You are on page 1of 16

Engineering Fracture Mechanics 71 (2004) 1891–1906

www.elsevier.com/locate/engfracmech

Size effects in concrete gravity dams: a comparative study


Gabriella Bolzon
Dipartimento di Ingegneria Strutturale, Politecnico di Milano, piazza Leonardo da Vinci 32, 20133 Milano, Italy
Received 13 May 2003; received in revised form 27 October 2003; accepted 27 November 2003

Abstract
This paper discusses the applicability of the size effects law which originates from linear elastic fracture mechanics to
concrete gravity dams. Comparison is made between the predictions, in terms of overall structural strength and failure
mechanism, obtained from simplified limit state methods, from equivalent elastic analyses and from cohesive crack
approaches. A benchmark problem proposed by ICOLD (International Committee on Large Dams) is chosen as a
reference test. The results lighten the peculiar role of the self-weight contribution, which stabilizes crack propagation for
typical structural sizes in dam construction.
Ó 2003 Elsevier Ltd. All rights reserved.

Keywords: Size effects; Cohesive fracture; Structural strength; Gravity dams

1. Introduction

The size effect law which derives from linear elastic fracture mechanics (LEFM) permits to relate the
nominal strength of self-similar structures to one dimensional parameter, say D, the material characteristics
being captured by fracture energy only; see, e.g., the recent book by Baant [1] and the references listed therein.
The application of the size effect law to quasi-brittle situations (e.g., to medium-size structures made of
concrete) entails the consideration of a material length, say cf , besides the structural size D (see, e.g., the
comprehensive treatise [2] and the abundant literature quoted therein); LEFM theory and expression are
recovered in this extended framework as the ratio cf =D vanishes. However, while the reliability of the size
effect law is acknowledged for large homogeneous structures, its use is still controversial at lower scale (for
instance, in laboratory tests) and for inhomogeneous and polycrystalline materials with large grain size; see,
e.g., the recent research work by Guinea et al. [3] and by van Mier and van Vliet [4,5] focusing on the
identification of the fracture properties of quasi-brittle materials, and the papers by Dempsey et al. [6,7] on
scale effects in large ice formations. In particular, the in situ experiences carried out by Dempsey et al. [6]
on samples of large grained freshwater ice, in the size range 1:81, show marked dependence of the strength
on the specimen size to the grain size ratio. The results of the experimental work carried out by van Vliet
and van Mier [5] on self-similar specimens of six different sizes in the range 1:32, made of concrete and
sandstone under uniaxial tension, suggest that the material strength is significantly influenced by the

E-mail address: gabriella.bolzon@polimi.it (G. Bolzon).

0013-7944/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfracmech.2003.11.006
1892 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

Nomenclature

a crack length
a0 initial value of the crack length
ac critical value of the crack length for crack propagation
aec equivalent crack length
B interface length
cf effective crack extension
D characteristic structural dimension
E YoungÕs modulus
E0 equivalent elastic modulus for plane strain condition
f geometry function
fc0 derivative of the geometry function evaluated for the critical crack length
f FR geometry function relevant to full reservoir hydrostatic pressure
f OT geometry function relevant to overtopping pressure
f SW geometry function relevant to self-weight contribution
f UL geometry function relevant to uplift pressure
Gf mode I fracture energy
H water level
Hmax maximum water level
KI stress intensity factor for mode I fracture
KIc critical stress intensity factor for crack propagation
KIFR stress intensity factor relevant to full reservoir hydrostatic pressure
KINc apparent fracture toughness for quasi-brittle fracture
KIOT stress intensity factor relevant to overtopping pressure
KISW stress intensity factor relevant to self-weight contribution
KIUL stress intensity factor relevant to uplift pressure
lch material characteristic length
Mt matrix collecting the interpolation functions for equivalent tractions
Mw matrix collecting the interpolation functions for nodal opening displacements
p uplift water pressure
P nodal force vector equivalent to uplift pressure
T internal nodal force vector (equivalent tractions)
TE internal nodal force vector in the absence of opening displacements (equivalent elastic tractions)
SW
TE equivalent elastic tractions relevant to self-weight contribution
FR
TE equivalent elastic tractions relevant to full reservoir hydrostatic pressure
bE
T equivalent elastic tractions independent of the water level H
E
T equivalent elastic tractions evaluated for unit water level H (equivalent unitary elastic
tractions)
EOT
T UL equivalent unitary elastic tractions relevant to overtopping pressure
E
T equivalent unitary elastic tractions relevant to uplift pressure
Tu ultimate nodal force vector (equivalent tractions relevant to tensile strength)
W nodal opening displacement vector
w opening displacement
wc critical opening displacement
z influence (Green) function defined over C
G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906 1893

Z influence matrix
a dimensionless crack length
ac dimensionless critical crack length
aec dimensionless equivalent crack length
cc concrete density
cw water density
C interface
f local coordinate running along C
m PoissonÕs ratio
n local coordinate running along C
q coefficient modulating fluid pressure
r stress component orthogonal to C
rE stress distribution in the absence of displacement discontinuity
rN nominal stress
rNc critical nominal stress
ru tensile strength
u crack activation function
U equivalent nodal values of the crack activation function

specimen shape and by material inhomogeneity and load eccentricity, so that the observed size effect is
mainly statistical (Weibull-type).
The study to be presented herein concerns dams which, actually, can be considered large size structures
in concrete construction. Their laboratory testing, when available, is necessarily performed on scale models
of prototype blocks, the main dimensions of such specimens ranging from some decimeters (for centrifuge
experiments) to a few meters, see e.g. [8–10]. In this context, the transfer of the prototype results to the
actual dam design by proper size effect law would be particular desirable for the purpose of construction,
retrofitting or repair, although discrepancies are almost unavoidable due to the fact that fracture properties
of concrete are related to the aggregate dimensions and these, in turn, depend on the casting volumes and
are therefore different in laboratory and in construction.
The applicability of the size effect approach to concrete dams is discussed here in the light of some
computational results referring to a benchmark problem proposed by the International Commission for
Large Dams (ICOLD) [11]. The benchmark structure is schematized as a plane strain problem, and crack
propagation at the foundation interface is defined by normal opening (mode I) and in-plane sliding (mode
II). The outcome of some parametric studies carried out on this mixed-mode fracture problem can be found
in [11]. The results to be presented in this paper refer instead, for simplicity, to mode I analyses only, the
conclusion to be drawn being independent of the actual fracture mechanism. The peculiar role of self-
weight contribution in this context is especially lighten.

2. Theoretical background

2.1. Linear elastic fracture mechanics

Within the assumption of LEFM, at the tip of a crack propagating in pure opening mode (mode I) in a
brittle material, a stress singularity arises which is characterized by the stress intensity factor (SIF) KI . For
self-similar structures, to be considered in what follows, KI can be expressed by
1894 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

Fig. 1. Schematic representation of the case test and of possible interface discretization, with explanation of some symbols.

pffiffiffiffi
KI ¼ rN Df ðaÞ ð1Þ

where rN represents the nominal stress; D defines the characteristic dimension of the structure, see e.g. the
schematic drawing of Fig. 1; f ðaÞ is a dimensionless geometry function, depending on dimensionless crack
length a ¼ a=D.
Explicit analytical expression of the function f ðaÞ relevant to some specific geometry can be found in
handbooks [15,16] especially for the shape of specimens to be used in standardized laboratory tests;
otherwise, the geometry function f ðaÞ can be evaluated numerically, by boundary integral equations,
compliance methods or displacement extrapolation techniques; see e.g. [17–21].
According to LEFM criterion, crack propagation occurs as soon as the SIF KI reaches its critical value
KIc , which is assumed to be a material property. In plane strain condition, KIc can be related to the fracture
energy Gf as follows:
pffiffiffiffiffiffiffiffiffiffi
KIc ¼ E0 Gf ð2Þ

where E0 ¼ E=ð1  m2 Þ, E and m respectively representing YoungÕs modulus and PoissonÕs ratio of the
considered material.
Unstable crack propagation (usually leading to ultimate structural failure) occurs when KI ¼ KIc for the
critical value of the crack length, say ac , such that:

0 df 
fc  >0 ð3Þ
da a¼ac

Condition (3), in fact, implies that any increase of the crack length due to fracture propagation would
increase the SIF beyond its value even for fixed load multiplier, see (1).
The structural strength can then be identified with the critical nominal stress, rNc :
KIc
rNc ¼ pffiffiffiffi ð4Þ
Df ðac Þ

In many significant situations (i.e. for ‘‘positive geometry’’, in common jargon [1]), f ðaÞ is a mono-
tonically increasing function of its argument; in this case, the critical crack extension ac coincides with the
initial crack length a0 and the nominal strength rNc depends on the structural size only for given material
toughness.
G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906 1895

Fig. 2. Piecewise linear and exponential softening law.

2.2. Cohesive crack models for concrete

Peculiar of quasi-brittle materials such as concrete is the possibility of transmitting some residual
(cohesive) stresses across fracture surfaces (here assumed to develop in pure mode I) up to some critical
value wc of the crack opening displacement (Fig. 2).
The actual stress versus crack opening displacement relationship depends on the material characteristics,
but it is not always univocally determined. For instance, for mode I fracture of concrete, a piecewise linear
relationship with a break-point, see Fig. 2, is getting increasing popularity [22–26] but alternative conti-
nuous formulations are also used; see, e.g., [27–32]. It is however expected that the actual shape of the
cohesive crack law is less and less influent on the overall response of a structure as its size and (conse-
quently) its brittleness increase, so that fracture energy becomes the dominant parameter and LEFM can be
applied.
An analytically simple softening expression is the exponential one which depends on the material tensile
strength and fracture energy only, as follows:
ru
G w
r ¼ ru e f ð5Þ

Relation (5) gives the value of the normal stress acting on the interface, C, in a progressive fracture
process, when w > 0. The whole description of the interface behavior can be given in the following compact
complementarity form:
ru
G wðnÞ
uðnÞ ¼ ru e f  rðnÞ P 0; wðnÞ P 0; uðnÞwðnÞ ¼ 0; n2C ð6Þ

where the so-called activation function uðnÞ has been introduced, as earlier done in [12,33–36]. Notice that,
according to (6), w > 0 implies u ¼ 0 while the condition of perfect adhesion w ¼ 0 implies r 6 ru .
The overall mechanical response of a quasi-brittle structure is determined by coupling the cohesive law
(6) with equilibrium, compatibility and constitutive equations written for the bulk material, often con-
sidered as linear elastic. In this hypothesis, superposition can be exploited to condense all the information
about the solid on the interface variables which govern the cohesive behavior, namely:
Z
E
rðnÞ ¼ r ðnÞ þ zðn; fÞwðfÞ df ð7Þ
C

where n, f are coordinates running along the locus of possible crack propagation, C (Fig. 1); zðn; fÞ is an
influence (Green) function; rE ðnÞ represents the interface response in the absence of displacement dis-
continuities, i.e. under an assumed elastic regime.
The structural problem formulated by the coupling of relations (6) and (7) is usually solved by numerical
techniques, based on finite element (FE) or boundary element (BE) methods, see e.g. [33–36].
Uplift forces acting along the crack and on the initial notch, are included by introducing some functional
dependence of the fluid pressure, say pðnÞ, on the opening displacement, for instance as follows:
1896 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

Fig. 3. Exponential pressure distribution in the cohesive crack model.

pðnÞ ¼ cw H ð1  eqwðnÞ Þ ð8Þ


Notice that q ¼ 0 returns the perfectly drained situation (pðnÞ ¼ 0), while the uniform (rectangular) pres-
sure distribution (pðnÞ ¼ cw H ) is obtained in the limit as q ! 1; q ¼ 1000 mm1 can be satisfactorily
introduced in this latter case for the purpose of numerical computations, see Fig. 3.
The hydraulic pressure p can be conceived as a pure mechanical loading on the crack surface, accounted
for by replacing, e.g. in (6), the normal stress component r with the stress r þ p acting on the solid skeleton
[12,37]. The term r þ p acquires therefore the meaning of ‘‘effective’’ stress, r being the corresponding
‘‘total’’ stress.
This approach is a clearly crude simplification from reality which is however currently accepted in dam
practice; see, e.g. [11,13,14,38].

2.3. Large scale size effects in quasi-brittle fracture

In quasi-brittle fracture, only an apparent fracture toughness can be defined, which is not a material
property but depends on the specimen size and shape and on the notch size. This apparent toughness can be
related to a nominal stress at the peak load as follows:
pffiffiffiffi
KINc ¼ rNc Df ðac Þ ð9Þ
Alternatively (see, e.g., [2,39,40]), the equivalent elastic crack concept can be introduced and relation (9) can
be rewritten in a format similar to (4):
pffiffiffiffi
KIc ¼ rNc Df ðaec Þ ð10Þ
where aec ¼ aec =D, aec being the so-called ‘‘equivalent’’ crack length which permits to relate the nominal
strength rNc to the toughness KIc , and hence to the fracture energy Gf (2), recognized as a material property,
independent of the specimen geometry.
The difference between the critical crack length ac and the equivalent crack length aec is expected to
become independent of the structural geometry and size as this latter becomes large enough, that is:
lim aec  ac ¼ cf ¼ blch ð11Þ
D!1

where cf represents the critical effective crack extension; b reflects the shape of the softening curve;
lch represents the so-called material characteristic length, which is a measure of the material ductility
and depends on the tensile strength, the fracture energy and the elastic modulus, as follows:
E 0 Gf
lch ¼ ð12Þ
r2u
G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906 1897

Notice that, clearly, due to the above definitions (11) and (12), aec ðaec =DÞ tends to ac ðac =DÞ as the structural
dimension D ! 1, thus implying the coincidence of relations (9) and (10) and of KINc and KIc . The concept
of effective crack extension is however essential to describe the structural response in a unified manner,
through relation (10), also at intermediate dimensions, for non-vanishing ratio cf =D.
A good quantitative approximation of the lower bound of the critical effective crack extension cf (11) has
been determined by Planas and Elices [39,40], as a function of the critical crack opening displacement wc :
p 2 E0
cf ¼ w ð13Þ
32 c Gf
wc representing the value of the critical crack opening at which cohesive traction vanishes, see Fig. 2.
Since cf can be considered a material constant for large structural size, function f ðaÞ can be replaced by
its linearization about ac by means of its 1st-order TaylorÕs series expansion, as follows:
cf
f ðaec Þ ffi f ðac Þ þ fc0 ð14Þ
D
c2f
Then, substituting (14) into (10), rearranging and neglecting the higher-order term D2
, it is possible to get:
KIc
rNc ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð15Þ
Dfc2 þ 2fc fc0 cf
being fc  f ðac Þ.
Clearly, as the material brittleness and/or the structural size increase, the ratio cf =D tends to zero and
relation (15) recovers the LEFM expression (4).

3. Case study

The practical usefulness of the alternative approaches briefly summarized in Section 2 has been checked
on a benchmark problem proposed by ICOLD (International Committee on Large Dams) [11].
A concrete gravity dam is modeled as a plane strain problem; the geometry of the structure, schematized
in Fig. 1, is defined by the following aspect ratios: B=D ¼ 3=4; d=D ¼ 1=16; a=B ¼ 1=10; D ¼ 80 m in the
ICOLD benchmark. The bulk material is supposed to be linear elastic, characterized by: YoungÕs modulus
E ¼ 24 GPa; PoissonÕs ratio m ¼ 0:15; density cc ¼ 24 kN/m3 . Crack propagation along the foundation
interface is hypothesized, starting from the initial notch, promoted by hydraulic pressure and resisted by the
self-weight of the structure beside by the interfacial tensile strength. An exponential decay of normal
traction with increasing opening displacements is assumed, as shown in Fig. 2, with tensile strength ru ¼ 1
MPa and fracture energy Gf ¼ 90 J/m2 .
Safety against imminent flood failure is to be evaluated; this factor can be quantified by the maximum
overtopping water level sustainable by the structure. Both the dam concrete and the foundation rock are
considered to be impervious; uplift pressure may eventually arise at the dam–foundation interface, due to
crack formation. Two extreme alternatives have been considered for uplift forces: either they act as a
rectangular distribution without any reduction of the hydrostatic value, as schematized in Fig. 1, or they are
zero everywhere as it would be in the presence of a perfect drainage system.

3.1. Linear elastic fracture analysis

The SIF KI relevant to the present study can be obtained by superimposing the effects of the following
basic load conditions: self-weight (SW), full reservoir hydrostatic pressure (FR), overtopping pressure (OT)
and uplift pressure (UL) in the notch:
1898 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

KI ¼ KISW þ KIFR þ KIOT þ KIUL ð16Þ


Eq. (16) can be rewritten in the format, analogous to (1):
pffiffiffiffi pffiffiffiffi pffiffiffiffi pffiffiffiffi
KI ¼ cc D Df SW ðaÞ þ cw D Df FR ðaÞ þ cw ðH  DÞ Df OT ðaÞ þ cw H Df UL ðaÞ ð17Þ
where cw represents the density of water and H > D indicates the water level in the case of flooding, see
Fig. 1.
Dimensionless shape functions for the simplified dam geometry drawn by dashed line in Fig. 1 have been
evaluated by Plizzari [13,14] who gave analytical expressions based on the interpolation of numerical re-
sults. The functions employed for the present study are given in Appendix A, and are drawn in Fig. 4. The
dimensionless crack length a ¼ a=B here, the ratio B=D being fixed for self-similarity.
Notice that the geometry function relevant to self-weight contribution f SW is negative, therefore the
combination (17) can return a non-monotonic function, as shown in Fig. 5 where KI is drawn versus a for
H ¼ D, i.e. for full reservoir in the absence of flooding. The horizontal dashed line indicates the material
toughness (i.e. the critical SIF value, KIc ’ 1470 kN/m3=2 ) resulting from the above assumed parameters.
Monotonic SIF increase is observed for a > 0:1 and pressurized fracture for several structural dimensions,
see Fig. 5(a). Observe that LEFM predicts that the structure is not able to withstand the full reservoir water
pressure for D ¼ 80. On the contrary, Fig. 5(b) shows that the SIF decreases as a increases in the absence of
uplift pressure (KIUL ¼ 0). In this case, the self-weight contribution dominates and promotes stable crack
growth along most part of the foundation interface; see, e.g., Fig. 6. Therefore, failure is expected for a-
values which are beyond the range of proven validity of the provided geometry functions (namely:
0:01 6 a 6 0:4 [13,14]; extrapolation up to a ¼ 0:5 has been admitted in the present work). No reliable
predictions can then be drawn in this situation.
In the case of pressurized fracture, condition (3) is satisfied at the initial notch length and the nominal
strength rNc can then be easily expressed in terms of the maximum sustainable overtopping water level,
Hmax  D, by equating the SIF obtained by (17) setting ac ¼ a0 ¼ 0:1 to the toughness (2). Results are
represented in Fig. 7 versus the size D (dam height).

3.2. Equivalent elastic analysis

LEFM is expected to give reliable quantitative results for large structural size, but predictions can be
improved in quasi-brittle situations by equivalent elastic analyses (EEA) based on the effective crack

10
SW
8 FR
geometry function f I

OT
6
UL
4

-2

-4
0 0.1 0.2 0.3 0.4 0.5
normalized crack length α

Fig. 4. Geometry functions after Plizzari [14] for self-weight (SW), full reservoir (FR), overtopping (OT) and uplift (UL) pressure.
G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906 1899

D = 80 m
(a)

KI [MN/m ]
2

3/2
60 m

1 40 m

20 m

0
0 0.1 0.2 0.3 0.4 0.5
normalized crack length α

0
0 0.1 0.2 0.3 0.4 0.5
KI [MN/m3/2]

20 m
-5
40 m
-10
60 m
-15
(b) D = 80 m
-20
normalized crack length α

Fig. 5. SIF distribution for some structural size, for the condition of full reservoir without overtopping (H ¼ D): (a) pressurized crack;
(b) no uplift contribution (ideal perfect drainage). The dashed horizontal line indicates the assumed critical value KIc .

10
H — D = 25 m

5
K I [MN/m ]
3/2

0
0 0.1 0.2 0.3 0.4 0.5
-5 H — D = 15 m

-10
H—D=5m

-15
normalized crack length α

Fig. 6. SIF distribution in the absence of uplift pressure, for increasing overtopping water level and structural dimensions D ¼ 60 (thin
lines) and D ¼ 80 (thick lines). The dashed horizontal line indicates the assumed critical value KIc .

concept introduced in Section 2.3. In this case, the ultimate value of the nominal strength is computed
according to relation (15), based on both the geometry functions and their derivatives besides the critical
effective crack extension, cf .
1900 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

100

LEFM

H max - D [m]
10

EEA LS
1

0.1
1 10 100 1000
D [m]

Fig. 7. Maximum overtopping water level versus the structural size as predicted by LEFM, by equivalent elastic analysis (EEA), by
simplified limit state predictions (LS) and by the cohesive crack model (rhombs), in the presence of uplift force.

As already said, the geometry functions are available in analytical form for the present application, while
the critical effective crack extension (cf ) can be replaced quite effectively by its lower bound, easily
obtainable as shown by (13) [39,40].
For computational convenience in view of the discrete cohesive crack analysis to be presented next, an
exponential softening law has been hypothesized. Therefore, a critical crack opening displacement at which
traction vanishes is not defined, and wc has been substituted by the opening displacement value at which the
residual strength reduces to 5% of the tensile strength; namely, it has been set:
Gf
wc ’ 3 ð18Þ
ru
With these assumptions and the assumed data set, listed above, the critical crack extension cf results equal
to about 2 m.
The results obtained by the EEA in the case of pressurized fracture are drawn in Fig. 7, which allows
immediate comparison with LEFM predictions. In particular, as expected, it can be observed that the two
curves tend to merge as the structural dimension D increases: for this application, EEA and LEFM pre-
dictions can be confused when D is larger than some 50 m, whence cf =D is smaller than 4%.

3.3. Cohesive crack analysis

The cohesive crack approach [22,25,30,32] is usually employed in numerical analyses based on FE or BE
modeling; see, e.g., [33–36]. The primary unknowns, i.e. the fields rðnÞ and wðnÞ defined over the interface
C, are then replaced by the corresponding discretization variables (usually nodal variables), say T and W,
interpolated through suitable functions collected by matrices Mw ðnÞ, Mt ðnÞ:
wðnÞ ¼ Mw ðnÞW; rðnÞ ¼ Mt ðnÞT ð19Þ
Interpolations Mw ðnÞ and Mt ðnÞ can be defined over C in interrelated way:
Z
MTw Mt dn ¼ I ð20Þ
C

so that:
Z Z
T¼ MTw ðnÞrðnÞ dn; W¼ MTt ðnÞwðnÞ dn ð21Þ
C C
G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906 1901

where I represents the identity matrix of order ðnD  nN Þ  ðnD  nN Þ, nD being the spatial dimension of the
problem (here nD ¼ 2) and nN the number of the discretization points (nodes) along the interface [33–36]
(nN ¼ 120, equally spaced, for the present application).
With these provisions, the discrete counterpart of the linear integral equation (7) reads:
T ¼ TE þ ZW ð22Þ
where T and TE acquire the meaning of nodal forces, while the influence matrix Z is symmetric and pre-
serves the essential features of the integral operator it approximates for FE and for symmetric Galerkin BE
discretization [33,36]. In a FE context, pursued here, matrix Z can be derived from the usual stiffness matrix
by condensation of all the out-of-C discretization variables; see, e.g., [35,36]. Analysis of self-similar (and
similarly discretized) structures can be easily performed since matrix Z is independent of the actual
dimensions; only vectors T and TE need to be scaled proportionally to D or to D2 .
Analogously, the discrete counterpart of the interface law (6) acquires the format:
U ¼ Tu ðWÞ  T P 0 WP0 UT W ¼ 0 ð23Þ
where Tu ðWÞ collects the integral form of the exponential function (5), weighted by means of the shape
functions Mw ðnÞ according to (21). The explicit expression of Tu ðWÞ is given in Appendix B.
When uplift pressure has to be taken into account, then relationship (8) can be substituted by its
weighted average over the whole interface, as done for the exponential function in (5), see Appendix B;
namely:
Z Z
T
PðWÞ ¼ Mw ðnÞpðwðnÞÞ dn ¼ MTw ðnÞpðMTw ðnÞWÞ dn ð24Þ
C C

Clearly, the components of the vector valued function PðWÞ are zero in correspondence of the bonded part
of the interface (w ¼ 0), where pðwÞ ¼ 0.
Vector T, see (23), can be then replaced by T þ PðWÞ, so that the coupling of equations (22) and
relations (23) returns the discretized version of the problem in point, in the format of a finite-dimensional
complementarity problem (CP):
U ¼ Tu ðWÞ  PðWÞ  T P 0 WP0 UT W ¼ 0 ð25Þ
Besides performing standard evolutionary non-linear analyses up to the maximum sustainable water
pressure, alternative direct methods can be used to determine the ultimate load carrying capacity of the
structure [11]. These techniques rest on the simplifying assumption of holonomic (i.e. path-independent)
progressive fracture, and can lead to accurate predictions at reduced computing costs. In particular, the
structural problem in point can be reformulated as a constrained optimization problem aiming at the direct
determination of the maximum load amplifier, as follows.
The assumed linear background allows to decompose the nodal force vector relevant to the interface C,
in the form:
b E þ H TE
TE ¼ T ð26Þ
and the contribution of the single loading conditions can be evidenced as:
OT
EOT EUL
b E ¼ TESW þ TEFR  DTE ;
T
E
T ¼T þT ð27Þ
ESW EFR
where T and T represent, respectively, the nodal force vectors resulting from the dead-load contri-
bution and from the triangular water pressure distribution acting on the upstream wall of the dam in the
EOT EUL
case of full reservoir, while T and T represent the nodal force vectors resulting from a uniform
distribution of overtopping and uplift unit pressure.
1902 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

Analogously, see (8) and (24):

PðWÞ ¼ H PðWÞ ð28Þ

The holonomic problem given by the combination of (22) and (25) can then be written as:

b E  H TE  ZW P 0
UðH ; WÞ ¼ Tu ðWÞ  H PðWÞ  T WP0
T
UðH ; WÞ W ¼ 0 ð29Þ

The maximum water level, Hmax , can be obtained by the direct solution of the following optimization
problem under complementarity constrains:
T
Hmax ¼ minfH jUðH ; WÞ W ¼ 0; UðH ; WÞ P 0; W P 0g ð30Þ
H ;W

The specific format of the above optimization problem (30) is the subject of current research in
mathematical programming, see e.g. [41], but its solution in the present context can be sought by commonly
available optimization software, as shown in [12].
The results derived through this approach for some characteristic structural dimension D in the case of
pressurized fracture are summarized in Fig. 7 where the maximum overtopping water level obtained as
solution of the optimization problem (30) is indicated by means of markers (rhombs). These values are close
to the formerly obtained ones up to a threshold value of D roughly equal to 60 m.
The whole structural response in terms of overtopping water level versus the corresponding crack mouth
opening displacement (CMOD) and versus the normalized crack length is represented in Figs. 8 and 9,
respectively, for some characteristic dimensions D chosen astride the above indicated threshold value. These
results have been obtained by solving the holonomic complementarity problem (29) for different H values,
and visualize the alternative failure mechanism which are returned by the calculations. In particular, it can
be seen that instability under load control, leading to catastrophic failure, is clearly met after a short crack
propagation for relatively small structural sizes (e.g., D ¼ 50 m), while for large D values (e.g., D ¼ 80 m)
crack propagation starts before the reservoir is completely filled up, but advancement requires increasing
water levels. Intermediate behavior is shown by intermediate structural dimensions: a phase of stable
propagation follows the first, more or less pronounced, peak of the overtopping water level.
The CP (25) can easily accommodate the case of zero uplift forces, setting PðWÞ ¼ 0. The maximum
sustainable overtopping water level, obtained by the solution of the corresponding optimization problem
(30), are drawn in Fig. 10 (circles) for some structural size D. A threshold value discriminating the structural

D = 50 m
H-D [m]

1
60 m

70 m
80 m

0
0 2 4 6
CMOD [mm]

Fig. 8. Overtopping water level versus crack opening displacement for different structural size.
G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906 1903

D = 50 m

H-D [m]
1 60 m

70 m 80 m

0
0.1 0.4 0.7 1
normalized crack length α

Fig. 9. Overtopping water level versus normalized crack length for different structural size.

100

LS
EEA
Hmax - D [m]

10

LS
α cr
1.2
1
0.8

0.4

0 D [m]
1 10 100 1000
0.1
1 10 100 1000
D [m]

Fig. 10. Maximum overtopping water level versus the structural size as predicted by the cohesive crack model in the presence (rhombs)
and in the absence (circles) of uplift forces. The normalized crack length at maximum load resulting from the discrete crack approach is
visualized in the insert. The straight lines represent simplified limit state predictions.

responses in terms of different failure mechanism can be clearly identified again, the transition size being
now D  15 m.

3.4. Discussion

The results of the application of the alternative approaches and computing tools introduced in previous
Sections to the same benchmark problem give rise to the following remarks.
LEFM and EEA have been applied to the case of pressurized fracture setting ac ¼ a0 in relationships (4)
and (15) since condition (3) is fulfilled, see the SIF distribution represented in Fig. 5(a), and unstable
propagation is therefore expected starting from the initial crack. Predictions of the structural strength, in
terms of the maximum sustainable overtopping water level, are compared in Fig. 7. The relevant results are
qualitatively and quantitatively similar in the large size range while, as expected, the sustainable water level
evaluated by EEA is more conservative for smaller dimensions. A dramatic drop of the load carrying
capacity under the effect of uplift forces is predicted by both approaches when the dam height is above some
60 m.
1904 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

The results of the cohesive-crack approach (discrete crack), obtained as the solution of the constrained
minimum problem (30) for a discrete set of the structural sizes D, are also shown for comparison in Fig. 7,
by rhombs. A good match between the computed maximum overpressure (in water meters) and the EEA
prediction is met at intermediate scale, but an opposite trend is observed as D increases; in particular, the
discrete-crack analyses return increasing load multiplier at failure as D > 60 m. This result can be attributed
to the growing stabilizing role of the self-weight contribution which promotes stable crack growth along
most part of the interface, as shown by Figs. 8 and 9.
The straight lines in the log–log plots of Figs. 7 and 10 represent safety factors predicted by simplified
limit state (LS) analyses, based on the consideration of the rigid-body momentum around the downstream
base-point Q (Fig. 1) of the external forces acting on the dam, consisting of self-weight and hydrostatic
pressure only (no cohesion). The maximum overtopping water level (namely, 0.59 D) is thus found for no
fluid penetration (dashed line in Fig. 10), while the minimum (0.0184 D) results for uniform (rectangular)
distribution of uplift pressure all along the foundation interface (continuous line in Figs. 7 and 10).
For large structural size, the cohesive crack model predicts longer cracks along the foundation interface,
permeated by water pressure, while crack propagation occurs in a progressive stable way: this explains the
reason why a fairly good estimation of the structural strength can be obtained by LS predictions based on
rigid-body idealization of the structure, which are widely used in dam engineering and accepted as design
rules in some national standards, e.g. [38].
The normalized crack length at the maximum load, acr , evaluated through the cohesive crack model, is
plotted versus the characteristic dimension D in the insert of Fig. 10. Observe that acr initially reduces as the
dam height D increases, thus suggesting an increase of the structural brittleness; however, beyond some
structural size which depends on the possible presence of uplift forces, stable fracture propagation occurs
along most part of the interface and acr ’ 1 at failure.

4. Summary and conclusion

Alternative computing methods for evaluating the structural strength of an idealized gravity dam of
different structural sizes have been considered to the purpose of checking the range of applicability of the
size effect law to this kind of structures and the consequent possibility of transferring the results of
experiments on prototypes to the actual design, despite some unavoidable discrepancies related to the
different characteristics of the materials used in laboratory and in construction.
The results of this investigation show that information can be effectively transferred though relationships
based on EEA up to some intermediate size while, for larger dimensions which are more typical in dam
construction, simplified limit state predictions are more suited to the purpose. This transitional size sep-
arates structural situations characterized by different failure mechanisms: either propagation which be-
comes unstable close to crack initiation, or progressive fracture advancement under increasing load up to
complete material separation, which reflects the dominant effect of gravity.

Acknowledgements

The author wishes to thank Prof. G. Maier for stimulating discussions, and A. Pasquazzo and G.
Cocchetti for their help in some preliminary calculations. This work has been carried out within the Cofin02
research project ‘‘Concrete dam–foundation–reservoir systems: integrity assessment allowing for interac-
tions’’. Financial support by the Italian Ministry of Education and Research (MIUR) is gratefully
acknowledged.
G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906 1905

Appendix A. Geometry functions

Dimensionless geometry functions for evaluating the SIF contribution due to self-weight, full reservoir,
overtopping and uplift pressure for the current geometry (B=D ¼ 0:75) have been obtained by Plizzari
[12,13] interpolating obtained numerical results in the range 0:01 6 a 6 0:4 (a ¼ a=B, see Fig. 1):
2 1 1
f SW ¼ ð0:32255 þ 1:03623 ln a þ 0:087235ðln aÞ Þ ; f FR ¼ ð0:970159  1:41421aÞ
 1
pffiffiffi 1 0:928425
f OT ¼ ð0:411525  0:437345 aÞ ; f UL ¼ pffiffiffi  1:01127 ðA:1Þ
a
The above functions are drawn in Fig. 4.

Appendix B. Discretized exponential law

The exponential function introduced by relation (5):


ru
G wðnÞ
ru e f ðB:1Þ
can be given the following analytical discretized version when the opening displacement wðnÞ is interpolated
by linear shape functions, Mw ðne Þ, defined over each partition Ce of the interface C (see Fig. 1) as follows:
ne le  n e
wðne Þ ¼ Mw ðne ÞW ¼ WA þ WB 0 6 n e 6 le ðB:2Þ
le le
WA and WB being the discretization variables (opening displacements) at nodes A and B, respectively.
The weighted average over C of the function (B.1), Tu ðWÞ, is defined according to (21):
Z X Z
ru r n l n
Mw ðne Þ e Gf ð le A le B Þ dne
u T G wðnÞ T  u eW þ e eW
T ðWÞ  Mw ðnÞ ru e f dn ¼ ru ðB:3Þ
C e Ce

u
The expression T ðWÞ can be obtained explicitly by the assembly of the following integrals which result
from the substitution of (B.1) and (B.2) into (B.3):
Z le ( r
 u ðWB WA Þ ru
e Gf 1 Gf WB
r u ne
G ð WA 
le ne
WB Þ r u l e e if WB 6¼ WA
ru e f e l l e dne ¼ ru
WB WA ðB:4Þ
G WB
0 ru le e f if WB ¼ WA
8   r
 u ðWB WA Þ
Z le
>
< ru
1 1G ðWB WA Þ e Gf
ru
ru r u ne
G ð l n
WA  e le e WB Þ dn ¼ ru le
f
e
G WB
if WB 6¼ WA
ne e f le
e ðWB WA Þ2
f
ðB:5Þ
le 0 >
:1 ru
G WB
r le
2 u e
f if WB ¼ WA

References

[1] Bazant ZP. Scaling of structural strength. London, UK: Hermes Penton Science; 2002.
[2] Bazant ZP, Planas J. Fracture and size effect in concrete and other quasibrittle materials. Boca Raton, Florida: CRC Press; 1997.
[3] Guinea GV, Elices M, Planas J. Assessment of the tensile strength through size effect curves. Engng Fract Mech 2002;65:189–207.
[4] van Mier JGM, van Vliet MRA. Uniaxial tension test for the determination of fracture parameters of concrete: state of the art.
Engng Fract Mech 2002;69:235–47.
[5] van Vliet MRA, van Mier JGM. Experimental investigation of size effect in concrete and sandstone under uniaxial tension. Engng
Fract Mech 2000;65:165–88.
1906 G. Bolzon / Engineering Fracture Mechanics 71 (2004) 1891–1906

[6] Dempsey JP, Defranco SJ, Adamson RM, Mulmule SV. Scale effects on the in-situ tensile strength and fracture of ice. Part I:
Large grained freshwater ice at Spray Lakes Reservoir, Alberta. Int J Fract 1999;95:325–45.
[7] Dempsey JP, Defranco SJ, Adamson RM, Mulmule SV. Scale effects on the in-situ tensile strength and fracture of ice. Part II:
first-year sea ice at Resolute, N.W.T. Int J Fract 1999;95:347–66.
[8] Renzi R, Ferrara G, Mazza G. Cracking in a concrete gravity dam: a centrifugal investigation. In: Bourdarot E, Mazars
J, Saouma V, editors. Dam fracture and damage. Rotterdam: Balkema; 1994. p. 103–10.
[9] Plizzari G, Waggoner F, Saouma VE. Centrifuge modelling and analysis of concrete gravity dams. ASCE J Struct Engng
1995;121:1471–9.
[10] Barpi F, Valente S. Numerical simulation of prenotched gravity dam models. ASCE J Engng Mech 2000;126:611–9.
[11] ICOLD. Theme A2: imminent failure flood for a concrete gravity dam. Denver CO: Fifth International Benchmark Workshop on
Numerical Analysis of Dams, 2–5 June 1999.
[12] Bolzon G, Cocchetti G. Direct assessment of structural resistance against pressurized fracture. Int J Numer Anal Meth Geomech
2003;27:353–78.
[13] Plizzari GA. LEFM applications to concrete gravity dams. ASCE J Engng Mech 1997;123:808–15.
[14] Plizzari GA. On the influence of uplift pressure in concrete gravity dams. Engng Fract Mech 1998;59:253–67.
[15] Tada H, Paris PC, Irwin GR. The stress analysis of cracks handbook. St. Louis, Missouri: Paris Productions Incorporated; 1985.
[16] Murakami Y, editor. Handbook of stress intensity factors. Oxford: Pergamon Press; 1987.
[17] Matsumto T, Tanaka M, Obara R. Computation of stress intensity factors of interface cracks based on interaction energy release
rates and BEM sensitivity analysis. Engng Fract Mech 2000;65:683–702.
[18] Wang Q, Noda N-A, Honda M-A, Chen M. Variation of stress intensity factor along the front of a 3D rectangular crack by using
a singular integral equation method. Int J Fract 2001;108:119–31.
[19] Bolzon G. An approximate method for fatigue-life prediction of framed structures. Fatigue Fract Engng Mater Struct
1996;19:1481–91.
[20] Guinea GV, Pastor JY, Planas J, Elices M. Stress intensity factor, compliance and CMOD for a general three-point-bend beam.
Int J Fract 1998;89:103–16.
[21] Guinea GV, Planas J, Elices M. KI evaluation by the displacement extrapolation technique. Engng Fract Mech 2000;66:243–55.
[22] Wittmann FH, Hu X. Fracture process zone in cementitious materials. Int J Fract 1991;51:3–18.
[23] CEB-FIP. Model Code 1990: design code. London: Thomas Telford Ltd.; 1993.
[24] Phillips DV, Binsheng Z. Direct tension test on notched and un-notched plain concrete specimens. Mag Concr Res 1993;45:25–35.
[25] Guinea GV, Planas J, Elices M. A general bilinear fit for softening curve of concrete. Mater Struct 1994;27:99–105.
[26] Bolzon G, Maier G, Novati G. Some aspects of quasi-brittle fracture analysis as a linear complementarity problem. In: Bazant ZP,
Bittnar Z, Jir
asek M, Mazars J, editors. Fracture and damage in quasi-brittle structures. London: E&FN Spon; 1994. p. 159–74.
[27] Cornelissen HAW, Hordijk DA, Reinhart HW. Experiments and theory for the application of fracture mechanics to normal and
lightweight concrete. In: Wittmann FH, editor. Fracture toughness and fracture energy. Elsevier; 1986. p. 565–75.
[28] Foote RML, Mai YW, Cotterel B. Crack growth resistance curves in strain-softening materials. J Mech Phys Solids 1986;34:
593–607.
[29] Cedolin L, dei Poli S, Iori I. Tensile behaviour of concrete. ASCE J Engng Mech 1987;113:431–9.
[30] Planas J, Elices M. Nonlinear fracture of cohesive materials. Int J Fract 1991;51:139–57.
[31] Carol I, Prat PC, L opez CM. Normal/shear cracking model: application to discrete crack analysis. ASCE J Engng Mech
1997;123:1–9.
[32] Elices M, Guinea GV, G omez J, Planas J. The cohesive zone model: advantages, limitations and challenges. Engng Fract Mech
2002;69:137–63.
[33] Maier G, Novati G, Cen ZZ. Symmetric Galerkin boundary element method for quasi-brittle fracture and frictional contact
problems. Comput Mech 1993;13:74–89.
[34] Maier G, Frangi A. Symmetric boundary element method for discrete crack modelling of fracture processes. Comp Assist Mech
Engng Sci 1998;5:201–26.
[35] Bolzon G. Hybrid finite element approach to quasi-brittle fracture. Comp Struct 1996;60:733–41.
[36] Bolzon G, Corigliano A. A discrete formulation for elastic solids with damaging interfaces. Comp Meth Appl Mech Engng
1997;140:329–59.
[37] Slowik V, Saouma VE. Water pressure in propagating concrete cracks. ASCE J Struct Engng 2000;126:235–42.
[38] D.M.24.03.82. Norme tecniche per la progettazione e la costruzione delle dighe di sbarramento. G.U. n. 212, 04.08.82.
[39] Planas J, Elices M. Asymptotic analysis of a cohesive crack: 1. Theoretical background. Int J Fract 1992;55:153–77.
[40] Planas J, Elices M. Asymptotic analysis of a cohesive crack: 2. Influence of the softening curve. Int J Fract 1993;64:221–37.
[41] Luo ZQ, Pang JS, Ralph D. Mathematical programs with equilibrium constraints. Cambridge, UK: Cambridge University Press;
1996.

You might also like