Professional Documents
Culture Documents
net/publication/237553405
Article
CITATIONS READS
17 2,690
3 authors, including:
All content following this page was uploaded by Ed Mcroberts on 29 December 2016.
Abstract
Within the entire range of failure modes that have occurred at tailings impound-
ments static liquefaction is likely the most common, and at the same time likely the least
understood. As design practice in many mining regions has in fact discounted the possi-
bility of the mechanisms and criteria for this failure mode, the possibility of its occurrence
has often been overlooked in the search for other causes of failure. Static liquefaction,
and the resulting flowslide of liquefied tailings materials, is shown to be a relatively com-
mon phenomenon among the more dramatic tailings impoundment failure case histories.
Static liquefaction can be a result of slope instability issues alone, or can be triggered as a
result of other mechanisms.
The fundamentals of the liquefaction phenomena are summarized. Liquefaction is
a term most often associated with seismic events. However, mine tailings impoundments
have demonstrated more static liquefaction events than seismic induced events. The
summary of the fundamentals includes particular emphasis on static liquefaction.
Several static liquefaction case histories are described to demonstrate various ways
in which this failure mechanism has manifested itself. From an understanding of the fun-
damentals and the lessons learned from the case histories, basic guidelines to minimize
the concern for tailings impoundments are presented.
Introduction
Classical soil mechanics as found in many textbooks still being used today, pres-
ents a simplistic and erroneous view of the loading of saturated cohesionless granular
particles (usually lumped together as “sands”) and water systems – that is for example
most tailings. The simplistic view is that by defining the friction angle and pore pressure of
the sand we can predict the strength of that sand, the drained strength. The exception
these references allow for sands is during an earthquake when the sand may become ‘liq-
uefied’. Clays on the other-hand are deemed to be cohesive and have an undrained
strength. Those readers who have benefited from a more enlightened geotechnical edu-
cation may not find this a credible proposition, but it is clear to the authors that even as we
st
enter the 21 century, a range of educators, regulatory and quasi-governmental groups,
and an alarming number of geotechnical consultants still have not un-learned their first se-
ries of lectures in soil mechanics based on textbooks expounding the views noted above.
Until these simplistic models have been un-learned by all involved with the design, licens-
ing, and construction of tailings impoundments, a major contributor to failures,
i.e. inappropriate and incorrect designs based upon a lack of understanding of the tailings
strength, will continue.
There is a wide range of specialized literature on the subject of the strength of co-
hesionless soils and their interactions with shear-induced pore pressures. However, little
of this is to be found in a few textbooks, it is mostly in technical journals and specialized
publications. Recent useful discussions can be found, for example, in Martin and
McRoberts (1999), Carrier (1991), and Been (1999). These are written from the perspec-
tive of geotechnical engineers with a thorough understanding of tailings materials and also
provide a starting point for the newcomer to the field of the considerable, and often mis-
leading, literature that exists.
The most fundamental of the “new” lessons on cohesionless soil (sand and most
silts) strength is that like a clay, rapidly loaded saturated sand can have an undrained
strength, and like clay this strength can be stress and strain path dependent. Loose
sands/silts such as those deposited in an underwater tailings beach can have a very low
strength; they contract during shear just like a sensitive clay. However, unlike clays that
have a unique void ratio compression state, sand has wide ranges in its void ratio com-
pression state. The wide ranges in the initial void ratio of sands, and of the fabric of field-
scale deposits of these sands, means that predictions of the in-situ undrained strength for
these materials is highly uncertain.
The undrained strength of sand becomes a fundamental issue whenever there is
rapid loading that triggers significant shear-induced pore pressure rise. Rapid loading is
subjectively defined as a rate of increase in shear stress and resulting pore pressures that
cannot drain or dissipate sufficient rapidly such that these higher pore pressures [and not
the pre-triggering event pore pressures measured in a piezometer] define the sand
strength. The most readily identified of these rapid loading conditions, at least from a de-
sign perspective, is the transient loading from seismic events. Whether limited-
deformation or eventual flowslide development, the effects of transient seismic loads on
mine tailings are well documented in the literature and well recognized by current engi-
neering standards.
However, there are many other rapid or undrained shear loads that affect mine
tailings. These potential triggers of undrained response can be of equal importance to
seismic loads due to their more common occurrence at mine sites in comparison to seis-
mic events. Included in these common loads are incremental impoundment raise con-
struction and episodic tailings slurry placement. The former can lead to relatively rapid in-
creases in stress levels and undrained conditions in susceptible materials while the latter
can cause temporary changes to the amount of tailings saturated in a given section of an
impoundment. Conversely, traditional static loads are taken to be those in place for a
considerable period. Other mechanisms, such as a transient saturation of the down-
stream shell of a tailings structure, can also trigger liquefaction due to rapid reductions in
effective stress.
Regardless of loading condition, the most dramatic effect a transient load can have
on mine tailings is to impart liquefaction of those tailings over a sufficient volume that then
leads to a “failure” event. “Failure” can mean different things but non-intentional release of
tailings solids or supernatant fluid(s) is the most dramatic failure mechanism and the one
most typically set as the design “upset” condition. The term transient load is chosen to
avoid the confusion between seismic and static liquefaction events because, though the
loading conditions are different, the resulting concern to the mine operator is identical.
The mechanisms at the root of either static or seismic liquefaction are the same. That
dams liquefy during the construction phase due to non-seismically induced transient loads
may belie the assurances offered in some cases as to the actual seismic stability of these
structures.
Liquefaction Fundamentals
Over the past two decades, issues related to liquefaction have become one of the
more heavily researched and published sub-disciplines of soil mechanics. Liquefaction
flow failures of mine tailings represent some of the more dramatic case history contribu-
tions to the database of actual liquefaction events that have occurred at full-scale. The
definition developed by the NRC (1985) for liquefaction and its related physical phenom-
ena is both basic and complete and is the definition used in this paper. No attempt to du-
plicate the extensive literature on liquefaction will be attempted herein. Tailings will be as-
sumed to have one of four characteristics upon shear loading:
1. Brittle strain softening (full liquefaction with the potential for limitless deforma-
tion) – contractant behavior upon shear up to the steady-state condition.
2. Limited strain softening (limited liquefaction with limited deformation) – some
initial contraction followed by dilation of the tailings skeleton;
3. Ductile behaviour with undrained shearing but no significant degree of strain-
softening (no liquefaction); and
4. Strain hardening (no appreciable liquefaction or deformation) - essentially pure
dilation.
The liquefaction equivalence noted above for each loading condition is consistent
with the NRC (1985) nomenclature.
Figure 1 presents schematic representations of strain softening response to both
monotonic and cyclic shear loading conditions. Identical ultimate responses, although with
more complex loading histories, can result from static and cyclic shear loading.
Figure 1 includes the concept of collapse surface introduced by Sladen et. al.
(1985) used here in its most general sense. Consider an element of sand at an initial
stress state τst; for loose sand the transient loading, either a rapid monotonic load of fast
construction, or the cyclic loading of a seismic event, will cause the sand to engage the
collapse surface. Once so engaged, there is a dramatic and uncontrollable loss of
strength down to sus the steady state or residual strength. The collapse surface concept
provides a useful framework that provides linkage between the seismic and static stress
paths that can trigger liquefaction, and basically says that it does not matter how you get
there, the ‘liquefied’ strength is the same.
There is debate regarding the position, linearity and even existence of a unique
collapse surface or a unique critical state line, see McRoberts and Sladen (1992) and
more recently Been (1999) for discussion. Laboratory work on stress paths recently sum-
marized by Vaid and Sivathayalan (2000) show that both stress path and fabric heavily
control the low strain undrained strength of sand. This laboratory work indicates that the
way sand grains are packed together [the fabric] and the direction of the major principal
stress during shear relative to this fabric or grain imbrication can produce design-
significant variability in strength predictions. However, following Been (1999), this does
not mean that there is not a unique critical state line. Both McRoberts and Sladen (1992)
and Been (1998) have noted that the critical state may simply not have been reached in
low strain tests, a difficulty inherent in triaxial cell testing.
Experimental difficulties with accurate definition of moisture content [or void ratio]
re-distribution in globally undrained tests is a major problem on relying upon steady state
strength determination from laboratory tests carried out under both static and cyclic load-
ing conditions. Ayoubian and Roberston (1998) report on a series of medium-loose ex-
tension tests on Ottawa sands in which undrained tests were frozen at different state of
shear and sectioned. The local void ratios were from 0.03 to 0.05 higher than the overall
or global value. This is a significant difference given that a void ratio change of 0.15 em-
braces the entire practical range of the critical state line for Ottawa sand. Desrues et. al
(1996) reports on shear band localization in triaxial compression tests for dense of critical
sands using tomography. These tests provide results that seriously question the validity of
relying on dilatant stress paths to establish the ultimate state. Drained tests in dense
sands indicated a difference in localized to global void ratio increase of about 0.10, a very
significant amount. Equivalent undrained tests are not yet reported, but it seems reason-
able to infer that moisture content redistribution such that the usual procedure in interpre-
tation of dilatant data is fundamentally flawed.
Figure 2 correlates a typical series of laboratory tests in which the void ratio of the
tailings sand at stress state τst is plotted against the corresponding steady state strength.
From a design perspective a unique state line offers the possibility of measuring the sand
void ratio and then predicting the ‘worst case’ strength and then using this to design
against this eventuality, and attempts to do this date back to at least Poulos et. al. (1985).
However, measuring the void ratio of sand to the precision required to make this a reliable
exercise is far from straight-forward as noted by McRoberts and Sladen (1992), Been and
Jefferies (1993), Jefferies and Davies (1993), and Martin and McRoberts (1998). Also
shown on Figure 2 is the definition of the state parameter (Ψ), per Been and Jefferies
(1985). The state parameter is a convenient way of expressing the soils relative stress
and density state relative to the steady state/ultimate state line. For example, the initial
state shown in Figure 2 is above the steady state line implying contractant behaviour and
a positive state (Ψ>0).
e Ultimate State in
Undrained Loading
CONTRACTANT ( ∆ e = 0)
( ψ >0)
Void Ratio
e
ψ
eus
DILATANT
(ψ <0) USL
(ψ = 0)
Figure 3 shows a schematic t (shear) vs. s’ (normal) stress plot as adapted from
Sladen et al. (1985). Three zones were indicated as A (liquefaction “impossible”),
B (liquefaction susceptibility under cyclic loading), and C (liquefaction susceptibility under
static loading).
"Drained" Failure
Line t/s'=sin φ'
t= ( σ' - σ' ) /2
Point C
Collapse Surface
B Stable States
t SS
aL A
Case Examples
A relatively minor rainstorm caused the limited freeboard to be overcome and water
leaving the impoundment caused toe erosion, which, in turn, initiated the flow failure. The
Harmony tailings were quite fine-graded with more than 60% finer than 74 µm. However,
these fines were also essentially cohesionless and once an area of the dam toe was
eroded and local slopes were increased to the range of 2H:1V, static liquefaction and the
massive flowslide was initiated soon after. Fourie et al. (2001) stated that a large portion
on the tailings had Ψ >0.1.
Much of the post-failure laboratory testing exhibited dilatant behavior, leading a
number of well-published engineers to suggest that the failure mode was uncertain. The
fact that contractant behavior could not be easily coaxed from the tailings in a laboratory
setting yielded the flawed conclusion that they must then be dilatant in both the laboratory
and field settings. This conclusion was reached even in light of in-situ cone data that
clearly indicated the potential for an in-situ contractant response to rapid transient loading.
Terzaghi noted, “nature has no contract with mathematics – she has even less of an obli-
gation to laboratory test procedures and results”. The authors have encountered too
many geotechnical projects in general, and tailings dam projects in particular, in which sci-
ence was revered as king, and his loyal subjects viewed with distrust and skepticism any-
thing that could not be repeatedly demonstrated in laboratory testing. In the giant stress-
controlled test represented by the dam itself, contractant, undrained behavior clearly re-
sulted, and Figure 5 is unambiguous in this respect. With Fourie et al. (2001) finally con-
firming that substantive portions of the tailings were highly contractant (Ψ>0.1 for exten-
sive portions of the tailings), undrained response clearly occurred and static liquefaction
was triggered resulting in the massive flow failure.
The Merriespruit case record provides a good example of field evidence being mis-
interpreted due to its apparent non-conformity with laboratory data. Blight (1997), pro-
duced figures illustrating dilatant behavior of gold tailings. He then describes field vane
tests, carried out slowly “for the shear strengths to be representative of the drained condi-
tion”. It is probably reasonable to assume that those tests were carried out under drained
to partially drained conditions. Blight then noted that the “remoulded strengths correspond
to values of φ’ as low as 6°”, and that “the undisturbed vane shear strengths are drained
and therefore should allow for the effect that dilation has of drawing water into the pores
on the failure surface to relieve the reduced pore pressure. It is thus difficult to explain
why the remoulded strengths are so much lower.” Setting aside the dilatancy suggested
by the laboratory testing, it is the opinion of the authors that the explanation is clear and
self-evident: the vane rotation induced a significant level of strain softening, which was
clearly indicative of contractant response, consistent with the nature of the failure. Piezo-
cone data was equally compelling in terms of the contractant nature of the tailings.
Figure 7 Sullivan Tailings Pond Showing Approximate Outline of 1948 Failure Scarp
The Sullivan event resulted in extensive liquefaction and a flowslide but, fortunately,
another tailings dyke contained the flow and no off-site impact was experienced. The dam
had been built on a foundation of older tailings that were placed as BBW material. The
failure of the upstream constructed facility was triggered by the initiation of shear stresses
in the foundation tailings in excess of their shear strength. As the material strained, the
pore pressures rose and drainage was impeded leading to liquefaction event. The down-
stream slopes of the dyke average roughly 3H:1V, imposing stresses in excess of the col-
lapse surface for the foundation tailings. The failure was very brittle and sand boils, water
expressed from standpipes and other “classic” liquefaction expressions were evident.
Figure 6 shows a ground view of the post-liquefaction appearance of the Sullivan tailings
dyke. The only trigger to the liquefaction failure was the slope geometry; a pre-failure
dyke slope of about 2.5 H to 3.5H:1V with a maximum dyke height of about 25 m.
Post failure investigations in the area of the liquefaction failure indicated very con-
tractant in-situ states in the range of Ψ = +0.01 to +0.12. Back calculated strengths for the
liquefied tailings yielded an average Su/p’ of about 0.08.
The Sullivan failure also served to illustrate a phenomenon termed “tailings dam
amnesia” (TDA) in previous work by the authors (e.g. Davies et al., 2000). The 1991 fail-
ure bore a striking resemblance to a failure that occurred at the facility in 1948, during up-
stream raising (Figure 7). That event, unlike its 1991 descendant, involved significant off-
site release from a flow failure, an obvious undrained event. However, the design analy-
ses prior to the 1991 failure were based on peak-drained strengths with no regard for
shear-induced pore pressures. Here the site-specific lesson from 43 years before was
forgotten – history did indeed repeat itself. As another example of TDA, the authors are
aware of a tailings impoundment in the United States where no fewer than three failures
involving static liquefaction have occurred in the same general area of the impoundment.
Stava Mine, Italy, 1985
Perhaps the most tragic tailings dam failure to date occurred on July 19, 1985. A
fluorite mine, located near Stava in northern Italy, had both of its tailings dams fail sud-
3
denly and release approximately 240,000 m of liquefied tailings. The liquefied mass
moved up to speeds of 60 km/h obliterating everything in its path for a stretch of some
4 km. The flowslide destroyed the village of Stava and also caused considerable damage
at Tesero, at the junction of Stava Creek and the Avisio River at the 4 km point from the
mine.
The tailings dams were both nearly 25 m high with one directly upstream of the
other. Figure 8 shows views of the impoundments pre and post-failure, while Figure 9
shows a schematic section of the two impoundments. The failure mechanism began with
failure of the upper dam that in turn overtopped and failed the lower dam as well. The
dams were upstream constructed with outer slopes from 1.2 to 1.5H to 1V. Based upon
the likely state of the in-situ tailings and the aggressively steep slopes, the soil mechanics
curiosity with this failure is that the dams could have attained such a height prior to failure
achieving states of in-situ stress that were likely far in excess of the collapse surface irre-
spective of stress path. Clearly drained, or at least largely drained, loading was sustained
for the life of the facility prior to the failure. When the failure did occur, it was dramatic and
spontaneous with an apparent collapse so sudden that it “sounded as though a cannon
went off”. There is no question that the design of these dams was not consistent with
even the most elementary of engineering principals available at the time. There are a
number of "rules" for upstream tailings dam engineering (Davies and Martin, 2000) that
were understood on both a theoretical and empirical basis for many years prior to the
Stava failure. The Stava dams both broke far more of these rules than they followed.
Morgenstern (2001) suggested that a rising phreatic surface within the upper dam
caused an initial failure of the outer sand shell. Morgenstern (2001) states that the failure
started off under drained conditions, but subsequent straining induced positive pore pres-
sures, causing the sand shell to flow, which then triggered liquefaction in the retained
slimes as they suddenly lost confinement. A significant aspect of this sequence is that it
illustrates that fully drained conditions can exist up until the very moment of liquefaction
triggering. Failure was initiated not because of increasing shear stresses and undrained
loading, but because of decreasing effective stress due to the rising phreatic surface un-
der drained conditions.
The operative friction angle, e.g. residual, at failure along the marl was estimated to
be between 11° and 15°. The peak strength of the marl is considerably higher, but the
material is very brittle when sheared parallel to its bedding. There were also construction-
induced pore water pressure increases in the foundation. The dam had a downstream
slope of up to 1.3H:1V and was about 30 m high when it failed. The section that failed
was the highest section of dam, the depth of tailings the greatest and the thickness of al-
luvium overlying the marl relatively thin in comparison the rest of the impoundment.
Without noting what type of investigation and design work was carried out prior to
dam construction, or during the life of the facility, it is clear that the dam geometry prior to
failure and the foundation conditions were not compatible. The initial movement of the
dam was the triggering mechanism that allowed the tailings to liquefy and this liquefaction
exerted a thrust against the dam that contributed to the relatively rapid progression and
large lateral displacement of the overall failure event.
Figure 10 Aftermath of the Los Frailes (Anzacollar) Tailings Dam Failure
This postulated failure sequence bears great similarity to that of the Fort Peck dam
in 1938 (see Figure 4). Casagrande (1965) presented information in support of the hy-
pothesis that liquefaction of the hydraulic fill was triggered by spreading along a weak
plane in the shale foundation. Smith (1969) discusses another static liquefaction tailings
dam failure in northern Spain believed triggered by foundation spreading. Case records
such as these illustrate the dangers involved when brittle hydraulic fills are constructed on
foundations that are themselves brittle and/or are susceptible to sufficient movement as to
induce undrained loading in the tailings structures they support.
It is interesting to note that the pore pressure contours indicate below hydrostatic
pore pressures in the slimes section of the dam, but the factor of safety was based on hy-
drostatic conditions (i.e. higher pre-failure pore pressures than actually existed). It is also
interesting to note that the slimes zone was actually assigned higher shear strength than
the outer shell. The basis for this judgment is unknown, but anyone who has ever walked
(or attempted to do so) on slimes would find such a judgment literally insupportable. The
flowslide nature of the failure also confirmed the designer’s judgment as insupportable,
and the designer’s stability analyses non-representative of the actual safety of the dam.
The designer’s analyses (notwithstanding the paradox of slimes being stronger than
shell tailings) were relevant only so long as no triggering mechanism for undrained be-
havior existed. As such, the safety of dams susceptible to static liquefaction is perhaps
better expressed in terms of the cumulative probability of potential triggering mechanisms.
Despite the obvious difficulties in quantifying this probability, the approach has the virtue
of at least recognizing the potential for static liquefaction. A factor of safety of 1.7 pro-
vides no such recognition and in fact provides a false sense of security.
Carrier (1991) provides one of the most comprehensive discussions of the drained
versus undrained shear strength issue for tailings dams. It is the opinion of the authors
that Carrier’s paper should be mandatory reading for any engineer involved in the analy-
sis, design, and construction of tailings dams, particularly those of the upstream variety.
Unfortunately, as illustrated by the next case record, the distribution and appreciation of
this paper has not been as widespread as it should have been.
Figure 12 Tailings Dam Pre-Failure Geometry (after Carrier, 1991)
“Although more and more research studies are being made on the phenomenon of
liquefaction, the findings of such work, while increasing our knowledge on the subject, also
seem to be extending the “zone of ignorance” surrounding the problem – generating fur-
ther questions rather than providing answers. Very simply, two basic lessons that have
been learned from several well-publicized failures resulting from liquefaction are: (i) ensure
that the density of the dike tailings is greater than “critical” and (ii) provide positive drain-
age so that all tailings within the retaining structure are not in a saturated condition”
(Smith, 1972).
Thirty years later, it is debatable whether the “zone of ignorance” is in a mode of
dissipation or expansion. Regardless, Smith’s basic lessons are as valid today as they
were 30 years ago.
Table 1 implies that materials with a state more dilatant than about Ψ<-0.1 will not be a
concern for undrained shear phenomena. This is purely from the author’s experience and
has been noted elsewhere (e.g. Davies, 1999). It is interesting to note that independent
experience of others, e.g. Been (1999) and Jefferies (1999), suggest values of –0.08 and
–0.10, respectively, as their practical limits of minimum state to ensure satisfactory engi-
neering performance provided the drained strength of the material is sufficient for all
loading conditions. In more simple language, the best way to deal with a problem is to
avoid creating it in the first place.
Concluding Remarks
The concepts and lessons offered by this paper are not new and should not be
seen in that light. The fundamental soil mechanics related to static liquefaction have been
evident for decades. The list of case histories, both older and more recent, together with
the rate at which case histories continue to be added to the failure database, demon-
strates that undrained shear response of mine tailings has not been appropriately appreci-
ated by a sufficient number of tailings dam designers. A thorough understanding of the
fundamentals and the lessons offered by the case histories should be mandatory training
for all engineers working with tailings dams, or any other hydraulic fill structures. This
training will probably be at odds with some of their education and/or the literature they
have viewed as professionals; but is necessary to correct a trend that requires reversal
sooner rather than later.
References