direction. It is also important to notice that:

vi
ui (2.2)
I

where ui is the average velocity in the pores and I the porosity. In this rigorous form of
Darcy’s law the permeability is solely dependent on the detail geometry of the porous
material and has the unit m2. In the literature describing embankment dams it is also
common to give the permeability the unit m/s, conductivity or velocity, Fell et al. (2005).
This implies that the viscosity and the gravitational constant are incorporated in the actual
permeability value. The advantage of this approach is that a velocity-like component is
explicitly given. The draw-back is that the actual value is dependent on the conditions for
the measurements.

5
Figure 2.1 Set-up by Darcy.

The permeability of a porous material is generally obtained from measurements although

numerous models often based on the micro geometry of the porous media have been
derived. The models are developed for special cases and measurements need to be carried
out in order to calibrate the models Gebart (1992). Many methods for the experimental
assessment of the permeability tensor have been proposed following the experiments by
Darcy; Kershaw (1972), Adams et al. (1988), Pikulik et al. (1991), Parnas and Salem
(1993), Parnas et al. (1995), Young and Wu (1995), Gebart and Lidström (1996), Lekakou
et al. (1996), Weitzenböck et al. (1998), Lundström et al. (1999) and Lundström et al.
(2000). Most of these methods are based on either of two fundamentally different
principles: (i) parallel flow, by which the liquid is made to flow in a controlled direction
through the porous sample (the Darcy experiment, see Figure 2.1); and (ii) radial flow, by
which the liquid is injected at a “point” and flows freely in all directions (usually in a
plane). Either of these principles may be applied in a variety of ways, e.g. by driving the
liquid by a constant flow rate or a constant pressure drop. The advantage of the radial flow
techniques is that a full permeability tensor, including principal directions, can be obtained
in a single experiment Ahn et al. (1995) and Weitzenböck et al. (1998). The disadvantage
is that they are based on tracking a flow front, i.e., the flow must be unsaturated, with the
risk of including various transient effects, such as capillary action and void formation at the
wetting flow front. Parallel flow measurements, on the other hand, can be carried out
under steady state conditions Lundström et al. (2000) and edge effects can be avoided
6
Lundström et al. (2002). Hence there are by all means several ways to determine the
overall permeability of a porous material.

As Reynolds number (Re) increases over 1-10, the value is depending on material
properties such as porosity and packing (quadratic, hexagonal or other), it is well-known
that experimental data starts to deviate from Darcy’s law. The overall effect is that the
pressure drop becomes higher than what are predicted by equations 2.1-2 due to inertia
and turbulence and an additional term is introduced in the so called Forchheimer equation

m
1 'p Q §Q·
 K  b¨ ¸ (2.3)
P L A © A¹

where b is a property of the porous media and m is a measure of the influence of fluid
inertia. The equation was later modified by Ergun by fittings to experimental data
according to

2
Q §Q·
P U¨ ¸
'p
g 150
1  H 2 A  1.75 1  H  © A ¹ (2.4)
L H3 Dp
2
H3 Dp

where H is the fractional void volume in the bed and Dp is the effective diameter of
particles Ergun (1952). The Ergun equation does not have any micromechanical basis and
the geometrical parameters shaping the form of the equation are therefore unknown.
Recent studies indicate that the parameters of the Ergun equation should be modified for
certain geometries Papathanassiou et al. (2001) and Nemec and Levec (2005) which was
also confirmed by a study by Hellström and Lundström (2006).

7
3. INTERNAL EROSION PROCESSES
The soil fractions that are considered as most susceptible to erosion are relatively uniform
coarse silt and fine sand. Cohesive soils as clays are more resistant to erosion as long as the
chemical bonds are not destroyed, Srbulov (1988). It seems like some core materials of
glacial origin can be particularly susceptible to internal erosion. Glacial till is of special
interest since it is a common core material in Swedish dams. Erosion susceptibility of
glacial till is described in e.g. Foster et al. (2000a), Ravaska (1997), Norstedt and Nilsson
(1997) as well as Sherard (1979) and discussed further in Section 4.1.

According to Fell et al. (2005), in their excellent textbook about geotechnical engineering
of dams, four conditions must exist for internal erosion and piping to occur. These are: ”1)
there must be a seepage flow path and a source of water; 2) there must be erodible
material within the flow path and this material must be carried by the seepage flow; 3)
there must be an unprotected exit, from which the eroded material may escape and 4) for
a pipe to form, the material being piped, or the material directly above, must be able to
form and support ´roof´ for the pipe”.

In a central core earth and rockfill dam there are mainly three processes (Fell et al., 2005)
which can initiate piping: backward erosion, concentrated leak and suffusion. Backward
erosion is initiated at the exit point of seepage and the erosion is gradually progressing
backward forming a pipe. Concentrated leak initiates a crack or a soft zone emanating
from the source of water to an exit point. Erosion gradually continues along the walls of
the erosion hole intensifying the concentrated leak. Suffusion is the process where the fine
particles of the soil wash out or erode through the voids formed by the coarser particles.
This can be prevented if the soil has a well graded particle size distribution with
sufficiently small voids. Soils are called internally instable if suffusion takes place and
internally stable if particles are not eroding under seepage flow.

Piping can occur in the embankment, through the foundation and from the embankment
into the foundation. Conceptual models for development of failure by piping for these
three cases are presented in Figure 3.1 to Figure 3.3. In addition, a failure path diagram for
failure by piping through the embankment is shown in Figure 3.4. Similar failure path
diagrams for failure by piping through the foundation and from the embankment into the
foundation can be found in Fell et al. (2005) or Foster (1999). External erosion problems
are described in e.g. Skoglund and Solvik (1995).

9
Figure 3.1 Conceptual model for development of failure by piping in the embankment:
(a) backward erosion and (b) concentrated leak (Foster, 1999).

Figure 3.2 Conceptual model for development of failure by piping in the foundation
(Foster, 1999).

10
Figure 3.3 Conceptual model for development of failure by piping from the embankment into
the foundation (Foster, 1999).

Figure 3.4 Failure path diagram for failure by piping through the embankment (Foster, 1999).

3.1. Uneven settlements and hydraulic fracturing

Embankment dams are subjected to deformations as compression, extension and shear
distortion. Normally, embankment dams are constructed in zones of different materials.
Each zone has an important function for the structure as a whole. Different
compressibilities of the various zones might lead to internal redistribution of stresses
compared to the distribution in a homogeneous embankment. Deformations in a dam
might lead to cracks or “soft zones” where internal erosion can be initiated. Fill materials
have in general low tensile strength, so cracking can result from even very small tensile
stresses. As will be seen below, there is often various types of uneven settlement that is the
main cause for cracking. Even if the deformations are of too low magnitude and character

11
as to cause cracking immediately, cracking may occur later on by hydraulic fracturing.
Drying and shrinkage can also cause low internal embankment pressures, Sherard (1986).

In Kjærnsli et al. (1992), results from measurements of displacements and stresses in dams
are presented for some case studies. All the dams included have been sufficiently well
instrumented and monitored to give a full account of developments within the dam during
construction, the simultaneous or subsequent impoundment as well as during operation.
Measurements in dams with central earth core point out that the greatest deformations,
and the corresponding redistribution of stresses, take place during the construction of the
dam and not during the first filling of the reservoir. Impoundment is, however, the main
reason for transverse displacement of the core. Longitudinal displacements tend to be
directed towards the centre of the valley bottom implying that horizontal compressional
stresses increase in the central regions and decrease towards the abutments. Even if most of
the settlements arise during the construction phase of a dam, settlements will continue for
many years after the dam is completed. To maintain freeboard during the operation
period, the crest of the dam is commonly built a little bit higher than the design level (a
camber is provided). Detailed information about construction settlement and post
construction settlement magnitudes in embankment materials can be found in e.g. Fell et
al. (2005).

Figure 3.5 to 3.10 present several schematic examples of differential settlement induced
cracks in the embankment: transverse differential settlement cracks, longitudinal differential
settlement cracks, internal differential settlement cracks and potential cracks between the
dam fill and the spillway wall. If internal erosion through the core zone is initiated by a
crack in the transversal direction, it may create a problem to the safety of the dam.
Longitudinal cracks oriented in a foremost vertical direction in a dam, do not have as
obvious effect on leakage as transverse cracks. However, in some cases, longitudinal cracks
can be the route of leakage if these are connected with transverse cracks. Causes of
cracking in embankment dams can be further studied in e.g. Kim et al. (2004), Fell et al.
(2005), Zhang and Du (1997), Ng and Small (1999), Sherard (1986) as well as in Lo and
Kaniaru (1990).

In Figure 3.5 two common types of transverse differential settlement cracks are shown.
The cracks are located at the dam crest. Since the deformations are directed towards the
centre of the valley bottom, horizontal compression stresses increase in the middle regions
and decrease closer to the abutments, where extension and cracking might occur.
Obviously, larger changes in the abutment slope increase the risk for tensile stresses and
cracking in areas above the points where the variations occur, as in Figure 3.5b.

12
Figure 3.5 Ordinary transverse differential settlement cracks.

The core in Figure 3.6 is less compressible than the rest of the embankment and cracks
develop at the boundary between the core and the rest of the fills. Further, in this
situation, the weight of the fills on each side of the core is partly transferred to the stiffer
core, thus increasing the stresses of the core.

Figure 3.6 Longitudinal differential settlement cracks between embankment zones

(Sherard et al., 1963).

Seepage through a permeable natural foundation can be reduced by constructing a cutoff

with low permeability through the permeable material. When the depth of the permeable

13
soil is relatively small, a cutoff trench can be excavated through the permeable layer and
filled with relatively incompressible rolled earth, as in Figure 3.7. The natural foundation
on both sides of the cutoff trench is more compressible than the rolled earth fill. Therefore
the settlements will be larger at the upstream and downstream slopes than at the dam crest
region and cracks as shown in Figure 3.7 might appear.

Figure 3.7 Longitudinal differential settlement cracks over foundation discontinuities

(Sherard et al., 1963).

Figure 3.8 presents three cases for which internal differential settlement cracks can arise. In
the example in Figure 3.8a the core is less stiff than the fills on each side. The result of this
is that the core settles more than the other materials and the weight of the core are
partially transferred to the adjacent zones. The core is thus “hanging” on the more rigid
zones, a phenomenon called arching, and as a consequence of this the vertical stress on
horizontal planes in the core will be reduced compared to homogeneous embankment
conditions. An increase in water pressure can lead to a fracture, as the water pressure lower
the effective stresses in the core even further. If the vertical effective stress becomes low
enough for cracks to develop, cracks on the upstream face of the core might initiate
concentrated leak while cracks on the downstream face of the core might initiate
backward erosion. The case in Figure 3.8b displays tension cracks above a very
compressible clay lens in the foundation. Apparently, it is important to perform careful
geotechnical investigations for such a case. Figure 3.8c shows tension cracks in the soil
near the top of a concrete diaphragm wall illustrating that for dam applications, concrete is
often too rigid. When the soil mass surrounding the wall settles, additional load will be
transferred to the wall by negative skin friction. If the wall is rigid, this can also cause
crushing of the wall and penetration of the wall into the dam fill. One way to overcome
this problem is to use plastic concrete that is easier to deform (Fell et al., 2005).

14
Figure 3.8 Typical internal differential settlement cracks in embankments
(Sherard et al., 1963).

The interface between earthfill and concrete structures is often a potential region for
cracking and internal erosion processes. In Figure 3.9 two similar situations are
demonstrated, where settlement of the dam fill can cause a crack to develop between the
spillway retaining wall and the embankment fill.

15
Figure 3.9 Situations where a crack might form between the dam fill and the spillway wall:
(a) steep foundation adjacent to spillway wall; (b) change in slope of the retaining
wall (Fell et al., 2005).

Differential settlement cracks can expand and develop further by the hydraulic fracturing
process. If differential settlements have lead to “soft zones” rather than cracks, hydraulic
fracturing might later on form cracks in such “soft zones”. Hydraulic fracturing is the
process when water pressure splits a soil with a low permeability, which might happen if
the minor total principal stress at any point within the soil becomes equal to the pore
water pressure at the same level (then the minor effective principal stress reduces to zero).
Fracturing may proceed continuously following roughly the plane of the minor principal
stress. For hydraulic fracturing to take place, the adjacent material must be sufficiently
impervious so that water can wedge its way along opening cracks more rapidly than its
permeates the pores. Therefore, hydraulic fracturing occurs only in rather impervious
materials, such as a dam core of clay or moraine (Sherard, 1986) and not in coarse pervious
materials such as gravel and stone.

In a central core earth and rockfill dam, internal stress redistribution might takes place in
both directions, i.e., by transverse arching of the core between the upstream and
16
downstream shells (Figure 3.8a) and longitudinally between dam sections of different
height (Figures 3.5 and 3.10). A critical zone for potential leaks is over the abutments,
where longitudinal stretching of the dam and hydraulic fracturing are most probable. The
internal embankment stress on the transverse planes marked with a dotted line in Figure
3.10 can be expected to be low and a leak on either of these planes is particularly likely.

Figure 3.10 Planes where hydraulic fracturing is likely to occur (Sherard, 1986)

For a high central core dam, with rockfill or gravel shells, the conditions for stress transfer
and hydraulic fracturing are commonly more severe than in relatively low homogenous
dams. The reason for this is that the impervious section, the core, is relatively thinner and
the total settlements are greater (Sherard, 1986). It should be noticed that a thin vertical
core is more sensitive to fracture than a sloping core where the weight of the upstream
rockfill shell helps to keep the core in compression.

Zhang and Du (1997) studied the effects of abutment slopes on the performance of high
central core earth and rockfill dams by three-dimensional nonlinear finite element
modelling. It was e.g. concluded that: ”(1) As a result of the longitudinal soil arching, the
steeper an abutment slope is, the more susceptible to hydraulic fracturing the core will be.
(2) Large changes in abutment slopes will result in large tensile stresses in the core contact
areas and therefore should be avoided. (3) Application of a 5 m thick high-plasticity clay in
the core contact areas may markedly decrease the tensile stresses near the interface, and
reduce the risk of contact cracking because of its good strain hardening performance after
yielding. (4) The compaction of the core is shown most effectively in improving the stress-
strain conditions in the core. The risk of hydraulic fracturing can be greatly alleviated if
the gradation and compaction of the core material are properly carried out.”

It was for a long time not completely accepted in the profession that concentrated leaks
can develop through dams by hydraulic fracturing. This could be because hydraulic
fractures cannot be visually observed since they are inside the dam, erosion processes

17
destroys evidence of them and different interested parties, normally, have different vested
interests in the result of dam incident investigations (Sherard, 1986). In the paper by
Sherard (1986) evidence is gathered to support the conclusion that concentrated leaks by
hydraulic fracturing occur frequently without being noticed, even in dams not exposed to
large differential settlement. The evidence is presented in the following main categories:
”(1) concentrated leaks which appear soon after the first reservoir filling through well-
constructed homogeneous dams with no internal filter or drain, sometimes leading to
breaching failure; (2) erosive leaks through central core dams with inadequate filters,
usually occurring soon after the first relatively rapid reservoir filling; (3) records of well
constructed and designed central core dams in which nearly full reservoir pressure was
measured in piezometers at the downstream face of the core, and in which exploratory
borings in the dam made to study this condition showed conclusively that near horizontal,
water-filled cracks existed in the core; (4) FEM calculations showing that only relatively
small differential settlements will easily create stress conditions in most dams under
ordinary conditions which will allow hydraulic fracturing to occur; and (5) the discovery
of ´wet seams´ inside impervious dam sections, with water content higher than can be
accounted for reasonably by any other mechanism than the entry of water into an open
crack.” Different scenarios that might take place in an embankment dam after initial leaks
have developed by hydraulic fracturing are also thoroughly described in the paper by
Sherard (1986). Notice, however, that the mechanical background to hydraulic fracturing
implies that the initiation of cracks by hydraulic fracturing has nothing to do with the
erodibility of the embankment material and that the development of initial leaks is not
limited to dams with inadequate filters.

In numerical software, it is obviously vital to properly model hydraulic fracturing since this
phenomenon might initiate the erosion processes. Lo and Kaniaru (1990) derived an
analytical expression for prediction of the hydraulic fracturing pressure

uf C cos I  1  sin I V t 2  1  sin I V H (3.1)

where uf is the hydraulic fracturing pressure, Vt is the tensile strength, VH the total minor
principal stress, C is the cohesion and I is the angle of internal friction. The expression in
equation 3.1 was also evaluated against experimental hydraulic fracturing tests. It was
found that the hydraulic fracturing pressure depends on the type of test performed and
good agreement against the theoretical result was obtained only for one type of test;
saturated-consolidated hydraulic fracturing tests. In numerical modelling, constitutive
relations have to be general, therefore equation 3.1 should not be implemented without

18
modifications; it might, for instance, be appropriate to try to reformulate the expression as
a function of effective stresses instead of total stresses.

3.2. Methods to examine and measure internal erosion

The statistics of dam failures and accidents reveal that it is essential to establish systems for
monitoring and surveillance of dams. Fell et al. (2005) states, ”it is apparent that many
accidents would have become failures if they had not been detected by monitoring and
surveillance and some action taken”. The amount of monitoring and surveillance that is
necessary depends on a number of factors, such as: the consequences of a failure, the type
of dam and its size, known deficiencies, the age of the dam, the reservoir level etc. (Fell et
al., 2005). It is especially important to have sufficient inspection during first filling of the
reservoir and when the reservoir level is in the vicinity of historic high levels. Some
common warning signs during piping incidents in embankment dams are listed in Fell et
al. (2005) as: increase in leakage, muddy leakage, sinkholes, settlements, cracking,
whirlpool in reservoir, increase in pore pressure and sand boils. According to Fell et al.
(2005), the most important detection methods for identification of deterioration of
embankment dams are, in order of efficiency: inspection by trained observers, seepage
flow measurements, vertical and horizontal surface displacement measurements, pore
pressure measurements, reservoir water level and rainfall measurements (as these influence
the other factors) as well as internal deformation measurements.

Regular visual inspections are most important in a dam surveillance program. Historically,
damage has mainly been detected by direct observation. This is also the case for Swedish
dams, see Nilsson (1995). By visual inspection of the dam and its surrounding,
deteriorations that are not easily detected by instruments (fissures, leakages, dam spots etc.)
might be found. The main drawback with visual inspection to detect internal erosion is
that the erosion process might have progressed much too far when it becomes visual.

Seepage flow information, by visual surveillance or measurements, is one of the best

indicators of the effectiveness of the impervious elements of a dam. If the seepage water
contains dispersed soil particles, which can be noticed by discolouration of the water,
immediate attention is required since it might indicate an internal erosion process. To
measure the quantity of seepage, special seepage collection and measuring systems can be
built in to a dam, see e.g. Fell et al. (2005) or Kjærnsli et al. (1992).

Results from surface displacement measurements of the dam embankment can be a useful
indication for developing problems that might be associated with internal erosion, such as:

19
local larger settlements, slope instability etc. In order to measure horizontal as well as
vertical displacements a network of markers is installed on the surface of the dam. To
determine the coordinates of the markers by triangulation reference pillars are placed on
firm ground on the sides of the dam, Kjærnsli et al. (1992).

Measurements of pore pressures in the embankment as well as the foundation of a dam can
provide important information for stability analyses, detect potential ”blow-up” conditions
in the foundation and observe unusual seepage pressures. Different types of instruments
can be utilised to measure this pressures (Fell et al., 2005), as: observation wells,
Casagrande type piezometers, hydraulic piezometers, pneumatic piezometers and vibrating
wire piezometers. Much engineering skill is needed to select appropriate types of
instruments and design where to install them to obtain an overall picture of the pore
pressures. Normally, it is not practically or economically feasible to install the instruments
so closely together that the pore pressure measurements can be expected to capture local
deviations or impairments in the dam, Kjærnsli et al. (1992). In that case, pore pressure
measurements will obviously not be a reliable precursor to internal erosion and piping.

Internal displacements and deformations are often monitored on larger embankment dams.
This can give information about long-term deformations as well as probable arching effects
and extensional strains that might cause cracking followed by internal erosion. An
overview of the types of instruments that might be used can be found in e.g. Fell et al.
(2005) or Kjærnsli et al. (1992). Vertical displacements can be measured by installing a
telescopic tube system with cross-arm anchors or magnetic markers. Lateral movements
can be measured by inclinometers. An inclinometer can be installed at any inclination.
Displacements resulting from extension can be measured by extensometers, which
normally are installed horizontally in the longitudinal direction of the dam.

Installation of instruments for monitoring and surveillance into the dam core results in a
potential weakness, since internal erosion and piping might initiate along the instruments
installed. The mechanisms for internal erosion are similar as for piping along a conduit,
discussed in Fell et al. (2005). From internal erosion and piping perspectives it is more
dangerous to make penetrations through the core in the transverse direction. In order to
form an opinion if instruments should be installed in the core or not, the value of the
information gathered must be weighted against the risk to damage the core. Some senior
dam engineers, as the authors in the reference Fell et al. (2005), seems to be generally
reluctant to bring any potential weaknesses into a dam core.

In Fell et al. (2003) the capability of standard methods of surveillance and monitoring to
detect internal erosion and piping is discussed. There point of view in the matter might be
20
best summarized by the following quotation from the paper: ”Many dam owners rely
heavily on surveillance, and monitoring of pore pressures and seepage to warn of potential
internal erosion and piping problems for many older dams which were not provided with
filters designed and constructed to control internal erosion. It has been the authors
perception that many may have placed an over reliance on monitoring and surveillance
rather than carrying out remedial works, given the potential for rapid failures, and the
likelihood that some (or most) modes of initiation of internal erosion may not be easily
detectable. Detection may however be possible once erosion has progressed, provided the
monitoring is well designed, and read frequently enough.”

A potential problem that should be addressed in this context is that standard instruments
for monitoring and surveillance normally have a shorter life length than the dam structure
itself. If instruments cease to function, they must be replaced, which is a process that might
cause damage to the dam during drilling and backfill grouting operations or the dam
owner must go on without the information.

An interesting complement to standard detection methods is geophysical methods, e.g.

temperature measurements, resistivity measurements, streaming potential measurements,
acoustic methods, seismic methods and electromagnetic methods. Many geophysical
methods can be classified as non-destructive (or just cause minor intrusions) and, when
that is not the case, existing standpipes might be possible to utilise for the measurements.
Geophysical methods have the theoretical potential to register small changes in the seepage
flow rate, globally, along the entire dam and thus detect internal erosion at an early stage.
The essential objective for the measurements is to detect changes and the absolute
accuracy becomes less important as compared to the relative accuracy. In some cases,
geophysical methods are most suitable for regular measurements since seasonal variations
can be considered then, single measurements at investigations might be difficult to evaluate
if too many of the variables are unknown or uncertain (Johansson, 1997). Further, it is
advantageous to use different geophysical methods together as far as possible in order to
improve the quality and reliability of the evaluation, Johansson et al. (2004).

Still much research and development are needed to be able to trust on the results from
geophysical methods if they are to be utilised for permanent monitoring and surveillance
of dam structures, Elforsk (2001). For example, it is essential to understand more about the
accuracy of which geophysical measuring techniques can be expected to capture
measurable parameters related to internal erosion (e.g. density, seepage flow, hydraulic
conductivity, temperature, seismic velocity, dielectricity and resistivity) and how such
measurable parameters are affected by the mechanical processes of internal erosion, see e.g.
Johansson (1997) and Elforsk (2001).
21
Notice, when internal erosion processes can be successfully modelled numerically and the
behaviour is included in general software, such software might be utilised to study how
sensitive measurable parameters in geophysical methods are to internal erosion processes in
a dam structure. In general, it seems like numerical modelling is a great complement to
geophysical methods in order to detect internal erosion at an early stage. A vision can even
be to use measurements and numerical tools in-situ for direct control of the dam safety.

22
4. EMBANKMENT DAM FAILURES AND
ACCIDENTS
It is often difficult to precisely determine the cause to a dam accident or failure. Several
types of processes might be involved in an accident and multiple modes in a failure.
Internal erosion has as previously mentioned a tendency to destroy evidence of initial leaks
which might have existed. As Sherard (1986) noticed, ”another common problem is that
different interested parties have different vested interests in the outcome of the
investigations. It is natural that the designer desires to show that the design was not at fault,
that the engineer who controlled the construction wants to show that the construction
supervision was done carefully, and that the owner wants to establish that someone is
clearly responsible, etc.” This might obstruct the investigation process and lead to that
definite conclusions, agreed upon by all parties, and are not reached. However, Peck
(1980) states that a failure of a dam is seldom the consequence of a single shortcoming.
”Usually there is at least one other defect or deficiency, and the failure occur where two
or more coincide. This inference supports the principle of designing to provide defence in
depth, the ´belt and suspenders´ principle long advocated by Arthur Casagrande. It
postulates that if any defensive element in the dam or its foundation should fail to serve its
function, there must be one or more additional defensive measures to take its place.” Still
more research in the fields of internal erosion processes as well as dam monitoring and
surveillance might provide the knowledge necessary for a better understanding of the
reasons for dam incidents.

A statistical analysis of embankment dam failures and accidents is thoroughly presented in

Foster et al. (2000a), based on a large database of embankment dam incidents as well as a
population database for embankment dams. The study is specifically focusing on internal
erosion and slope stability problems. The result provides useful insights into the factors
which contribute to dam incidents.

In Table 4.1 overall statistics of failure for different failure modes is presented. The
historical average frequency of failure is estimated to be 1.2 % over the life of the dam.
Failure by overtopping and piping are the most common modes of failure while failure by
slides is less common. In the case of piping failure, the incidence of piping through the
embankment is two times higher than piping through the foundation and twenty times
higher than piping from the embankment into the foundation. Downstream slide failures
are more common than upstream ones. Most of the failures occur during operation of the
dam. Further, it was noticed in Foster et al. (2000a) that about half of all piping failures
through the embankment are associated with the presence of conduits. The different
23
modes of piping associated with conduits are: piping into the conduit, along and above the
conduit or out of the conduit (Fell et al., 2005).

Foster et al. (2000a) studied the factors affecting the frequency of incidents of: piping
through the embankment, piping through the foundation, piping from the embankment
into the foundation, downstream sliding and upstream sliding. All types of incidents were
examined with regard to dam characteristics as e.g. dam zoning, year of construction, dam
heights, filters, core soil types, compaction, location of conduits, foundation geology and
foundation cutoff. The paper contains a great number of important observations and is
highly recommended for further reading.

In Foster et al. (2000b) the statistical data presented in Foster et al. (2000a) is utilized to
estimate the relative likelihood of failure of embankment dams by internal erosion and
piping. The method described in Foster et al. (2000b) is called the University of New
South Wales (UNSW) method. In the method ”the likelihood of failure of a dam by
piping is estimated by adjusting the historical frequency of piping failure by weighting
factors which take into account the dam zoning, filters, age of the dam, core soil types,
compaction, foundation geology, dam performance, and monitoring and surveillance”.
The UNSW method ”is only suitable for preliminary assessments, as a ranking method for
portfolio risk assessments to identify which dams to prioritise for more detailed studies”.
The sequence of events leading to failure by piping can be divided into: initiation,
continuation, progression and breach. In the UNSW method it is not possible to assess
what influence each of the weighting factors has on the different events. It is
recommended in Foster et al. (2000b) that an event tree method is used to obtain a better
understanding of how each of the factors influences initiation, continuation, progression
and breach. Use of event trees to estimate the probability of failure of embankment dams
by internal erosion and piping is discussed in Foster and Fell (2000). The recommendation
is to compare results from event tree analyses with results from the UNSW method to find
out if there are large differences. If this is the case, the reasons to it should be examined
further.

The time to develop failure and the warning signs which may be evident prior to failure
are important factors on assessing whether actions to prevent failure are possible or what
warning time will be available for evacuation of the population at risk downstream of the
dam. In Fell et al. (2003) ”a method is presented for the approximate estimation of the
time for progression of internal erosion and piping, and development of a breach leading
to failure in embankment dams and their foundations. The method accounts for the nature
of the soils in the dam core, the foundation, and the materials in the downstream zone of
the dam. Guidance is also provided on the detectability of internal erosion and piping,
24
taking account of the mechanism of initiation, continuation, and progression to form a
breach, for internal erosion and piping in the embankment, the foundation and from the
embankment to foundation”. The development of piping is often a very fast process,
leaving not much time for proper actions. The case studies of piping failures and accidents
included in Foster et al. (2000b) revealed that ”in the majority of failures, breaching of the
dam occurred within 12 h from initial visual indication of piping developing, and in many
cases this took less than 6 h”. About two thirds of the failures occurred on first filling or in
the first 5 years of operation. For the piping accidents it was found that ”the emergency
situation often lasted several days, with piping reaching a limiting condition, allowing
sufficient time to draw the reservoir down or carry out remedial works to prevent
breaching”.

Fell et al. (2003) noticed ”that in most cases it has not been possible to identify the time of
initiation of erosion, and the first signs of erosion tend to be at the progression phase, with
a concentrated, often muddy leak”. The problem to detect that internal erosion has been
initiated relates to the typical processes of initiation: backward erosion, concentrated leak
and suffusion. Fell et al. (2003) points out that backward erosion ”would not necessarily be
expected to be preceded by large increases in seepage during the time the erosion is
gradually working back from the downstream exit point. When the erosion has progressed
to within a short distance of the reservoir/foundation interface, it breaks through rapidly
or very rapidly”. Concentrated leak piping ”could be expected to initiate very rapidly or
rapidly once the reservoir level reaches the critical level at which erosion begins in cracks
or high permeability zones; or the critical level at which hydraulic fracture initiates”.
Suffusion ”is likely to be a more slowly developing process, accompanied by more gradual
increases in seepage, and changes in pore pressure with time”. Further, it is much to learn
about dam failures and accidents by reading case histories, see e.g. Vogel (2004), Watts et
al. (2002) and Von Thun (1996).

25
Table 4.1 Overall failure statistics for large embankment dams up to 1986, excluding dams constructed in Japan
pre-1930 and in China (Foster et al., 2000a).
4.1. Internal erosion susceptibility of glacial till
Foster et al. (2000a) have studied if the geological origin of the core has any influence on
the incidence of piping. It was concluded ”that dams with core materials of glacial origin
have experienced more piping accidents (i.e., initiation of piping) than those built from
other materials but have experienced fewer failures”. In many of the incidents have
sinkholes appeared on the crests and slopes of the embankment dams. Milligan (1976)
discusses geotechnical aspects of glacial tills and states that ”because of the dense and
impermeable nature of many tills, the term ´till´ is generally considered to represent an
´ideal´ material for the foundations of heavy structures and for rolled earth fill as in dams.
The heterogeneous nature of till deposits in situ and their wide variation in engineering
properties, make till, on occasion, far from a ´ideal´ material.” According to Sherard
(1979) has the troubles occurred in dams comprised of remarkably similar coarse, broadly
graded soils that are frequently of glacial origin, within the typical gradation shown in
Figure 4.1. Milligan (2003) mentions that ”the characteristic feature of these glacial tills is
that they are reasonably similar in gradation throughout the world, but the character of the
fines content depends on the source rock”. Case studies of dam incidents could be found
in e.g. Milligan (2003), Norstedt and Nilsson (1997) and Sherard (1986). Many completely
successful dam cores have been constructed of coarse grained, broadly graded glacial soils.
Problems seem to develop only under unusual circumstances.

Figure 4.1 Problem soils according to Sherard (Milligan, 2003).

A number of explanations to the internal erosion susceptibility of glacial till can be found
in the literature. Sherard (1979) states that the problems ”can be explained by the fact that

27
the broadly graded coarse soils are internally unstable”. The self filtering of the base soil
does not work satisfactory and the fine particles will erode through the coarse particles.
The general process leading to this type of incidents was described by Sherard (1979) as
follows: ”a concentrated leak developed through the core which caused a type of internal
erosion in which the soil fines are eroded selectively and carried out of the core, leaving
the coarse sand and gravel particles behind to act as a pervious drain. The volume of fine
material eroded was larger than the volume of the void spaces between the coarser soil
particles causing progressive collapse of the material above the initial leakage channel,
which action finally reached the dam surface as manifested by the sinkhole or crater.”
Observe that this implies that the coarser particles are floating in a matrix of fines, which is
common for glacial moraine (Sherard and Dunnigan, 1989), and the fines are carrying the
embankment compressive stresses. In addition, it was noticed by Sherard (1979) that
”probably sinkholes in the downstream shell are more hazardous than in the upstream shell
if the dam has a central core, because the sinkhole could remove lateral support for the
core allowing it to move locally downstream under the water pressure”. Milligan (2003)
means that ”many tills, except when badly gap-graded, are highly stable to seepage in situ”
and explains that ”internal erosion (piping) of broadly graded till core materials may result
more often as a consequence of segregation during placement”. Foster et al. (2000a) writes
that ”it is possible that these glacial soils are more erodible because their fine silt and clay
size fractions are finely ground rock, rather than more common clay minerals”. Ravaska
(1997) obtained test results that pointed out that ”the higher the specific surface area is the
better the capability of the till to resist erosion”. Milligan (2003) states that ”most of the
incidents, not exclusively, apparently occur in tills of lower plasticity”. The term plasticity
is a measure of the ability of a soil to undergo unrecoverable deformation without cracking
and is therefore also important for the ability of the soil to resist internal erosion.

28
5. FILTER DESIGN
It is widely accepted that the safety of large embankment dams is mainly dependent on the
reliability of the performance of their critical filters. Filters have two basic functions:
prevent erosion of soil particles from the soil they are protecting (named the base soil) and
allow drainage of seepage water. A summary of the chronological development of filter
design criteria is found in ICOLD (1994). In Fell et al. (2005) a brief review of available
methods for designing filters is presented. Fell et al. (2005) also recommend a modified
method for designing critical filters with flow normal to the filter; the method is based on
Sherard and Dunnigan (1985, 1989) as well as USDA-SCS (1994).

In general, filter design criteria are worked out with the base soils grouped under various
categories (e.g. four categories) based on their fines content, and a different filtering
criterion is given for each base soil category. The quantity D15F (particle size of the filter
material for which 15 % by weight is finer) is supposed to reflect the particle capture
ability of the filter material. The basic idea is to design the particle size distribution of the
filter so that the voids in the filter are sufficient small to prevent erosion of the base soil
and the void size in the filter is controlled by the finer particles, for which D15F is a
measure (Sherard et al., 1984a). In most filter design methods, the quantity D85B (particle
size of the base material for which 85 % by weight is finer) represents the particle size and
the erosion parameter of the base soil. No sufficiently adequate physical explanation for
the use of D85B was found in the literature during the work with this literature survey (see
B

e.g. Sherard et al., 1984a as well as 1984b and Sherard & Dunnigan, 1989). As an example
of a filtering criterion, it is prescribed in the recommended method in Fell et al. (2005)
that for base soils of sand and gravel with less than 15 % material finer than 0.075 mm, the
filter quantity D15F must be smaller or equal to 4×D85B (valid for base soil after regrading).
B

The filter should be permeable enough for the seepage flow to pass through it without
building up high pore pressures. This is normally accounted for in filter design by
prescribing a smallest allowed value of the quotient D15F/D15B (e.g. D15F/D15B> 4 or 5
B

ensures a much higher permeability of the filter than of the base soil) and restrict the
amount of fines in the filter. Low fines content is also desirable in order to avoid cracks in
the filter. Internal instability or suffusion in the filter material should be avoided, a number
of different criteria to handle this are proposed based on the analysis of the grain size
distribution, see e.g. Fell et al. (2005). Segregation at placement of the filter material
should be prevented, e.g. by specifying a maximum particle size. The issue of segregation
is of high importance in the context of internal erosion, and discussed further in Section
5.1.

29
The current filter design criteria are mostly empirical, based on statistical correlations of
experimental observations. These criteria are constantly being questioned and revised as
long as new experimental data becomes available. Inherent in the design methods are for
example the assumptions that percent fines content reflect the erodibility of the base soil,
D85B reflects the erosion parameters of the base soil and D15F reflects the particle capture
capacity of the filter material. According to Reddi and Kakuturu (2004) have these
assumptions not been validated thus far. On the other hand, Sherard and Dunnigan (1989)
express themselves like this: ”the filter boundary D15F determined in the no erosion filter
(NEF) test for each given impervious soil is a unique value, which can be considered a
fundamental property of the soil in the same sense that the Atterberg limits or the effective
shear strength envelope parameters are fundamental properties of impervious soil”.
Anyway, it feels like more research in the field of filter design would be valuable since it is
so important, and then with a physical sound mechanical approach instead of constantly
applying empirical methods. A recent alternative filter criterion is proposed by Indraratna
et al. (2007), based on a mathematical procedure to determine the controlling constriction
(opening) size. The main idea is that it is the constriction sizes rather than the particle sizes
that influence filtration. Constriction sizes have been studied before by e.g. Kenney et al.
(1985).

If filters in existing dams do not satisfy the modern filter criteria there is always a potential
of internal erosion. The filter performance mainly affects the continuation of erosion.
There are many dams built with filters not meeting the accepted filter criteria. To
exemplify, a study of case histories of piping incidents, discussed in Foster and Fell (2001),
pointed out that the dams included in the study were ”generally constructed from the
1960s to 1970s, which coincides with a period when there was a trend away from the use
of uniformly graded multiple filters and towards the use of a single filter of substantial
width and broad gradation”. The filters in the study had wide gradings and low
proportions of sand sizes, which can make them susceptible to segregation and potentially
internally unstable. Dams with filters that do not fulfil modern design criteria rarely fail but
damage to the dam in the form of sinkholes and large leakage is rather common. Foster
and Fell (2001) presents a method for assessing embankment dam filters that do not satisfy
design criteria. ”The method can be used to determine whether filters that are coarser than
required by modern filter design criteria will eventually seal or experience continuing
erosion leading to possible failure of the dam in the event piping initiates.” The method is
empirical, based on laboratory test results and characteristics of dams that have experienced
piping incidents. An interesting outcome of this study is that it seems like it is a
considerable margin between no-erosion and continuing erosion criteria for most soils;
except for fine base soils where the margin is much less (Fell et al., 2005). This might help
to explain why self-healing so frequently occur at dam incidents. Empirical filter design
30
criteria do not normally take into account the variability in the particle size distributions of
the base soil and filter. Mínguez et al. (2006) present a new general method for assessing
the probability of fulfilling any empirical filter design criteria considering base soil and filter
heterogeneity by means of first-order reliability methods. The method can be used by
engineers to estimate the safety of existing filters in terms of probability of fulfilling their
design criteria.

More research would be desirable in the field of ageing effects of dams and uncertainties of
core/filter appearance after accidents, incidents etc. This is a difficult research subject since
it is not easy to ”look inside” a dam without damage it; detection with installations and
geophysical methods might not be sufficient so the authors´ of this literature survey suggest
numerical modelling as a complementary approach. Results from the no erosion filter tests
reported by Sherard and Dunnigan (1989) gave interesting information about the soil
behaviour at the interface between the base specimen and the filter. This result might at
least partly reflect the real behaviour in a dam. The laboratory test setup of the no erosion
filter test is shown in Figure 5.1. Series of this type of tests were made and a filter
boundary size Db15F was found between successful and unsuccessful tests for each base soil
in the study, where ”successful” was defined as no visible erosion of the walls of the
preformed hole in the base specimen and ”unsuccessful” when the erosion was visible.

Figure 5.1 The no erosion filter test (Sherard & Dunnigan, 1989).
31
The condition observed at the end of successful tests on fine silt or clay as base soil (that
have more than 85 % by weight of particles finer than the 0.075 mm sieve) is presented in
Figure 5.2. The seal of the filter did not take place only at the small area at the end of the
preformed leakage hole. The water flowing in the leakage hole turned 90Ý at the filter face
and progressively eroded out radially from the hole, creating a thin slot and sealing the
filter face over the entire diameter. In unsuccessful tests, for filters with D15F sizes up to
two to five times larger than Db15F, essentially the same thing happened; a thin slot
appeared and the filter face was finally sealed, however, the preformed hole was enlarged
since eroded material from the walls of the initial leakage channel was needed for the self
healing.

Figure 5.2 Condition at end of successful test on fine silts and clays
(Sherard & Dunnigan, 1989).

The condition at the end observed on silty and clayey sands with low fine contents (15 %
or less by weight finer than the 0.075 mm sieve) as base soil in a successful no erosion filter
test is fundamentally different from the behaviour of the clays and silts previously
discussed, see Figure 5.3. For the sandy soils with low contents of fines no eroded slot
developed at the bottom of the base specimen. It was believed that the main reason for this
was that the coarser particles are in contact and form a granular skeleton with high
intergranular friction forces that resist movement of particles by erosion. In the
unsuccessful tests on silty and clayey sands a large crater immediately developed through
the base specimen.

32
Figure 5.3 Condition at end of successful test on silty and clayey sands with low fine contents
(Sherard & Dunnigan, 1989).

Reddi and Kakuturu (2004) proposed a mathematical method to analyse the self healing
nature of concentrated leaks at core-filter interfaces. The relative degree of self healing of
various filters for a given base soil can be studied with the method. An experimental
method was also developed to determine the erosion properties of the base soil and obtain
input values to the parameters of the mathematical model. It was noticed that the entire
particle size distribution of the filter governs particle accumulation at the interface between
the base soil and the filter. ”It is suggested that a wide range of particle accumulation
shapes (ranging from a rapid reduction in particle accumulation to a relatively uniform
particle accumulation) is possible for soils when only D15F of the filter is specified.”

5.1. Segregation of filter materials

It has been noticed that dams with good filter performance have filters with characteristics
that would tend to make them less susceptible to segregation, Foster and Fell (2001).
Segregation is the process that causes separation of a graded soil material into finer and
coarser zones. It might occur at placement of material by downslope discharge during
dumping and spreading operations. The larger particles have a tendency to accumulate at
the bottom of the slope. All broadly graded materials that do not have sufficient fines to
provide some degree of cohesion are difficult to place without segregation. If a segregated
zone of sufficient coarseness arises in the filter there is a risk for loss of fines from the core
by internal erosion. Since it is the large soil particles that are most sensitive to segregation,
it is especially important to restrict the maximum particle size included in a filter. Milligan
(2003) warns that the issue of segregation is a critical uncertainty, which is generally poorly

33
recognized and dealt with in dam engineering. In many projects the problem with
segregation is handled by a declarative statement in the specifications telling that “materials
must be placed without any segregation taking place”. This is, however, in many practical
cases very difficult to accomplish and the specifications thereby places an obligation on the
contractor that cannot be met.

Kenney and Westland (1993) performed rotary-drum segregation tests to study the process
of segregation in different granular soils, including granular filter materials. It was found
that: 1) all dry soils consisting of sands and gravels will readily segregate; 2) the patterns of
segregation are repeatable in repeated tests and the amount of segregation might be
predictable; 3) water effectively prevents segregation of finer soils (e.g. sandy soils) while it
has no influence on the segregation behaviour for coarser soils (e.g. gravels). Water in
unsaturated soils reduces segregation because the sorting mechanism is suppressed by the
increased effective stresses caused by the capillary tensions acting at points of contact
between wetted particles. Soils with a high content of small particles are most strongly
influenced by these effects of capillary stresses. In Figure 5.4, an approximate gradation
limit for wetting effects is presented. Further, Kenney and Westland (1993) carried out
tests at different water content and the results showed that the ability of water to inhibit
segregation was not changed even at water contents equal to one half of the ”field
capacity” values.

Figure 5.4 Approximate gradation limit for preventing segregation by wetting

(Milligan, 2003).

34
Milligan (2003) concludes that ”much of the uncertainty related to appropriate filter
design and placement may be largely resolved by specifying the use of wide, narrowly
graded, sand-rich filters, of a gradation finer than the limit suggested in Figure 5.4 and
placed wet in thin lifts. Then the specification clause that ´segregation will not be
permitted´ is a practical reality.”

35
6. MICROMECHANICAL MODELS OF HYDRO
DYNAMICAL LOADS ON SINGLE PARTICLES
AND SYSTEM OF PARTICLES
Continuum formulations of flow through porous media often results in equations
consisting of a few unknown parameters such as the permeability. The physical
background of such parameters can often be traced to the detailed flow in the pores and it
is therefore in place to study the flow on this level, as well. To start with we can scrutinize
models for single particles. The viscous drag force on a single sphere in an infinite fluid at
Re < 1 is, for instance, described by Stokes law:

f 6SRPv (6.1)

where R is the radius of the sphere and v the velocity of it. This expression was early
extended by Oseen to be valid for Re up to about 10 according to Lamb (1932):

§ 3 ·
f 6SRPv¨1  Re ¸ (6.2)
© 16 ¹

and later on by Happel and Brenner (1957) to be valid for the case of a sphere moving
through a tube with diameter D to be:

§ R·
f 6SRPv¨1  2.1 ¸ . (6.3)
© D¹

The drag force has also been derived for particles with other geometries than spherical. For
slender rods the drag force per unit length along the rod is, Emersleben (1925):

f 4SPv . (6.4)

While the following expression has been derived for flow perpendicular to the rod, Lamb
(1932):

4S
f Pv . (6.5)
2  ln(Re)

37
The obvious conclusion is that the drag on a particle is linearly dependent on the averaged
velocity as long as Re is less than unity. The importance of velocity then increases to 2nd
order.

For very dense systems and at low Re the flow follows Darcy’s law and the porous media
can be characterised by their permeability through measurements or by the usage of
permeability models based on, for instance, the detailed rod geometry: cf. Speilman and
Goren (1968), Bear (1972), Scheidegger (1972), Sangani and Acrivos (1982), Jackson and
James (1986), Dullien (1992), Wang et al. (1994) and Higdon and Ford (1996). Generally
permeability models may be divided into the following models, phenomenological,
conduit flow (Kozeny-Carman for instance), empirical, network, deterministic (directly
from Navier-Stokes) and models based on flow around submerged objects (models
proposed by Brenner and Brinkman, for instance) Dullien (1992).

The permeability is often represented by the Kozeny-Carman relation, according to:

1 H3
K (6.6)
B S 2 1  H 2

where H is the porosity, B a constant and S the specific particle surface. The Kozeny-
Carman relation is based on a tube-like representation of a porous medium. The constant
B has to be determined experimentally, but has never been determined explicitly, Williams
et al. (1974). The Kozeny-Carman relation assumes an isotropic porous medium. For rods
other equations have been derived that distinguish between the permeability along and
perpendicular to the direction of the rods, according to:

8 H3
K R2 (6.7)
c 1  H  2

5/ 2
§ H ·
KA C ¨¨ min  1¸¸ R2 (6.8)
© H ¹

where R is the radius of the rods and c and C are constants depending on the arrangement
of the rods while Hmin denotes the minimum pore volume for this arrangement. Also in this
case forces on individual particles can be calculated. Mei and Auriault (1991) derived, for
instance, the average drag acting on a particle in a porous medium with fore-aft symmetry
at small Reynolds number to be:
38
 f Pv k 0  k 2 Re 2  (6.9)

for a two-dimensional array. Here k0 and k2 are non-dimensional constants to be

determined which is, for instance, done for both ordered and random arrays of spheres by
Hill et al. (2001). To summarize the forces on individual particles have been exploited for
certain geometries and for a number of flow conditions. We however need to investigate
further higher Reynolds number flows, more complex geometries and instationary
conditions.

39
7. MICROMECHANICAL MODELS FOR FRICTION
BETWEEN SINGLE GRAINS AND SYSTEM OF
GRAINS FOR NON-COHESIVE SOILS
The forces outlined in the previous chapter must be balanced by gravity and forces that
emanates from particle interactions for the dam to be stable. As a first, and most simple,
criteria the size distribution of the particles being subjected to hydrodynamic pressure is
compared to the pore size distribution of the porous medium. Hence the particles will
move if they are small enough independent on magnitude of the force on it. In reality,
however, the hydraulic forces must exceed particle interaction forces and or gravitational
forces keeping the particles in place at normal conditions. The force due to gravitation is
given from:

Fg z mgn z (7.1)

if gravity, g is directed along the z-axis and m is the mass of the particle. The type of the
dominating particle interaction force is dependent on the typical size of the particles. For
very small particles, around one micron and smaller cohesion dominates. For larger
particles up to fractions of millimetres adhesion becomes the major mechanism for particle
interaction while for even larger particles, that is in the range of one millimetre and larger
friction forces are major mechanisms for the particle interaction. The static frictional force
that is directly related to the normal forces on the particle is directed in the opposite
direction to the applied forces and has a magnitude according to:

0 d F fz d P s N (7.2)

as long as the static friction factor, Ps is isotropic and can be treated as a scalar. Once the
drag force on the particle becomes higher than the expression on the right-hand side of
this expression the magnitude of the friction force can be expressed as

F fz Pk N (7.3)

where the kinetic friction factor, Pk often is smaller than Ps. The friction factor in this form
is usually determined by experiments.

41
8. CONCEPT FOR NUMERICAL MODELLING OF
INTERNAL EROSION
In recent years, numerical methods have been used to an increasing degree for solving
more complicated problems in geotechnical engineering. A wide range of problems have
been analysed with the finite element method (FEM), the finite difference method (FDM)
and more recently the boundary element method (BEM). By utilizing e.g. the finite
element method, based on a continuum approach, stresses, strains, deformations and pore
pressures in an embankment dam can be theoretically analysed and it might be possible to
draw conclusions from that information about the initiation of internal erosion processes.
An approach worth considering, when initiation is established, is to interpret internal
erosion numerically as a type of localisation and describe the constitutive behaviour with
micromechanical models in localised zones. The mathematical description of the strain
localisation phenomenon is nowadays an intensive research area (see e.g. Adachi et al.,
1997) and, hopefully, models for strain localisation could give valuable input to the
formulation of internal erosion localisation.

The failure of a soil material is usually accompanied by formation of narrow zones with
highly localised deformation. The development of such localised deformation zones might
cause significant stress redistribution and strength reduction by softening, which can lead
to a progressively developing slip line that induce failure of the entire soil structure. A
typical example in geotechnical engineering is the progressive development of the slip
surface in a slope stability failure. With conventional finite element techniques is it difficult
to capture deformation that is localised to narrow zones. If a very dense element mesh is
used, it would be theoretically possible to capture the localised deformation. However, this
is normally not practicable due to large computational costs. In addition, is the
conventional FEM technique not suitable for describing the kinematics at localised failure
and the constitutive laws included might not be representative for the material in the
localised zone. There exist a lot of different strategies for numerical modelling of
localisation, Tano (2001). In e.g. discrete crack formulations the shear bands are following
the boundaries of the elements and a remeshing algorithm is necessary, when the path is
not known in advance, in order to be able to get cracks in the proper directions. In
formulations based on element-embedded discontinuities, e.g. the inner softening band
method described by Tano (2001), remeshing is normally avoided and relatively large
elements can be used.

The choice of a suitable strain localisation model for reformulation in to an internal

erosion localisation model might not be an easy task. For example Tano (2001) write the
43
following sentences about strain localisation models. ”Many methods have been developed
and their popularity is switching continuously. No method seems to lead to correct results
for all types of localisation analyses. Some of the methods are mathematically elegant but
may be impractical in large scale modelling due to the requirement of very dense finite
element meshes.”

One way forward is to develop a numerical model for internal erosion processes based on
micromechanical soil properties and implement it mathematically as a type of localisation
in e.g. finite element software. Besides the ordinary application for analysis of deformation
and stability in geotechnical engineering projects, such a software could be useful for:
increasing the knowledge about internal erosion processes, evaluating the risk for dam
incidents caused by internal erosion, estimating the time for progression of internal erosion
and piping, studying self-healing of leaks, changes in filter behaviour subsequent to particle
accumulation and ageing effects in dams, analysing the amount of instrumentation needed
in a dam and the proper location for monitoring and surveillance, designing dams etc. In
Figure 8.1, numerical modelling of backward erosion treated as a localisation in the finite
element method is illustrated.

Figure 8.1 Numerical modelling of internal erosion.

A fundamental understanding of the physical processes involved in internal erosion is

required to be able to successfully formulate models for the constitutive behaviour in
localised zones. Theoretical formulations of a model for internal erosion should be
complemented by performing laboratory tests. In the theoretical formulations, the
following matters should be paid attention to:

x Performing a literature survey of strain localisation models that could be useful as a

mathematical base for micromechanical internal erosion localisation models.

44
x Examining which physical quantities that are of importance and should be included
in the internal erosion model.
x Formulating a micromechanical model for internal erosion (or utilising an existing,
see Section 6 and 7).

It is of considerable importance that the localisation model, the micromechanical model

and the finite element model are mathematically consistent with each other. Otherwise,
the whole exercise of developing models that should be used for practical purposes is
utterly pointless. Appropriate laboratory tests on internal erosion must be designed and
tests performed. The test result is expected to give useful information about internal
erosion processes as well as important physical quantities. It is desirable that foremost
measurable quantities are included as parameters in the models. Internal erosion models
should be based on test results and the type of tests needed is influenced of the type of
models chosen, so it is very important that the theoretical work and the practical
laboratory work are performed simultaneously and interactive.

When the internal erosion model is developed and included in software it is essential to
evaluate its behaviour. Since dams are complex structures it is not advisable to evaluate the
model directly on a dam; some simpler structure would be preferable. The evaluation of
the model may be carried out in two simplified steps before real dams are studied. First,
laboratory tests performed, according to the discussion above, could be numerically
simulated and the numerical result could be compared to the laboratory test result. It
would be advantageous if this kind of evaluation could be performed continuously, if
possible, during the development of the model. As a second step, small scale internal
erosion tests ought to be performed on model dams. These tests will be numerically
simulated and numerical and laboratory test results compared. An example of a small scale
test on a model dam at Vattenfall Research and Development in Älvkarleby is shown in
Figure 8.2.

45
Figure 8.2 Model test at Älvkarleby.

46
ACKNOWLEDGEMENT
The literature survey presented here was carried out within the framework of “Swedish
Hydropower Centre – SVC”. SVC has been established by the Swedish Energy Agency,
Elforsk and Svenska Kraftnät together with Luleå University of Technology, The Royal
Institute of Technology, Chalmers University of Technology and Uppsala University.

Alstom Hydro Sweden, CarlBro, E.ON Vattenkraft Sverige, Fortum Generation, GE

Energy (Sweden), Jämtkraft, Jönköping Energi, Mälareenergi, Skellefteå Kraft,
Sollefteåforsens, Statoil Lubricants, Sweco VBB, Sweco Energuide, SweMin, Tekniska
Verken i Linköping, Vattenfall Research and Development and Vattenfall Vattenkraft,
Waplans, VG Power and Öresundskraft are also participating in SVC.

47
REFERENCES
1. Adachi, T., Oka, F. & Yashima, A. (Eds.). (1997), Localisation and bifurcation theory for
soils and rocks. Balkema, Rotterdam, ISBN 9058090043.
2. Adams, K. L., Russel, W. B. & Rebenfeld, L. (1988), Radial penetration of a viscous
liquid into a planar anisotropic porous media. International Journal of Multiphase Flow,
14 (2): 203-215.
3. Ahn, S. H., Lee, W. I. & Springer, G. S. (1995), Measurement of the Three-
Dimensional Permeability of Fibre Preforms Using Embedded Fiber Optic Sensors,
Journal of Composite Materials, 29 (6): 714-733.
4. Bear, J. (1972), Dynamics of fluids in porous media, Dover Publications Inc., ISBN: 0-
486-65675-6.
5. Day, R. A., Hight, D. W. & Potts D. M. (1998), Finite element analysis of
construction stability of Thika Dam. Computers and Geotechnics, 23(4), 205-219.
6. Dullien, F. A. L. (1992), Porous media: Fluid transport and pore structure, Academic press,
ISBN: 0-12-223651-3.
7. ELFORSK. (1999). Dammsäkerhet, Comparative evaluation of CFX and
PHOENICS codes in modelling flow in porous media, James Yang & Bengt
Hemström. Elforsk report 99:49, (in English).
8. ELFORSK. (2001). Dammsäkerhet, Scientific workshop on internal erosion
monitoring 13-15 June 2000, Edited by M. Bergman and S. Johansson. Elforsk report
01:21, (in English).
9. ELFORSK. (2005). Dammsäkerhet, Dam-break project in Norway, CFD evaluation
of test results, Karin Eriksson & James Yang. Elforsk report 05:02, (in English).
10. Emersleben, O. (1925). Das Darcysche filtergesetz. Physik. Zeitschr., 26, 601-610.
11. Ergun, S. (1952). Fluid flow through packed columns. Chemical Engineering Progress,
48(2), 89-94.
12. Fell, R., MacGregor, P., Stapledon, D. & Bell, G. (2005), Geotechnical engineering of
dams. Balkema, Leiden, ISBN 041536440x.
13. Fell, R., Wan, C. F., Cyganiewicz, J. & Foster, M. (2003), Time for development of
internal erosion and piping in embankment dams. Journal of Geotechnical and
Geoenvironmental Engineering, 129(4), 307-314.
14. Foster, M. A. (1999). The probability of failure of embankment dams by internal erosion and
piping. PhD Thesis. School of Civil and Environmental Engineering, The University
of New South Wales, Sydney.
15. Foster, M. & Fell, R. (2000), Use of event trees to estimate the probability of failure
of embankment dams by internal erosion and piping. In Proceedings of the 20th ICOLD

49
 Congress on Large Dams, Beijing, Vol. 1, Q76, R16, 237-260, International
Commission on Large Dams, Paris.
16. Foster, M. & Fell, R. (2001), Assessing embankment dam filters that do not satisfy
design criteria. Journal of Geotechnical and Geoenvironmental Engineering, 127(5), 398-
407.
17. Foster, M., Fell, R. & Spannagle, M. (2000a), The statistics of embankment dam
failures and accidents. Canadian Geotechnical Journal, 37(5), 1000-1024.
18. Foster, M., Fell, R. & Spannagle, M. (2000b), A method for assessing the relative
likelihood of failure of embankment dams by piping. Canadian Geotechnical Journal,
37(5), 1025-1061.
19. Gebart, B. R. (1992), Permeability of unidirectional reinforcements for RTM, Journal
of Composite Materials, 26: 1100-1133.
20. Gebart, B. R. & Lidström, P. (1996), Measurement of in-plane permeability of
anisotropic fiber reinforcements, Polymer Composites, 17 (1): 43-51.
21. Happel, J. & Brenner, H. (1957), Low Reynolds number hydrodynamics, ISBN: 90-247-
2877-0.
22. Hellström, J. G. I. & Lundström, T. S. (2006), Flow through Porous Media at
Moderate Reynolds Number, Proc. 4th Int. Scientific Colloquium - Modelling for Material
Processing, Riga, pp. 129-134.
23. Higdon, J. J. L. & Ford, G. D. (1996), Permeability of three-dimensional models of
fibrous porous media, Journal of Fluid Mechanics, 308: 341-361.
24. Hill, R. J., Koch, D. L. & Ladd, A. J. C. (2001), The first effects of fluid inertia on
flows in ordered and random arrays of spheres, Journal of Fluid Mechanics, 448: 213-
241.
25. Huang, T. K. (1996), Stability analysis of an earth dam under steady state seepage.
Computers & Structures, 58(6), 1075-1082.
26. ICOLD (1994), Embankment dams granular filters and drains. International Commission
on Large Dams, Paris, Bulletin 95.
27. Indraratna, B., Raut, A. K. & Khabbaz, H. (2007), Constriction-based retention
criterion for granular filter design. Journal of Geotechnical and Geoenvironmental
Engineering, 133(3), 266-276.
28. Jackson, G. W. & James, D. F. (1986), The permeability of fibrous porous media, The
Canadian Journal of Chemical Engineering, 64: 364-374.
29. Johansson, S. (1997), Seepage monitoring in embankment dams. Doctoral Thesis, Royal
Institute of Technology, Stockholm, ISBN 91-7170-792-1.
30. Johansson, S., Dahlin, T., Friborg, J. & Sjödahl, P. (2004), Leakage detection ability
using resistivity, self potential, temperature, and induced polarization measurements –
experiences from field tests. ICOLD 72nd Annual Meeting, Seoul, 16-22 May 2004, In

50
 Proceedings of the Workshop on Dam Safety Problems and Solutions-Sharing Experience.
Korea National Committee on Large Dams, W1-06-A053.
31. Kenney, T. C., Chahal, R., Chiu, E., Ofoegbu, G. I., Omange, G. N. & Ume, C. A.
(1985), Controlling constriction sizes of granular filters. Canadian Geotechnical Journal,
22(1), 32-43.
32. Kenney, T. C. & Westland, J. (1993), Laboratory study of segregation of granular
filter materials. Filters in Geotechnical and Hydraulic Engineering (Eds: Brauns, Heibaum
& Schuler). Balkema, Rotterdam, pp. 313-319, ISBN 9054103426.
33. Kershaw, T. N. (1972), The three dimensions of water flow in press felts, Tappi
Journal, 55 (6): 880-887.
34. Kim, H., Ahn, S. R. & Shin, E. W. (2004), A Study on the causes of cracking on fill
dam and the remedial measures. ICOLD 72nd Annual Meeting, Seoul, 16-22 May
2004, In Proceedings of the Workshop on Dam Safety Problems and Solutions-Sharing
Experience. Korea National Committee on Large Dams, W1-08-A077.
35. Kjærnsli, B., Valstad, T. & Höeg, K. (1992), Rockfill Dams. In the series Hydropower
Development, volume 10, Norwegian Institute of Technology, Trondheim, ISBN 82-
7598-014-3.
36. Lamb, H. (1932), Hydrodynamics, ISBN0-521-05515-6.
37. Lekakou, C., Johari, M. A. K., Norman, D. & Bader, M. G. (1996), Measurement
techniques and effects on in-plane permeability of woven cloths in resin transfer
moulding, Composites: Part A, 27A: 401-408.
38. Lo, K. Y. & Kaniaru, K. (1990), Hydraulic fracture in earth and rock-fill dams.
Canadian Geotechnical Journal, 27(4), 496-506.
39. Lundström, T. S. (2000), The permeability of non-crimp-stitched fabrics, Composites:
Part A, 31: 1345-1353.
40. Lundström, T. S., Gebart, B. R. & Sandlund, E. (1999), In-plane permeability
measurements on fibre reinforcements by the multi-cavity parallel flow technique,
Polymer Composites, 20 (1): 146-154.
41. Lundström, T. S., Stenberg, R. S., Bergström, R., Partanen, H. & Birkeland, P-A.
(2000), In-plane permeability measurements: A Nordic round-robin study,
Composites: Part A, 31: 29-43.
42. Lundström, T. S., Toll, S. & Håkanson, J. (2002), Measurement of the permeability
tensor of compressed fibre beds, To be published in Transport in Porous Media.
43. Mellah, R., Auvinet, G. & Masrouri, F. (2000), Stochastic finite element method
applied to non-linear analysis of embankments, Probabilistic Engineering Mechanics, (15):
251-259.
44. Mei, C. C. & Auriault, J. -L. (1991), Effect of Weak Inertia on Flow Through a
Porous Medium, J. Fluid Mech., 222, 647-663.

51
45. Milligan, V. (1976), Geotechnical aspects of glacial tills. Spec. Publ. No. 12, Royal
Society of Canada, Ottawa, 269-291.
46. Milligan, V. (2003), Some uncertainties in embankment dam engineering. Journal of
Geotechnical and Geoenvironmental Engineering, 129(9), 785-797.
47. Mínguez, R., Delgado, F., Escuder, I. & G. de Membrillera, M. (2006), Reliability
assessment of granular filters in embankment dams. Int. j. numer. anal. meth. geomech.,
30(10), 1019-1037.
48. Muir Wood, D. (2004), Geotechnical modelling. Spon Press, Oxfordshire, ISBN 0-419-
23730-5.
49. Nemec, D. & Levec, L. (2005). Flow through packed bed reactors: 1. Single phase
flow. Chemical Engineering Science, 60, 6947-6957.
50. Ng, A. K. L. & Small, J. C. (1999), A case study of hydraulic fracturing using finite
element methods. Canadian Geotechnical Journal, 36(5), 861-875.
51. Nilsson, Å. (1995), Åldersförändringar i fyllningsdammar. VASO Dammkommitte´s
rapport nr 16, Stockholm ISSN 1400-7827.
52. Norstedt, U. & Nilsson, Å. (1997), Internal erosion and ageing in some of the
Swedish earth and rockfill dams. Proc. 19th Int. Congress on Large Dams, Florence,
Q73, R20, 307-319. International Commission on Large Dams, Paris.
53. Papathanasiou, T. D. (1996), A structure-orientated micromechanical model for
viscous flow through square arrays of fibres, Composites Science and Technology, 56:
1055-1069.
54. Papathanasiou, T. D., Markicevic, B. & Dendy, D. (2001). A computational
evaluation of the Ergun and Forchheimer equations for fibrous media. Physics of
Fluids, 13(10), 2795-2804.
55. Parnas, R. S., Howard, J. G., Luce, T. L. & Advani, S. G. (1995), Permeability
characterization. Part 1: A proposed standard reference fabric for permeability, Polymer
Composites, 16: 429-445.
56. Parnas, R. S. & Salem, A. J. (1993), A comparison of the unidirectional and radial in-
plane flow of fluids through woven composite reinforcements, Polymer Composites,
14(5): 383-394.
57. Peck, R. B. (1980), ”Where has all the judgement gone?”. Canadian Geotechnical
Journal, 17(4), 584-590.
58. Pikulik, I. I., Gilbert, D., McDonald, J.D. & Henderson, J. R. (1991), A new
instrument for measuring the permeability of paper-machine clothing, Tappi Journal,
74 (4): 169-176.
59. Potts, D. M. & Zdravkovi´c L. (1999), Finite element analysis in geotechnical engineering:
theory. Thomas Telford, London, ISBN 0727727532.
60. Potts, D. M. & Zdravkovi´c L. (2001), Finite element analysis in geotechnical engineering:
application. Thomas Telford, London, ISBN 0727727834.
52
61. Ravaska, O. (1997), Piping susceptibility of glacial till. Proc. 19th Int. Congress on Large
Dams, Florence, Q73, R30, 455-471. International Commission on Large Dams,
Paris.
62. Reddi, L. N. & Kakuturu, S. P. (2004), Self-healing of concentrated leaks at core-
filter interfaces in earth dams. Geotechnical Testing Journal, 27(1), 1-10.
63. Sangani, A. S. & Acrivos, A. (1982), Slow flow past periodic arrays of cylinders with
application to heat transfer, International journal of Multiphase flow, 8 (3): 193-206.
64. Scheidegger, A. E. (1972), The physics of flow through porous media, University of
Toronto Press, ISBN: 978-0802018496.
65. Sharif, N. H., Wiberg, N. E. & Levenstam, M. (2001), Free surface flow through
rock-fill dams analyzed by FEM with level set approach. Computational Mechanics,
27(3), 233-243.
66. Sheng-Hong, C. (1996), Adaptive FEM analysis for two-dimensional unconfined
seepage problems, Journal of hydrodynamics, Ser: B, (1), 60-66.
67. Sherard, J. L. (1979), Sinkholes in dams of coarse, broadly graded soils. 13th Int.
Congress on Large Dams, New Delhi, Q49, R2, 25-35. International Commission on
Large Dams, Paris.
68. Sherard, J. L. (1986), Hydraulic fracturing in embankment dams. Journal of Geotechnical
Engineering, 112(10), 905-927.
69. Sherard, J. L. & Dunnigan, L. P. (1985), Filters and leakage control, in embankment
dams. In Seepage and Leakage from Dams and Impoundments. ASCE Geotechnical
Engineering Division Conference, 1-30.
70. Sherard, J. L. & Dunnigan, L. P. (1989), Critical filters for impervious soils. Journal of
Geotechnical Engineering, 115(7), 927-947.
71. Sherard, J. L., Dunnigan, L. P. & Talbot, J. R. (1984a), Basic properties of sand and
gravel filters. Journal of Geotechnical Engineering, 110(6), 684-700.
72. Sherard, J. L., Dunnigan, L. P. & Talbot, J. R. (1984b), Filters for silts and clays.
Journal of Geotechnical Engineering, 110(6), 701-718.
73. Sherard, J. L., Woodward, R. J., Gizienski, S. F. & Clevenger, W. A. (1963), Earth
and earth rock dams. John Wiley & Sons.
74. Skoglund, M. & Solvik, Ø. (1995), External and internal erosion in rockfill dams.
International Journal on Hydropower & Dams, 2(3), 44-47.
75. Speilman, L. & Goren, S. L. (1968), Model for predicting pressure drop and filtration
efficiency in fibrous media, Environmental Science and Technology, 3: 279-287.
76. Srbulov, M. (1988), Estimation of soil internal erosion potential. Computers and
Geotechnics, 6, 265-276.
77. Tano, R. (2001), Modelling of localized failure with emphasis on band paths. Ph.D. Thesis
2001:08, Luleå University of Technology, Luleå.

53
78. Tucker III, C. L. & Dessenberger, R. B. (1994), Governing equations for flow and
heat transfer in stationary fiber beds, Flow and Rheology in Polymer Composites, edited
by Advani, S. G., Elsevier Science B.V., ISBN: 0-444-89347-4.
79. USDA-SCS (1994), Gradation design of sand and gravel filters. United States Department
of Agriculture, Soil Conservation Service, Part 633, Chapter 26, National
Engineering Handbook.
80. Vogel, A. (2004), Lessons from dam failures caused by seepage or internal erosion.
ICOLD 72nd Annual Meeting, Seoul, 16-22 May 2004, In Proceedings of the Workshop on
Dam Safety Problems and Solutions-Sharing Experience. Korea National Committee on
Large Dams, W1-24-A078.
81. Von Thun, J. L. (1996), Understanding seepage and piping failures – the No. 1 dam
safety problem in the west. In Proceedings of the Association of Dam Safety Officials
(ASDSO) Western Regional Conference, Lake Tahoe, Nevada, pp. 3-21.
82. Wang, T. J., Wu, C. H. & Lee, L. J. (1994), In-plane permeability measurement and
analysis in liquid composite molding, Polymer Composites, 15 (4): 278-288.
83. Watts, R., Burk, K., McLaren, M., Wolfe, J. & Zender, K. (2002), Failure of the
Teton dam: geotechnical aspects. International Water Power and Dam Construction,
54(7), 30-31.
84. Weitzenböck, J. R., Shenoi, R. A. & Wilson, P. A. (1998), Measurement of three-
dimensional permeability, Composites Part A, 29 A: 159-169.
85. Williams, J. G. Morris, C. E. M. & Ennis, B. C. (1974), Liquid flow through aligned
fiber beds, Polymer Engineering and Science, 14 (6): 413-419.
86. Wörman, A. & Xu, S. (2001), Stochastic analysis of internal erosion in soil structures
– implications for risk assessments. Journal of Hydraulic Engineering, 127(5), 419-428.
87. Zhang, L. & Du, J. (1997), Effects of abutment slopes on the performance of high
rockfill dams. Canadian Geotechnical Journal, 34(4), 489-497.
88. Young, W. B. & Wu, S.F. (1995), Permeability measurement of bidirectional woven
glass fibers, Journal of Reinforced Plastics and Composites, 14 (October): 1108-1119.

54