International Journal of Solids and Structures 50 (2013) 32743296

Contents lists available at SciVerse ScienceDirect

International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr

Compatible model for herringbone bond masonry: Linear elastic

homogenization, failure surfaces and structural implementation
G. Milani a,, A. Cecchi b
a
Department of Architecture, Built Environment and Construction Engineering (ABC), Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milan, Italy
b
Dipartimento di Architettura Costruzione Conservazione (DACC), IUAV University of Venice, Dorsoduro 2206, Ex Convento Terese, 30123 Venezia, Italy

a r t i c l e i n f o a b s t r a c t

Article history: A simplied kinematic procedure at a cell level is proposed to obtain in-plane elastic moduli and macro-
Received 26 October 2012 scopic masonry strength domains in the case of herringbone masonry. The model is constituted by two
Received in revised form 14 April 2013 central bricks interacting with their neighbors by means of either elastic or rigid-plastic interfaces with
Available online 15 June 2013
friction, representing mortar joints. The herringbone pattern is geometrically described and the internal
law of composition of the periodic cell is dened.
Keywords: A sub-class of possible elementary deformations is a-priori chosen to describe joints cracking under in-
Masonry
plane loads. Suitable internal macroscopic actions are applied on the Representative Element of Volume
Herringbone bond
Kinematic model
(REV) and the power expended within the 3D bricks assemblage is equated to that expended in the mac-
Linear elastic moduli roscopic 2D Cauchy continuum. The elastic and limit analysis problem at a cell level are solved by means
Limit analysis of a quadratic and linear programming approach, respectively.
Strength domains To assess elastic results, a standard FEM homogenization is also performed and a sensitivity analysis
Simplied compatible model of regarding two different orientations of the pattern, the thickness of the mortar joints and the ratio
homogenization between block and mortar Young moduli is conducted. In this way, the reliability of the numerical model
is critically evaluated under service loads.
When dealing with the limit analysis approach, several computations are performed investigating the
role played by (1) the direction of the load with respect to herringbone bond orientation, (2) masonry
texture and (3) mechanical properties adopted for joints.
At a structural level, a FE homogenized limit analysis is performed on a masonry dome built in herring-
bone bond. In order to assess limit analysis results, additional non-linear FE analyses are performed,
including a full 3D numerical expensive heterogeneous approach and models where masonry is substi-
tuted with an equivalent macroscopic material with orthotropic behavior and possible softening. Reliable
predictions of collapse loads and failure mechanisms are obtained, meaning that the approach proposed
may be used by practitioners for a fast evaluation of the effectiveness of herringbone bond orientation.
2013 Elsevier Ltd. All rights reserved.

1. Introduction failure marked orthotropy along material axes, strictly dependent

A huge amount of ancient buildings and monuments character-
izing the historical cities and archeological sites are made by ma-
sonry. Their conservation and seismic upgrading is becoming
more and more relevant, but, at present, the evaluation of their
vulnerability still remains very challenging.
The complexity to perform reliable static analyses is connected
to masonry heterogeneity, the variety of unit arrangement which
may be regular or irregular, the little tensile strength and the
marked frictional behavior of the mortar joints, the softening
exhibited in both tension and compression, etc. In addition, when
a regular arrangement of units is considered, masonry exhibits at
Corresponding author. Tel.: +39 022399 4290; fax: +39 022399 4220.
E-mail address: gabriele.milani@polimi.it (G. Milani).
on units geometry and bond pattern.
Basically, masonry may be analyzed by means of micromechan-
ical models, macroscopic approaches or adopting homogenization
techniques, based either on averaging procedures or simplied
models of direct identication (compatible or equilibrated). In this
framework, different constitutive masonry models have been pro-
posed in the literature. Generally speaking, the assumptions to
make for a correct macroscopic characterization are strongly con-
nected to masonry geometry (texture), as well as bricks and mortar
mechanical behavior. In the literature, blocks have been modeled
as rigid (Luciano and Sacco, 1997), deformable with either linear
(Anthoine, 1995) or non-linear response, featuring softening and
damage (Massart et al., 2007; Sacco, 2009), whereas for mortar
either a full continuum representation was adopted (Gambarotta
and Lagomarsino, 1997; Milani et al., 2006a; Massart et al., 2007)

0020-7683/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijsolstr.2013.05.032
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296
or the interface concept was preferred to improve computational
efciency (Loureno and Rots, 1997).
When dealing with bricks disposed in a regular arrangement,
probably the most investigated case is constituted by masonry in
running-bond (e.g. Anthoine, 1995; Loureno and Rots, 1997; Luci-
ano and Sacco, 1998; Cecchi and Sab, 2004; Milani et al., 2006a;
Mercatoris et al., 2009; Salerno and de Felice, 2009; Bacigalupo
and Gambarotta, 2011, 2012a, etc.). Recently, different arrange-
ments have been studied, as for instance English bond or quasi
periodic masonry (e.g. Cecchi and Sab, 2009; Milani and Loureno,
2010a; Milani and Loureno, 2010b; Cecchi and Milani, 2008), but
the research appears still somewhat fragmented when dealing
with herringbone bond masonry, even if some preliminary studies
have been very recently proposed (e.g. Bacigalupo et al., 2012).
Despite its limited application for at walls, such disposition of
the units has proven its usefulness many times in the past for
curves structures, as for instance cross vaults and masonry domes.
Indeed, many historical examples may be found in both Middle
East and European monuments, Fig. 1. As a matter of fact, herring-
bone bond shows increased load bearing capacity when compared
to standard running bond and the advantage of allowing the con-
struction of large curved structures without the centering support.
The present paper aims to obtain a better insight into the elastic
and limit state behavior of herringbone bond masonry in-plane
loaded by means of a novel compatible identication approach.
When a discrete system is modeled as a continuum, a crucial
question is how the kinematic, static and constitutive descriptors
are transferred to the continuum model. In particular, two impor-
tant questions are (1) which type of continuum has to be used to
represent the discrete system and (2) what is the most suited pro-
cedure to characterize the continuum model starting from the dis-
crete system.
Here the attention is devoted to the particular case of the in-
plane behavior. In particular, the 3D continuum is modeled by
means of a 2D approach, where the contemporary presence of
bricks and mortar is accounted for by means of a rigid internal
microstructure. Different models may be potentially chosen,
including Cauchy and Cosserat (Stefanou et al., 2010; Addessi
and Sacco, 2012) continua or higher order (Bacigalupo and Gam-
barotta, 2012a) homogenization. In particular, Addessi and Sacco
(2012) and Stefanou et al. (2010) developed a Cosserat homoge-
nized model with deformable and rigid blocks, respectively, while
Bacigalupo and Gambarotta (2011) compared micropolar and sec-
ond order homogenization approach for masonry available in the
literature and nally Bacigalupo and Gambarotta (2012a) applied
a second gradient homogenization for running and English-bond
masonry in both the static and dynamic elds.
The second question requires some preliminary remarks. The
transfer of the constitutive descriptors from a heterogeneous body
Fig. 1. Masonry dome built using a herringbone texture. Arslan Jadhib Mausoleum,
Sang Bast, Iran, 11th century.
3275
to a homogeneous continuum is often performed referring to stan-
dard homogenization methods (Sanchez-Palencia, 1980; Suquet,
1987). Procedures based on homogenization have been widely
used in the past to provide homogenized constitutive descriptors
for masonry panels subject to in- and out-of-plane actions, see Zuc-
chini and Lourenco (2002), Anthoine (1995), Cecchi and Sab (2002),
etc. Generally, a suitable eld problem is dened and solved, on a
Representative Elementary Volume (REV), before using average
operations, to determine the constitutive homogenized functions.
Other procedures may be found in the literature, as for instance
methods based on equivalence relations between a discrete system
and a continuum model (e.g. Salerno and de Felice, 2009; Masiani
et al., 1995; Masiani and Trovalusci, 1996; Cecchi, 2010; Addessi
and Sacco, 2012), not necessarily based on the solution of an elastic
eld or non-linear problem. In order to link the behavior on the mi-
cro (discrete) level to the macro (continuum) level, the internal
power of the discrete system has to be written as a function of
the deformation variables for the continuum. This requires, in prin-
ciple, the choice of a kinematic correspondence between the mo-
tion in the two models and to write the internal power of the
discrete system in terms of the strain in the continuum. Alterna-
tively, in a dual form, the internal power of the discrete system
may be written in terms of the stress in the continuum, by select-
ing an appropriate correspondence between the stress elds in the
two systems. Therefore, both approaches generally are based on an
approximation, due respectively to equilibrium or compatibility
assumptions and depending on the correspondence postulated.
In this framework, in the present paper, a kinematic approach in
which (a) blocks are supposed either rigid or deformable and (b)
joints are reduced to interfaces, is proposed in order to have a real-
istic prediction of the actual behavior of herringbone bond ma-
sonry in the linear elastic range and near collapse (ultimate
behavior). Then, a numerical procedure of identication between
the 3D discrete Lagrangian system and a continuum equivalent
model is imposed in terms of power dissipated in the 3D discrete
model and in the continuum. A Cauchy continuum is adopted un-
der either the assumption of elastic and rigid-plastic (limit analy-
sis) behavior for mortar joints reduced to interfaces.
When dealing with the elastic case, since deformation can take
place only at the interface between bricks when bricks are as-
sumed rigid, a simple quadratic programming problem in few vari-
ables is obtained to evaluate homogenized elastic moduli. In this
way, macroscopic masonry mechanical characteristics are evalu-
ated as a function of the macroscopic forces. A comparison with
standard FE homogenization is also performed to evaluate the reli-
ability of the approach proposed. The case of bricks elastically
deformable is also discussed through an upper bound superposi-
tion approach.
The extension to rigid-plastic materials requires the combina-
tion of homogenization concepts and classic limit analysis theo-
rems, it is relatively simple and allows to study the behavior at
failure of entire structures disregarding the actual arrangement
of bricks and mortar at a structural level. Furthermore, limit anal-
ysis has the important advantage of requiring only a reduced num-
ber of material parameters, at the same time providing as output
information limit multipliers of loads, failure mechanisms and, at
least on critical sections, the stress distribution at collapse.
Despite the theoretical simplicity of the homogenized limit
analysis, it is worth noting that, at present it has been mainly
applied to running bond textures subjected either to in-plane
(e.g. de Buhan and de Felice, 1997; Milani et al., 2006a) or
out-of-plane (e.g. Cecchi et al., 2007; Sab, 2003) loads. In both
cases, an associated ow rule for joints was adopted, despite
the fact that frictional phenomena may require the adoption of
non-associated plasticity for the constituent materials (e.g Ferris
and Tin-Loi, 2001).
3276 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

Nevertheless, at present, we have not at disposal theorems
regarding homogenization of rigid plastic materials with non-asso-
ciated ow rule. Furthermore, the adoption of this hypothesis im-
plies the lack of the uniqueness of the solution (Ferris and Tin-Loi,
2001). On the contrary, when associated plasticity is adopted for
the constituent materials, homogenization theory can be used in
combination with limit analysis (Suquet, 1983) and simple linear
programming problems (easily manageable by means of standard
packages) are obtained for the evaluation of the homogenized
strength domains.
Following the general procedure adopted by the authors for
running (Cecchi et al., 2007) and English bond (Cecchi and Milani,
2008) textures and using the compatible model utilized for deduc-
ing homogenized elastic moduli, herringbone masonry is studied at
failure within the classic hypothesis holding for limit analysis to
provide upper bound estimates of the homogenized masonry fail-
ure surfaces. It is also shown that the REV constituted by two
bricks interacting with their neighbors provides results identical
to those obtained by different REVs, for example constituted by 4
portions of bricks with interfaces disposed all internally to the
REV boundary. The theoretical formulation provided allows a com-
prehensive study of the behavior of the REV under in-plane loads,
with the future possibility to study thick walls out-of-plane loaded.
Several numerical examples are analyzed, to evaluate the sensi-
tivity of the results to (1) direction of the load with respect to her-
ringbone bond inclination, (2) masonry texture and (3) mechanical
properties adopted for joints.
At a structural level, a FE homogenized limit analysis is per-
formed on a masonry dome built in herringbone bond. In order
to assess limit analysis results, additional non-linear FE analyses
are performed, including a full 3D numerical expensive hetero-
quence, here the internal law of composition represented in
Fig. 3(a) is considered for the sake of simplicity.
The elementary pattern is univocally dened by the ai axes, and
the herringbone bond wall may be generated by repetition and
translation of the chosen pattern in the y1 ; y2 plane (Fig. 3(a)).
Hence, the periodicity may be characterized by a frame of refer-
ence (a1, a2), where a1 and a2 are two independent vectors having
the property that the mechanical characteristics of the 3D assem-
blage are invariant along any translation m1a1 +m2a2, where m1
and m2 are integers (Anthoine, 1995; Stefanou et al., 2010), i.e.:
a1 ae1 e2
1
a2 be1  e2
In Eq. (1), e1 and e2 are the unit vectors of the Euclidean space par-
allel to middle plane of the plate and shell 2D model, while e3 is the
unit vector orthogonal to the middle plane.
According to Fig. 2, the position of a block centroid is univocally
characterized by two indices. In particular, a block with its b length
disposed parallel to the vertical direction is dened by the indices
(i; j, while a block with its b length disposed parallel to the hori-
zontal direction is dened by the indices (i + 1/2, j + 1/2). In this
way, the position of all blocks may be dened according to the rep-
resentation of Figs. 2 and 3(a).
Hence, the gi;j centroid of the block Bi;j is dened as:
   
ia1 ja2 e1 ia jb
gi;j 2
ia1 ja2 e2 ia  jb
whereas the gi1;j1 centroid of the block Bi1;j1 is dened as:
2 2 2 2

geneous approach and models where masonry is substituted (      ) (    )

1 1
with an equivalent macroscopic material with orthotropic i 2 a1 j 2 a2 e1 i 12 a j 12 b
gi1;j1         
behavior and possible softening. Reliable predictions of collapse 2 2 i 12 a1 j 12 a2 e2 i 12 a  j 12 b
loads and failure mechanisms are obtained, meaning that the
approach proposed may be used by practitioners for a fast eval- 3
uation of the effectiveness of a herringbone disposition of the
bricks. Thus, the basic pattern is represented, see Fig. 2, by two blocks. The
pattern exhibits 1 internal interface and 10 external interfaces, 5 for
the block Bi;j and 5 for the block Bi1;j1 .
2 2

2. Geometric description: the herringbone bond pattern Let us indicate with the symbol Ii;j the internal interface be-
tween Bi;j and Bi1;j1 , with Iik1 ;jk2 the external interfaces of Bi;j
2 2
A masonry wall constituted by blocks arranged in herringbone block and with Ii1k1 ;j1k2 the external interfaces of Bi1;j1 block.
2 2 2 2
bond is considered, as schematically sketched in Fig. 2. The block The introduction of k1 and k2 parameters, which may assume val-
1
dimensions are a; b (height and length) and t (thickness). ues equal to 1;  2 and 0, allows to dene all external interfaces.
The 3D assemblage of blocks (heterogeneous
  model) under con- In particular, for an EXTERNAL Iik1 ;jk2 INTERFACE:
sideration occupies a domain X   2t ; 2t where X is the middle
surface of the wall and t is its thickness, that coincides, in this case, if k1 1 and k2 0, then Iik1 ;jk2 Ii1;j represents a vertical
with the thickness of a single block. interface between vertical blocks;
The arrangement is periodic in the space, according to the if k1 1 and k2  12, then Iik1 ;jk2 Ii1;j1 represents a verti-
2
repetitive module represented in Fig. 3. The Representative Ele- cal interface between the Bi;j vertical block and a horizontal
mentary Volume (REV) is chosen in order to contain all the infor- block;
mation necessary to geometrically describe the entire masonry if k1 0 and k2  12, then Iik1 ;jk2 Ii;j1 represent horizontal
2
wall. interfaces between the Bi;j vertical block and two horizontal
 
The module is denoted by Y x   2t ; 2t , where Y  R3 and blocks.
x  X is the middle plane of the module. The boundary of Y is de-
noted by @Y @Y l [ @Y   t
3 [ @Y 3 ; @Y 3 x   2. For an EXTERNAL Ii1k1 ;j1k2 INTERFACE:
2 2
It has to be mentioned that, generally, the elementary cell is not
unique and that its choice in principle may affect the obtained if k1 1 and k2 0, then Ii1k1 ;j1k2 Ii11;j1 represent hor-
2 2 2 2
homogenized continuum. In Fig. 3(a) and (b), for instance, two pos- izontal interfaces between horizontal blocks;
sible different cells are represented. The cell shown in Fig. 3(a) has if k1 1 and k2 0; 1, then Ii1k1 ;j1k2 Ii11;j or
2 2 2
1 internal and 10 external interfaces, with two entire blocks consti- Ii1k1 ;j1k2 Ii11;j1 represent horizontal interfaces between
2 2 2
tuting the whole cell. Conversely, the cell shown in Fig. 3(b) has the Bi1;j1 horizontal block and two vertical blocks;
2 2
only 5 internal interfaces and it is constituted by several portions if k1 0 and k2 12, then Ii1k1 ;j1k2 Ii1;j1 represents a verti-
2 2 2
of different blocks. If the adopted model is a Cauchy continuum, cal interface between the Bi1;j1 horizontal block and a vertical
2 2
the choice of the cell does not affect the nal results. As a conse- block.
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3277

y2
3a+ b
4
y 2

b y 1
y1

a+b
4
a

a b
Fig. 2. REV assumed in the simulations for herringbone bond masonry.

For the sake of simplicity, the following interface denitions will 3. Compatible model for herringbone bond masonry
be used, with reference to the Bi;j block centroid and a frame of ref-
erence y1  y2  y3 centered on gi;j , with y1 horizontal and y2 verti- A wide literature basing on compatible identications between
cal, see Fig. 2: a discrete system and a continuous model exists, but mainly de-
voted to running bond (see for instance Cecchi and Sab, 2004,
(1) INTERNAL INTERFACE between Bi;j and Bi1;j1 blocks: 2009) and less frequently to English bond masonry (Cecchi and
2 2
Milani, 2008). According to such literature, the kinematic of the
discrete system is described with reference to a generic couple of
8 a blocks. The compatible equivalent model bases on a correspon-
< y1 2
>
Ii;j  2b 6 y2 6  2b a 4 dence between equivalent class of motions in the discrete blocks
>
: t system and a plane continuous model.
 2 6 y3 6 2t
A portion of a H masonry panel (assumed constituted by a
(2) EXTERNAL INTERFACES of Bi;j block: homogenized material and hereafter called continuous model)
8 8 with the same dimensions of the REV (discrete block system mod-
a a
<y
>

1 2
 < y1  2
>
 
el) is considered. This portion is chosen so that its center gc coin-
Ii1;j a  2 6 y2 6 2 Ii1;j  2 6 y2 6  a  2b
b b b cides with the center of the REV. A portion of panel P, with the
>
: t >
: t same edge is considered, so that the x point of P coincides with
 6 y3 6 2t  2 6 y3 6 2t
82 a
gc (this is the center of pattern represented in Fig. 2).
< y1  2
> In the discrete system, the motion of a generic couple of blocks
Ii1;j1  a  2b 6 y2 6 2b Bi;j and Bi1;j1 may be described as a function of their center veloc-
2 >
: t 2 2
 6 y3 6 2t _ i;j ; w
ity w
1 1
_ i2;j2 and their angular velocity Xi;j ; Xi2;j2 . In what fol-
1 1

8 a 2 a
8 a a
<  2 6 y1 6 2
> <  2 6 y1 6 2
> lows, for the sake of simplicity, the generic interface (either
Ii;j1 y2 2b Ii;j1 y2  2b internal or external) will be indicated with I symbol. Let p be the
2 >
: 2 >
: center of the I interface between Bi;j and Bi1;j1 . The velocity of
 2t 6 y3 6 2t  2t 6 y3 6 2t 2 2

the material points x of Bi;j and Bi1;j1 in contact in a position

5 2 2

n 2 I, may be written as:

_ i;j p Xi;j n  p
_ i;j x w
w
(3) EXTERNAL INTERFACES of Bi1;j1 block: 7
2 2 1 1 1 1 1 1
83 8a w _ i2;j2 p Xi2;j2 n  p
_ i2;j2 x w
a a
< 2 a 6 y1 6 b 2
> < 2 6 y1 6 b  2
>
The strain rate may be written as function of jump of the velocity
b b
Ii11;j1 y2  2 a Ii11;j1 y2  2 _ n between Bi;j and Bi1;j1 in a point n 2 I:
eld w
2 2 >
: t 2 2 >
: t 2 2
 2 6 y3 6 2t  2 6 y3 6 2t 1 1
8 _
bwnc _ i2;j2 n  w
w _ i;j n
a
< y1 b 2
> 1 1 1 1
_ i;j p Xi2;j2 n  p  Xi;j n  p
_ i2;j2 p  w
w
Ii1;j1  2b 6 y2 6  2b a
2 >
: t _ p Xp n  p
 6 y3 6 2t w 8
8a 2 3a
8 a a i12;j12 i;j i12;j12 i;j
where: w_pw _ _ p and Xp X
p  w X .
< 2 6 y1 6 2
> < b  2 6 y1 6 b 2
>
Ii11;j y2  2b a Ii11;j1 y2  2b Side by side for the panel, a 2D model is dened independently
2 >
: t 2 >
: t from the discrete assemblage of blocks. In particular, reference is
 2 6 y3 6 2t  2 6 y3 6 2t
made to a bi-dimensional continuum identied by its middle plane
6 S of normal e3 (Fig. 4).
3278 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

j-1 j-1/2

j+1/2
j+1/2

j+1

-a

e2
i+1
e1 i i+1/2
i-1/2
i-1

mortar joint
(b-a)/2 (b+a)/2
(b+a)/2
a/2
(b-a)/2

a -b
a/2
a/2

a/2 b/2

brick
a

Fig. 3. Internal law of composition for a herringbone texture (a) and (b) alternative geometry of REV with only internal joint interfaces used for the limit analysis simulations.

Analogously to what done for the discrete model, the kinematic From the above considerations, the generic motion is described by
descriptors of a generic point belonging to the 2D continuum are the elds: w_ : S ! V; X : S ! SkwV that completely describe the
represented by the following elds: velocity and the angular velocity of all points belonging to S.
In the 2D case, the static counterpart is fully described by the
_
wx
9 eld N, collecting the in-plane macroscopic internal actions and
Xx by the eld M, collecting micro couples in the case of micro-polar
_
where wx and Xx are the velocity and angular velocity and angu- continua.
lar velocity of the point x respectively. wx _ is a vector with 2 compo- For the continuum, set N and M actions, the mechanical power
nents, namely w _ 1 x1 ; x2 and w
_ 2 x1 ; x2 , whereas the angular velocity on S may be written as:
is a skew matrix with 1 component 0, dened as follows: p N  grad w_ X Me3 grad X 11
0 1 where grad represents the gradient operator on S.
0 x3 0
B C If the adopted continuum follows Cauchys hypotheses, the in-
X @ x3 0 0A 10
plane couple is assumed equal to zero (Sulem and Mhlhaus,
0 0 0 1997; Stefanou et al., 2008). The total internal power dissipated
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3279

Fig. 4. Identication between REV and a 2D continuum.

Z
can be thus evaluated as the power dissipated by membrane ac- 1 1 1 1
p ti;j n  w
_ i;j n ti2;j2 n  w
_ i2;j2 n
tions as follows: I
Z h i
1 1
p N  symgrad w
_ 12 tn  w_ i2;j2 n  w
_ i;j n
I
In the paper we assume that symgrad w _ D,_ where D_ is the Z
  _ p Xp  skwt  n  p
in-plane membrane strain rate tensor and D_ ab 1=2 wa;b wb;a , tp  w 15
I
with the Greek index a = 1, 2.
R R
From the above considerations, it is possible to assign a corre- Set tp I tn and Mp 2 I skwt  n  p, the previous Eq. (15)
spondence between a class of regular motions in P and H. In par- can be re-written as follows:
ticular, it is assumed that the velocity and angular velocity of the
center of the bricks Bi;j and Bi1;j1 in the discrete system and the 1
2 2 pp tp  w_ p Mp  Xp 16
velocity and angular velocity of the center of the REV in the contin- 2
uum model are equal: According to the kinematic description adopted, let us dene the
T
vector tp as tp t 1p t 2p 0  . Taking into consideration correspon-
_ i;j x wx
w _ _
grad wxg i;j
 x dent motion tests, from Eqs. (13) and (14), Eq. (16) may be split into
13
Xi;j x Xx grad Xxgi;j  x two parts and re-written as follows:

and  1 1 1
_ p tp  gi2;j2  gi;j  grad w
tp  w _ X Mp  Xp
 1 1 2
w
1 1
_ i2;j2 x wx
_ _
grad wx gi2;j2  x Z  1 1
 1 1 14 tn  dp  dp  tn  grad X gi2;j2  gi;j
1 1
Xi2;j2 x Xx grad Xx gi2;j2  x I
Z
1  1 1
1 1 d3p tn  gi2;j2  gi;j grad Xe3 17
where gi;j and gi2;j2 are the centroids of Bi;j and Bi1;j1 2 P gener- 2 I
2 2
ic couple of bricks and a rst order Taylor approximation (rst order
identication) in the velocity and angular velocity is used. where the distance vector dp can be written as dp n  p and d3p is
In the discrete system, the contact forces between blocks Bi;j and the distance between n  p along y3 axis, see Fig. 4.
1 1 At this stage, for a chosen REV and a given class of regular mo-
Bi1;j1 are ti;j n and ti2;j2 n for n 2 I , being n a generic point on
2 2 tions, we impose that the mechanical power dissipated by the con-
the interface I. Equilibrium condition requires that ti;j n tact actions on P and H coincides. Under these assumptions, if the
1 1 1 1
ti2;j2 n. Hence, set ti2;j2 n tn, the power dissipated at the chosen continuum is a Cauchy model, only the symmetric part of p
interface is: is considered, hence the membrane tensor N may be expressed as a
3280 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

function of the vector tp , i.e. as a function of the measure of the It is interesting to notice that the tensor K has, in this case, a diag-
stress in the micro-mechanical model. onal form and that, in linear elasticity, the formulation (16) is
 1 1 substituted with a work-based approach, assuming displacements
1 X
N symtp  gi2;j2  gi;j 18 instead of velocities as kinematic variables.
2A n
In analogy to what done previously in the general case, the eld
P problem may be dened on the Y characteristic module. It must be
where A is the area of the chosen REV and the symbol indicates a
summation extended to all the interfaces to which the chosen REV noted that the eld problem is written exclusively as a function of
is in contact. It must be noted that the part of p associated to the block size.
skwgrad Xe3 is not taken into account. In fact, in the adopted plane The internal work of the interfaces formally is identical to the
model such kinematic elds characterize neutral (rigid) motions. power relation (17) and becomes:
The 1/2 coefcient appearing in the above expressions for N is Z h i
1 1
relative only to the external interfaces of the REV, because such pk1;k2 tn  wi2;j2 n  wi;j n
interfaces are shared by contiguous REVs, while in the case of ZI h i h i
1 1 1 1
internal interfaces, the coefcient is equal to 1. wi2;j2 n  wi;j n  K wi2;j2 n  wi;j n 23
I
After this preliminary characterization of powers expended on
the continuum model and the regular assemblage of blocks, the By minimizing the internal work expressed in (23) by means of a
constitutive homogenized functions for masonry may be intro- standard
 quadratic
F programming routine, the constitutive functions
duced. In particular, for rigid blocks and linear elastic cohesive
 F AF Aabjd representing the homogenized membrane elastic ten-
interfaces, the AF Aabjd homogenized membrane elastic tensor sor may be evaluated numerically.
(corresponding to a prescribed D _ macroscopic strain eld rate) may
be dened as follows: 4.2. Elastic blocks and elastic interfaces: the in-plane case
N  iY AF D_
 thr 19
  The AH homogenized membrane constants may be also found
 Nab is the
Where hiY is the average operator dened in Y and N when the blocks are isotropic linear elastic bodies connected by
macroscopic in-plane (membranal) actions eld for the homoge- cohesive interfaces made with isotropic mortar. Through the vari-
nized shell. ational formulation, it is possible to build an upper bound of the
homogenized constitutive function.
4. The linear elastic problem The bound on AH may be dened, according to Cecchi and Sab
(2002)) as:
In this Section, the previously presented compatible identi-
E  AH E 6 E  AR E 24
cation model is applied to some cases of technical interest of
R
herringbone bond masonry in-plane loaded. Homogenized elastic where A is the homogenized in-plane tensor:
moduli are evaluated numerically by means of a quadratic pro-
 1  1  1
gramming approach with few variables, both in presence of rigid AR AB AF 25
and deformable blocks, in this latter case adopting a bounding
strategy described later in the detail. Results so obtained are
where AB is the plane stress elasticity tensor of blocks and AF is the
compared with those provided by a standard FEM homogeniza-
homogenized membrane tensor for rigid blocks connected by elas-
tion procedure performed on a different unit cell (Anthoine,
tic interfaces. Two orientations for the herringbone pattern are
1995), at two different orientations of the blocks with respect
hereafter considered, namely u = 0 (see Fig. 5(a)) and u p/4
to the horizontal direction.
(Fig. 5(b)).
The homogenized constitutive function may be obtained sim-
4.1. Rigid blocks and elastic interfaces
ply from the rotation of the REV axes by means of well-estab-
lished formulas, once that the constitutive functions on a
If blocks are assumed as rigid bodies and mortar joint is reduced
rotated frame of reference are known. For u p/4 an orthotro-
to an interface with linear elastic behavior, the interaction between
pic homogenized material is obtained, whereas for u = 0 the
a generic couple of Bi;j and Bi1;j1 blocks may be dened by the con-
2 2 homogenized material formally shows a form identical to an
stitutive elastic tensor K between the t tractions at the I interface
anisotropic material. It can be however shown that only four
and the w jump of displacement eld on I, as:
of the total six elastic constants appearing in the tensor are
1 M independent.
K ij a nk nl 20
e iklj Indeed, the homogenized constitutive function for u = 0 is ob-
Here e is the real thickness of the joint, aM is the mortar constitutive tained through a rotation of the principal axes of orthotropy. The
functions and n is the normal to the interface. In the isotropic case, linear transformation, apt at representing the orthotropic constitu-
the above expression becomes: tive constants in new reference system, is:
^
1 M  AH T1 AH TT 26
K l I lM kM n  n 21
e ^
H
where A is the elastic constitutive function in the new reference
where lM and kM are the Lam constants of the mortar, and I is the system, and T is the rotation operator in which c cosu and
identity tensor (Klarbring, 1991; Avila-Pozos et al., 1999). Eq. (21) s cosu:
may be easily re-written as a function of EM and mM mortar Young
and Poisson ratio as follows: 8 9 8 9
2
" < c
> s2 2cs = > <c
> 2
s2 2cs = >

#
1 EM 1 T s 2
c2 2cs ; T1 s 2
c2 2cs 27
K I n  n 22 >
: >
; >
: >
;
e 21 mM 1  2mM cs cs c2  s2 cs cs c2  s2
^
In particular, the components of AH become:
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3281

^
AH1111 AH1111 c4 2AH1122 2AH1212 c2 s2 AH2222 s4 BOUNDARY CONDITION 1:
^
AH2222 AH1111 s4 2AH1122 2AH1212 c2 s2 AH2222 c4 _ 1 0; y2 w
w _ 1 B; y2
^ _ 2 y1 ; 0 w
w _ 2 y1 ; H 0 29
AH1122 AH1111 AH2222 4AH1212 c2 s2 AH1122 c4 s4 
^ 28 _ 1 y1 ; 0 w
w _ 1 y1 ; H
2
AH1212 AH1111 AH2222  2AH1122 c2 s2 AH1212 c2  s2 
^ the third condition of this rst set is imposed to ensure periodicity
AH1112 AH1111 c3 s  AH2222 cs3 AH1122 2AH1212 cs3  c3 s of homologous points in the REV boundary.
^
The corresponding homogenized moduli are:
AH2212 AH1111 cs3  AhH 3 H hH 3 3
2222 c s  A1122 2A1212 cs  c s
Z
1
AH11ab rab dA 30
As can be easily deduced from (28), the obtained homogenized con- A A
stitutive function is expressed in a form identical to the one of an
anisotropic material, but only four of the total six elastic constants BOUNDARY CONDITION 2:
that appear in the tensor are independent.
_ 1 0; y2 w
w _ 1 B; y2 0
w _ 2 y1 ; H
_ 2 y1 ; 0 w 31
4.3. FEM model
_ 2 0; y2 w
w _ 2 B; y2

The third condition of this last set is imposed to ensure periodicity

In this section, a FEM approach is discussed, in order to com-
of homologous points in the REV boundary.
pare the results obtained by means of the compatible identica-
The corresponding homogenized moduli are:
tion model proposed, with alternative, more standard
procedures, but less suited for repeated computations or large Z
1
scale simulations. Following what stated previously, two orienta- AH22ab rab dA 32
A A
tions of the REV are considered, namely u = 0 (Fig. 5(a)) and
u p/4 (Fig. 5(b)). BOUNDARY CONDITION 3:
The adopted discretization is shown in Fig. 6. It is constituted by
4866 2D four-noded elements in plane stress and 4987 nodes and _ 2 0; y2 w
w _ 2 B; y2
appears a good compromise between reliability of the results ob- _ 1 y1 ; 0 w
w _ 1 y1 ; H
tained and computational efciency. The results refer to a herring- 33
_ 1 0; y2 w
w _ 1 B; y2
bone bond with angle of orientation of the blocks with respect to
_ 2 y1 ; 0 w
w _ 2 y1 ; H
the horizontal direction u equal to p/4.
A numerical homogenization procedure may be performed on
The third and fourth conditions are imposed to ensure periodicity of
the unit cell in analogy to what done in Anthoine (1995) for run-
homologous points in the REV boundary.
ning bond textures. However, in the case of herringbone bond,
The corresponding homogenized moduli are:
the REV does not have two orthogonal axes of symmetry
(Fig. 6). As a consequence, in this case, the elementary eld Z
1
problem on the REV must be solved considering the overall cell. AH12ab rab dA 34
A A
According to Fig. 7, boundary conditions to impose on the REV
are the following: As it will be shown from numerical simulations results hereafter,
the obtained homogenized elastic constitutive function will result
orthotropic.
Simulations are performed for u p/4, however the compo-
nents of AH may be numerically determined by means of Eqs.
(26)(28) for u equal to zero (Fig. 5(a)).

-a

-b
=
4

Fig. 5. Pattern orientation. (a) u=0 and (b) u p/4. Fig. 6. FE Model of the REV utilized in the simulations as reference solution.
3282 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296
Fig. 7. REV global dimensions.
5. Elastic numerical results
Several numerical results are reported in this Section varying
EM =EB ratio between mortar and brick Young moduli, thickness of
the mortar joints (two limit cases are considered, namely innitely
thin and 10 mm thick joints) and u orientation of the REV with re-
spect to the horizontal direction. Standard Italian UNI bricks of
dimensions 55 mm  120 mm  250 mm (height  thick-
ness  length) are assumed for all the simulations.
5.1. Discussion on results obtained
As already pointed out in the previous Sections, a model based
on the assumption of linear mortar interfaces and elastic blocks
may be deduced from the less general case of elastic thin mortar
joints and rigid blocks. In this case, it can be shown that an upper
bound of the solution is obtained by superposition of the solution
with rigid blocks to an elastic homogeneous solution. In this man-
ner, also the sensitivity of the results with respect to mortar joint
thickness may be investigated.  F
In Fig. 8, the trends of the homogenized AF Aabjd mem-
brane moduli are represented varying the EM =EB ratio, when the
Young modulus EB of the block is assumed equal to 10000 MPa
and EM varies from 1000 to 5000 MPa. Poisson ratios of both mor-
tar and block (mM and mB respectively) are assumed identically
equal to 0.2. As can be noted, the sensitivity of the results to mortar
thickness decreases considerably when EM =EB increases, and, as ex-
pected, the difference tends to vanish for EM =EB 1 (homogeneous
case).
In Fig. 9, the percentage error er , dened as
AF e1mmAFabjd e10mm
er abjd AF e10mm
 100 is represented as a function of
abjd
EM =EB ratio. It must be noted that the most relevant error of the
innitely thin model is experienced for AF1122 modulus, i.e. the mod-
ulus not belonging to the principal diagonal. This behavior was lar-
gely expected, because the constitutive function chosen for mortar
interfaces is unable to take into account transversal contractions.
Hence, the model is expected to be unreliable for thick (10 mm)
mortar joints. In addition, for the case under study, the ratio be-
tween the a size of the block and the e thickness of the mortar is
equal to 5.5, which, in common practice, corresponds to a mortar
joint 10 mm thick. As a consequence, it appears clear that the
hypothesis of zero thickness for mortar joints may be affected by
quite remarkable errors, at least in selected cases of practical inter-
est, providing systematically quite stiff predictions of membrane
homogenized elastic values.
5.2. Comparison between discrete and FEM model: homogenized
elastic moduli
As already mentioned, in the FEM numerical simulations, Stan-
dard Italian bricks of dimensions 55 mm  120 mm  250 mm
(height  thickness  length) and 10 mm thick mortar joints are
analyzed.
An extensive numerical investigation is carried out to compare
results provided by the model proposed with those obtained
through FEM simulations, for the aforementioned xed geometry,
when the mortar Young modulus EM is assumed to vary in a wide
range of technical interest, and the block Young modulus EB re-
mains constantly equal to 10000 MPa. For both mortar and bricks,
the Poissons ratio is assumed equal to 0.2 (mM mB . In Fig. 10,
ABabjd membrane elastic moduli obtained either by the analytical/
numerical model (the model with elastic blocks is considered) or
by the FEM approach are represented as a function of the EM =EB ra-
tio, varying mortar Young modulus on a wide range of technical
interest (from 1000 to 5000 MPa). Furthermore, the percentage er-
jAF AFEM j
ror estimation eabjdr abjAdF abjd  100 of homogenized moduli ob-
abjd
tained through the approach proposed (AFabjd in comparison with
FEM moduli (AFEM M B
abjd at different E =E ratios is represented in
Fig. 11, for the orientation investigated and one with blocks rotated
by u p/4 (see following section). As it is possible to notice, very
good agreement is found accounting for the actual deformability of
the blocks. Some non-negligible errors (exceeding 15%) are still
present for A1122 module with blocks orientation u p/4 and weak
mortar, probably consequent to the well-known inability of mod-
els with joints reduced to interfaced to reproduce transversal (Pois-
son) deformations (Cecchi et al., 2005).
After a detailed analysis of simulations results, some crucial is-
sues related to the homogenization approach proposed should be
pointed out:
The error introduced using the identication model proposed when
compared to FEM decreases when EM =EB ratio increases. This is
quite obvious because the limit case of EM =EB 1 corresponds to
a homogeneous material. However, the differences between the
models become negligible even for ratios around 0.3.
The differences between the two identication models (thick or
thin mortar joints) are strictly related to the quite large thick-
ness of the mortar joint considered in the numerical application.
The interface model (Cecchi and Sab, 2002) provides less reli-
able results when the ratio between the thickness of the joints
and the dimension of the blocks increases. This is due to the fact
that the homogenized moduli are obtained averaging internal
stresses exclusively in the volume of the blocks (a consequence
of the hypothesis of zero thickness of the mortar joints assumed
before). As obvious consequence, it can be stated that this sim-
plied model is consistent for a small ratio between joint thick-
ness and block dimensions and for ratios between mortar and
block Young modulus not lower than 0.5.
When compared to FE results, the interface model exhibits
higher stiffness, exception made for the A1122 homogenized
modulus. This behavior is typical and strongly connected to
the upper bound approach adopted in the eld solution, see
Eq. (24). A critical discussion on this systematic behavior has
been provided in Cecchi et al. (2005) and the reader is referred
there for further details.
The assumption of a diagonal mortar constitutive tensor
increases slightly the discrepancies between identication
model and FEM results, but can be rather inuent for those
moduli that are affected by a transversal Poisson deformation
of the mortar joint, as for instance A1122 .
5.3. Sensitivity on mortar joint thickness and orientation of the
elementary cell
An additional numerical analysis is carried out here to evaluate
the sensitivity of the homogenized elastic moduli with respect to
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3283

F
A1111 (MPa ) 1.110 4 F
A2222 (MPa ) 1.110 4

4 4
110 110

3 3
910 910

3 3
810 810

3 3
710 710

3 3
610
0.1 0.18 0.26 0.34 0.42 0.5
EM 610
0.1 0.18 0.26 0.34 0.42 0.5 EM
EB EB
F
A1212 (MPa ) 3
4.510
F
A1122 (MPa ) 2.5103
3 3
410 2.110

3 3
3.510 1.710

3 3
310 1.310

3
2.510 900

210
3
E M
500 EM
0.1 0.18 0.26 0.34 0.42 0.5 0.1 0.18 0.26 0.34 0.42 0.5
EB EB
 F
Fig. 8. Trends of the homogenized AF Aabjd membrane moduli as a function of EM =EB ratio.

151

125.833
[A F
1111
(e = 1mm ) A F (e = 10mm )]
1111
100
A F (e = 10mm )
1111
100.667
[A F
2222 (e = 1mm ) A F (e = 10mm )]
2222
100
A F (e = 10mm )
2222

75.5
[A F
1212 (e = 1mm ) A1212
F
(e = 10mm )] 100
A F (e = 10mm )
1212
50.333
[A F
1122 (e = 1mm ) A (e = 10mm )] 100
F
1122
A F (e = 10mm )
1122 25.167

0
0.1 0.18 0.26 0.34 0.42 0.5 E M
EB
AFabjd e1 mmAFabjd e10 mm
Fig. 9. Percentage error AFabjd e10 mm
 100 as a function of EM =EB .

REV orientation. According to Fig. 5, two REV orientations are con-
sidered and the corresponding homogenized constant in the case
of the orientation u 0 may be obtained through Eq. (28).
In Fig. 12, the homogenized membrane moduli obtained with
the identication model proposed are represented varying EM =EB
ratio, and compared with FEM results for the aforementioned ori-
entations u 0 and u p/4.
In Fig. 13, the percentage relative difference between elastic
constants obtained for the two orientations of the REV are re-
ported, assuming as relative difference measure the function
Aabjd up4Aabjd u0
d Aabjd  100.
Aabjd up
4 nated by the construction sequence of massive materials (see
As can be noted, the differences are more evident for the iden-
tication model. In any case, both models provide results with
engineering negligible errors. It is nally interesting to notice that
the bulk moduli are greater for u p/4 orientation, while shear
modulus is greater for u 0. More relevant differences are experi-
enced for the identication model when dealing with AH1122 homog-
enized modulus, for the reasons discussed above.
6. Limit analysis case
Limit analysis is a rather simple and effective procedure com-
monly used in practice to obtain fast estimates of macroscopic ma-
sonry strength domains. In addition, limit analysis has the
important advantage of being independent on self-stresses origi-
Bacigalupo and Gambarotta, 2012b).
Analogously to what done by the authors for running bond ma-
sonry (Cecchi et al., 2007), a MohrCoulomb failure criterion with
tension cut-off ft and a linearized cap in compression (fc ; U2 , see
3284 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

A1111 (MPa ) 110

4
A2222 (MPa ) 110
4

3 3
910 910

3 3
810 810

3 3
710 710

3 3
610 610
_ _
3 EM 3 EM
510 510
0.1 0.18 0.26 0.34 0.42 0.5
EB 0.1 0.18 0.26 0.34 0.42 0.5 EB

A1212 (MPa ) 410

3
A1122 (MPa ) 1.810
3

3 3
3.610 1.5610

3 3
3.210 1.3210

3 3
2.810 1.0810

3
2.410 840
_ _
3 EM EM
210 600
0.1 0.18 0.26 0.34 0.42 0.5
EB 0.1 0.18 0.26 0.34 0.42 0.5 EB

Fig. 10. AFabjd homogenized moduli as a function of EM =EB ratio.

30 15

24 11

18 7

12 3

6 1

0 5
0.1 0.18 0.26 0.34 0.42 0.5 0.1 0.18 0.26 0.34 0.42 E M 0.5
EM
EB E0 B5

[A F
1111
A FEM
1111
] 100 [A F
2222 A FEM
2222
] 100 [A
F
1122 A1122FEM
]
100
[AF
1212 A1212
FEM
]
100
F
F
A 1111
F
A 2222 A F1122 A 1212

Fig. 11. error estimation eabjd of AFabjd homogenized moduli obtained through the model proposed when compared with FEM results and varying EM /EB ratio.

Loureno and Rots, 1997, is assumed for joints. A classic Mohr It can be easily shown that 3  nlin independent plastic multiplier
Coulomb failure criterion may be obtained as particular case when rates have to be assumed as optimization variables for each inter-
ft c= tan U and fc ! 1. face, dealing with a simple set of 3 linear equations involving plas-
Furthermore, it is worth noting that, while the REV shown tic multiplier rates elds k_ Ii n1 ; n3 and velocity jump w
_ n1 ; n3 ,
Fig. 2 is considered in the numerical simulations, alternative ap- that may be written in each point n n1 n3  2 I as:
proaches are equally convenient. For instance, in Fig. 3(b), an
nlin
X
alternative REV is sketched, which, as experienced by the @G
_ 1 ; n3 
wn k_ Ii n1 ; n3 35
authors, provides identical results but has the advantage that i1
@ r
plastic dissipation is allowed exclusively on interfaces laying in-
side the elementary cell (absence of plastic dissipation on inter- In Eq. (35), we assume that w _ n1 ; n3  Dw
_ 2 Dw _ 3 T is the
_ 1 Dw
faces shared by two contiguous REVs). jump of velocity eld (linear in n1 ; n3 on the Ith interface and
A piecewise linear approximation of the failure surface G Gr Dw _ j corresponds to the jump along the direction j, whereas
is adopted for each interface I of area AI , constituted by nlin planes k_ Ii n1 ; n3 is the i-th plastic multiplier rate eld (linear in n1 ; n3
T
of equation AIi r cIi 1 6 i 6 nlin , where r r22 r21 r23 ; r22 of the interface I, associated to the ith linearization plane of the fail-
is the normal stress on the interface and r21 and r23 are tangential ure surface.
stresses along two assigned perpendicular directions, see Cecchi In order to satisfy Eq. (35) for each point of the interface I, nine
et al., 2007 for a detailed description of the numerical model. equality constraints for each interface have to be imposed.
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3285

4 4
110 110

3 3
910 910

3 3
810 810

3 3
710 710

3 3
610 610

3 3
510 510
0.1 0.18 0.26 0.34 0.42 0.5 EM 0.1 0.18 0.26 0.34 0.42 0.5 E M
EB EB
3
410
3 210

3
3.610
3 1.710

3
3.210
3 1.410

3 3
2.810 1.110

3
2.410 800

3
210 500
0.1 0.18 0.26 0.34 0.42 0.5 0.1 0.18 0.26 0.34 0.42 0.5
EM EM
EB EB
B
Fig. 12. Aabjd homogenized moduli as a function of EM =EB ratio, at different REV orientations.

_ A2222 5
_ A1111 6

4
4.8

3
3.6

2
2.4

1
1.2

0
0 0.1 0.18 0.26 0.34 0.42 0.5 E M
0.1 0.18 0.26 0.34 0.42 0.5 EM
EB
EB
0
_ A1212 0 _ A1122
8
3

16
6

24
9

32
12 _
_
40 EM
15 EM 0.1 0.18 0.26 0.34 0.42 0.5
0.1 0.18 0.26 0.34 0.42 0.5 EB
EB

Fig. 13. Percentage relative difference of ABabjd homogenized moduli for the two orientations analyzed, as a function of EM =EB ratio.

Z Z X
nlin  T nlin
Internal power dissipation occurs only on interfaces. For a gen- T I @G 1X X3
pIint _ rdA
w k_ Ii n1 ; n3 rdAI cIi k_ Ii nk1 ; nk3 AI 36
eric Ith interface, such dissipation is dened as the product of the AI AI i1
@ r 3 i1 k1
interface stress vector for the jump of velocities eld, i.e. from where cIi is the right hand side of the ith linearization plane of the
Eq. (35): interface I failure surface.
3286 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

Fig. 14. Elementary homogeneous deformations applied to the representative volume element. (a) D_ 11 , (b) D_ 12 , (c) D_ 21 and (d) D
_ 22 .

External power is evaluated applying the macroscopic deforma-

~_
_ which is hereafter re-arranged in a 3  1 vector D 12
tion tensor D,
~
(where shear contribution D_ 3 12 D12 D21 to facilitate numerical
computations. A graphical representation of single components of D _ n macroscopic strength domain
tensor (elementary deformations applied to the REV) is schemati-
cally sketched in Fig. 14. As it is possible to notice and differently
to the running bond case (Cecchi et al., 2007) each elementary defor-
mation involves interfaces in both shear and pure normal tension.
In the in-plane case, homogenized plate and shell model, exter-
 ~
nal power dissipated can be written as pext RT0 kRT1 D, _ where
22
R0 is the vector of permanent loads, k is the load multiplier, RT1 is
the vector of loads dependent on the load multiplier. As the ampli-
tude of the failure mechanism is arbitrary, a further normalization
~_ 11
condition RT1 D 1 is usually introduced. Hence, the external power
~_
becomes linear in D. Fig. 15. Meaning of nR in the plane-stress case.
Due to the linearity of all the constraints, a linear compact rela-
~_ _ n1 ; n3  may be written for each interface I as:
tion between D and w
~_
_ 1 ; n3  SI n1 ; n3 D After a series of algebraic passages not reported here for the
wn 37
sake of conciseness (a detailed description of the numerical model
where SI n1 ; n3 is a 3  3 matrix which depends only on the geom- is provided in Cecchi et al. (2007) and the reader is referred there),
etry of the interface under consideration. the following linear programming problem is obtained:
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3287

Table 1
Mechanical properties adopted for the numerical simulations at a cell level (standard Italian UNI bricks).

Model A Model B
Mohr Coulomb failure criterion Linearized Loureno Rots failure criterion
U 36 c ft c fc
0.1 (N/mm2) 3.5
2
c 0.1 (N/mm ) U U2
36 30

0.2
Homogenized stress 22 [MPa]

22
0.15

0.1 11
-a
=0
=22.5
0.05
=45

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Homogenized stress 11 [MPa]

0.2

22
Homogenized stress 22 [MPa]

0.15

11
0.1 -b

0.05 =0
=22.5
=45
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Homogenized stress 11 [MPa]

Fig. 16. In plane failure surface, tensiontension region for model A at different orientation of principal axes with respect to horizontal direction. (a) herringbone bond and (b)
running bond.

8
> X nI tion rates and 3  nlin plastic multiplier rates per interface, with a
>
>
> minfkg
>
> pIint  RT0 D~_ total of 1 internal and 10 external interfaces).
>
> I1
< From a numerical point of view, macroscopic masonry failure
~ 38
> nTR D_ 1 nTR nk 0 8 k i; j surfaces may be obtained solving repeatedly a suitable linear pro-
>
>
>
> Xnlin gramming problem derived from the previous, more general, linear
>
> I ~_
>
: G n k D wn k  k_ Ii nk1 ; nk3 @G
@r programming problem (38).
i1

where: 7. In-plane homogenized 11 22 failure surfaces at different

orientation of the load with respect to horizontal direction
k represents the collapse load when a direction nR in the R space
(see Fig. 15) is assigned; In this section, some cases of technical interest are discussed in
nk is a versor such that Rk RT nk ; detail, with the aim of testing both the differences in terms of
i and j represent the axes of projection of the homogenized fail- strength domain when passing from a running bond to a herring-
ure surface. bone texture and the role played by the failure criterion adopted
for mortar joints.
The computation effort to evaluate in-plane masonry failure Two constitutive laws are utilized and critically compared, con-
surfaces for a herringbone texture results thus very limited, since sisting of a classic MohrCoulomb (called here Model A) and a lin-
the kinematic variable are a few (3 macroscopic in-plane deforma- earized Loureno and Rots (1997) failure criterion (denoted as
3288 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

22
V tan( )

(n-1)a+a/2

V
n-1

a
2
1

22
(n-1)a+a/2

Fig. 17. Left: failure mechanism of the elementary cell for vertical stretching obtained from the numerical code implemented. Right: at hand calculation to validate numerical
results assuming joints obeying a MohrCoulomb failure criterion, vertical stretching.

0.2
0.18

0.16
Homogenized stress 22 [MPa]

0.14
22
0.12

0.1
-a
0.08 11
=0
0.06
=22.5
0.04 =45

0.02

0
0 0.05 0.1 0.15 0.2 0.25 0.3
Homogenized stress 11 [MPa]

0.2

0.18

0.16
Homogenized stress22 [MPa]

22
0.14

0.12

0.1 11
-b
0.08

0.06 =0
=22.5
0.04
=45
0.02
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Homogenized stress 11 [MPa]

Fig. 18. Model B. In-plane failure surfaces at different orientation of principal axes with respect to horizontal direction, tensiontension region. (a) herringbone bond and (b)
running bond.

Model B). A validation, by means of at hand calculations, of some interfaces are considered. Mechanical properties adopted for Mod-
numerical results found through the compatible model presented el A and Model B are summarized in Table 1.
is also provided for joints obeying a MohrCoulomb failure crite- In Fig. 16(a), strength domain sections in the tensiontension
rion on two points of the strength domain. region obtained by the homogenization approach assuming a
Standard Italian UNI bricks of dimensions 5.5 cm  12 cm  MohrCoulomb failure criterion are depicted. Three different ori-
25 cm (height  thickness  length) and mortar joints reduced to entations (0, 22.5 and 45) of the principal stress directions
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3289

22

Homogenized stress 22 [MPa]

1.5 11

=0
1 =22.5
=45
-a
0.5

0
-5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0
Homogenized stress 11 [MPa]
Homogenized stress 22 [MPa]

1.5
22

1
=0 11
-b
0.5 =22.5
=45

0
-5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0
Homogenized stress 11 [MPa]

Fig. 19. Model B. In-plane failure surfaces at different orientation of principal axes with respect to horizontal direction, compressiontension region. (a) herringbone bond
and (b) running bond.

(R11 with respect to horizontal axis are plotted. Results may be
directly compared with those relevant for a running bond pat-
tern, reported in Fig. 16(b), where exactly the same mechanical
properties for the joints (Table 1) and the same brick geometry
have been assumed. As it is possible to notice, the differences in-
duced by the utilization of different patterns are rather marked,
especially when dealing with the orthotropy ratio along horizon-
tal and vertical direction. For a running bond texture it is equal
to around 3.55, whereas it settles down to one for herringbone
bond. Generally, herringbone bond is much more resistant along
vertical direction (around 1.45 times) and less resistant along
horizontal direction.
In Fig. 17, the deformed shape at collapse obtained from the
numerical model proposed are represented in the case
R11 R12 0 with only R22 0 and equal to the failure multiplier
k. As it is possible to notice, a stepped failure mechanism is ob-
tained, with cracks zigzagging on horizontal and vertical joints
having all length a, i.e. the height of the brick.
A detailed analysis of the failure mechanism may help in under-
standing why the orthotropic ratio is equal to one. It is interesting
to notice, indeed, that the value of the failure multiplier k can be
evaluated at hand by means of the upper bound theorem of limit
analysis and assuming the failure mechanism represented in
Fig. 17 on the right.
In particular, the assumption of an associated ow rule for the
joints with a MohrCoulomb failure criterion imposes that at fail-
ure the eld of velocities of the joints subjected to plasticization
has a vertical and a horizontal component, as indicated symboli-
cally in Fig. 17, respectively equal to V= tan U and V, where V is
the mutual vertical velocity of the upper and lower part of the wall
forming the failure mechanism and U is joint friction angle.
Assuming to deal with a large wall constituted by 2n bricks in her-
ringbone bond, internal power dissipation for such structure is the
following:
Pin tn  1a a=2ft V tan U tn  1a a=2cV
Having dened with ft and c joint tensile strength and cohesion
respectively, with c ft tan U.
External power expended by the homogenized stress R22 0 is:
Pex t n  1a a=2R22 V 40
Equating internal dissipation and external power expended by the
failure load, R22 is evaluated as:
tn  1a a=2ft tan U tn  1a a=2c
R22
tn  1a a=2
ft tan U c 2c 41
Eq. (41) shows that the vertical ultimate strength of a masonry wall
with bricks disposed in herringbone bond is twice the cohesion of
the joint, when a MohrCoulomb failure criterion is used.
The same formulas may be re-written for the homogenized hor-
izontal strength, meaning that the texture under consideration
exhibits the same resistance along both vertical and horizontal
direction. Such a behavior and the exact values of strength along
material axes are well reproduced by the numerical approach pre-
sented, see Fig. 16.
In addition, it must be emphasized that such mechanism shows
that the overall strength is dependent on both the mortar tensile
and shear strength. As well known, masonry exhibits a brittle
behavior in tension. On the other hand, limit analysis deals with ri-
gid-perfectly plastic and innitely ductile materials. These hypoth-
eses cannot be circumvented and intrinsically represent a
limitation of the model with regard to the actual behavior of any
masonry structural element that cannot be eliminated.
The same simulations are repeated assuming for mortar joints a
linearized LourenoRots failure criterion (Model B), in order to
have an insight into the effect induced by a reduced tensile
strength of the joints and a limited compressive resistance. In
Fig. 18, the homogenized failure surfaces in the tensiontension re-
gion so obtained are shown at different orientations of the bed
39 joint with respect to horizontal direction, and compared with those
3290 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

0 8. Structural application on masonry curved structures

-0.5
Herringbone bone has been widely used in the past to build ma-
22
-1 sonry curved structures, as for instance some typologies of cross

Homogenized stress 22 [MPa]

vaults and masonry domes, both in Europe and Middle East, see
-1.5
Fig. 1. The most indicated structural application to consider for a
11
-2 validation of the numerical model proposed at the macro-scale ap-
pears therefore a masonry dome with small thickness and made of
-2.5 homogeneous material. Since the homogenization model discussed
-a
-3 in the paper assumes rigid plastic constituent materials and limit
analysis theorems are used at the meso-scale, the structural appli-
-3.5 cation is devoted to the evaluation of collapse loads and failure
=0 mechanisms of a curved structure by means of a nite element dis-
-4
=22.5 cretization with rigid plastic elements.
-4.5 =45 When dealing with masonry domes and vaults, it has been re-
cently demonstrated (Milani et al., 2008, 2009a,b) that the most
-5
-6 -5 -4 -3 -2 -1 0 indicated homogenization approach is based on the study of
Homogenized stress 11 [MPa] curved REVs at the meso-scale, which allows a quantitative deter-
mination of the strength reduction induced by the curvature. How-
0
ever, it can be shown that, at least in a preliminary phase, the
-0.5 utilization of curved REVs may be avoided, thus implicitly disre-
22
garding curvature effects, without introducing in the model mean-
-1 ingful errors from an engineering standpoint.
Homogenized stress 22 [MPa]

Another important issue to consider when curved structures are

-1.5
11 considered is the combined presence of in- and out-of-tangent-
-2 plane actions. Out-of-plane actions include exural and torsional
actions, as well as out-of-plane shears. It has been extensively
-2.5
-b shown (e.g. Milani et al., 2006b) that an acceptable approximation,
-3 at least for structures with small thickness is to obtain exural and
torsional strengths by means of an integration along the thickness
-3.5
(layered approach) of in-plane failure surfaces. The relatively small
-4 =0 thickness of the dome allows also disregarding plastic dissipation
=22.5 due to out-of-plane shear, thus implicitly assuming a Kirchhoff
-4.5 Love hypothesis effective. Despite the fact that the model proposed
=45
at the meso-scale allows in principle a separate evaluation of in-
-5
-6 -5 -4 -3 -2 -1 0 and out-of-plane strengths, a so called layered approach is adopted
Homogenized stress 11 [MPa] here for the sake of simplicity, i.e. exural strengths are derived
from integration along the thickness of membrane failure surfaces.
Fig. 20. Model B. In-plane failure surfaces at different orientation of principal axes
The hemispherical dome considered as structural validation has
with respect to horizontal direction, compressioncompression region. (a) herring-
bone bond and (b) running bond. an inner radius equal to 2.2 m, a thickness equal to 12 cm, and it is
built using a herringbone texture. The same dome was experimen-
tally tested by Foraboschi and co-workers in absence (Foraboschi,
relevant for a running bond pattern with exactly the same mechan-
2006) and presence of FRP (Faccio et al., 1999), but using a running
ical properties for the joints (Table 1) and the same brick geometry.
bond disposition. A number of different numerical results for
The differences between the two patterns are quite evident, with
bricks in running bond are also at disposal, e.g. from Milani et al.
an extra-resistance provided by herringbone bond in vertical direc-
(2008) and Milani and Loureno (2012).
tion equal to 1.8. The same comparisons are replicated in Figs. 19
The implementation of the failure surfaces obtained in the pre-
and 20 in the compressiontension and compressioncompression
vious section by means of the meso-mechanical approach pro-
region respectively. From an overall analysis of strength domain
posed is done within an existing limit analysis code rstly
sections, it appears that globally (the term is intended as along
proposed by Milani et al. (2008) in a homogenization framework.
the majority of the directions), the strength exhibited by herring-
The code relies into a FE discretization of the structure by means
bone bond may be rather higher than that provided by a running
of four noded innitely resistant quadrilateral elements and ri-
bond disposition of the bricks. Especially in the compressionten-
gid-plastic interfaces. Both in- and out-of-plane failures are ac-
sion region, the difference appears drastic.
counted for, with the possibility to model also thick structures
Simulations results appear in good agreement with direct build-
(ReissnerMindlin hypothesis).
ing experience and consolidated rule of thumbs. As a matter of fact,
The original dome experimentally tested by Foraboschi (2006)
whilst more difcult to realize in practice, a herringbone pattern
in running bond was built using common Italian bricks of dimen-
may be particularly indicated in all the cases where it is necessary
sions 120  250  55 mm3, with joints thickness approximately
to increase tension and exural resistance of masonry along either
equal to 10 mm. The same bricks are considered here when dealing
vertical or meridian (in the case of domes) directions.
with the herringbone texture. The ratio between brick length and
Finally, the examples discussed underline that masonry macro-
dome diameter is roughly equal to 1:10, which further reduces
scopic failure surface results dependent both on the geometrical
considering the ratio between brick length and dome circumfer-
and mechanical characteristics assumed for the components and
ence (1:55). As recently shown by de Felice et al. (2010) for shear
that the proposed model is able to reproduce different macroscopic
walls, a 1:10 ratio is adequate for the successful application of the
strength domains whenever different failure behaviors for the
homogenization theory. In this case, indeed, it was found that
components are taken into account.
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3291

Loaded
area

2.2 m

-a -b
Fig. 21. (a) Hemispherical dome, geometry and loading conditions. (b): FE discretization utilized for the heterogeneous 3D non-linear analysis and the homogenization
approach.

Table 2
Hemispherical dome tested by Foraboschi (2006). Elastic and inelastic mechanical properties assumed for joints and bricks in the non-linear simulations.

Joint interface Brickbrick interface

E 800 2500 MPa Young Modulus
G E/2 E/2 MPa Shear Modulus
c 1.2ft MPa Cohesion
ft 0.1 MPa Tensile strength
fce 1/3fcp MPa Compressive hardening/softening behavior
fcp 1.8 MPa
fcm 1/2fcp MPa
fcr 1/7fcp MPa
jp 0.009
jm 0.049
U 20 Friction angle
U2 45 Angle of the linearized compressive cap
GIf 0.0065 (CASE A: actual softening case)1 N/mm Mode I fracture energy
(CASE B: elasto-plastic case)
GIIf 0.0050 (CASE A: actual softening case) 1 N/mm Mode II fracture energy
(CASE B: elasto-plastic case)

homogenization provides a limit multiplier lower than the experi-
mental one, with a percentage difference of around 6%.
In Fig. 21, the geometry, the loading condition and the FE dis-
cretization used in the simulations are represented. A 3D heteroge-
neous FE discretization, utilized within existing non-linear FE
software (Milani and Loureno, 2012), is also considered to com-
pare with collapse loads provided by the homogenization model.
The existing code allows to consider joints with both a perfectly
plastic and a softening behavior in tension and shear (described
by Mode I and II fracture energies), as well as a C1 hardening/soft-
ening model in compression. Full details of the non-linear ap-
proach proposed are discussed in Milani and Tralli (2012) and
Milani and Loureno (2012), where the reader is referred to for fur-
ther details.
The dome is loaded until failure by means of a concentrated
vertical increasing load applied at the top of the structure. A steel
plate with dimensions reported in Fig. 21 is placed between the
load and the external loaded surface in order to diffuse vertical
stresses and avoid stress concentrations.
Existing results available from the literature, see e.g. Milani and
Tralli (2012), are all referred to a disposition of the bricks in run-
ning bond and include (1) experimental load displacement curves
(Foraboschi, 2006), (2) collapse loads obtained by means of a limit
analysis approach (Milani et al., 2008), (3) numerical loaddis-
placement curves obtained with non-linear FE simulations per-
formed by means of the DIANA commercial code (Milani et al.,
2009b; DIANA, 2008), where the dome is discretized with 3D ele-
ments and a macroscopically homogeneous material exhibiting
3292 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296

Table 3 achieved. In agreement with the limit analysis model proposed at

Hemispherical dome tested by Foraboschi (2006). Mechanical characteristic assumed
for joints and bricks in the limit analysis simulations.
Joint (LourenoRots failure criterion)
ft N=mm2 Tensile strength
fc N=mm2 Compressive strength
c Cohesion
U Friction angle
U2 Angle of the linearized compressive cap
softening is adopted for masonry and (4) two numerical loaddis-
placement curves obtained using for masonry the non-linear soft-
ening homogenization model proposed in Milani and Tralli (2012).
In this latter case, two different approaches were used at a struc-
tural level, called respectively QP and DSM model and essentially
differing one each other for the number of non-linear interfaces
utilized at a structural level. Again, the reader interested in the de-
tails of such models is referred to Milani and Tralli (2012).
The same QP model is adapted here to study the dome in her-
ringbone disposition with full 3D elements and a heterogeneous
approach, i.e. modeling separately bricks with 8-noded elements
and joints with quadrilateral interfaces.
Elastic and inelastic properties utilized for bricks and mortar
interfaces within the non-linear heterogeneous simulations are
summarized in Table 2, whereas limit analysis parameters are
summarized in Table 3. As it is possible to notice from Table 2,
two different sub-cases are tested, the rst relying into a softening
model with realistic fracture energies in tension and shear (CASE
A), the second with large fracture energies, approximating an elas-
tic-perfectly plastic material (CASE B). While for CASE A, a drop of
the global loaddisplacement curve is expected, for CASE B a better
approximation of the homogenized limit analysis prediction is
the meso-scale, joints are assumed obeying a MohrCoulomb fail-
ure criterion combined with tension cut-off and cap in compres-
sion, Loureno and Rots (1997), whereas bricks are supposed
0.1 innitely resistant (elastic behavior).
1.8 In Fig. 22, a comparison among collapse loads and load-maxi-
1:2f t mum vertical displacement curves provided by all the different
20 models previously mentioned is reported.
45
As it is possible to notice, the homogenized limit analysis ts
reasonably peak load provided by the heterogeneous non-linear
approach, meaning that useful predictions of the load bearing
capacity of the structure may be obtained using the procedure pro-
posed. However, for CASE A, an error equal to 33% is reached.
In addition, it is interesting to notice the effect induced by the
bricks disposition on both collapse loads and initial elastic stiff-
ness, with a clear increase of the load bearing capacity.
In Fig. 23, the deformed shapes at peak obtained by means of
non-liner heterogeneous and homogenized limit analysis are com-
pared. As it is possible to notice, in the non-linear heterogeneous
deformed shape some interfaces between rigid elements within
the same block exhibit a small opening, not related to an inelastic
deformation of the interface, but associated to the elastic contribu-
tion. To prevent bricks interfaces opening, an innite Young mod-
ulus should be assumed, causing however numerical instabilities.
Apparently, the discrepancy between limit analysis collapse
load and the heterogeneous model (Case A) do not fully justify
its applicability for practical purposes.
However, it is worth noting that, in the heterogeneous model,
the behavior of the interfaces is modeled by means of equivalent
normal and shear non-linear springs, according to Kawai (1978)).
The exural behavior is derived by integration of the non-linear ax-
ial springs stressstrain diagrams along the thickness and an iter-
ative procedure is adopted to take into account the increase of peak

120

100
External load P [kN]

80

60

40

20

0
0 1 2 3 4 5 6 7 8
Max displacement [mm]
Herrigbone bond: present limit analysis homogenization model
Herrigbone bond: present model with finite out-of-plane shear strength
Herrigbone bond: present model, finite out-of-plane shear strength, triangular mesh
Herrigbone bond: Heterogeneous softening 3D model
Herrigbone bond: Heterogeneous elasto-plastic 3D model
Running bond: experimental data
Running bond: Milani & Tralli (2012) QP model
Running bond: Milani & Tralli (2012) DSM model
Running bond: DIANA (Homogeneous)
Running bond: DIANA (Heterogeneous)
Running bond: limit analysis Milani et al. (2009)

Fig. 22. Load displacement curves and limit analysis collapse loads provided by the different models in presence of bricks disposed in running bond and herringbone bond.
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296 3293

Fig. 23. Deformed shape at peak of the heterogeneous elasto-plastic model and failure mechanism provided by limit analysis.

out-of-plane strength and ductility due to the normal pre-com-
pression of the interface. In order to utilize a mathematical pro-
gramming approach, the actual stressstrain behavior of the
springs is approximated with a linear piecewise constant dia-
gram (Milani and Tralli, 2012). Such approximation allows solv-
ing the incremental problem by means of quadratic
programming. In order to deal with the drop of the load carrying
capacity of the interfaces, a sequential scheme is proposed in
Milani and Tralli (2012). The main advantage of the approach
proposed is essentially linked to the extreme conceptual simplic-
ity. Drawbacks connected to the mathematical programming
procedures are that (1) the repeated call of the quadratic pro-
gramming routine cannot compete with commercial software
in efciency, (2) there are some limitations on the maximum
optimization variables to use and (3) the linear piecewise con-
stant approximation of the stressstrain curves has to be quite
coarse. By default, an approximation with 4 constant values of
residual strength per spring is assumed. An area equivalence
principle with the original stressstrain diagram is adopted, with
a slight underestimation of the peak load, which actually is one
of the causes of the less resistant behavior of the heterogeneous
model.
3294 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296
700
600
Bending moment Mnn [N*mm/mm]
500
400 Elasto-damaging
Elastic perfectly plastic
300
200
100
0 a.
0 1 2 3 4 5
curvature [1/mm]
Fig. 24. (a): curvature bending moment for the material of Table 2 with softening
and without softening. (b) Effect of out-of-plane sliding, detail near the point of
application of the external load.
Further causes of the discrepancies found between numerical
simulations and limit analysis approach are the following:
(1) For Case A, mortar joints with softening and small fracture
energy in tension are used. In Fig. 24, the meridian bending
moment for interfaces with low pre-compression (0.05 MPa)
and assuming material properties assigned to joints in the
paper is shown. Two diagrams are shown, one assuming
an innite fracture energy in tension (elastic perfectly plas-
tic material, CASE B), the other with the actual low fracture
energy adopted in the heterogeneous simulations (CASE A).
As it is possible to notice, a 25% of difference between the
peak moments is experienced, which is at the base of great
part of the discrepancy.
(2) To assess the conclusion raised in the previous point, non-
linear heterogeneous simulations are repeated on the same
discretization, assuming for the interfaces an elastic-per-
fectly plastic behavior with innite ductility, simply increas-
ing Mode I and II fracture energies of mortar joints, see
Table 2 CASE B. Simulation results conrm an increase of
the load carrying capacity of the structure up to around
100 kN, very near to homogenized limit analysis
simulations.
(3) Out-of-plane shear limited strength is not considered in the
homogenized model. Such a simplistic assumption is in
agreement with the in-plane approach proposed, where
bending moment is derived by direct integration along the
thickness of in-plane actions. As it is possible to notice from
a detail of the deformed shape at collapse near the point of
application of the load, Fig. 24(b), a small but perceivable
out-of-plane sliding of the bricks in the heterogeneous
model exists, meaning that an approach without out-of-
plane shear dissipation tends to slightly overestimate the
collapse load by its intrinsic nature.
(4) The approximation induced by homogenization when deal-
ing with a problem exhibiting a ratio between brick length
and dome diameter roughly equal to 1:10 is hardily evalu-
able from a quantitative point of view. As a matter of fact,
such ratio seems to represent an upper bound for the appli-
-a
cability of the homogenization theory, according to results
reported in de Felice et al. (2010).
(5) Comparing with simulation results found in a sensitivity
analysis conducted by Milani et al. (2008) on the same
example but for running bond masonry, the following addi-
tional aspects are worth noting:
For joints tensile strength and friction angle assumed
6 7 8
-4 respectively equal to 0.1 MPa and 20, the running bond
x 10
FE limit analysis collapse load found was 58 kN. Within
the homogenization model, in presence of small negative
annular stress, Fig. 19, the meridian tensile strength remains
essentially equal to the mono-axial vertical strength in run-
ning bond, but increases considerably for herringbone bond.
b. In the post processing phase of the upper bound homoge-
nized limit analysis (i.e. after proper solution of the dual
-b problem), it is found a meridian compressive stress roughly
equal to 0.2 MPa in correspondence of the zone undergoing
plastic deformation. Such value corresponds to a meridian
tensile strength of about 0.2 MPa in the homogenized
model. According to Milani et al. (2008) sensitivity analyses
for running bond dispositions, a limit load equal to 105 kN is
found. It is therefore expected that a similar value is found
for the same dome in herringbone bond.
c. A collapse load equal to 118 kN is provided by the
homogenization model in the herringbone bond case.
The slightly higher value found (when compared to the
simplied prediction based on Milani et al. (2008) simula-
tions, see previous point) is connected both to the innite
out-of-plane shear strength assumed in the homogenized
model (see deformed shape detail in Fig. 24(b)) and to
the FE limit analysis model utilized at a structural level.
As a matter of fact, the FE discretization is constituted
by rigid innitely resistant quadrilateral elements (thick-
ness is represented only in the post-processing phase for
the sake of clearness) with plastic dissipation allowed
only at the interfaces between adjoining elements. No
diagonal interfaces are present, where it is actually expe-
rienced some plastic dissipation (Milani et al., 2008). Sim-
ulations were re-run with an arbitrary out-of-plane
strength equal to joint cohesion and a triangular discreti-
zation, conrming the slight drop of the failure load
(103 kN, see Fig. 22).
A comparison with literature deformed shapes, see Creazza
et al. (2000, 2002) and Milani et al. (2009b), shows very convincing
analogies between present model and commercial FEM. Consider-
ing that present simulations are performed at a small fraction of
processing time needed by commercial FEM in the inelastic range,
it seems reasonable to conclude that the approach proposed may
be considered interesting for practical purposes.
No drastic differences between deformed shapes for herring-
bone bond and running bond dispositions are found by the authors
using the homogenized limit analysis. This was largely expected,
since failure is essentially due to exural actions distributed on a
crown plastic hinge at approximately 2/3 of the height, Fig. 23. In
addition, the difference in shape of the in-plane strength domains
does not justify evident discrepancies in the qualitative behavior in
exion.
 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296
Fig. 25. Normalized power dissipated patch (P max is the maximum nodal power dissipation value).
Conversely, the greater collapse load provided by the herring-
bone texture is a consequence of the larger resistance of the
homogenized material along parallels, when compared to running
bond. This behavior has been quantitatively assessed in the previ-
ous section and is conrmed by intuition.
Internal plastic dissipation patch is nally represented in
Fig. 25. As it possible to note, internal dissipation is concentrated
along a circular crown, with a quite clear formation of one annular
bending hinge, again in agreement with the failure mechanism
represented in Fig. 23. Finally a minor amount appears along the
meridians, with a slight crack opening between contiguous in-
clined bricks, as kept by the heterogeneous model.
9. Conclusions
A compatible identication model for the analysis of herring-
bone bond masonry under service loads (linear elastic analysis)
and near failure (limit analysis) has been presented.
After a general discussion of the model assumed from a kinematic
point of view, the approach has been specialized to in-plane loads
and either elastic or rigid-plastic materials, with an identication
of the 3D assemblage of blocks with a 2D Cauchy continuum.
When dealing with the elastic case, two different models have
been proposed, the rst constituted by rigid blocks and mortar re-
duced to interfaces, the second characterized by elastic blocks. Two
orientations of the herringbone patter have been discussed,
namely with blocks disposed horizontally and vertically or along
the bisecting line of the rst and third quadrant.
The elastic moduli obtained by means of the approaches dis-
cussed have been extensively compared with those provided by a
standard FE discretization of the unit cell, at different ratios be-
tween the elastic moduli of mortar and brick.
When dealing with the rigid-plastic case, joints have been re-
duced to interfaces obeying a MohrCoulomb failure criterion
eventually with limited tensile and compressive strength, whereas
bricks have been assumed innitely resistant. Within such hypoth-
eses, a kinematic limit analysis formulation has been presented to
obtain upper bound in-plane homogenized failure surfaces, to
implement at a structural level within a 3D FE limit analysis code,
suitable to provide estimates of collapse loads and failure mecha-
nisms of real structures.
Several numerical examples have been analyzed, to evaluate
the sensitivity of the results at the meso-scale (i.e. failure surfaces)
with respect to (1) direction of the load against herringbone bond
disposition, (2) masonry texture and (3) mechanical properties
adopted for joints.
At a structural level, a FE homogenized limit analysis has been
performed on a masonry dome built in herringbone bond and the
3295
results so obtained have been compared to a number of additional
non-linear and limit analysis FE simulations, to evaluate the reli-
ability of the approach proposed.
Acknowledgments
The research project reported in this paper was conducted
thanks to the nancial support of ReLUIS 2 Task A-1.1 and
PRIN2010-2011.
References
Addessi, D., Sacco, E., 2012. A multi-scale enriched model for the analysis of
masonry panels. International Journal of Solids and Structures 49 (6), 865880.
Anthoine, A., 1995. Derivation of the in-plane elastic characteristics of masonry
through homogenization theory. International Journal of Solids and Structures
32 (2), 137163.
Avila-Pozos, O., Klarbring, A., Movchan, A.B., 1999. Asymptotic model of orthotropic
highly inhomogeneous layered structure. Mechanics of Materials 31, 101115.
Bacigalupo, A., Cavicchi, A., Gambarotta, L., 2012. A simplied evaluation of the
inuence of the bond pattern on the brickwork limit strength. Advanced
Materials Research 368373, 34953508.
Bacigalupo, A., Gambarotta, L., 2011. Non-local computational homogenization of
periodic masonry. International Journal for Multiscale Computational
Engineering 9 (5), 565578.
Bacigalupo, A., Gambarotta, L., 2012a. Computational two-scale homogenization of
periodic masonry: characteristic lengths and dispersive waves. Computer
Methods in Applied Mechanics and Engineering 213216, 1628.
Bacigalupo, A., Gambarotta, L., 2012b. Effects of layered accretion on the mechanics
of masonry structures. Mechanics Based Design of Structures and Machines 40,
163184.
Cecchi, A., 2010. Procedures to build plate micromechanical models for composites
like periodic brickwork: a critical review. Composites: Mechanics,
Computations, Applications 1 (4), 287313.
Cecchi, A., Milani, G., 2008. A kinematic FE limit analysis model for thick English
bond masonry walls. International Journal of Solids and Structures 45, 1302
1331.
Cecchi, A., Milani, G., Tralli, A., 2007. A ReissnerMindlin limit analysis model for
out-of-plane loaded running bond masonry walls. International Journal of
Solids and Structures 44 (5), 14381460.
Cecchi, A., Milani, G., Tralli, A., 2005. Validation of analytical multiparameter
homogenization models for out-of-plane loaded masonry walls by means of the
nite element method. Journal of Engineering Mechanics ASCE 131 (2), 185
198.
Cecchi, A., Sab, K., 2002. Out of plane model for heterogeneous periodic materials:
the case of masonry. European Journal of Mechanics A-Solids 21, 249268.
Cecchi, A., Sab, K., 2004. A comparison between a 3D discrete model and two
homogenised plate models for periodic elastic brickwork. International Journal
of Solids and Structures 41 (910), 22592276.
Cecchi, A., Sab, K., 2009. Discrete and continuous models for in plane loaded random
elastic brickwork. European Journal of Mechanics A-Solids 28, 610625.
Creazza, G., Saetta, A., Matteazzi, R., Vitaliani, R., 2000. Analyses of masonry vaulted
structures by using a 3-D damage model. In: European Congress on
Computational Methods in Applied Sciences and Engineering, ECCOMAS 2000,
Barcelona, SP.
Creazza, G., Saetta, A., Matteazzi, R., Vitaliani, R., 2002. Analyses of masonry vaults:
a macro approach based on three-dimensional damage model. Journal of
Structural Engineering 128 (5), 646654.
3296 G. Milani, A. Cecchi / International Journal of Solids and Structures 50 (2013) 32743296
de Buhan, P., de Felice, G., 1997. A homogenisation approach to the ultimate
strength of brick masonry. Journal of the Mechanics and Physics of Solids 45 (7),
10851104.
de Felice, G., Amorosi, A., Malena, M., 2010. Elasto-plastic analysis of block
structures through a homogenization method. International Journal for
Numerical and Analytical Methods in Geomechanics 34, 221247.
DIANA 9.3.2, 2008. Users Guide. TNO building and construction research, The
Netherlands.
Faccio, P., Foraboschi, P., Siviero, E., 1999. Masonry vaults reinforced with FPR strips.
LEdilizia 7 (8), 4450 (In Italian: Volte in muratura con rinforzi in FRP).
Ferris, M., Tin-Loi, F., 2001. Limit analysis of frictional block assemblies as a
mathematical program with complementarity constraints. International Journal
of Mechanical Sciences 43, 209224.
Foraboschi, P., 2006. Masonry structures externally reinforced with FRP strips: tests
at the collapse. In: Proc I Convegno Nazionale Sperimentazioni su Materiali e
Strutture, Venice (in Italian).
Gambarotta, L., Lagomarsino, S., 1997. Damage models for the seismic response of
brick masonry shear walls part ii: the continuum model and its application.
Earthquake Engineering and Structural Dynamics 26, 441462.
Kawai, T., 1978. New discrete models and their application to seismic response
analysis of structures. Nuclear Engineering and Design 48 (1), 207229.
Klarbring, A., 1991. Derivation of model of adhesively bounded joints by the
asymptotic expansion method. International Journal of Engineering Science 29,
493512.
Loureno, P.B., Rots, J., 1997. A multi-surface interface model for the analysis of
masonry structures. Journal of Engineering Mechanics ASCE 123 (7), 660668.
Luciano, R., Sacco, E., 1997. Homogenization technique and damage model for old
masonry material. International Journal of Solids and Structures 34 (24), 3191
3208.
Luciano, R., Sacco, E., 1998. Variational methods for the homogenization of
periodic heterogeneous media. European Journal of Mechanics-A/Solids 17
(4), 599617.
Masiani, R., Rizzi, N., Trovalusci, P., 1995. Masonry as structured continuum.
Meccanica 30 (6), 673683.
Masiani, R., Trovalusci, P., 1996. Cauchy and Cosserat materials as continuum
models of brick masonry. Meccanica 31 (4), 421432.
Massart, T.J., Peerlings, R.H.J., Geers, M.G.D., 2007. An enhanced
multi-scale approach for masonry wall computations with localization of
damage. International Journal for Numerical Methods in Engineering 69 (5),
10221059.
Mercatoris, B.C.N., Massart, T.J., Bouillard, P., 2009. Multi-scale detection of failure
in planar masonry thin shells using computational homogenisation.
Engineering Fracture Mechanics 76 (4), 479499.
Milani, G., Loureno, P.B., 2010a. Monte Carlo homogenized limit analysis model for
randomly assembled blocks in-plane loaded. Computational Mechanics 46 (6),
827849.
Milani, G., Loureno, P.B., 2010b. A simplied homogenized limit analysis model for
randomly assembled blocks out-of-plane loaded. Computers & Structures 88,
690717.
Milani, G., Loureno, P.B., 2012. 3D non-linear behavior of masonry arch bridges.
Computers & Structures 110111, 133150.
Milani, G., Loureno, P.B., Tralli, A., 2006a. Homogenised limit analysis of masonry
walls. Part I: Failure surfaces. Computers and Structures 84 (34), 181195.
Milani, G., Loureno, P.B., Tralli, A., 2006b. Homogenization approach for the limit
analysis of out-of-plane loaded masonry walls. ASCE Journal of Structural
Engineering 132 (10), 16501663.
Milani, G., Milani, E., Tralli, A., 2008. Limit analysis of masonry vaults by means of
curved shell nite elements and homogenization. International Journal of Solids
and Structures 45 (20), 52585288.
Milani, G., Milani, E., Tralli, A., 2009a. Upper bound limit analysis model for FRP-
reinforced masonry curved structures. Part I: Unreinforced masonry failure
surfaces. Computers & Structures 87 (2324), 15161533.
Milani, G., Milani, E., Tralli, A., 2009b. Upper bound limit analysis model for FRP-
reinforced masonry curved structures. Part II: Structural analyses. Computers &
Structures 87 (2324), 15341558.
Milani, G., Tralli, A., 2012. A simple mesomacro model based on SQP for the non-
linear analysis of masonry double curvature structures. International Journal of
Solids and Structures 49 (5), 808834.
Sab, K., 2003. Yield design of thin periodic plates by a homogenisation technique
and an application to masonry walls. Comptes Rendus Mecanique 331, 641
646.
Sacco, E., 2009. A nonlinear homogenization procedure for periodic masonry.
European Journal of Mechanics-A/Solids 28 (2), 209222.
Salerno, G., de Felice, G., 2009. Continuum modeling of periodic brickwork.
International Journal of Solids and Structures 46 (5), 12511267.
Sanchez-Palencia, E., 1980. Non Homogeneous Media and Vibration Theory.
Springer, Berlin.
Stefanou, I., Sulem, J., Vardoulakis, I., 2008. Three-dimensional Cosserat
homogenization of masonry structures: elasticity. Acta Geotechnica 3 (1), 71
83.
Stefanou, I., Sulem, J., Vardoulakis, I., 2010. Homogenization of interlocking
masonry structures using a generalized differential expansion technique.
International Journal of Solids and Structures 47, 15221536.
Sulem, J., Mhlhaus, H.-B., 1997. A continuum model for periodic two-dimensional
block structures. Mechanics of Cohesive-Frictional Materials 2, 3146.
Suquet, P., 1983. Analyse limite et homogeneisation. Comptes Rendus de lAcademie
des Sciences Series IIB Mechanics 296, 13551358.
Suquet, P., 1987. Elements of homogenization for inelastic solid mechanics. In:
Homogenization Techniques for Composite Media. Springer-Verlag, Berlin.
Zucchini, A., Lourenco, P.B., 2002. A micro-mechanical model for the
homogenisation of masonry. International Journal of Solids and Structures 39,
32333255.