You are on page 1of 10

Engineering Structures 28 (2006) 1430–1439

www.elsevier.com/locate/engstruct

Numerical simulation of collapse scenarios in reduced scale tests of


arch dams
Sérgio Oliveira a , Rui Faria b,∗
a Department of Concrete Dams, Laboratório Nacional de Engenharia Civil, Avenida do Brasil, 101, 1700-066 Lisboa, Portugal
b Civil Engineering Department, Faculdade de Engenharia da Universidade do Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal

Received 6 June 2005; received in revised form 26 January 2006; accepted 30 January 2006
Available online 22 March 2006

Abstract

This paper is dedicated to the numerical simulation of failure scenarios of concrete dams, reproduced experimentally on reduced scale models.
As far as the concrete material is concerned, a Continuum Damage Mechanics model that incorporates two independent scalar damage variables,
one for tension and the other for compression, is adopted and briefly described. Both in tension and compression material softening is reproduced,
and the inherent localization phenomenon is regularized through a consistent characteristic length. Two applications involving mock-ups of large
concrete dams recently built are presented: one with respect to a scenario where concrete strength is gradually reduced until failure, and the other
concerning a scenario where a relative displacement is prescribed on a foundation fault. Performance of the numerical model is therefore checked
on rather dissimilar conditions, that is, when collapse is reached due to crushing of the crest arch, or influenced by a dominant crack crossing the
dam.
c 2006 Elsevier Ltd. All rights reserved.

Keywords: Damage model; Strain localization; Concrete dams; Reduced scale models; Numerical modelling

1. Introduction phenomenon of localization, characterized by the tendency of


local cracks to occur in vanishing volumes [1,2]. If, as in the
Safety re-evaluation of concrete dams that are currently in present paper, a continuum finite element (FE) discretization is
operation, of which many of them were built in the middle of adopted for the dam, numerical simulation of cracks is usually
the 20th century, is actually a field of great concern. The most performed also via distributed (or continuous) approaches,
important static and seismic failure scenarios of concrete dams a strategy that with regards to tensile softening must be
need therefore to be identified, and properly devised numerical implemented wisely in view of the localization phenomenon,
models to realistically simulate behaviour under such extreme to reduce the tendency from the numerical solutions to depend
load conditions are necessary to access the actual safety of these drastically on the refinement of discretization. Accordingly, for
structures, in order to undertake rehabilitation measures in case very refined meshes localization limiters are usually adopted,
of necessity. whereas for very coarse meshes special softening branches are
Regarding the specific material of which these dams are frequently used [1–4].
made up — massive or unreinforced concrete — numerical Besides the capability for reproducing tensile induced
simulation of failure scenarios entails the use of constitutive
cracks, the numerical model to be used for the analysis of
models suitable for predicting the formation and growth of
concrete dams must account also for the nonlinear behaviour
cracks. This points to the need for the numerical model
under compression, coping with phenomena like three-
to cope with tensile softening, a well known feature that
dimensional (3D) confinement. In this work a 3D constitutive
entails considerable numerical difficulties engendered by the
model based on Continuum Damage Mechanics (CDM) is
presented, which incorporates two independent scalar damage
∗ Corresponding author. Tel.: +351 22 508 1950; fax: +351 22 508 1835. variables: one for tension and the other for compression. By
E-mail address: rfaria@fe.up.pt (R. Faria). assuming that the evolution of the tensile and compressive

0141-0296/$ - see front matter 


c 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2006.01.012
S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439 1431

damage occurs separately, with dissimilar softening branches at local level, that is, d = 1 − An /A0 , and accordingly
for tension and compression, this CDM model is presented in it evolves between ‘zero’ (intact material) and ‘one’ (when
Section 2. material reaches collapse).
Under tensile conditions the adopted constitutive model In a 3D problem the effective stress concept is generalized
resembles the Smeared Crack Approach (SCA), and accord- to σ = Dε, where D stands for the elastic constitutive tensor,
ingly each crack is thought to spread over a band inside each and ε is the strain tensor. Besides, as concrete features under
FE, being the material fracture energy dissipated on the ‘char- tension are quite dissimilar than under compression, we will
acteristic length’, concept originally applied to concrete struc- adopt here two independent scalar damage variables d + and d − ,
tures by Bazant and Oh [2]. Having in mind the applications to reproduce the internal degradations induced by the tensile
to be presented in Section 4, Section 3 details the implemented and compressive stress tensors σ + and σ − that arise from the
strategy for computing this characteristic length, according to decomposition of the effective stress tensor σ in the space of
the consistent methodology suggested by Oliver in [5]. At each the principal stresses and directions (see [9] for further details).
FE this approach allows an adequate dissipation of the concrete Furthermore, in the present model a free energy potential
fracture energy, and strongly attenuates problems induced by with the form [9]
the crack localization tendency [6].
1 1
As real scale tests are obviously unfeasible for failure ψ= (1 − d + )(σ + )T D−1 σ + (1 − d − )(σ − )T D−1 σ ≥ 0
2 2
scenarios of concrete dams, in this paper validation of the
(1)
numerical model predictions is made in Section 4 through
reduced scale experimental tests undertaken at the Laboratório is postulated. Since during a loading process the dissipated
Nacional de Engenharia Civil (LNEC), in Lisboa, Portugal. energy can never decrease, which is the thermodynamic coun-
Two applications concerning large concrete dams recently built terpart of the entropy tendency to grow, the Clausius–Duhem
in Portugal are presented: (i) the performance of the Alto inequality −ψ̇ + σ T ε̇ ≥ 0 must be fulfilled [10,11], which with
Lindoso arch dam under a scenario where concrete strength regard to the free energy potential assumed in (1) leads to the
is gradually reduced until failure, and (ii) the performance classical requirements in CDM
of the Alqueva arch dam under a scenario where a relative
displacement is prescribed on a foundation fault. In application ḋ + ≥ 0 ḋ − ≥ 0 (2)
(i) numerical results are compared to the ones from a failure
as well as to a constitutive law with a rather attractive format
test performed on a gypsum–diatomite reduced scale model,
(see details in [9,12]):
being collapse strongly influenced by tensile cracking of the
arch dam, as well as by compressive crushing of the crest σ = (1 − d + )σ + + (1 − d − )σ − . (3)
arch. For application (ii) a comparison between the predictions
from the CDM model and the results obtained from a failure Inequalities (2) impose the damage variables to have
test performed on a cement mortar reduced scale model is monotonously increasing evolutions, a classical condition that
presented; in the mock-up collapse was greatly influenced is fulfilled by the evolution laws to be defined in Section 2.3.
by a dominant crack, which started close to the fault and
propagated throughout the body of the dam. These applications 2.2. Damage criteria
are intended to demonstrate the adequacy of the proposed
numerical model to trace the response of concrete dams up In order to compare general states of stress it is worth using
to failure, putting also into evidence the advantages of using scalar norms τ + and τ − , one for the tensile and the other
both reduced scale tests and numerical analyses to reproduce for the compressive effective stress tensors, here postulated as
the relevant collapse scenarios of these structures. follows [9]

2. Damage model τ + = (σ + )T D−1 σ + (4a)
√
τ− = 3(K σ − −
oct + τ oct ) (4b)
2.1. Constitutive law
where K is a material property, and σ − −
oct and τ oct are the normal
An essential idea in CDM [7] is that if at each material and tangential components of the octahedral stress obtained
point an initial elemental area A0 is considered, subsequent from tensor σ − .
application of an external loading will engender initiation and If one designates by rt+ and rt− the greatest values (here
growth of micro-cracks, and consequently the ‘net’ area An termed as thresholds) ever reached until an instant t by the
will reduce. On An the actual stress is termed ‘effective’ and equivalent stresses τ + and τ − , increments of the damage
denoted by σ , to be distinguished from the usual Cauchy σ variables during time interval [t, t + t] are assumed to occur
stress commonly adopted in structural applications. Following only when τ ± ±
t +t > rt [see [8] for details]. Accordingly, the
this interpretation σ is physically more representative than following two separate damage criteria [13] are adopted for
σ , since it acts on the actual existing area An [8]. On this tension and compression
simple 1D interpretation the damage scalar d may be viewed
as the instantaneous surface density of material defects existing τ+ − r+ ≤ 0 τ − − r − ≤ 0. (5)
1432 S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439

Fig. 1. Process of fracture: (a) real process; (b) smeared FE approach.

Based on Eqs. (4) and (5), the initial elastic domain is defined
√ of deformations on a band of FE oriented with the crack (see
by conditions
√ τ + = ro+ and τ − = ro− , where ro+ = f o+ / E Fig. 1(a) and (b)).

and ro− = 3/3(K − 2) f o− , with f o+ and f o− being the Combining the constitutive Eq. (3) and the evolution law
stresses that define the onset of nonlinearity on 1D tension defined in Eq. (6a) for d + , under tensile conditions the
and compression characterization tests, and E representing the described damage model performs conceptually very similarly
Young’s modulus. to the SCA, in the sense that no discrete representation
of a crack is performed, and that a ‘continuous’ FE
2.3. Evolution laws for the damage variables displacement based formulation is adopted throughout, both
for the ‘undamaged’ or ‘damaged’ finite elements. Since d +
In strict dependence to the definitions given above for the is not associated to any specific direction, the exact concepts
thresholds r ± , appropriate evolution laws need to be introduced of ‘localized crack’ or ‘crack orientation’ are not directly
for the damage variables d ± to reproduce both the tensile foreseeable; yet, on a refined mesh the alignment of the
softening and the compressive hardening/softening observable damaged FE may provide a clear identification of the dominant
in concrete. With such a goal the following evolution laws are cracks, even for large scale structures like arch dams, as will be
proposed demonstrated in the applications.
ro+ A+ (1−r + /ro+ ) Tensile induced cracking is the macroscopic counterpart to
d+ = 1 − e (6a)
r+ the descending branch schematically reproduced on Fig. 1(a),
r− which occurs in concrete when the imposed strain exceeds the
d − = 1 − o− (1 − A− ) − A− e B (1−r /ro )
− − −
(6b)
r deformation corresponding to the peak stress on the 1D σ –ε
curve. Since Bazant and Oh [2] it was recognized that the use
where the parameter A+ is fixed so as to guarantee an
appropriate dissipation of the tensile fracture energy G f of constitutive laws with softening in the SCA leads to solutions
of concrete inside each FE (to be discussed in Section 3), strongly dependent on the mesh refinement, as on such an
and the parameters A− and B − are defined by fitting the approach a crack is viewed as introducing large gradients on
numerical σ –ε curve with an experimental one obtained from the displacement field, as well as concentrations of strains
a 1D characterization test in compression. Being softening in inside the FE. The interpolation functions of continuity Co
compression relevant as well, parameters A− and B − should usually adopted in FE approaches are unable to reproduce such
accordingly be defined so that an adequate dissipation of the high gradients on the displacement field, but as one goes into
compressive fracture energy is ensured, in the same manner increasingly refined FE meshes the singularity associated to a
adopted for the tensile fracture energy. crack tends to be captured, yet without reaching a physically
converged solution since the former is likely to concentrate on
3. Strain localization and objectivity of FE solutions almost vanishing regions.
To overcome this drawback, known as ‘lack of objectivity
3.1. Introduction of the numerical solutions regarding the FE mesh refinement’,
and to ensure that during the formation of a crack the available
In many applications simulation of cracking may be
tensile fracture energy G f of concrete (assumed as a material
performed by combining a standard FE discretization with a
Smeared Crack Approach [1]. The original idea of such a property) is adequately dissipated, within the SCA the area g f
strategy consists in analyzing the whole structure with the finite below the σ –ε curve (see Fig. 1(a)) should fulfill the condition
elements usually adopted for the continuum, and assuming that Gf
on such a mesh a crack would be identified by a localization gf = (7)
w
S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439 1433

Fig. 2. (a) Cracking band; (b) displacement evolution along normal x1 .

where w expresses the width of the Fracture Process Zone normal to the crack (see Fig. 2(a)). According to [5] within the
(FPZ), that is, the region where fracture occurs. The FE cracking band displacements may be defined by
counterpart of Eq. (7) is (see Fig. 1(b))
u(x ) = u− (x 2 , x 3 ) + φ(x )[u+ (x 2 , x 3 ) − u− (x 2 , x 3 )] (10)
Gf
g ∗f = ∗ (8) where u− (x 2 , x 3 ) denotes the displacement vector on a
w
point located on the crack boundary Γ − , u+ (x 2 , x 3 ) is the
where w∗ is the relevant dimension of the finite element where displacement vector on the homologous point located on
a crack is to be represented (usually different from w), and g ∗f the crack boundary Γ + , and φ(x ) is a continuous and
is the artificial specific fracture energy to be attributed to the differentiable scalar function that assumes value ‘zero’ at Γ −
1D σ –ε tensile curve adopted on such an element, in order and value ‘one’ at Γ + .
to guarantee that the available (and physical) energy G f is Within a volume V ∗ inside the band the energy dissipated by
dissipated at the end of the cracking process. cracking is computed according to (see [5])
During a 1D tensile test the evolution law expressed in  
Eq. (6a) induces a σ –ε softening branch that is asymptotic to W∗ = g f dV ∗ = G f φ(x ) dΓ ∗ (11)
the strain axis. Consequently, parameter A+ must be fixed to V∗ Γ∗
comply with the requirement expressed in Eq. (8), which leads where Γ ∗ is the contour of V ∗ . Having in mind Green’s theorem
to [14] this equation may be transformed into
 −1  
GfE 1 ∂φ
+
A =  2 − ≥ 0. (9) W∗ = g f dV ∗ = G f ∗
 dV . (12)
w∗ f o+ 2 V∗ V ∗ ∂ x1

The local form of Eq. (12) is therefore


3.2. Consistent characteristic length in 3D FE meshes Gf ∂φ
g ∗f = = Gf  (13)
w∗ ∂ x1
Inherently to what was stated above, in the SCA realistic
simulations of concrete fracture are only possible if a consistent and consequently the cracking bandwidth is necessarily
definition for the characteristic length w∗ of each FE under  
consideration is adopted. This means that the definition to be ∗ ∂φ −1
w = . (14)
selected would have to take into account the relevant dimension ∂ x 1
of the FE, as well as the path followed by the crack within For a complete definition of the w∗ function φ needs to
it, conditions that in the present paper will be accomplished be postulated for the intended 3D applications. If cubic-like
by adopting the consistent approach originally developed in finite elements are adopted (see Fig. 3), irrespective of the
Oliver [5], here generalized to 3D conditions. specific number of nodes that characterize the FE in use for
The occurrence of a crack can be associated with the the structural analysis, a straightforward definition for φ is
formation of: (i) a discontinuity on the gradients of the
displacement field, and (ii) a jump on the displacements
8

across the crack [15–17]. Within the SCA framework, where φ( y P ) = Ni∗ ( y P )αi (15)
i=1
FE with continuity Co are adopted, only discontinuity (i) is
directly reproduced; therefore a crack should be understood as where Ni∗ are the interpolation functions of a linear FE whose
a fracturing band, with a bandwidth w∗ tending to zero (see nodes are coincident with the eight vertices of the finite element
Fig. 2(a) and (b)). Let us introduce on such a cracking band selected for the structural analysis, y P represent the local
a referential with local axes x j ( j = 1, 2, 3), with x 1 being coordinates of a point P on the FE (for instance a Gauss point),
1434 S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439

Fig. 3. Cracking band and crack orientation on a 3D FE mesh.

and αi are parameters that assume value ‘one’ on the vertices


located above a plan that passes through the centre of the FE
and it is perpendicular to the crack normal at P (on the other
vertices αi assumes value ‘zero’).
The characteristic length at point P is finally given by
 
  −1
∂φ −1 8 3
∂ Ni∗ ( y P )  T

w = = (e1 ) e j αi (16)
∂ x 1 i=1 j =1
∂x j

where e j denotes the unit vector of the global coordinate axis


x j and e1 represents the unit vector of direction x 1 normal to
the crack. This definition for w∗ is consistent with the finite
element approximation adopted in the numerical discretization,
and it will be adopted in the applications to be presented in the
next section. Fig. 4. Alto Lindoso arch dam.

4. Applications

4.1. Alto Lindoso dam

The Alto Lindoso dam (see Fig. 4), finished in 1991 in the
north of Portugal, is a double curvature arch with a maximum
height of 110 m, a total crest length of 297 m, and a central
cantilever with a variable thickness ranging from 4 m on the
crest and 21 m close to the foundation.
Safety evaluation of this dam, regarding a scenario where
concrete strength is thought to be gradually reduced until
failure, was carried out at LNEC by using a reduced scale
model built up at scale 1:250 with gypsum–diatomite (see
Fig. 5). Since during lifetime progressive deterioration is Fig. 5. Gypsum–diatomite mock-up of Alto Lindoso dam (scale 1:250).
unquestionably engendered in all concrete structures, for many
decades this failure scenario was selected at LNEC to compare the intended scenario by forcing material strength to reduce
the designs and performances of many large dams [18]. continuously until failure. Thus, the maximum load factor
Although for the present scenario one might think it as (or safety coefficient) λ S for this scenario is experimentally
commendable to reproduce all the loads that are feasible to determined via an indirect and dual strategy, which consists
act almost permanently, for simplification it is usually assumed in keeping constant the material properties, but increasing
that only the dead load from the concrete dam and the gradually the relevant loads applied to the reduced scale model.
hydrostatic pressure are of relevance, and so the influence of This is accomplished by increasing the oil pressure on a set of
the thermal action is neglected. Furthermore, and regarding hydraulic jacks that are fixed against the upstream face of the
obvious practical difficulties, it is unfeasible to reproduce mock-up (see Fig. 5), according to a configuration where each
S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439 1435

Fig. 6. FE meshes for Alto Lindoso dam: (a) coarse; (b) refined.

Fig. 7. 1D constitutive laws.

jack induces a force that is locally the resultant of the dead load
and the hydrostatic pressure.
In order to numerically simulate the present scenario the Fig. 8. Radial displacement of a point on the central cantilever.
dam and the adjacent foundation were discretized with 20-
noded 3D isoparametric FE, using either the coarse or the up an envelope for the softening conditions in compression, two
refined meshes reproduced in Fig. 6; in both discretizations variants were adopted for the 1D post peak branch (see Fig. 7
three FE were considered across the thickness. As far as for clarification): (i) one reproducing the typical softening
the boundary conditions are concerned, only the foundation on gypsum–diatomite admixtures, with a compressive fracture
bottommost nodes of the FE meshes of Fig. 6 were fixed, energy around 38 kN m/m2 , and (ii) one comparable to the
with the displacement components inhibited along the three softening usually found in concrete, with a compressive fracture
directions, as in the mock-up set-up (see Fig. 5). Nonlinear energy of circa 105 kN m/m2 ; a complementary one with no
performance of the gypsum–diatomite was simulated with softening at all, and assuming a horizontal branch after the
the CDM model described in Section 2, in compliance with peak strength, was also considered for illustration purposes.
the procedure referred to in Section 3 to cope with tensile In Fig. 8 the experimentally obtained evolution of the radial
induced strain localization. Calibration of the CDM model displacement on a point located on the central cantilever, and
parameters was performed by fitting a σ –ε curve to the average as a function of the load factor λ, is compared to the ones
stress–strain diagram obtained during the characterization tests predicted numerically with the coarse mesh reproduced in
for the gypsum–diatomite mixture (see Fig. 7). Transformed Fig. 6(a), and for the softening alternatives just mentioned for
into the real scale of the dam material properties were the compression. This figure emphasizes that neglecting softening
following: Young’s modulus E = 30 GPa, Poisson’s ratio in compression leads to a quite unsafe prediction of the ultimate
υ = 0.2, compressive strength fc = 30 MPa, tensile strength load factor, in the present scenario 50% superior to the value
ft = 7 MPa, G f = 5 kN m/m2 and density ρ = 2400 kg/m3 . λ S = 10.7 obtained in the experimental test; as regards
For the foundation a linear elastic behaviour was assumed, with the displacements at collapse, unrealistically large predictions
the same elastic properties as for the dam. were also obtained with such a hypothesis, which underlines
In order to assess the influence of the descending branch in the considerable relevance of softening even in compression.
compression on the ultimate behaviour of the dam, and setting Conversely, under the softening hypotheses (i) and (ii) the
1436 S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439

numerical predictions stood quite close to the experimental test


results, both in terms of the load factor (λ S = 11.1 and λ S =
12.1, respectively) and the radial displacement at collapse.
Since for the reference load configuration (λ = 1) quite low
compressive stresses are installed on this dam, typically less
than 3.4 MPa, the large values here reported for λ S are to be
considered normal and expected, since damage in compression
started only above the load factor λ = 7. The ultimate load
factor obtained numerically for the concrete type softening
branch in compression was slightly superior to the experimental
one, which is consistent with other reported failure tests of
reduced scale tests of dams, where the safety factors obtained
with cement mortar mock-ups are usually greater than those
obtained with gypsum–diatomite models [19].
Switching to the refined mesh depicted in Fig. 6(b), the
load factor predicted numerically at collapse for the concrete
type softening branch in compression changed to λ S = 11.3,
which compared to the value λ S = 12.1 obtained with the
coarse mesh corresponds to a variation of 6.6%. This change
may be considered quite small — particularly if compared to
some unavoidable larger errors involved in the experimental
campaign — which demonstrates that the strategy described
in Section 3 really succeeded in significantly reducing the
numerical solution dependency on mesh refinement when
material softening is a subject of concern. Accordingly,
subsequent results will refer only to numerical applications
obtained with the FE mesh illustrated in Fig. 6(a).
Fig. 9 reproduces the distributions of the compressive and
tensile damages at collapse on the present scenario, as well
as the principal stresses on both faces of the Alto Lindoso
dam. The failure regions predicted by the numerical model
are almost coincident with those observed in the reduced scale
model: crushing due to excessive compression on the centre of
the crest arch is evidenced in Fig. 9 by the concentration of Fig. 9. Principal stresses and damages at collapse.
high values of d − , and on detail A of Fig. 10 (mock-up), and
the same applies to the dam insertion into the foundation, on
the downstream face (compare Fig. 9 and detail B of Fig. 10).
Therefore, predictions from the described numerical model are
demonstrated to be realistic for the present load scenario, both
in what concerns the safety factor of the dam and the ultimate
strength mechanism.

4.2. Alqueva dam

The Alqueva dam (see Fig. 11), recently built in the south of
Portugal, is a symmetric double curvature arch with a maximum
height of 96 m, a chord length of 317 m at the crest and a central
cantilever with a thickness varying from 7 m on the crest to
33.5 m on the base. The foundation consists mainly of green Fig. 10. Collapse of the mock-up.
schist, with elastic properties identical to the concrete ones. At
the left bank the dam insertion into the foundation is intersected at LNEC. For this purpose a 1:250 reduced scale model
by the so-called Alqueva fault, reproduced in Fig. 11, which of cement mortar was built (see Fig. 12). In this mock-up
presents an angle of 40◦ with the horizontal plane, and a fault the dead load and the hydrostatic pressure were simulated
trace forming an angle of 25.5◦ with the upstream–downstream with the same hydraulic jack system already referred to for
direction. the Alto Lindoso dam. As regards to the fault movement,
A failure test concerning a scenario where a horizontal and since only a horizontal component was envisaged, for
sliding was prescribed to the borders of this fault was performed simplification in the mock-up it was decided to reproduce the
S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439 1437

Fig. 13. FE mesh for the mock-up of Alqueva dam.

Fig. 11. Alqueva arch dam.

Fig. 14. 1D constitutive laws.

foundation. Fig. 14 reproduces the 1D σ –ε numerical curve


Fig. 12. Cement mortar mock-up of Alqueva dam (scale 1:250). that was fitted to the average stress–strain diagram obtained
from the characterization tests performed for the cement mortar
material adopted in the mock-up, showing the following main
fault as vertical. This assumption allowed us to install a mobile features, when scaled to the real size of the arch dam: (i) in
base on the foundation block, at the left bank, which during compression the peak strength is 25 MPa, and softening is
the experimental test was forced to move horizontally, via a characterized by a fracture energy of about 750 kN m/m2 ;
screwed bolt aligned with the fault trace (see the indication (ii) in tension the peak strength is f t = 3.5 MPa, and the
on the right of Fig. 12). A few problems were detected during asymptotic softening branch reproduced in Fig. 14 encloses a
the test with this mechanical system, since in addition to the domain inherent to a fracture energy G f = 0.8 kN m/m2 .
intended component parallel to the fault and towards upstream, The real components of the fault movement registered
a spurious component normal to the fault, and towards the left during the experimental test were reproduced in the numerical
bank, was also introduced. The amplitude of this unexpected model. This was accomplished by assuming that the foundation
normal component was observed to be 3/4 of the intended bottommost nodes standing between the fault and the right
sliding one parallel to the fault. bank were fixed in all directions; with regards to the foundation
For the numerical simulations both the dam and the bottommost nodes standing between the fault and the left bank
foundation were discretized with the 3D FE mesh reproduced the vertical displacements were prevented, whilst along the fault
in Fig. 13. Similarly to the assumption made for the mock- trace and normal to it horizontal displacements were prescribed
up, the fault was represented on the numerical model by a to simulate the intended movements. In overall terms, and
vertical discontinuity with a 10 m gap, reproducing the average for a 0.3 m sliding along the fault, Fig. 15 demonstrates that
Alqueva fault thickness at the real scale of the dam. The the arch displacements computed numerically agree with the
concrete Young’s modulus and Poisson’s ratio were taken as ones obtained in the reduced scale model: minor differences
E = 35 GPa and υ = 0.2, the same adopted also for the in the direction and the amplitude of the experimental and
1438 S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439

Fig. 16. Principal stresses and damage for a 0.1 m fault sliding ( f t = 3.5 MPa).

Fig. 15. Displacements for a 0.3 m fault sliding.

numerical displacements are detected. Even so, for the crest


arch the numerical displacements are greater than those
obtained experimentally.
Figs. 16 and 17 reproduce the numerical principal stresses
and damage distributions on the dam for the following values
of the prescribed fault sliding: 0.1 and 0.3 m. Significant
concentration of compressive stresses is observed on the dam
on the left bank, close to the fault, leading to damage in
compression that spread considerably at this location on the
foundation when the fault movement increases from 0.1 to
0.3 m. Collapse is mainly due to shear and tensile failures that
occur at local level, which coalesce on a dominant crack that
initially develops horizontally across the dam, starting on the Fig. 17. Principal stresses and damage for a 0.3 m fault sliding ( f t = 3.5 MPa).
fault and pointing towards the opposite bank (see the contours
of tensile damage in both the upstream and downstream faces in
Figs. 16 and 17); afterwards this crack changes its propagation
trend by rotating to the vertical direction, pointing towards the
crest arch, as indicated by the contours of tensile damage on in
the downstream face of Fig. 17.
Having in mind the final propagation tendency of this
dominant crack, which is aligned with a vertical construction
joint where tensile properties are expected to be lower than
for the rest of the dam, the scenario with a 0.3 m fault sliding
was also simulated by adopting the following reduced values
for the tensile strength and fracture energy: f t = 1.5 MPa
and G f = 0.4 kN m/m2 . As expected, the new damage
distributions reproduced in Fig. 18 indicate that the domains
with relevant tensile damage are now substantially increased,
and the dominant crack reaches the crest, following a vertical
alignment on the right bank.
Propagation of the bands with tensile damage reported in Fig. 18. Principal stresses and damage for a 0.3 m fault sliding ( f t = 1.5 MPa).
Figs. 17 and 18 are rather consistent with the final collapse
configuration obtained in the mock-up, reproduced in Fig. 19.
In fact, in the mock-up a major crack began at the foundation reaching the crest by following a vertical contraction joint that
close to the fault, and propagated horizontally across the dam, opened from the bottom up to the top on the opposite bank, as
S. Oliveira, R. Faria / Engineering Structures 28 (2006) 1430–1439 1439

movement on a tectonic fault crossing the foundation, the


collapse being influenced by a dominant crack that propagated
horizontally across the dam and vertically up to the crest
arch. This failure mechanism on the reduced scale model was
successfully captured by the numerical model, which indicated
the occurrence of collapse for a 0.3 m fault sliding.

References

[1] Bazant ZP, Cedolin L. Blunt crack band propagation in finite element
analysis. ASCE J Eng Mech Div 1979;105(EM2):297–315.
[2] Bazant ZP, Oh BH. Crack band theory for fracture of concrete. Matér et
Fig. 19. After failure, mock-up downstream view. Constr 1983;16(93):155–77.
[3] Planas J, Elices M. Nonlinear fracture of cohesive materials. Int J Fract
1992;51:139–57.
reproduced also in Fig. 18 for the numerical model predictions. [4] Jirásek M. Objective modelling of strain localization. Rev Française de
Furthermore, a vertical contraction joint aligned with the fault Génie Civil 2002;6:1119–32.
started to open in the mock-up (reported as “second crack” [5] Oliver J. A consistent characteristic length for smeared cracking models.
in Fig. 19), with a propagation from the crest towards the Int J Numer Methods Eng 1989;28:461–74.
foundation, which was also predicted in the numerical model [6] Bazant ZP, Jirásek M. Non-local integral formulations of plasticity
and damage: survey of progress. ASCE J Eng Mech 2002;128:
in spite of the hypothesis of continuity that was assumed (see 1119–49.
relevant locations in Figs. 17 and 18). [7] Kachanov L. Introduction to continuum damage mechanics. Dordrecht:
A final remark is made about the extreme severity of the fault Martinus Nijhoff Publishers; 1986.
movement that is expected to produce the collapse of the dam, [8] Lemaitre J. How to use damage mechanics. Nucl Eng Des 1984;80:
which is able to withstand a relative displacement as high as 233–45.
[9] Faria R, Oliver J, Cervera M. A strain-based plastic viscous-damage
0.3 m, considerably greater than the one that is likely to occur model for massive concrete structures. Int J Solids Struct 1998;35(14):
at the Alqueva fault. 1533–58.
[10] Lubliner J. On the thermodynamic foundations of non-linear solid
5. Conclusion mechanics. Int J Non-Linear Mech 1972;7:237–54.
[11] Lubliner J. Plasticity theory. NY (USA): MacMillan Publishing
This paper presents a numerical strategy suitable for Company; 1990.
[12] Faria R, Oliver J, Cervera M. Modeling material failure in concrete
reproducing different failure scenarios of concrete dams, which structures under cyclic actions. ASCE J Struct Eng 2004;130(12):
is validated via experimental tests performed on two reduced 1997–2005.
scale mock-ups loaded up to failure. It combines a constitutive [13] Simo J, Ju J. Strain- and stress-based continuum damage models. I:
model with two independent scalar damage variables to Formulation. Int J Solids Struct 1987;23(7):821–40.
efficiently reproduce concrete nonlinear performances in [14] Oliver J, Cervera M, Oller S, Lubliner J. Isotropic damage models and
smeared crack analysis of concrete. In: Proc. 2nd international conference
tension and compression, and a strategy to regularize strain on computer aided analysis and design of concrete structures. 1990. p.
localization based on the “consistent characteristic length” 945–57.
originally developed by Oliver. In the first application, [15] Jirásek M. Embedded crack models for concrete fracture. In: Borst et al.,
concerning the Alto Lindoso arch dam, the numerical model editors. Proc. EURO-C 1998 computer modelling of concrete structures.
was validated by reproducing a collapse scenario that induced 1998. p. 291–300.
[16] Jirásek M, Zimmermann T. Analysis of rotating crack model. ASCE J Eng
the crushing of the mock-up when both the self weight Mech 1998;124:842–51.
and the hydrostatic pressure were increased up to failure. [17] Chaves E. A three dimensional setting for strong discontinuities modelling
Independently of the mesh refinement adopted in two different in failure mechanics. Ph.D. thesis. Spain: Technical University of
FE discretizations, the experimental load factor at failure Catalonia; 2003.
was captured almost exactly by the numerical model, which [18] Rocha M, Serafim L. The problem of safety of arch dams. Rupture studies
on models. In: Proc. VI congress of large dams. New York; 1958.
predicted also the mock-up’s deformed configuration at [19] Oliveira SB. Models for the analysis of concrete dams, with allowance to
collapse. In the second application the mock-up of the Alqueva cracking and time effects. Damage formulations. Ph.D. thesis, Portugal:
arch dam was analyzed under a scenario involving a sliding University of Porto; 2000 [in Portuguese].

You might also like