A Finite Element Analysis of the Fundão Dam Failure

Reza Moghaddam, Guillermo Riveros, Siavash Farhangi

Golder Associates, Mississauga, Ontario, Canada

ABSTRACT
The Fundão tailings dam failure of November 2015 in Brazil is one of the deadliest and most environmentally damaging
tailings dam breaches in recent history. Roughly 32 million cubic meters (Mm3) of iron mine tailings were accidentally
released in this catastrophic collapse, claiming the lives of nineteen villagers, and causing major environmental concerns
after polluting local water systems. As part of the forensic investigation that followed, a finite difference analysis (FDA)
simulation using the NorSand constitutive model in FLAC software was conducted by the Panel to test the hypothesis of
lateral extrusion triggered failure. The purpose of the present study is to simulate the triggering of static liquefaction failure
of the Fundão dam using a finite element analysis (FEA) approach in which the NorSand constitutive model is adopted in
Rocscience RS2 software. A selected number of the Panel’s laboratory test results were simulated in a series of FEA
models to determine the validity of the numerical results and the strain-softening behaviour of the tailings. Drained triaxial
compression tests were simulated to confirm calibration of the constitutive model. A computer model of the failing section
of the dam’s left abutment was subsequently generated for numerical analyses following the depositional details provided
in the Panel’s report. The results of the FEA are compared with the FDA shear stress-strain behaviour of the tailings
reported by the Panel on a laboratory scale and for the dam’s failing section, thus supporting the hypothesis of a slope
failure triggered by a lateral extrusion mechanism. The implications and limitations associated with the FEA are further
discussed in this paper.

RÉSUMÉ
La rupture du barrage Fundão en novembre 2015 au Brésil est une des ruptures plus meurtrières et plus dommageables
pour l'environnement de l'histoire récente. Environ 32 millions de mètres cubes (Mm3) de résidus miniers de fer ont été
accidentellement libérés lors de cet effondrement catastrophique, tuant dix-neuf villageois et causant de graves problèmes
environnementaux après avoir pollué les systèmes d'eau locaux. Dans le cadre de l'enquête scientifique qui a suivi, une
simulation numérique de l'analyse des différences finies (FDA) utilisant le modèle constitutif NorSand dans le logiciel FLAC
a été menée par le Comité pour tester l'hypothèse d'une défaillance générée par l'extrusion latérale. Le but de la présente
étude est de simuler le déclenchement de liquéfaction statique du barrage Fundão en utilisant une approche d'analyse par
éléments finis (FEA) dans laquelle le modèle constitutif NorSand est adopté dans le logiciel Rocscience RS2. Une sélection
de résultats d’essais en laboratoire du Comité a été simulés dans une série des modèles réalisés à base d'analyse FEA
pour déterminer la validité des résultats numériques et aussi que la anti-écrouissage des résidus. Des essais de
compression triaxiale drainée ont été simulés pour confirmer l'étalonnage du modèle constitutif. Un modèle de la section
défaillante de la culée gauche du barrage a ensuite été généré pour des analyses numériques à la suite des détails de
dépôt fournis dans le rapport du Comité. Les résultats de la FEA sont comparés au comportement contrainte-déformation
de cisaillement de la FDA des résidus rapporté par le Panel à l'échelle du laboratoire et pour la section défaillante du
barrage, soutenant ainsi l'hypothèse d'une rupture de pente déclenchée par une extrusion latérale. Les implications et les
limites associées à la FEA sont discutées plus en détail dans cet article.

1 INTRODUCTION Following the failure, a review panel (the Panel) was

established to investigate the cause of the failure, issuing
The failure of the Fundão Dam in Brazil was undoubtedly a a report in August 2016 (e.g., Morgenstern et al. 2016).
rude awakening to the mining and mine-waste There have also been additional analyses of the Fundão
management industries about the assumptions made Dam such as Riveros and Sadrekarimi (2019) and Reid
during the design, construction, and maintenance of mining (2019). For the purpose of this study, the approach
dams. After accidentally releasing roughly 32 million m3 of adopted by the Panel (Morgenstern et al. 2016) was used
iron mine tailings on November 5, 2015, amounting to to perform a series of deformation analyses on the failing
about 61% of the impoundment’s contents, the breach section of the dam prior to its failure and to simulate the
buried the nearby town of Bento Rodrigues, claimed the displacements reported to have triggered the static
lives of 19 villagers, and considerably polluted the local liquefaction failure of the dam. Conversely to the analysis
natural water systems. using Fast Lagrangian Analysis of Continua (FLAC)
performed by the Panel, deformations are calculated in a loading on the concrete conduit. This resulted in placing the
series of four finite element analyses (FEA) in Rocscience crest directly over previously deposited slimes at the left
RS2 software in this study. The first iteration consists of a abutment.
relatively simple elastic analysis to determine the elastic
deformations within the dam. The second iteration includes 2.2 Panel’s Dam Failure Conclusions
non-associated Mohr-Coulomb properties for all materials
to evaluate the effect of yielding. Subsequently, with one Following a thorough investigation into the probable
iteration building on the Mohr-Coulomb model and another causes of the breach, Morgenstern et al. (2016) concluded
one on the elastic model, the third and fourth iterations that failure was due to static (flow) liquefaction triggered in
include the critical state constitutive model NorSand the left abutment where the necessary triggering conditions
(Jefferies 1993; Jeffries and Shuttle 2005) for the beached prevailed. The more compressible slimes that had
sand tailings in order to analyze the influence of state- encroached the sand beach area beneath the setback
dependent strength variations and yielding in the triggering crest inhibited the free drainage required to control the
of liquefaction. Lastly, to evaluate the implications phreatic surface, which in turn enhanced saturation of the
associated with the FEA conducted in this study, the loosely deposited sand tailings. The Panel hypothesized
horizontal displacement results obtained are compared that lateral spreading of the slimes induced a drained relief
with those reported by the Panel in their base analyses. of the effective mean stresses in the liquefiable sands at
constant deviatoric stresses – a mechanism they termed
2 FUNDÃO DAM FAILURE “lateral extrusion,” which ultimately led to the instability of
2.1 History of the Fundão Dam the tailings sand matrix.
With the goal to assess the influence of slope
The Fundão Dam was the main containment structure in deformations on the triggering of liquefaction at the Fundão
an iron tailings storage facility (TSF) built in the state of Dam, the Panel carried out a set of deformation analyses
Minas Gerais, Brazil. Two types of slurry tailings (i.e., sand on a cross-section through the region where the failure
tailings and slimes) were produced because of the flotation began. The deformation models were developed using
concentration method used in iron mining (Vick 1990). The FLAC finite difference software and simulated the staged
tailings were separately conveyed to the TSF. The sand construction of the dam in roughly four-month time intervals
tailings consisted of a mixture of sand-sized with finer silt throughout most of its operational history. As previously
particles that were free-draining, but susceptible to described, three distinct analyses were performed: An
liquefaction when loose and saturated. The slimes, in analysis employing solely elastic material properties
contrast, were much finer and clay-like in nature with low estimated from seismic cone penetration tests (SCPT); a
permeability (Morgenstern et al. 2016). Mohr-Coulomb analysis considering the strain-weakening
The initial design of the embankment consisted of a response of the encroached slime layers; and a critical
compacted earth fill starter dam that was to be raised with state analysis making use of the NorSand constitutive
an upstream configuration. The sands deposited upstream model to evaluate the effects of density-dependent shear
would in turn help retain water-borne slimes deposited strength and yielding. Their results showed that horizontal
separately farther upstream and prevent them from displacements were concentrated in two regions located at
reaching the dam, as long as a minimum sand beach- the interface of the sand tailings with the predominantly
distance of 200 m was maintained. Based on this initial slime layer. With the largest horizontal deformations being
design, saturation of the sands was to be prevented by a in a region downstream of the dam crest, the Panel
high-capacity drainage system built at the base of the concluded that this was the most plausible location where
starter dam that would allow the free-drainage of water liquefaction was triggered.
from the deposited tailings.
A series of incidents beginning shortly after 3 DEFORMATION ANALYSES MODEL
construction in 2009 required certain modifications to the PROPERTIES
initial plans and are believed to have promoted the
eventual triggering of static liquefaction in the sand tailings. FLAC utilizes an explicit finite difference formulation that
The first of these was the discovery of construction defects can model complex behaviors, such as problems that
in the base drain, which forced the closure of the entire consist of several stages, large displacements and strains,
drainage system. The revised design made use of a non-linear material behavior, or unstable systems.
blanket drain that was built on the surface of the sand However, compared to the FE methods, FLAC analyses
tailings already contained inside the TSF, thus allowing the are generally time consuming and not routinely carried out
widespread saturation of the sands below. The second in many geotechnical applications. As part of this study, the
incident involved the mismanagement of the sand beach deformation analyses of the Fundão Dam left abutment
distance required to prevent slime deposits and water were carried out with finite element (FE) method in
encroachment near the dam, which allowed slimes to settle Rocscience RS2 software. The base abutment analyses
in areas closer to the crest than originally designed. The conducted by the Panel were emulated to the best extent
third and last incident was the structural deficiency of a possible, using the triggering of static liquefaction failure to
concrete conduit beneath the dam's left abutment, which draw comparisons of the results. This section describes
was found incapable of sustaining any further loading both the development of the numerical model of the failing
generated by raising of the dam. Consequently, the dam’s slope, including the material properties employed, and the
alignment was pushed upstream with the intent to reduce series of deformation analyses conducted.
3.1 Fundão Tailings Deposition Cycle and Material The materials identified are namely the compacted and
Properties deposited sand tailings; graded sand-slime mixtures
The Panel’s numerical model was developed to simulate consisting of predominantly slimes (100% slimes), mixed
the staged construction of the dam following the sand and slimes (50% slimes), interbedded slimes (20%
construction history from survey data and internal dyke slimes), and isolated slimes (100% sand); and bedrock.
stratigraphy. Despite thoroughly describing the The layer of graded sand-slime mixtures represents the
methodology employed to define the spatial reconstruction relative proportions of these materials that the Panel
of the abutment, public records of the investigation do not considered to be credibly present throughout the cross
include clearly identifiable and scalable dyke profiles section based on deposition records and slimes mass
covering the complete timeline from March 2013 to balance analyses (Morgenstern et al. 2016). The unit
November 2015 in the same 4-month intervals used in the weights of all materials were considered 22 kN/m3. All
development of the Panel’s model. As a result of the limited material properties were obtained from the Panel’s report,
amount of information available, a simple two-stage model and their evaluation is briefly described in the following
was built for this study, covering the construction stages of sections.
the left abutment that occurred between August 2014 and
November 2015. Owing to the unequal stage 3.1.1 Elastic Properties
reconstruction timeline and understanding the effects this
has on the absolute horizontal displacements produced, The large-strain elastic properties of the deposited sand
only the incremental displacements from the latter timeline tailings were estimated by the Panel from 2015 SCPT data
are used in the comparison of the results. through the beach of the dam by dividing the calculated
Figure 1 shows the location of the section evaluated small-strain shear modulus (Gmax) values by a factor of 3.
and the numerical model developed for the series of FEAs Similarly, the elastic properties of the slimes were
conducted. empirically estimated from one-dimensional compression
The phreatic levels for each stage were based on testing, and those for the sand-slimes mixtures were
piezometric data reported by the Panel, and zero and full estimated as combinations based on their slimes-to-sand
saturation were respectively assumed above and below the proportions. Whereas the elastic property inputs in the
piezometric lines. The material properties of each of the Panel’s FLAC model were in the form of a bulk modulus, K,
geotechnical units in the model are presented in Tables 1 relationship as a function of effective vertical stress, ’vc
and 2 for the three types of FEA conducted. (assuming a Poisson’s ratio of 0.3), the FEA model in RS2
requires elastic inputs in the form of Young’s modulus (E).

(a) (b)

Figure 1. (a) FE model of the Fundão Dam’s left abutment and (b) plan view
 This relationship conversion was readily calculated moduli and undrained shear strength was employed. While
from elastic theory principles for plain-strain boundary the Panel used a strain-weakening Mohr-Coulomb
conditions. relationship involving peak undrained strength linearly
reducing to residual values at 20% shear strain, the Mohr-
Table 1. Summary of FEA Material Properties Coulomb model in RS2 is an elasto-brittle-plastic material
Mohr-Coulomb model, which means that the strength of the material
Material Elastic Properties1 instantly drops from its peak at initial yielding to the
Parameters
constant residual value thereafter. Since the brittle drop in
′𝑣𝑐 0.4
shear strength was noticed to result in considerably larger
Compacted 𝐸 = 52 ( ) MPa,
100 ’=35o, c’=5 kPa displacements than those in the strain-weakening model, a
Sands
simple elastic perfectly plastic model employing equal peak
E*=76 MPa, =0.3
and residual undrained strength ratios was adopted for the
analyses in this study.
′𝑣𝑐 0.4 An additional difference to the Panel analyses is that
Deposited 𝐸 = 52 ( ) MPa,
100 ’=33o, c’=0 kPa due to convergence challenges, constant values of E were
Sands
E*=100 MPa, =0.3 employed, corresponding to the 85th-percentile of the
Su/’vc=0.22 stiffness values estimated for the range of ’vc observed.
This percentile was selected to eliminate the otherwise
′𝑣𝑐 0.7
Predominantly 𝐸 = 9.4 (
100
) MPa, (peak) weighing effect of the low E-values for ’vc < 100 kPa in the
slimes arithmetic mean, which would result in the underestimation
Su/’vc=0.22
E*=18 MPa, =0.3 of stiffness at higher depths.
(residual)
Su/’vc=0.47
′𝑣𝑐 0.42 (peak) 3.1.3 NorSand Parameters and Calibration
Interbedded 𝐸 = 43.3 ( ) MPa,
100
slimes
Su/’vc=0.47
E*=70 MPa, =0.3
(residual) The critical state constitutive model NorSand was used to
Su/’vc=0.52 evaluate the effects of state-dependent shear behavior of
′𝑣𝑐 0.4
the deposited sand only. This particular constitutive model
𝐸 = 52 ( ) MPa, (peak)
Isolated slimes 100 requires the input of several parameters associated with
Su/’vc=0.52 the critical state (Eq. 1), stress- and state-dilatancy
E*=55 MPa, =0.3 (residual) relationships (Eq. 2 and 3, respectively), and plastic
Su/’vc=0.39 hardening (Eq. 4), all to be obtained from a comprehensive
′𝑣𝑐 0.47 (peak) program comprised of laboratory testing and iterative
Mixed sand 𝐸 = 30.4 ( ) MPa,
and slimes 100 numerical modelling.
Su/’vc=0.39
E*=75 MPa, =0.3
(residual)
𝑒𝑐𝑠 =  − 𝑒 ln σ′𝑚 [1]
Bedrock E=500 MPa, =0.3 - 𝑚𝑎𝑥 = 𝐷𝑚𝑖𝑛  + 𝑀𝑡𝑐 [2]
1
E* values correspond to the 85th percentile of those calculated
for the ranges of effective stress observed in the model 𝐷𝑚𝑖𝑛 = 𝑡𝑐 𝐷 [3]
𝑚𝑖𝑛

Table 2. Critical State Parameters

𝐻 = 𝐻𝑜 − 𝐻𝑦  [4]
Material NorSand Parameters
𝑒𝑐𝑠 : void ratio at critical state
Deposited =0.865, 𝑒 =0.024, 𝑀𝑡𝑐 =1.33, =0.38, 𝛤: reference void ratio on CSL
Sands
𝐻𝑜 =156, 𝐻𝑦 =756, 𝜒𝑡𝑐 =7.2 : slope of the critical state line (CSL) in e-ln(σ’m) space
̅ 𝑚 : mean effective stress (= p′ under triaxial
𝜎′
conditions= (𝜎̅1 + 𝜎̅2 + 𝜎̅3 )/3)
3.1.2 Mohr-Coulomb Parameters
𝜂: dimensionless shear measure as ratio of stress
For the Mohr-Coulomb model, the compacted sand Mohr-
invariants 𝜂 = 𝜎̅𝑦 /𝜎̅𝑚
Coulomb parameters were obtained from the dam’s
original design (Morgenstern et al. 2016). In contrast, the 𝐷𝑚𝑖𝑛 : maximum dilation rate (the subscript min reflects
deposited sand was modeled using the calculated critical
state friction angle of φ' = 33° with c’=0 kPa, as established convention of positive compression)
by the Panel from the triaxial compression tests completed 𝑁: volumetric coupling coefficient
as part of its investigation. Because the slimes and sand-
slimes mixtures were conceptualized to mobilize undrained 𝑀𝑡𝑐 : critical friction ratio in triaxial compression
strength immediately after each loading increment, 𝜒𝑡𝑐 : dilatancy constant in triaxial compression
successively dissipating excess pore water pressures
between loading stages, a combination of drained stiffness 𝜓 : state parameter (=ec – ecs)
𝐻𝑜 : plastic hardening modulus FEA. As explained by Morgenstern et al. (2016), “the
upstream zone is a result of material settling above the
𝐻𝑦 : change in hardening modulus with  highly compressible zone of predominantly slimes and
sliding along the interface with the bedrock”, while “the
As part of its investigation, the Panel carried out
downstream zone is a result of the dam’s geometry”.
extensive analyses to determine the NorSand parameters
However, it should be noted that the magnitudes shown for
shown in Table 2. In addition to those, the Panel estimated
the Panel’s results are different from those shown for this
the initial state parameter (ψ) for the analysis to equal -0.02
study due to the difference in stage reconstruction
based on a series of numerical simulations to match the
previously mentioned. As will be discussed later, a better
80th percentile state parameter within the model, estimated
comparison can be made by juxtaposing the incremental
from previous CPT investigations completed at the Fundão
displacements from August 2014 to November 2015
Dam. All the same parameters were used in this study; and
reported by the Panel with those obtained here.
as presented in Figure 2, finite element simulations of
isotropically consolidated drained triaxial compression
tests (IC-TxC) performed in RS2 exhibited relatively close
fidelity to the laboratory test results (Figures 2a and 2b), as
well as the Panel’s NorSand simulations in FLAC (Figure
2a). Note that because liquefaction triggering was
concluded by the Panel to have occurred under drained
conditions, only drained triaxial tests were simulated here.
Therefore, from the results observed, it was confirmed that
the NorSand constitutive model in RS2 was capable of
capturing the stress-strain relationship and volumetric
response of the deposited sands appropriately.

Figure 3. Comparison of horizontal displacement contours

between the elastic (a) FEA and (b) FLAC (adapted from
Morgenstern et al. 2016)

4.2 Mohr-Coulomb Analysis

The contours of the horizontal displacement from the Mohr-

Coulomb FEA obtained in this study are shown in Figure
4a. As identified in the previous elastic analysis, regions of
concentrated lateral deformations are located mostly in
areas within and above the interbedded slimes layers. The
highest displacements of 60-70 mm are still located in the
region below and downstream from the dam crest.
Comparably, the Panel’s absolute horizontal displacement
distribution, shown in Figure 4b, exhibits contours closely
matching those obtained from the FEA. As mentioned
before, however, the magnitudes shown for the Panel’s
results are different from those shown for this study as a
Figure 2. Comparison of the drained shearing behavior of result of the differences in stage reconstruction adopted.
Fundão sand tailings using (a) laboratory test data, and
FEA and FLAC simulation results; and (b) laboratory test
data and FEA results

4 DEFORMATION ANALYSES RESULTS

4.1 Elastic Analysis

The contours of the horizontal displacement obtained from

the simple elastic FEA identified regions of concentrated
lateral deformations, as shown in Figure 3a. These regions
are located mostly in areas within and above the
interbedded slimes layers, with the highest displacement Figure 4. Comparison of horizontal displacement contours
magnitudes (highlighted in red) ranging from 60-70 mm in between the Mohr-Coulomb (a) FEA and (b) FLAC
a region below and downstream from the dam crest. For (adapted from Morgenstern et al. 2016)
comparison, the absolute horizontal displacement
distribution obtained by the Panel is reproduced in Figure
3b, exhibiting similar contours as those obtained from the
4.3 Critical State Analysis observed with a few differences in magnitudes at specific
sections as is discussed next.
The critical state analysis employing the NorSand
constitutive model was performed as an extension of the
Mohr-Coulomb analysis in order to consider the
dependency of the deposited sand’s shear strength on its
density – or more precisely, its state of stress (i.e. the state
parameter). The contours of the horizontal displacement
from this FEA are shown in Figure 5a. Broadly speaking,
the distribution of concentrated lateral deformations is
somewhat similar to those in the previous analyses.
However, a slight upward shift of the region below and
downstream from the dam crest is observed, while also
exhibiting higher magnitudes of displacement (~122 mm).
In the same manner, as shown in Figure 5b, the Panel’s
absolute horizontal displacement distribution shows
contours generally comparable with those obtained from
the FEA.

Figure 5. Comparison of horizontal displacement contours

between the critical state (a) FEA and (b) FLAC (adapted
from Morgenstern et al. 2016)

5 DISCUSSION OF RESULTS

In the absence of a common loading-stage reference

profile between the simulations performed in this study and
those by the Panel (i.e. the one 15-month stage in this
study from August 2014 to November 2015 and the seven
4-month stages from March 2013 to November 2015),
resulting from the limited amount of information publicly
available, direct comparisons of the absolute horizontal
displacement magnitudes were not feasible. For this
reason, only the incremental changes in horizontal Figure 6. Comparisons of horizontal displacement trends
displacements from August 2014 to November 2015 at El. 856 m between FEA and FLAC for (a) elastic
reported by the Panel were extracted from the report and analyses, (b) Mohr-Coulomb analyses, and critical state
used for direct juxtaposition with the displacement results analyses.
obtained here; thus using the dam profile from August 2014
as the zero-displacement reference. As shown in Figure 6a, the largest horizontal
Because the largest horizontal displacements were displacements in the elastic FEA occur in a region beneath
shown to concentrate in a zone that was beneath and and slightly downstream of the dam crest, at a chainage
downstream of the dam crest, centered approximately distance of approximately 540 m. With a magnitude of 63
along El. 856 m, data in this section were used to compare mm, the maximum displacement obtained from the FEA
the results. Figure 6 presents an assessment of the compares well with that from the Panel’s FLAC model (67
incremental horizontal displacements obtained in each of mm). In the downstream direction, both trends seem to
the FEA. These plots also include the accumulative follow each other up to a chainage distance of about 700
horizontal displacements from the FLAC models used by m, where they begin to steadily diverge, although still
the Panel for the period of August 2014 to November 2015. somewhat paralleling one another. This segment
Altogether, consistent trends along the section can be corresponds to a region of sands deposited on interbedded
and isolated slimes. Differences between the square
discretization in FLAC and the 3-noded triangle meshing entirety of the section, including the segment overlying the
may have caused this slight variability. The trends are seen zone of predominantly slimes. However, there seems to be
to follow one another very closely upstream of the peak up a rather consistent offset in magnitude ranging from 10-15
to a chainage distance of 425 m with a displacement mm, with the FEA yielding the upper values. For reference,
difference of ~4 mm, in a region where the sands begin to this is equivalent to approximately 7-11% of the maximum
overlie the zone of predominantly slimes. Beyond this absolute horizontal displacement reported by the Panel
point, the trends show entirely different tendencies and (135 mm) in its critical state analysis.
magnitudes, with the FEA generally exhibiting lower To investigate this issue further, a new critical state
displacements than the FLAC analysis, with a maximum FEA was performed, coupling the NorSand constitutive
deviation of 14 mm. For reference, this corresponds to model for the deposited sands with simple elastic models
approximately 12% of the maximum absolute displacement for all other materials rather than the Mohr-Coulomb
reported by the Panel (118 mm) from its elastic analysis. models. As shown in Figures 7a and 7b, the contours and
Despite the difference in elastic parameter input from displacement magnitudes are more analogous to those of
the elastic FEA, the Mohr-Coulomb analysis shows very the Panel’s FLAC model, despite exhibiting moderately
similar results between the displacements estimated here smaller values. With a magnitude of 79 mm, the maximum
and those from the FLAC analysis. displacement obtained from the modified critical FEA is
As shown in Figure 6b, the largest horizontal now 6 mm lower than that from the FLAC analysis. In the
displacements in the Mohr-Coulomb FEA also occur at a downstream direction from the peak values, both trends
chainage distance of approximately 540 m with a seem to follow each other very closely. Likewise, both
magnitude of 67 mm, correlating properly with that from the trends are observed to begin diverging slightly while still
FLAC analysis (66 mm). Furthermore, both trends seem to paralleling one another upstream of the peak
continue following each other, with a slightly higher displacements, up to a chainage distance of 450 m.
deviation of ~7 mm at a distance of approximately 630 m, Although the trends do not exactly replicate one another
up to a chainage distance of about 730 m where they begin beyond this point, the leading tendency of horizontal
to diverge slightly. As before, and based on the results displacements of 20 mm in magnitude is approached by
obtained, the parameters used in this segment of the cross- this modified critical state FEA. Thus, these new results
section seem to be relatively similar despite the difference suggest that the yielding produced in the Mohr-Coulomb
in input mode – namely, the elastic perfectly plastic model materials in the first critical state FEA (Figure 8) is the
with constant stiffness used in the FEA. Conversely to the reason the horizontal displacements were initially
elastic FEA, however, both trends are observed to continue overestimated.
following each other very closely in the segment upstream
of the peak displacement, up to a chainage distance of 425
m, with a maximum displacement difference of ~7 mm
where the sands begin to overlie the zone of predominantly
slimes. Once again, the trends show entirely different
tendencies and magnitudes past this point, with the FEA
exhibiting much lower horizontal displacements than the
FLAC analysis. As a reference, the maximum deviation of
27 mm in this region corresponds to approximately 25% of
the maximum absolute horizontal displacement reported
by the Panel (106 mm).
This disparity of results is plausibly due to the axiomatic
effect in material stiffness produced by the buildup of
effective stress induced by an increasing overburden. In
other words, volumetric strains are not only affected by the Figure 7. (a) Contours of horizontal displacement obtained
confining effective stresses applied, but also by the from the critical state FEA employing NorSand for the
changes in void ratio produced with each loading/unloading deposited sands and elastic material properties for all other
cycle. Consequently, it seems reasonable that, since the deposited materials; (b) comparisons of the horizontal
E-’vc relationship could not be successfully implemented, displacement trends at El. 856 m between FEA and FLAC
the constant stiffness values assigned in the FEA may for the modified (left) and the initial (right) critical state
have been overestimated in the zone of predominantly analyses.
slimes, resulting in the smaller horizontal displacement
obtained. 6 CONCLUSIONS AND LIMITATIONS OBSERVED
Lastly, Figure 6c presents the results of the critical state
FEA. As shown in the graph, the largest horizontal Finite element simulations were performed on the failing
displacements occur at a chainage distance of section of the Fundão Dam. These analyses paralleled the
approximately 560 m, consistent with the slight shift of the approach followed by the investigation Panel using FLAC
concentrated region of displacements beneath the dam in its evaluation of the deformations that caused the
crest observed in both the FEA and the FLAC analysis triggering of static liquefaction failure. The methodology
shown in Figure 5. Unlike the two previous FEAs, the adopted consisted of a series of elastic, Mohr-Coulomb,
horizontal displacement trend in the critical state FEA is and critical state FEA deformation analyses.
observed to parallel that of the FLAC analysis for the
Figure 8. Contours of the yielded regions of initial FEA

For the most part, the FEA results exhibited regions of Vick, S. G. (1990). Planning, Design, and Analysis of
concentrated horizontal displacements upstream and Tailings Dams. John Wiley and Sons.
downstream of the crest similar to those produced by the
Panel’s FLAC models. Furthermore, the trends of Jefferies, M G (1993). Nor-Sand: a simple critical state
horizontal displacements at El. 856 m were observed to model for sand. GeotechniqueV43, N1, March 1993,
largely replicate the Panel’s results, albeit two appreciable P91–103. International Journal of Rock Mechanics and
differences. The use of the relationship of E as a function Mining Sciences and Geomechanics Abstracts, 30(5),
of ’ in the FEA produced convergence issues when used 276–276.https://doi.org/10.1016/0148-9062(93)92281-
in combination with Mohr-Coulomb parameters. As a result T.
of this limitation, and despite giving acceptable results for
the deposited sands and similar materials, the use of a Jefferies, M.G., & Shuttle, D.A. (2005). NorSand: features,
constant Young’s modulus in the elastic and Mohr- calibration, and use. In Soil constitutive models:
Coulomb FEAs for the regions of predominantly slimes evaluation, selection, and calibration. Edited by J.A.
produced considerable deviations in horizontal Yamamuro and V.N. Kaliakin. ASCE Geotechnical
displacement near these zones when compared with those Special Publication 128, pp. 204–236.
obtained from FLAC. Another observation is that yielding in
the Mohr-Coulomb materials produced horizontal Reid, D. (2019). Additional Analyses of the Fundão Tailings
displacements 10-15 mm larger in magnitude than the Storage Facility: In Situ State and Triggering
Panel’s results in the critical state FEA despite following the Conditions. Journal of Geotechnical and
same overall tendencies. Notwithstanding these few Geoenvironmental Engineering. 2019, 145(11):
limitations, and considering the scale of the deviations 04019088 ASCE, ISSN 1090-0241.
obtained with respect to the absolute displacements
reported by the Panel, this series of FEA was able to 8 ACKNOWLEDGEMENT
adequately replicate the deformations and regions of
critically-stressed tailings sands concentrating in the area The authors thank Dr. Joe Carvalho (Golder Associated
beneath and downstream of the dam crest. Thus, it can be Ltd.) for providing numerical analysis technical support.
conjectured, in agreement with the Panel’s conclusion, that
this is the region where liquefaction would most likely have
initiated.

7 REFERENCES

Galindo da Fonseca, P., and Galindo da Fonseca, I.

(2016). Brazil’s Greatest Environmental Catastrophe –
Samarco’s Fundão Tailings Dam. Environmental Policy
and Law, 46/5: 334-337.

Morgenstern, N., Vick, S. G., Viotti, C. B. and Watts, B. D.,

2016. Report on the Immediate Causes of the Failure
of the Fundão Dam, Gottlieb Steen and Hamilton LLP.

Riveros, G.A., & Sadrekarimi, A. (2019). Static liquefaction

analysis of the Fundão dam failure. Proceedings of the
ICOLD 2019 Symposium, (ICOLD 2019)/ Publications
du symposium CIGB 2019. Ottawa, Canada.