Sie sind auf Seite 1von 9

Available online at www.sciencedirect.

com

Advances in Engineering Software 39 (2008) 249257


www.elsevier.com/locate/advengsoft

A numerical model for the description of the nonlinear behaviour


of multi-leaf masonry walls
Marcio Antonio Ramalho a, Alberto Taliercio b,*, Anna Anzani b,
Luigia Binda b, Enrico Papa c
a

S. Carlos School of Engineering, USP, Av. Trabalhador Sancarlense, 400, 13566-590 Sao Carlos (SP), Brazil
Department of Structural Engineering, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milan, Italy
Department of Environment, Health and Safety, Universita` dellInsubria, Via J.H. Dunant 3, 21100 Varese, Italy

b
c

Available online 15 February 2007

Abstract
A nonlinear nite element model was developed to simulate the nonlinear response of three-leaf masonry specimens, which were subjected to laboratory tests with the aim of investigating the mechanical behaviour of multiple-leaf stone masonry walls up to failure. The
specimens consisted of two external leaves made of stone bricks and mortar joints, and an internal leaf in mortar and stone aggregate.
Dierent loading conditions, typologies of the collar joints, and stone types were taken into account. The constitutive law implemented in
the model is characterized by a damage tensor, which allows the damage-induced anisotropy accompanying the cracking process to be
described. To follow the post-peak behaviour of the specimens with sucient accuracy it was necessary to make the damage model nonlocal, to avoid mesh-dependency eects related to the strain-softening behaviour of the material. Comparisons between the predicted and
measured failure loads are quite satisfactory in most of the studied cases.
 2007 Elsevier Ltd. All rights reserved.
Keywords: Multi-leaf masonry; Testing; Damage; Anisotropy; Finite elements; Non-local damage; Mesh-sensitivity

1. Introduction
The historical-monumental and residential patrimony
diused over the Italian territory, declared seismic by the
National Code, as well as in the rest of Europe, mostly consists of masonry constructions, often made of multiple-leaf
stone walls dating back to the Roman times, to the middleage (see Fig. 1), and to later times, up to the XIX century.
According to an extensive survey of the masonry cross
sections carried out in dierent Italian regions, the most
common typologies turned out to be single-, double- and
triple-leaf walls, characterized by various types of mutual
connections ranging from simple juxtaposition to good
interlocking. The partial or total collapse of a wall is
hampered by the eectiveness of the connections, that is,
*

Corresponding author. Tel.: +39 02 23 99 42 41; fax: +39 02 23 99 42

20.
E-mail address: alberto.taliercio@polimi.it (A. Taliercio).
0965-9978/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advengsoft.2007.01.003

the capacity of yielding a monolithic response under outof-plane actions by preventing the leaves from detaching
easily. Some striking events which took place in seismic
areas (damage of historical centres after an earthquake in
Umbria and Marche in 1997 [1]; collapse of the Cathedral
of Noto, Sicily, in 1996 [2]) further highlighted the importance of understanding the mechanical behaviour of similar
structures, the vulnerability of which to seismic actions can
be enhanced by creep-induced damage originated by elevated dead loads. This was actually the case with the pillars
of the Cathedral of Noto, which collapsed six years after
being hit by an earthquake without revealing any serious
evidence of damage. Only after plaster removal, damage
became visible, in the form of deep vertical cracks in the
outer limestone leaf, whereas the rubble inll material,
made of pebbles and a very weak mortar, appeared to be
scarcely damaged [2].
Following a previous research on the behaviour of
multiple-leaf walls [3], an experimental investigation was

250

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257

Fig. 1. Examples of multi-leaf masonry walls: (a) St. Antimo Abbey, Tuscany, Italy and (b) Durham Castle, Yorkshire, UK.

carried out in the laboratory to better understand the


mechanical response of this kind of structures [4]. A number of specimens were expressly built, with the aim of
reproducing the non-homogeneous stone masonry of
which the pillars of the Cathedral of Noto were made,
and subjected to mechanical tests: some details on the testing program are given in Section 2.
According to the high degree of heterogeneity of multiple-leaf masonry, made of irregularly shaped stones and
mortar joints, the development of reliable mathematical
models to predict its behaviour under vertical and horizontal actions is known to be a dicult task [3]. Here, a numerical model is proposed based on a nonlinear 3D constitutive
law allowing for mechanical damage, assuming each layer
to be homogeneous (Section 3). The experimental tests
performed on the three-leaf specimens were simulated by
prescribing suitable displacement boundary conditions to
a nite element model. In Section 4, the main numerical
results obtained are presented and compared with the
experimental ones.
2. Description of the experimental tests
The materials, the construction technique, the morphology and the dimensions of the specimens were selected so
as to match multiple-leaf walls and piers like those constituting the Cathedral of Noto and other churches in Sicily.
Two external leaves, made of regularly cut stone units with
horizontal and vertical mortar joints, are connected to an
internal leaf in rubble masonry, made of pebbles of the
same kind of stone, merged into mortar. Twelve three-leaf
wallettes, 310 510 790 mm3 in size, were built (Fig. 2);
two types of connection between the leaves were adopted
(namely, at and keyed collar joints) in order to study
the inuence of the connections on the wall behaviour.
In addition to the Noto stone (a sandstone tu), specimens made of a medium grained sandstone (Pietra Serena),
characterized by a higher strength and a smaller porosity,
were also prepared. The mechanical characteristics of the
stones used for the preparation of the samples, obtained

Fig. 2. Typical geometry of the tested walls: (a) keyed collar joints and (b)
at collar joints.

on cylindrical specimens 80 mm in diameter and 160 mm


in height, are reported in Table 1.
The same ready-to-mix hydraulic lime mortar was used
to build both the external and the internal leaves of all the
masonry specimens.
Two kinds of mechanical tests were performed: compression tests, with all the leaves loaded at the top and
restrained at the base (Figs. 3a,b), and (pseudo-) shear
tests, with the central leaf loaded at the top and the base
of the external leaves restrained (Fig. 3c). Compression
tests on specimens with at and keyed collar joints will
be labelled type 1 and type 2, respectively. Shear tests
on walls with keyed collar joints will be labelled type 3.
Shear tests on walls with at collar joints caused the separation of the leaves at their interface at very low loads,
and will not be considered in the continuation. These leaves

Table 1
Properties of the stones used for the preparation of the samples
Stone

Load vs.
bed joint direction

E
(N/mm2)

rc
(N/mm2)

rt
(N/mm2)

Noto
Noto
Serena
Serena

Normal
Parallel
Normal
Parallel

9476
8526
18218
23293

0.10
0.09
0.19
0.21

20.6
17.6
104.2
89.0

_
2.06
_
6.07

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257

251

general case, orthotropic. In nite form, the nonlinear


stressstrain law of the material reads:
e CD : r;

Fig. 3. Types of loading conditions: (a) compression test, at collar joints,


(b) compression test, keyed collar joints and (c) shear test, keyed collar
joints.

were individually tested in compression to determine their


average mechanical properties.
Compression tests on walls in Serena stone did not bring
the specimens to failure, as their bearing capacity exceeded
the maximum load applicable by the testing machine
(Pmax = 2500 kN see Figs. 5b and 6b). The experimental
loaddisplacement plots obtained in the tests are shown in
Section 4, together with the results of the numerical simulations. More details on the mechanical properties of the
materials employed and the experimental results can be
found elsewhere [4].
3. A damage model for brittle materials
3.1. Local damage model
To simulate and analyze the experimental tests briey
described in Section 2, a damage model was employed
which was originally developed by some of the authors to
interpret the time evolution of mechanical damage in brittle
materials such as concrete and masonry, under either
increasing or sustained stresses of high intensity [5]: readers
are referred to the original paper for details on the model.
The model was implemented in two nite element codes
(ABAQUS and FEAP), both of which allow users to
employ particular constitutive laws through expressly
developed subroutines. Accordingly, nonlinear structural
analyses of two ancient masonry towers were performed,
which gave useful information on the causes of the collapse
of one of the towers [6] and the current state of damage in
the other one [7,8].
The essential features of the model, when applied to
stress analyses involving quickly increasing loads and negligible creep eects, are as follows: at the virgin (undamaged) state, any material element is supposed to be
isotropic, linearly elastic: the Youngs modulus and the
Poissons ratio of the material are denoted by E and m,
respectively. The damage phenomena are macroscopically
taken into account through a symmetric, second-order tensor D; accordingly, the damaged material is, in the most

where C = the fourth-order exibility tensor of the damaged material. The eigenvalues and the normalized eigenvectors of the damage tensor will be denoted by Da and
na (a = I, II, III), respectively. Note that nI, nII and nIII
are the unit vectors normal to the local planes of material
symmetry of the damaged material. Any one of the planes
of damage-induced orthotropy is somehow associated to a
plane microcrack that forms in the solid. Once any damage
direction is activated, its orientation is supposed to remain
xed throughout the rest of the stress history. Thus, the
ensuing model can be qualied as a non-rotating, smeared
crack model.
In this model, the damage process driving variable is
supposed to be an equivalent strain measure, y =
1/2 (eel)2. As the maximum eigenvalue of y attains a critical
value (y0T or y0C, according to the sign of the associated
strain), the rst damage direction (nI) is activated. An additional damage direction, nII, can activate in the plane
orthogonal to nI if the maximum direct component of y,
that is, yhh = nh (y nh), with nh ? nI, attains the damage
threshold. The third possible damage direction is necessarily nIII = nI ^ nII.
In the case of increasing stresses, each principal value of
the damage tensor is supposed to evolve according to a law
similar to that presented in [5] for concrete:
!
1
; a I; II; III:
2
Da C H 1 
B
1 AH hy hh  y 0H i H
Here, h*i are McAuley brackets and AH, BH and CH are
material parameters, which take dierent values according
to the sign of the strain component that activates damage
(H = T for tension; H = C for compression). In the model
version implemented in ABAQUS, CC = CT = 1.
Note that permanent (plastic) strains are disregarded in
the proposed version of the model. Also, crack-closure
eects (that is, the recovery of stiness upon closure of
an existing crack due to a stress reversal) are neglected.
These are not deemed to be serious limitations as far as
the model is applied to the simulation of tests involving
monotonically increasing loads or displacements. If the
local stress rate changes in sign in signicant regions of
the analyzed structure, the model results are likely to be
unreliable, unless the above simplications are removed.
For nite element applications, it is expedient to have
the constitutive law in matrix form. Let {e} and {r} be
the arrays gathering the six engineering (innitesimal)
strains and stresses, referred to any (global) Cartesian reference frame (x, y, z):
T

feg fex ; ey ; ez ; cxy ; cyz ; czx g ;


T

frg frx ; ry ; rz ; sxy ; syz ; szx g :

3a
3b

252

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257

Also, let f^eg and f^


rg be the homologous arrays collecting the stress and strain components referred to the reference frame of the local principal directions of damage.
These arrays are related to {e} and {r} by
feg qT f^eg;

frg q1 f^
rg;

where
2

nxI 2

6 x 2
6 nII
6
6 x 2
6 n
q 6 III
6 nx nx
6 I II
6 x x
4 nII nIII
nxIII nxI

nyI 2

nzI 2

nzII

nyII

2nxI nyI

2nyI nzI

2nxII nyII

2nyII nzII

nyIII 2

nzIII 2

2nxIII nyIII

2nyIII nzIII

nyI nyII

nzI nzII

nxI nyII nyI nxII

nyI nzII nzI nyII

nyII nyIII
nyIII nyI

nzII nzIII
nzIII nzI

nxII nyIII nyII nxIII


nxIII nyI nyIII nxI

nyII nzIII nzII nyIII


nyIII nzI nzIII nyI

with nia = cos(xi, xa), i = 1, 2, 3; a = I, II, III (see e.g., [9]).


At a given time, ti, the strain array in the global reference frame is related to the current stress by
fei g C i fri g;

where [C(i)] = the current secant exibility matrix of the


damaged material element. Similarly, the stress and strain
arrays in the local reference frame at time ti are related by
^ i f^
f^ei g C
ri g;

where
2
6
6
6
6
6
6
1
^
C i  6
E6
6
6
6
6
4

1
wI;I

m
wI;II

m
wI;III

1
wII;II

m
wII;III

1
wIII;III

21m
wI;II

symm:

0
21m
wII;III

7
0 7
7
7
0 7
7
7
0 7
7
7
7
0 7
5

21m
wI;III

tti1

8
with wa,b = [(1Da) (1Db)]1/2 (a, b = I, II, III). This
matrix can be rotated to the global Cartesian reference
frame (x, y, z) to get the exibility matrix involved in Eq.
(6) using the classical transformation rule:
^ i q:
C i  qT C

3.2. Non-local damage model


In local damage procedures, the consistent damaged
tangent or secant matrix is evaluated considering the local
strain-state. In nite element based procedures (e.g.,
FEAP), this means to evaluate the matrix according to
the strain at an integration point. However, for problems
in which the strain eld is not constant and the material
exhibits strain softening, this leads to spurious results and

strong mesh sensitivity (see e.g., [10,11]). Typically, the


inelastic strains are concentrated in narrow bands, whereas
the major part of the structure remains nearly unstrained.
This is illustrated in Figs. 8a and 9a, where meshes with elements of the same height before loading are shown after
deformation: strains are apparently localized in some elements, or rows of elements.
2nzI nxI

7
7
7
7
2nzIII nxIII
7
7
z x
x z 7
nI nII nI nII 7
7
nzII nxIII nxII nzIII 5
2nzII nxII

nzIII nxI nxIII nzI


There are two general groups of non-local procedures to
avoid this strain localization. In the so-called integral-type
models [12,13], a non-local state variable is computed as
the weighted average of the local state variable over a
neighborhood of the point being examined. On the other
hand, in gradient-type models [14,15], higher-order
(usually, second-order) derivatives are added to the partial
dierential equation that describes the evolution of the
non-local variable. Both approaches lead to similar results
and, in most cases, can be considered to be equivalent.
In this paper, the non-local procedure implemented in
the nite element code FEAP is based on an integral-type
procedure. In fact, several quantities may be transformed
into the corresponding (integral) non-local quantities
(e.g., the damage parameter itself, the strain and, more
recently, the displacement eld), and a number of proposals can be found in the literature [16]. In this work, a
weighted strain-state averaging proceeding was selected,
as it can be easily implemented in a nite element code.
A displacement averaging procedure would be preferable
if the nite element had the number of Gauss points greater
than the number of nodes [16]. However, it is not the case
here, as eight-node 3D nite elements with full integration
are employed, and both procedures would be equivalent
[17].
In this paper, the mentioned strain averaging procedure
is performed as follows:
R
f x  xs exs dV
ea x V R
;
10
f x  xs dV
V
where ea is the average strain at the current point x,
f (x  xs) is a weighting function that depends on the distance from x to xs, e(xs) is the strain at any source point
xs, and V is the volume of the dened neighbourhood.
The weighting function f (x  xs) can vary from one
application to another, but it is often taken as [11,18]:
 

x  xs 2
f x  xs exp 
;
11
2k

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257

where k is the internal length of the non-local neighbourhood, representing the diameter of a sphere centered at
the current point x. The value of k to be employed in
non-local damage models is not a trivial choice (see e.g.,
[19]). For concrete, there is a certain agreement upon taking three times the diameter of the largest aggregate. No
indication was found in the literature for stone masonry,
and some theoretical work about this topic should be performed. Thus, the choice made in the present work
(k = 100 mm for Noto stone and 60 mm for Serena stone)
is quite arbitrary.
There are a few basic requirements that the averaging
procedure has to meet. First, if the local strain would be
always simply replaced by the average strain this could
aect incorrectly the results as long as the material behaviour is still in the elastic range. So it is important that averaging is performed only when the material is actually
damaged. Secondly, the model should give a realistic
response even in simple loading situations, such as uniaxial
tension or compression. Then, being the strain eld constant, the stresses should be constant as well.
In nite element analyses, strains and stresses are usually evaluated at the Gauss points. Taking this fact into
account, one can write:
P
xj 2V f xi  xj xxj J xj exj
ea xi P
;
12
xj 2V f xi  xj xxj J xj
where x(xj) is the Gauss integration weight at point j
and J(xj) is the Jacobian of the isoparametric transformation evaluated at point j. According to Eq. (12), it is possible to evaluate the average strain at any Gauss point i
based on the strain eld values at neighbouring Gauss
points.
As FEAP identies that the user material model for
non-local damage is being used, it calls a routine that performs a loop over all the active elements in order to identify, for each element, its neighbourhood according to
two dierent procedures. The rst one, performed when
the neighbourhood radius (RNL) is null, identies all the
elements that have common nodes with the current element. The second, performed when RNL is not null, identies all the elements that have the centre inside a sphere of
radius RNL that has the origin at the centre of the current
element. For both procedures, an array is allocated in
order to store this information.
When the material model routine is called (for each element and for each Gauss point), it receives the local strain
evaluated in the usual way by the routines of FEAP. Then,
inside the material model and just when a damage direction
is activated, the new non-local strains are evaluated according to Eq. (12) and used to compute the secant matrix.
There are two options for the weighting function. The rst
one is Eq. (11); the second one is f = 1 (constant), which
means to perform an arithmetic averaging over the neighbourhood. The second option may be important for analyzing simple cases, like uniaxial loading.

253

It is worth to emphasize that the non-local strains


should be used only to evaluate the tangent or secant
matrix. However, to compute the stresses it is necessary
to consider the local strains, because they must be compatible with the real displacement eld; otherwise, convergence would not be achieved or, at least, would be hard
to obtain.
4. Numerical results comparisons with experiments
The nite element mesh employed in the numerical analyses is shown in Fig. 4 and consists of 2880 eight-noded isoparametric brick elements: only one quarter of the walls
was discretized, accounting for symmetry boundary conditions at the vertical mid-planes of the specimens. Only displacement boundary conditions were prescribed, with the
lower base of the model being completely or partially
restrained, and the upper base undergoing increasing displacements. Similarly to [4], each leaf was assumed to be
homogeneous, thus neglecting the horizontal and vertical
mortar joints between two adjacent stones. The external
leaves were given average mechanical properties derived
from tests on the individual leaf. The model parameters
employed in the analyses are summarized in Table 2. The
leaves are assumed to be perfectly bonded, except for one
of the analyses (see below).
In Figs. 57, experimental and numerical loaddisplacement curves are compared, for tests type 1, 2 and 3, respectively. The origin of the numerical plots was shifted to
discard the initial stage of the laboratory tests, where an
adjustment takes place between machine platens and
specimens.
For Type 1 tests, FEAP and ABAQUS yield similar
results, as can be seen in Fig. 5. In the simulation of the
tests on the specimen in Noto stone it is possible to observe
that the peak-load and the corresponding displacement are
predicted with a very good degree of accuracy. Regarding
the post-peak softening behaviour, however, only FEAP,
where a non-local damage procedure is implemented, yields
good results. ABAQUS, on the contrary, predicts an excessively brittle response because of the local nature of the

Fig. 4. Finite element mesh employed in the numerical analyses; the


discretized inner and outer leaves are also separately shown in the case of
specimens with keyed collar joints.

254

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257

Table 2
Model parameters employed in the FE analyses
Noto stone

Serena stone

outer leaf

inner leaf

outer leaf

inner leaf

E (MPa)

FEAP
Abaqus

2200
2500

1100
1800

3800
3800

1300
1875

FEAP/Abaqus

0.15

0.15

0.15

0.15

y0C

FEAP
Abaqus

0.001
0.5E6

0.001
0.5E6

0.1
0.5E6

0.0001
0.5E6

AC

FEAP
Abaqus

6.68E+6
0.34E+6

1.78E+6
0.30E+6

23.42E+6
1.0E+6

0.545E+6
0.50E+6

BC

FEAP
Abaqus

1.27
1.2

1.2
1.11

1.5
1.3

1.2
1.2

CC

FEAP
Abaqus

2.8
1

2.0
1

6.0
1

1.05
1

Fig. 5. Loaddisplacement plots for Type 1-tests: (a) Noto stone and (b) Serena stone.

Fig. 6. Loaddisplacement plots for Type 2-tests: (a) Noto stone and (b) Serena stone.

damage model implemented. For Serena stone, the results


are not conclusive, as the specimen did not fail during
the test (see Section 2); the relevant loaddisplacement plot
does not exhibit any signicant nonlinearity.
Similar remarks apply to the simulation of Type 2 tests
(Fig. 6). For Noto stone, both codes predict the peak-load
and the associated displacement with good accuracy,
although the estimated values are slightly lower than the

experimental ones. Contrary to ABAQUS, FEAP is also


capable of capturing the post-peak behaviour of the specimen. For Serena stone, the local damage model implemented in ABAQUS predicts a peak-load which is not
matched by the experimental plot, as the specimen did
not fail during the test; this is not the case with FEAP.
Finally, regarding Type 3 tests, (Fig. 7) FEAP overestimates the peak-load for specimens both in Noto and

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257

255

Fig. 7. Loaddisplacement plots for Type 3-tests: (a) Noto stone and (b) Serena stone.

_________________
STRESS 3

_________________
STRESS 3

-2.41E+00

-1.79E+00

-2.26E+00
-2.11E+00

-1.68E+00

-1.96E+00

-1.47E+00

-1.80E+00
-1.65E+00

-1.37E+00

-1.50E+00

-1.16E+00

-1.35E+00

-1.05E+00

-1.20E+00

-9.48E-01

-1.04E+00

-8.44E-01

-8.91E-01
-7.39E-01

-7.39E-01

-5.86E-01

-5.30E-01

Time = 1.20E+01

-1.58E+00

-1.26E+00

-6.35E-01

Time = 1.20E+01

Fig. 8. Type 1-tests on walls in Noto stone: contour plots for the axial stress in the post-peak regime and deformed mesh obtained with the local (a) and
the non-local (b) damage model.

_________________
STRESS 3

_________________
STRESS 3

-3.86E+00

-3.31E+01

-3.44E+00
-3.01E+00

-2.87E+01

-2.59E+00

-1.98E+01

-2.16E+00
-1.73E+00

-1.54E+01

-1.31E+00

-6.54E+00

-8.81E-01

-2.12E+00

-4.54E-01

2.30E+00

-2.83E-02

6.73E+00

3.98E-01
8.24E-01

1.11E+01
1.56E+01

1.25E+00

2.00E+01

Time = 1.60E+01

-2.42E+01

-1.10E+01

Time = 1.60E+01

Fig. 9. Type 3-tests on walls in Serena stone: contour plots for the axial stress in the post-peak regime and deformed mesh obtained with the local (a) and
the non-local (b) damage model.

256

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257

Serena stone. Also, both codes overestimate the stiness of


the specimens in Serena stone subjected to Type 3 tests
(Fig. 7b). Both results are likely to be due to the assumption of perfect bonding at the interface between the inner
and the outer leaves made in the analyses. Indeed, the
bonding between stones and mortar is negligible, as indicated by the separation of the leaves which took place in
the tests at very low load values [4,20]. To check the validity of this assumption, a numerical analysis was performed
with ABAQUS assuming a purely frictional Coulombs law
at the interface between the leaves (friction coe. = 0.866).
The results are shown in Fig. 7b: note that, whereas the
stiness of the specimens is correctly reproduced, the
peak-load is denitely underestimated, indicating that
the interfacial cohesion cannot be completely neglected.
Thus, it is possible to conclude that it would be extremely
important to employ a joint element to correctly simulate
the imperfect interfacial bonding between the leaves, so
as to match the real behaviour of the specimens.
Figs. 8 and 9 show the contour plots of the axial stress
(rz, in MPa) obtained in the post-peak regime for specimens in Noto stone subjected to type 1 test and for specimens in Serena stone subjected to type 3 test,
respectively. The contour plots are drawn on the deformed
nite element meshes obtained using the damage model in
its local and non-local versions. When the local model is
employed, damage (and strains) localize either in a strip
of elements (in the simulation of test type 1 Fig. 8a), or
just below the loaded side of the nite element model (in
the simulation of test type 3 Fig. 9a), which makes the
numerical solution strongly mesh-dependent. The use of
the non-local damage model denitely regularizes the
obtained deformation mode (see Figs. 8b and 9b). Note,
however, that in type 3-tests the experimental crack pattern, which mostly runs along the interface between the
inner and the outer layers [4,20], is not matched by the
numerical predictions.
5. Concluding remarks
The proposed numerical model captures dierent features of the nonlinear response of multi-leaf walls (stiness,
peak-load, and damage pattern) in several instances. The
use of a non-local damage model alleviated the meshdependency of the nite element solution and allowed also
the post-peak response of some of the tested walls to be
correctly matched by the numerical simulations. Some
details still remain to be settled, such as the correct description of cracking at the interface between adjacent leaves.
Interfacial brittleness was neglected in the present work,
as perfect bonding was assumed between adjacent layers.
In several instances (compression tests on specimens with
at and keyed collar joints) this is not a severe approximation, as damage mostly involves the individual layers (refer
to [4,20] for a detailed description of the crack pattern in
the specimens). In (pseudo-) shear tests, interfacial cracking
is the dominant failure mode. If perfect bonding is

assumed, both the overall stiness of the specimens and


the peak-load are overestimated, as shown by the plots
reported in Fig. 7; this is especially true for wallettes in Serena stone, which has a very low porosity.
The prosecution of this research will be focused on the
mathematical modeling of interfacial debonding, taking a
weak cohesion into account prior to cracking. Also, the
extension of the model to accommodate creep-induced
damage under high-intensity sustained loads will be dealt
with, with the aim of applying the model to the safety
assessment of massive ancient multi-leaf masonry buildings.
Acknowledgements
This work was developed within the framework of a research program supported by MIUR the Italian Ministry
for Education, University and Research. The contribution
of Dr. Paolo Taranto in performing a part of the numerical
analyses is gratefully acknowledged.
References
[1] Binda L, Cardani G, Saisi A, Modena C, Valluzzi MR, Marchetti L.
Guidelines for restoration and improvement of historical centers in
seismic regions: the Umbria experience. In: Modena C, Lourenco PB,
Roca P, editors. Proceedings of the fourth international seminar on
structural analysis of historical constructions. Leiden (NL): Balkema; 2004, 2: 10618.
[2] Binda L, Tiraboschi C, Baronio G. On site investigation on the
remainings of the Cathedral of Noto. Constr Build Mater
2003;17(Special Issue):54355.
[3] Binda L, Fontana A, Anti L. Load transfer in multiple-leaf
masonry walls. Proceedings of the ninth IB2MaC, Berlin 1991;3:
148897.
[4] Pina-Henriques J, Lourenco PB, Binda L, Anzani A. Testing and
modelling of multiple-leaf masonry walls under shear and compression. In: Modena C, Lourenco PB, Roca P, editors. Proceedings of
the fourth international seminar on structural analysis of historical
constructions, Balkema: Leiden (NL); 2004. 1: p. 299310.
[5] Papa E, Taliercio A. A damage model for brittle materials under nonproportional monotonic and sustained stresses. Int J Numer Anal
Met 2005;29(3):287310.
[6] Papa E, Taliercio A. Prediction of the evolution of damage in ancient
masonry towers. In: Proceedings of the IASS-MSU international
Symposium on bridging large spans from antiquity to the present,
Istanbul; 2000. p. 13544.
[7] Papa E, Taliercio A. Binda L. Safety assessment of ancient masonry
towers. In: Arun G, Seckin N, editors. Proceedings of the second
international congress on studies in ancient structures (SAS2001),
Istanbul, 2001; 1: p. 34554.
[8] Papa E, Taliercio A. Creep modelling of masonry historic towers. In:
Brebbia CA, editor. Structural studies, repairs and maintenance
of heritage architecture VIII. Southampton (GB): WIT Press; 2003.
p. 13140.
[9] Mehrabadi MM, Cowin SC. Eigentensors of linear anisotropic elastic
materials. Q J Mech Appl Math 1990;43:1541.
[10] Bazant ZP. Instability, ductility, and size eect in strain-softening
concrete. J Eng Mech Div-ASCE 1976;102(2):33144.
[11] Jirasek M. Nonlocal models for damage and fracture: comparison of
approaches. Int J Solids Structures 1998;35(3132):413345.
[12] Pijaudier-Cabot G, Bazant ZP. Nonlocal damage theory. J Eng Mech
Div-ASCE 1987;118(10):151233.

M.A. Ramalho et al. / Advances in Engineering Software 39 (2008) 249257


[13] Bazant ZP, Pijaudier-Cabot G. Nonlocal continuum damage, localization instability and convergence. J Appl Mech-T ASME
1988;55(2):28793.
[14] de Borst R, Pamin J, Peerlings RHJ, Sluys LJ. On gradient-enhanced
damage and plasticity models for failure in quasi-brittle and frictional
materials. Comput Mech 1995;17(12):13041.
[15] Comi C, Perego U. A generalized variable formulation for gradient-dependent softening plasticity. Int J Numer Meth Eng
1996;39:373155.
[16] Rodriguez-Ferran A, Morata I, Huerta A. A new damage model
based on non-local displacements. Int J Numer Anal Met
2005;29:47393.

257

[17] Jirasek M, Mara S. Non-local damage model based on displacement


averaging. Int J Numer Meth Eng 2005;63:77102.
[18] Rodriguez-Ferran A, Morata I, Huerta A. Ecient and reliable
nonlocal damage models. Comput Method Appl M 2004;193:343155.
[19] Comi C. A non-local model with tension and compression damage
mechanisms. Eur J Mech A-Solid 2001;20:122.
[20] Anzani A, Binda L, Ramalho MA, Taliercio A. Historic multi-leaf
walls: experimental and numerical research. Masonry Int 2005;18(3):
10114.

Das könnte Ihnen auch gefallen