Download as pdf or txt
Download as pdf or txt
You are on page 1of 63

Proceedings of the 20th ICSMGE-State of the Art and Invited Lectures– Rahman and Jaksa (Eds)

© 2022 Australian Geomechanics Society, Sydney, Australia, ISBN 978-0-9946261-6-5

Energy geo-engineering
Geotechnics of mine tailings: a 2022 State of the Art
Geo-ingenierie pour l’energie
Géotechnique des résidus miniers: État de L’Art 2022
J. Carlos Santamarina
Energy Resources and Petroleum Engineering, KAUST, Saudi Arabia, carlos.santamarina@kaust.edu.sa
Andy Fourie
School of
Adnan Engineering, University of Western Australia, Australia, andy.fourie@uwa.edu.au
Aftab
WASM: Minerals, Energy, and Chemical Engineering, Curtin University, Australia
Ramon Verdugo
D. Nicolas
CMGI, Espinoza
Santiago, Chile, rverdugo@cmgi.cl
Petroleum and Geosystems Engineering, U. Texas at Austin, USA

Annika Bjelkevik
Maurice Dusseault
Earth and
Tailings Environmental
Consultants Sciences,
Scandinavia U. Sweden
AB, Waterloo, Canada

Antonio Gens
Luis and
Civil Alberto Torres-Cruz
Environmental Engineering, U. Politecnica de Catalunya - CIMNE, Spain
School of Civil and Environmental Engineering, University of the Witwatersrand, South Africa
Hussein Hoteit
Energy Resources
Dobroslav and Petroleum Engineering, KAUST, Saudi Arabia
Znidarcic
Department of Civil, Environmental and Architectural Engineering, University of Colorado Boulder, USA
Seunghee Kim
Civil and Environmental Engineering, U. Nebraska-Lincoln, USA

Joo Yong Lee


ABSTRACT: Global interest in the geotechnical stability of tailings storage facilities (TSFs) has burgeoned over the past decade.
In particular,
Climate the widely
Change accessible
Response videoKorea
Division, of the Institute
catastrophic failure of the and
of Geoscience Feijão TSF inResources,
Mineral Brazil in 2019 was a definite
Republic hinge point.
of Korea
Enormous efforts are now focussed on the assessment and improvement of existing facilities as well as the design of new ones. A
Liang
focus ofLeithis paper has been on problems associated with contractive and brittle tailings, how to determine liquefaction
susceptibility,
School and whetherWestlake
of Engineering, currently U.,
available
Chinamonitoring technologies are able to provide meaningful and timely information on a
potential, developing failure. We also believe there are other, important aspects of tailings behaviour that must be understood, and
thus the paper
Guillermo also deals with the consolidation, desiccation and creep behaviour of highly compressible tailings, the importance of
Narsilio
correctly accounting
Department for seepage,
Infrastructure Eng.,andU.the analysis and
Melbourne, design for seismic loading conditions. Although new management
Australia
technologies such as filtered tailings are seeing increasing use, we contend that comprehensive understanding of tailings
geotechnics remains
Jean-Michel paramount to ensuring the stability of all tailings facilities.
Pereira
Navier, Ecole des Ponts, Univ Gustave Eiffel, CNRS, France
RÉSUMÉ : L'intérêt mondial sur la stabilité géotechnique des parcs de stockage de résidus miniers (PSRM) a eclaté au cours de la
dernière décennie.
Marcelo Sanchez En particulier, la vidéo largement diffusée de l´effondrement dévastateur du PSRM de Feijão au Brésil en 2019 a
constitué
Civil andun point de convergence
Environmental certain. Texas
Engineering, D'énormes
A&M,efforts
USA sont désormais concentrés sur l'évaluation et l'amélioration des
installations existantes ainsi que sur la conception de nouvelles installations. Cet article s'intéresse aux problèmes liés aux résidus
miniers compressibles
Kenichi Soga et fragiles, à la déterminaison de la susceptibilité à la liquéfaction et la possibilité des actuelles technologies de
surveillance
Civil disponibles à fournir
and Environmental des informations
Engineering, appropriées
UC Berkeley, USAet opportunes sur une potentielle rupture en cours. Il y a, également,
d'autres aspects importants du comportement des résidus qui doivent être appréhendés, et donc le document traite en outre la
consolidation,
María la dessiccation
Victoria Villar et le fluage de résidus hautement compressibles, l'importance de prendre en compte correctement les
infiltrations, ainsi que les analyses et la conception pour des conditions sismiques. Bien que de nouvelles technologies de gestion,
CIEMAT, Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas, Spain
telles que les résidus filtrés, soient de plus en plus utilisées, une compréhension complète de la géotechnique des résidus demeure
primordiale pour assurer la stabilité de tous les parcs à résidus.
Marie Violay
KEYWORDS:
Lab tailings,
Experimental Rock liquefaction,
Mechanics, brittle,
EPFL, consolidation,
Switzerlandseismic.

ABSTRACT: The geo-science and engineering fields have critical roles to play towards a sustainable energy future. This state-of-
1 the-art review focusses on five areas where the geotechnical community
INTRODUCTION has been
Despite involved
the focus theunacceptable
on the most: the oilrate andofgas sectorofwith
failures TSFs
emphasis on methane hydrates, carbon geological storage, geothermal, energy
provided by geo-storage,
ICOLD Bulletin and 121,
nuclearandwaste
manystorage.
subsequentExtensive
studies
tables
1.1 in unacceptable
The the appendix identify potential
rate of failures of geotechnical
Tailings Storagecontributions to(e.g.,
all energy resources
Azam and (mining
Li, 2010), is critically
failures continue needed to supply
to occur. Indeed,rawnot
materials for the energy sector, yet is not addressed in this review).only
Facilities Energy-related
have failuresapplications
continued to involve
occur at a wide range of high
unacceptably time rates,
and
length scales, coupled thermo-hydro-chemo-mechanical processes, but multi-phase fluids, highofpressure-temperature-stress
the consequences many of the more recent failures conditions,have
The ICOLD
fines (International
migration, Commission
reactive fluid transportonand
Large Dams)
phase Bulletin
transformations, and multi-physics
been devastating, repetitive loads. Analysis
both to human life and and design
to the require
downstream
#121 documented
careful some 221
experimentation case records
(including of either
field-scale incidents
studies) at a
and advanced numerical simulations.
environment. Educational
Two particularly programs
catastrophic musthave
failures address the
occurred
tailings storage facility (TSF) or indeed
knowledge needs in energy geo-engineering. outright failure of a TSF. in the past few years in Brazil. In 2015, the Fundão TSF collapsed,
Published in 2001, this bulletin concluded that, “effective releasing some 43 million m3 of tailings, which flowed into the
reduction of the cost of risk and failure can only be achieved by a
Les domaines des géosciences et de l’ingénierie ont unDoce River, eventually
rôle essentiel reaching
à jouer pour the Atlantic
un avenir Ocean
énergétique some 660
durable. Cettekm
commitment from Owners to the adequate and enforced away (Fonseca do Carmo et al, 2017). This failure resulted in 19
revue de l’état de l’art se concentre sur cinq domaines
application of available engineering technology to the design, dans lesquels la communauté géotechnique a été le plus impliquée : le secteur
pétrolier avec fatalities and was followed by detailed forensic investigations that
construction andun focus of
closure surtailings
les hydrates de méthane,
dams and impoundmentsle stockage
over géologique
identified
de dioxyde
theenprimary
de carbone,
causes
la géothermie,
of the failure,
le géostockage
detailscontributions
of which are
d’énergie et le stockage des déchets
the entire period of their operating life”. radioactifs. Des tableaux détaillés donnés annexe identifient les potentielles
discussed later in this paper.

121
95
The Fundao failure resulted in a study commissioned by the stability of the facility. On the other hand, tailings facilities
ICMM (International Council on Mining and Metals) that set out constructed using the downstream method are totally supported
ten principles to be followed to maintain integrity of TSFs and on the foundation soil and tailings only act as a load against the
minimise the risk of catastrophic failure (ICMM, 2020). Despite retaining embankment. The centerline option is an intermediate
this document being adopted by the ICMM (an organisation that structure, with the downstream shoulder resting on the foundation
represents some 27 large multinational mining companies), and soil and supporting the lateral forces generated by the retained
other initiatives that followed the Fundão failure, in January 2019 tailings and water.
the collapse of the Feijão TSF in Brazil resulted in over 260 Although many upstream tailings facilities have performed
fatalities and widespread downstream environmental damage. The satisfactorily for many years and continue to do so, we note that
disturbingly graphic video footage that was published on the analyses of TSF performance records generally identify the
internet has probably done more for the improvement in tailings upstream construction method as the least reliable one (Islam and
management practice than any single previous initiative. It clearly Murakami 2021, Franks et al. 2021). We believe that this result,
demonstrated the extreme brittleness of the tailings retained in this coupled with the previous observation that upstream dams rely on
TSF and the speed with which the resulting flowslide occurred. adequate strength development in the impounded tailings,
The Feijão failure led to the establishment of an independent, warrants the claim that upstream construction carries implicitly
multi-stakeholder review, and, finally in August 2020, to the higher risks than other methods. Indeed, upstream TSFs do require
publication of the GISTM (Global Industry Standard on Tailings much greater oversight and attention to measurable performance
Management). The GISTM provides a blueprint for a step change than downstream facilities.
in the way TSFs are designed, constructed and operated and is
essential reading for anyone working in the field.
Aside from the two catastrophic TSF failures discussed above,
that occurred in Brazil, the past decade has seen more than thirty
TSF accidents or failures (https://www.wise-
uranium.org/mdaf.html). Notable among these are the failures that
occurred in Canada in 2014 at the Mount Polley TSF, and the
Cadia TSF failure that occurred in Australia in 2018, although
neither of these failures resulted in fatalities. In the case of Cadia,
the tailings release was very confined and limited to the mining
lease, whereas the Mount Polley failure resulted in off-lease
release, with flow occurring into the nearby Hazeltine Creek and
Quesnel Lake. The reason these two failures are specifically
flagged here is that they both occurred in well-regulated mining
environments. Mount Polley is in British Columbia, which has
one of the best regulated tailings management regimes in the
world, and Cadia is in New South Wales, also a well-regulated
and mature mining state. These two failures, while not resulting
in the same quantum of devastation as the Brazilian TSF failures,
certainly required a major re-think of the level of understanding
of key geotechnical principles within the tailings industry. How
could such significant failures occur in jurisdictions with a mature
and technically enlightened regulator, with TSFs designed by
competent and skilled geotechnical engineers? Reasons for this
situation are many and varied, and many of these are discussed in
this paper. It is recognised at the outset that we will not be able to
cover all pertinent tailings related geotechnical topics in this
single paper but believe a state-of-the-art review such as this Figure 1. Conventional TSFs. Common types of dam construction.
provides a much-needed opportunity for reflection and
recalibration of the adequacy of our current understanding of the 1.2 Increasing tailings volumes
engineering behaviour of mine tailings.
Despite various recycling initiatives around the world, demand for
Depending on how tailings are processed, transported,
discharged, and stored, different types of TSFs are generated, most minerals remains almost insatiable. Clearly, mining as a
which have significant variations in terms of cost, volume and human activity is not going away. The volume of tailings
type of materials required for construction, rate of construction, produced annually is increasing significantly, as the grade of
and inherent stability. The current practice of tailings management available ore bodies declines. Robertson (2018) suggested that
is mostly associated with tailings deposits of what may be termed every 30 years the volume of tailings worldwide increases about
the ‘conventional’ type, due to their lower cost and simpler ten-fold, the area covered by TSFs increases five-fold and average
implementation. The conventional system of tailings disposal heights of TSFs doubles. He pointed out that the maximum height
consists of one or several dams that confine and retain the of TSFs in the decade 2001 to 2010 was about 240m, with even
saturated loose, hydraulically deposited tailings (slimes) and higher TSFs planned. Improved understanding of the geotechnical
process water in an impoundment. The retaining component of behaviour of mine tailings will thus continue to be imperative, and
tailings facilities (e.g., the perimeter embankment) are usually to evolve to include the effect of increasingly high confining
constructed using the sandy fraction of the tailings, which may be stresses on tailings characteristics. Furthermore, based on data
accomplished using cyclones or, more often, conventional disclosed by mining houses upon the request of institutional
earthmoving equipment. According to the construction procedure investors, it is estimated that 50 billion m3 of tailings will require
used, it is possible to distinguish between upstream, downstream storage during the five-year period extending from 2019 to 2023
and centerline methods of construction (see Figure 1). The (Franks et al. 2021). Tailings storage will remain a necessity for
upstream method of construction requires a minimum volume of the foreseeable future.
coarser tailings (or borrow material) for embankment
construction, but the geotechnical properties of the tailings,
sometimes including the finer fractions, are involved in the overall

122
1.3 Changes in response to catastrophic failures
As already inferred above, we believe the Feijão failure that
occurred in Brazil in 2019 will be seen as an inflection point in
the attention paid to improved tailings management worldwide.
The publication of the GISTM will ensure the focus of owners and
operators of TSFs, government regulators, financiers and
insurance companies are retained on trying to ensure no future
catastrophic failures occur ever again. Having said that, there are
many existing TSFs that have been (and unfortunately some still
are) poorly managed historically and the potential for failure of
such deposits cannot be ruled out, unless placed under appropriate
operational management.
The GISTM covers many topics aimed at vastly improving
tailings management, many of them non-technical. However,
embodied in the guiding ‘Requirements’ is one that we believe
will have (and indeed already is having) a profound influence on
the way many TSFs are operated. Associated increases in cost,
sometimes substantial, are going to be unavoidable. The
Requirement we refer to is Requirement 4.6 which states,
“Identify and address brittle failure modes with conservative
design criteria, independent of trigger mechanisms, to minimise
their impact on the performance of the tailings facility” (the
italicised words are according to the GISTM). Unfortunately,
many TSFs contain extensive zones of potentially brittle tailings,
a topic we discuss in more detail in section 8.1.1. Ideally, going
forward TSFs would be operated to prevent the accumulation of
brittle tailings, but sometimes this is extremely difficult to Figure 2 Relationship between solids content and yield stress for tailings,
achieve. Operational practices that can be adopted to produce showing indicative ranges of properties and appropriate dewatering
dilative rather than contractive (and brittle) tailings are touched technologies (after Jewell and Fourie, 2015; courtesy Australian Centre
upon throughout this paper. for Geomechanics).

1.4 Alternative and emerging technologies. 2 THE CONTRAST BETWEEN WATER DAMS AND
TAILINGS STORAGE FACILITIES.
As discussed in ICOLD Bulletin 121, many, if not all these
failures could be largely attributed to the effects of too much Water dams are usually built to store water for power generation,
water. There are additionally many other drivers to reduce water irrigation, water supply etc. Historically, water dams have been
consumption in tailings management, e.g., in Chile costs of up to built since time immemorial. The history of tailings dams is, in
$US4/m3 have been reported for water. contrast, much more recent (of the order of 100 years) and they
Over the past two to three decades, significant advances have are built to store tailings, which contain different proportions of
been made in technologies that utilise less water to transport sand, silt and clay-sized particles, and related process water.
tailings to the TSF. Figure 2, taken from Jewell and Fourie (2015), Tailings dams form an impoundment, and the entire facility is
demonstrates the qualitative effects of significantly enhanced commonly referred to as tailings storage facility (TSF), tailings
dewatering of tailings. Solid’s concentration is plotted against a
management facility (TMF), or more recently (the termed used in
measure of strength, in this case shear yield stress, as measured
the GISTM), a tailings facility. For purposes of continuity with
typically using a vane rheometer, usually reported in Pascals.
Most operational TSFs today would classify as ‘conventional current literature, we use the former term in this paper (TSF). A
tailings’ (indicated by ‘segregating’ in Figure 2), with no TSF may also include structures such as spillways and
measurable yield stress of the as-deposited tailings. Additional clarification ponds. This section provides a perspective of the
thickening, using much deeper thickeners than previously used, differences between water dams and tailings dams.
results in higher solids concentrations and thus higher yield
stresses, requiring much more attention to detail of tailings 2.1 Ongoing construction and increases in height
transportation, as the higher solids concentration may sometimes Water dams are normally constructed as a conventional civil
require the use of positive displacement pumps rather than engineering project, i.e. with a defined start and end. Normally the
conventional centrifugal pumps. There are also implications for construction phase lasts from a year to a few years, depending on
deposition practices when using high-density thickened tailings. the height, length and type of dam etc. Once construction is
Added impetus to the adoption of increased dewatering has been completed, the dam is handed over to the owner to operate.
provided by recommendations arising from some of the forensic Tailings dams are designed for life of mine (LOM) (in addition to
investigations of major TSF failures to minimise storage of water a period of post-closure, which may be extremely long, as
on TSFs. Referring again to Figure 2, further thickening requires discussed in section 11) and commonly built in stages over that
use of filters, producing a cake-like material when successful. The time. Firstly, a starter dam(s) is constructed, generating an initial
production of filtered tailings usually comes at significantly capacity for typically one to three years storage of tailings and
increased cost, particularly operational costs due to increased process water. The dam is then raised in discrete increments at
energy use. regular time intervals, or continuously as during cycloning.
Regardless of whether the dam is raised discretely or
continuously, the rationale is to progressively increase the storage
capacity as the demand requires. Construction is therefore an
ongoing process throughout the whole operating phase. As dam
raises often include tailings as a construction material, operations

123
such as deposition of tailings is often part of construction on a stationary cyclone, several small cyclones on the crest,
continual basis. mobile cyclone or a set of cyclones and they can be
LOM may vary from around approximately 10 years up to 50+ horizontal or vertical.
years and, more importantly, LOM will likely change over time Subaqueous deposition means deposition under water.
as a) exploration continues and more ore is found and b) This method is mainly used when the tailings contain
commodity prices change leading to more tailings to store if prices sulphur and are prone to oxidation. The TSF is then a
rise or less to store, or even an early close of operations, if prices water impoundment where tailings is deposited under
decrease. Due to changes in LOM the final height for a tailings water. As the natural slope of tailings deposited under
dam will change with time, which must be considered in the initial water is much steeper compared to subaerial deposition,
design, i.e., that the final crest elevation may increase. the discharge point must be moved regularly to utilize the
capacity of the impoundment effectively. The density of
2.2 Importance of operational procedures the tailings is also looser than subaerial (above water)
deposition.
Tailings dams require ongoing construction during operation, Thickened and paste deposition means deposition of
when the design engineer and the owner/operator should work in tailings slurry at a higher solids content than conventional
parallel to ensure that design, construction and operation become tailings, where the latter often have only some 10-50%
an integrated ongoing process. Close collaboration, solids by weight. To reach a higher solids content a deep
communication, knowledge and experience sharing become very thickener is used to further remove water. If enough water
important. Operational activities of a tailings dam include is removed, it goes from a slurry to paste (with a
planning and management of deposition of tailings in the consistency like toothpaste). The higher the solids
impoundment as well as management of water (process water and content, the steeper the beach slope and the lower the
precipitation). Depending on the design chosen for a particular propensity for particle segregation. Thickened tailings or
tailings dam, the influence of tailings deposition may become paste is used when water is scarce and/or when a steeper
critical, e.g., upstream construction totally relies on the deposited beach slope can increase the capacity of the facility to
tailings beach immediately upstream of the crest of the dam being more effectively use the land footprint.
a part of the structural zone of the dam. Conversely, a dam using Filtered tailings deposition is taking thickened tailings one
downstream construction is more analogous to a conventional step further as even more water is removed, filtering the
earth- or rockfill dam and is thus not reliant on the strength of the tailings after it has been thickened. Filtered tailings are too
deposited tailings. For the upstream construction method, the dense to be pumped and must be transported to the TSF
density and the state of the deposited tailings in the structural zone using, for example, conveyor belts or trucks. In the TSF
are critical for stability. The tailings in the structural zone should the filtered tailings usually need to be compacted to create
be non-liquefiable (see also Section 5.1.1). a stable structure. Dust problems are potentially an issue
Deposition of tailings also influences how efficiently the and in wet climates, logistics and compaction must be
storage capacity is utilized as deposition affects the in-situ dry planned to not saturate the filtered tailings after
density of the stored tailings. Economically, dry density is placement. The resulting deposit is commonly referred to
critically important, as it directly affects the utilisation of storage as a “dry stack” and since it may not include free water, a
volume (air space) and thus the footprint of a tailings dam. TSF failure is likely to result in more limited damage
Different methods of deposition of tailings are used depending on compared to a conventional TSF containing tailings and
the desired outcome. The following are the most common. See process water. However, a saturated “dry stack” is still
also Figure 3. prone to liquefaction.
Co-deposition is here used as an umbrella term for all
Single discharge deposition, i.e., a single slurry pipeline manner of integrated mine waste management, e.g.
from the process plant to the TSF. The outlet may methods where tailings and other mine waste materials,
occasionally be moved to even out the build-up of tailings. like waste rock, are deposited in the same facility. Other
Spigot deposition is a method where the main slurry terms used include co-placement (placed, but not
pipeline is divided into several outlets (like a sprinkler). transported together) and co-mingling (placed and
Normally the main slurry pipeline has several discharge transported together). Renaming of existing technologies
points of which some are open simultaneously. To get an such as these includes, for example, the terms eco-tails
even beach, uniform thickness of deposited layer and and paste rock.
improve consolidation, spigots are rotated frequently,
which also prevents dusting if the tailings beach above
water is kept wet.
Drop-pipe is an extension of a deposition point, a spigot,
where there is a large height difference between the
discharge point and the beach. Too high a difference in
height usually results in erosion on the beach and possibly
on the upstream face of the dam. Using a drop pipe helps
dissipate the energy in the flow.
Spray bar deposition is a refinement of spigot deposition
as each spigot is in turn divided into several outlets. This
reduces flow and velocity even more, which increases the
amount of tailings particles settling early on the beach,
almost invariably producing a steeper beach.
Cyclone deposition is when coarse and fine particles are
separated using a cyclone. In the cyclone, the combined
action of gravity and centrifugal forces separates the
coarse fraction from the fines and most of the water. The
coarse fraction is then used for dam embankment
construction and the fines deposited further out on the
Figure 3. Examples of different tailings deposition methods. Photographs
beach. Cyclone schemes may consist of one large
courtesy of Boliden, LKAB and FL Schmidt.

124
For tailings dams relying on deposited tailings for stability, i.e., virtually independent; thus, when changes to the design are
where deposited tailings are part of the structural zone of the dam, introduced, a clear understanding of the resulting implications and
a suitable deposition method is important, as is cycle time and operational requirements are crucial.
adequate drainage. Cycle time refers to how often a specific
deposition point is moved and/or how long it takes to do a full 3. APPLICATION OF GEOTECHNICAL ENGINEERING
cycle around the perimeter of the TSF. Deposition for a short time PRINCIPLES – DO TAILINGS DIFFER FROM NATURAL
results in a thin layer of tailings being deposited, which allows for SOILS?
rapid drying and consolidation. For a specific facility, deposition
for a short time means moving on to the next deposition point Tailings, from an engineering point of view, can be considered as
faster and therefore coming back to the same/first point quickly, man-made soils. Tailings derived from different parent ores, or
i.e., a short cycle time for the full facility, which means less time from different mineral extraction processes may be extremely
to consolidate before deposition takes place again. Therefore, to different and may vary from clayey to sandy soils, sometimes
ensure adequate consolidation an ideal would be a short exhibiting properties outside the framework of understanding
deposition time, but a long cycle time, which are contradictory developed in conventional geotechnical engineering (e.g. laterite
requirements. An optimization is necessary. Additionally, there is nickel tailings). However, tailings can usually be best
the challenge of managing the rate of rise (RoR), which refers to characterised and analysed as geotechnical materials. In this
the rate of raising of the beach of deposited tailings. If the RoR is context, it is necessary to point out that the main difference
too high, consolidation will be negatively affected, and stability between soils and other engineering materials (for example,
may be jeopardized. The only opportunity to affect these concrete, steel and wood), is associated with the development of
conditions (deposition time, cycle time and RoR), is the area, or volumetric strains when subjected to shear stresses.
capacity, of the impoundment. Namely, the area needs to be big Tailings, like any un-cemented soil, is essentially a particulate
enough to facilitate a long enough cycle time and low enough medium made up of mineral particles with interstices (or voids)
RoR. The current tendency to minimize the footprint of a TSF can, and therefore undergo volume changes when subjected to shear
in this perspective, have a negative impact on stability and safety. stresses due to the movements between particles (rotation,
To maximize consolidation and retain a low phreatic surface sliding), which interact with the voids, inducing in them either a
within the structural zone of the dam, drains are normally used. reduction or an increase of their volume.
Upstream dams are sometimes referred to as “drained dams”, The volume change of sandy soils was first observed by
emphasising that drainage is important. It is important for stability Reynolds (1885), who showed that dense sands tend to expand,
directly, but also indirectly, as good consolidation improves shear increasing their total volume when subjected to shear stress.
strength and drainage speeds up the consolidation process. Reynolds called this phenomenon "dilatancy". However, it was
Drainage can be designed in many ways; as horizontal layers or only in the 1930s that Casagrande realized the real importance of
finger drains at the bottom of the dam or under the entire structural volumetric deformation in the mechanical response of soils,
zone for the upstream construction method. Alternatively, developing the concept of "critical density or critical void ratio"
horizontal layers or finger drains can be built at intervals at (Casagrande, 1936). Later, Roscoe and his co-workers presented
increasing elevations as the dam is raised. In some cases, vertical a conclusive study proving the concept of critical void ratio and
drains can be used to connect horizontal drains at different extended it to clayey soils (Roscoe et al, 1958). Typical test results
elevations, although this option is not common. The drains can be obtained on 1 mm steel balls using simple shear are shown in
constructed using natural gravel, crushed rock, perforated or Figure 4 in terms of void ratio and horizontal displacement for a
slotted pipes or different bespoke products using various constant vertical stress of 1.41 kg/cm2 (138 kPa). As can be
geosynthetic options. Collected water can be drained passively or observed, the volumetric strain can be either positive or negative,
actively pumped from a collection sump. depending on the initial void ratio and the magnitude of
deformation, but when the ultimate state is achieved, the volume
2.3 Changes to design intent change tends to cease and the soil deforms under a constant-
volume condition reaching the critical void ratio associated with
As the conditions at a particular tailings dam may change over the normal stress under which the test is performed.
time, the design details may likewise need to change. Conditions
that typically may lead to the need for a design change include
changes in production rate (higher rates of tailings deposition),
changes in grind size (almost invariably meaning production of
tailings with a finer particle size distribution than at the start of the
project), and availability of construction material, where lack of
adequate waste rock, for example, may precipitate a switch from
downstream to upstream construction. Changes in mine staff,
operators and designers may also result in changes to the design.
Such changes in themselves are not a problem. However, it is
crucially important that everyone with operational responsibilities
understand the geotechnical principles of the structure and how
changes to design may affect the overall functioning and stability
of the structure. The mantra is: ‘operate according to the design
intent’. As already mentioned, an upstream constructed tailings
dam relies on provision of suitable underdrainage to ensure a low
phreatic surface which in turn assures a well consolidated
structural zone of the deposited tailings beach. Conversely,
downstream construction relies on the strength of the deposited
tailings to a much lesser extent, if at all. A particular tailings dam Figure 4. Results of simple shear tests on 1 mm steel balls with normal
will be somewhere between 100% dependent on management of stress 140 kPa. Republished with permission of ICE Publishing, from
Roscoe et al. (1958); permission conveyed through Copyright Clearance
tailings deposition providing an adequate structural zone to
Center, Inc.

125
initial settlements can be missed unless displacement data
With some limited exceptions, tailings are amenable to the acquisition is automated.
application of theories and frameworks of understanding
developed for natural soils. These exceptions are discussed 3.2 Quantifying particle shape
subsequently. Being essentially a two phase (sometimes three
phase) material, i.e. solid particles in a liquid matrix, the key Approaches to measuring particle size distribution are well
building block of geotechnical engineering, namely the effective developed and widely used, e.g. sieve and hydrometer methods.
stress principle, is usually applicable and relevant. Additionally, Not so for measuring particle shape. There is no universally
conventional geotechnical engineering laboratory and in-situ tests accepted procedure, although one of the most widely used
that aid in the mechanical characterisation of natural soils are techniques is that used by Cho et al., 2006. This method is based
generally useful to characterise tailings. on the chart shown in Figure 5 and is based on two parameters,
the sphericity, S, and the roundness, R.
3.1 So what are the differences?
When differences occur, they are usually attributable to the way
tailings originate. Unlike natural soils, which essentially evolve
through a combined process of weathering, transportation, and
deposition (sometimes the former totally dominating of course, as
in the case of residual soils), tailings are invariably derived from
some form of industrial process. The most common of these
processes include some (or all) of the following: crushing,
grinding, chemical treatment, and hydraulic transport in a pipeline
(occasionally in a channel). The resulting particles are therefore
often significantly more angular than most natural soils.
In addition, tailings are often deposited at void ratios that are
much higher than the void ratios encountered in natural soils
Figure 5. Method of characterising particle shape showing: a) sphericity,
familiar to most geotechnical engineers. Indeed, when deposited
S and roundness, R, and b) measurements used to determine S and R.
as a slurry, tailings undergo sedimentation before a true soil
Republished with permission of ASCE, from Cho et al. (2006); permission
deposit is formed for which the effective stress principle is valid.
conveyed through Copyright Clearance Center, Inc.
Even after such a tailings layer is formed, it continues to undergo
significant volume changes under self-weight stresses and Sphericity, S, characterises the gross shape of a particle and
exhibits highly nonlinear compressibility and large strains. Most reflects the similarity between a particle’s length, height and
of the volume change takes place at low effective stresses and is width. It is generally defined as the ratio of the radius of the largest
associated with a significant reduction in the hydraulic circle inscribed in the particle, ris, to the radius of the smallest
conductivity of the material. It is not unusual that, after circle circumscribed around the particle, rcs.
sedimentation, a tailings layer is formed at a void ratio as high as Roundness, R, which is the conceptual inverse of angularity,
5 (for relatively low plasticity materials) and up to 20 or even 30
represents the scale of a particle’s major surface features (such as
for materials with high plasticity (e.g. phosphatic clays, Abu-
Hejleh et al 1996). Furthermore, volumetric strains that exceed
corners and edges). R is quantified using the equation, =
50%, as the applied effective stress increases from 0.1 kPa to 1
kPa are not rare. A reduction of the hydraulic conductivity by five
or more orders of magnitude is common for the effective stress where ri is the radius of each major surface feature and Nsf is the
range expected in the field (e.g. Abu-Hejleh et al 1996). number of surface features (Figure 5). A limitation of the chart
Considering these significant changes, traditional linear, small depicted in Figure 5 is that it is based on a 2D projection of the
strain consolidation theory is not suitable for such materials and particle and as such is unable to capture 3D aspects of particle
the assumption of a constant coefficient of consolidation and a shape.
constant compression index is not always justified. A nonlinear, Altuhafi et al (2013) provide three shape measures, namely
finite strain consolidation theory that accounts for the effect of convexity, sphericity and aspect ratio, for characterising particles
self-weight of the material is thus essential for a rational approach using digital image analysis (DIA). The method appears more
to analyze the behavior of tailings as they transition from a high robust than the Krumbein and Sloss approach, as the parameter R
water content slurry upon deposition to a soil deposit in which the is somewhat scale dependent, an undesirable feature when
principle of effective stress holds true (e.g. Gibson et al 1967). characterising shape. Recently Yang et al (2019) have also used
Consequently, conventional consolidation tests are not suitable DIA techniques to investigate shapes of tailings particles, with
for determining the consolidation parameters of such tailings and promising results. However, much of the available literature
testing procedures and test analyses consistent with the nonlinear correlating particle shape to engineering characteristics is based
consolidation theory must be used for characterization. Such on the parameters R and S, and this paper is thus restricted to the
experimental procedures have been developed and are available existing literature bank.
for routine use as discussed in section 4.2.2.
We emphasize that many tailings streams, particularly those 3.3 Correlation of sphericity and roundness to geotechnical
derived from hard-rock mining, settle out (sediment) very rapidly properties.
and thus do not necessarily require testing according to the
procedures discussed later in this paper and conventional 3.3.1 Maximum and minimum void ratio values
consolidation testing methods are appropriate. Although tailings There is now overwhelming and consistent evidence that S and R
derived from many hard rock ore deposits may contain correlate strongly with the maximum and minimum void ratios,
appreciable fines contents (more than 20% by mass), they still emax and emin respectively. Both these extreme void ratio values
have high hydraulic conductivity (and thus coefficient of increase with decreasing S and R, i.e. as the particle becomes more
consolidation) values because the fines are non-plastic, being rock angular and irregular (or less rounded). As noted by Santamarina
flour. In fact, some (coarser) tailings have such high coefficients and Cho (2004), for mono-sized spheres the loosest stable packing
of consolidation that using a conventional oedometer may be (emax) that can be achieved is 0.91 and the densest (emin) is 0.34. If
problematic because consolidation takes place so rapidly that a range of different sized spherical particles were present, emin

126
could be even lower. Cho et al (2006) presented data from 16 Where m is the slope of the correlation and is necessarily >1 since
natural sands from around the world, 17 crushed sands from by definition emax > emin. The range of void ratios can thus be
Georgia, USA (being crushed, these sands are likely similar to expressed as:
tailings sands and silts), as well as a few other materials such as
glass beads, granite powder and tailings from the Syncrude
emax – emin = m* emin – emin = emin (m – 1) (2)
oilsands operation in Alberta, Canada. Figure 6 clearly illustrates
the inverse relationship between the extreme void ratios and both
sphericity and roundness. A third parameter, , which was
suggested by Cho et al (2006) who called it the ‘regularity’; and
defined it as = (R+S)/2, is also shown in Figure 6. The regularity
is intended to capture both sphericity and roundness in a single
parameter. As can be seen in Figure 6, it has the same inverse
correlation with the extreme void ratios as S and R (which is not
unexpected).

Figure 7. Approximately proportional correlation between emax and emin for


a wide variety of soil types. (after Cubrinovski and Ishihara (2002)).

Since m > 1, it follows that (m – 1) > 0. So (m – 1) is a direct


proportionality constant between the void ratio range (emax – emin)
Figure 6. The effect of particle shape on extreme void ratios. Republished and the minimum void ratio emin. A similar analysis shows that
with permission of ASCE, from Cho et al. (2006); permission conveyed (emax – emin) is also directly proportional to the maximum void
through Copyright Clearance Center, Inc. ratio emax. We can then conclude that an increase in emin (or emax,
as they are proportional) also increases the void ratio range (emax
Doubt is sometimes expressed about the value and relevance – emin). This in turn increases the potential for volume change
of extreme void ratios emax and emin for soils with a high upon shearing of a very loose packing. If shearing occurs under
percentage of silt sized particles (or finer), given that the standard undrained conditions, the volume reduction will generate large
techniques used to determine these parameters were originally positive excess pore pressure, a factor closely linked to the
intended for clean sands. However, the data presented by Cho et mechanism of static liquefaction, discussed in Section 4.3.
The results from Cho et al. (2006) highlight that high particle
al (2006) were all for material with d50 values more than 0.12mm angularity, such as that of tailings, promotes high values of emin,
(most being in the range of 0.3 to 0.5mm), with Cu (coefficient of emax, and therefore of (emax – emin). However, it is worth
uniformity) values less than or equal to 2.5, thus obviating this remembering that particle size distribution also has a significant
concern. Furthermore, the technical literature now contains effect on the extreme void ratios. Namely, broad gradations
multiple examples of experimental programmes that have adopted promote tight packings while narrow gradations promote loose
emin and emax to aid in the characterisation of non-plastic sand-silt ones (Biarez and Hicher 1994). So, the final value of extreme void
mixtures (e.g. Zlatovic and Ishihara 1995, Cubrinovski and ratios, and associated volume compressibility, of tailings will be
Ishihara 2002, Thevanayagam et al. 2002, Papadopoulou and Tika both a function of particle size distribution and particle shape.
2008, Yang et al. 2006, Carrera et al. 2011, Torres-Cruz and
3.3.2 Critical state (constant volume) friction angle
Santamarina 2020). The consistency of the results yielded by
The ratio of deviator stress (q) to mean effective stress (p') that
these programmes strongly suggests that the extreme void ratios
soils reach at critical state is termed the critical stress friction ratio
are useful indices to understand the mechanical behaviour of silt-
(M) (also discussed in Section 4.3.3). M values are affected by
sized non-plastic tailings.Figure 6 shows that particles with lower
particle shape, with particle angularity promoting higher values of
regularity have a larger difference between the extreme void
M. Tailings that result from rock crushing processes have angular
ratio values, e.g. for , the difference is about 0.2, whereas for
shapes and can thus be expected to exhibit relatively high M
= 0.3, this difference is about 0.45. This observation is also
values. Figure 8 presents the M values for a variety of tailings and
supported and expanded by the extreme void ratio data for over
compares them to values of other soil types that include natural
300 soils reported by Cubrinovski and Ishihara (2002). Their data
soils and sands from rock crushers. The median particle size D50
covers a wide variety of soil types, including gravels, clean sands
is a convenient parameter to plot the data although no correlation
and silts. Although Cubrinovski and Ishihara (2002) calculated
is observed between M and D50. There is significant overlap
separate emax vs emin regression lines for different soil types within
between the M values of tailings and that of other types of soils.
their database, the database as a whole reveals a roughly
This is somewhat expected given that angular particles that
proportional correlation (linear and passing through the origin)
produce high M values are also found in soils that are not tailings.
between emin and emax (Figure 7).
Furthermore, the M values of tailings span a relatively wide range
emax = m*emin (1) from 1.12 to 1.78. The data from glass beads serves to highlight
the role of particle shape, with their spherical geometries leading
to the lowest M values of approximately 0.85.

127
1.8 when depositing the tailings hydraulically, as very significant
particle segregation occurs along the tailings beach. This results
1.5 in a steep beach of sandy material near the deposition point, and a
relatively flat (<1% slope) beach of fine tailings, with these fine
Friction ratio M

1.2 tailings showing extremely low rates of consolidation.

0.9 Differences in Gs for different commodities mean that


Other BAU DCU different deposition strategies may need to be adopted. For
HMS HVC MER
0.6 OIL PFE SAU Glass beads example, tailings with a very high Gs will settle very quickly after
PLA STA discharge from the delivery pipe, forming a relatively steep beach;
0.3 this in turn means the height of the discharge point requires
1 10 100 1000 3000 frequent adjustment, adding to costs of tailings management.
Median particle size D50 [ m] Another implication is that the use of ‘general rules’ are not
always applicable. For example, for gold tailings an in-situ dry
density of 1.7t/m3 should provide tailings that are dilative upon
Figure 8. Critical friction ratio (M) against median particle size (D50) for shearing, a highly desirable outcome as discussed later in this
tailings and other soil types. Symbols for tailings are defined as: BAU = paper. Adopting this value of dry density for an iron ore tailings
Brazilian gold tailings, DCU = Deixing copper tailings, HMS = Hilton
with Gs of 3.8, however, results in a void ratio of 1.24. This may
Mines, HVC = Highland Valley Copper tailings, MER = Merriespruit gold
not be sufficiently dense to ensure dilative behaviour.
tailings, OIL = Syncrude oil sand tailings, PFE = Panzhihua Iron tailings,
SAU = South African gold tailings, PLA = platinum tailings, STA = Stava
3.4.2 Water chemistry
fluorite tailings. Data as compiled by Torres-Cruz and Santamarina
The mineral processing options used for extraction of the
(2020).
commodity of interest vary widely across industries. For example,
extraction of heavy minerals from mineral sands ores is usually
3.4 Some other notable differences between tailings and done by gravity separation, which is made possible by the large
natural soils difference in Gs between mineral and non-mineral (i.e. the sand
and clay fractions), whereas commodities such as gold and copper
3.4.1 Specific gravity require utilisation of a chemical processing agent, such as cyanide
Specific gravity, Gs, is the particle density normalised by the that is often used for gold extraction. Perhaps even more extreme
density of water which, although dependant on temperature, is are the processing of nickel laterite deposits, and of bauxite to
commonly approximated to 1 g/cm3. For natural soils, Gs is often produce alumina.
between 2.65 and 2.7, the former value being that of quartz sand. When extracted from a sulphide deposit (in general this
Occasionally a natural soil may be encountered with Gs values indicates a hard-rock deposit), the process for nickel typically
well outside this range, but it is unusual. Tailings, on the other utilises some form of chemical processing agent. However, nickel
hand, routinely exhibit values well outside this range. Table 1 contained within laterite deposits (which are generally near
below provides some examples of the variability encountered. surface and soft rock to soil consistency) is much more difficult
to extract. The most widely used process utilises high pressure,
Table 1. Range of specific gravity values of mine tailings acid leach. The resulting waste stream is more a chemical process
residue than a conventional tailing. For example, Azam et al
Commodity Specific gravity, Gs Reference
(2005) report results from tests on pressure acid leach (PAL)
Coal (fine refuse) 1.5 to 1.8 Soderberg and Busch, 1977 nickel laterite slurries which all had pH values less than unity and
Lead-Zinc 2.8 to 3.4 Soderberg and Busch, 1977 electrical conductivity values more than the upper limit of the
Phosphate 2.5 to 2.8 Bromwell and Oxford, 1977 measuring device (10 mS/cm). The clay minerals originally
Gypsum 2.3 to 2.4 Vick, 1977 present in the ore were completely decayed by this process, while
Gold 2.8 to 3.1 Fourie et al, 2021; Li et al, magnesium sulphates were formed by the reaction between
2018 sulphuric acid and the magnesium released from the ores. The
Iron ore 3.1 to 3.8 Fourie et al, 2021; Li and resulting slurry was thus radically transformed by the extraction
Coop 2019 process. Sedimentation tests were carried out on PAL slurries
Bauxite 2.9 Fourie et al, 2021 from four different countries; the final settled void ratios varied
Copper 2.7 to 3.75 Fourie et al, 2021; Li 2017 between 4 and 13.
Platinum 3.0 to 3.6 Fourie et al, 2021; Torres- Similarly, the extraction of alumina from bauxite ore produces
Cruz and Santamarina, 2020 a process residue, rather than a tailing. Here the process (most
typically the patented Bayer process) comprises heating the ore to
Table 1 shows the high variability across commodities and about 150 to 200°C in a pressure vessel to which sodium
highlights the importance of experimentally measuring Gs as hydroxide is added, resulting in a process residue that has a pH of
opposed to relying on assumed values. The value of Gs is affected around 12, with high levels of calcium and sodium hydroxide. The
process residue consists of two distinct components, ‘red sand’
by the particulars of the mining process. For example, the Gs value
and ‘red mud’. The red mud residue may contain haematite,
of gold and copper tailings will be closer to the upper end of the goethite, gibbsite, rutile, calcite, sodalite and other complex
range provided when significant quantities of pyrite are silicates, but there is usually a complete absence of either quartz
discharged in the tailings stream. In addition to the across- or clay minerals (Newson et al, 2006). Particles are generally silt
commodity variation, there is often also distinct variability for a sized, with some specific residues having appreciable quantities
particular commodity. For example, many mineral sands deposits of clay-sized particles. Red muds typically have low plasticity and
(from which heavy minerals such as zircon (Gs = 3.9 to 4.7) are high specific gravity values (up to 3.3). Despite the lack of clay
recovered) often contain two distinctly different tailings streams, minerals, and low plasticity, red muds demonstrate some
the sand fraction and the very fine fraction (commonly referred to geotechnical properties that are characteristic of clay-rich tailings,
as slimes). Not only are the particle size distributions (PSDs) very such as relatively high compressibility. In contrast, the angle of
internal friction for red muds tends to be similar to sand-like
different, but the Gs values are often also different. This
geomaterials, with reported values being typically 38 to 42°. Red
combination (different PSD and different Gs) presents challenges muds thus exhibit both clay-like and sand-like characteristics,

128
illustrating the dangers of blindly applying empirical correlations
that have been developed over many years for natural soils
(examples being correlations between undrained shear strength
and Atterberg limits). Furthermore, there are strong indications
that many red muds are cemented or aggregated, with in-situ
undrained shear strengths being very high compared to
uncemented, natural clayey soils at equivalent liquidity indices
(Newson et al, 2006). It is also not unusual for red muds to exhibit
high sensitivities (values from 5 to 15 are common), that is, high
ratios of peak to residual undrained strength. Based on chemical
analyses, Newson et al (2006) attribute the cementation to the
action of hydroxysodalite, which is only slightly water soluble but
becomes more soluble in alkaline conditions.
Figure 9 shows results from a CPTu (cone penetration test
with pore pressure measurement) performed at a red mud site in
Australia. Red mud has been deposited for many years at the site,
and until recently was operated solely through cycles of
deposition in relatively thick layers, with minimal time allowed Figure 10. Illustration of increased yield stress due to bonding; above a
for desiccation drying. The few spikes of higher tip resistance (e.g. critical value of vertical effective stress, the bonding is destroyed,
at about 5.8 to 7m and again at 7.8m are likely periods when some resulting in very large decreases in void ratio under small increments of
drying occurred, a practice that is now more common, as stress (after Leroueil and Vaughan, 1990).
evidenced by the higher strengths in the top 1m or so. What is
particularly important in this result (which is not atypical in red Bitumen extraction from the oilsands deposits in northern
muds) is the very gradual, if any, increase in tip resistance with Alberta utilises a process that combines sodium hydroxide and
depth. It would be expected that, as for natural soils, the tip heat. The clay fraction of the resulting tailings tends to segregate
resistance would increase with depth, due to the effect of from the dominant sand fraction (a function of the bimodal
overburden. Equilibrium pore pressures at this site were particle size distribution) and because the bitumen extraction
hydrostatic, so the observed effect (i.e. no increase in tip process causes dispersion of the clay particles, the resulting
resistance with depth) cannot be attributed to excess pore water deposit experiences extremely slow consolidation rates. As an
pressures. Rather, it is suggested here that this is consistent with example, Jeeravipoolvarn et al (2008) discuss results from a 10m
the observation by Nagaraj et al (1998) that the stress-carrying highly experimental column containing fine oilsands tailings,
capacity of bonds in cemented soils does not change during where after 20 years consolidation settlements were still
compressive loading, manifesting as very little, if any, void ratio occurring.
change with increases in effective stresses. This is illustrated in In terms of impact on the engineering characteristics of
Figure 10, which is a result from two natural, lightly cemented tailings, the biggest impact of the above-mentioned chemical
clays reported by Leroueil and Vaughan (1990) which contrasts processes (that produce residue with chemically altered pore
the behaviour in compression of the intact clays compared with fluids) is probably most pronounced in the initial rates of
the behaviour when prepared as a slurry to the same void ratio as sedimentation and consolidation. This is due to the altered states
the intact samples. of components of the residue, primarily clay minerals, such as the
An explanation for the CPTu result shown in Figure 9 dispersion that occurs with oilsands tailings and some coal
therefore is that the void ratio is almost constant with depth, as is tailings, where dispersion occurs because inter-particle repulsive
the undrained shear strength. The cone detects this constant shear forces dominate. The resulting suspensions are very slow settling,
strength, but the unknown corollary is what happens when the sometimes even remaining indefinitely as a suspension (a state
cemented bonds are destroyed, as they almost certainly are during termed ‘stable’ in some industries, whereas in a geotechnical
advancement of the cone. engineering context the term ‘stable’ is of course the exact
opposite). A good example is provided by Blight et al (1995),
where diamond tailings retained such a high suspended solids
content that the water could not be re-used in the processing plant.
If such repulsive interparticle forces are not overcome, the process
of self-weight consolidation is severely inhibited.
In cases where self-weight consolidation, and other processes
that accelerate an increase in density (such as evaporation) do
occur, the effect of pore fluid chemistry begins to diminish. The
shear strength of the residue is governed by factors such as tailings
density, and in particular density relative to the critical density (or
critical state). Having said that, the lingering impact of inhibited
consolidation may result in a material state above the critical state,
so behaviour upon shearing remains impacted by the pore fluid
chemistry.
An effect that is often overlooked in laboratory testing of
tailings is the salinity of the pore fluid. Small salt concentrations
have little to no effect, but at highly elevated salt concentrations
the effects cannot be ignored. Many of the gold mining areas of
Western Australia (WA) utilise hypersaline groundwater for ore
processing. The resulting tailings have extremely high salinity.
Values of total dissolved solids (TDS) of borewater used in some
WA mines exceed 200 g/l. Aside from engineering characteristics,
such high salinity affects even simple characterisations such as
Figure 9. Results from cone penetration test in red mud. Note in particular
measured water content. This occurs because oven drying causes
the relatively constant tip resistance with depth, e.g., below 15m depth. salts to precipitate as solids, and the conventional definition of

129
gravimetric water content w = Mw/Ms, where Mw and Ms are the deposition of thickened, high-density tailings, the yield stress
masses of water and solids content respectively, incorrectly enables the resulting profile of deposited tailings, this profile
includes the salt mass in the value of Ms, whereas it is actually in referred to as the beach, to be steeper than is usually achieved with
solution in the tailings pore fluid. The correct water content (given conventional tailings, potentially increasing the storage capacity
by Mundle et al, 2012) is: of a given footprint of a TSF. The higher as-placed density of
thickened tailings compared with conventional tailings further
= (3) improves the in-situ density of the resulting TSF.
( ) A drawback with many flocculant-treated tailings is that the
resulting material (the underflow) has a loose, fluffy texture, the
where f is the density of the pore fluid. structure of which can be very sensitive to shearing. The
underflow tailings are usually pumped to the TSF, using either
To illustrate the importance of this correction, for a tailings centrifugal or positive displacement pumps, with the transport
with hypersaline pore water (a TDS of 210 g/litre), at a particular distances often being many kilometres. During such transport the
void ratio the correct water content would be 79%, whereas the thickened tailings are thus subjected to relatively significant
conventional (incorrectly determined) water content would have shearing, potentially resulting in a decrease in the yield stress of
been 67%. The discrepancy flows through to calculations of void the tailings deposited at the TSF. Given the importance of yield
ratio and thus dry density. stress on beach slope development, this alteration of yield stress
is critically important to TSF operators. Houman et al (2007)
3.4.3 Effect of addition of flocculants sampled and tested tailings from ten points along the distribution
As noted in 3.4.2, some tailings may exhibit extremely slow system, from the thickener to the TSF, more than 5km away. Their
settling rates, an effect compounded by the dispersion of clay results are shown Figure 11, which illustrates the change in the
yield stress due to different shear inducing actions. The greatest
particles due to the process used to extract the mineral of interest.
effect is the centrifugal pumping from the thickener to the
Ways to deal with this problem have followed two distinct paths. distribution sump with an approximate four-fold drop in yield
The first (and most widely adopted) is to introduce additives to stress. It should be noted that this large drop was bound to occur
the tailings water in large thickening tanks, the second being to early in the distribution process, when the aggregated structure
introduce (additional) additives at the point of deposition, often was at its most fragile. Of importance is the final yield stress of
referred to as in-line thickening. the tailings as deposited, which is about 25Pa, compared with the
underflow value of 250Pa.
3.4.3.1 In-tank thickening
To overcome the inter-particle repulsive forces that occur in some
dispersed clay-rich tailings in particular, the tailings stream from
the process plant is often first diluted (to solids contents of the
order of 8%) before being fed into the feedwell component of a
tailings thickener. Additives, usually flocculants but with pre-
treatment using coagulants not uncommon, are also introduced
into the feedwell, providing the opportunity for thorough contact
with the dilute tailings, enhancing the subsequent interaction
between tailings and additives.
A comprehensive discussion of tailings clay mineralogy and
treatment options is provided by Vietti and Coghill (2015). For
present purposes it is sufficient to note that the addition of high
molecular weight, synthetic polymeric flocculants (commonly
used products include polyacrylamides and its derivatives) adhere
to clay particles, providing a bridging between and agglomeration
of particles, increasing the size of the resulting aggregate, also
termed a ‘floc’, thus increasing the rate of settlement of the
suspension.
Thickeners for providing more rapid settlement of treated
Figure 11. Measured change in partially sheared yield stress throughout
tailings have evolved rapidly over the past two decades or so, and
the tailings distribution system, where SP1 indicates immediately after the
a comprehensive discussion is provided by Bedell et al (2015). thickener and SP10 at the TSF (from Houman et al, 2007; courtesy
These thickeners, sometimes referred to as deep cone thickeners, Australian Centre for Geomechanics).
or paste thickeners, can be very large diameter, e.g. >50m, and
more than 20m high. The height enables a reasonably substantial Accounting for the shear-induced drop in yield stress is trivial,
overburden stress to develop within the thickener, generating if samples are taken at an appropriate location, e.g. the point of
some self-weight consolidation of the newly aggregated tailings.
deposition. During the design phase for a new TSF, the obvious
The aggregation process is further enhanced by the provision of
vertical rakes (pickets) that rotate slowly through the settling precaution is to test fully sheared samples of thickened tailings
slurry, enhancing the drainage and consolidation of the slurry. The prior to undertaking geotechnical tests such as settling and
thickened, or high-density tailings, is extracted from the base of consolidation tests. The effect of partial or no shearing of
the thickener, this being referred to as the underflow. At this stage flocculated suspensions is more acute in applications of the
of the process, the tailings have a measurable, but rather low yield second generic technique of tailings thickening, i.e. in-line
stress, unlike conventional tailings, which usually undergo flocculation.
minimal thickening and certainly insufficient thickening to impart
any measurable yield stress. 3.4.3.2 In-line flocculation
Yield stress is essentially just the undrained shear strength of This technique is now known by various names (e.g. secondary
a geomaterial, but measurements are in the pascal range rather flocculation or polymer treatment); we retain the term in-line
than kilopascal range. Indeed, the preferred method of measuring flocculation. In this technique, additional flocculant is introduced
yield stress has emerged as the vane test, as borrowed from into the tailings stream as it travels through the pipeline, with the
geotechnical engineering practice (Gawu and Fourie, 2004). Upon point(s) of flocculant addition being close to the pipeline end,

130
mainly to minimise shearing of the flocculated material. In-line
flocculation has many challenges, such as differences in the
viscosity of the flocculant and the tailings hindering mixing and
achieving the optimum mixing energy to maximise the
flocculation process but minimise post-flocculation shearing.
Reported applications of in-line flocculation almost invariably
report extremely high flocculant dosages, with values of 250g/t
reported by Wells et al (2015).
The primary reason for using in-line flocculation is rapid
water release and more rapid increase in the shear strength of near-
surface tailings. One of the pioneering applications of in-line
flocculation was to treat the problematic ‘mature fine tailings’, or
MFT that occur within oilsands tailings due to segregation of the
tailings stream upon deposition. Successful reported applications
of the technology are discussed by Wells et al (2015). After an
initial period of enthusiasm for in-line flocculation about ten years Figure 12. Difference between flocculated (PT) and untreated (UT)
ago, this enthusiasm has waned slightly, as operational challenges material (after Reid and Fourie, 2017).
became more apparent, examples being difficulties controlling
beach profile, in addition to the high costs, given the required 4 LABORATORY CHARACTERISATION OF TAILINGS
flocculant dosage concentrations. Nevertheless, the technology BEHAVIOUR.
continues to find some applications, with designers now aware of
potential pitfalls. 4.1 Index tests
Assuming successful implementation of an in-line As already noted, the specific gravity (SG) of mine tailings varies
flocculation deposition system, interest turns to potential changes widely. When preparing samples for SG testing, precautions may
in the geotechnical properties of the resulting tailings, particularly be necessary, for example tailings containing salts or other
as the structure (or fabric) of the material is so different from concentrated chemicals. This becomes even more important when
conventional tailings. working with process residues, such as bauxite red muds.
Most researchers who have investigated changes in hydraulic Similarly, pre-testing treatment may be required when carrying
conductivity report either an increase or no change, for tests out particle size analyses. Some industries have bespoke
undertaken at the same density for both thickened and
techniques that find relatively little application outside of those
conventional tailings (Jeeravipoolvarn et al, 2009, Beveridge et al,
specific industries. One example is the testing of oilsands tailings,
2015, Znidarcic et al, 2015 and Reid and Fourie, 2017). There are
thus two factors potentially benefitting water management and where the methylene blue index (MBI) test is used as an indicator
recovery with an in-line operation, i.e. the initial rapid water of clay content; this is used principally with the tailings fraction
release on deposition, and the increased rate of self-weight referred to as mature fine tailings. We limit further discussion on
consolidation provided by the higher hydraulic conductivity. the topic of industry-specific index tests and rather focus on
What is not so clear are potential changes in long-term strength. geotechnical characterisation.
The polymers used to date in in-line flocculation applications are,
to the authors knowledge, all biodegradables. Over time, the 4.2 Sedimentation, consolidation, desiccation and creep.
fabric induced by the high flocculant dosages used will likely be Laboratory testing techniques to evaluate volume change behavior
retained, as the flocculants degrade. The question is whether this and determine the consolidation parameters to be used in
induced fabric has the potential to be more sensitive (where modeling settlement and void ratio distributions of the disposed
sensitivity reflects the change from peak to post-peak strength, tailings in a TSF continue to evolve. For the design phase of a
usually measured in terms of undrained shear strength) than project, samples for testing should be obtained from a pilot mill
conventional tailings. processing plant which mimics the operational conditions in the
To determine possible long-term shear strength of highly future plant. However, upon starting the mine operation,
flocculated tailings, it is of course first necessary to determine additional samples should be collected and tested to confirm that
whether there are any short-term changes. Reid and Fourie (2017) the consolidation parameters in the design phase were realistic. If
clearly show that the critical state line (CSL) for heavily not, the consolidation analyses from the design phase should be
flocculated material lies significantly above the CSL of repeated and subsequently used to monitor the performance of the
equivalent, untreated material, with the difference in void ratio at disposed tailings. It is noted that some tailings exhibit various
100kPa mean effective stress, p', for example, being levels of particle segregation upon deposition, depending on their
approximately 0.1 (see Figure 12). This difference in void ratio is gradation and water content upon deposition. Particle segregation
significant, particularly when defining the boundary between creates a heterogeneous spatial distribution of the consolidation
potentially contractive and potentially dilative behaviour. For properties upon deposition and the laboratory testing program and
example, material deposited using in-line flocculation that is at a subsequent analyses should recognize this. It is noted again that
p' value of 100kPa and a void ratio of 0.5 would be well below its consolidation parameters of highly compressible tailings are
corresponding CSL; however, it is also well above the CSL of the highly nonlinear, with significant changes in compressibility and
untreated tailings. The question that arises is, if degradation of the hydraulic conductivity values over the range of stresses to which
flocculant results in an unaltered fabric and void ratio, is the the tailings will be subjected, so any analyses using “average”
tailings then potentially highly collapsible (contractive), a natural values of these parameters may often be meaningless. To address
analogy perhaps being the Rissa landslide in Norway in 1978, or this problem rationally, the sample collection and testing
is the process of degradation of flocculant accompanied by a strategies should include samples of various gradations from the
gradual decrease in void ratio, perhaps moving the state condition finest to the coarsest that are collected from the pool and beach
below the CSL of untreated material? This question appears to areas of the impoundment. Thus, a range of consolidation
remain unresolved. parameters is obtained enabling the settlement and void ratio
distribution analyses to “bracket” the expected behavior in the
impoundment or provide a rational method of assessing the
settlement rates and magnitudes in different parts of the
impoundment.

131
The selected samples are then subjected to testing in the The slurry behaves like a soil if the initial void ratio is below
laboratory to get the model parameters necessary for predicting 8. For a higher initial void ratio the material will sediment until
tailings behavior in the impoundment upon deposition. Depending the zero effective stress void ratio is reached. The zero effective
on the consistency or water content of the deposited material and stress void ratio for the kaolin clay is between 8 and 20, depending
the tailings management strategies adopted, upon deposition the on the water content at which the slurry was mixed. This is not a
processes controlling tailings behavior include: sedimentation, surprising result since the particle interaction at the very low
consolidation, desiccation and possibly creep. The goal of the effective stress will be dependent primarily on the characteristics
laboratory testing protocol should be to obtain the relevant of the diffuse double layer attached to the clay surface and the
modeling parameters for all these processes. Subsequently we chemical characteristics of the pore liquid, so appropriate process
discuss the testing protocol for each of these processes. water should be used in settling tests. When more free liquid is
available for each particle in more dilute suspensions, the double
4.2.1 Sedimentation layer will be thicker and the clay particles will reach equilibrium
In typical hard-rock tailings and other low compressibility
at a larger distance apart. This difference in the initial void ratio
tailings, particles tend to settle out along the beach soon after
creates different compressibility curves at low effective stresses.
deposition and thereafter the process of consolidation proceeds.
If a more detailed analysis of the sedimentation part of the process
However, in highly compressible tailings, consolidation is
preceded by sedimentation. is desired, a more elaborate testing protocol needs to be employed
Sedimentation is the process in which the soil particles settle following a procedure similar to the one described by Pane (1985)
in water until at the bottom of the settling column they create a in which he proposes a fluidization test in addition to the
soil layer with developing effective stresses and corresponding consolidation test to characterize the material behavior from
void ratio reduction. Unlike the settling process described by sedimentation to consolidation. In many practical applications the
Stokes’ law, in which the settling velocity of the particles is detailed sedimentation modeling is not critical and can be ignored
proportional to their size, during sedimentation all the particles in the analyses if the volume change associated with the transition
settle at the same velocity proportional to the local concentration from the initial void ratio at which tailings are deposited to the soil
of suspension. The process is properly described by the classical formation void ratio is properly accounted for. This is also valid
Kynch’s (1952) theory of sedimentation which was later for rapid-settling tailings as often very substantial bleed water is
augmented by work of Pane (Pane 1985, Pane and Schiffman released upon deposition on the beach, and the initial void ratio of
1997) who analyzed the combined sedimentation-consolidation the tailings layer is significantly lower than it was upon
process and developed experimental techniques for the deposition. Although not discussed in detail in this paper, properly
determination of soil sedimentation characteristics. accounting for the sedimentation process becomes extremely
The theory provides a convenient tool for the analysis of the important when sizing deep thickeners for clayey (i.e. highly
sedimentation process. However, from the practical point of view compressible) tailings used to produce high density tailings.
in tailings disposal problems the sedimentation process does not
usually play a major role and could be neglected in the analysis. 4.2.2 Consolidation
Namely, compared to the consolidation process that takes place Once a tailings layer is formed in an impoundment, some pore
over a period of several years, sedimentation is usually completed water will be expelled from the material during the consolidation
within several days. However, with the evolving interest in the use process. As stated earlier, the process is described by a nonlinear
of in-pit deposition, recognizing the different approach required finite strain consolidation theory (Gibson et al 1967) and the
for tailings where sedimentation is important is essential. material parameters must be determined by laboratory testing and
Nevertheless, the consolidation analysis will often be the only test analyses procedures that are consistent with the theory to be
analysis needed in the design of highly compressible used for both test interpretation and consolidation modeling. The
impoundments as regards volume changes. It is important to know consolidation process depends on both compressibility and
at which soil density or void ratio the sedimenting slurry becomes permeability characteristics of the soil. Power functions, fitted to
a soil and the effective stress principle applies. Unfortunately, this the experimental data, best represent constitutive properties of soft
value, which was called the fluid limit by Monte and Krizek
soils under one-dimensional compression (Somogyi et al 1981).
(1976), is not a material constant but depends on the initial water
Compressibility:
content of the slurry (Liu, 1990; You, 1993). This void ratio can
be easily determined in a simple experiment in which the slurry is e = A ´B (4)
allowed to sediment in a laboratory beaker and the surface soil is
sampled for the void ratio determination. This void ratio Hydraulic conductivity:
corresponds to zero effective stress and gives the upper bound to
the void ratio at which the mixture exists as a soil. Figure 13 shows k = C eD (5)
the results for a laboratory mixture of kaolin clay with water.
in which e is the void ratio, k is hydraulic conductivity, ´ is the
effective stress and A, B, C and D are material parameters that
need to be determined in laboratory testing.
In addition to these relations the zero effective stress void ratio
must be defined, bringing the number of parameters for
compressibility to three. Liu and Znidarcic (1991) expanded the
compressibility relationship to the form:

e = A( ´+Z)B (6)

so that the zero effective stress void ratio becomes e0 = AZB. There
is general agreement that the void ratio-hydraulic conductivity
relationship for highly compressible materials is well represented
by the stated power function, although several alternative forms
Figure 13. Relationship between initial void ratio and void ratio at onset for the compressibility relationship have been proposed over the
of development of effective stresses (i.e. zero effective stress void ratio) years. Any form of these functions is acceptable if they properly
for a kaolin clay (data from Liu, 1990).

132
model compressibility and hydraulic conductivity relationships
observed in the experiments.
The objective of the experimental work is to determine the
five parameters A, B, Z, C and D. Conceptually, the simplest
approach to determining these parameters is to take a tailings
sample and load it incrementally with several load increments
covering the stress range of interest and then directly measure
hydraulic conductivity at each load increment. This procedure
results in several void ratio-effective stress and void ratio-
hydraulic conductivity data points to which the analytical
expressions could be fitted and the corresponding parameters
determined. This simple procedure is often referred to as the
Slurry Consolidation Test and many variants of the procedure
have been implemented in various laboratories. There are two
challenges with this procedure. First, the test must provide reliable
data in the low effective stress range because for many slurries the
dominant volume change is observed in the low effective stress
range, (less than 10 kPa) and often even less than 1 kPa. Second,
to measure the hydraulic conductivity, a hydraulic gradient must
be created across the sample, which results in a variable effective
stress distribution, and a corresponding variation in void ratio and
variable hydraulic conductivity within the sample. This is
particularly critical at low effective stresses where the hydraulic
conductivity could easily change by an order of magnitude due to
the imposed seepage forces and self-weight of the material. These Figure 14 Seepage induced consolidation test equipment.
challenges often require imposition of low hydraulic gradients and
long testing times in highly compressible, low permeability
tailings to minimize these undesirable effects.
To address these challenges Imai (1979) suggested that
seepage forces could be used to trigger consolidation of the slurry.
By adding pore water pressure measurement needles in the
sample, he could account for the hydraulic conductivity variation
in the test analysis. The procedure required a specialized
apparatus and a very skilled operator to obtain reliable results.
Following the idea of using seepage forces to trigger consolidation
in slurry samples, Liu (1990), Abu-Hejleh and Znidarcic (1994)
and Abu-Hejleh et al (1996) developed a testing procedure and the
corresponding analysis to determine the consolidation
characteristics of slurries from a two-step test and an inverse
process solution approach to determine all five parameters for the
compressibility and hydraulic conductivity relationships. It
requires an initial permeation of the sample under a relatively high
gradient and a final step load test and direct hydraulic conductivity
measurement at the maximum stress level expected in the field.
The testing procedure significantly shortened the testing time and
the inverse problem analysis procedure determined the
Figure 15 Equipment used for the SICTA procedure.
consolidation parameters using the finite strain consolidation
theory. Thus, the experimental efforts and the data analysis Figure 16 presents the compressibility and hydraulic
method are completely consistent with the numerical models to be
conductivity test results for several tailings samples ranging from
subsequently used to analyze behavior of tailings upon deposition.
fine to coarse materials. Figure 16b presents both the step load test
The schematic of the laboratory equipment for the SICTA
results with the direct measurement of hydraulic conductivity at
procedure is presented in Figure 14. It consists of three major
components: the testing cell containing a slurry sample with a each load increment and the seepage induced consolidation testing
piston and load frame, a flow pump for controlling the imposed and analysis (SICTA) results (Abu-Hejleh and Znidarcic 1994).
flow rates and a differential pressure transducer connected to a Both procedures yield very similar test results, but the advantages
data acquisition system. Figure 15 shows an example of the of the SICTA procedure is evident in the results for fine-grained
system components in a laboratory. tailings. For the step load test results an assumption is made that
the effective stress, void ratio and hydraulic conductivity within
the sample are uniform and the data points represent the average
values for these quantities without any regard for the self-weight
stresses or induced seepage forces within the sample. The seepage
induced consolidation test analysis on the other hand correctly
accounts for the variation of these quantities within the sample
throughout testing. This is best illustrated by the data points in the
compressibility plot for fine tailings at the effective stress of 0.1
kPa. The upper data point represents the average void ratio at
steady state under the applied stress of 0.1 kPa, self-weight of the
sample and a very small seepage force during hydraulic
conductivity measurements under a low flow rate. The lower data
point represents the average void ratio at steady state under the

133
applied stress of 0.1 kPa, self-weight of the sample and a seepage shrinkage where the soil remains saturated from the initial void
force due to hydraulic conductivity measurements involving a ratio to the shrinkage limit void ratio. The second stage involves
flow rate four times higher than the rate used for the upper data zero shrinkage, where the soil does not show any volumetric
point. The higher flow rate increased the effective stress at the change from the shrinkage limit void ratio to the end of the
bottom of the sample to about 4 kPa. The SICTA procedure used desiccation process. To account for these two shrinkage stages, an
the data at steady state under the higher flow rate and the data at extension of the nonlinear finite strain consolidation theory as
the maximum effective stress to calculate all five model developed by Gibson et al (1967) was established by Abu-Hejleh
parameters A, B, Z, C and D and generate the continuous curves and Znidarcic (1995), Oliveira Filho (1998) and Yao et al (2002).
covering the whole range of effective stresses and void ratios The overall consolidation and desiccation process of soft soils is
observed in the test. modeled in this theory with consecutive phases which correspond
chronologically to the phases that a soft soil layer undergoes in
the field after deposition. These phases are consolidation under
one-dimensional compression, one-dimensional shrinkage,
propagation of desiccation cracks with tensile stress release, and
three-dimensional shrinkage. Several material functions are
required in the solution process for a given field situation. In
addition to the compressibility and hydraulic conductivity
relations for consolidation analysis, the shrinkage relations under
one- and three-dimensional conditions are needed. Additional
laboratory tests are required to determine these relations, but for
slurries they are often found to be very similar to the
compressibility relations used in the consolidation analysis and
additional tests might not be necessary (Oliveira Filho 1998). If a
detailed analysis of desiccation crack propagation is desired,
parameters for a cracking function must be determined. The
function that relates the void ratio at which the crack will open for
a given depth (total stress) and its parameters can be obtained from
a desiccation test conducted in a geotechnical centrifuge under
increased self-weight stresses. However, given the relative
scarcity of geotechnical centrifuge testing facilities, this step is not
commonly conducted.

Figure 16 Compressibility (a) and hydraulic conductivity (b) relationships


for several tailings samples.

The effect of seepage forces during permeation of the sample


to get the hydraulic conductivity characteristics is significant for
fine grained tailings under low effective stresses (< 1 kPa).
Neglecting these forces in the test analysis could lead to incorrect
representation of the consolidation characteristics of tailings. In
addition, the SICTA procedure requires a much shorter testing
time than the step load test with the direct hydraulic conductivity
measurement at each load increment. Depending on the hydraulic
conductivity of the material, a step load type of test may take
several months to complete while the seepage induced Figure 17 Shrinkage data for china clay (after Abu-Hejleh and Znidarcic,
consolidation test is typically completed with a couple of weeks 1995).
for fine grained materials (Abu-Hejleh et al 1996).
4.2.4 Creep
4.2.3 Desiccation In some instances, tailings processing includes treatments that
When the upward flow of water, due to the consolidation process create flocculated structures with significant viscous deformation
is completed, the desiccation process starts. Although we usually of the flocs. Examples include treated fine-grained oilsands
associate desiccation with evaporation, it is important to realize tailings (often referred to as mature fine tailings, or MFT), and
that the lowering of the groundwater table could cause desiccation fine-grained mineral sands tailings (‘slimes’). The nonlinear finite
as well. In general, the desiccation process may be defined as the strain consolidation theory does not properly capture this behavior
process in which soil compression takes place under increasing and an extended consolidation model that includes creep is needed
suction or negative pore water pressure. Thus, any time the pore (Murphey et al, 2021). It is also possible that some tailings,
water pressure drops below the atmospheric pressure the soil especially of high plasticity, exhibit time dependent compression
undergoes desiccation. As suction increases, the soil shrinks but even without any additional chemical treatment that would create
remains saturated until the air entry value is reached, after which a flocculated structure. In such cases the laboratory testing must
desaturation begins. This simplifies the analysis significantly. be expanded to include time dependent compressibility
Experimental shrinkage data, i.e. void ratio versus water content, characterization for modeling purposes. By necessity, such tests
are shown in Figure 17. They suggest that soil shrinkage can be require prolonged testing times, and a proper balance should be
modeled with two stages. The first stage involves normal found between test duration and the trustworthiness of the data. In
any case, any observed and documented creep effects in the

134
laboratory must be extrapolated by the creep model to longer provide some redundant data as well as creep data collection at all
times to accommodate time scale in the field. four load levels. The other extreme would be to test only one
A creep model recently developed for tailings accounts for the sample and collect creep data at each load level, implying a four-
time dependent compressibility by changing the three fold increase in testing time. Assuming that a minimum of three
compressibility parameters A, B and Z into time dependent weeks of creep data collection is desired, the four parallel tests
equivalents defined by the following relations (Gjerapic et al would require about six weeks of testing time, while for a single
2021): sample, the test would take six months to complete. It is also noted
that any creep model will rely on extrapolation beyond the testing
time in the laboratory up to the times of interest in the field. The
Af, Bf and Zf parameters in the proposed model (Eqs. 7 to 9) should
be considered as parameters to be used in sensitivity analyses
rather than the “true” parameter values at infinite time, as clearly
they can never be truly verified.
In a soil exhibiting creep, the total void ratio change is an
additive decomposition of the void ratio change due to the
effective stress change and the void ratio change due to creep per
se. The continuity condition requires that this total void ratio
change be consistent with the imposed hydraulic gradient and the
hydraulic conductivity of the material. The creep-induced void
ratio change is not associated with changes in the hydraulic
gradient and does not speed up the pore pressure dissipation
process. Creep only causes void ratio changes to be larger, but it
Equations (7), (8) and (9) allow for the time-dependency of does not significantly affect the void ratio change rate until the
constitutive properties A, B and Z by introducing additional consolidation process ceases. This does not imply that there are
(creep) parameters. At time equal to zero, parameters A, B and Z no creep induced void ratio changes during the consolidation
are defined by their initial values A0, B0 and Z0. Experimental process, as will be demonstrated later. To better elucidate this
evidence suggests that there is a limit in the change of parameters peculiarity a thought experiment is helpful. Let us consider a real
A, B and Z. Therefore, parameters Af, Bf and Zf are introduced to soil with infinite hydraulic conductivity. Such a soil would, under
define the final (limiting values) of parameters A, B and Z at the imposed stress change, such as self-weight or externally
infinite time. Parameters a1, a2, b1, b2, z1, and z2 are found by applied load, undergo instantaneous consolidation followed by
fitting laboratory data to Equations (7), (8), and (9) creep induced void ratio change directly correlated to elapsed
A set of laboratory tests on a sample, or multiple samples, of time. In that case the two separate void ratio changes would be
a slurry are required to collect data for the estimation of these completely uncoupled and easily distinguishable from each other.
constitutive parameters. In order to determine all parameters in the In a real soil however, the void ratio change will progress
proposed model including A0, B0, Z0, C, D, Af, Bf, Zf, a1, a2, b1, b2, gradually whereby the rate of change is controlled by Darcy’s law,
z1, and z2, a sample must be compressed under several load and any creep induced void ratio change will be undistinguishable
increments. The hydraulic conductivity should be measured at from the effective stress induced change. The creep induced void
each load increment and the loads should be sustained for a long ratio change will only very slightly affect the initial consolidation
enough period to collect sufficient data to model the long-term
compressibility, i.e. creep. From our experience with testing soft theoretically never ends, the effective stress induced void ratio
soils, we determined that for routine testing a set of four load changes become negligible on a relatively small laboratory
increments is sufficient to properly characterize the specimen after a period controlled by the hydraulic conductivity
compressibility behavior of these materials. At the low end of the of the material. In the past, 24 hours or the time required for
stress range, a seating load of 0.1 kPa is considered the lowest dissipation of the measurable excess pore water pressure was
value for which reliable data could be collected. The highest stress often considered to be long enough to assume the consolidation
of interest should correspond to the highest effective stress process to be completed. However, for slurries, taller samples, and
expected in the project. In between these two extreme values, two very low hydraulic conductivity materials that time can be
additional load increments are desirable to better characterize the substantially longer. Nevertheless, at some point the void ratio
material in the mid-stress range and to provide some data change with time can be considered to be caused by creep only
redundancy. Likewise, measuring the hydraulic conductivity at and the experimental data collected after that time can be used to
each of these loads provides sufficient data to correctly fit the creep model and determine the model parameters. The time
characterize the void ratio – hydraulic conductivity relationship. after which the creep model is fitted to the experimental data
When measuring the hydraulic conductivity the imposed should be determined by performing consolidation analysis for
hydraulic gradient should be as low as practical to minimize the each load increment to find at what approximate time the
void ratio variation within the sample during testing. This is consolidation process is essentially completed. Note, only a rough
particularly critical at low effective stress range at 0.1 and 1 kPa estimate of this time is sufficient.
load increments. As pointed out above, the SICTA procedure Once the creep model parameters are determined from the
(Abu-Hejleh et al. 1996) accounts for variations in effective stress data in the creep observation period, the model is used to
due to seepage. However, this procedure is only applicable for extrapolate the creep void ratio change to infinite time (AF, BF and
materials that do not exhibit creep and cannot be applied for this ZF) and to zero time (A0, B0 and Z0). While the parameters at
model without further development to account for creep in the infinite time should be used as variables for parametric studies,
analysis. the parameters at zero time have a role of separating the creep
The long duration tests required to capture creep behavior induced void ratio changes from those caused by the effective
come with a set of challenges that may compromise the quality of stress change during the consolidation process when experimental
the data. This includes potential gas generation, bacterial growth separation of the two effects is not possible. It is noted that while
or other long-term changes that might take place in the laboratory. this might appear to be a somewhat undiscriminating process, its
Thus, in each case a compromise must be found between the effect on any subsequent consolidation/creep analyses will be
quantity and quality of the data. In ideal circumstances, four minimal as only the total void ratio change is relevant for realistic
simultaneous tests would be performed, and each test would be predictions.
targeted for a prolonged testing time at one load level. This would

135
An example of the creep effect and its modeling for a water pressure. Loose soils and tailings will, when sheared
laboratory prepared bentonite slurry is shown in Figure 18. undrained, produce large positive increments in excess pore water
pressure, this often being referred to as shear-induced pore
pressures. More discussion on what constitutes ‘loose’ follows in
the section immediately after this.
Undrained triaxial compression tests on loose samples will
generally mobilise a peak deviator stress well before the effective
stress strength envelope is reached. This is best explained by
reference to a stress path plot, as shown in Figure 19, which plots
mean effective stress p´ against deviator stress q. The sample
shown has been isotropically consolidated to a mean effective
stress p0´ and then loaded in undrained compression. Being loose,
the undrained effective stress path follows the path shown,
reaching a peak deviator stress at a ratio that is much less than
would be achieved in an undrained compression test on a denser
specimen (indicated by the dashed line in the figure). A straight
line through the origin and the peak deviator stress value (with
multiple triaxial tests done at different effective confining stresses
to produce more points, preferably) defines the instability line,
often defined by:

= (10)

Undrained effective stress stability analyses are often carried


out using the vertical effective stress-dependent undrained shear
strength ratio,
Figure 18. Example of creep tests and associated modeling results
(parameters A, B and Z given in equations 7 to 9).

4.3 Shear strength under monotonic loading


Testing of tailings to determine appropriate strength parameters
has evolved rapidly over the past decade or so. Before discussing,
in some detail, the application of the critical state soil mechanics
(CSSM) framework to tailings testing, we discuss other aspects of
monotonically loaded tailings specimens.

4.3.1 Effective stress (drained) strength parameters


Stability evaluations of TSFs were for many years confined to
effective stress slope stability analyses, where key inputs were the
strength parameters c´ (apparent cohesion) and the angle of
internal friction, and the elevation of the phreatic surface.
Effective stresses were evaluated based on the phreatic surface
level. No account was paid to pore water pressures generated due
to shearing, which we later discuss in some detail, which are
crucial when considering the stability of loose, contractive and
particularly brittle tailings.
Unfortunately, it is still not uncommon to see stability Figure 19. Illustration of stress path of an undrained triaxial compression
evaluations based solely on effective stress (sometimes termed test on specimen that was initially looser than the critical void ratio.
drained) considerations, ignoring undrained stability evaluations,
despite the dangers of this approach having been emphasised Many national guidelines for TSF design and analysis now
many times over. Fortunately, this is changing. specifically call for both drained (effective stress) and undrained
The drained shear strength parameters are almost invariably stability analyses (ANCOLD, 2012, CDA, 2020), with the same
derived from results of triaxial compression tests, which may be Factor of Safety (1.5) being required for both conditions. In the
drained or undrained. An issue of concern, even with these case of ANCOLD it is specifically applicable for TSFs where loss
relatively simple tests, is the use of multi-stage loading tests, of containment is possible should a slope instability occur.
where a single sample is tested at three values of confining stress, The first experimental results showing the exceptional stress-
and only the peak deviator stress is determined. The justification strain response of a loose sandy soil subjected to an undrained
often made for multi-stage loading tests is limitations on the loading is shown in Figure 20 (Castro 1969). This sample of sandy
number of samples available for testing. However, it must be soil was first isotropically consolidated under 400 kPa, reaching a
recognised that any post-peak strength loss and the strain over void ratio after consolidation equal to 0.714, which is associated
which it occurs is thus lost, rendering the data useless for with a relative density of 37%. Over 14 minutes the sample was
implementation in any form of numerical modelling. axially loaded by means of small dead-load increments that
yielded a typical stress-strain curve with a peak vertical stress of
4.3.2 Undrained strength parameters around 230 kPa at approximately 1% of axial strain. Thereafter,
When the intrinsic tendency of volume change is suppressed in a the next small load increment triggered a sudden failure of the
saturated soil mass, for example due to fast loading due to seismic specimen and within a fraction of a second (0.18 seconds) it
disturbances, the soil response is undrained, and therefore, the reached about 20% axial deformation and the strength dropped to
volume change tendency is transformed into changes in pore a value of 80 kPa.

136
= log (12)

There is no fundamental physical basis for Equation 11. It is


merely intended to provide an empirical fit to the underlying data
points. Indeed, Equation 11 predicts physically inadmissible
values at very low stresses (e e < 0).
Empirical alternatives to Equation 11 are available to model CSLs
that exhibit curvature in e-log(p') space (e.g. Li and Wang 1998,
Bauer 1996). Naturally, the greater versatility of these alternative
equations comes at the expense of introducing a third fitting
parameter. Choosing between Equation 11 or more elaborate
alternatives largely depends on visual inspection of the data and
the envisaged application of the CSL. Ultimately, it is a matter of
Figure 20. Load-controlled CIU (isotropically consolidated, undrained) how closely the idealised CSL needs to follow the underlying data
triaxial test on loose sand (modified after Castro 1969). points, and whether the scatter and curvature of such points
warrants adopting a more complex alternative to Equation 11. It
As can be observed, the pore water pressure increases is interesting to note that even when characterising the exact same
significantly during the test, reaching a constant value close to tailings, some laboratories adopt Equation 11, while others adopt
90% of the confining pressure. From this type of test result, the more elaborate alternatives, such as a power law alternative (Reid
true liquefaction or flow failure was established, as well as the et al. 2021).
concept of critical state that provides the framework to describe
the behaviour of sandy soils and low-plasticity silts. Experimental 4.3.3.1 The Steady State of Deformation
results have shown that it is possible to establish an instability line The terms ‘steady state’ and ‘critical state’ are both used to
(Figure 19), which, if reached under an in-situ drained loading describe behaviour of soils at large strains. It is often suggested
regime (stress-controlled), will trigger a sudden increase in that they are one and the same, but in this paper we use the
strains, and generate positive excess pore pressures. Because the opportunity to take a short diversion to point out that there are
rate of pore pressure generation is greater than the time required differences. Whether these differences are of consequence when
to dissipate these shear-induced pore pressures, the drained characterising the engineering behaviour of tailings was
condition is almost instantaneously modified into an undrained extensively debated by the co-authors of this paper.
response of the soil, reducing the stresses to the undrained residual The term “steady state of deformation” was introduced by
strength. Poulos (1971) and it was defined as the state at which a particulate
A distinction has been made between ‘loose’ and ‘dense’ material deforms continuously under a constant state of effective
samples. Since these are relative terms, we must clarify what stress at constant velocity and at constant void ratio. The steady
constitutes loose and dense. Invoking the concept of CSSM state of deformation is achieved only after all particle orientation
provides a convenient way to make this distinction and to utilise has reached a statistically steady-state condition and after all
the resulting differences in behaviour in routine design and particle breakage, if any, is complete, so that the shear stress
analysis and interpretation of laboratory and field data. needed to continue deformation and the velocity of deformation
remains constant (Poulos 1981).
4.3.3 Application of critical state soil mechanics principles Roscoe and collaborators (Roscoe et al, 1958) introduced the
Critical state soil mechanics (CSSM) is commonly adopted as a critical void ratio line to refer to the locus in e-p'-q space
framework within which to understand the strength and volume associated with the ultimate state reached by a soil that is sheared
change characteristics of tailings (Jefferies and Been, 2015, and continues to deform at constant stress and constant void ratio,
Robertson et al. 2019, Bedin et al. 2012). The key tenet of CSSM as already discussed in this paper.
is that when soils are subjected to large shearing deformations, Later, Schofield and Wroth (1968) coined the term "critical
they eventually reach their critical state line (CSL) which is state soil mechanics", stating that the kernel of this concept is that
typically defined in a three-dimensional space: void ratio (e), soil and other granular materials, if continuously distorted until
mean effective stress space (p'), and deviator stress (q). It is they flow as a frictional fluid, will reach a well-defined critical
common to model the e-p' projection of the CSL by adopting a state that can be represented by two straight lines: one in the q-p´
semi-logarithmic linearization: plane and the other in the e-log p´ plane.
According to these definitions, both conditions (steady state
and critical state) seem to be identical in that both emphasize the
condition of an ultimate state at large strains of continuous
= log (11) deformation at constant void ratio and constant stresses. However,
the definition given by Poulos (1981) for the steady state has more
conditions to be satisfied, being:
where 10 is the slope and thus a measure of compressibility, p* is
a reference stress that ensures a dimensionless argument for the
- Constant velocity
logarithm, and p* is the void ratio when p' = p*. Equation 11 is
often used without explicitly including p*, which effectively - Particle orientation to reach a specific fabric
amounts to adopting p* as one unit of stress. The value of p* is - All particle breakage, if any, is completed
immaterial to the resulting CSL, although stronger correlations
between p* and soil properties emerge when p* takes on p' values According to these definitions, the critical state would be a
at which the underlying triaxial tests reached critical state (Cho et state that does not include the rate of deformation and orientation
al. 2006, Torres-Cruz 2019). As for the base of the logarithm, of particles, which are important factors that affect the mechanical
besides the base 10 used in Equation 11, it is also common to use behavior of clayey materials. It is worth mentioning that because
a natural logarithm. Changing the base b of the logarithm the critical state concept was applied successfully to develop the
well-known Cam-Clay model, its applicability for clayey soils is
produces a different slope b
often considered completely valid, regardless of the effects of

137
deformation rate and particle re-orientation. This situation needs Additionally, an example presented by Schofield and Wroth
some clarification, which likely is associated with the complex (1968) is reproduced in Figure 23, where it is clear that the
behavior of clayey materials, part of which is summarized below. residual strength developed by clayey soils with platy particles
The Japanese Kawasaki clay (Gs = 2.69, LL = 55.3, PI = 29.4, differs from the strength established at the critical state.
FC = 83.9%) has been tested at different rates of deformation
(Nakase and Kamei, 1986). Figure 21 shows the results of a series
of undrained triaxial tests started from a K0-consolidated
condition (zero lateral strain) and loaded in either compression or
extension. Three different rates of deformations were applied
during loading: 7x10-1, 7x10-2 and 7x10-3 %/minute.

Figure 23. Critical state strength and residual strength. Republished with
permission of ICE Publishing, from Skempton (1964); permission
conveyed through Copyright Clearance Center, Inc.

Regarding the requirement associated with the completion of


the particle breakage, if it exists, we deem it to not be relevant in
the case of fine-grained materials.
According to the above discussion of the behavior of clayey
materials, it is possible to assert that the critical state is applied to
these soils without actually accomplishing steady state behavior.
Therefore, the critical state is a less rigorous framework, but is
sufficiently satisfactory even for practical applications in fine-
grained materials.
Figure 21. Undrained response of Kawasaki clay at different rates of
Conceptually, the steady state definition is a better
deformation (after Nakase and Kamei, 1986).
representation of the fundamental behavior of soils. However, if
These results clearly show that the rate of deformation has an applied rigorously, a non-unique (rate dependent) steady state will
important effect on the stress-strain curves and on the undrained be obtained in the case of fine-grained soils, although in the case
strength; the higher the rate of deformation the higher the of sandy soils its application is more direct, because rate effect
undrained strength. However, the angle of friction mobilized at and re-orientation of platy particles are not particularly relevant to
the ultimate state or the critical state is not influenced by the rate these soils. Particle breakage may occur in sandy soils, and if it
of deformation. This means that the critical state line in the q-p´ occurs this effect is reflected in the resulting steady state line.
plane is unaffected by the rate of deformation, but the critical state Therefore, it may be concluded that the critical state only has
line in the e-log p´ plane is definitely influenced by the rate of two requirements: the ultimate state of a soil can be represented
deformation. For clayey soils, the systematic dependence of the by one straight line in the q-p´ plane and another in the e-log p´
undrained strength on the rate of strain has been reported by plane. This simplicity is probably the key to the practical
different researchers (e.g. Bjerrum 1969; Sheahan et al, 1996; applicability of the CSSM, even in fine-grained soils.
Díaz-Rodríguez et al, 2009; Chow et al, 2012). In the case of tailings, the applicability of either of these
On the other hand, particle orientation in platy particles of frameworks depends on whether behavior is closer to a fine-
clays has been identified as the cause of the development of a grained soil or to a granular soil. Considering that the most
“residual strength” that is reached at significantly large catastrophic failures that have affected tailings dams are
deformations and under very low rates of deformation (Bishop et associated with liquefaction, the tailings that behave as sandy
al, 1971; Lupini et al, 1981; Skempton, 1985). As an example, in materials are those that are of greatest interest in terms of their
Figure 22 the stress ratio-displacement curve obtained from a ring mechanical response when assessing physical stability of tailings
shear test is presented where, after about 30 days of shearing, a deposits.
residual strength associated with a friction angle, r = 8.6o , was A comprehensive study showing the robust framework
mobilized in Kalabagh clay. provided by the steady state of deformation was reported by
Ishihara (1993) using the Japanese standard Toyoura sand. In
Figure 24 a series of CIU triaxial test results are presented in terms
of stress-strain and effective stress-path curves for samples of
different void ratios consolidated at the same isotropic pressure of
1 MPa. The tremendous effect of the density is apparent in the
mobilized undrained residual shear strength, which in this case
varies from zero to 2 MPa, associated with soil responses from
contractive to dilative.

Figure 22. Development of residual strength in Kalabagh clay.


Republished with permission of ICE Publishing, from Skempton (1985);
permission conveyed through Copyright Clearance Center, Inc.

138
plotting the experimental data with a logarithmic horizontal axis.
For instance, Ishihara (1993) reports the e-p' projection of the SSL
(essentially the same as the CSL, as discussed previously) of
Toyoura Sand for p' values that exceed 3 MPa (Figure 26). When
p' is plotted on an arithmetic scale, the SSL is almost linear with
a slightly concave up shape. There is certainly no evidence of an
inflection point. Conversely, if p' is plotted on a logarithmic scale,
the data in Figure 26 plot as a distinctly curved SSL. In this case
Figure 24. CIU triaxial test results at a confining pressure of 1 MPa and at least (and likely in many more), the SSL (or CSL) curvature is
different void ratios (after Verdugo, 1992). not necessarily related to particle breakage. This assertion is
backed up by results of a numerical study using the Discrete
Another series of CIU triaxial test results are presented in Element Method (DEM) reported by Nguyen et al (2018). They
Figure 25, where the samples were tested with the same void ratio, simulated drained and undrained triaxial compression tests to
e = 0.833, (after consolidation) and at consolidation pressures of determine the CSL of a collection of varying sized ellipsoid
0.1, 1, 2 and 3 MPa. These experimental results show that for a particles. The resulting particle size distribution of the particles
given void ratio, the undrained strength mobilized at large strains had a d50 = 1mm and a coefficient of curvature of 1.5, producing
is unique regardless of the initial confining pressure. Additionally, a simulated material not unlike a uniformly sized tailings material
the effective stress-paths vary from dilative to contractive (aside from the particle shape). The key aspect of this work is that
depending on the initial confining pressure with respect to the particle breakage was not allowed. Nevertheless, the distinctly
ultimate state, or steady state of deformation. curved CSL shown in Figure 27 was produced; another indicator
that curvature of the CSL may have nothing to do with particle
breakage.

Figure 25. CIU triaxial test results at different confining pressure and same
Figure 26. Steady state projections of Toyoura sand (after Verdugo, 1992).
void ratio after consolidation. Republished with permission of ICE
Publishing, from Ishihara (1993); permission conveyed through Copyright
Clearance Center, Inc.

4.3.3.2 Implications of a curved CSL


Although it is often reasonable to use the semi-logarithmic
representation of the CSL (equation 11), accounting for curvature
of the CSL, when it clearly occurs, does have important
implications. As hypothesised by Robertson (2017), tailings
exhibiting a curved CSL are less brittle at higher effective stresses
than at low stresses. When sheared undrained, an element of loose
tailings at a high mean effective stress, p´, will decrease to a lower,
but still substantial value of p´, whereas a loose element at an
initially low value of p´ reduces to a very small, perhaps even zero
value, i.e. true liquefaction.
Curvature of the CSL has often been attributed to particle
breakage (Bedin et al, 2012 and Schnaid et al, 2013, among
others). Valenzuela (2015) suggests that at pressures (presumably
vertical stresses) greater than 1MPa, breakage of the more angular Figure 27. Curved CSL derived from triaxial simulations (using DEM) of
a collection of ellipsoidal particles (after Nguyen et al, 2018).
edges of tailings particles occurs. However, the curvature of the
CSL may, in many cases, be nothing more than an artefact of

139
4.3.3.3 The q-p' projection of the CSL. CSL variability. Admittedly, such under estimation is unlikely to
Selection of an equation to model the q-p' projection of the CSL be significant if at least one intermediate gradation is also tested.
is straightforward because unequivocally the data define a In contrast to adopting particle size distribution to guide the
proportional trend: sample selection process, the analysis of a database of over 150
soils suggests that, regardless of silt content, the minimum void
ratio (emin) is a suitable index property to assess the variability of
CSLs in non-plastic soil deposits including tailings (Torres-Cruz
= (13) 2019, Torres-Cruz and Santamarina 2020). Namely, emin
correlates linearly to p* and to the upper bound of 10. The
determination of emin in tailings may seem an unusual proposition
where the slope M is termed the critical state friction ratio. As M because standardised emin protocols are generally designed for
depends on particle shape (Cho et al. 2006, Sadrekarimi and Olson sands, as opposed to the often silt-dominated gradations of
2011), it is conceivable that the CSL may deviate from the simple tailings. However, there are numerous precedents in the technical
linear trend proposed by Equation 13 when the stresses are high literature that support the reliability of emin determinations in silts
enough to induce particle breakage. However, we are not aware (e.g. Zlatovic and Ishihara 1995, Cubrinovski and Ishihara 2002,
of any practical cases in which this has been observed. Thevanayagam et al. 2002, Papadopoulou and Tika 2008, Yang et
M depends on the magnitude of the intermediate stress al. 2006, Carrera et al. 2011, Torres-Cruz and Santamarina 2020).
relative to the other two principal stresses. Typically, the triaxial
2 3) condition is taken as a reference and the 4.3.3.4 CSL Experimental procedures.
resulting slope denoted Mtc. This reference condition can then be Considerable effort has been devoted to refine the procedures used
scaled to other stress conditions (Jefferies and Been 2015). to measure the CSL. The following are some of the most salient
aspects of the available guidance.
4.3.3.4 Sample selection
CSLs depend on intrinsic particle characteristics such as a. In principle, both dilatant and contractive specimens
mineralogy, particle shape, and particle size distribution. approach the same critical state line (Been et al. 1991,
Specifically, the e-p' projection is sensitive to variations in any of Verdugo and Ishihara 1996). However, dilatant specimens
these characteristics while the q-p' projection is only sensitive to are prone to strain localisations that imply the routinely
variations in mineralogy and particle shape (Sadrekarimi and measured global void ratio of the specimen may not be
Olson 2011, Li et al. 2015, Torres-Cruz 2019). Investigations into representative of the localised zone of the specimen that
the variability of CSLs within a single tailings deposit generally reached critical state (Desrues et al. 1996). Accordingly,
show that the different gradations exhibit different e-p' CSL determinations should rely on contractive specimens
projections, but essentially the same q-p' projection (Fourie and (Been and Olivera 2015).
Papageorgiou 2001, Carrera et al. 2011, Li and Coop 2019, b. The large shear strains required to reach the critical state
Torres-Cruz and Santamarina 2020). The positive practical destroys the effect of the initial fabric of the specimen (Been
implication of this is that any potential variations in mineralogy et al. 1991, Verdugo and Ishihara 1996). Indeed, numerical
and particle shape within the deposit are sufficiently constrained simulations indicate that, for a constant intermediate
to prevent a significant effect on the q-p' projection of the CSL.
principal stress ratio, soils reach a constant fabric at critical
This seems reasonable in TSFs that receive tailings from a single
state (Zhao and Guo 2013). Consequently, the CSL is
plant or at least from a single ore body. On the negative side, there
considered independent of the initial fabric (i.e. preparation
is the implication that thoroughly characterising the e-p'
projections of the CSLs of any given tailings dam requires testing method and consolidation stress path) of the specimen. The
multiple gradations. This immediately leads to the question: what preferred preparation method becomes the one that provides
is a robust approach to select the samples whose CSLs will be greatest control over the initial void ratio, as opposed to the
experimentally determined? This requires careful consideration one that best mimics the in-situ depositional environment.
because CSL determination is a fairly resource intensive The moist tamping method has proven very effective for this
enterprise, with each point of the CSL requiring one triaxial test purpose (Been and Olivera 2016). It is worth noting, though,
with careful void ratio measurements (Been et al. 1991, Verdugo that the thin layering that often occurs during hydraulic
and Ishihara 1996, Reid et al, 2021). The investigations into recent deposition may lead to differences between the CSLs of
tailings dam failures suggest that, for practical purposes, CSLs are undisturbed and moist-tamped specimens (Verdugo, 2009,
defined with anywhere between 4 and 8 points (Morgenstern et al. Reid and Fanni, 2022). We also note that although the
2016, Jefferies et al. 2019, Robertson et al. 2019). Occasionally, undercompaction method (Ladd 1978) is often used to
even more tests may be required to achieve a well-defined CSL. promote greater density uniformity across the specimen, the
Clearly, samples must be chosen carefully before committing to method does not appear to have a systematic effect on the
such comprehensive testing. resulting CSL (Reid et al. 2021).
Ideally, the gradations selected for CSL testing should enable c. Both drained and undrained specimens are useful in CSL
assessment of the variability of CSLs within the tailings deposit. determinations (Been et al. 1991, Verdugo and Ishihara
As the focus is on the variability of the e-p' projection, and 1996, Been and Olivera 2016). In general, the low stress
assuming Equation 11 is valid, the selected gradations should portion of the CSL can be more easily determined with
disclose the variability of p* and 10 within the tailings deposit. undrained specimens, while the CSL at higher stresses can
A common approach is to select samples that span the be readily determined using drained tests. Once M is known
predominant range of gradations in the deposit, from the finest to from initial tests, drained tests have the advantage of
the coarsest material (Jefferies and Been 2016, Jefferies et al. allowing estimates of the initial consolidation stress p'i
2019, Robertson et al. 2019). In plastic tailings, this approach required to reach a target p' value at critical state. Indeed,
seems reasonable because the finer material will typically be more for a specimen that is initially consolidated anisotropically
10
to a stress state (p'i, qi), tested along a straight effective
(Schofield and Wroth 1968). However, the silt content of non- stress path with a slope sl, and which aims at having a mean
plastic soils does not have a systema
effective stress p'cs at critical state; the required value of p'i
exhibit a monotonic correlation with p* (Thevanayagam et al.,
is given by:
2002). Accordingly, testing only the coarsest and finest gradation
of a tailings deposit may lead to a systematic under estimation of

140
+ g. The large strains required to reach a critical state (i.e.
= (14) constant e, p', and q) imply that triaxial tests often do not
reach true critical state conditions, or perhaps steady state
conditions, as discussed in section 4.3.3.1 (Verdugo and
For the common case of isotropic consolidation (qi = 0) and Ishihara 1996, Murthy et al. 2007). Accordingly, assessment
shearing under constant cell pressure (sl = 3), equation 14 of critical state conditions may be slightly improved by
becomes: extrapolating triaxial test results to conditions of zero
volume change or zero pore water pressure change, for
drained and undrained tests, respectively (Carrera et al.
(3 ) 2011). While the extrapolations are logically sound, there
= (15)
3 appear to be no systematic investigations into the change of
CSL parameters that result from performing these
extrapolations.
Equations 14 and 15 are particularly useful when running
tests that aim at populating specific portions of the p' 4.3.3.5 Reproducibility of CSL measurements
domain of the CSL. As the engineering of tailings dams becomes increasingly reliant
d. Reliable void ratio measurements are essential for CSL on critical state soil mechanics concepts, it becomes critically
determination. Given the difficulties in measuring the void important to assess the reliability with which the CSL can be
ratio changes that occur during specimen saturation, the measured. For the practicing engineer, the question becomes one
recommended approach is to determine the void ratio by of assessing the level of agreement that can be expected amongst
measuring the water content at the end of the test. This can the CSLs determined by different laboratories. That is, the
be accomplished by freezing the specimen at the end of the reproducibility of CSL determinations. The issue of CSL
test (Sladen and Hanford 1987, Been et al. 1991), or by reproducibility was recently investigated by means of a round
increasing the cell pressure at the end of the test to squeeze robin testing scheme in which 15 soil laboratories, both
out as much water as possible, thus making the specimen commercial and academic, from around the world tested the same
easier to handle upon disassembly of the triaxial cell tailings (Reid et al. 2021). The laboratories were instructed to
(Verdugo and Ishihara 1996). determine the CSL of a sample of sandy silt gold tailings using
e. The use of enlarged and lubricated end platens has been between 4 and 7 triaxial tests and focusing on the p' stress range
recommended to promote constant radial strains along the between 50 and 800 kPa. No additional instructions on how to
height of sand specimens (Been and Olivera 2015). measure the CSL were given to the laboratories. As such, the
round robin programme provided unique insights into how the
However, these modified platens are not entirely effective
variability in CSL testing protocols affect reproducibility. Herein
(Rees 2010) and we are not aware of any investigation
we merely comment on the differences in reproducibility of the e-
showing that they influence the resulting CSL. Quite the
p' and q-p' projections.
opposite, an investigation into CSL reproducibility showed
that enlarged and lubricated end platens do not significantly
affect the CSL (Reid et al. 2021).
f. Detailed descriptions of CSL experimental procedures
generally do not address the connection between the loading
ram and the top cap (Been et al. 1991, Verdugo and Ishihara
1996, Been and Olivera 2015). However, a recent study has
specifically recommended the adoption of an embedded
connection between the top cap and loading ram as
illustrated in Figure 28 (da Fonseca et al. 2021). The
rationale for recommending this type of connection is that it
eliminates tilting of the top cap and misalignment between
the loading ram and the specimen. This type of connection
is different from the more widely adopted connection in
which the top cap is allowed to rotate (Reid et al. 2021).
Importantly, the embedded connection appears to yield
slightly, but consistently, higher values than those yielded
by rotating connections (da Fonseca et al. 2021, Reid et al.
2021). This issue thus appears to be worthy of further
investigation.

Figure 28. Loading ram embedded into top platen: a) Sketch and b)
Figure 29. e-p' data points from a CSL testing round robin exercise. a) All
photograph. Source: da Fonseca et al. (2021).
13 laboratories with available data and b) 11 screened laboratories. Data
source: Reid et al. (2021).

141
Figure 29 shows the critical state e-p' data points measured by CSLs, the effect of the scatter can be formalised by computing the
the 13 laboratories whose detailed results were included in the standard error of p* and of 10 which provide a statistical
supplementary files of Reid et al. (2021). As might be expected, assessment of the precision of p* and of 10 (Torres-Cruz and
the scatter of the dataset as a whole is significantly larger than the Santamarina 2020, Reid et al. 2021). Since the data points are
scatter of the data of any individual laboratory. Indeed, most of centred around p' = 100 kPa, we adopt p* = 100 kPa, and therefore
the data points fall within a band that has a height of = 0.1. we work with 100. As shown in Figure 31, the two data points of
Reid et al. (2021) noted that three of the laboratories adopted the Fugro dataset with the largest values of p' are excluded from
procedures that are known to significantly compromise the quality the regression as they may suggest a curvature that cannot be
of CSL determinations, and one laboratory generated a clearly captured with Equation 11. The resulting standard errors of 100
outlying dataset. On this basis, Reid et al. (2021) excluded the data (SE- 100) and of 10 (SE- 10) are shown in Figure 31. As expected
from these four laboratories from further analysis and focused on from the scatter, the values of SE- 100 and of SE- 10 of the BGC
the 11 datasets shown in Figure 29b. The screening reduced the dataset are lower than the corresponding values of the Fugro
scatter in the data, with most data points now falling within a band dataset. That is, the BGC dataset enables more precise estimates
that has a height of = 0.07. However, the data suggests that of 100 and 10.
there is still room for improvement, as single laboratories are
generally able to measure CSLs with a scatter of less than =
0.02. See for example the data from UWA, BGC, Golder Perth, 0.8
KCB, UPorto, and Golder Vancouver in Figure 29. Another
important lesson to be taken from this data is the need for BGC:
consultants to query the experimental procedures used by 100 = 0.63, SE- 100 = 0.002

laboratories since deviation from good practice can result, as 0.7 10 = 0.082, SE- 10 = 0.004

detected by Reid et al. (2021), in spurious CSLs.


Figure 30 shows all the available critical state points in q-p'
space. Even without excluding data from any of the laboratories,
there is much better convergence in q-p' space than in e-p' space 0.6

Void ratio e
(compare Figures 29a and 30). Since there are no void ratio Fugro:
calculations involved in the q-p' projection of the CSL, it appears 100 = 0.60, SE- 100 = 0.008
that the lower reproducibility of the e-p' CSL projection is due to 10 = 0.044, SE- 10 = 0.012
0.5
difficulties in reliably measuring the relevant volumes, and the
associated changes, of the specimens.
Excluded from
0.4 regression
3000 BGC
UWA BGC
UC Lab A Fugro All
Golder Perth KCB 0.3
2500
UPorto UofT 1 10 100 1000 10000
PUC UC Berkeley
Fugro Golder Vanc. Mean effective stress p' [kPa]
Deviator stress q [kPa]

2000 UTFSM
Figure 31. CSL data points that exhibit different degrees of scatter. Data
source: Reid et al. (2021).
1500
Comparisons between SE- 100 and SE- 10 require
normalisation of the values. A possible normalisation is to divide
1000 SE- 100 and SE- 10 100 10,
respectively, that can be observed in soils in general. This yields
the normalised values NSE- 100 and NSE- 10 (Torres-Cruz and
Santamarina 2020). In the absence of a database that covers the
500
full spectrum of soil types, we rely herein on a CSL database of
160 non-plastic soils (Torres-Cruz 2019, Torres-Cruz and
Santamarina 2020). This database suggests that, for non-plastic
0
100 10 varies from
0 500 1000 1500 2000
virtually 0 to 0.3 (range of 0.3). Previous analyses had put the
Mean effective stress p' [kPa] 100 at 0.2 (Torres-Cruz and Santamarina 2020,
100
Figure 30. All available q-p' data points from a CSL testing round robin
numerical models, not in actual soils. So the slightly revised lower
exercise. Data source: Reid et al. (2021).
bound of 0.3 is adopted herein.
The NSE- 100 and NSE- 10 of the BGC and Fugro datasets are
4.3.3.6 Confidence in estimated CSL parameters. plotted in Figure 32 along with the values that correspond to 91
An issue that is often overlooked in the determination of CSL CSLs analysed by Torres-Cruz and Santamarina (2020). The data
parameters is the confidence with which they can be estimated show that the NSE- 10 is always larger than NSE- 100. That is, the
considering the scatter of the underlying points. This is 100 estimates is better than the relative
particularly relevant to the e-p' projection of the CSL since this is 10 estimates. Indeed, most of the data points indicate
the projection prone to greater scatter (compare Figures 29a and that NSE- 10 is between three to twelve times NSE- 100.
30). To illustrate this point, consider two of the datasets from the
CSL round robin programme mentioned above (Reid et al. 2021):
BGC and Fugro. Figure 31 shows the critical state data points of
these two datasets in e-p' space. Less scatter is apparent in the
BGC dataset, suggesting that these data points lead to a more
precise definition of the CSL. Adopting Equation 11 to model the

142
Figure 32 100 10. Data
sources: Torres-Cruz and Santamarina (2020) and Reid et al. (2021).

Figure 32 provides a useful benchmark against which to


assess the effect of scatter on the fitting parameters of Equation
11 100 10. In Figure 32, the closer to the lower left corner a Figure 33. Illustration of the effect of assumed residual strength ratio on
100 10. the resulting cost of a stabilising buttress (after Sarantonis et al, 2020).
However, it is important to keep in mind that precision and
accuracy do not always go hand in hand. That is, low scatter in the There are two generic approaches to determining the residual
e-p' points does not necessarily mean that the CSL has been shear strength of a material. One is the use of empirical
measured correctly. This point was also illustrated in the Reid et correlations based on back-analyses of documented failures, and
al. (2021) round robin exercise in which several CSL entries were the other is direct measurements using either in-situ tests or
correctly disregarded due to lack of adherence to proper testing laboratory tests (sometimes both). There are strong advocates of
protocols, regardless of scatter. Clearly, accuracy should take both generic approaches, e.g. advocates of the former insist that
precedence over precision. As the advice of the statisticians goes: void ratio localisation in most laboratory element tests render
it is better to be approximately right, than exactly wrong. interpretation of such results potentially misleading (Weber,
2015). Laboratory testing advocates point to the enormous
4.3.4. Residual strength of tailings. uncertainties in mobilised shear strength in a failing mass, e.g. due
As already mentioned, there is a growing call to evaluate the to varying rates of shearing, together with many of these analyses
stability of existing TSFs on the basis of residual (large strain) not accounting for effects of momentum when evaluating the
strengths, when potentially liquefiable tailings exist in the zone of shear strength that is required to produce the observed failed
the TSF that may be incorporated in a potential failure. If geometry. Additionally, Robertson (2010) pointed out that many
liquefiable tailings only exist in the centre of a TSF, distant from failure case histories involved retrogressive sliding, the
the retaining perimeter embankment(s), then this issue is Merriespruit TSF failure in South Africa in 1994 being a good
example (Fourie et al., 2001), which is rarely accounted for in the
irrelevant. What is of concern is the presence of potentially
backanalyses. Evaluation of the liquefied shear strength based on
liquefiable tailings near the perimeter. The emerging approach is
such examples is thus questionable. From results presented by
thus, ‘if the tailings can liquefy, assume they will’. For example,
Been and Jefferies (2006), estimating residual strength ratios
some time ago, Silvis and de Groot (1995) suggested that based on backanalyses of failures provides a (sometimes very
triggering should always be assumed if the soils (tailings) are significantly) conservative estimate when compared with
susceptible to strength loss, and Morgenstern (2018) strongly predictions using laboratory test data and a critical state
counselled along similar lines, suggesting, “that liquefiable constitutive model. Since the work of Been and Jefferies (2006)
deposits that can liquefy be assumed to do so and that containment there have been further attempts to reconcile backanalysed
be provided by a buttress of non-liquefiable unsaturated tailings strengths with laboratory test data, but a unified approach appears
and/or compacted dilatant material”. to remain elusive. Perhaps the complexity of the stress-strain
Emphasis is moving away from trying to engineer out all behaviour of very brittle, loose soils and tailings and the attendant
plausible mechanisms that may trigger liquefaction and rather to implications of realistically analysing CPTu data, as discussed
assume that some unknown (even potentially unknowable trigger) later, will make such a rapprochement impossible, at least in the
will occur at some time. In many cases this might simply be the short to medium term.
MDE (maximum design earthquake) or even lower magnitude Laboratory data, primarily from triaxial compression tests and
earthquakes. However, the cost implications of such an approach direct simple shear (DSS) tests are currently widely used in
can be significant, with the authors aware of many recently practice for residual strength assessments. Some examples, drawn
completed and currently under construction buttresses that cost in from experience with copper tailings in Chile are provided here to
the order of tens-of-millions of USA dollars. Furthermore, as illustrate the range of factors to consider when testing.
demonstrated so well by Sarantonis et al (2020), the importance Figure 34 presents values of the residual undrained strength
of the assumed post-liquefaction strength ratio on the resulting as a function of the void ratio for copper tailings, considering
buttress cost is very significant. They provide a summary of their different fines contents (where ‘fines’ refer to particles finer than
analyses of a hypothetical TSF, buttressed using rockfill, and plot highlight the effect of particle size
the cost (in millions of dollars) per 100m length of buttress against distribution on mechanical response.
the assumed liquefied (residual) strength ratio (Figure 33). As
noted by Sarantonis et al (2020), as the residual strength ratio is
varied between 0.05 and 0.2 (a plausible range), there is a three-
fold increase in cost. Clearly, accurate and reliable determination
of residual strength of tailings is a key challenge to the industry.

143
particularly by Dyvik et al (1987), is that changes in vertical stress
required to ensure constant volume conditions (i.e. no change in
height) are equal to the measured pore water pressure in truly
undrained simple shear tests. This means that tests in the DSS can
even be carried out on dry specimens and if constant volume
conditions are maintained (by adjusting the vertical total stress as
shearing progresses) the pore pressures that would be developed
during undrained tests can be determined.
The DSS device does, however, have drawbacks. In
particular, effective stress strength parameters cannot be correctly
determined as the principal stresses are unknown (and
unknowable explicitly). In fact, principal stresses rotate during
shearing, with the extent of rotation a function of the friction angle
(Budhu, 1984). Atkinson et al (1991) showed that the friction
Figure 34. Residual undrained strength of Chilean copper tailings. angle obtained in a DSS is smaller than that obtained from a
triaxial test, results confirmed by Dounias and Potts (1993).
These results are particularly relevant in applications that use Despite these limitations the DSS test has, for many years
cyclones to reduce the fines content in the sandy tailings, as used been widely used in Scandinavia in the design of TSFs, with the
in the construction of tailings sand dams, which is a widely view apparently being that it provides a conservative estimate of
practiced in Chile and Peru. shear strength (Knutson, 2018). However, this extra degree of
Although widely used in practice, the strain limitations of ‘safety’ cannot be quantified, and use of the triaxial test for
typical triaxial and direct simple shear devices are such that it is determining effective stress strength parameters for TSF design is
not always possible to reach steady state (or indeed critical state) now becoming more widespread in Sweden, particularly after the
conditions in these tests. There is also emerging evidence that comprehensive work of Bhanbro et al (2017) on sandy tailings,
boundary conditions, particularly in the DSS may be such that which clearly showed the extent of these differences.
post-peak behaviour measured in these devices is not Notwithstanding its limitations, the DSS remains a tool
representative. Although a ring shear device, particularly such as worthy of consideration particularly for determining undrained
described by Suits et al (2009), holds promise in determining strength ratios for the horizontal and near-horizontal sections of
residual strengths at very large strains, the authors were not aware potential failure planes in slope stability analyses. It is worth
of any measured undrained residual (or large strain) shear noting that DSS tests have been extensively used in many of the
strengths of tailings using such equipment. This we regard as an recent forensic evaluations of TSF failures (Morgenstern et al,
important research need. 2016, Jefferies et al, 2019).

4.3.5 Simple shear testing 4.3.5 Loading conditions other than triaxial compression and
Considering that a significant portion (often the majority) of simple shear
potential failure surfaces may be sheared under conditions much Although not common in tailings engineering, other loading
closer to simple shear, there is an overreliance on triaxial conditions that have been used include triaxial extension, plane
compression testing. Ladd (1991) showed how soil tested under strain and hollow cylinder torsional shear tests.
different loading conditions (e.g. triaxial compression, triaxial Triaxial extension tests provide strength data that are
extension and simple shear) resulted in different values of considered relevant to a relatively small length of a typical failure
undrained strength ratio. He presented results as a function of plane in a TSF slope stability analysis, typically the downstream
plasticity index (PI) and showed the divergence increased as PI tail of potential failure surface, see Sadrekarimi (2016). As shown
decreased. In other words, the difference will be greatest for by Ladd (1991), the undrained shear strength ratio in triaxial
non-plastic tailings, typically silty and sandy tailings, exactly the
extension is lower than both triaxial compression (TC) and SS,
type of material particularly prone to brittle behaviour if loose of
but difficulties in carrying out triaxial extension tests (typically
critical.
stress concentrations causing sample necking, which renders area
An alternative to triaxial compression tests, particularly to
characterise undrained strengths along horizontal or near- calculations nigh impossible, and some difficulties ensuring
horizontal sections of a potential failure surface was suggested as sample contact with the loading top cap) have usually meant that
being simple shear (SS) testing. In a true SS device, the sample is strength data from either TC or SS tests are used for this section
sheared under conditions of pure shear, with no complementary of the failure surface.
shear stresses on any of the boundaries. The Cambridge SS device Intuitively, a plane strain loading apparatus is highly suited
as described by Roscoe (1953) comes close to imposing true for determining shear strengths relevant to TSF slope stability
simple shear loading conditions. However, practicality has analyses, as these are usually well represented using a two-
resulted in other SS devices being developed, with pioneering dimensional simulation. However, plane strain devices are not
work completed at both the Norwegian Geotechnical Institute common, and usually only available in research institutions. Much
(NGI) and Swedish Geotechnical Institute (SGI). In these devices, valuable work on plane strain testing of static liquefaction of loose
a cylindrical specimen is confined in the horizontal direction by sands (not specifically tailings) has been carried out by
either a reinforced rubber membrane or a stack of thin, teflon Wanatowski and his co-workers (e.g. Wanatowski and Chu, 2007,
rings. The intention is to provide restraint while assembling a 2011).
specimen, but minimise confinement during application of shear. The most versatile loading apparatus is the torsional shear
These devices, and various modified versions, are commonly hollow cylinder (TSHC) (often referred to simply as a hollow
referred to as direct simple shear (DSS) devices, a term that is used cylinder device), although its use for testing tailings has been
in this paper. limited. The TSHC can reproduce a range of stress paths,
In a DSS test the known stresses are the (imposed) vertical including TC, true SS and plane strain. A drawback is the cost of
stress and the shear stress developed on a horizontal plane during the device and the complexity of operation (particularly compared
the shearing process. The shear strength along a horizontal plane with SS and triaxial devices). However, some suppliers are now
through the specimen is thus determined (i.e. the undrained shear producing compact and fully automated devices. Given the
strength) and the vertical total stress is known. An important importance of evaluating a range of possible stress paths for
assumption, which has been verified by several researchers, but

144
stability of TSF slopes, we foresee much more widespread uptake Tailings Dams failed due to the Valparaiso Earthquake which
of the TSHC in the next decade, if not sooner. occurred on March 3, 1985, of Mw = 8.0 (Castro and Troncoso,
Section 4.3 has focused on various approaches to determining 1989), the Antiguo Tailings Dam failed due to the Punitaqui
the shear strength of mine tailings under static loading conditions. Earthquake on October, 1997, Ms = 7.0, generating a flow of
More detailed discussion on the importance of various stress paths nearly 60 000 m3 of tailings (Villavicencio et al, 2014), the Las
followed during testing is provided in section 6, where we briefly Palmas Tailings Dam collapsed causing 4 casualties during the
discuss some of the results from well-documented forensic Maule Earthquake, Mw = 8.8, which occurred on February 27,
investigations. 2010. It is important to mention that all these tailings dams were
constructed using the upstream method.
4.4 Shear strength under dynamic loading The tailings dams that have experienced seismic failures have
been grouped considering the earthquake magnitude in
The experience gained in regions of high seismicity clearly shows
combination with the distance to the epicenter (Conlin 1987). For
that tailings dams are sensitive to earthquakes; in fact, several
seismic failures have been reported, as discussed below. However, earthquakes of moment magnitude greater than 8.0, the epicenter
almost all of these seismic failures occurred in tailings dams can be a large area (rupture zone), with a more representative
constructed with the upstream method (Lyu et al., 2019). In this measure being the minimum distance to the rupture surface.
regard, it is important to note that properly designed tailings dams Figure 36 summarises data from failures due to earthquakes,
constructed with the downstream or centerline methods have providing guidance on when potential seismic failure may occur
experienced stable behavior under strong seismic loading (Valenzuela et al, 2020) and, accordingly, special attention must
conditions (Valenzuela et, 2020). be considered in the design of TSFs.
Figure 35 shows the location of the world's mining areas and
of recorded earthquakes. It is evident that there are many countries
that have an important mining industry and at the same time have
active seismicity. Among these countries are: New Zealand,
Indonesia, Japan, China, Canada (west coast), United States (west
coast), Philippines, Mexico, Colombia, Peru and Chile).

Figure 36. Seismic failures of tailings sand dams (after Valenzuela et al.
2020).

On the other hand, there are a significant number of countries


with an important mining industry and with a relatively negligible
seismicity, for example Brazil, Australia, many African countries
and several European countries. In these countries, with low or
practically zero seismicity, it is common to find TSFs constructed
with tailings that are in a contractive state, and that can be
seriously affected by disturbances that mobilize the undrained
response. There are many examples, such as the catastrophic
failures of Los Frailes in Spain, Samarco and Brumadinho in
Brazil, Merriespruit in South Africa and Cadia in Australia,
among many others. The Los Frailes and Cadia failures have
strong similarities; first, the foundation soil failed and generated
significant displacement that triggered liquefaction of the stored
Figure 35. Location of a) mining areas and b) major earthquakes (where tailings, which contributed significantly to the dam failure
black dots are magnitude >9, red dots magnitude between 8 and 8.9, mechanism (Gens et al, 2006; Jefferies et al, 2019). Thus, when
orange between 7 and 7.9 and yellow between 6 and 6.9). contractive materials are present, there is an elevated risk of
failure if the residual undrained resistance is triggered, and the
In Japan, the failure of Dikes 1 and 2 of Mochikoshi TSF were stability analyses have not considered the possibility of the
associated with the 1978 Izuoshima-Kinkai Earthquake (Ishihara, mobilisation of this resistance.
1984). More recently, also in Japan, the seismic failures of the
Kayakari, Zenigame and Gengorou Tailings Dams were due to the 4.4.1 Liquefaction in terms of Cyclic Mobility.
2011 Great East Japan Earthquake of Mw = 9.0 (Ishihara et al, The term liquefaction was coined by Hazen (1920) to describe the
2014). In Chile, several tailings dams have experienced collapse failure of the hydraulic fill sand of the Calaveras Dam on 24
due to seismic disturbances: the Barahona Tailings Dam suffered March 1918. In this failure, the upstream toe of the Calaveras
a breach due to liquefaction on December 1, 1928, induced by the Dam, which was under construction near San Francisco,
Mw = 7.6 Talca Earthquake (Aguero 1929), the El Cobre Tailings California, failed suddenly, and approximately 700,000 m3 of
Dam failed catastrophically due to the La Ligua Earthquake, Mw material flowed for around 90m. Apparently no special
= 7.4, on March 28, 1965, causing more than 200 casualties disturbance was noticed at the time of the failure, indicating that
(Dobry and Alvarez, 1967), the Cerro Negro and Veta del Agua this phenomenon can occur in the absence of earthquakes. Since

145
that failure, the term liquefaction has been used in a broad sense amplitude of cyclic stress. Figures 38 and 39 present typical
to describe two different phenomena that may occur in saturated results of CRR curves obtained in tailings samples, considering a
cohesionless soils. These phenomena have in common a strain level criterion. Figure 40 presents a CRR curve considering
significant pore-pressure buildup and large ground deformations. the 100% pore water pressure build-up criterion.
Nevertheless, to understand the actual soil behavior, it is
important to distinguish between the so-called flow failure, or
static liquefaction, where a sudden loss of strength takes place,
and the term cyclic mobility, which is associated with a
progressive strain softening with cycles of loading, without any
loss of strength.

Figure 38. Cyclic stress ratio for ±2.5% axial strain vs number of cycles.
Tailings retrieved from Kayakari dam after the 2011 earthquake.
Reprinted from Ishihara et al. (2014), with permission from Elsevier.

Figure 37. Undrained cyclic response of loose (upper plots) and dense
(lower plots) sandy soils (Ishihara 1985).

Typical results of undrained cyclic torsional shear tests are


presented in terms of both stress-strain curves and effective stress-
paths in Figure 37 (after Ishihara 1985). In the case of the loose
specimen (Dr = 47%), there is a gradual migration of the effective
stress-path toward the origin, which is more pronounced in the
first cycle and after the phase transformation line is met. When the
cyclic effective stress path touches the phase transformation line,
a significant change in the cyclic response takes place. During
loading, the effective stress path turns upwards and to the right,
indicating a dilative response, while during unloading it turns left,
toward the origin, which indicates a strong contractive behavior Figure 39. Cyclic resistance ratio at = 3.75% obtained in cyclic simple
associated with a large increase in pore water pressure. Once the shear for gold tailings (James et al, 2011).
phase transformation line is crossed, this phenomenon is repeated
and at each cycle of loading and unloading the effective stress path
moves upward and downward closely along the failure line until
eventually it starts to pass though the origin, indicating a condition
of zero effective stress at the instant of zero acting shear stress.
That is, a pore water pressure build-up equal to the confinement
stress only occurs when the shear stress is zero. Accordingly, this
condition is momentary and does not induce failure of the soil.
Although this condition is associated with zero effective stress, it
should not be considered as a condition of zero shear resistance.
As soon as there is an acting shear stress, the pore water pressure
decreases and the shear resistance of the soil is again mobilised.
The same general pattern is observed for the sample with Dr
= 75%, except that the phase transformation line is crossed during
the first cycle and thereby the large change in the effective mean
stress occurs much earlier than in the specimen of loose sand.
The cyclic responses outlined above can be classified as
"cyclic mobility" according to Casagrande (1970; 1975). This is Figure 40. Cyclic resistance ratio for 100% pore water pressure build-up,
because they only involve degradation of stiffness, not strength, in simple shear for tailings samples retrieved with Shelby tube (Been
which can be significant or moderate depending upon the density 2016).
of the sandy soil and the level of the applied cyclic loading, among
other factors. 4.4.2 Effect of fines content on cyclic resistance ratio.
If the specimen is strained sufficiently, there is a tendency for Experimental evidence indicates that the mechanical properties of
the soil to dilate and regain stiffness and strength. In this context, silty sands are largely controlled by the fines content and by the
when cyclic mobility is the phenomenon under evaluation, the plasticity of these fines (Troncoso et al., 1985; Polito et al., 2001;
cyclic strength is normally defined as the Cyclic Resistance Ratio Verdugo et al., 2004; Ni et al, 2004, among others). The results of
(CRR), which is evaluated as the number of cycles required to a series of cyclic triaxial tests performed on a copper tailings sand
reach a certain amount of shear deformation, or 100% of pore using samples prepared at the same void ratio, with different low
water pressure build-up, when a sample is subjected to a certain

146
plasticity fines content, are shown in Figure 41 (after Troncoso et Plasticity Index equal or greater than 7. However, in tailings
al., 1985). materials further research is needed to confirm this criterion.

Figure 43. Strain softening in clayey material. Republished with


permission of ASCE, from Boulanger and Idriss. (2006); permission
conveyed through Copyright Clearance Center, Inc.

Figure 41. Effect of low-plasticity fines on the cyclic strength of copper For practical purposes, fine-grained soils with a sand-like
tailings (after Troncoso et al, 1985). behavior can be treated as sands to assess their liquefaction
potential, so the empirical charts based on SPT and CPT are
A more comprehensive study was carried out by Verdugo and
applicable. On the other hand, fine-grained soils with a clay-like
Viertel (2004) using a copper tailings sand. The tailings sand was behavior could undergo strain-softening during the application of
mixed with various contents of low-plasticity tailings fines: 2; 10; cyclic loads. It is recommended to evaluate the potential
18 and 28%. For each mixture, different densities were selected. occurrence of strain softening by means of cyclic tests in the
The results are presented in Figure 42. laboratory. For fine-grained tailings this is considered the most
reliable procedure.

4.4.3 Accounting for earthquake magnitude.


Since the cyclic resistance to liquefaction is a function of the
number of loading cycles, it is necessary to establish the number
of cycles that corresponds to the design earthquake. Research
suggests that the effective number of equivalent uniform cycles is
predominantly related to the earthquake magnitude, as shown in
Figure 44. This issue is discussed in section 8.3.1.

Figure 42. Cyclic strength as a function of void ratio for different fines
contents (after Verdugo and Viertel (2004).

For the range of fines content tested, the cyclic strength


decreases consistently as the fines content increases, regardless of
the void ratio of the samples.
Extremely large volumes of tailings are produced from the
extraction processes of copper, gold and nickel. These tailings
vary mainly from sandy silt to silty clay, with the fine tailings
usually being non-plastic or of low plasticity. There are other mine
tailings that include significant fine-grained particles of medium
to high plasticity, such as tailings from the extraction of alumina
from bauxite ore (red mud), many iron ore tailings, and the oil Figure 44. Number of equivalent stress cycles vs earthquake Magnitude.
sand fine tailings resulting from the extraction of crude bitumen Reprinted from Idriss et al. (2004), with permission from Elsevier.
from oil sands, among others.
5 FIELD TESTING.
It has been recognized that fine-grained soils can undergo
liquefaction or significant deformation during earthquakes (Bray Concerns about the potential instability of upstream TSFs is
et al., 2004, 2006; Boulanger and Idriss, 2006). Depending on the rightly currently focussed on existing TSFs. Sampling of tailings,
type of fines, the cyclic behavior of these soils can be separated particularly cohesionless silty or sandy tailings (which constitute
into two groups: the "sand-like" and the "clay-like" fine-grained a very significant proportion of existing TSFs) without disturbing
soils. The first exhibits liquefaction susceptibility characteristics the very characteristic that is of interest, i.e. void ratio, is
similar to those observed in sandy soils. Although clay-like fine- extremely difficult. Although interesting work is currently
grained soils, when subjected to cyclic loading, produce some underway using, for example, a modified Sherbrooke-type
increase in pore pressure and can deform significantly, they do not sampling device, such techniques are far from commonplace or
liquefy like sandy soils, rather they undergo a cyclic softening, proven effective. Furthermore, even if truly undisturbed sampling
"strain-softening" phenomenon, as observed in Figure 43 were feasible, there would still be considerable uncertainty
(Zergoun et al., 1994; Boulanger and Idriss, 2006). A simple regarding the in-situ stress levels which also affect mechanical
criterion proposed by Boulanger et al., (2006) to identify fine- response. Such uncertainty particularly affects estimates of lateral
grained soils with a clay-like behavior is that they present a stresses (see Section 5.4). Recourse is therefore made to the use
of in-situ testing.

147
Until relatively recently, the SPT (standard penetration test) of soils. Furthermore, Ottawa Sand is a clean and uniformly
was used extensively for estimating the in-situ strength and even graded sand with D50 = 0.35 mm, sub-rounded particles, and
susceptibility to liquefaction of tailings. As with many procedures composed primarily of quartz (Cunning 1994). These
used in tailings, many of the SPT correlations in use were derived characteristics are significantly different from those of many
from work on essentially clean (low fines content) sands. With tailings which tend to be siltier, more broadly graded, with angular
recent rapid and significant advances in CPT (cone penetration particles, and possess a more complex mineralogy (Torres-Cruz
test) technology, the SPT is not recommended for tailings and Santamarina 2020). Figure 45 illustrates the effect that K0 and
characterisation. The primary methods currently in use are the particle grain characteristics can have on the contractive-dilative
CPT and field vane, with lesser but increasing use of techniques boundary as inferred from Fear and Robertson (1995). To this end,
such as full-flow penetrometers, dilatometers and very Figure 45 considers K0 = 0.5 and K0 = 0.8, and two distinctly
occasionally, pressuremeters. The discussion that follows is different sands: Ottawa Sand and Alaska Sand. The latter is an
skewed toward the two former techniques, with less coverage of angular and highly compressible carbonate marine tailings sand
the other methods, given their relative sparsity of use.
(Fear and Robertson 1995). It is worth noting that in Figure 45,
Equation 16 does not match the curve of Ottawa Sand at K0 = 0.5.
5.1 Cone Penetration Testing of Tailings.
This is due to differences in converting from shear wave velocity
The cone penetration test (CPT) is the most used in-situ test to to cone tip resistance between Fear and Robertson (1995) and
characterise tailings. Early versions of the CPT were limited to Olson and Stark (2003). The variety of contractive-dilative
measurements of tip resistance (qc) and sleeve friction (fs). boundaries in Figure 45 highlights the potential shortcomings of
Modern CPTs almost invariably include pore water pressure using Equation 16 as a generally applicable contractive-dilative
measurements (known as CPTu) and routinely include seismic boundary.
wave velocity measurements (SCPTu). More recently,
qc1 (MPa)
incorporation of P-wave (i.e. compressional wave) measurements
0 2 4 6 8 10 12 14
have become more common (thankfully with no further change to 0
the acronym). The most common location for the pore water
pressure sensor is on the shoulder of the cone (u2), although
alternative locations have also been used mostly for research
100
purposes (Lunne et al. 1997).
Given that the CPT does not perform direct measurements of
the mechanical behaviour of the soil, its use relies on
'v0 (kPa)

correlations between CPT results and the soil behavioural 200


aspects of interest. In tailings engineering, particularly in
assessments of liquefaction potential, the focus is generally on
the state, the peak (or yield) undrained strength, and the residual 300
(or liquefied) undrained strength. In this context, state refers to
whether the material has a contractive or dilative tendency when
subjected to shearing. Contractive materials are prone to 400
significant strength loss under undrained conditions whereas
dilative materials are not. Different methodologies are available Figure 45. Non-uniqueness of contractive- v0-qc1
to conduct these three stages of a liquefaction potential space. Notes: Boundaries of Alaska and Ottawa Sands as per Fear and
assessment. Herein we critically review some of the most Robertson (1995). Dilative soils are expected to plot to the right of the
common approaches. boundaries and contractive soils to the left.

5.1.1 Assessment of state Robertson (2010) proposed the criterion Qtn,cs = 70 as a


Olson and Stark (2003) proposed a generally applicable conservative contractive-dilative boundary (Figure 46), where
contractive-dilative boundary defined as Qtn,cs is the normalised clean sand equivalent tip resistance defined
.
below. This criterion was considered in the post-failure
= 1.10 × 10 ( ) (16)
investigation of the Fundao Dam (Morgenstern et al. 2016) and,
although not the preferred approach, in the post-failure
where v and qc1 have units of kPa and MPa respectively and v investigation of the Brumadinho tailings dam (Robertson et al.
has an upper bound of about 350 kPa. qc1 is the overburden- 2019).
normalised tip resistance defined in Olson and Stark (2002) as
1000
1.8
Normalised cone resistance Qtn

q = (17)
0.8 +
(Qtn - 11)(1 + 0.06F)17 = 70
100

where pa
have noted the limitations of using Equation 16 as a generally
applicable contractive-dilative boundary (Olson 2009, Robertson 10
2010). Notwithstanding, Equation 16 was considered in the post-
failure investigation of the Fundão Dam (Morgenstern et al. 2016)
Qtn,cs = 70
and it is thus deemed pertinent to address this equation herein. The 1
limitations of Equation 16 stem from its origins (Torres Cruz 0.1 1 10
2016). This boundary was initially proposed, in terms of SPT Normalised friction ratio F
results, by Fear and Robertson (1995) specifically for Ottawa
Sand under K0 = 0.5 stress conditions. That is, Equation 16 was Figure 46. Comparison of the contractive-dilative boundaries proposed by
not originally intended to be generally applicable to a wide variety Robertson (2010) and Robertson (2016), shown by the solid and dashed
lines respectively.

148
Development of the Qtn,cs = 70 criterion relied on a database This implies that, at the same Qtn, sands with fines are less likely
of 36 flow liquefaction case histories that were grouped into five to classify as contractive (Qtn,cs
different classes of CPT data quality. In descending order of rationale for attributing a greater liquefaction resistance to silty
reliability, the five classes were defined as follows. Class A sands than to clean sands was "the greater compressibility and
corresponded to cases with electric CPT results that included decreased permeability of silty sands" (Robertson and Wride
sleeve friction readings. Class B included cases with mechanical 1998). However, non-plastic soils, which include many types of
or electric CPT results but no sleeve friction readings. Classes C, tailings, do not exhibit a direct correlation between fines content
D, and E included cases for which CPT results were not available and compressibility as indexed by the slope of the critical state
but rather inferred from SPT, relative density, or otherwise line (Reid 2015, Torres-Cruz and Vermeulen 2018). Furthermore,
estimated, respectively. Robertson (2010) relied mostly on the even when containing significant amounts of fines, the
Class A case histories to justify the Qtn,cs = 70 criterion. permeability of non-plastic soils may remain high enough to
enable virtually drained (Bq
These observations suggest that the use of Kc in Equation 18 to
The normalised clean sand equivalent tip resistance Qtn,cs is
compensate for the assumed higher compressibility and lower
defined as
permeability of non-plastic silty tailings can lead to potentially
unconservative assessments of state.
, = (18) Robertson (2010) indicated that the case history database
agreed with the Qtn,cs = 70 boundary because most flow
where Qtn is the stress-normalised tip resistance liquefaction case histories, including all case histories with high
quality CPT data, exhibited Qtn,cs
however, is that since flow liquefaction case histories do not
provide information on the performance of dilative soils, the data
= (19)
supports the adoption of Qtn,cs = 70 as an upper bound for
contractive soils, without necessarily implying that it is also a
lower bound for dilative soils. That is, the case history data
qt is the cone tip resistance corrected for unequal area effects and
presented by Robertson (2010) is not sufficient to establish that
and n is a stress exponent capped at 1 calculated with the
Qtn,cs = 70 is indeed a boundary. We acknowledge, however, that
following equations (Robertson 2009):
without additional information, the cautious approach is to treat
Qtn,cs = 70 as a boundary. That is, to assume that all soils with Qtn,cs
= 0.381 ( ) + 0.05 ( ) 0.15 (20) indeed contractive.

Robertson (2016) updated the contractive-dilative boundary


to Equation 9 and noted that it is mainly applicable to soils with
little or no microstructure (e.g. cementation).
= (3.47 log ) + (log + 1.22) (21)
70 = ( 11)(1 + 0.06 ) (24)

Figure 46 compares Equation 24 to the Qtn,cs = 70 criterion.


= 100% (22) The two criteria bear strong resemblance for Fr
corresponds to values often measured in tailings deposits (Reid
2015). Accordingly, the objections raised herein to the Qtn,cs
criterion, are likely to remain valid to Equation 24 when applied
Ic is a soil behaviour type (SBT) index and F is the sleeve friction
to tailings. However, we note that Equation 24 is written in terms
ratio. Given the interdependence between Qtn, n, and Ic an iteration
of Qtn, as opposed to Qtn,cs. This reflects the fact that Robertson
process is required to find their final values. While Robertson
(2016) did not adopt Equations 18 and 23. We believe this is a
(2009) explained the rationale for Equation 20, we note the
step in the right direction given the difficulties noted above to
absence of supporting experimental data points that would enable
correlate fines content to compressibility and permeability.
an assessment of its reliability.
Robertson (2016) does retain Equation 20, however, for which we
noted the lack of validation data. Despite this limitation, we
The second factor in Equation 18, Kc, is defined as
believe that its agreement with the case history database presented
1.64, = 1.0 by Robertson (2016), makes Equation 24 a useful empirical
> 1.64, = 0.403 + 5.581 21.63 + 33.75 17.88 (23) criterion.
State is also commonly characterised by the state parameter
which is defined as the void ratio difference between the current
It is worth pointing out that although Equation 23 includes a state of the soil and the critical state line (CSL) at the same mean
fourth degree polynomial whose coefficients involve 19 effective stress p' (Been and Jefferies 1985). Positive values of
significant figures, the equation is approximate, was proposed indicate states that plot above the CSL (contractive), whereas
based on qualitative arguments and lacks any supporting data negative values of indicate states that plot below the CSL
points (Robertson and Wride 1998). Similar to Equation 20, the (dilative). It has been suggested that soils with -0.05 should
uncertainty of Equation 23 is thus not quantifiable. be deemed contractive (Shuttle and Cunning 2008). While
Equation 23 defines a direct correlation between Kc and Ic. In calculating for a laboratory specimen is relatively
general, coarse-grained soils take on low values of Ic whereas fine straightforward, inferring from CPT readings is anything but.
grained soils take on high values, with Ic = 2.6 considered an Been et al. (1987) first proposed a systematic framework to
approximate boundary between sand-like and clay-like soils
(Robertson and Wride 1998). Accepting this premise, soils with may be of limited use in tailings wherein drained penetration
high fines contents are expected to yield high values of Ic, which cannot be generally guaranteed. Plewes et al. (1992) suggested a
lead to high values of Kc, and consequently high values of Qtn,cs. simplified screening procedure to assess from CPT readings

149
while accounting for excess pore water pressures. The method, based estimates of are "broadly aligned." Improved agreement
which was slightly updated by Jefferies and Been (2016), relies is observed in the data from Jefferies et al. (2019), although
on a correlation between the normalised sleeve friction ratio F and unfortunately the data clusters in a small portion of the graph,
the slope of the critical state line CSL. However, recent additions limiting the potential of this data to validate the approach.
to the -F dataset indicate that the uncertainty of the correlation is Shortcomings in the widget method to characterise loose sands (
higher than reflected by the original dataset (Reid 2015). > 0) may stem from the fact that it is anchored on calibration
Furthermore, error propagation analyses show that even when the chamber testing programmes that involved, almost exclusively,
Plewes method is implemented with values measured in the dense sands ( < 0) (Shuttle and Jefferies, 2016) and sands that
laboratory, the variability of within tailings deposits impedes were both clean (low to zero fines) and nearly all quartz sands
estimates of that are accurate enough for detailed (Been, 2016).
characterisation (Torres-Cruz 2021). These observations highlight Appraisal of Figure 47 requires acknowledgement that the
the screening nature of the procedure. More rigorous methods to measured from a sample is not a perfect validation benchmark.
estimate the state parameter from CPT results can be broadly Shuttle and Cunning (2007) mention sample disturbance as a
divided into those that assume drained penetration and those that potential explanation for the mismatch reflected in Figure 47. This
assume undrained penetration. It is the latter that is often used in would have required the disturbance to consistently induce
tailings because, as noted above, drained CPT results cannot be dilation of the samples. However, it appears unlikely that samples
guaranteed at the standard penetration rate of 20 mm/s. Within this that were deemed contractive by both the widget method and by
undrained CPT category, the method that appears to have gained water content logging would have been prone to dilation due to
the widest acceptability is the one proposed by Shuttle and disturbance. This is borne out by the results presented by Mohajeri
Cunning (2007), further developed by Shuttle and Jefferies and Ghafghazi (2012) for a natural silt from Vancouver Island,
(2016), and also described in Jefferies and Been (2015). The Canada, that was sampled and handled with extreme care. Despite
procedure relies mostly on CPT results, in-situ measurements of
shear wave velocity (Vs), knowledge (or estimates) of lateral stress which corresponded to about one third of the void ratio difference
conditions, laboratory testing of disturbed samples, and numerical between the loosest and densest possible void ratios of this soil.
simulations of triaxial tests and of spherical cavity expansion. Shuttle and Jefferies (2016) estimated the of the high-quality
Because the spherical cavity expansion simulations can be samples by simulating relevant triaxial tests and determining the
performed with a publicly accessible widget, we will refer herein best fit . They noted that uncertainty was introduced because the
to this approach as the "widget method." in-situ void ratio of the triaxial specimen was taken as equal to the
The widget method was used in the investigation of the Cadia average void ratio over the 0.8 m length of the sampler. It is
NTSF embankment failure (Jefferies et al. 2019) and was unlikely, though, that this difference between average and local
regarded as the "most reliable" approach to assess the in-situ state void ratio can account for the systematic underestimation of
of the Brumadinho Dam (Robertson et al. 2019). As such, the observed in Figure 47. Jefferies et al. (2019) computed the of
method is likely to be viewed as the best alternative by many high-quality samples by comparing the sample void ratio to that
practicing geotechnical engineers. To assess the capabilities of the of the critical state line at the same mean effective stress. Two
method, we collected data from three different sources that report CSLs were considered because layering in the samples introduced
values inferred using the widget and values estimated from uncertainty regarding which was the applicable one. The
high quality samples. The comparison is shown in Figure 47, in horizontal bars in Figure 47 reflect this. Estimates of lateral
which the vertical bars are centred at the mean estimated by the stresses (see Section 5.4) are another source of uncertainty when
widget method over the length of the sampler. The upper and inferring from high-quality samples. Despite limitations in
lower ends of the vertical bars are placed at one standard deviation 47 suggests that there may be
from the mean. Of the three references considered in Figure 47, an unconservative bias in the widget method and is a reminder that
the length of the sampler was only reported in Shuttle and Jefferies the validation of this method in real silt deposits is a pending task
(2016) as 0.8 m. The same length was assumed for the other two for the geotechnical community.
references. It is worth noting that the three references considered in
Figure 47 used different versions of the widget method to estimate
0.4 . For example, Shuttle and Jefferies (2016) updated the mapping
Shuttle and Cunning (2007) of the limiting cavity expansion pressure to CPT resistance, and
0.3
Shuttle and Jefferies (2016) Jefferies et al. (2019) reported that the pore water pressure
Jefferies et al. (2019) measurement location considered by the widget changed from the
face of the cone to the shoulder of the cone. The widget version
from widget

0.2 used by Jefferies et al. (2019) made the predictions that plot
closest to the identity line in Figure 47. However, validation data
0.1 from additional case histories is clearly required to reliably assess
the predictive capability of the widget.
0 A recent analysis of the 1974 Tar Island slump in Canada aims
at validating the computation of liquefaction using finite elements
coupled with critical state soil mechanics (Shuttle et al. 2022).
-0.1
Although this analysis uses the widget method to infer from
-0.1 0 0.1 0.2 0.3 0.4
CPT measurements, it does not constitute a validation of the
from sample
widget because no comparisons are presented between as
measured in high-quality samples and predicted by the widget.
Figure 47.
and based on high quality samples. 5.1.2 Assessment of undrained yield strength
When sheared under undrained monotonic conditions, contractive
Figure 47 shows that for two of the datasets (Shuttle and soils mobilise a yield (or peak) strength Su that, upon further
Cunning 2007, Shuttle and Jefferies 2016) the widget method shearing, drops to a residual value Sr (Verdugo and Ishihara 1996,
significantly and systematically underestimates the Olson and Stark 2003). It is common to normalise the yield
contractiveness of the samples. This is at odds with the undrained strength Su by the pre-failure vertical effective stress
observation by Shuttle and Jefferies (2016) that CPT and sample- v, thus leading to the yield strength ratio (YSR). When a

150
contractive soil reaches its YSR, liquefaction is triggered and the 5.1.3 Assessment of residual undrained strength
strength of the soil drops to its residual value Sr. Static loads, Olson and Stark (2002) presented an analysis of residual (or
deformations, or dynamic loads can all move a soil towards its liquefied) strength Sr based on the same database of 33 flow
YSR and, importantly, reaching the YSR does not necessarily failure case histories considered by Olson and Stark (2003) for
require the loading conditions to reach Su (Olson and Stark 2003). their yield strength analysis. They identified 21 case histories with
Yield strength assessments are often disregarded in liquefaction enough information to assess the mobilised residual shear strength
potential assessments because designers and owners adopt the by means of a rigorous back-analysis which on occasions included
cautious view, warranted in our opinion, that if liquefaction can kinetic considerations. Based on this data they proposed a
be triggered (soil is saturated and contractive) then it will occur liquefied strength ratio (LSR)-qc1 correlation which they
(e.g. Robertson 2010). This appears to have led to much more complemented with upper and lower limits to account for
emphasis in the literature on the liquefaction susceptibility and variability. Figure 49 presents these 21 case histories in LSR-qc1
residual strength analyses than on the yield strength analysis. One space and shows that there is only a weak (R2 = 0.21) linear
of the few methods available to assess the yield strength ratio of correlation. That is, 79% of the variation in LSR is unaccounted
contractive soils was proposed by Olson and Stark (2003). This for by qc1. Furthermore, of the 21 case histories, CPT readings
approach was considered in the post-failure investigation of the were only available for seven of them. For the remaining 14 case
Brumadinho tailings dam (Robertson et al. 2019). histories, qc1 was assessed from correlations to SPT, relative
Olson and Stark (2003) back analysed 33 flow failure case density, or otherwise estimated. As noted above, this inevitably
histories involving mostly non-plastic and low plasticity soils. introduces uncertainty. It is also apparent from Figure 49 that the
They identified eight case histories whose pre-failure conditions LSR-qc1 correlations proposed by Olson and Stark (2002) are
corresponded to yield strength conditions and on this basis steeper than the linear regression computed herein. That is, the
proposed a YSR-qc1 correlation which they complemented with regressed line suggests a weaker dependence of LSR on qc1 than
upper and lower limits to account for variability. Figure 48 indicated by the correlation proposed by Olson and Stark (2002).
summarises these eight case histories in YSR-qc1 space and shows Again, the implication for LSR assessments is that qc1 can be
that there is only a weak (R2 = 0.15) linear correlation. That is, bypassed as an explanatory variable and should instead take note
85% of the variation in YSR is unaccounted for by qc1. that Olson and Stark (2002) reported LSR values that ranged from
Furthermore, of the eight case histories shown in Figure 48, CPT 0.027 to 0.12 with a mean value of 0.083. Wider upper and lower
readings were only available for two of them. For the remaining bounds result from considering the uncertainties of each case
six case histories, qc1 was assessed from correlations to SPT, history as reported by Olson and Stark (2002). The applicable
relative density, or otherwise estimated. Such correlations and LSR value must then be chosen within these bounds, or beyond if
estimations further reduce the reliability of the correlation. It is warranted, while keeping in mind that qc1 does not provide robust
also apparent from Figure 48 that the YSR-qc1 correlations guidance. As for the YSR, we note that while our comments do
proposed by Olson and Stark (2003) are steeper than the linear not provide additional guidance in the selection of LSR, they will
regression computed herein. That is, the regressed line suggests a hopefully prevent undue overreliance on the LSR-qc1 correlation
weaker dependence of YSR on qc1 than indicated by the proposed by Olson and Stark (2002).
correlation proposed by Olson and Stark (2003).
The implication for YSR assessments is that instead of 0.25
attempting to correlate YSR to qc1, it should be recognised that the Data point labels indicate case history ID as per Olson & Stark (2002).
flow failure case history database analysed by Olson and Stark
(2003) yielded YSR values that ranged from 0.24 to 0.28 with a 0.2 Average and limit trends (Olson & Stark 2002):
Liquefied strength ratio LSR

YSR = 0.0143qc1 + 0.03 ± 0.03


mean value of 0.26. Wider upper and lower bounds result from
considering the uncertainties of each case history as reported by
Olson and Stark (2003). The applicable YSR value must then be 0.15 Best fit line: YSR = 0.0093qc1 + 0.053
(R2 = 0.21)
chosen within these bounds, or beyond if warranted, while
3
keeping in mind that qc1 does not provide robust guidance. 26 8 2 15
Doubtless a significant challenge. While these comments do not 0.1 31 33 25 30
29 32 19
provide additional guidance to estimate the actual YSR, we hope 28 6 20
that they will prevent undue overreliance on the YSR-qc1 12 22
9
correlation proposed by Olson and Stark (2003). 0.05 24
10 21 Lacks CPT data
Has CPT data
0
0 1 2 3 4 5 6 7
Corrected tip resistance qc1 (MPa)

Figure 49. Flow failure case histories analysed by Olson and Stark (2002)
with adequate to back-calculate the liquefied strength ratio.

Based on 13 flow failure case histories with CPT records (data


classes A and B), Robertson (2010) proposed the following
correlation for the lower bound of liquefied strength ratio LSR
which is valid for Qtn,cs :
0.0003124 , + 0.02199
= = (25)
0.0001783 , 0.02676 , + 1

Figure 48. Flow failure case histories analysed by Olson and Stark (2003)
Equation 25 has a lower bound of 0.03 and an upper bound of
and deemed to be close to their yield strength envelope.
tan . Given the uncertainties involved in back-analysing case

151
histories, it is a positive feature of Equation 25 that it aims at correlations that fit soils with different compressibilities: stiff,
providing a lower bound of Sr, rather than a best estimate. intermediate, and compressible (Figure 51a).
However, Equation 25 relies on Qtn,cs and thus inherits the There are at least three points of concern in the approach
objections that were previously made regarding this parameter. It suggested by Jefferies and Been (2016). Firstly, like Olson and
is also worth noting that Equation 25 was considered during the Stark (2002, 2003), most of the case histories (8 out of 12) used
post-failure investigation of the Brumadinho TSF but ultimately to develop the correlations do not have electric CPT readings. This
not used due to a lack of match with laboratory results. The leads to reliance on results from mechanical CPT, SPT, and, in
discrepancy was attributed to the mineralogy and bonding of the one case, assessment of tip resistance from relative density. That
Brumadinho tailings (Appendix E of Robertson et al. 2019) and is, there is a strong reliance on the data classes B, C, and D as
may point to limitations in the applicability of Equation 25. defined by Robertson (2010) (Figure 51b). Secondly, the
Robertson (2010) noted that Equation 25 can be overly approach relies on in-situ determinations of , the feasibility of
conservative in sensitive clays and for such soils proposed: which remains unproven even when high quality CPT data are
available (Figure 47) and seems even less likely when back
= = (26) analysing case histories that fall into data classes B, C, and D.
Lastly, because the CSL of 10 of the 12 case histories is unknown,
Jefferies and Been (2016) relied on the assumption that finer
10. However, this
Robertson (2010) justified Equation 26 with the observation assumption is at odds with experimental data (Papageorgiou and
that for most clay soils fs Sr. The supporting reference cited in
Fourie 2001, Reid 2015, Torres-Cruz and Vermeulen 2018,
Robertson (2010) leads to two datasets: one from the Santa
Torres-Cruz 2019).
Barbara Channel site originally reported by Quiros and Young
(1988) and later reproduced by Lunne et al. (1997), and another
a) 0.3
from the Scoggins Dam in Oregon, USA originally reported by
Compressible ( = 0.17) Compressible
Farrar et al. (2008) and later reproduced by Robertson (2009). At 10
Intermediate
the Santa Barbara Channel site, the undrained residual strength
v
Stiff
was evaluated with triaxial tests, whereas at Scoggins Dam it was
Liquefied strength ratio Sr/ Intermediate
evaluated using field vane testing. Figure 50 summarises these 0.2 ( 10 = 0.12)
datasets and provides an indication of the degree of scatter around
the fs = Sr line. We note the absence of supporting data from clayey
tailings deposits and suggest that such datasets would be a useful
addition to the literature, especially considering that the fs Sr 0.1
assumption has been challenged on the grounds that there are
factors independent from Sr that significantly influence fs. Such
Stiff ( = 0.05)
factors include the pore water pressure on the end areas of the 10

sleeve, the roughness of the sleeve, and the extent of soil 0


remoulding (Lunne, 2007). -0.1 0 0.1 0.2 0.3
Characteristic state parameter k

160
Residual undrained strength Sr (kPa)

Santa Barbara Channel


Scoggins Dam b) 0.3
120 Compressible ( 10 = 0.17) Class A
Class B
v

Class C
Liquefied strength ratio Sr/

Intermediate Class D
80 0.2 ( 10 = 0.12)

40
0.1

0
0 40 80 120 160 Stiff ( 10 = 0.05)
Sleeve friction fs (kPa)
0
-0.1 0 0.1 0.2 0.3
Figure 50. Comparison between the sleeve friction fs and the residual
shear strength Sr from two non-tailings sites. Sources: Lunne et al. Characteristic state parameter k

(1997) and Robertson (2009).


Figure 51. Best practice LSR- k correlations proposed by Jefferies and
Acknowledging that perfectly undrained conditions cannot be
Been (2016) with data discriminated by a) assumed soil compressibility,
guaranteed in the field, Jefferies and Been (2016) resorted to a and by b) data class.
database of 12 flow failure case histories to develop correlations
between the characteristic state parameter k and the liquefied 5.1.4 Partial drainage during CPTu
strength ratio LSR. In their framework, k represents the value of Analyses of CPTu test data are usually carried out assuming either
that effectively controls the mechanical behaviour of a soil fully drained penetration (no generation of excess pore water
deposit. Jefferies and Been (2016) suggest a value between the pressure) or fully undrained penetration (zero volume change
80th and 90th percentile of for liquefaction analyses. That is, they during penetration). Perhaps this is a hangover from the desire to
suggest that the weakest 20% or 10% of a soil deposit determines think of soils as either sands (drained) or clays (undrained).
its stability. Their approach accounts for the effect of soil However, as emphasised many times in this paper, many mine
properties, especially compressibility, on the LSR- k correlation. tailings are either silts or silty sands, with permeability often in
Accordingly, they use the case history database to develop three the range from 10-5 to 10-8 m/s (Schnaid, 2021). The traditional

152
cone penetration rate of 20mm/s may not be appropriate for such Distinction between fully drained, partially drained, and fully
tailings, potentially leading to partial drainage during the undrained conditions can also be aided by the CPTu-based soil
penetration process. This is especially important when testing classification chart proposed by Schneider et al. (2008, 2012). The
loose, contractive materials. When compared to undrained chart is defined in Q-U2 space and its regions are given by the
conditions, partial drainage increases penetration resistance and dotted lines in Figure 53. Schneider et al. (2008) explain that the
affects the excess pore water pressure generated during Bq = 1 contour line is not a boundary, but rather is intended as a
penetration and the form of any subsequent dissipation test carried link to the Q-Bq chart proposed by Robertson (1991). Q, U2, and
out. A key implication is that undrained strengths may be Bq are normalised measures of CPTu results and are defined as:
overestimated.
The issue of partially drained CPTu penetration has been = (28)
investigated by several researchers, including Finnie and
Randolph (1994), who suggested a normalised velocity, V, to
account for variables that affect penetration bearing resistances. 2= (29)
Their equation was
= (30)
= . (27)

where v is the penetration velocity (usually 20mm/s for field


CPTu tests), d is the diameter of the cone and cv is the vertical
coefficient of consolidation measured in a 1-D consolidation test.
Results are often presented schematically in a form such as that
shown in Figure 52, where a logarithmic scale is usually used for
the horizontal axis. The tip resistance is usually normalised by
either a reference tip resistance value (in Figure 52 the drained tip
resistance is used) or by vertical effective stress. As indicated,
below a certain threshold normalised velocity the CPTu test is
fully drained; above another threshold velocity, the test is fully
undrained. Between these thresholds, penetration generates partial
drainage. From our experience, insufficient acknowledgement of
the potential for a CPTu test in tailings (particularly silts or silty
sand tailings) to be in the region between drained and undrained,
i.e. partially drained, has been given in many testing programmes.

Figure 53. Q-U2 soil classification chart. Note: Dotted lines correspond to
original boundaries (Schneider et al. 2008), and solid lines correspond to
Bq contours that approximate or replace the original boundaries. Plotted
data correspond to sounding CPTU-05-05N_Updated reported in
Robertson et al. (2019).

Figure 52. Illustration of effect of normalised penetration velocity on The Q-U2 chart has five regions: one drained region, two
normalised cone tip resistance. undrained regions (clays and sensitive clays), one region for
transitional soils, and one region shared by low rigidity clays and
Finnie and Randolph (1994) suggested the transition from transitional soils. The term "transitional soils" refers to the wide
drained to partially drained penetration occurred at V 0.01 and variety of soil types in which the CPTu is partially drained and
from partially drained to undrained at V 30. includes clayey sands and silts, silts, and residual soils (Schneider
During a CPTu it is common to occasionally stop the test and et al. 2008).
allow excess pore water pressures to equalise, referred to as a Contours of constant Bq provide reasonable approximations to
dissipation test. From the rate of dissipation, the coefficient of the curved boundaries of the Q-U2 chart (Robertson 2016).
consolidation due to horizontal drainage, ch, may be obtained, Furthermore, results from non-plastic tailings and coal ash
using e.g. the procedure suggested by Teh and Houlsby (1991). indicate that the boundaries of the drained region can be replaced
The parameters cv and ch are usually not equivalent, due to by Bq contours (Torres-Cruz and Vermeulen, 2018). Combined,
anisotropy, although this effect is often considered insubstantial these two observations suggest that the Q-U2 chart as proposed
enough to ignore. An alternative approach to determine whether by Schneider et al. (2008, 2012) can be simplified if all the
cone penetration is undrained is to use the time to 50% dissipation boundaries are replaced by Bq contours as shown in Figure 53. It
of excess pore pressure, i.e. t50, as an indicator. DeJong and follows that Bq can be used to readily verify whether, according to
Randolph (2012) and Robertson (2012) have suggested that when the Q-U2 chart, CPTu soundings are being conducted under
t50 is greater than 50 seconds the penetration is essentially drained, undrained or partially drained conditions. Table 2
undrained.

153
summarises the expected drainage conditions for different values 5.2 Field vane testing
of Bq.
As mentioned earlier, field vane testing is used extensively for
fine-grained soils and is appropriate for clayey tailings. However,
Table 2. Approximate correspondence between Bq and drainage conditions use of the vane for silty and sandy tailings is problematic if the
based on the Q-U2 chart proposed by Schneider et al. (2008, 2012). standard testing rate of typically 0.1° per second is used. Blight
(1962) realised this problem when working with silty gold tailings
Bq range Drainage conditions and derived a relationship between degree of consolidation U
<-0.01 Partial drainage
during shearing and a dimensionless time factor T, given by:
-0.01 to 0.01 Fully drained
0.01 to 0.05 Partial drainage = (31)
0.05 to 0.2 Partial drainage or undrained
>0.2 Undrained where tf is the time to failure, i.e. time to reach peak strength and
D is the vane diameter. Blight (1962) derived this relationship
both empirically and theoretically and the validity of Blight’s
Figure 53 includes data from a CPTu sounding performed at approach has been confirmed by Chandler (1988). Using his
Dam I at the Córrego do Feijão iron mine near Brumadinho, Brazil approach, Blight (1962) suggests a T value smaller than about
(Robertson et al. 2019). Regardless of whether one considers the 0.05 is required to ensure undrained shearing occurs. This is easily
original Q-U2 boundaries or the Bq contours, the chart indicates achieved for clayey tailings, but for a typical silty tailings, which
that partial drainage conditions prevailed during some portions of may have a cv value of about 100 m2/year, the required time to
the sounding. Naturally, plots of Bq versus depth can aid in failure to ensure undrained conditions using a 5 cm diameter vane
locating the layers in which partial drainage is occurring. is about 40 seconds, i.e. much faster than achieved using the
Based on the discussion in this section, there appears to be a standard rate of 0.1° per second. Using a larger diameter vane
need for greater awareness amongst tailings engineers that it is increases the time to failure, but for the silty tailings used in the
often warranted and necessary to modify the standard cone example, even this would not ensure undrained shearing for any
penetration rate of 20 mm/s to produce results that are compatible reasonably sized vane. Clearly, increased rates of shearing are
with interpretation procedures that assume either fully drained or required to ensure truly undrained shear strengths are obtained. If
fully undrained conditions. CPTu data are available, it is possible to substitute ch for cv in the
above equation, with little loss of relevance.
5.1.5 CPT testing in partially saturated tailings Given the enormous number of cone penetration tests and
It is crucially important to remember that CPTu testing, and field vane tests that have been carried out on silty and silty sand
indeed other field tests such as vane or full-flow penetrometer tailings worldwide, the majority of which were likely carried out
tests, have all been developed for application to saturated at inappropriate testing rates, one can only wonder about the
conditions, which for all practical purposes can be assumed to validity of some of the stability analyses carried out using such
begin below the phreatic surface. Above the phreatic surface, data and the millions of dollars that have perhaps been wasted as
where negative pore water pressures relative to atmospheric exist, a result.
there can be a substantial contribution to strength provided by A related issue that the authors have often encountered in
suction effects. Simplistically, suction increases the effective practice is the use of field vane data to ‘calibrate’ cone
stress (being a negative pore water pressure) and thus shear penetrometer data. If a CPTu test is carried out at a rate such that
strength. Interpretations of material parameters from tests such as the test is truly undrained, the peak undrained shear strength, Cu,
CPTu tests done above the phreatic surface can thus be misleading. is often estimated from the equation
Unfortunately, most currently available interpretation software
does not adequately account for this issue. = (32)
The suggestion by Robertson (2010) that ‘soils with a constant
value of Qtn,cs have essentially a similar state parameter and hence where qc is the cone tip resistance, v0 is the total vertical stress
a similar response to loading’, should not be extended to and Nkt is a bearing capacity factor, usually taken to be between
conditions above the phreatic surface. Doubtless Robertson 11 and 20.
(2010) never intended for such extrapolations to be made, but However, it is not uncommon to see the undrained shear
unfortunately they sometimes are. Aside from work such as that strength obtained from field vane tests, which are implicitly
by Pournaghiazar et al (2011) who describe a calibration chamber assumed to be truly undrained, to back-calculate Nkt values using
suitable for testing tailings, there has, until recently, been limited the known v0 and measured qc values. Values of Nkt higher than
research on cone penetration testing of partially saturated silts or 40 are not uncommon when this approach is adopted and the value
silty sand tailings. Miller et al (2018) provide a useful summary derived is nothing more than an artefact of either the penetrometer
of tests done on natural soils, varying from sands to clays. They or the vane, or both, not being truly undrained tests. We strongly
conclude that cone tip resistance generally increases with discourage this practice unless it can be confirmed that both the
increasing matric suction (as would be expected), so ignoring this CPTu and vane tests were truly undrained.
effect would result in an overestimate of soil strength and Field vane tests provide another option for determining
underestimate the state parameter. They further conclude that the residual strengths. Although traditionally used primarily for
tip resistance dependency on suction is greater for soils with measuring peak shear strength, increasing interest in residual
significant plasticity compared to non-plastic soils. This is strength determination has meant the vane is now widely used in
consistent with more plastic soils being capable of sustaining this regard as well. Improvements such as high precision control
higher suctions than sandy and silty soils. This observation is of rate of rotation and down-hole torque measurement (thus
perhaps something of a saving grace when dealing with common eliminating rod friction effects) have improved confidence in vane
silty and sandy tailings – the suction hardening effect may be test results. However, given the above discussion about the rate of
relatively muted and the potential for incorrectly evaluating shearing, particularly when dealing with silty materials, it is
strength parameters not as great as with clayey tailings. conceivable that even if the vane were rotated sufficiently fast to
ensure measured peak strength values were truly undrained

154
values, the same may not automatically be true for residual As noted above, measurement of lateral stresses is one of the
strength values. The much greater time (related to much greater applications of the PMT. Such measurements are necessary for a
degree of rotation) required to mobilise the large strain strength thorough interpretation of the CPT (Jefferies and Been, 2015).
may also be enough time to enable some degree of consolidation Furthermore, the relative magnitudes of the vertical and horizontal
to occur, thus providing an overestimate of strength. effective stresses determine the levels of deviatoric stress which,
in turn, affects stability. Despite the importance of measuring
5.3 Full-flow penetrometers lateral stresses, a review of the literature and consultations with
geotechnical practitioners suggest that such measurements are
As discussed by Schnaid (2021), an alternative for estimating
seldom made. Indeed, Graham (1992) loosely cited Jamiolkowski
residual shear strength is full-flow penetrometers, the two most
as saying that horizontal stress measurements are the 'black hole'
common being ball penetrometers and T-bar penetrometers (see
of geotechnical engineering. Fast forward to 2022 and horizontal
Stewart and Randolph, 1994). These were initially developed
stresses remain as elusive as ever.
primarily for the offshore oil industry, to measure strengths of soft
Figure 54 presents plots of v vs h corresponding to an oil
seabed soils. Recommended procedures and interpretation
sand tailings deposit (Shuttle et al. 2022). Given their similarities
methods for determining peak and remoulded strengths (and thus
with tailings deposits, measurements in hydraulic sand fills are
sensitivity, alternatively expressed as Bishop’s Brittleness Index
also included in Figure 54 (Graham and Jefferies 1986). All h
IB, discussed in Section 8.1.1) are discussed by Randolph and
measurements in the figure were performed with a SBPMT. The
Andersen (2006) and Zhou and Randolph (2009) for tests in clays.
figure shows that although it is standard practice to assume a
With suitable adjustments, such as rate of cycling, these full-flow
constant value of the h/ v ratio throughout an entire tailings
penetrometers are likely to see increased use for testing of some
deposit (e.g. Jamiolkowski and Masella 2015, Lipinski and
tailings. However, the authors’ experience with a ball
Wdowska 2018, Robertson et al. 2019), significant deviations can
penetrometer has been that in slightly stiff tailings, the borehole
occur from the mean value.
behind the penetrometer does not always close (violating the
concept of ‘full-flow’) and upon cycling of the device, very low Horizontal effective stress 'h [kPa]
or even zero readings are detected (because there is no resistance 0 200 400 600 800
to extraction). Although potentially very useful, indiscriminate 0
use of full-flow penetrometers is inadvisable. Vertical effective stress 'v [kPa]
Hydraulic fill (Alerk)
Mahmoodzadeh and Randolph (2014) suggest a similar 50 Hydraulic fill (Amauligak)
relationship between normalised resistance and normalised
Oil sand tailings (dyke crest)
penetration for the ball penetrometer as for a cone penetrometer. 100 Oil sand tailings (beach)
Relevant values of ch may be determined directly from dissipation
tests carried out using the ball penetrometer as most ball 150
penetrometers include a pore pressure sensor. However, the
location of this sensor is key, as sometimes it is located at mid- 200
height (referred to as the equator location), sometimes on the face
of the ball and sometimes on the shaft, the latter being most 250
similar to the location on a typical cone penetrometer. Using data
from Mahmoodzadeh and Randolph (2014), a suitable penetration
300
rate can be determined to ensure fully undrained penetration
occurs with the ball penetrometer.
Figure 54. h v in hydraulically placed oilsand tailings and

5.4 Pressuremeter testing in tailings. other hydraulic fills. Data sources: Shuttle et al. (2022) and Graham and
Jefferies (1986).
Pressuremeter tests (PMT) are conducted by expanding a
membrane against the walls of a borehole. There are three types Further to the scarcity of lateral stress measurements in
of PMTs: 1) prebored (PBPMT), in which the borehole is first tailings reported in the literature, part of the problem appears to
made and the PMT device subsequently introduced; 2) self-boring lie with the inherent difficulty of measuring lateral stress in soils
(SBPMT), in which boring and device insertion occur in general. There are significant uncertainties in assessing the in-
simultaneously; and 3) full-displacement (FDPMT), in which the situ value of h from PMT results as well as from other in-situ
PMT device is fitted with a cone that allows it to be pushed into tests such as the Iowa stepped blade, and the flat dilatometer
the ground (Mair and Wood, 1987). (Marchetti 2015, Mair and Wood 1987). The uncertainties stem
PMTs have been used in a variety of ways to characterise both from the fact that lateral stresses are very sensitive to the
deposited tailings and the underlying foundation soils. Its disturbance that necessarily results from the insertion of any kind
applications include assessment of the friction angle (Gomes et al. of measurement probe into a formed soil. That is, the act of
2003, Chakraborty et al. 2020), flow slide potential (Debasis et al. measuring alters the parameter being measured. Engineered fills
2002), state parameter (Wride et al. 2000, Ghafghazi and Shuttle such as tailings deposits allow the positioning of a measurement
2008), elastic moduli (Morgenstern et al. 2015, Chakraborty et al. probe, such as a total stress cell, before depositing the tailings,
2020), and lateral stress (Graham and Jefferies 1986, Wride et al. thus eliminating the need for disturbance to make the
2000, Chakraborty et al. 2020, Shuttle et al. 2022). Clearly some measurement (e.g. Chen et al. 2021). However, even in this case,
advantage has been taken of the potential of pressuremeter tests to the effect of the relative stiffness of the stress cell and the soil
better characterise TSFs. However, our consultations with results in the measurement of lateral stresses not being
industry practitioners show that, overall, the use of PMTs in TSFs straightforward (Clayton and Bica, 1993).
remains limited. This is also reflected by the fact that the post-
failure geotechnical investigations of the embankment at CADIA
Valley Operations (Jefferies et al. 2019) and of the Brumadinho 5.5 Relative density as a screening tool.
dam (Robertson et al. 2019) did not consider PMT results. The Relative density, RD, is defined as
question as to whether PMTs can contribute more to the
geotechnical characterisation of TSFs thus remains open.
= (33)

155
where emax and emin are the maximum and minimum void ratios caution against unquestioning adherence to the use of empirical
determined using standard laboratory testing techniques and e is correlations when determining tailings design parameters, such as
the void ratio of interest (e.g. the in-situ void ratio Therefore, a state and strength. Empirical correlations remain useful as
min screening tools and as a reality check, but our strong preference is
emax). Relative density is used universally as a state for a mechanics-based approach rather than slavish use of
index parameter for non-plastic soils, despite its inability to empirical correlations. The techniques for deriving correlations
normalise mechanical response. However, it is not a good between CPTu results and soil state, as advocated by Shuttle and
indicator of behaviour for silts. As discussed by Shuttle and Cunning (2007) and Shuttle and Jefferies (2016) amongst others,
Jefferies (1998), one sand (Hilton Mines sand) produced the same implicitly account for influences of aspects such as particle shape.
CPT resistance at 60% relative density as another sand (Monterey For example, the Norsand parameters that are used to calculate
sand) at 40% relative density, both sands having a few percent limiting pressures from spherical cavity expansion solutions
fines. (which are subsequently used to estimate the relationship between
The limitations of relative density as an indicator of state are normalised tip resistance and the state parameter) will be
further illustrated by the results presented by Liu and Lehane influenced by particle shape. Predictions of limiting cavity
(2012). They report results from centrifuge tests on four uniformly expansion pressure will therefore also change with changes in
graded silica materials, each with its own characteristic particle constitutive model input data.
shape (one being spherical glass beads, SGB). A result is shown
in Figure 55 for three of the four materials tested, which is for a 6 MODES OF FAILURE.
centrifuge test carried out at 50g with samples prepared at a
relative density of 20% (i.e. expected to be contractive). As can There are a variety of potential failure modes to consider when
be seen, there are significant differences, with the cone resistance evaluating the stability of both proposed and existing TSFs. To
increasing with the degree of angularity of the particles. Using the systemize risk analysis and operation of TSFs, failure modes are
terminology discussed in Section 3.2, the roundness, R, and generally divided into categories. These categories can be
sphericity, S, are summarised in Table 3, together with index different from site to site. However, the following generalized
properties that illustrate the otherwise similar nature of the three eight failure modes are commonly used:
materials tested.
Overtopping
Slope instability
Internal erosion
External erosion
Foundation
Environmental
Human
Other

A risk analysis considering relevant failure modes can, and


should, be used as a basis for the design of surveillance and
monitoring programs. The aim is to be able to detect a potential
failure scenario as well as initiate and execute measures to prevent
failure. Below follows a description of what each failure mode
suggested here may comprise.

6.1 Overtopping
Figure 55. CPT end resistances (qc) observed in centrifuge tests run at 50g
on samples prepared at 20% relative density. Republished with permission Water management, e.g. water balance and management of
of ICE Publishing, from Liu and Lehane (2012); permission conveyed extreme flows, is crucial for all dam structures. Tailings dams
through Copyright Clearance Center, Inc. Table 3. Properties of three typically have a smaller catchment area compared to water dams,
materials used in centrifuge tests (after Liu and Lehane, 2012). but water discharge may be restricted due to water quality, for
example, leading to positive water balances. There are, however,
Table 3. Properties of three materials used in centrifuge tests (after Liu instances, where a TSF has a significant upstream catchment, such
and Lehane, 2012). as valley TSFs, or facilities storing high or extreme rainfall events
Material d50 (mm) Cu R S
(e.g. facilities without spillway) and these facilities may be more
susceptible to overtopping. Freeboard needs to be designed to
SGB 0.12 1.05 1 1 retain extreme design floods and water management under
UWA sand 0.15 1.4 0.7 0.62 extreme conditions should preferably be automatic, i.e. not require
GB1 0.45 1.13 0.62 0.54 human intervention.
Examples of overtopping failures include Baia Mare,
Rumania in 2000 and Merriespruit, South Africa in 1994.
Considering Figure 55, below about 50mm (equivalent to
2.5m depth at full scale), there is more than 100% difference 6.2 Slope instability
between the SGB and GB1 materials, with the natural UWA sand Material properties in the structural zone of the dam, pore
material also being substantially lower than the GB1 results. pressures, deformations and static and seismic loads affect TSF
Similar differences were seen in centrifuge tests run at 100g and slope stability. Deposition practices, consolidation, compaction
200g. and drainage affect the strength of the tailings and must be
Cone tip resistance can therefore be a poor indicator of in-situ understood to correctly estimate material properties. The
state if a crude framework such as relative density is used. It is elevation of the phreatic surface is equally important, and a
too dependent on unquantified influences, such as particle shape realistic assessment is crucial. As many tailings dams (typically
and compressibility. As argued in several places in this paper, we upstream dams) are drained structures (or at least designed to be

156
so) the pore pressure is not necessarily hydrostatic. In such cases, Examples of foundation failures are Aznalcollar, Spain in
the use of piezometers to adequately assess the porewater pressure 1997, Mount Polley, Canada in 2016 and Cadia, Australia in 2018.
regime is an absolute necessity. Seismic hazard/response analyses
also need to be done correctly, as dealt with in detail in section 6.6 Environmental
8.2.
The thinner the structural zone (i.e. the narrower the beach of Environmental failure modes include situations where the facility
unsaturated tailings) the more sensitive the dam becomes to exceeds the expected impact on the environment. For example,
potential instability. This is particularly relevant in upstream dust generation, uncontrolled seepage, emergency release of water
dams, where the operation of the dam has a significant impact on during extreme events, limited settling time before discharge
the width of the structural zone. causing increased levels of contaminants in the discharge water
Examples of slope instability failures include Brumadinho, Brazil etc. This failure mode, typically, does not lead to catastrophic
in 2019 and Fundao, also Brazil, in 2015. failures such as a dam breach, but is nevertheless important to
consider.
6.3 Internal erosion (piping)
6.7 Human
Seepage occurs through virtually all TSFs and if the hydraulic
gradient is high enough, the water may erode fine particles if Intervention, or lack of intervention, by mine personnel may cause
adequate filters are not in place. Ongoing erosion can lead to a critical situation, worsen the situation or in worst case a dam
piping, and an open channel may develop from the water pond to failure. Education, training and emergency preparedness planning
the downstream side through the dam and/or the foundation and is therefore crucial.
ultimately cause failure.
An example of internal erosion failure is Omai, Guyana in 6.8 Other
1992. This includes failure modes not categorized above. For example,
natural hazards including rock, soil, or snow avalanches into the
6.4 External erosion
TSF impoundment leading to overtopping, or instability of the
Extreme floods in adjacent streams may erode the toe of the dam. dam.
High intensity rainfall may also cause erosion on the downstream Actual failures show that a combination of factors is normally
slope. Both external erosion events can potentially lead to dam the reason for failure, and in hindsight the cause of the failure
instability. If a large free water pond exists, or excessive wave could, in many cases, have been predicted. (ICOLD, 2001).
action occurs, it could possibly cause erosion and instability of the
upstream slope leading to overtopping. 6.9 Trigger mechanisms.

6.5 Foundation In order to initiate a failure mode a trigger is required. Figure 56


shows a number of different triggers, such as: 1) high pond water
Instability of the foundation can be due to undetected weak layers, levels, 2) inhomogeneous tailings mass, 3) low density in the
incorrect strength assumptions, pore pressure generation, artesian tailings mass, 4) heavy rain, 5) low permeability layers in the
water pressures and/or incorrect seismic hazard/response analysis. structural zone, 6) weak zones in the foundation and 7) material
Complex geologic formations (including clay layers, weak transport in the dam.

bedding planes etc.) increase the risk as well as soils that may be Figure 56. Mechanisms and processes that contribute to the failure of
sensitive to static and/or cyclic liquefaction, such as loose tailings dams. From Santamarina et al. (2019). Reprinted with permission
granular soils. from AAAS.

157
6.10 Lessons from forensic investigations of TSF failures downstream of the TSF received inadequate warning of the
developing failure.
As discussed in section 9 (monitoring), various combinations of Morgenstern et al. (2016) present an in-depth forensic
instrumentation and visual observation may be utilised to gain evaluation of the Fundão failure, working through several
advance warning of one or more of the above generic failure hypothetical failure modes and steadily eliminating all but one
modes developing. However, failures of actual TSFs provide based on the weight of evidence gathered. The process followed
additional lessons on failure modes that are sometimes overlooked is a lesson in itself; some forensic investigations appear to have
or not anticipated. It is therefore useful to evaluate the likelihood rushed into deciding on a particular failure mode explanation
of a particular failure mode occurring, but it is not a substitute for without working through a process of elimination as Morgenstern
a good understanding of geotechnical engineering principles. A and his co-investigators did. The failure was attributed to lateral
few of the recent TSF failures and associated forensic extrusion of slimes at depth causing horizontal extension (and thus
investigations illustrate this point. lateral unloading) of overlying, loosely placed sand tailings.
Given these sandy tailings were extremely brittle, the amount of
6.10.1 The Fundão TSF failure in Brazil in 2015 strain required to initiate collapse due to static liquefaction was
On the afternoon of 5th November 2015, the Fundão TSF in small. It appears these movements certainly did not manifest as
Brazil collapsed. The resulting flowslide of 32 million m3 of cracks on the surface of the TSF or noticeable bulging at the toe
tailings flowed for some 660 km downstream (after entering a of the TSF.
river), engulfing the town of Bento Rodrigues and resulting in 19
known fatalities. It was probably the worst industrial accident in 6.10.1.2 Lateral extrusion as a failure mechanism.
Brazil until that time. Figure 57 shows a lateral extrusion type stress path, such as may
The Fundão TSF was a cross-valley embankment that abutted have been experienced in the brittle sandy tailings overlying the
natural slopes on either end. The original plan was to deposit ductile slimes. Strain compatibility between the slimes and the
sandy tailings behind a compacted earthfill starter dam and then sandy tailings as the slimes deforms causes a reduction in
raise the TSF height through a sequence of upstream raises. horizontal stress but the vertical total stress remains unaltered; the
Deposition of sandy tailings from the embankment crest was to resulting stress changes produce a decrease in mean effective
continue while deposition of a separate tailings stream, consisting stress and an increase in deviator stress. Failure is triggered at
of finer, clay-like tailings, often referred to as ‘slimes’, took place. point A, well below the effective stress failure envelope.
The intention was to use the deposited sand tailings to retain the
slimes, maintaining a wide zone of free-draining sands adjacent to
the embankment. This goal is consistent with good practice when
operating upstream TSFs, i.e. maintaining a large, well-drained
and unsaturated tailings beach adjacent to the perimeter
embankments. Unfortunately, this design intent was not adhered
to.

6.10.1.1 Findings from the forensic investigation


At the time of the failure there were several workers either on the
slope of the embankment or on the embankment abutments. From
the various eyewitness accounts, Morgenstern et al. (2016)
reported the following sequence of events:
Figure 57. Illustration of lateral extrusion stress path showing triggering
of failure when instability line is reached.
On the afternoon of the failure, November 5th,
2015,
several workers were on the facility and in a position to see
Prior to the failure developing, pore pressure measurements
along the length of the embankment crest. The first thing
were stable; loading appeared to be proceeding in a drained
noticed by many workers was a cloud of dust rising from
fashion, i.e. no generation of excess pore water pressure.
the left (looking downstream) abutment.
A worker reported seeing waves developing in the central However, when failure was triggered, the failure was undrained.
portion of the reservoir, accompanied by cracks forming Shear-induced excess pore pressures in the contractive sandy
on the left side and blocks of sand moving up and down on tailings could not be sufficiently quickly dissipated, despite the
the left abutment setback. tailings having a relatively high permeability. It illustrates the
Another worker saw a crack open along the crest of a step- dangers of adopting an attitude that was still widespread in the
in embankment that had been constructed adjacent to the tailings community at the time, that ‘sands cannot fail undrained’.
left abutment, propagating in both directions. With the enormous increase in research activity over the five years
At another location, workers experienced an avalanche of since Fundão, hopefully attitudes such as this no longer exist and
mud-like tailings cascading down from the left abutment. the risks posed by brittle tailings are now acknowledged.
Other observers reported seeing a sudden jet of dirty water
“explode” out of an underdrain. 6.10.2 The Feijão TSF failure in Brazil in 2019.
A worker standing on the tailings beach reported the The Feijão failure, which occurred on 25th January 2019, has
ground begin to move beneath him and crack around him, already been mentioned in the introduction to this paper. It caused
detaching from the step-in slope and moving downstream. over 260 fatalities, many of them members of downstream
communities. The graphic video that emerged of the failure and
The report by Morgenstern et al. (2016) contains additional its immediate aftermath has doubtless been viewed many millions
eyewitness accounts, providing an excellent timeline of events as of times. It clearly demonstrates the suddenness with which the
they unfolded. Only the above observations, which were the first failure occurred. A vast volume of the tailings was clearly brittle.
in a sequence of many, are included here to illustrate a specific Failure of the TSF occurred at about 12:30 pm local time.
point. At Fundão it is clear the failure evolved over a few minutes. Unfortunately, this meant there were several people in a meal
Initiating an Emergency Response Plan (ERP) was not possible. room located a short distance downstream of the TSF when it
Some workers on the TSF managed to scramble to safety, but failed. The video footage shows the failure occurred over about
unfortunately, others did not. Additionally, communities 80% of the face of the TSF in only five seconds, with the release

158
of tailings occurring through a sequence of retrogressive failures 6.10.2.3 Reconciling the EPR and CIMNE findings
in only about five minutes. Once again, timeous implementation The findings from the two forensic reports are very different and
of an ERP was not possible. sometimes contradictory. For example, CIMNE conclude, “The
Unusually, the Feijão failure has seen two detailed forensic set of numerical analyses performed lead to the conclusion that
investigations carried out by different investigators. The first of the drilling of borehole B1-SM-13 is a potential trigger of the
these was commissioned by the TSF owners Vale and released in liquefaction that caused the collapse of the dam” and furthermore,
December 2019 (Robertson et al, 2019). This report is referred to “The analyses carried out have not been able to identify other
as the Expert Panel Review (EPR) in following discussions. The liquefaction triggers. In particular, the calculations performed
second report resulted from a Term of Cooperation signed by Vale incorporating only the effects of enhanced rainfall and creep,
with the Federal Public Prosecutor’s Office (MPF) to contribute either alone or in combination, have not resulted in a general dam
to and collaborate with the authorities on a second, independent failure”.
forensic investigation. This second investigation was carried out As noted earlier, EPR on the other hand attribute the failure to
by the Centro Internacional de Métodos Numéricos (CIMNE) and a combination of creep and loss of suction due to rainfall.
their report was released in August 2021 (CIMNE, 2021). This Furthermore, they specifically discard borehole over-
will be referred to as the CIMNE report. pressurisation as a trigger, concluding, “The localized drilling of
The Feijão TSF was built as an upstream valley impoundment the borehole on the day of the failure would not have triggered the
and raised in 15 stages over 37 years, starting in 1976. No wall observed global failure of the dam, as confirmed by computer
raises occurred after 2013 and tailings deposition ceased in 2016; simulations”.
the facility was thus inactive at the time of the failure. The fourth Design engineers would, quite reasonably, be concerned
wall raise include a step-in, much like the portion of the Fundão about this disparity in conclusions from the two forensic
TSF referred to above. investigations. Although the authors of the current paper have
personal opinions on which of the trigger mechanisms is most
6.10.2.1 Findings from the EPR forensic investigation plausible, we cannot comment in detail, not being privy to all the
The EPR concluded that the sudden loss of strength and the available data and investigations. A rational approach to the
resulting failure w ere a result of a critical combination of dilemma resulting from the different conclusions is perhaps to
ongoing internal strains, i.e. creep, as well as a strength reduction adopt the approach suggested some time ago by Silvis and de
due to loss of suction in the (surface and near-surface) unsaturated Groot (1995) and more recently by Morgenstern (2018), amongst
zone due to intense rainfall towards the end of 2018. others, which is – if there are tailings that are liquefiable, assume
The EPR notes that operations allowed ponded water to get they will liquefy. Rather than trying to predict plausible trigger
close to the crest of the TSF at times, resulting in the deposition mechanisms, simply assume liquefaction will be triggered.
of weak tailings near the crest, the above-mentioned step-in Although this is certainly a seemingly logical suggestion, it comes
resulting in upper portions of the slope being over weaker fine at a sometimes-substantial cost, for example when stabilising
tailings (slimes) and a lack of significant internal drainage buttresses are required to address stability under conditions of
resulting in a persistently high phreatic surface. They also flag the assumed mobilised large-strain strengths (see Figure 33).
likelihood of bonding between tailings particles occurring due to
the iron content, and this bonding producing a stiff, but collapsible 7 FIELD BEHAVIOUR AND ASSOCIATED MODELING.
structure which could manifest as a very brittle material if sheared
undrained. 7.1 Consolidation

6.10.2.2 Findings from the CIMNE forensic investigation Modeling of the field consolidation behavior of tailings serves
The CIMNE investigators commissioned additional laboratory several purposes. It provides information on the storage capacity
and field tests and undertook extensive, detailed numerical of a tailings impoundment as well as the rate of rise of the
modeling of the TSF. The CIMNE findings are very different from resulting tailings dam. Analyses also provide information on
those of the EPR and specifically discard both creep and bonding spatial and temporal distributions of the void ratio and densities
as contributory causes of the failure. within the impoundment. The information is then used to evaluate
Their report unequivocally identifies over-pressurisation of a both static and seismic stability of the resulting dam at various
borehole on the day of the failure as the trigger mechanism that stages of its construction. The void ratio distribution is also used
resulted in the catastrophic failure that resulted. Problems to establish the hydraulic conductivity values within the
associated with attempted installation of horizontal drains in 2018 impoundment to perform a realistic seepage analysis. Finally,
had been reported, with the compressor pressurising fluid during knowing the pre-closure void ratio distribution is essential to
drilling causing pressures of as much as 2MPa. In the CIMNE evaluate long-term post closure settlements in magnitude and
study they triggered (numerically) a very localised pore water rates. Detailed consolidation process modeling provides all these
over pressurisation at the tip of the borehole that was being drilled distributions through a rational approach, rather than through
on the day of the failure (borehole B1-SM-13). This overpressure empirical correlations.
propagated throughout the TSF perimeter zone, resulting in a very To gain an insight into the field behavior of tailings
rapid progressive failure. The predicted locations of rapid and consolidating under self-weight, several simple analyses are
significant movement match closely the observations provided by presented. The first example is a 5 meter tall column filled with
the video footage referred to previously. Interestingly, they tried tailings instantaneously. The material characteristics for the
triggering a failure by over-pressurising other boreholes that had analysis are presented in Figure 16 for both fine and coarse
been installed, but in all these cases the zone of overpressure grained tailings and the model parameters are listed in Table 4.
remained relatively localised. In their concluding discussions they The material parameters of the fine tailings are representative of
suggest, “It is likely the occurrence of liquefaction and its highly compressible and plastic tailings such as some oil sand
subsequent uncontrolled propagation requires a particularly tailings or kimberlite tailings. Conversely, the material parameters
unfavourable combination of circumstances”. Perhaps we can of the coarse tailings are representative of low compressibility and
take some solace from this suggestion (but still not ignore risks low plasticity tailings such as those characteristic of hard rock
associated with brittle tailings). mining. The analyses assume an impervious bottom boundary in
which settlement is caused by self-weight only.

159
Table 4. Parameters for finite strain modeling of consolidation example. change from a concave upward shape to a convex upward shape for both
fine and coarse tailings.
Parameter A B Z C (m/day) D Gs
(kPa)
7.2 Desiccation and creep
Fine tailings 8.00 -0.254 0.164 1.89x10-6 4.00 2.60
Coarse 1.48 -0.188 0.296 5.48x10-3 2.69 2.60 7.2.1 Desiccation
tailings Desiccation of tailings is widely acknowledged to improve the
mechanical characteristics of soils and tailings through suction-
Figure 58 presents the results of two analyses in terms of induced overconsolidation (Daliri et al. 2014). However,
column heights versus time for both fine and coarse grained experimental data that quantifies the magnitude of desiccation-
materials, for instantaneous filling. In both cases, an impervious induced suction as well as its effect on the mechanical response of
bottom boundary is assumed and the settlement is caused by the the tailings is relatively scarce. Efforts to conduct real-time
self-weight only. The settlement rate and the time to consolidated monitoring of the development of suction pressures at a South
state is controlled by the hydraulic conductivity values for the two African platinum tailings dam have been recently reported
tailings, while the ultimate height is controlled by their (Basson et al. 2021). The approach partly relied on the use of low-
compressibility characteristics. The physical processes cost tensiometers that enable the measurement of negative pore
controlling the consolidation behavior upon deposition are better water pressures as low as -1.7 MPa (Jacobsz 2019). Regarding
understood by looking at the void ratio distributions at selected laboratory assessments of the beneficial effects of desiccation on
times presented in Figure 59. strength, Daliri et al. (2014) reported improvements under both
monotonic and cyclic loading. Daliri et al. (2014) also noted that
mechanical overconsolidation yielded a mechanical response that
was qualitatively different from that of desiccation-induced
overconsolidation. Importantly, they cautioned that simulating
desiccation through mechanical overconsolidation can result in
unconservative designs. More recently, results from monotonic
direct simple shear tests on undisturbed samples of desiccated
tailings have quantified the increase in the peak undrained
strength ratio (Su v) induced by desiccation (Reid et al. 2018a,
b). The results from Reid et al. allow an estimation of the
preconsolidation vertical effective stress beyond which the
benefits of desiccation-induced overconsolidation become
negligible. Their results also highlight that the effective
preconsolidation pressure is lower in silt-sized materials (e.g. gold
tailings) than in more clayey materials (e.g. iron ore and bauxite
tailings) (Reid et al. 2018b).
To demonstrate the desiccation process we analyze the same
5 m columns (i.e. initially saturated and with an impervious
bottom boundary) that were previously considered when
Figure 58. Height vs time for 5m tall columns of fine and coarse tailings. discussing the consolidation procedure. Again, both fine and
coarse tailings are considered (Table 4) and an evaporation rate of
0.3 mm/day is assumed that eventually leads to the development
of capillary forces. Figure 60 presents the height vs time
relationships for both the fine and coarse tailings, while Figure 61
presents the void ratio distributions at various times during the
consolidation and desiccation process.

Figure 59. Void ratio distributions at selected times for 5 m tall columns
of fine and coarse tailings.

The consolidation process starts at the bottom boundary, despite it being


impervious. Initially an upward hydraulic gradient equal to the critical
gradient for the material is created in the slurry. The upward flow
maintains the condition of zero effective stress and a uniform void ratio Figure 60. Height vs time for 5 m tall columns of fine and coarse tailings
throughout the column, except at the impervious bottom where the flow subjected to surface evaporation at 3 mm/day.
rate is zero. This condition persists in the consolidating column until the
consolidation front reaches the top of the column. During that time, the
surface settlement rate is also constant and equal to the critical gradient
multiplied by the hydraulic conductivity of the slurry at the initial void
ratio. In a settling column, this initial velocity can be used to calculate the
hydraulic conductivity at the beginning of the consolidation process. Once
the consolidating front reaches the top boundary, the void ratio profiles

160
profile develops above that point. Figures 62 and 63 present the
results of such an analysis for both fine and coarse tailings.

Figure 61. Void ratio distributions at selected times for 5 m tall columns Figure 62. Height vs time for 5m tall columns of fine and coarse tailings
of fine and coarse tailings subjected to surface evaporation at 3 mm/day. subjected to drainage to atmospheric pressure at the bottom.

Comparing Figures 58 with 60 and 59 with 61, several


observations can be made. Firstly, in the initial stages of 5
Fine tailings - 20 days Fine tailings - 100 days
consolidation, there is no difference in settlement rates and void 4.5
Fine tailings -150 days Fine tailings - 350 days
Fine tailings 500 days Fine tailings - 700 days
ratio distributions between the self-weight consolidation and self- Fine tailings 1000 days Fine tailings 4344 days
weight consolidation with surface evaporation of 3 mm/day. 4 Coarse tailings 20 days
Coarse tailings - 60 days
Coarse tailings - 40 days
Coarse tailings - 80 days
During that time, the self-weight consolidation expels the pore 3.5 Coarse tailings 100 days Coarse tailings - 559 days
water, for both materials, at a rate that is higher than the
evaporation rate. Thus, evaporation has no effect on the tailings’ 3
Height (m)

behavior. Once the settlement rate drops below the evaporation 2.5
rate the desiccation process starts, and evaporation “pulls” water
out at the constant rate of 3 mm/day and consequently the 2

settlement rate equals the evaporation rate. The time at which the 1.5
evaporation rate becomes larger than the settlement rate due to
1
self-weight consolidation can be discerned from the change in
slope in the curves of Figure 60. For the fine tailings, this change 0.5

in slope occurs at approximately 230 days, whereas for the coarse


0
tailings it occurs at approximately 90 days. The settlement process 0 2 4 6 8 10 12 14
continues until the shrinkage limit is reached, at which point the Void ratio

material starts desaturating and the evaporation efficiency


significantly decreases or completely stops. After the desiccation Figure 63. Void ratio distributions at selected times for 5 m tall columns
process starts, the void ratios at the tailings surface reduce due to of fine and coarse tailings subjected to drainage to atmospheric pressure
suction development and a crust is formed. The crust thickness at the bottom.
depends on the relative values of hydraulic conductivity at the
tailings surface and the evaporation rate. The higher the The results show larger settlements due to larger increases in the
evaporation rate relative to the hydraulic conductivity, the thinner effective stresses when compared to self-weight consolidation
the crust. This is seen in Figure 61 where a distinct desiccated only. The void ratio profiles do not show the development of a
crust is formed for fine-grained material with lower hydraulic distinct crust at the surface as the suction development is more
conductivity, while for coarser material with higher hydraulic gradual than in the case of high evaporation rate at the surface.
conductivity the same evaporation rate produces more uniform
drying with depth. 7.2.2 Creep
The desiccation process also leads to tension crack To demonstrate the creep effect on the settlement and void ratio
development at the tailings surface due to the release of tensile distributions of a bentonite slurry (a highly plastic material) a five-
stresses created in the desiccation process. The thinner the crust, meter high column, impervious at the bottom, was analyzed using
the shallower and more closely spaced the cracks, while deeper the CONCREEP model (Gjerapic et al 2021). The results are
and broader spaced cracks are observed when the evaporation rate presented in Figures 64 to 66. Figure 64 compares the height-time
is lower relative to the hydraulic conductivity of tailings. Detailed results for both consolidation with creep and consolidation-only
description of the cracking process analysis is beyond the scope predictions for the material. The time-dependent compressibility
of the present paper, but a rational experimental approach and relations are presented in Figure 18 and the corresponding model
numerical model development has been presented elsewhere parameters are listed in Table 5.
(Abu-Hejleh and Znidarcic 1995, Oliveira-Filho, 1998 and Yao et
al 2002). Table 5. CONCREEP model parameters for a bentonite slurry.
Tailings can be subjected to suction stress development upon Creep and compressibility Hydraulic conductivity Specific Initial
deposition even in the absence of significant evaporation at the (m/day) gravity void ratio
(kPa)
surface. If the surface water is readily removed from the tailings
and the bottom boundary is drained to the atmosphere (i.e. the A0
Af
17.47
10.53
B0
Bf
-0.44
-0.35
Z0
Zf
0.29
0.16
C 2.62x10-7

2.40 30
bottom boundary constitutes a water table), suction develops near a1 3.00 b1 1.50 z1 7.00
D 2.43
a2 -0.50 b2 -0.30 z2 -0.60
the tailings surface as the groundwater level is gradually lowered
below the surface. Ultimately, the groundwater level reaches the
water table at the bottom of the deposit and a hydrostatic suction

161
7.3 Tailings storage facility filling analyses.
The nonlinear finite strain consolidation theory developed by
Gibson et al (1967) is a one-dimensional theory with spatial
variability in the vertical direction. Although it rigorously treats
the consolidation process in slurries, it cannot account for the
three-dimensional nature of tailings storage facilities. However, it
should be recognized that any water drainage in horizontal
directions in such facilities will be minimal as the depth to the area
ratio is very small in all cases. Even when the depth of such
facilities exceeds a few hundred meters, the horizontal lengths are
several times larger and any horizontal flow due to consolidation
can be neglected. Still, the three-dimensional geometry must
sometimes be considered, as the depth of tailings will vary across
the facility. Gjerapic et al (2008) developed an analysis algorithm
Figure 64. Height vs time for a 5m tall column of a bentonite slurry in which a tailings facility was discretized into several columns
analyzed with and without creep effects. and filling of each column analyzed using the one-dimensional
nonlinear finite strain consolidation theory. The algorithm
As expected, the creep effect causes larger settlement than
accounts for the interaction of the adjacent columns during the
consolidation only. However, the initial rate of settlement,
filling process in the sense that while they are filled
controlled by hydraulic conductivity, is the same in both analyses
simultaneously, the taller column will undergo larger settlement
and the creep effect cannot be separated from consolidation only,
for the first sixty years. Figures 65 and 66 present the and will require more tailings to be placed in it to maintain a
corresponding void ratio distributions at selected times for both leveled tailings surface. The algorithm is implemented into the
consolidation and creep, and consolidation only analyses. Again, numerical model FILCON that provides the filling curves for
there is very little difference in the void ratio distributions up to several scenarios of the filling process. It first provides the tailings
sixty-one years, but the creep effects are significant beyond that elevations if the tailings are deposited at an initial solids content
point. The creep effects are especially noticeable at the low and no water is “bled” from the slurry in the sedimentation
effective stress range (shallower depth) as expected from the time process. Thus, this is the filling rate for the case when an
dependent compressibility curves in Figure 18 for which the creep incompressible liquid is filling the facility. The second filling
effect diminishes with the increase in the effective stress. For this curve presents the case when the solids content in the slurry is
example, the maximum effective stress at the bottom is 2.26 kPa, increased instantaneously due to sedimentation and the freed
creating a creep induced reduction in void ratio from the effective water is bled out and removed from the facility. For that case, any
stress induced ultimate value of 11.6 to the value of 7.78 after consolidation beyond the zero effective stress void ratio is
2740 years. neglected. The third curve in the output presents the actual
projected tailings elevations under the given production rate that
accounts for simultaneous filling and consolidation. Finally, the
filling curve for a 100% consolidated tailings is also given. That
would be a filling scenario when the production rate is so slow
that any deposited tailings instantaneously consolidates under its
self weight.
Figure 67 presents an example of elevation-area curve for a
tailings storage facility that is discretized into four columns to
perform the filling analysis. The total volumes for the actual
facility and the discretized model are identical at each discretized
elevation.

Figure 65. Void ratio distributions for a 5m tall column of a bentonite


slurry with creep.

Figure 67. Elevation-area data for a tailings storage facility discretized into
four columns with equivalent volume for filling analyses.

Figure 68 presents the results of the filling analysis assuming


an ore production rate of 2000 tonnes per day and the tailings
disposal at an initial solids content of 10% (initial void ratio of
Figure 66. Void ratio distributions for a 5m tall column of a bentonite 23.4). The material characteristics are given in Table 4 for Fine
slurry without creep. Tailings. The results show that the 120m deep facility would be

162
filled in about 1800 days if no water is bled out during filling, differences were particularly acute near the toe of the slope,
while it would take about 3200 days to reach maximum elevation exactly where failures often initiate.
if all the water released in the sedimentation process is removed The analyses presented by Gjerapic and Znidarcic (2020)
simultaneously. When accounting for self-weight consolidation assumed isotropic hydraulic conductivity values for the various
during filling, the top elevation will be reached in about 4350 layers in the analysis. Beckett et al (2016) compared phreatic
days. If the production rate were much lower so that the tailings surface profiles measured in geotechnical centrifuge tests with
could completely consolidate under its own weight during filling, those predicted using standard software (SEEP/W). The soil
the facility could accommodate 2.8 times larger tailings volume chosen for testing was a relatively uniform silt (coefficient of
(12200 days/4350 days). This is clearly seen from the post-filling uniformity of five) and thus almost incompressible above about
settlement curve for the given production rate where the ultimate 25kPa vertical effective stress. As shown in Figure 69, the
tailings height due to the self-weight only is significantly reduced. agreement was generally extremely good for all the equilibrium
For this example, the ultimate height was 41.4 meters that would water levels that were set in the constant head reservoir on the left
of the centrifuge.
be reached after two thousand years.

Figure 68. Filling analyses results for the tailings storage facility
discretized into four columns.
Figure 69. Predicted and measured results for steady state seepage, after
7.4. Seepage Beckett et al (2016).

The prime focus of seepage analyses of TSFs is usually the The good correlation observed in Figure 69 gives confidence
potential environmental impacts of contaminated water on water in using standard software for seepage analyses. However, the
resources. This paper does not discuss these issues as it focuses conditions analysed in this comparison were idealised, in
on geotechnical aspects of tailings behaviour. However, correct particular the isotropic nature of hydraulic conductivity. However,
seepage analyses are also crucial for rational evaluation of TSFs, as many natural soils, may have anisotropic and
geotechnical stability. Shear strength of tailings (whether heterogeneous hydraulic conductivity characteristics. This is
characterised in terms of effective stresses or through undrained partly due to effects of particle shape (e.g. more clayey tailings
stability analyses) is dictated by effective stresses; correct will have plate-like particles rather than being bulky, resulting in
evaluation of pore water pressures is thus critical. As these are particle fabric governed by settling conditions) and partly due to
usually obtained by estimates of the phreatic surface evaluated flow conditions on a tailings beach, where meandering flow
through interpretation of spot measurements of pore pressure (e.g. produces variable rates of particle deposition. True hydraulic
using piezometers), assumptions inherent to such interpretations conditions may thus sometimes best be characterised as
should be challenged and verified. anisotropic (higher horizontal hydraulic conductivity).
As discussed in Section 9, pore water pressures are usually Furthermore, where tailings compressibility is significant, it is
measured at specific locations (and depths) using electronic necessary to characterise a reduction in hydraulic conductivity
piezometers, which are increasingly being automated rather than with depth, as illustrated previously in Figure 16. Another often-
requiring manual readings. The pressures are converted to head, overlooked factor is the effect on hydraulic conductivity of
i.e. water level, to define the location of the phreatic surface. particle segregation during flow down a beach. Blight (1997)
Somewhat inexplicably, at many TSFs the authors are aware of, a illustrates this effect, where hydraulic conductivity decreases
single piezometer is installed along a given cross-section. This is exponentially down the beach according to the observed decrease
of no use when trying to compare predicted versus measured pore in particle size. The resulting phreatic surface, when compared
pressure profiles. There are some TSFs where two piezometers are with a phreatic surface derived from assumptions of homogeneous
installed on a given cross-section and, albeit rarely, sometimes hydraulic conductivity was starkly different, as illustrated in
three or more. When two piezometers are available for a given Figure 70.
cross-section, it is common for the phreatic surface to be derived
by simply joining the inferred water levels. As noted by Gjerapic
and Znidarcic (2020), this is a fundamentally incorrect approach
as it violates continuity of mass. They illustrate the implications
further by taking two approaches to the interpretation of the
phreatic surface shape; these were the commonly-used linear
interpolation and an interpretation based on seepage analyses that
predicted pore pressures that matched the measured values. The
two approaches provide distinctly different outcomes, with the
linear approach underestimating the height of the phreatic surface
and thus overestimating effective stresses and strength. The

163
saline tailings. They found that with inhibited evaporation,
tailings consolidation was due to self-weight consolidation alone.
This invariably results in only low densities being achieved and
consolidation may take extremely long. Initial evaporation from
newly deposited tailings occurs through loss of pure water, often
resulting in precipitation of a distinct salt crust on the tailings
surface responsible for inhibiting further evaporation. The top
layer of the deposit does not achieve any significant strength
under such circumstances, with tailings below the crust remaining
at an extremely high water content.
With relatively fresh water, diligent management of
deposition layer thickness and cycle time can produce dense
tailings. However, even relatively moderate levels of salinity in
Figure 70. Comparison of phreatic surface resulting from assumption of the tailings water can cause actual evaporation rates to fall
homogeneous, isotropic hydraulic conductivity with assumption that significantly below the potential evaporation rate, Ep. This
hydraulic conductivity decreases exponentially down the beach. reduced evaporation rate occurs even with an open body of saline
Republished with permission of ICE Publishing, from Blight (1994); water and is exacerbated by crust formation. Firstly, the crust is
permission conveyed through Copyright Clearance Center, Inc. almost always white and the surface reflectivity (albedo) values
as high as 0.83 (Fujiyasu, 1997), meaning most of the incoming
In Figure 70, the assumption of isotropic permeability solar radiation is reflected. Secondly, once the salt crust dries, it
produces a clear overestimate of the phreatic surface height and acts somewhat like a surface mulch, meaning moisture flow must
thus underestimate of effective stress and strength. Although this occur in vapour form, further reducing the rate of evaporation.
may be a conservative outcome, it is surely no way to design The vastly constrained reduction in evaporation that occurs
major structures, i.e. blind faith in implied conservatism. It also due to crust formation means the tailings consolidate extremely
illustrates the folly of trying to fit a seepage model that is based slowly and in-situ void ratios remain high for some time, a highly
on assumptions of homogeneous, isotropic hydraulic conductivity undesirable scenario. Some mining operations have therefore
to measured pore water pressures of a system that is more akin to instituted intervention techniques to mechanically break up the
the exponential relationship evaluated in Figure 70. salt crust, often referred to as ‘mud farming’. Equipment has been
Another critical aspect of seepage analyses is the presence of specifically developed for this role, such as amphiroles. Newson
unsaturated zones within TSFs and the associated reduction in and Fahey (1998) demonstrated the benefit of crust removal by
hydraulic conductivity. While the presence of unsaturated zones measuring actual evaporation rates both before and after salt crust
might in some instances improve the stability of tailings (as removal at a Western Australian gold mine TSF. Figure 71 shows
suction increases effective stress and thus shear strength), seepage the results, and compares the measured evaporation rates with
of the pore-water towards downstream slopes will cause the measured pan evaporation rates (using tailings water). As can be
initially unsaturated zones to re-saturate and create zones with seen, the salt crust causes a decrease in evaporation of about 5 or
entrapped air. The effects of air entrapment on hydraulic 6 times, thus significantly reducing loss of water and strength
conductivity in the zones of positive pore-water pressures have gain.
been studied and reported by Hwang (1999, 2002), Bicalho
(1999), Bicalho et al (2011) and Lee (2011). The air entrapment
creates zones of reduced hydraulic conductivity, especially in the
areas where pore-water pressure is lower near the phreatic line and
towards a downstream slope. This lower hydraulic conductivity in
turn restricts the pore-water flow and increases the phreatic line
especially near downstream slopes of TSFs. Collins (1997) and
Collins and Znidarcic (2004) demonstrated the importance of
accounting for unsaturated seepage conditions when analysing
rain infiltration and its effects on triggering landslides. Similar
mechanisms may affect the stability of TSFs.
We suggest that a great deal more attention needs to be placed
on rational seepage analyses, taking account of measurable
hydraulic conductivity variations, including anisotropy and
unsaturated conditions, where possible. This is not a trivial
expectation, but enormous attention to aspects such as in-situ Figure 71. Measured evaporation at a gold mine TSF in Western Australia
strength and state characterisation have greatly improved current (after Newson and Fahey, 1998).
practice; perhaps it is time similar attention was paid to improving
seepage predictions and interpretations of in-situ measurements. 7.6 Aging

7.5 Crust formation. The increase over time of the resistance and/or stiffness of soils
has been identified as aging (Mitchell et al., 1984; Skempton
The salinity of tailings water in most TSFs is generally relatively 1986; Schmertmann, 1991). Furthermore, it has also been
low. However, there are many operations where water salinity is recognized that the resistance to liquefaction increases with the
extremely high, an excellent example being many mining
age of the deposit (Kramer et., 1998; Leon et., 2006). In particular,
operations in the Western Australian goldfields, where salinity
may be as high as 300 g/l total dissolved solids. Other examples Troncoso and co-workers studied the effect of aging on the
include many process residues, such as bauxite red mud, which liquefaction resistance of a copper tailings sand using cyclic
also have high process water salinities. triaxial tests (Troncoso et al., 1988). This study was carried out
The key concern with high salinity tailings water is the based on a series of cyclic triaxial tests performed on
inhibition of evaporation. Newson and Fahey (1998) reported on ‘‘undisturbed’’ samples retrieved from an old tailing dam at
an extensive study of various mining operations in Western different depths, which basically means different age of
Australia, with a prime focus being on effects of evaporation from deposition. Additionally, tests on fresh samples reconstituted in

164
the laboratory were conducted. The test results are summarized in 7.7.2 Tailings Deposition.
Figure 72, where the cyclic stress ratio required to produce a state Tailings deposition in cold temperatures may:
of softening with 5% double amplitude strain has been used as
criterion of liquefaction. Restrict the length of the slurry pipelines to prevent freezing
of the pipeline, if measures are not taken (e.g. isolating the
pipeline).
Limit the possibility of regularly changing the deposition
point, resulting in deposition occurring for extended periods
from the same deposition point.
Cause freezing of the beach, potentially resulting in embedded
permafrost layers if overly thick layers are deposited before
thawing occurs.

The first two issues are more of an operational nature, which


Figure 72. Curves of cyclic resistance ratio (a), and increase in cyclic can be addressed through appropriate procedures, whereas the
resistance with time (b), (after Troncoso et al., 1988). third may have a significant impact on stability of the tailings dam.
Embedded frozen layers will restrict permeability, or even result
in effectively impermeable layers, which will affect the rate of
It can be observed that cyclic resistance ratio tends to increase consolidation and result in layers of loose, saturated, contractive
by a factor of 3.5, 2.4 and 2 for the samples of 30, 5 and 1 years materials in the deposited tailings. If these layers are in the
of deposition, respectively. These experimental results indicate structural zone of the tailings dam, stability will be compromised.
that for stability analyses during the period of post closure, the If the layer stays frozen, the strength of the layer will be high, but
aging effect should be considered. This type of study can be done possible future thawing will cause an increase in pore water
when the tailings dam has been in operation for several years, pressure, reduced effective stress, and increased water content
when it is possible to retrieve samples at different depths, which within the material. For these reasons, the risk associated with
are associated with different years of deposition. Testing this impermeable layers such as strength loss during thawing, frozen
batch of ‘‘undisturbed’’ samples, it should be possible to establish layers, or permafrost, should be avoided in deposited tailings near
or within the structural zone of the tailings dam.
the variation of the cyclic strength with the age of deposition,
Knutsson (2018) examined tailings deposition in cold regions
which allows an estimation of the improvement of the cyclic
and did a numerical study of the potential formation of permafrost.
resistance with time.
The risk of generating permafrost due to tailings deposition exists
even in regions with no natural permafrost, as material being
7.7 Freezing
frozen during winter might not fully thaw during the following
Cold weather is a condition affecting tailings management in summer, due to added height of the tailings on the surface (e.g.
several ways with regards to both dam construction and deposition continued tailings deposition). The study simulated conditions in
of tailings. northern Sweden above the arctic circle with an average yearly
temperature of 0.1 C. Tailings deposition was studied for an
7.7.1 Dam Construction annual raise of 1, 2 and 3m, as well as for the yearly raise being
Freeze and thaw processes may reduce the density of dam deposited during 4, 8 and 12 months. For the first two scenarios,
construction material to a depth related to the material properties, deposition starting in January (winter) and July, respectively,
the temperature over time as well as the depth of snow cover. were studied. The following tailings characteristics were
Effects to depths of up to 3-4 m, if not more, at locations with low considered:
temperatures (below 0 °C) over long periods of time (6 months or
more) are not unusual. This means that the zone of material Specific gravity (Gs) – 2.8
affected by freeze and thaw must be re-compacted before the next Quartz content (q) – 40%
raise can be done. If the freeze depth is more than the rate of rise, Degree of saturation (Sr) – 33 and 100%
this may mean re-compaction is required for material at a depth Dry density ( d) – 1.3 and 1.8 t/m3
below the new tailings surface (Figure 73). A solution may be to
place a layer of “frost protection material” on the dam crest, which for four different scenarios, “Wet & Loose”, “Wet & Dense”,
is then removed before the next raise. With frost protection, only “Dry & Loose” and “Dry & Dense” which together cover a wide
the top surface of the material (dam crest) needs re-compaction range of in-place tailings possibilities. The results from the study
before the next raise is constructed. demonstrated:

Decreased water content results in less permafrost generation,


which has a larger impact than density.
Thickness of permafrost layers is nearly proportional to the
deposition rate (RoR).
A shorter deposition period generates more permafrost when
starting in winter.
Deposition starting (for 4 or 8 months deposition periods) in
January (northern hemisphere winter) results in more
permafrost compared to starting in July.
Figure 73. Illustration of a possible situation where freeze depth (here
When considering a 12 month deposition period, “Wet &
illustrated as 3m) is more than a yearly rise (here illustrated as 2m) results
Loose” material always generated permafrost for the studied
in a zone affected by freeze and thaw (and therefore reduced compaction)
deposition rates (1,2 and 3 m/a). Conversely, “Dry & Dense”
corresponding to 1.5 dam raises, e.g. 1.5 dam raises need to be re- material never generated permafrost (see Table 6).
compacted (zone to be re-compacted) before next raise can be done.

165
Table 6. Permafrost thickness [m] for 12 month deposition (e.g. constant considered as a first approximation in the verification of the TSF
RoR of tailings surface with time). stability and should be carried out with full knowledge of the
ultimate strengths of the materials involved, including those of the
Material Deposition rate (RoR) foundation soils.
1 m/a 2 m/a 3m /a Recently, Sadrekarimi (2016) proposed a modification to the
Wet & Loose 0.35 0.8 1.2 conventional, widely used limit equilibrium method of analysis to
Wet & Dense 0 0.425 0.875 account for various factors, including modes of shear (plane strain
Dry & Loose 0 0.25 0.65
compression, simple shear and plane strain extension). His
method allows for re-assigning shear strengths of regions (‘slices’
Dry & Dense 0 0 0
in the vernacular of the limit equilibrium method) where the peak
shear strength has been mobilised, to values equal to the large
strain shear strength. The focus was particularly on undrained
Freezing and thawing may, however, be used to improve shear strengths. The possibility offered by the modification
consolidation and shear strength of wet, soft tailings. Cyclic proposed by Sadrekarimi was to incorporate some degree of
freeze-thaw has been shown to significantly dewater the tailings. progressive failure into the simple limit equilibrium method.
During freezing, ice lenses develop and when they thaw, they Although the method does assist in providing insight into
leave “channels” in the material through which water can be vulnerabilities of some TSF slopes, it is overly brutal in the
drained. These “channels” then collapse and a volume decrease is instantaneous switch from peak strengths to large strain (residual)
achieved. For maximum effect it is crucial to make sure the strengths. No allowance for strain compatibility is made. The
released water is drained away from the tailings. Practically, this issue of brittleness of some tailings materials is over-simplified
is achieved by letting the top surface of wet, soft tailings freeze, using this approach.
thaw and drain a number of cycles (years) until maximum
consolidation is achieved. This method has been studied and used 8.1.1 Brittleness
in northern Canada to facilitate and reduce costs of TSF closure As noted in Section 1.3, the GISTM requires that brittle potential
capping (Beier and Sego, 2008, Proskin et.al., 2010). failure modes be identified and addressed. No clarity is provided
on what constitutes brittle behaviour however. There are two
8 METHODS OF ANALYSIS. aspects that we believe must be considered to address this
This section discusses methods of analysis considering static question, the first being the amount (percentage) of loss of
loading and dynamic loading conditions, with analyses strength post-peak, and the strain over which this strength loss
approaches that vary from relatively straightforward (e.g. limit occurs. Unfortunately, it seems that too often only the former
equilibrium) through empirical approaches, to advanced aspect is considered.
numerical modelling. However, each of these topics, particularly One outcome of the forensic investigations into the Aberfan
the latter, warrant a state-of-the-art paper in their own right, so this disaster in Wales in 1966, in which a flowslide of colliery residue
discussion is necessarily kept as brief as possible. destroyed large parts of a nearby village and resulted in 144
fatalities, was the definition by Bishop (1967) of what he termed
8.1 Static loading conditions the Brittleness Index, IB:

The limit equilibrium method assumes a potential slip surface and


the mobilised shear strength along the surface is determined and = (34)
compared with the available shear strength. The soil (tailings) is
assumed to be a perfect rigid-plastic material, and therefore, when
materials of quite different stress-strain behavior are involved, an where is the resistance to shear on the potential failure surface
engineering judgment is required to establish the most probable and the subscripts f and r relate to peak (failure) strength and
strength that is mobilised simultaneously by these materials along residual strength respectively. Thus, a material that exhibits no
the considered slip surface. The most common method is to post-peak loss in strength has a brittleness index of zero.
discretize the soil mass into a series of slices that are analyzed as
free bodies. To satisfy both the equilibrium of forces and moments
in the slices, it is necessary to transform an inherently
indeterminate problem into a determinate one. Depending on the
simplifications and assumptions made to satisfy the equilibrium
of forces and moments in the slices, different procedures have
been proposed, among which the one developed by Morgenstern
and Price (1965) is probably the most widely used in tailings
engineering.
Along the slip surface under analysis, both the available
mobilized shear strength and the shear stress required for
equilibrium (driving stresses) are evaluated. Then, the evaluation
of a Factor of Safety (FoS) is straightforward as the ratio between
these quantities. The FoS is a parameter, or an index, that provides
a simple indication of the state (in terms of its stability) of the
earth structure. Figure 74. Illustration of brittleness, which is a function of both the amount
The reports on the failures of Aznalcóllar (Los Frailes), Mount of post-peak strength loss, and the strain over which this strength loss
Polley, Samarco, Brumadinho and Cadia tailings dams (Gens, occurs.
2019; Hoffman, 2015; Morgenstern et al, 2016; Robertson et al,
2019; Jefferies et al, 2019) allow us to point out that when the Figure 74 illustrates this concept, which can be applied to
ultimate (large strain) strengths of the materials are properly either drained or undrained representations of soil strength. In the
established, the FoS obtained from the limit equilibrium analyses latter case the undrained strength is often normalised with respect
were systematically lower than one, consistent with the failures to the relevant vertical effective stress. Material A in this figure
that have occurred. In this context, the authors' opinion is that the shows a significant loss of strength (with a brittleness index close
use of the limit equilibrium analysis is useful, but it should be to unity) over a very small increment of strain. Such behaviour

166
was demonstrated for Merriespruit tailings (where failure of a TSF
in South Africa in 1994 caused 17 fatalities) by Fourie and
Tshabalala (2005). Tests were carried out on K0 consolidated
specimens subjected to undrained triaxial compression, with
results shown in Figure 75. As can be seen, the peak strength was
mobilised at axial strains less than ½%, followed by very rapid
strength loss, with IB values ranging from about 0.4 to 0.8. These
tailings exhibit both a high brittleness (high IB) and a rapid loss of
strength.

Figure 75. Deviator stress vs axial strain for undrained triaxial Figure 76. Variation of deviator stress with axial strain for undrained
compression tests on K0 consolidated specimens of Merriespruit tailings triaxial compression test on anisotropically consolidated bauxite residue.
(after Fourie and Tshabalala, 2005). Numbers in the legend refer to the
void ratio prior to shearing. Both the Merriespruit (gold) tailings and the bauxite residue
referenced above were anisotropically (K0) consolidated prior to
Material B in Figure 74 shows approximately the same values shearing. One exhibited extreme brittleness, with very rapid loss
of peak and residual strength as material A, but the strain over of strength, whereas the other was more ductile. However, it is not
which the strength loss occurs is many times greater than that of only material type that influences brittleness. The same material,
material A. Is material B therefore brittle? In terms of the consolidated along different stress paths prior to being subjected
percentage strength loss, yes it is. However, the loss occurs to some form of undrained loading, may exhibit different degrees
gradually (in terms of strain, not necessarily time). Perhaps a key of brittleness. This is illustrated in Figure 77, after Been (2016).
distinction between material A and material B is that post-peak Figure 77 shows the same material, prepared loose, subjected to
behaviour of material B is likely to provide some indicators – three different consolidation stress paths prior to being sheared.
visual and otherwise – of the loss of available strength, because Path A was isotropic consolidation. Paths B and C both involved
significant movements will occur prior to the terminal (residual) drained, constant mean effective stress consolidation, but sample
strength being mobilised. Measurements using inclinometers, B was not loaded into the ‘instability zone’ whereas sample C was.
shape accelerator arrays (SAA’s), or InSAR techniques are likely The instability zone is the region between the line defining the
to detect movements prior to residual strength being mobilised critical friction ratio (i.e. ‘M’) and the instability locus. As can be
over the entire potential slip surface. More about monitoring seen from Figure 77b, the slightest perturbation applied to sample
techniques in Section 9. C results in immediate loss of strength, whereas in sample B a
Behaviour similar to that of material B is shown in Figure 76,
small amount of strain is required to fully mobilise peak available
which is for an undrained triaxial compression test on a sample of
strength; for sample A, even more strain is required. Similarly,
bauxite residue (red mud) anisotropically consolidated to a
experimental data suggests that the brittleness index of nonplastic
vertical effective stress of 600kPa. Once again, the peak strength
is mobilised at a very small axial strain (less than 1%), but even soils is a function of their state parameter normalised by the slope
after 25% strain, the residual strength had not yet been fully of their CSL (Jefferies and Been 2015).
mobilised.
There is thus a great deal of difference between material A
and material B in Figure 74, in terms of the potential risk of a
catastrophic failure. Material A is likely representative of the
Feijão tailings, which exhibited both high brittleness and rapid
strength loss. A challenge facing TSF designers is to determine
what the threshold of truly brittle behaviour is. Assuming there is
the potential for significant loss of strength (high IB), over what
strain does this become a significant concern? Is it 1% strain
(probably expressed as shear strain in order to enable comparison
with numerical models – see section 8.2), is it 5%; 15%? This is a
major challenge for the industry. Simply designing a stabilising
buttress for any material that exhibits some strength loss (i.e.
strain softening), irrespective of the amount of strain required to
fully mobilise residual strength does not seem logical. For less Figure 77. Illustration of importance of consolidation stress path on
brittle materials exhibiting a degree of strength loss, it may be subsequent stress-strain behaviour, particularly the strain at which stress
more prudent to complement limit equilibrium analyses with loss is triggered (after Been, 2016; courtesy Australian Geomechanics
appropriately sophisticated numerical modelling (where Society).
‘appropriate’ refers to the ability to capture key features of
According to Robertson (2010), cohesive soils (tailings) with
behaviour, in this case being strength loss with strain).
a shear strain to reach peak undrained strength greater than about
5% and a gradual drop-off in resistance after reaching peak

167
strength are less likely to experience flow liquefaction, which
applies, for example, to many high-PI clays. As with any guidance
rule, the 5% shear strain indicator mentioned here may be
misleading. For example, when consolidated isotropically, rather
than anisotropically, the bauxite residue shown in Figure 76
reached peak strength at a much higher axial strain than the K0
consolidated specimen shown.
There is also the issue of structure (or fabric) and its impact
on brittleness. According to Leroueil and Hight (2003), a feature
of sensitive fine-grained soils involved in flow failures is their
relatively low plastic limit and small strain required to reach peak
undrained shear strength. Although some guidance does exist, it
may be expecting too much of the CPTu test to provide
meaningful information on brittleness of a tailings material.
Shuttle and Cunning (2007) and Jefferies and Been (2006) suggest
that when a soil has a state parameter greater than -0.05, strain
softening and strength loss in undrained shear can be expected.
They do not speculate on rate of post-peak strength loss however.
In a similar vein, Robertson (2010) suggests that soils with a
Qtn,cs>70 are likely dilative and strain hardening in undrained
shear is likely. Gens (2019) provides some extremely valuable
insights to this problem. Using a finite element approach with the
constitutive model CASM, as described by Yu (1998), he reports
the response to CPTu of soils having the same properties, except
for the strain required to reach peak strength and the strain over
which strength loss occurs. The various idealisations are
illustrated in Figure 78, where sample G has an IB value close to
zero and sample A an IB of about 0.66. Significant detail is
provided on the resulting profiles of tip resistance and excess pore
pressure generated during the simulations, a summary of which
Gens (2019) provides in Figure 79, which shows an increase in Qt
with decreasing IB, i.e. the more brittle the soil, the smaller the tip Figure 79. Variation of normalised tip resistance (Qt) and pore pressure
resistance generated, even for the same value of state parameter. parameter Bq with brittleness index (after Gens, 2019).
Also note, the peak undrained shear strength is the same for all the
simulations, whereas the resulting Qt value is not. Gens (2019) As before, the best way to address this issue is using a
emphasises that these results are preliminary, but the implications mechanics-based approach to analysing CPTu results. The
are potentially extremely important, as it throws into question our ‘widget’ approach of Shuttle and Jefferies (2016), as discussed
understanding of the limitations of the CPTu test. We expect this before, uses analyses of the spherical cavity expansion problem
topic is likely to be a focus of active research in coming years. with realistic soil models to derive a surrogate of CPTu behaviour,
with a scaling factor being required to make the transition from
spherical cavity to CPTu. Although the ability of the widget to
predict in situ remains a pending task (Figure 47), perhaps a
focus on an approach such as this provides a way to accommodate
materials with various rates of strength loss.

The role of bonding in determining brittleness


No discussion of brittleness would be complete without some
discussion of potential bonding (or cementation) in tailings
producing a brittle response to undrained loading, particularly in
light of the suggestions made by the ERP (Robertson et al, 2019)
relating to the Feijão TSF failure that the iron ore tailings
developed a bonded structure after deposition. As summarised by
Schnaid (2021), the bonding may have rendered the tailings stiff
and potentially brittle, which he suggests is consistent with the
lack of observable deformations prior to failure and the sudden
and rapid behaviour once failure was triggered. However, the
CIMNE report (2021) comes to the opposite view, as mentioned
before, and concludes there was no evidence of bonding.
To provide guidance on whether a soil is structured (whether
due to bonding, cementation or even aging – see for example
Figure 78. Stress-strain curves of soils used in CPTu simulations, as Troncoso et al, 1998), Schneider and Moss (2011) suggested a
reported by Gens (2019). rigidity index KG, which was slightly modified by Robertson
(2016) to , where

=( )( ) . (35)

and G0 = (Vs)2, Vs is the shear wave velocity and is soil (tailings)


density.
Using this approach, it is suggested that if KG > 330, the soil
can be classified as structured, and generalised CPTu based

168
empirical correlations may have less reliability (Robertson, 2016). example, is often defined for a 2% probability of exceedance
However, the situation is further complicated if the material is within a lifespan of 50 years, which corresponds to a return
partially saturated. As discussed by Robertson et al (2017), period of 2475 years.
suction hardening increases the small strain stiffness (and thus Maximum Credible Earthquake (MCE): This is the largest
Vs), an effect that could erroneously be interpreted as possible earthquake that could occur at the project site. It is
cementation. recommended to establish the MCE using the deterministic
approach. This earthquake is associated with the post-closure
8.2 Seismic stability analyses condition of the TSF and represents the ultimate possible
seismic loading under which the tailings facility must remain
8.2.1 Defining loading conditions in place without collapsing. Nevertheless, in countries with
To carry out seismic stability analyses of a tailings dam, it should very high seismicity such as Perú, Chile, Japan and Indonesia,
be recognised that there are two clearly different stages: operation the MCE must be considered during the operation of the TSF,
and post-closure. The operational phase, which often lasts a few but accepting important damages that will not compromise the
decades (20 to 40 years in many cases, with some mining confinement of the stored tailings and water.
operations being over one hundred years old), is practically
always linked to the construction of the facility and the global The Global Industry Standard on Tailings Management
growth of the TSF. During this stage it is possible to foresee the (GISTM) recommends the seismic design criteria summarised in
existence of a routine situation (normal operation) and special Table 1. It is worth noting that the application of an earthquake
situations, e.g. associated with some critical but brief construction with a return period of 10,000 years is associated with a 1%
issue (such as excavation at the toe of the existing dam to remove likelihood of probability of exceedance in a lifespan of 100 years,
weak foundation soils for the new stages). To face these practical which can be considered unnecessarily conservative. A return
situations two seismic events are normally considered: period of 5,000 years is associated with a 2% likelihood of
Operational Basis Earthquake (OBE) and the Maximum Design probability of exceedance in a lifespan of 100 years, equivalent to
Earthquake (MDE). The MDE is more demanding than the OBE, a 1% exceedance in 50 years, which is commonly referred to as
and it must be ensured that for this seismic event, considering the the Safe Shutdown Earthquake that is applied in seismic design of
special conditions of the TSF, the tailings and water remain critical structures such as chemical industries, liquefied natural
perfectly stored. In Australia and New Zealand, the MDE is gas industries and power plants, among others. Even though it is
termed the SEE (Safety Evaluation Earthquake); the associated an extreme seismic event, its use in highly sensitive facilities
descriptions are that the OBE only results in minor damage and makes sense, and therefore, it can also be considered applicable to
the TSF and appurtenant structures and equipment remain TSFs.
functional, whereas the SEE causes damage, which is acceptable,
but there is no uncontrolled release of water (or tailings) from the Table 7. Seismic design criteria.
facility.
Consequence classification Seismic criteria – annual exceedance probability
Additionally, the long-term stability of decommissioned (also Operations and closure Passive – closure (passive
termed ‘closed’) TSFs as well as those that have been truly (active care) care)
‘abandoned’, i.e. there is no identifiable owner who can safely Low 1/200 1/10 000
close the facility – hopefully fast becoming a problem consigned Significant 1/1 000 1/10 000
to history, constitutes another design consideration. Seismic High 1/2 475 1/10 000
stability must be guaranteed in the long term, and therefore, long Very high 1/5 000 1/10 000
Extreme 1/10 000 1/10 000
return periods in the selection of the earthquakes must be
considered. Accordingly, the Maximum Credible Earthquake
(MCE) must be estimated and used in the stability analysis. The seismic hazard analysis has to consider the different
The probabilistic seismic hazard analysis (PSHA) and potential seismic sources in the area of the project. For example,
deterministic (DSHA) methods for the assessment of seismic in subductive seismic environments, there are three potential
hazard analysis are usually presented as antagonistic approaches. seismic sources: interplate (thrust), intraplate (intermediate depth
However, they may in fact complement one another when in-slab earthquakes caused by down-dip tension in the subducting
estimating the ground motions for design. In regions of high plate) and cortical (shallow crustal earthquake). Because
seismicity, where continental active faults and/or tectonic intraplate and cortical earthquakes attenuate much faster than
boundaries generate large seismic events, which have been interplate earthquakes, the appropriate selection of the attenuation
systematically reported in the last 200 - 300 years, the DSHA is law becomes relevant. Additionally, when considering intraplate
more suitable. This is because in regions of high seismicity, the earthquakes, the resulting seismic disturbance at the site of the
empirical evidence of extreme seismic events is reliable and dam is highly sensitive to the definition of the hypocenter
provides valuable real information about the ground motion locations. This is an important issue where no consensus has been
effects (Wang, 2010). However, in areas where seismicity is low achieved yet, and in the meantime the worst case is usually
or diffuse, the PSHA is more appropriate for establishing the adopted resulting in the selection of an intraplate earthquake just
potential earthquakes that can occur at the project site. In general, below the site of the project.
the different seismic disturbances that must be considered are: It is common to select a time history that provides the largest
Operational Basis Earthquake (OBE): This is a probable peak ground acceleration. However, the frequency content of the
earthquake to which a tailings facility can be expected to be selected earthquake could be a more relevant criterion, especially
exposed to during its design life; therefore, the tailings when analyzing high dams that tend to have a low predominant
disposal facility is expected to remain fully operational when vibration frequency (or large period). In this case, it is
subjected to such an earthquake. This earthquake could be recommended to first estimate the natural period of the tailings
defined, for example, for a 10% probability of exceedance dam, and then the seismic hazard analysis should identify those
within a 50 year period, which corresponds to a return period potential earthquakes that contain intensive energy at around the
of 475 years. natural frequency of the dam. These can be earthquakes of distant
Maximum Design Earthquake (MDE), also termed SEE, as epicenters that generate and impose surfaces waves, which result
discussed previously: The main purpose of designing against in seismic events containing low frequencies.
this earthquake is to safeguard loss of life, uncontrolled A balance must be struck between an earthquake with the
failure, collapse and loss of confinement, which may lead to highest possible acceleration and with a frequency content as
catastrophic consequences. This strong ground motion, for close as possible to the predominant frequency of the TSF. In

169
summary, two different design earthquakes must be defined: one There are several commercial software packages that can be
with maximum PGA and another with a frequency content that used to perform seismic analyses, but it is important to point out
matches, or it is close enough to, the predominant frequency of that the constitutive laws that are adopted to model the stress-
the dam. Therefore, the seismic analysis should consider at least strain behavior of the materials have a significant effect on the
six records (OBE x 2; MDE (SEE) x 2; MCE x 2). results. The discussion of the different constitutive laws that have
been developed is beyond the scope of this paper. We stress
8.2.2 Pseudo-Static Analysis – Limit Equilibrium however, that the selected constitutive law must show a
This procedure was very popular before computers were compromise between the predictive capacity of the model and the
developed to the level that made possible a dynamic stress-strain practicality of evaluating the input parameters required by the
analysis in only a few hours. In this context, many engineers see model. Interesting reviews on constitutive models have been
no reason to continue using the pseudo-static approach to verify reported by Lade (2005) and Prevost and Popescu (1996).
the stability of TSFs. Although perhaps true, it is necessary to Under static loading conditions, the mechanical response
retain “the big picture” before entirely eliminating the pseudo- depends mainly on the geometry, the magnitude of the acting
static analysis approach as one of the alternatives for assessing the forces and the stress-strain behavior of the materials involved,
stability of TSFs. with the physical stability ultimately controlled by the distribution
Seismic stability evaluated by pseudo-static analysis requires of stresses and their magnitudes relative to the mobilized
the use of seismic coefficients, which represent inertial forces strengths. On the other hand, in the case of seismic disturbances
through the application of permanent forces using a limit of earth structures, the shaking induces inertial forces that are
equilibrium approach as discussed in section 8.1. Strictly largely dependent on the amplitude and frequency content of the
speaking, the replacement of a seismic disturbance by a static seismic load and on the predominant natural frequency of the earth
force is theoretically incorrect. Nevertheless, from an engineering structure. The well-known phenomenon of resonance occurs
design point of view, it has been used successfully in countries when there is a perfect match between the frequency of the load
with high seismicity, in both building design and in the stability and the natural frequency of the structure. This issue is important
assessment of earth structures (Sano, 1906; Housner, 1984; in the selection of the design earthquakes as discussed above.
Towhata 2008). The seismic coefficients (horizontal and vertical) In addition, every dynamic response is affected by the loss of
were traditionally estimated from the expected peak horizontal energy that normally takes place during vibrations. This
acceleration of the site, but more recently the allowable level of phenomenon is called damping, which for simplicity is usually
seismic displacements of the slopes has also been included (Bray characterized as viscous type behaviour, and therefore increases
and Travasarou, 2009; Papadimitriou et al, 2014). Considering the as the strain rate increases. It is necessary to point out that soils
high seismicity of Chile, Saragoni (1993) proposed the following develop a loss of energy that is of the hysteretic type (it does not
relationships for the horizontal seismic coefficient, kh, for depend on the speed of deformation). In soils, the damping ratio
interplate (shallow thrust) earthquakes: is defined as the area of one stress-
normalized by the area of the equivalent input linear elastic energy
(Figure 80a, after Kokusho 1980, Ishihara 1996). This hysteretic
damping ratio is strongly dependent on the level of deformation
as shown in Figure 80b, (after Kokusho 1980, Ishihara 1996). At
(36) large deformations close to failure, the damping ratio reaches
values as high as 25%, which means a significant loss of energy,
which in general, is favorable for the attenuation of the seismic
response.

The vertical seismic coefficient is commonly assumed as 1/2


or 2/3 the horizontal seismic coefficient. In general, the
incorporation of this vertical seismic coefficient does not have a
real effect of the resulting factor of safety.
The pseudo-static analysis procedure is certainly attractive
because of its simplicity and because empirically it has been
shown to provide a reasonable estimation of the level of stability
of an earth structure subjected to earthquakes. Nevertheless, due
to the rather crude assumptions inherent to this method, its main
disadvantage is associated with the need for engineering
judgment, which is not assured if applied by inexperienced
professionals. Additionally, for tailings dams where liquefaction Figure 80. Damping ratio definition (a), and damping ratio of Toyoura
can be a major problem, or where foundation soils are complex, sand (b), (after Kokusho 1980).
pseudo-static analysis must be used with careful attention to the
strength of weak materials. Similarly, however, in these The output of a dynamic analysis can be greatly affected by
situations, strain-strain numerical models have also to be applied the mathematical formulation adopted to model the damping. In
with the same care, as the variety and complexity of input the dynamic analysis of earth dams (including tailings dams), it is
parameters and the range of available modeling options reduces common to use the so-called Rayleigh Damping (RD), which is
the likelihood of similarity of outcomes when undertaken by mathematically very attractive, but is difficult to estimate the
different engineers (including those with extensive experience). values that represent the real problem under analysis. The RD is
essentially a viscous damping that is established as a linear
8.2.3 Numerical modelling combination of mass and stiffness. Thus, the damping matrix C is
With the development of computers with large memory and given by:
computing capacity, the numerical methods of finite elements and
finite differences in 2D or 3D can feasibly be used to analyse and
predict the behavior of TSFs under both static and seismic stress [ ]= [ ]+ [ ] (37)
loading conditions, including different scenarios of seepage and
stages of construction.

170
where, and are constants of proportionality and M and K are It is of great importance to note that in the data used in this
the mass and stiffness matrices respectively. chart, which comes from different earthquakes that occurred in
different parts of the world, there are no cases of liquefaction for
varies with the response frequency. Therefore, it is important to values of qc1-Ncs greater than 170. This allows the conclusion to be
adjust this damping ratio to the fundamental frequency of the drawn that there is a threshold value of the tip resistance (qc1-Ncs =
analyzed problem, otherwise the results could be significantly 170) above which liquefaction would not occur, regardless of how
overdamped in terms of lower accelerations and smaller strong the earthquake is. A corollary for areas of high seismicity
deformations. Importantly, the RD is different from the hysteric is that compaction of non-plastic materials (e.g., tailings) to qc1-
damping ratio indicated in Figure 80a (after Kokusho 1980, Ncs > 170, prevents the occurrence of liquefaction.

Ishihara 1996), and, therefore, the number that is adopted for the For earthquakes of magnitude other than 7.5, overburden
RD requires experience and a deep understanding of the variables pressures different than 1 atm and nonzero driving static shear, the
of the problem under analysis. It is not an exaggeration to say that CRR obtained from this chart must be modified using the
Magnitude Scaling Factor, MSF, the effective overburden stress
in a dynamic problem, damping is a fundamental parameter,
factor, K , and the shear stress factor, K , respectively (National
which is usually not well characterized. Nevertheless, when
Academies US, 2016). Accordingly, once the representative value
constitutive models are used that reproduce the stress-strain curve
of penetration resistance qc1-Ncs has been established, the cyclic
during loading and unloading, damping results automatically from resistance ratio, CRRM=7.5, is obtained from the corresponding
the load-unload loops that the model generates. In the absence of chart. This value is then corrected as follows to obtain the cyclic
liquefaction, the analysis of these loops allows verification of the resistance ratio, CRR, representative of the actual conditions:
capacity of the model to reproduce the behavior of the soil or
tailings. = × × × (38)
.
The use of stress-strain analyses is a clear improvement on the
limit equilibrium analysis method, but it is extremely important to As can be observed from the above expression, the effects of
keep in mind that stress-strain analyses must always be verified earthquake magnitude, overburden pressure and static shear
by alternative means. A first verification of the capability of the modify the CRRMw=7.5, which means that the curve of Figure 81
model is to check how well it reproduces the results of the moves vertically when these effects are considered. Therefore, the
performed laboratory tests. Also, it is recommended to verify the threshold value of the tip resistance (qc1-Ncs = 170) discussed
model in simple conditions where the soil response is known. In above holds valid for higher confining pressure, static shear and
addition, it should be reviewed by independent experts to verify earthquake magnitude.
the quality of the analysis.
8.3.1 Magnitude Scaling Factor, MSF.
8.3 Empirical approaches To account for increased likelihood of liquefaction with an
Given the difficulty in recovering high quality undisturbed increase in the magnitude of an earthquake, Seed and Idriss (1982)
introduced the Magnitude Scaling Factor, MSF, which is based on
samples from the field for laboratory testing, for the estimation of
the concept that the greater the magnitude of the earthquake the
the CRR the use of in-situ tests is usually recommended, for
greater its duration, and therefore the larger the number of cycles
example, SPT, CPTu, or the vane shear test. In this context, a
of the seismic disturbance. It has been recognized that the MSF
significant amount of research has been carried out using the depends on several factors such as type of earthquake, distance to
CPTu to evaluate the onset of liquefaction through the CRR. the epicenter, soil type and characteristics of the site, among
Figure 81 presents a chart for estimating the CRR to magnitude others. However, the actual effect of these factors is still under
7.5 earthquakes based on the cone tip resistance normalized by investigation. Boulanger and Idriss (2014) incorporated the soil
overburden pressure and fines content, qc1-Ncs. characteristics by considering the values of the tip resistance qc1Ncs
and/or (N1)60cs in the MSF as shown in Figure 82.

Figure 81. CRR for Mw = 7.5, 'v = 1 atm and no static shear stress, as a
function of the equivalent clean sand qc1Ncs, (after Boulanger and Idriss Figure 82. MSF as a function of the penetration resistance. Reprinted from
2014). Boulanger and Idriss (2014), with permission from Elsevier.

This chart has been obtained from sites than are not associated 8.3.2 Overburden stress factor, K .
with tailings materials. Therefore, its use for tailings should be Results of laboratory cyclic tests have systematically shown that
carried out with due caution. The authors´ opinion is that for liquefaction resistance decreases with increasing confining stress,
tailings that classify as sandy soils or silty soils of low plasticity, as shown in the test results on tailings sand presented in Figure
the chart provides a good estimate of the CRR. However, in 83. Because the empirical field-performance database utilised
tailings containing plastic fines, the chart is likely to give cases associated with overburden pressures of around 1 atm (100
misleading results.

171
kPa), Seed and co-workers introduced the overburden correction However, as shown in Figure 85, it has been recognized that
factor, K , defined as the ratio between the CRR at a given K is a function of the relative density (Vaid et al., 2001). Different
effective vertical stress to the CRR at an effective vertical stress studies have provided more data that show that other factors, such
of 1 kg/cm2 (1 atm), considering the same conditions of density as the stress history, also influence the value of K (Boulanger
and other variables than may affect the CRR (Seed 1983). 2002, Hynes et al., 1998, 1999; Montgomery et., 2012; Vaid et al.,
2001; Seed et al., 1990, among others).
It is important to mention that there is experimental evidence
indicating that tailings materials have a less pronounced decrease
in cyclic stress ratio with confining pressure (Riemer et al, 2008;
Verdugo and Santos 2009; Verdugo 2011). This condition can be
explained by the greater compressibility shown by tailings
materials, that results in a fairly pronounced increase in density
with the confining pressure. Figure 86 presents the isotropic
consolidation curves of three natural sands and one copper tailings
sand. For a better comparison between these sands, the void ratios
have been normalized by the initial preparation void ratio of each
sand. From these results, the higher compressibility of copper
tailings sands is clear, at least in the range of pressures used up to
5 MPa.

Figure 83. Effect of confining pressure on the cyclic strength of a Peruvian


tailings sand. Republished with permission of ASCE, slightly modified
from Riemer et al. (2008); permission conveyed through Copyright
Clearance Center, Inc.

For natural sands, reported values of K are presented in


Figure 84, where significant scatter can be observed (Montgomery
et al., 2012). Figure 86. Compressibility of natural and tailings sands (after Maureira et
al., 2012).

A particular characteristic of tailings particles is their


angularity and sharp edges, which may crush during shear as
shown in Figure 87. Therefore, it is possible to postulate that the
angularity of the tailings sands results in a greater compressibility
generated by the breakage of the sharp edges, which increases the
density and packing, favouring a better arrangement of the
particles and their contacts.

0.1 mm
0.1 mm
Figure 84. Reported values of K .(after Montgomery et al., 2012).
Figure 87. Illustration of angularity of particles of copper tailings, before
triaxial test (left), and after test (right).

Accordingly, for tailings materials it is highly recommended to


assess the effect of the overburden pressure by means of suitable
laboratory tests.

8.3.3 Shear stress factor, K .


The chart developed by Seed and co-workers to estimate the
potential of liquefaction based on SPT measurements was
established for level ground conditions in which there is no static
driving shear stress. However, it is well accepted that the static
shear stress has a significant effect on the generation of pore
pressure, as well as on the development of cyclic strains (Vaid et
al., 1983, Pillai, 2001, Boulanger 2002, Idriss et al., 2003, Hosono
et al., 2008, Verdugo 2011, Park, 2020). Given this, a significant
amount of research has been done on this topic. Nevertheless, the
Figure 85. Effect of Dr on K values (after Vaid et al., 2001) reported results are not definitive, and it has even been considered
to jointly include the effects of confining pressure and static shear,
for which the coupled factor K has been proposed. Considering

172
that the cyclic resistance is sensitive to the static shear stress, and
it varies from sand to sand, in the case of tailings materials is
recommended to perform suitable laboratory tests.

9 MONITORING.

The Global Industry Standard on Tailings Management (GISTM)


devotes a substantial section of the document to issues of
monitoring (Principle 7). Of relevance to the discussions
presented in this paper, Requirement 7.2 requires, “Design,
implement and operate a comprehensive and integrated
engineering monitoring system that is appropriate for verifying
design assumptions and for monitoring potential failure modes.
Full implementation of the Observational Method shall be
adopted for non-brittle failure modes. Brittle failure modes are
addressed by conservative design criteria”.

9.1 Displacement monitoring.

9.1.1 Remote monitoring.


The GISTM provides no illumination on what constitutes ‘brittle’
behaviour, probably reflecting the lack of consensus on this issue).
However, the distinction made in the above quote is critically
important. It reflects the understanding provided by the Feijão
forensic reports that failure of extremely brittle tailings is unlikely
Figure 88. Experimental data from modified direct simple shear device
to be preceded by sufficient warning. Although there are a number
showing minimal vertical strain prior to rapid collapse of specimen (after
of publications that claim to have successfully predicted the Feijão
Reid and Fourie, 2019).
failure, mainly using Interferometric Synthetic Aperture Radar
(InSAR) satellite imagery, these studies have one thing in Despite the likely limitations of InSAR discussed above, there
common. Their predictions were done with the benefit of is likely significant value to be gained by using InSAR for
hindsight. None of the authors of this paper are aware of accurate monitoring of more ductile systems, as long as these systems
predictions of failure of a TSF made (and published) before the (where ‘system’ is used to indicate the combination of tailings,
failure. If such papers do exist, we would warmly welcome underlying foundation soils and any other soil or rock zone of
notification of such work. relevance, or even an embedded structure such as a decant facility)
In their forensic investigation of the Feijão failure, CIMNE are correctly characterised.
highlight a number of potential difficulties in using InSAR
imagery to detect an early indication of potential instability of 9.1.2 Local measurements of displacement.
brittle material, including the line-of-sight (N-S) of the satellites Although many mines utilise an array of surface survey
were in the same direction as the movement of the TSF, they do monuments for monitoring displacements, our experience is that
not separate different motion components, and the histories of such systems are rarely useful, particularly if used in isolation.
line-of-sight displacement measurements from the different Great value can be obtained from instruments such as
studies noted above vary quite significantly. It is noted that it is inclinometers, which are installed in vertical casings, or the more
unclear (using InSAR alone) if the area of the TSF of interest recent technology known as shape accelerator arrays (SAAs);
moved by 5, 15 or 30mm in the year prior to the failure. Holden both of these systems enable detection of horizontal movement
et al (2020) point out the challenges of correctly accounting for (primarily) with depth, thereby providing a continuous profile of
signal-to-noise ratios of InSAR measurements in many cases. horizontal movement with depth. These devices are particularly
InSAR has many potentially useful applications when dealing useful when a well calibrated numerical model has been used to
with ductile materials, but the application of the technology to produce predictions of expected movement, ahead of the
more brittle material remains questionable. measurements being made. This constitutes a true application of
the much-advocated Observational Method (Peck, 1969), which
Reid and Fourie (2019) report on a laboratory study using a
is rarely correctly implemented.
purpose-built direct simple shear (DSS) device in which stress
control of both horizontal (shear) and vertical stresses was
9.2 Pore pressure monitoring.
possible. Specimens were prepared at a positive state parameter
and consolidated under an applied shear stress (sometimes Although there are still numerous TSFs where pore pressures are
referred to as a stress bias). Samples were then gradually brought monitored using standpipe piezometers, which need to be read
to failure (drained conditions were maintained during this manually, our observations are that vibrating wire piezometers
process) by gradually reducing the vertical total stress (referred to (VWPs) are increasingly being used, with data acquisition being
in the paper as a CSD stress path test). A typical result is provided automated. This is the preferred option; obtaining continuous
in Figure 88, which shows vertical strain as a function of time. records of pore water pressure, with the ability to check data
Initially, there is a slight expansion (negative vertical strains) due whenever required, is immensely beneficial to a robust tailings
to elastic unloading, but when failure is triggered, contractive management system. However, it is extremely important that the
strains occur extremely rapidly, too rapidly to track accurately, data from VWPs are interrogated by an experienced engineer,
and complete liquefaction occurs. It is highly unlikely that such with issues such as regular checks of calibration values being
small pre-failure deformations could be detected, even by a made, and interpretation of data being made in the light of a
sophisticated system such as InSAR. reasonable seepage model. Given the discussions provided in
section 7.4, it is unlikely that interpretations made on the
assumption of homogeneous and isotropic hydraulic
conductivities will automatically be rational. Emphasis must be

173
placed on developing and maintaining a calibrated seepage model dam-break studies. The results obtained from a dam-break
wherever possible. analysis have been shown to be sensitive to the following
parameters and conditions (this list is not necessarily exhaustive):
9.3 Other instrumentation.
Amount of both supernatant tailings liquor and free water in
Monitoring systems based primarily on a combination of the impoundment.
displacement monitoring and/or pore pressure sensors are Extension, shape, and location of the considered breach.
currently common at many TSFs worldwide. Recent high-profile Residual strength of stored tailings.
failures of TSFs, as discussed at some length in this paper, have Topography of the zone downstream of the dam
galvanised the introduction of many more, largely unproven
technologies. Some of these may well prove highly applicable and The strength (viscosity) of tailings released in a flowslide
economically viable to implement, such as some of the event as well as the downstream topography are variables to be
geophysical based techniques, but until a solid database of proven quantified when undertaking a dam break study. The viscosity is
success becomes available, caution is warranted. The University usually characterised as being deterministic in nature. However,
of Western Australia currently hosts a website focussed on the amount of fluid retained within an impoundment is
‘Meaningful Monitoring’, accessible at: probabilistic as well as being perhaps the most important
www.tailingsmonitoring.com, where advantages and condition that controls the run-out distance of the tailings, should
disadvantages of existing as well as emerging technologies are a flowslide occur (Rourke et al., 2015). The characteristics of the
discussed. breach that forms, enabling tailings release, cannot be measured
but rather requires a prediction that will have to rely to some
10 GOVERNANCE extent on engineering judgement. Therefore, a dam-break analysis
This paper focussed on geotechnical engineering aspects of safely can range from excessively conservative to excessively
unconservative. Given this, we strongly recommended that dam-
managing TSFs. However, we are cognisant that such
break analyses be done that include sensitivity studies allowing
considerations form only a part of the decision-making processes
for the variations noted above.
that most mining companies are currently implementing. There
The consequence category may, however, sometimes be
are increasingly onerous compliance issues, whether these be chosen without a dam break analysis if it is obvious which
regulatory (e.g. imposed by a sovereign government) or sector led, category the facility falls under. A dam break analysis should,
such as the GISTM, which originated from actions initiated by the however, be carried out as a basis for emergency preparedness and
Church of England pension fund. The intention of this short response planning (EPRP).
section is not to go into any detail about such governance issues, Different countries have different systems of consequence
but primarily to highlight the framework that increasingly governs classification, and we do not explore this issue here. It is also a
tailings management. rapidly evolving endeavour, with changes to regulations occurring
on an ongoing basis.
10.1 Consequence classification
Approaches to safely managing TSFs, from both regulators and 11 CLOSURE AND RESIDUAL LIABILITIES.
interested sector groups (such as the GISTM), are increasingly Closure of TSFs is probably the most challenging and most
based on the concept of a Consequence Category difficult aspect of overall mine site closure. It propels tailings
classification. The intention of this approach is to understand the management into a design category similar to that of management
potential damage that a TSF failure would inflict on its of nuclear waste, as the time perspective becomes ‘in perpetuity’.
environment regardless of the likelihood of such failure. This GISTM (2020) reinforces the concept of “designing for closure”,
approach has been used in some countries for many years and is which influences the project from inception (e.g. site and
often used to differentiate between facilities posing high risk from technology selection) as well as influencing design alternatives
those posing less risk to population, environment, society, considered for the operational phase, based on likely impacts on a
infrastructure, economies etc. successful closure strategy.
Recognising the extremely long design life of a closed TSF
A consequence classification is undertaken in order to: (particularly when compared with typical civil engineering
Inform selection of design loading criteria infrastructure) a number of challenges face designers of these
Inform dam safety management operations and programs facilities, as well as owners, including:
Meet regulatory and stakeholder requirements
Loading cases
The classification is based on the consequences of a possible Land use after mining
TSF failure and the associated runout, should a flow-type failure Responsibility for land after mining
occur. The likelihood of a TSF failure is not considered, but the Approval of closure design
failure mode needs to be physically possible, referred to as a
‘credible failure mode’. It is important to understand that the TSF The current strong focus on mine closure is relatively new (the
consequence classification category is not a measure of risk first International Mine Closure Seminar was held in Perth in
(which considers likelihood). On the contrary, the dam 2006). Few TSFs have been closed and relinquished, i.e.
classification category is used to support design and operating including transfer of responsibility from the mining company to
decisions intended to reduce the likelihood of failure and hence the state. Nevertheless, planning for closure is today, in most
reduce risk. A dam owner can, if so inclined, base the design and countries, a regulatory obligation and a closure plan is a
surveillance program on a higher consequence category, which in requirement for getting an operating permit. Knowledge and
some cases may be relevant based on a cost-benefit evaluation. competence, as well as requirements and expectations, are
Not always, but often, the potential runout from a TSF failure constantly increasing, placing more and more attention and focus
is determined using a dam break analysis which in turn aids with on closure as well as changing the perspective from merely
consequence classification. Until recently such analyses were, at closing a facility to designing for closure.
best, rudimentary. However, recent efforts by various research Planning for closure up front includes input from a wide range
groups and consulting engineering companies have seen of specialists in addition to the tailings and/or mine engineers. It
significant improvements in our ability to conduct meaningful should include specialists in surface and groundwater,

174
hydrologists, geochemists, biologists, social planners, community isotropic, homogeneous hydraulic conductivity values. In reality,
relations planners, landscape architects, and others who may hydraulic conductivity may be anisotropic (usually more
provide valuable input to the closure plan. Additional to including permeable horizontally than vertically), may decrease with depth,
a broader range of specialists, the closure process will invariably and may decrease with proximity to the decant pond. All, or some
benefit from being integrated with the design and operation of these factors could render an oversimplified seepage model
processes. useless. Calibration of a seepage model, using monitored and
We flag the issue of closure in this paper, because we strongly correctly interpreted piezometer data should be the starting point
believe that sustainable closure of a TSF will require for correctly determining phreatic conditions. Anything else may
implementation of sound geotechnical engineering principles. provide a false sense of security.
Topics discussed in this paper, such as correct laboratory and field Whilst many countries in South America, as well as some
testing, methods of analysis and appropriate application of other parts of the world have faced the challenges of designing
empirical approaches apply to ensuring the stability of closed and TSFs to withstand very high seismic loading conditions, for other
relinquished TSFs as well as to operating facilities. Given current
parts of the world (e.g. Australia and South Africa) it remains
climate variations, TSFs are likely to be subjected to extreme
largely an unlikely, but possible event. The paper therefore walks
weather events beyond what occurs during the operational life of
a line between a fully mechanistic, laboratory-informed approach
a facility. Geotechnical robustness is essential to ensure future
generations are not faced with problems of repairing and to seismic design and analysis, and one that reflects current
remediating inherently inappropriate geotechnical structures. practice in seismically active countries, which includes the
widespread use of empirical approaches, such as charts relating
12 DISCUSSION penetrometer resistance to liquefaction susceptibility. Another
aspect that is flagged is the continuing use of pseudo-static
At present, the focus of many, perhaps most tailings engineers is coefficients in seismic analyses, at least for screening-level
on how to best deal with upstream TSFs containing brittle tailings, evaluations. Retaining such empirical approaches respects the
and rightly so. The spate of recent failures, particularly in Brazil philosophy espoused by Burland (1989) when discussing his
in 2015 and 2019 has highlighted how sudden these failures can concept of the ‘soil mechanics triangle’, at the centre of which is
occur and how wide-reaching the impact can be. Before delving what he refers to as ‘well winnowed experience’.
into a deeper discussion about this specific topic, we reflect on a Operating a TSF in an extremely cold climate, such as that in
few other issues we believe should not be lost sight of when northern Sweden and northern Canada present special challenges,
designing a new TSF, or indeed evaluating the stability of an including embedment of frozen layers during winter operations.
existing TSF. These topics include volume change of soft tailings, While still frozen, these layers become effectively impermeable,
seepage considerations, stability in highly seismic regions, and potentially resulting in preferential horizontal flow of seepage
impacts of climatic conditions on TSF stability. water, with attendant stability implications. If these frozen layers
Many TSFs are now more than 1km2 in area, sometimes many subsequently thaw, they may result in increased pore water
square kilometres. A large percentage of such TSFs is therefore pressures and thus a reduction in shear strength. At the opposite
essentially subject to one-dimensional consolidation, with end of the climatic spectrum, many TSFs are in arid or semi-arid
principal stress directions remaining constant throughout the locations, where evaporation dominates over precipitation. This
growth of such TSFs. Tailings derived from the processing of water imbalance can contribute to strength gain through the
sulphide ore (i.e. hard rock) and composed of silty and/or sandy process of desiccation noted above. However, crust formation due
particles will usually settle rapidly upon deposition and to desiccation, which is exacerbated by use of highly saline
conventional methods of consolidation analysis are adequate. groundwater, results in restricted drying of tailings below the crust
However, for a large range of mine tailings, including mineral layer. This has negative implications on TSF stability as gains in
sands slimes, fine iron ore tailings, oilsands fine tailings (or shear strength that would otherwise occur are inhibited.
mature fine tailings, MFT), and many phosphate tailings, Let us now return to the key issue addressed in this paper, how
deposition takes place at extremely high void ratios and the to address the design of upstream TSFs and how to evaluate the
tailings are highly compressible. Properties such as hydraulic stability of existing TSFs containing potentially liquefiable
conductivity vary over orders of magnitude with increasing tailings.
effective stress, and conventional approaches to calculating The greatest of these two challenges is evaluating the stability
volume change are deficient. The paper demonstrates laboratory of existing TSFs. There is often a paucity of relevant information
testing techniques to characterise these properties and numerical about the operational history of many TSFs, with issues such as
modelling procedures to predict likely field behaviour. Such poor decant pond control, decant pipe breakages, or slope
evaluations are relevant not only to volume calculations but also instabilities often undocumented. Obtaining a reliable evaluation
to stability assessments. There are instances where the volume of existing conditions is thus difficult. We contend that in-situ
decrease renders the tailings dilative (located below the critical testing techniques are the preferred first option, as extracting
state line) and thus inherently less susceptible to liquefaction. As samples for laboratory testing may destroy the very property of
embodied in critical state soil mechanics, volume change and interest, i.e. brittleness. Of the available in-situ testing methods,
shear strength gain cannot be separated. the preferred option currently is the CPTu. This is probably due to
Another, often overlooked aspect of volume change is that the great deal of both empirical and theoretical information to
induced by desiccation. Solar drying, which induces desiccation, facilitate estimation of properties such as undrained strength and
can produce large volume changes, particularly in fine-grained, state. However, practitioners should be aware of the limitations of
compressible tailings, providing large gains in shear strength. these methods some of which we covered under Section 5.1.
Through manipulation of deposition layer thickness desiccation Furthermore, the vast majority of studies using calibration
can be enhanced, providing both improved shear strength and chamber tests to correlate CPTu readings with known, controlled
increased storage of tailings (as the achieved in-situ dry density is state conditions were carried out on clean, quarzitic sands. Many
higher). tailings of interest when it comes to potential brittleness are
Seepage is an issue we suggest is often inadequately predominantly silts or silty sands. Recently, research has begun to
addressed. Given the critical importance of the shape and location focus on this category of tailings using calibration chamber tests,
of the phreatic surface to the stability of upstream TSFs, it is e.g. Ayala et al (2020). Existing limitations in the interpretation of
surprising that many seepage models are based on assumptions of in-situ tests highlight the need of proactively managing a TSF to

175
prevent it from reaching a state where it is susceptible to failure, where they identify the primary trigger for liquefaction as
catastrophic failure. Measurable indicators that can inform this being over-pressurisation of a vertical borehole during
type of proactive management have been described by Boswell installation. This particular trigger mechanism has not previously
and Sobkowicz (2018). been documented, to our knowledge, but required detailed
Evaluations of the potential for static liquefaction at an laboratory testing and numerical modelling.
existing TSF are currently predominantly dependent on CPTu Quite rightly, the need for better monitoring techniques of
testing, backed up with interpretations utilising fundamental TSFs to provide sufficient early warning of developing, or
mechanics, such as critical state constitutive models in numerical impending failure likelihood is currently attracting a lot of
simulations of the cone penetration process. Evaluations of attention and research effort. Unfortunately, there has also been a
potential seismic instability are different. Although evaluation of great deal of shameless opportunism, with some companies
in-situ state is carried out, as with static liquefaction assessments, making outrageous and unproven claims of their proprietary
there is much greater reliance on empirical correlations, an technology’s ability to provide the required early warning. We
example being Figure 81 in this paper. These empirical counsel scepticism when considering implementation of unproven
correlations are understandable, as there have been many technologies to such important structures as TSFs containing
earthquakes in which liquefaction of sands has occurred, as well brittle tailings. We recognise that some of the emerging
as many where liquefaction has not occurred, with cone technologies may ultimately provide the desired accuracy,
penetration data also being available. Drawing correlations using relevance and reliability, but this is not currently the case for many
such data pairs is thus an obvious strategy. However, once again, technologies. Traditional instrumentation, such as vibrating wire
the majority of these correlations are based on earthquake-induced piezometers for pore pressure measurement, and inclinometers or
liquefaction of relatively clean sands. Extrapolation to silty shape acceleration arrays for displacement measurements, should
materials usually requires accounting for any relevant differences be retained (where appropriate). Coupling these conventional
between silts and sands. monitoring approaches with some of the emerging solutions is a
Laboratory testing is of course the only way to characterise sensible approach.
tailings behaviour during the design of a new facility, but can also Although this State of the Art paper is lengthy, there are many
be useful for evaluations of existing facilities, as long as it is clear topics we could still have covered. These include more detailed
what can reasonably be measured. Disturbed samples are widely discussions on high-density thickened tailings and filtered
used to develop the critical state line, but may not provide tailings, as these tailings management strategies are generating
representative information about brittleness (particularly in terms increased interest as owners tackle issues such as risk of failure
of both strength loss and the strain over which strength loss and competing demands for available water. Other emerging
occurs). Undisturbed samples are necessary to determine these solutions include the use of blended, or co-mingled waste
characteristics, and these are notoriously difficult to obtain in silts (combining waste rock with tailings). We suggest that most, if not
and silty sands. all of the approaches to characterisation of tailings discussed in
In characterising brittle (i.e. strain-softening) soils, focus is on this paper are also applicable to thickened tailings, filtered
both the peak strength and the residual, large strain strength, with tailings, or co-mingled waste. An aspect not discussed in our
many designers now regarding the latter as the governing stability paper which will certainly find increasing application is
assessment criterion. This posed a question for the authors of this characterisation of unsaturated properties, particularly for projects
paper, ‘are we (our industry) prepared to do whatever it takes to using filtered tailings for example.
ensure stability’? We have also not specifically addressed the topic of
Many currently operating TSFs, if evaluated using the transitional materials, where the term is used to define soils
residual (or liquefied) shear strength (Sr), would be found to have (tailings) that are characterised by incomplete convergence to a
a factor of safety against slope instability of less than unity. We unique normal compression line and/or critical state line in simple
say this based on our personal experience with reviews of many laboratory tests (Todisco and Coop, 2019). Although Li and Coop
existing facilities. What course of action should be taken in such (2019) found no evidence of transitional behaviour in their tests
cases? Many of the larger mining companies have chosen as first on iron ore tailings from the Panzhihua TSF in China, it is a
resort the construction of a stabilising buttress on the downstream consideration we believe warrants further investigation for a range
slope of the existing TSF, with some going even further and of tailings. Given the current reliance on evaluations of the critical
constructing large earthen ‘deflector’ walls, to cater for a potential state line for tailings, any dependence of the CSL on initial void
flowslide failure. However, is this necessarily an appropriate ratio must be identified and quantified.
blanket solution (buttress construction), or should we adopt a Finally, we note the extreme importance of disseminating
more nuanced approach for materials that require very large knowledge to those operating TSFs in resource-poor countries,
strains to fully mobilise Sr (e.g. material B in Figure 74)? Clearly where we use ‘resource’ in this context to denote access to
an answer to this question depends on factors that include cost, information, availability of suitable laboratories and skilled,
potential consequences of a failure, the risk appetite of different competent contractors (e.g. for conducing CPTu tests). The
companies and whether a credible failure mode for a particular performance record of TSFs leaves no doubt that significant
TSF truly exists. failures can occur in well-established mining jurisdictions in
Although case studies of the various TSF failures that have which access to competent engineering services is available (e.g.
been well-documented provide invaluable lessons to all tailings Australia, Brazil and Canada). Clearly the occurrence of high-
engineers, relying on a reactive approach such as this is not profile accidents in resource-poor countries is also a distinct
sufficient. There is a clear need for trained geotechnical engineers possibility. Preventing a TSF failure in any country should be a
who understand the subtleties of mine tailings, and have a sound concern of the entire mining industry. As evidenced by the
fundamental understanding of geotechnical engineering international reaction to the Feijão failure, everyone becomes
principles. Ideally, this would equip the next generation of tailings tarred with the same brush when such a failure occurs.
engineers with the ability to identify possible failure mechanisms,
or triggers for a failure, based on an understanding of geotechnical 13 CONCLUDING COMMENTS
engineering rather than just by comparison against a check-list of
previously documented failures. An excellent example of the In the discussion section preceding these concluding comments
former is the report by CIMNE into the Brumadinho (Feijão) TSF we reviewed the main points of our paper and discussed these in
a little more detail. Here we offer three overriding concluding

176
comments. They are based on the numerous discussions held used empirical correlations are based, a cautionary approach is
between the five co-authors over a period of many months. appropriate. We suggest that empirical correlations are a useful
We believe that, in general, the key underlying principles that starting point for tailings characterisation, but should be
govern the behaviour of mine tailings (geotechnically speaking; confirmed with a comprehensive suite of testing, including both
geochemical considerations are beyond our scope for this paper) laboratory and in-situ tests wherever possible.
are well understood. The concerns generated by failures such as
Feijão, where spontaneous, or static liquefaction occurred with 14 ACKNOWLEDGEMENTS
such sudden and devastating consequences, caused consternation
for many engineers, who perhaps wondered if there was an The authors would like to thank Mr Edi Nuryatno for assistance
underlying principle that had until now remained unrecognised. with drafting many of the figures and for assistance with final
The relatively poor appreciation, at least until recently, of static formatting of the paper, Mr Christopher O’Donovan for assistance
liquefaction may be due to the training at most universities of the with obtaining copyright clearances and for authors who kindly
principles of soil mechanics, which often endorse the assumption granted us copyright clearance to reproduce figures of theirs,
that ‘loading is drained in sands and undrained in clays’. With being Antonio Gens, Ross Boulanger, Gonzalo Castro and
many tailings being sand-like in that they consist of nonplastic Michael Etezad. The authors also gratefully acknowledge many
silty mixtures, it may have been tempting to assume that they of our students and colleagues who have contributed through
always exhibit drained behaviour. However, if we pause to numerous discussions over the years to the development of ideas
consider the definition of drained loading, which is loading that is presented in the paper.
applied at a rate slow enough that excess pore pressures dissipate
quicker than they are generated, it becomes clear that sands can 15 REFERENCES
fail in an undrained fashion, which is the essence of static
liquefaction. A key to understanding static liquefaction is that Abu-Hejleh, A.N. and Znidarcic, D. 1994. Estimation of the Consolidation
shearing of contractive soils induces positive excess pore water Constitutive Relations, Computer Methods and Advances in
pressures. Once such shear-induced pore pressures are triggered, Geomechanics, Siriwardane & Zaman (eds) Balkema, Rotterdam,
they may propagate faster than even the most permeable sand is 499-504.
capable of dissipating. Abu-Hejleh, A. N. and Znidarcic, D. 1995. Desiccation Theory for Soft
Our second point relates to the first. We contend that the Cohesive Soils, Journal of Geotechnical Engineering, ASCE, New-
training of many tailings engineers is inadequate. Not recognising York, Vol. 121(6), 493-502.
the coupling between volume change and shear strength is a Abu-Hejleh, A. N., Znidarcic, D., and Barnes, B.L. 1996. Consolidation
deficiency of many undergraduate programmes. Although it is Characteristics of Phosphatic Clays, Journal of Geotechnical
certainly not faultless, critical state soil mechanics provides an Engineering, ASCE, New-York, 122(4), 295-301.
elegant connection between these two aspects of soil behaviour. Agüero, G. 1929. Formation of Tailings Deposits in El Teniente Mine.
Inclusion of critical state soil mechanics, non-linear finite strain Anales del Instituto de Ingenieros de Chile. 164-187. (in Spanish).
consolidation theory, and a better appreciation of the strengths and Altuhafi, F., O’Sullivan, C. and Cavarretta, I. 2013. Analysis of an image-
limitations of widely used laboratory and in-situ tests would based method to quantify the size and shape of sand particles. J.
further aid in improving the quality of tailings engineering Geotech. Geoenviron. Eng., 139(8), 1290-1307.
practice. Likewise, seepage analyses for stability calculations Atkinson, J.H., Lau, W.H.W., and Powell, J.J.M. 1991. Measurement of
require far greater attention to the details of anisotropy, stress and soil strength in simple shear tests. Canadian Geotechnical Journal,
saturation dependent hydraulic conductivity in both formal 28, 255-262.
education and engineering applications. While routine seepage Ayala, J., Fourie, A.B. and Reid, D. 2020. Cone penetration testing on silty
models might be adequate for water balance calculations, often tailings using a new small calibration chamber. Geotechnique Letters,
they fail to identify critical issues that affect stability of TSFs. 10, 492-497.
Needless to say that if it were not for water in tailings, we would Azam, S. and Li, Q. 2010. Tailings dam failures: a review of the last one
not have stability problems. Given the rapid uptake of certain hundred years. Geotechnical News, December 2010, 50-53.
numerical modelling packages, many of which include highly Basson, J.A., Broekman A., and Jacobsz S.W. 2021. TD-DAQ: A low-
complex constitutive laws, we are concerned that a lack of cost data acquisition system monitoring the unsaturated pore pressure
understanding of fundamental soil mechanics principles will regime in tailings dams. HardwareX.
result in misleading, potentially dangerous interpretations of the https://doi.org/10.1016/j.ohx.2021.e00221
results generated by such packages. Bauer, E. 1996. Calibration of a Comprehensive Hypoplastic Model for
Our third, and final point is that there is still a great deal of Granular Materials. Soils and Foundations. 36(1): 13-26.
work to be done, techniques and theories to be developed and Beckett, C., Fourie, A.B. and O’Loughlin, C.D. 2016. Centrifuge
refined. Use of in-situ tests, particularly the cone penetrometer, to modelling of seepage through tailings embankments. International
tailings applications has greatly advanced the understanding of Journal of Physical Modelling in Geotechnics, 16(1), 18-30.
such tests, which has a benefit to all geotechnical engineering Bedell, D., Fawell, P., Slottee, S. and Schoenbrunn, F. 2015. Thickening.
practitioners. Issues of partial drainage during penetration, and the In: Paste and Thickened Tailings – A Guide, Third Edition. Ed. R.J.
challenge of interpreting cone test results when carried out in Jewell and A.B. Fourie, Australian Centre for Geomechanics.
partially saturated media are everyday experiences when working Bedin, J., Schnaid, F., da Fonseca, A.V., and de Costa Filho, L.M. (2012).
on TSFs. Similarly, use of conventional consolidation theory may Gold tailings liquefaction under critical state soil mechanics.
often be inappropriate and misleading, particularly when working Géotechnique, 62(3): 263–267. doi:10.1680/geot.10.P.037.
with soft, highly compressible mine tailings. Volume change Been, K. 2016. Characterising mine tailings for geotechnical design.
calculations, and thus estimates of storage capacity of a TSF, may Proceedings Geotechnical and Geophysical Site Characterisation 5,
be grossly in error if the consolidation behaviour of such tailings Sydney, Australia, Australian Geomechanics Society.
is not correctly characterised. We strongly caution against the Been, K., Jefferies, M.G. and Hachey, J.E. 1991. The critical state of
extrapolation of information on material behaviour from one site sands. Géotechnique, 41(3), 365–381.
to another without an evaluation of the validity of such Been, K. and Jefferies, M.G. 1985. A state parameter for sands.
extrapolation, preferably based on appropriate laboratory and/or Géotechnique, 35(2), 99–112.
in-situ testing. As discussed in our paper, tailings particle shape Been, K., Jefferies, M.G., Crooks, J.H.A. and Rothenberg, L. 1987. The
can be highly angular. Coupled with the fact that many tailings cone penetration test in sands: Part II, general inference of state.
have a wide distribution of particle sizes, unlike the clean, Géotechnique 37(3), 285-299.
relatively well-rounded sands on which many of our currently

177
Been, K., and Olivera, R. 2015. Appendix B: Laboratory testing to Carrera, A., Coop, M., and Lancellotta, R. 2011. Influence of grading on
determine the critical state of sands. In Soil Liquefaction: A critical the mechanical behaviour of Stava tailings. Géotechnique, 61(11):
state approach. CRC Press. 935–946.
Beier, N. and Sego, D. 2008. Cyclic freeze-thaw to enhance the stability Casagrande, A. 1936. Characteristics of Cohesionless Soils Affecting the
of coal tailings. Cold Regions Science and Technology. Elsevier 2008- Stability of Slopes and Earth Fills. Journal of the Boston Society of
08-16. Civil Engineering, January, pp. 13-32.
Beveridge, A., Mutz, P., and Reid, D. 2015. Tailings co-disposal case Casagrande, A. 1970. On Liquefaction Phenomena," Lecture, reported by
study – art or science? In Proceedings of the 18th International Green and Ferguson, Geotechnique, Vol. 21, No.3, (1971), pp. 197-
Seminar on Paste and Thickened Tailings, Cairns, Australia, 5–7 May 202.
2015. Edited by R.J. Jewell and A.B. Fourie. Australian Centre for Casagrande, A. 1975. Liquefaction and Cyclic Deformation of Sands - A
Geomechanics, Perth, Australia, 505–520. critical review," Fifth Panamerican Conference on Soil Mechanics
Bhanbro, R. 2017. Mechanical behaviour of tailings – Laboratory Tests and Foundation Engineering, Buenos Aires, Argentina.
from a Swedish Tailings dam. PhD thesis, Luleå University of Castro, G. 1969. Liquefaction of sands, PhD thesis, Harvard University,
Technology. USA.
Biarez, J., and Hicher, P.Y. 1994. Elementary mechanics of soil Castro, G. and Troncoso, J. 1989. Effects of Chilean Earthquake on three
behaviour: saturated remoulded soils. A. A. Balkema. tailing dams”. International Seminar on Dynamic Behavior of Clays,
Bicalho, K.V. 1999. Modeling Water Flow in an Unsaturated Compacted Sands & Gravel. Kitakyushu, Japan.
Soil, Ph.D. Thesis, University of Colorado Boulder. Chakraborty et al 2020. In Situ Investigation for Geotechnical
Bicalho, K.V., Znidarcic, D. and Ko, H.Y. 2011. One-dimensional flow Characterization of Consolidated Dry Mine Tailing Slurry Backfill.
infiltration through a compacted fine grained soil, Soils and In Geotechnical Characterization and Modelling.
Foundations, 51(2), 287-295. Chandler, R.J. 1988. The in-situ measurement of the undrained shear
Bishop, A.W. 1967. Progressive failure -- with special reference to the strength of clays using the field vane. Vane shear testing in soils: field
mechanism causing it. Proc. Geotech. Con, Oslo, 2: 142-150. and laboratory studies, ASTM STP 1014, A.F. Richards, Ed., ASTM,
Bishop, A. W., Green, G. E., Garga, V. K., Andresen, A. & Brown, J. D. Philadelphia, 1988, 13-44.
1971. A new ring shear apparatus and its application to the Chen, Z., Guo, X., Shao, L., Shunqun, L. and Tian, X. 2021. Design of a
measurement of residual strength. Geotechnique 21(4), 273-328. three-dimensional earth pressure device and its application in a
Bjerrum, L. 1969. Effect of rate of strain on undrained shear strength of tailings dam construction simulation experiment. Acta Geotechnica,
soft clays. Second contribution to panel discussion, Main Session 5, 16, 2203-2216.
7th International Conf. on Soil Mechanics and Found. Eng. Mexico. Cho, G-C, Dodds, J. and Santamarina, J.C. 2006. Particle shape effects on
Blight, G.E. 1962. A note on field testing of silty soils. Canadian packing density, stiffness and strength: natural and crushed sands. J.
Geotechnical Journal, 3, 142-149. of Geotechnical and Geoenv. Eng, 132(5), 591-602.
Blight, G.E., Boswell, J.E.S. and Zenon, A. 1995. Underwater Chow S., Alonso-Marroquin F. and Airey D. 2012. Viscous Rate Effects
construction of an embankment to extend the life of a tailings in Shear Strength of Clay. Proc. 11th Australia New Zealand
impoundment. Proc. Instn. Civ. Engrs. Geotech. Engng. 113, 80-85. Conference on Geomechanics.
Blight, G.E. 1997. The master profile for hydraulic fill tailings beaches. CIMNE 2021. Computational analyses of Dam I failure at the Corrego de
Proc. Instn. Civ. Engrs, Geotechnical Engineering, 107, 27-40. Feijao mine in Brumadinho, Final Report, CIMNE, August 2021.
Boswell J. and Sobkowicz J. 2018. Leading versus lagging indicators of Clayton CRI and Bica AVD 1993. The design of diaphragm-type
tailings dam integrity. In Proceedings of the 22nd International boundary total stress cells. Géotechnique, 43(4), 523-535.
Conference on Tailings and Mine Waste. Tailings and Mine Waste Collins, B.D., 1997. Landslide Triggering Mechanism, M.S. Thesis,
'18. Keystone, Colorado, USA, September 30–October 2, 2018. University of Colorado Boulder
Boulanger, R. W. 2002. Evaluating Liquefaction Resistance at High Collins, B.D., and Znidarcic, D., 2004, Stability Analyses of Rainfall
Overburden Stresses. Proc. 3rd U.S.-Japan Workshop on Advances Induced Landslides, Journal of Geotechnical and Geoenvironmental
Research on Earthquake Engineering for Dams, San Diego, CA., 1- Engineering, ASCE, New-York, 130(4), 362-372.
12. Conlin, B. 1987. A review of the performance of mine tailings
Boulanger, R. and Idriss, I. 2006. Liquefaction Susceptibility Criteria for impoundments under earthquake loading conditions. Proceedings
Silts and Clays. J. of Geotechnical and Geoenvironmental Eng., Vancouver Geotechnical Society Seminar on Earthquake. Geot.
132(11). Canada.
Boulanger, R. and Idriss, I. 2014. CPT and SPT based liquefaction Cubrinovski, M., and Ishihara, K. 2002. Maximum and minimum void
triggering procedures. Report No. UCD/CGM-14/01. University of ratio characteristics of sands. Soils and Foundations 42(6):65-78.
California at Davis. Cunning, J.C. 1994. Shear wave velocity measurements of cohesionless
Bray, J., Sancio, R., Durgunoglu, T., Onalp, A., Youd, L. Stewart, J., Seed, soils for evaluation of in situ state. MSc thesis. University of Alberta.
R., Cetin, O., Bol, E., Batuary, M., Christensen, C. and Karadayilar, Edmonton, Canada.
T. 2004. Subsurface characterization at ground failure sites in da Fonseca A.V., Cordeiro D., and Molina-Gómez F. 2021.
Adapazari, Turkey. Journal of Geotechnical and Geoenvironmental Recommended procedures to assess critical state locus from triaxial
Engineering. 130(7), 673–685. tests in cohesionless remoulded samples. Geotechnics. 1: 95-127.
Bray, J.D., and R.B. Sancio. 2006. Assessment of the liquefaction Daliri, F., Kim, H., Simms, P., and Sivathayalan, S., 2014. Impact of
susceptibility of fine grained soils. Journal of Geotechnical and Desiccation on Monotonic and Cyclic Shear Strength of Thickened
Geoenvironmental Engineering, 132(9), 1165–1177. Gold Tailings. Journal of Geotechnical and Geoenvironmental
Bray J, and Travasarou T. 2009. Pseudostatic coefficient for use in Engineering, 140(9).
simplified seismic slope stability evaluation. Journal of Geotechnical Debasis R., Campanella R.G., Byrne P.M., and Hughes J. 2002. Undrained
and Geoenvironmental Engineering, 135(9). anisotropic monotonic behavior of sand from in situ tests. J. Geotech.
Bromwell, L.G., and Oxford, T.P. 1977. Waste clay dewatering and Geoenviron. Eng. ASCE.
disposal. In: Proc. ASCE Specialty Conf. on Geotechnical Practice DeJong, J.T. and Randolph, M. 2012. Influence of partial consolidation
for Disposal of Solid Waste Materials, Ann Arbor, MI during cone penetration on estimated soil behavior type and pore
Budhu, M. 1984. Nonuniformities imposed by simple-shear apparatus. pressure dissipation measurements. J. Geotech. Geoenviron. Eng.,
Geotechnique, 20, 125-137. 138(7): 777-788.
Burland, J. B. 1989. Nash lecture: The teaching of soil mechanics – A Desrues J, Chambon R, Mokni M, and Mazerolle F. 1996. Void ratio
personal view. Proceedings of the ninth European Conference on Soil evolution inside shear bands in triaxial sand specimens studied by
Mechanics and Foundation Engineering, Dublin, 1987, Vol.3, pp computed tomography. Géotechnique 46(3):529-546.
1427-1447.

178
Díaz-Rodríguez, J., Martínez-Vasquez J. and Santamarina J. 2009. Strain- Methods, Tailings and Mine Waste ‘20, November 15-18, UBC
Rate Effects in Mexico City Soil. Journal of Geotechnical and Studios, University of British Columbia, Vancouver BC, 413-424.
Geoenvironmental Engineering, 135(2). Gjerapic, G., Znidarcic, D., Peric, D., and Murphy, F.D., 2021,
Dobry, R. and Alvarez, L. 1967. Seismic Failures of Chilean Tailings Development of an engineering model for consolidation and creep in
Dams. Journal of Soil Mechanics and Foundation Division, ASCE, tailings, Tailings and Mine Waste ’21, November 7-10, Banff, AB,
93(6), 237-260. Canada, ISBN: 978-1- 55195-466-0, 841-851.
Dounias, G.T. and Potts, D.M. 1993. Numerical analysis of drained direct Gomes, R.C., Ribeiro, L.F.M., Albuquerque Filho, L.H. and Rezende,
and simple shear tests. Journal of Geotechnical Engineering, 119(12), C.H. 2003. Geotechnical Aspects of Eroded Sediments Relevant to
1870-1891. Tailings Disposal Design, Natural Hazards Review, 4(2), 65-70.
Dyvik, R., Berre, T., Lacasse, S. and Raadim, B. 1987. Comparison of Graham, J. 1992. The key material properties in geotechnical engineering
truly undrained and constant volume direct simple shear tests. need to be measured instead of being specified. In Predictive Soil
Geotechnique, 37(1), 3-10. Mechanics.
Farrar, J.A., Torres, R., and Crutchfiedl, L.G. 2008. Cone penetrometer Graham, J. and Jefferies, M. 1986. Some examples of in situ lateral stress
testing, Scoggins Dam, Tualatin Project, Oregon. Engineering determination in hydraulic fills using the self-boring pressuremeter.
Geology Group Bureau of Reclamation, Technical Services Center, Hazen, A. 1920. Hydraulic-fill dams, ASCE Transactions, 83, 1713–1745.
Denver, Colo. Report No. 86–68320. Hoffman, A. 2015. Mount Polley Mine, tailings storage facility breach.
Fear, C.E., and Robertson, P.K. 1995. Estimating the undrained strength Investigation Report of the Chief Inspector of Mines. Ministry of
of sand: a theoretical framework. Canadian Geotechnical Journal, Energy and Mines.
32(5), 859-870. Holden, D., Donegan, S. and Pon, A. 2020. Brumadinho Dam InSAR
Finnie, I., and Randolph, M.F. 1994. Punch-through and liquefaction study: analysis of TerraSAR-X, COSMO-SkyMed and Sentinel-1
induced failure of shallow foundations on calcareous sediments. Proc. images preceding the collapse. Slope Stability 2020, Australian
Int. Conf. on Behavior of Offshore Structures, Boston (USA): 217- Centre for Geomechanics, Perth, Australia.
230. Hosono, Y. and Yoshimine, M. 2008. Effects of anisotropic consolidation
Fonseca do Carmo, Fl., Kamino, L. H. Y., Tobias Junior, R., de Campos, and initial shear load on liquefaction resistance of sand in simple shear
I., Fonseca do Carmo, Fe., Silvino, G., Xavier de Castro, K. J. d. S., condition. Geotechnical Engineering for Disaster Mitigation and
Mauro, M. L., Rodrigues, N. U. A., Miranda, M. P. d. S., Pinto, C. E. Rehabilitation. Liu, Deng and Chu (eds).
F. (2017). Fundão tailings dam failures: the environment tragedy of Houman, J., Paterson, A. and van Sittert, F. 2007. Quantifying the Effect
the largest technological disaster of Brazilian mining in global of Pump Shear on Slurry Rheology — A Case Study at the Combined
context. Perspectives in Ecology and Conservation, 15, 145-151. Treatment Plant. Proc. 10th International Seminar on Paste &
Fourie, A.B., and Papageorgiou, G. 2001. Defining an appropriate steady Thickened Tailings, March 2007, Perth, Australia. Ed. R. Jewell and
state line for Merriespruit gold tailings. Canadian Geotechnical A.B Fourie, Australian Centre for Geomechanics.
Journal, 38(4):695–706. Housner, G.W. 1984. Keynote lecture, 8th World Conf. Earthq. Eng., San
Fourie, A.B., Blight, G.E. and Papageorgiou, G. 2001. Static liquefaction Francisco, Post Conference. Volume 1, pp. 25–39.
as a possible explanation for the Merriespruit tailings dam failure. Hwang, C., 1999, Transient Seepage in Embankment Dams, M.S. Thesis,
Canadian Geotechnical Journal. 38(4), 707-719. University of Colorado Boulder.
Fourie, A.B. and Tshabalala, L. 2005. Initiation of static liquefaction and Hwang, C., 2002, Determination of Material Functions for Unsaturated
the role of K0 consolidation. Canadian Geotechnical Journal, 42(3), Flow, Ph.D. Thesis, University of Colorado Boulder.
pp 892-906. Hynes, M. E., Olsen, R. S. and Yule, D. E. 1998. The Influence of
Fourie, A., Reid, D., Ayala, J., Russell, A., Vo,T., Rahman, M. and Vinod, Confining Stress on Liquefaction Resistance. Proc. of the 30th Join
J. 2021. Improvements in estimating strengths of loose tailings: Meeting of the U.S.-Japan Cooperative Program in Natural
results from the TAILLIQ research project. Proc. Mine Waste and Resources Panel on Wind and Seismic Effects, NIST SP 931, 168-
Tailings, AusIMM, Brisbane, Australia, 207-217. 184.
Franks, D.M., Stringer, M., Torres-Cruz, L.A., Baker, E., Valenta, R., Hynes, M. E. and Olsen, R. S. 1999. Influence of Confining Stress on
Thygesen, K., Matthews, A., Howchin, J. and Barrie, S., 2021. Liquefaction Resistance. Proc. International Workshop on Physics
Tailings facility disclosures reveal stability risks. Scientific reports, and Mechanics of Soil Liquefaction, Rotterdam, The Netherlands,
11(1), 1-7. 145-152.
Fujiyasu, Y.1997). Evaporation behaviour of tailings. PhD thesis, Idriss and Boulanger, 2003. Estimating K for use in evaluating cyclic
University of Western Australia, Perth, Australia. resistance of sloping ground. In: Hamada, O’Rourke, Bardet (Eds.),
Gawu, S.K.Y. and Fourie, A.B. 2004. Assessment of the modified slump Proc. of the 8th US-Japan Workshop on Earthquake Resistant Design
test as a measure of the yield stress of high-density thickened tailings. of Lifeline Facilities and Countermeasures against Liquefaction,
Canadian Geotechnical Journal, 41(1), 39-47. Report MCEER-03-0003, MCEER, SUNY Buffalo, NY, pp. 449–
Gens, A. 2019. Hydraulic fills with special focus on liquefaction. Keynote 468.
Lecture. Proceedings of the XVII ECSMGE, Reykjavik, 1-31. Idriss I. and Boulanger R. 2004. Semi-empirical Procedures for Evaluating
Ghafghazi and Shuttle 2008. Evaluation of soil state from SBP and CPT: Liquefaction Potential During Earthquakes. Proc. 3rd International
A case history. Canadian Geotechnical Journal, 45(6). Conference on Earthquake Geotechnical Engineering. Berkeley,
Gibson, R.E., England, G.L., and Hussey, M.J.L. 1967. The theory of one- USA.
dimensional consolidation of saturated clays. 1. Finite Non-linear Imai, G., 1979, Development of a new consolidation test procedure using
Consolidation of Thin Homogeneous Layers. Géotechnique, 17, 261- seepage force, Soils and Foundations, Tokyo, Japan, 19(3), 45-60.
273. Ishihara, K. 1984. Post-Earthquake Failure of a Tailings Dam Due to
GISTM Global industry standard on tailings management. 2020. ICMM Liquefaction of the Pond Deposit. Int. Conference on Case Histories
(International Council on Mining & Metals), UNEP (United Nations in Geotechnical Engineering. St. Louis. USA.
Environment Programme) and PRI (Principles for Responsible Ishihara, K. 1993. 33rd Rankine Lecture: Liquefaction and flow failure
Investment). during earthquakes. Geotéchnique, Vol. 43, No. 3, pp. 351—415.
Gjerapic, G., Johnson, J., Coffin, J., and Znidarcic, D., 2008, Ishihara, K. 1996. Soil Behaviour in Earthquake Geotechnics. Clarendon
Determination of Tailings Impoundment Capacity via Finite-Strain Press, Oxford. ISBN 0-19-856224-1.
Consolidation Models, Proceedings of the Geo Institute Congress, Ishihara, K. 1985. Stability of natural deposits during earthquakes. Theme
New Orleans, 2008. 9-12 March 2008, New Orleans, Louisiana, CD- Lecture, Proceeding of the XI ICSMFE, Vol.2, 321-376.
ROM Ishihara K., Ueno K., Yamada S., Yasuda S. and Yoneoka, T. 2014.
Gjerapic, G., and Znidarcic, D., 2020, Assessment of Liquefaction Breach of a Tailing Dam in the 2011 Earthquake in Japan. 34th United
Triggering for Upstream Tailings Dams Using Limiting Equilibrium

179
States Society on Dams Annual Meeting & Conference Seismic Li, X. S. and Wang, Y. 1998. Linear representation of steady-state line for
Design of Tailings Dams. San Francisco, USA. sand. J. Geotech. Geoenviron. Engng., 124(12), 1215–1217.
Islam, K. and Murakami, S., 2021. Global-scale impact analysis of mine Lipinski and Wdowska 2018. Hybrid approach for evaluation of tailings
tailings dam failures: 1915–2020. Global Environmental Change, 70, state on the basis of shear wave velocity measurement. In Proceedings
102361. of SGEM Conference 2018.
Jacobsz SW 2019. TUKS tensiometer measures to -1.7 MPa. Civil Liu, J.C., 1990, Determination of Soft Soil Characteristics, Ph.D. Thesis,
Engineering Magazine – SAICE 27(1). University of Colorado Boulder.
James M., Aubertin M., Wijewickreme D. and Wilson G. 2011. A Liu, J.C., and Znidarcic, D., 1991. Modeling One-Dimensional
laboratory investigation of the dynamic properties of tailings. Can. Compression Characteristics of Soils. Journal of Geotechnical
Geotech. J. 48: 1587–1600. Engineering, ASCE, 117, No. 1, pp. 164-171.
Jamiolkowski and Masella 2015. "Geotechnical Characterization of Liu, Q.B. and Lehane, B.M. 2012. The influence of particle shape on the
Copper Tailings at Zelazny Most Site." Third International (centrifuge) cone penetration test (CPT) end resistance in uniformly
Conference on the Flat Dilatometer. graded granular soils. Geotechnique, 62(11), 973-984.
Jeeravipoolvarn, S., Chalaturnyk, R.J. and Scott, J.D. 2008. Consolidation Lunne, T., 2007. Can we measure remoulded strength or sensitivity with
modeling of oil sands fine tailings: history matching. Proc. CPTU sleeve friction and friction ratio? Norwegian Geotechnical
GeoEdmonton ’08, Edmonton, Canada, 190 - 197. Institute. Report No.: 20061023-1.
Jeeravipoolvarn, S., Scott, J.D., and Chalaturnyk, R.J. 2009. Geotechnical Lunne T, Robertson PK, and Powell JJM 1997. Cone penetration testing
characteristics of laboratory in-line thickened oil sands tailings. In in geotechnical practice. CRC Press.
Proceedings of Tailings and Mine Waste 2009, Banff, Alta., 1–4 Lupini, J. F., Skinner, A. E. and Vaughan, P. R. 1981. The drained residual
November 2009. Information Technology, Creative Media: strength of cohesive soils. 31, No. 2, 181-213.
Vancouver, B.C., 813–828. Lyu, Z., Chai, J., Xu, Z., Qin, Y. and Cao, J. 2019. A comprehensive
Jefferies M., and Been K. 2015. Soil Liquefaction: A critical state review on reasons for tailings dam failures based on case history.
approach. Second Edition. CRC Press. Advances in Civil Engineering, 2019, 1-18.
Jefferies, M., Morgenstern, N., Van Zyl, D. and Wates, J. 2019. Report on Mahmoodzadeh, H. and Randolph, M.F. 2014. Penetrometer testing:
NTSF Embankment Failure Cadia Valley Operations for Ashurst effect of partial consolidation on subsequent dissipation response. J.
Australia. Geotech. Geoenviron. Eng., 2014, 140(6): 04014022.
Jewell, R.J. and Fourie, A.B., Paste and Thickened Tailings – A Guide, Mair R.J. and Wood D.M. 1987. Pressuremeter testing: methods and
2nd edition, Australian Centre for Geomechanics, Perth, 2006. interpretation. CIRIA.
Knutsson, R., 2018. On the Behaviour of Tailings Dams – Management in Marchetti S. 2015. Some 2015 updates to the TC16 DMT report 2001.
Cold Regions. Doctoral Thesis, Luleå University of Technology, Third International Conference on the Flat Dilatometer.
Sweden. Maureira, S. and Verdugo, R. 2012. Cyclic response of tailings sands
Kokusho, T. 1980. Cyclic triaxial test of dynamic soil properties for wide tested at high pressure. VII Chilean Geotechnical Congress,
strain range. Soils and Foundations, 20, 45-60. Concepcion (in Spanish).
Kramer, C., and Arango, I. 1998. Aging effects on the liquefaction Miller, G.A., Tan, N.K., Collins, R.W. and Muraleetharan, K.K.2018).
resistance of sand deposits: A review and update. Proc., Eleventh Cone penetration testing in unsaturated soils. Transportation
European Conf. on Earthquake Engineering, Paris. Geotechnics, 17, 85 – 99.
Kynch, E.J., 1952, A Theory of Sedimentation, Transactions of the Mitchell, J. K., and Solymar, Z. V. 1984. Time-dependent strength gain in
Faraday Society, 48, pp. 168-176. freshly deposited or densified sand. Journal of Geotech. Eng., 11011,
Ladd, R. 1978. Preparing test specimens using undercompaction. 1559–1576.
Geotechnical Testing Journal, 1(1), 16–23. Mohajeri, M. and Ghafghazi, M. 2012. Ground sampling and laboratory
Ladd, C.C. 1991. Stability evaluation during staged construction. Jnl. testing on low plasticity clays. Proc. 15th World Conference on
Geotechnical Engineering, 117(4), 540 – 615. Earthquake Engineering, Lisbon, Portugal.
Lade, P. 2005. Overview of constitutive models for soils. ASCE Montgomery, J., Boulanger, R. and Harder, I. 2012. Examination of the
Geotechnical Special Publication No. 128. Center for
Lee, J., 2011, Limits to Continuity of Water Flow in Unsaturated Soils, Geotechnical Modeling, University of California, Davis, California.
Ph.D. Thesis, University of Colorado Boulder. Report No.UCD/CGM-12/02.
Leon, E., Gassman, S. and Talwani, P. 2006. Accounting for Soil Aging Monte, J.L., and Krizek, R.J., 1976, One-dimensional mathematical model
When Assessing Liquefaction Potential. Journal of Geotechnical and for large-strain consolidation, Geotechnique, 26(4), 495-510.
Geoenvironmental Engineering, 132(3). Morgenstern N.R., Vick S.G., Van Zyl D. 2015. Report on Mount Polley
Leroueil, S., and Hight, D.W. 2003. Behaviour and properties of natural Tailings Storage Facility Breach. Province of British Columbia.
soils and soft rocks. In Proc. of the 2nd International Workshop on Morgenstern, N.R. 2018. Geotechnical risk, regulation and public policy.
Characterisation and Engineering Properties of Natural Soils, Soils and Rocks, 41(2), 107 – 129.
Singapore, 2–4 December 2002. Ed. by K.K. Phoon, D.W. Hight, S. Morgenstern, N., Vick, S., Viotti, C. and Watts, B. 2016. Fundão Tailings
Heroueil, and T.S. Tan. A.A. Balkema, Rotterdam, the Netherlands. Dam Review Panel, Report on the Immediate Causes of the Failure of
Vol. 1, 29–254 the Fundão Dam.
Leroueil, S. and Vaughan, P.R. 1990. The general and congruent effects Morgenstern, N. and Price, V. 1965. The Analysis of the Stability of
of structure in natural soils and weak rocks. Geotechnique, 40(3), 467- General Slip Surfaces. Geotechnique, 15(1),
488. Mundle, C., Chapman, P., Williams, D., and Fredlund, D. 2012. The
Li, W. 2017. The mechanical behaviour of tailings [online]. Ph.D. thesis, impact of high density liquors on standard soil mechanics
Department of Architecture and Civil Engineering, City University calculations. Proc. 16th International Conference on Tailings and
of Hong Kong. Available from Mine Waste. Information Technology, Creative Media, Keystone,
https://scholars.cityu.edu.hk/en/theses/the-mechanicalbehaviour-of- Colorado. 281–290.
tailings(d3d2338a-6cf2-4e19-993b-a3f67dad274a).html. Murphy, F.D., Gjerapic, G., Znidarcic, D., and Sorta, A., 2021, Validation
Li, W., and Coop, M.R. 2019. The mechanical behaviour of Panzhihua of an engineering model for consolidation and creep in oil sand
iron tailings. Canadian Geotechnical Journal, 56(3): 420–435. tailings, Tailings and Mine Waste ‘21, November 7-10, Banff, AB,
Li, W., Coop, M.R., Senetakis, K., and Schnaid, F. 2018. The mechanics Canada, 833-840.
of a silt-sized gold tailing. Engineering Geology, 241: 97–108. Murthy, T.G., Loukidis, D., Carraro, J.A.H., Prezzi, M., and Salgado, R.
Li, G., Liu, Y.-J., Dano, C., and Hicher, P.-Y. 2015. Grading-dependent 2007. Undrained monotonic response of clean and silty sands.
behavior of granular materials: from discrete to continuous modeling. Géotechnique, 57(3): 273–288.
Journal of Engineering Mechanics, 141(6): 04014172.

180
Nagaraj, T.S., Pandian, N.S. and Narashima, R. 1998. Compressibility Quiros, G.W. and Young, A.G., 1988. Comparison of field vane, CPT, and
behaviour of soft cemented soils. Geotechnique, 48(2), 281-287. laboratory strength data at Santa Barbara Channel site. ASTM
Nakase, A. and Kamei, T. 1986. Influence of strain rate on undrained shear International.
characteristics of Ko-consolidated cohesive soils. Soils and Randolph, M. F., and Andersen, K. H. 2006. Numerical analysis of T-bar
Foundations, 26, 85-95. penetration in soft clay. Int. J. Geomech., 6(6), 411–420.61
National Academies of Sciences, Engineering, and Medicine. 2016. State Rees SD. 2010. Effects of fines on the undrained behaviour of
of the Art and Practice in the Assessment of Earthquake-Induced Soil Christchurch sandy soils. PhD Thesis. Department of Civil and
Liquefaction and Its Consequences. Washington, DC: The National Natural Resources Engineering. University of Canterbury. New
Academies Press. doi: 1017226/23474. Zealand.
Newson, T.A. and Fahey, M. 1998. Saline tailings disposal and Reid, D., 2015. Estimating slope of critical state line from cone penetration
decommissioning. MERIWA Project No. M241, Australian Centre for test—an update. Canadian Geotechnical Journal, 52(1), 46-57.
Geomechanics. Reid, D. and Fanni, R., 2022. A comparison of intact and reconstituted
Newson, T., Dyer, T., Adam, C. and Sharp, S. 2006. Effect of structure on samples of a silt tailings. Géotechnique, 72(2), 176-188.
the geotechnical properties of bauxite residue. Journal of Reid, D. and Fourie, A.B. 2017. Laboratory assessment of the effects of
Geotechnical and Geoenvironmentral Engineering, 132(2), 143-151. polymer treatment on geotechnical properties of low-plasticity soil
Nguyen, H.B.K., Rahman, M.M. and Fourie, A.B.2018). Characteristic slurry. Canadian Geotechnical Journal, 53: 1718–1730.
behavior of drained and undrained triaxial compression tests: DEM Reid, D. and Fourie, A.B. 2019. A direct simple shear device for static
study. Journal of Geotechnical and Geoenvironmentral Engineering, liquefaction triggering under constant shear drained loading.
144 (9). Geotechnique Letters, 9(2), 142-146.
Ni, Q., Tan, T., Dasari, G. and Hight, D. 2004. Contribution of fines to the Reid, D., Fourie, A., Ayala, J.L., Dickinson, S., Ochoa-Cornejo, F., Fanni,
compressive strength of mixed soils. Géotechnique, 54 (9) (2004), pp. R., Garfias, J., Da Fonseca, A.V., Ghafghazi, M., Ovalle, C., Riemer,
561-569. M., Rismanchian, A., Olivera, R., and Suazo, G. 2021. Results of a
Oliveira Filho, W.L., 1998, Verification of a Desiccation Theory for Soft critical state line testing round robin programme. Géotechnique,
Soils, Ph.D. Thesis, University of Colorado Boulder. 71(7), 616-630.
Olson SM. (2009). Strength ratio approach for liquefaction analysis of Reid, D., Fourie, A., Castro, J. and Lupo, J. 2018a. Undrained shear
tailings dams. Proceedings of the 57th Annual Geotechnical strength evolution with loading on an undisturbed block sample of
Engineering Conference. Minneapolis, Minnesota, USA. desiccated gold tailings. Proceedings of Tailings and Mine Waste,
Olson SM, and Stark TD. 2002. Liquefied strength ratio from liquefaction 2018.
flow failure case histories. Canadian Geotechnical Journal Reid, D., Fourie, A., and Russell, A. (2018b). Effects of desiccation on
39(3):629-647. shear strength of tailings – comparison of clayey and sandy tailings.
Olson SM, and Stark TD. 2003. Yield strength ratio and liquefaction Proceedings of Tailings and Mine Waste, 2018.
analysis of slopes and embankments. Journal of Geotechnical and Riemer M., Moriwaki, Y. and Obermeyer, J. 2008. Effect of High
Geoenvironmental Engineering, 129(8):727-737. Confining Stresses on Static and Cyclic Strengths of Mine Tailing
Pane, V., 1985, Sedimentation and Consolidation of Clays, Ph.D. Thesis, Materials. Proc. Geotechnical Earthquake Engineering and Soil
University of Colorado Boulder. Dynamics IV. GSP 181 ASCE.
Pane, V., and Schiffman, R.L., 1997, The permeability of clay Robertson P.K. 1991. Soil classification using the cone penetration test:
suspensions, Géotechnique, Vol. 47, 273-288. Reply. Canadian Geotechnical Journal 28(1):176-178.
Papadimitriou A., Bouckovalas G., Andrianopoulos K. 2014. Robertson P.K 2009. Interpretation of cone penetration tests – a unified
Methodology for estimating seismic coefficients for performance- approach. Canadian Geotechnical Journal 46:1337-1355.
based design of earth dams and tall embankments. Soil Dynamics and Robertson P.K. 2010. Evaluation of Flow Liquefaction and Liquefied
Earthquake Engineering. 56, 57–73. Strength Using the Cone Penetration Test. Journal of Geotechnical &
Papadopoulou, A., and Tika, T. 2008. The effect of fines on critical state Geoenvironmental Engineering 136(6):842-853.
and liquefaction resistance characteristics of non-plastic silty sands. Robertson, P.K. 2012. Evaluation of Flow Liquefaction using the CPT: an
Soils and Foundations, 48(5): 713–725. update. Keynote presentation at Tailings and Mine Waste Conference.
Park, S, Nong, Z. and Lee, D. 2020. Effect of vertical effective and initial Colorado.
static shear stresses on the liquefaction resistance of sands in cyclic Robertson P.K. 2016. Cone penetration test (CPT)-based soil behaviour
direct simple shear tests. Soils and Foundations. 60(6), 1588-1607. type (SBT) classification system – an update. Canadian Geotechnical
Peck, R.B. 1969. Advantages and limitations of the observational method Journal, 53(12):1910-1927.
in applied soil mechanics. Geotechnique, 19, 171-187. Robertson, P.K. 2017. Evaluation of flow liquefaction: influence of high
Plewes, H.D., Davies, M.P. and Jefferies, M.G. 1992. CPT based stresses. Proc. 3rd International Conference on Performance-based
screening procedure for evaluating liquefaction susceptibility. In: Design in Earthquake Geotechnical Engineering, Vancouver,
Proceedings of the 45th Canadian Geotechnical Conference. Canada, Canada.
Toronto. Robertson, P.K., Viana de Fonseca, A., Ulrich, B. and Coffin, J. (2017).
Polito, C. and Martin, J. 2001. Effects of Nonplastic Fines on the Characterisation of unsaturated mine waste: a case history. Can.
Liquefaction Resistance of Sands. Journal of Geotechnical and Geotech. J., 54, 1752-1761.
Geoenvironmental Engineering. 127(5). Robertson P.K., and Wride C. 1998. Evaluating cyclic liquefaction
Poulos, S.J. 1971. The stress-strain curves of soils. Geotechnical potential using the cone penetration test. Canadian Geotechnical
Engineers Inc., Winchester, Mass. USA. Journal, 35(3):442-459.
Poulos, S.J. 1981. The steady state of deformation. Journal of the Robertson, P.K., De Melo, L., Williams, D. and Wilson, G. 2019. Report
Geotechnical Engineering Division. 107(5). of the Expert Panel on the Technical Causes of the Failure of Feijão
Pournaghiazar, M, Russell, AR & Khalili, N. 2011, Development of a new Dam I (Brumadinho).
calibration chamber for conducting cone penetration tests in Roscoe, K. 1953. An apparatus for the application of simple shear to soil
unsaturated soils, Canadian Geotechnical Journal, 48(2), 314–321. samples. Proc. 3rd Int. Conf. Soil Mech. 1, No. 2, 186-191.
Prevost, J. and Popescu, R. 1996. Constitutive Relations for Soil Roscoe, K. H., Schofield, A N. and Wroth, C. P. (1958). On Yielding of
Materials. The Electronic Journal of Geotechnical Engineering. Soils. Geotéchnique, 8(1), 22-53.
Inaugural Volume. Rourke, H. and Luppnow, D. 2015. The Risks of Excess Water on Tailings
Proskin, S., Sego, D. and Alostaz, M., 2010. Freeze-thaw and Facilities and Its Application to Dam-break Studies. Tailings and
consolidation tests on Suncor mature fine tailings (MTF). Cold Mine Waste Management for the 21st Century. Sydney, Australia.
Regions Science and Technology, 63(3), 110-120.

181
Sadrekarimi, A. 2016. Static Liquefaction Analysis Considering Principal Skempton, A. 1985. Residual strength of clays in landslides, folder strata
Stress Directions and Anisotropy. Geotechnical and Geological and laboratory. Geotechnique 35, No. 1, 3-18.
Engineering. 34(4). Sladen, J.A. and Handford, G., 1987. A potential systematic error in
Sadrekarimi A. and Olson, S. 2011. Yield strength ratios, critical strength laboratory testing of very loose sands. Canadian Geotechnical
ratios, and brittleness of sandy soils from laboratory tests. Canadian Journal, 24(3), pp.462-466.
Geotechnical Journal, 48(3):493-510. Soderberg, R.L. and Busch, R. A. 1977. Design guide for metal and
Sano, T. 1906. Report on earthquake damage in California, USA. Journal nonmetal tailing disposal. Department of the Interior, Bureau of
of Architecture and Building Science. 238, 646–656 (in Japanese). Mines, USA.
Santamarina, J.C., Torres-Cruz, L.A., and Bachus, R.C. 2019. Why coal Somogyi, F., Keshian, B., and Bromwell, L.G., 1981, Consolidation
ash and tailings dam disasters occur. Science, 364(6440): 526–528. Behavior of Impounded Slurries, Presented at ASCE Spring
Santamarina, J. C., and Cho, G. C. 2004. Soil behaviour: The role of Convention, New York, New York.
particle shape. Advances in geotechnical engineering: The Skempton Stewart, D. P., and Randolph, M. F. 1994. T-bar penetration testing in soft
Conference, R. J. Jardine, D. M. Potts, and K. G. Higgins, eds., Vol. clay. J. Geotech. Engrg., 120(12), 2230–2235.
1, Thomas Telford, London, 604–617. Suits, L.D., Sheahan, T.C., Sadrekarimi A. and Olson, S. 2009. A new ring
Saragoni, R. 1993. Análisis del Riesgo Sísmico para la Reconstrucción del shear device to measure the large displacement shearing behavior of
Puerto de Valparaíso. Sextas Jornadas Chilenas de Sismología e sands. Geotechnical Testing Journal, 32(3).
Ingeniería Sísmica, Santiago (In Spanish). Teh, C.I. and Houlsby, G.T. 1991. An analytical study of the cone
Sarantonis, E., Etezad, M. and Ghafghazi, M. 2020. The effect of assumed penetration test in clay. Geotechnique, 41(1), 17-34.
residual strength on remediation cost of a typical tailings dam. Thevanayagam, S., Shenthan, T., Mohan, S., and Liang, J. 2002.
Proceedings Tailings and Mine Waste 2020, 671-678. Undrained fragility of clean sands, silty sands, and sandy silts.
Schmertmann, J. H. 1991. The mechanical aging of soils. Journal of Journal of Geotechnical and Geoenvironmental Engineering,
Geotechnical Engineering., 179, 1288–1330. 128(10): 849–859.
Schnaid, F. 2021. On the geomechanics and geocharacterisation of Todisco, M.C. and Coop, M.R. (2019). Quantifying “transitional” soil
tailings. Proceedings of 6th International Conference on behaviour. Soils and Foundations, 59, 2070-2082.
Geotechnical and Geophysical Site Investigation, 25pp. Torres Cruz LA 2016. Use of the cone penetration test to assess the
Schnaid, F., Bedin, J. Viana da Fonseca, A. J. P. and Costa Filho, L. M. liquefaction potential of tailings storage facilities. PhD Thesis.
2013. Stiffness and Strength Governing the Static Liquefaction of University of the Witwatersrand. Johannesburg, South Africa.
Tailings. J. Geotech. Geoenvironmental. Engng., 139 (12): 1943- Torres-Cruz, L.A. 2019. Limit void ratios and steady-state line of non-
5606. plastic soils. Proceedings of the Institution of Civil Engineers -
Schneider, J.A., Randolph, M.F., Mayne, P.W. & N.R. Ramsey 2008. Geotechnical Engineering, 172(3): 283–295.
Analysis of factors influencing soil classification using normalized Torres-Cruz, L.A., 2021. The Plewes Method: a Word of Caution.
piezocone tip resistance and pore pressure parameters. J. Geotech. Mining, Metallurgy & Exploration, 38(3), pp.1329-1338.
Geoenviron. Eng. 134(11): 1569–1586. Torres-Cruz LA and Santamarina (2020). The critical state line of
Schneider, J.A., Hotstream, J.N., Mayne, P.W. & Randolph, M.F. 2012. nonplastic tailings. Canadian Geotechnical Journal, 57(10).
Comparing CPTU Q-F and Q- 2 v0 soil classification charts. Torres-Cruz, L.A. and Vermeulen, N., 2018. CPTu-based soil behaviour
Géotechnique Letters 2:209–215. type of low plasticity silts. In Cone Penetration Testing 2018, 635-
Schneider, J.A. and Moss, R.E.S. 2011. Linking cyclic stress and cyclic 641. CRC Press.
strain based methods for assessment of cyclic liquefaction triggering Troncoso J.H., Verdugo R. 1985. Silt Content and Dynamic Behavior of
in sands. Geotechnique Letters, 1, 31-36. Tailings Sands. Proc. XI International Conference on Soil Mechanics
Schofield, A.N. and Wroth C.P. 1968. Critical State Soil Mechanics. and Foundation Engineering. Vol. 3. 1311-1314.
McGraw-Hill, London. Troncoso J.H, Ishihara K and Verdugo R. 1988. Aging Effects on Cyclic
Seed, H. 1983. Earthquake-resistant design of earth dams. Proc. Symp. Shear Strength of Tailings Materials. Proc. IX World Conference on
Seismic Des. of Earth Dams and Caverns, ASCE, New York, 41–64. Earthquake Engineering, Vol. 3, 121–126. Tokyo, Japan.
Seed, H. and Idriss, I. 1982. Ground motions and soil liquefaction during Towhata, I. 2008. Geotechnical Earthquake Engineering. Springer-Verlag
earthquakes. EERI. Monograph, University of California, Berkeley. Berlin Heidelberg. ISBN 978-3-540-35782-7
Seed, R. and Harder, L. 1990. SPT-Based Analysis of Cyclic Pore Vaid, Y.P., and Chern, J.C. 1983. Effect of static shear on resistance to
Pressure Generation and Undrained Residual Strength. In: Duncan, liquefaction. Soils and Foundations, 23(1): 47–60.
J.M., Ed., Proc. H.B. Seed Memorial Symposium, Vol. 2, BiTech Vaid, Y., Stedman, J. and Sivathayalan, S. 2001. Confining stress and
Publishers, Richmond, 351-376. static shear effects in cyclic Liquefaction. Can. Geotech. J. 38: 580–
Sheahan, T., Ladd, C. and Germaine, J. 1996. Rate-dependent undrained 591.
shear behavior of saturated clay. Journal of Geotechnical Valenzuela, L. (2015). Tailings Dams and Hydraulic Fills. The 2015
Engineering, 122(2), 99-108. Casagrande Lecture. 16th International Conference on Soil
Shuttle, D.A. and Cunning, J., 2007. Liquefaction potential of silts from Mechanics and Geotechnical Engineering. Buenos Aires, Argentina.
CPTu. Canadian Geotechnical Journal, 44(1), pp.1-19. Valenzuela L., Verdugo R., Campaña J. and Opazo C. 2020. Safety
Shuttle DA, and Cunning J. 2008. Reply to the discussion by Robertson performance of dams in Chile highly seismic environment. Proc. 4th
on "Liquefaction potential of silts by CPTu." Canadian Geotechnical International Dam World Conference, Lisbon, Portugal.
Journal 45:142-145. Verdugo R. 1992. Characterization of sandy soil behavior under large
Shuttle, D. and Jefferies, M. 1998. Dimensionless and unbiased CPT deformation. PhD Thesis. University of Tokyo, Japan.
interpretation in sand. Int. Jnl. Num. and Anal. Methods in Geomech., Verdugo, R. 2009. Seismic performance based-design of large earth and
22, 351-391. tailings dams. Proc. International Conference on Performance-Based
Shuttle, D. and Jefferies, M., 2016. Determining silt state from CPTu. Design in Earthquake Geotechnical Eng. – from case history to
Geotechnical Research, 3(3), 90-118. practice. Tsukuba, Japan.
Shuttle D., Martens S., and Jefferies M. (2022). Geostatic stress in oil-sand Verdugo, R. 2011. Seismic stability analysis of large tailings dams. Proc.
tailings. Proceedings of the Institution of Civil Engineers – Theme Lectures of 5th International Conference on Earthquake
Geotechnical Engineering. https://doi.org/10.1680/jgeen.21.00114. Geotechnical Engineering. Santiago., Chile.
Silvis, F. and de Groot, M.B. 1995. Flow slides in the Netherlands: Verdugo R, and Ishihara K. 1996. The steady state of sandy soils. Soils
experience and engineering practice. Can. Geotech. J., 32, 1086- and Foundations 36(2):81-91.
1092. Verdugo, R. and Santos, E. 2009. Liquefaction resistance of thickened
Skempton AW 1964. Long-term stability of clay slopes. Geotechnique, tailings of copper mines. Proc. 17th International Conference on Soil
14(2), pp.77-102. Mechanics and Geotechnical Engineering. Alexandria, Egypt.

182
Verdugo, R. and Viertel, P. 2004. Effect of density and fines content on
the cyclic strength of copper tailings. V Chilean Geotechnical
Congress.
Vietti, A. and Coghill, M. 2015. Slurry Chemistry. In: Paste and
Thickened Tailings – A Guide, Third Edition. Ed. R.J. Jewell and A.B.
Fourie, Australian Centre for Geomechanics.
Villavicencio G., Espinace R., Palma P., Fourie A. Valenzuela P. 2014.
Failures of sand tailings dams in a highly seismic country. Canadian
Geotechnical Journal. 51: 449–464.
Wanatowski, D. and Chu, J. 2007. Static liquefaction of sand in plane
strain. Can. Geotech. J., 44, 299-313.
Wanatowski, D. and Chu, J. 2011. Pre-failure instability behaviour of sand
in strain path testing under plane-strain conditions. Soils and
Foundations, 51(3), 423-435.
Wang, Z. 2010. Seismic Hazard Assessment: Issues and Alternatives.
Pure and Applied Geophysics. Published on line. DOI
10.1007/s00024-010-0148-3
Weber, J.P., 2015. Engineering evaluation of post-liquefaction strength.
PhD thesis, University of California, Berkeley.
Wells, P.S., Charlebois, L., Diep, J., Moyls, B., Omotoso, O., Revington,
A. and Weiss, M. 2015. Inline Flocculation. In: Paste and Thickened
Tailings – A Guide, Third Edition. Ed. R.J. Jewell and A.B. Fourie,
Australian Centre for Geomechanics.
Wride C.E., Robertson P.K., Biggar K.W., Campanella R.G., Hofmann
B.A., Hughes J.M.O., Kupper A., and Woeller D.J. 200. Interpretation
of in situ test results from the CANLEX sites. Canadian Geotechnical
Journal, 37(3).
Yang, S.L., Sandven, R., and Grande, L. 2006. Steady-state lines of sand–
silt mixtures. Canadian Geotechnical Journal, 43(11): 1213–1219.
Yang, Y., Wei, Z., Fourie, A.B., Chen, Y., Zheng, B., Wang, W. and
Zhuang, S. 2019. Particle shape analysis of tailings using digital
image processing. Environmental Science and Pollution Research,
26, 26397-26403.
Yao, D.T.C., 1998, Desiccation Process for Soft Soils, Ph.D. Thesis,
University of Colorado Boulder.
Yao, D.T.C , Oliveira-Filho, W.L., Cai, X.C. and Znidarcic, D., 2002.
Numerical solution for consolidation and desiccation of soft soils,
International Journal for Numerical and Analytical Methods in
Geomechanics, 26, (2), 139-161.
You, Z., 1993, Flow Channeling in Soft Clays and its Influence on
Consolidation, Ph.D. Thesis, University of Colorado Boulder.
Yu, H.S. 1998. A unified state parameter model for clay and sand.
International Journal for Numerical and Analytical Methods in
Geomechanics, 22, 621-653.
Zergoun, M., and Vaid, Y. 1994. Effective stress response of clay to
undrained cyclic loading. Can. Geotech. J., 31, 714–727.
Zhao J. and Guo N. 2013. Unique critical state characteristics in granular
media considering fabric anisotropy. Géotechnique, 63(8), 695-704.
Zhou, H., and Randolph, M.F. 2009. Resistance of full-flow penetrometers
in rate-dependent and strain-softening clay. Geotechnique, 59(2), 79-
86.
Zlatovic´, S., and Ishihara, K. 1995. On the influence of nonplastic fines
on residual strength. In Proceedings of the 1st International
Conference on Earthquake Geotechnical Engineering, Tokyo, Japan.
AA Balkema, Rotterdam, the Netherlands. 239–244.
Znidarcic, D., 2015, If it creeps, does it matter? Tailings and Mine Waste
‘15 conference proceedings, October 25-28, Vancouver, BC, Canada,
pp.229-235.
Znidarcic, D., Adkins, D., Utting, L. and Catling, M., 2015, Rheomax®
ETD technology, a laboratory study of application performance and
associated geotechnical characteristics for polymer assisted tailings
deposition of oil sands MFT, Tailings and Mine Waste ‘15, October
25-28, Vancouver, BC, Canada, 508-520

183

View publication stats

You might also like