International Journal of Architectural Heritage

Conservation, Analysis, and Restoration

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/uarc20

Efficient Constitutive Model for Continuous Micro-

Modeling of Masonry Structures

M. Petracca, G. Camata, E. Spacone & L. Pelà

To cite this article: M. Petracca, G. Camata, E. Spacone & L. Pelà (2022): Efficient Constitutive
Model for Continuous Micro-Modeling of Masonry Structures, International Journal of Architectural
Heritage, DOI: 10.1080/15583058.2022.2124133

To link to this article: https://doi.org/10.1080/15583058.2022.2124133

Published online: 18 Sep 2022.

Submit your article to this journal

Article views: 29

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=uarc20
INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE
https://doi.org/10.1080/15583058.2022.2124133

RESEARCH ARTICLE

Efficient Constitutive Model for Continuous Micro-Modeling of Masonry

Structures
a
M. Petracca , G. Camatab, E. Spacone b
, and L. Pelàc
a
ASDEA Software, Pescara, Italy; bDepartment of Engineering and Geology (Ingeo), Università G.d’Annunzio Chieti E Pescara, Pescara, Italy;
c
Department of Civil and Environmental Engineering, Universitat Politècnica de Catalunya (UPC-BarcelonaTech), Barcelona, Spain

ABSTRACT ARTICLE HISTORY

Masonry is a composite material often modeled as an equivalent homogenous material. However, Received 31 March 2022
the complexity of its micro-structure leads to complex mechanical responses, which are almost Accepted 6 September 2022
impossible to capture accurately with homogenous constitutive models. Micro-modeling can be KEYWORDS
used in these scenarios, allowing for the explicit modeling of microstructural components, leading Buildings; damage; impl-ex;
to an accurate capturing of their interaction. Its main drawback is the computational cost, which masonry; micro-modeling;
often makes this approach suitable only for the simulation of small specimens. This is especially true mixed implicit explicit
due to strain-softening leading to severe instabilities and non-convergence of the solution. The integration; parallel
objective of this work is to propose a simple yet effective constitutive plastic-damage model for the computing; partitioned
microstructural components of masonry. It is based on a damage model previously developed by mesh; plasticity; seismic
the authors. For a better representation of the cyclic response of masonry, plasticity is added using analysis
a simplified implementation that does not strictly follow the rules of standard elastoplasticity,
allowing an explicit computation of the stress tensor from the strain tensor without the need for an
iterative loop at the material level. To reduce the numerical issues related to strain-softening and
thus improve the stability of the solution, an IM-PLEX integration algorithm is adopted.

1. Introduction representing the so-called “size effect” (Barenblatt 2014;

From a macroscopic standpoint, masonry can be con­ Bažant 2004). A “size effect” arises every time a material
sidered a composite material consisting of units (bricks property does not appear to be the same for two geome­
or stones) and mortar joints. Their highly different trically similar structures with different sizes: In quasi-
mechanical properties, dimension ratios, and their brittle materials, in fact, both structural brittleness and
arrangement in the microstructure contribute to extre­ material strength are found to be scale-dependent.
mely complex non-linear responses characterized by Other approaches have been proposed in the litera­
several different failure mechanisms. Simulation of ture to overcome the shortcomings of macro-modeling
masonry structures with the Finite Element Method in accounting for the effect of the microstructure on
can be done at different levels of detail. the structural response. Fracture-Mechanics has been
Homogenous continuum models (Pelà et al. 2014; extensively used to study crack propagation in masonry
Pelà, Cervera, and Roca 2011, 2013) can be used in structures (Accornero, Lacidogna, and Carpinteri 2016,
a standard macroscopic approach, where the whole 2018; Accornero and Lacidogna 2020; Panian and
masonry structure is modeled as an equivalent homo­ Yazdani 2020; Yazdani and Habibi 2021). It is able to
geneous medium equipped with an equivalent homoge­ accurately describe the crack propagation, and the size
neous material that should be able to represent the main effects can be captured by resorting to the concepts of
features of masonry. This method is the fastest in terms scaling laws for strength and brittleness, dimensional
of both computational costs and model complexity. analysis and fractal geometry (Carpinteri 1994;
However, simple tensorial constitutive models may not Carpinteri and Chiaia 1997; Carpinteri and Puzzi
be able to accurately represent features such as strength 2009). On the other hand, Continuum-Mechanics has
orthotropy followed by damage-induced anisotropy and also been used to in this regard. To account for scale
the effects of the arrangement and size of micro- effects, the micro-structure should somehow be incor­
structural constituents. Furthermore, it is well known porated into the computational model. This can be
that standard tensorial constitutive models, either based achieved either with Micro-Modeling (Drougkas, Pelà,
on plasticity, damage, or a combination of them, fail in and Roca 2014; Lourenço 1996; Lourenço and Rots

CONTACT M. Petracca [email protected] ASDEA Software, Pescara, Italy

© 2022 Taylor & Francis
2 M. PETRACCA ET AL.

1997; Oliveira and Lourenço 2004; Petracca et al. computability and robustness of nonlinear consti­
2017b) or with Multi-Scale Computational- tutive models showing strain-softening.
Homogenization (De Bellis and Addessi 2009, 2011;
Massart 2003; Massart, Peerlings, and Geers 2007; All the numerical simulations shown in this work are
Mercatoris, Bouillard, and Massart 2009; Mercatoris carried out with the OpenSees (McKenna 2011) solver,
and Massart 2011; Petracca et al. 2016, 2017a; where the authors have implemented the proposed con­
Quinteros, Oller, and Nallim 2012; Zucchini and stitutive model. Pre- and post-processing are carried out
Lourenço 2002, 2009). with the STKO software (STKO Scientific ToolKit for
Among all the aforementioned approaches, this OpenSees).
work focuses on Micro-Modeling. Its main advantage
is the capability of capturing all the complex failure
mechanisms that can appear at the microstructural 2. Tension-compression plastic-damage model
level, the damage-induced anisotropy, and the inter­
The proposed tension-compression plastic-damage
action between microstructural components, even
model described herein is an extension of a tension-
when simple isotropic tensorial constitutive models
compression damage model previously formulated by
are used to model the constitutive behavior of the
the authors (Petracca et al. 2017b), which was in turn
microscopic components. However, micro-modeling
based on (Cervera, Oliver, and Faria 1995; Wu, Li, and
also has some drawbacks that limit its applicability to
Faria 2006). The original model was formulated in the
the simulation of small specimens, making it inap­
framework of classical continuum-damage mechanics,
plicable to real-life large-scale structures. When used
and therefore, inelastic permanent deformations were
to model large-scale structures, micro-modeling
not accounted for, rendering the model unsuitable for
obviously leads to extremely fine meshes and thus
simulating structures subjected to cyclic loading.
to costly analyses, both in terms of time and compu­
However, a nice feature of continuum-damage models
tational power. Furthermore, damage growth,
is that they can evaluate the stress tensor explicitly from
strength degradation, and strain localization intro­
the strain tensor, without the need for iterative loops at
duce the extra issue of poor convergence (or even
the constitutive level, which is common in the return
non-convergence at all) even when small time steps
mapping procedures in plasticity-based models. This
are used, increasing the computational cost even
makes the constitutive model calculations fast and
more. All the aforementioned drawbacks can be miti­
robust, which is a useful feature when analyzing large-
gated to a good extent using the right tools, such as
scale structures.
parallel computing and robust integration algorithms
The objective of the proposed new plastic-damage
for the constitutive models.
model is to introduce the description of plastic deforma­
This work presents a new plastic-damage model as an
tions in the existing damage model, keeping the simpli­
extension to a continuum damage model previously
city of the continuum-damage framework. For this
formulated by the authors (Petracca et al. 2017b). The
reason, plasticity has to be necessarily implemented in
proposed model adds two new aspects to the existing
a simplified way, such that the stress tensor can still be
model:
explicitly evaluated from the strain tensor as per stan­
dard continuum-damage models, without introducing
● Plasticity, to represent the inelastic deformation
iterative procedures at the constitutive level, thus keep­
and improve the representation of the masonry
ing the computational cost of the constitutive response
response under cyclic loadings. Since the focus of
as small as possible.
this work is to propose a fast and robust model for
the simulation of large-scale structures, plasticity is
introduced in a simplified fashion, avoiding itera­
2.1. Constitutive model
tive loops inside the constitutive model, thus
retaining the explicit evaluation of the stress tensor In the following description, all variables without sub­
from the strain tensor as in pure continuum scripts refer to the current time-step, while those with
damage models (Saloustros, Cervera, and Pelà the n and n 1 subscripts refer to the same variables at
2018). the two previous (known) time steps.
● IMPL-EX, a mixed implicit-explicit integration The nominal stress tensor σ is defined as
scheme (Oliver, Huespe, and Cante 2008) for the � �
constitutive models that aims at improving the σ¼ 1 dcþ σ þ þ 1 dc σ (1)
 INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 3

� � qffiffiffiffiffi � �
where σ þ and σ are the positive and negative parts, 1 ft
~τ þ ¼ H ðσ~max Þ α~I 1 þ 3~j2 þ βhσ~max i
respectively, of the effective stress tensor, while dcþ and 1 α fcp
dc are the positive and negative cracking damage vari­ (8)
ables. They account for stress reduction and stiffness
degradation of the effective stress due to the opening kb 1
of cracks. The effective stress tensor σ is defined as α¼ (9)
2kb 1
σ ¼ σþ þ σ (2)
fcp
β¼ ð1 αÞ ð1 þ αÞ (10)
� ft
�
σ� ¼ 1 dpl σ~� (3)
where ~τ þ and ~τ are the so-called equivalent (scalar)
where σ~þ and σ~ are the positive and negative parts of positive and negative stresses. ~I1 is the first invariant of
the trial (i.e., the elastic prediction) effective stress tensor the trial effective stress tensor, ~j2 is the second invariant of
σ~, while dpl þ and dpl are the positive and negative the trial effective deviatoric stress tensor, σ~max is the max­
plastic damage variables that account for stress reduc­ imum trial effective principal stress, fcp is the compressive
tion of the trial effective stress due to plastic flow. The peak stress, ft is the tensile strength, and kb is the ratio of
trial effective stress tensor σ~ is defined as the compressive bi-axial strength to the uniaxial compres­
sive strength. The constant k1 in equation (7) was pro­
σ~ ¼ σ n þ C0 : ðε εn Þ (4) posed by the authors to control the influence the
compressive criterion has on the dilatant behavior of the
X3 model (see Figure 1). For more information on this
σ~þ ¼ i¼1
hσ~i ipi � pi (5) aspect, the reader can refer to Petracca et al. 2017a.
The Heaviside functions of the maximum and mini­
mum eigenvalues, H ðσ~max Þ and H ð σ~min Þ, are required
σ~ ¼ σ~ σ~þ (6) to make sure the compressive surface is active only if at
least one eigenvalue is negative, and, accordingly, the
where σ n is the effective stress (i.e., only including plas­ tensile surface is active only if at least one eigenvalue is
ticity) at the previous time step n, while C0 : ðε εn Þ is positive. This is necessary for avoiding tensile damage
the elastic trial stress increment, σ~i is the ith eigenvalue of growth under purely compressive stress states and com­
σ~, and pi is its associated eigenvector. Note that in pressive damage growth under purely tensile stress
equation (4) the elastic trial predictor is performed in states. In fact, even if the positive and negative damage
an incremental way, while in the standard continuum- variables affect only the associated positive and negative
damage models the elastic prediction is made in total- parts of the stress, their failure surfaces, in equations (7)
strain. Note that equation (4) is equivalent to the more and (8), are a function of the whole stress tensor.
�
standard format σ~ ¼ C0 : ε εpn , where εpn is the
plastic strain tensor at the previous time step n.
However, due to the proposed simplified implementa­
tion of the plasticity part of the algorithm, the plastic
strain tensor does not necessarily need to be computed
and stored as an internal variable, so it is more conve­
nient to perform the elastic prediction as in equation (4).

2.2. Failure Criteria

The positive and negative failure criteria, or damage
surfaces, based on the work of Lubliner et al. (1989),
are defined as
� � qffiffiffiffiffi ��
1 ~ ~
~τ ¼ H ð σ~min Þ ~
αI 1 þ 3j2 þ k1 βhσ max i
1 α Figure 1. Damage surfaces in the principal stress space (2D plane
(7) stress case) (Petracca et al. 2017b).
4 M. PETRACCA ET AL.

Figure 2. Tensile hardening function.

It should also be noted that in equations (7) and (8),

the equivalent stresses ~τ þ and ~τ are functions of the
invariants of the trial effective stress tensor σ~, which is
computed incrementally as explained before. In this way
~τ þ and ~τ are not necessarily monotonically increasing
variables as required by the continuum-damage frame­
work. To be consistent with the standard continuum-
damage framework, the previously computed equivalent
stresses are corrected as
τ � ¼ ~τ � þ Eλn � (11)
where E is the Young’s modulus, and λn þ and λn are
the positive and negative equivalent plastic strains,
known from the previous time-step. In this way, the
equivalent stress measures are like the ones computed Figure 3. Compressive hardening function.
in a standard continuum-damage model from the total
strain.
nominal (including the effect of both plasticity and
damage) hardening variables, as
2.3. Plastic and cracking damage variables
r�
To impose the irreversibility of the damage process, the εtot � ¼ (14)
E
model introduces the so-called damage thresholds rþ
and r , two scalar variables that denote the largest values
ever reached by the equivalent stresses τ þ and τ during q� ¼ f � ðεtot � Þ (15)
the entire loading history for each time step t,
The hardening functions f þ and f of εtot � are reported
�
rþ ðt Þ ¼ max maxs2½0;t� τ þ ðsÞ; ft (12) in Figure 2 and Figure 3, respectively. For a detailed
description the reader can refer to Petracca et al. 2017a).
�
r ðt Þ ¼ max maxs2½0;t� τ ðsÞ; fc0 (13) The positive and negative plastic damage variables
can be computed as
where ft and fc0 are the elastic limits in tension and
compression, respectively. At this point, in view of the qpl �
dpl � ¼ 1 (16)
last term added in equation (11), rþ and r are two ~r�
scalar measures of the stress tensor as if they were
computed from the total strain. In fact, we can compute ~r� ¼ Eðεtot � λn � Þ (17)
their total-strain counterparts and the associated
 INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 5

With both plastic and cracking damage variables

correctly evaluated, the plastic effective stress tensor
and the nominal (damaged) stress tensor can be
finally calculated as per equations (3) and (1), respec­
tively, and the equivalent plastic strains can be
updated as

qpl �
λ� ¼ εtot � (20)
E
A schematic (uniaxial) representation of the above-
mentioned process is summarized in Figure 4, while
Figures 5 and 6 show some representative examples of
uniaxial and shear tests on a single integration point by
varying the magnitudes of ω� factors.

2.4. Key aspects of the simplified plastic-damage

algorithm
As described in the previous sections, the plastic part of
the proposed plastic-damage model has been implemen­
Figure 4. Schematic representation of the elastic predictor and ted in a simplified fashion, in order to achieve an explicit
the plastic and damage correctors. evaluation of the stress tensor from the current strain
tensor, minimizing the number of computations at the
qpl � ¼ q� þ ð1 ω� Þðr� q� Þ (18) constitutive level. Here is a summary of key points of the
simplified algorithm:
where ~r� are the positive and negative damage thresh­
olds associated with the trial effective stress tensor e σ, ● Some ingredients of standard plasticity theory, such
qpl � are the effective (plastic part only) hardening vari­ as yield surface and flow rule, are not present here.
ables, and ω� are the tensile and compressive plastic- Instead, the plastic corrector simply corresponds to
damage factors that can be used to calibrate the amount an isotropic scaling of the positive and negative
of plasticity to be considered. They range from 0 (pure parts of the trial elastic stress tensor, as in equation
damage) to 1 (pure plasticity). Once the plastic hard­ (3) via the plastic-damage variables dpl and dpl þ .
ening variables are computed, the cracking damage can ● Plastic (dpl � ) and cracking (dc � ) damage variables
be evaluated as share the same failure criteria in equations (7) and
(8), therefore they do not evolve independently.
q� ● How much one damage variable grows with respect
dc� ¼ 1 (19)
qpl � to the other is only controlled by the user-defined
tensile (ωþ ) and compressive (ω ) plastic-damage

Figure 5. Uniaxial tension-compression cyclic test on one integration point with (a) ωþ ¼ 0 and ω ¼ 0:0, and with (b) ωþ ¼ 0 and
ω ¼ 0:7.
6 M. PETRACCA ET AL.

Figure 6. shear cyclic test with a precompression on one integration point with (a) ωþ ¼ 0:5 and ω ¼ 0:5, and with (b) ωþ ¼ 1:0 and
ω ¼ 1:0.

factors. They can range from 0 (only the cracking global system matrix illconditioned. Instead, in the
damage evolves) to 1 (only the plastic damage explicit stage of the IMPL-EX algorithm, those internal
evolves). For materials such as mortar, the tensile variables are linearly extrapolated from the previous
factor can be lower than the compressive one to values at time tn 1 and tn as
reflect the more brittle nature of the tensile failure
with respect to the compressive failure. � Δtnþ1 � �
rnþ1 ¼ rn� þ r rn� 1 (21)
Δtn n
In this way, those variables and, in turn, the damage
2.5. Implementation of the IMPL-EX integration variables depend only linearly on the strain. There is,
scheme however, another source of nonlinearity: the tension-
It is well known that non-linear constitutive models compression split of the stress tensor as per equations
with strain softening lead to instabilities, and the con­ (5) and (6). Those equations can be re-written after
vergence of the iterative procedure becomes difficult to defining the rank-four positive and negative projectors
achieve or unachievable. To improve the stability and (Pelà, Cervera, and Roca 2011) as follows
robustness of this kind of problems, the so-called X3
IMPL-EX integration algorithm originally formulated Pþ ¼ i¼1
H ðσ~i Þpi � pi � pi � pi (22)
in Oliver, Huespe, and Cante (2008) is included in the
proposed tension-compression plastic-damage model. P ¼I Pþ (23)
The IMPL-EX algorithm is a mixed implicit/explicit
integration scheme for evaluating the internal variables where H ðσ~i Þ is the Heaviside function of the ith principal
of a constitutive model. The main idea is that the stress. Now equations (5) and (6) can be re-written as
computation of the constitutive model is divided into
σ~� ¼ P� : σ~ (24)
two main stages: an explicit extrapolation stage fol­
lowed by an implicit correction stage. In the explicit stage of the IMPL-EX algorithm, those
projectors are not computed as a function of the current
2.5.1. Explicit extrapolation trial stress but are set equal to their converged values at
During the global implicit iterative procedure to find the previous time step
equilibrium at time tnþ1 , the strain tensor is computed P� ¼ P� (25)
n
in each element, and it is sent to the constitutive model.
In the standard integration scheme, the internal vari­ At this point, all sources of nonlinearity are removed,
ables rþ and r are nonlinear functions of the trial and the global problem becomes step-wise linear.
effective stress tensor, as per equations (12) and (13), Furthermore, the consistent tangent stiffness matrix
and therefore, they depend nonlinearly on the current now coincides with the secant matrix:
trial strain tensor. This nonlinear dependence makes the n �� � �� � o
global problem nonlinear. The real issue, however, is the Ctan ¼ 1 dcþ 1 dplþ Pþ þ 1 dc 1 dpl P
presence of strain softening. In this case, the consistent : C0
tangent matrix required for the Newton-Raphson
(26)
scheme may have negative eigenvalues, rendering the
 INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 7

2.5.2. Implicit correction Table 1. Material parameters used in the numerical simulation.
Once the global implicit iterative procedure has con­ Parameter Description UnitMortar Brick
E Young’s modulus 2
verged and found equilibrium at time tnþ1 , all the trial N=mm 533.0 2171.0
ν Poisson’s ratio - 0.15 0.15
internal variables are saved as converged variables, i.e., ft Tensile strength N=mm2 0.1 1.62
the starting point for the next time step. Gt Tensile fracture energy N=mm 0.08 0.1
fcp Compressive strength at peak N=mm2 6.2 6.2
rn� 1 ¼ rn� (27) εcp Compressive deformation at peak - 0.015 0.008
Gc Compressive fracture energy N=mm 80.0 5.0
k1 Damage surface: shear - 0.16 0.16
rn� ¼ r� (28) compression reduction
ωþ Tensile plastic-damage factor - 0.6 0
ω Compressive plastic-damage factor - 1.0 0
λ�
n ¼λ
�
(29) Note that in the numerical model we used, for both bricks and mortar joints,
the compressive strength of the homogenized masonry. This is necessary
for 2D and Shell models, due to the plane-stress assumption: Due to the
In the IMPL-EX scheme, before doing the aforemen­ large difference in their elastic constants, when subjected to compressive
tioned internal variables swap, a standard implicit stress states, mortar is in triaxial-compression (thus increasing its strength
update of the current internal variables is performed. against the vertical stress), while brick is in biaxial-tension/uniaxial-
compression (thus increasing its strength against the vertical stress). In
This way, the error generated by the explicit extrapola­ plane-stress conditions this is not possible, therefore a simple remedy is to
tion scheme is mitigated, and due to the nature of the use, for both constituents, the compressive strength of the masonry itself.
explicit extrapolation, the time-step should be small
enough compared to standard implicit schemes to keep
the error under control.

3. Numerical application
The proposed model, in its original formulation based
purely on Continuum-Damage mechanics and
a standard implicit integration scheme, has been exten­
sively tested against benchmark problems and small
masonry specimens under monotonic loading condi­
tions. In Petracca et al. (2017b) the model has been
first tested in simple 1-element benchmarks to evaluate
the shear response against different values of vertical
pre-compression both in terms of shear strength and
dilatancy, and finally it has been used to simulate experi­
mental tests on the in-plane behavior of small masonry
shear walls. Instead, in Petracca et al. (2017a), it has been
tested against a masonry wall under monotonic out-of-
plane actions.
This work, instead, aims at assessing the capability of
the new model, equipped with the plasticity part to
Figure 7. Schematic representation of the tested structure and
represent permanent inelastic deformations and applied load, taken from .Magenes, Calvi, and Kingsley (1995)
a mixed implicit-explicit integration algorithm to
improve convergence and stability, to simulate large-
scale structures under cyclic loads. A two-story unrein­
forced masonry (URM) building tested at the University equal to 16 MPa, and mortar made of a mix of hydraulic
of Pavia, and described in detail in Magenes, Calvi, and lime and sand (1:3 volume) with a compressive strength
Kingsley (1995), is used and simulated numerically to ranging from 2 to 3 MPa. A complete description of
assess the effectiveness and robustness of the proposed material parameters is given in Binda et al. (1995).
plastic-damage model. For a complete description of the Material properties for the plastic-damage model used
model prototype and the testing procedure, the user can for bricks and mortar joints are given in Table 1.
refer to Magenes, Calvi, and Kingsley 1(995) and the The structure consists of four components named
references therein. Materials were chosen to match typi­ “Door Wall,” “Window Wall,” and two “Transverse
cal old urban buildings in Italy, i.e., solid fired-clay Walls,” subjected to a quasi-static cyclic loading proto­
bricks with a mean compressive strength on cubes col under displacement-control, as shown in Figure 7.
8 M. PETRACCA ET AL.

The numerical model was created in STKO pre- post-

processor (STKO Scientific ToolKit for OpenSees),
which automatically produced the TCL input files for
the OpenSeesMP parallel solver. OpenSeesMP is
a parallel version of OpenSees suitable for running either
parallel parametric analysis of small models or a single
analysis of a large-scale model. Even if the two structural
systems can be efficiently analyzed as two separate 2D
analyses due to the weak coupling between them, the
objective of this work is to show the performance and
robustness of the proposed model in large-scale cyclic
analyses. For this reason, we decided to model the whole
structure in a three-dimensional shell model, as shown
in Figure 8, by discretizing the units and mortar bed/
head joints of masonry separately as is usual in the
micro-modeling approach.
The numerical model consists of 4-node shell ele­
ments with a 2 × 2 Gauss integration, five through-the-
thickness integration points for bricks and mortar joints,
contact elements at the disconnected edges between the
Door Walls and the two Transverse Walls, and Force-
Figure 8. Micro-model developed in STKO’s pre and post- Based Beam elements for the flexible diaphragm.
processors. The adopted shell element is the ASDShellQ4
(ASDShellQ4 User Manual) element implemented by
The report describes how the Door Wall is substan­ the authors in OpenSees. The ASDShellQ4 element is
tially detached from the other three walls so that the a 4-node general purpose thick shell element. The mem­
Window Wall and the two Transverse Walls form brane behavior is enhanced with the AGQ6-I (Chen
a structural system that is almost independent from the et al. 2004) formulation, which makes the element
Door Wall. There is only a weak coupling represented by almost insensitive to geometry distortion, as opposed
the flexible floor steel beams. to standard iso-parametric elements. The drilling DOF

Figure 9. Partitioning of the mesh for parallel computing.

 INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 9

Figure 10. Force-Displacement curve for the Door Wall.

Figure 11. Force-Displacement curve for the Window Wall and the two Transverse Walls.

(rotation about the normal axis) is treated with the The analysis is performed using a Displacement-
Hughes-Brezzi (Hughes and Brezzi 1989) formulation, Control integrator with the cyclic displacement pro­
with special care to avoid membrane locking, using a 1 tocol described in Magenes, Calvi, and Kingsley
point quadrature plus hourglass-stabilization. Finally, (1995). The imposed displacement-increment for the
the plate bending part is treated using the MITC4 Displacement-Control integrator is about 2.5 mm,
(Dvorkin, and Bathe 1984; Bathe and Dvorkin 1985) which has proven to be sufficiently small to keep
formulation to avoid the well known transverse shear the integration error of the IMPL-EX scheme under
locking behavior of thick plate elements. an acceptable threshold.
The model is partitioned into 24 sub-domains to be Figure 10 shows the force-displacement curve
used with 24 processors in the parallel analysis. The obtained at the Door Wall, while Figure 11 shows the
whole mesh consists of roughly 177 thousand elements force-displacement curve obtained at the assembly made
and 60 thousand nodes, with about 7500 elements per of the Window Wall and the two Transverse Walls. The
process. The partitioning of the mesh is shown in overall prediction of the numerical model is in good
Figure 9. agreement with the experimental results, both in terms
10 M. PETRACCA ET AL.

Figure 12. Crack pattern obtained at the end of the analysis for the Door Wall (right), compared with the experimental result (left),
taken from Magenes, Calvi, and Kingsley (1995)

Figure 13. Crack pattern obtained at the end of the analysis for the Window Wall (right), compared with the experimental result (left),
taken from .Magenes, Calvi, and Kingsley (1995)

of maximum force and hysteresis loop. The main differ­ Magenes, Calvi, and Kingsley (1995). The observed fail­
ence that can be observed from the two curves is that the ure mechanisms are reproduced satisfactorily.
numerical model shows more plastic deformations dur­
ing the first cycles compared to the experimental results.
4. Computational costs
In the experimental results, it seems like the first cycles
produce almost only strength and stiffness degradation, The entire mesh of the case study consists of roughly 177
while plastic deformations are more predominant dur­ thousand elements and 60 thousand nodes, 6 DOFs per
ing the last cycles. This observation offers a starting node, for a total of about 360 thousand equations. The
point for future work, as in the current implementation mesh is partitioned into 24 sub-domains and analyzed
the plastic-damage factors ωþ and ω are constant dur­ with OpenSeesMP. The model ran on a 128-cores AMD
ing the analysis. A possible improvement could be an Ryzen Threadripper 3990 × 3.40 GHz CPU. The analysis
evolutionary law for plastic-damage factors so that the was subdivided into 3200 time-steps and ran in 2 hours
amount of plastic deformations can be controlled over and 53 minutes only, with a constant number of itera­
time. tions for each time-step equal to 2, thanks to the fact that
Figure 12 and 13 show the numerically obtained the IMPL-EX algorithm produces a step-wise linear
crack pattern at the end of the cyclic analysis, compared solution, that is, within each time step, the stress
with the experimental observations reported in response does not depend on the trial strain. The results
 INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 11

in terms of computational time are remarkable and (Multilevel evaluation of seismic vulnerability and risk mitiga­
clearly demonstrate the efficiency of the proposed con­ tion of masonry buildings in resilient historical urban centres,
stitutive model combined with a mixed implicit explicit ref. num. RTI2018-099589-B-I00).
integration algorithm.
Disclosure statement
5. Conclusions No potential conflict of interest was reported by the author(s).

This paper presented an efficient constitutive model based

on continuum damage mechanics and improved with the Funding
representation of inelastic deformations to better repre­
This work was supported by the European Regional
sent the plastic and cyclic response of masonry. The Development Fund.
simplified implementation of the inelastic deformations
allows an explicit computation of the stress tensor from
the input strain tensor without the need for an iterative ORCID
loop at the material level. The novel constitutive model
M. Petracca http://orcid.org/0000-0001-5083-6533
has been proposed for the continuous micro-modeling of E. Spacone http://orcid.org/0000-0002-8304-7028
masonry, i.e., by discretizing both units and mortar-joints
with continuum finite elements.
Micro-modeling is a tool that can accurately represent References
the complex nonlinear response of materials such as STKO Scientific ToolKit for OpenSees.https://asdeasoft.net/stko/.
masonry, with a highly heterogeneous micro-structure Accornero, F., and G. Lacidogna. 2020. Safety assessment of
where each constituent exhibits strongly nonlinear beha­ masonry arch bridges considering the fracturing benefit.
vior. The presence of the microstructure in the numerical Applied Sciences 10 (10):3490. doi:10.3390/app10103490.
mesh captures the complex geometric interactions and Accornero, F., G. Lacidogna, and A. Carpinteri. 2016.
Evolutionary fracture analysis of masonry arches: Effects
failure mechanisms that arise at the micro-scale extremely of shallowness ratio and size scale. Comptes Rendus
well. Unfortunately, the computational cost is prohibi­ Mécanique 344:623–30. doi:10.1016/j.crme.2016.05.002.
tively high and can drastically increase when standard Accornero, F., G. Lacidogna, and A. Carpinteri. 2018.
implicit algorithms are used to solve the nonlinear equa­ Medieval arch bridges in the lanzo valleys, Italy: case studies
tions due to the brittle nature of the masonry constituents on incremental structural analysis and fracturing benefit.
and, thus the ill conditioning of the system matrix. For Journal of Bridge Engineering (ASCE) 23 (7):05018005.
doi:10.1061/(ASCE)BE.1943-5592.0001252.
this reason, in the literature, micro-modeling has been ASDShellQ4. User Manual, https://opensees.github.io/
used mainly for the analysis of small-scale specimens. OpenSeesDocumentation/user/manual/model/elements/
However, the aforementioned drawbacks can be satisfac­ ASDShellQ4.html
torily mitigated by considering parallel computing, as it Barenblatt, G. I. 2014. Flow, deformation and fracture: Lectures
offers an increase in speed by partitioning the entire on fluid mechanics and the mechanics of deformable solids
for mathematicians and physicists, 49. Cambridge
domain into many smaller sub-domains that can be ana­
University Press.
lyzed simultaneously. The IMPL-EX algorithm removes Bathe, K. J., and E. N. Dvorkin. 1985. A four-node plate
the strong nonlinearity given by the plastic-damage bending element based on Mindlin/Reissner plate theory
model, thus improving convergence and robustness and and a mixed interpolation. International Journal for
consequently reducing the computation time. If, as in this Numerical Methods in Engineering 21 (2):367–83. doi:10.
case, the only source of nonlinearity is the constitutive 1002/nme.1620210213.
Bažant, Z. P. 2004. Scaling theory for quasibrittle structural
model (i.e., other nonlinearities are not considered, such failure. Proceedings of the National Academy of Sciences, U.
as large deformations), the nonlinear problem becomes S.A., 101:13400–07.
step-wise linear, and convergence is guaranteed. Binda, L., C. Tiraboschi, G. Mirabella Roberti, G. Baronio, and
G. Cardani. 1995. Experimental and numerical investiga­
tion on a brick masonry building prototype – Measuring
Acknowledgments masonry material properties: Detailed results from an
extensive research, Report 5.0. Dipartimento di Ingegneria
The last author gratefully acknowledges the financial support Strutturale, Politecnico di Milano.
from the Ministry of Science, Innovation and Universities of Carpinteri, A. 1994. Scaling laws and renormalization groups
the Spanish Government (MCIU), the State Agency of for strength and toughness of disordered materials. Int.
Research (AEI), as well as that of the ERDF (European J. Solids Struct 31 (3):291–302. doi:10.1016/0020-7683(94)
Regional Development Fund) through the project SEVERUS 90107-4.
12 M. PETRACCA ET AL.

Carpinteri, A., and B. Chiaia. 1997. Multifractal scaling laws in Mercatoris, B., P. Bouillard, and T. Massart. 2009. Multi-scale
the breaking behaviour of disordered materials. Solitons & detection of failure in planar masonry thin shells using
Fractals. 135–50. Chaos computational homogenisation. Engineering Fracture
Carpinteri, A., and S. Puzzi. 2009. he fractal-statistical Mechanics 76 (4):479–99. doi:10.1016/j.engfracmech.2008.
approach to the size-scale effects on material strength and 10.003.
toughness. Probabilistic Engineering Mechanics Mercatoris, B., and T. Massart. 2011. A coupled two-scale
24 (1):75–83. doi:10.1016/j.probengmech.2008.01.003. computational scheme for the failure of periodic
Cervera, M., J. Oliver, and R. Faria. 1995. Seismic evaluation of quasi-brittle thin planar shells and its application to
concrete dams via continuum damage models. Earthquake masonry. International Journal for Numerical Methods in
Engineering & Structural Dynamics 24:1225–45. doi:10. Engineering 85 (9):1177–206. doi:10.1002/nme.3018.
1002/eqe.4290240905. Oliveira, D. V., and P. B. Lourenço. 2004. Implementation and
Chen, X.-M., S. Cen, Y.-Q. Long, Z.-H. Yao. 2004. Membrane validation of a constitutive model for the cyclic behaviour of
elements insensitive to distortion using the quadrilateral interface elements. Computers & structures 82:1451–61.
area coordinate method. Computers & Structures. 82 Oliver, J., A. E. Huespe, and J. C. Cante. 2008. An implicit/
(1):35–54. doi:10.1016/j.compstruc.2003.08.004. explicit integration scheme to increase computability of
De Bellis, M. L., and D. Addessi. 2009. A Cosserat based multi- non-linear material and contact/friction problems.
scale technique for masonry structures. PHD Thesis. Computer Methods in Applied Mechanics and Engineering
De Bellis, M. L., and D. Addessi. 2011. A Cosserat based 197 (21–24):1865–89. doi:10.1016/j.cma.2007.11.027.
multi-scale model for masonry structures. International Panian, R., and M. Yazdani. 2020. Estimation of the service
Journal for Multiscale Computational Engineering load capacity of plain concrete arch bridges using a novel
9 (5):543–63. doi:10.1615/IntJMultCompEng.2011002758. approach: Stress intensity factor. Structures 27:1521–34.
Drougkas, A., L. Pelà, and P. Roca. 2014. Numerical modelling doi:10.1016/j.istruc.2020.07.055.
of masonry shear walls failure mechanisms. Proceedings of Pelà, L., M. Cervera, S. Oller, and M. Chiumenti. 2014.
9th International Masonry Conference, Guimarães, A localized mapped damage model for orthotropic
Portugal. materials. Engineering Fracture Mechanics 124:196–216.
Dvorkin E. N. and Bathe K. 1984. A continuum mechanics Pelà, L., M. Cervera, and P. Roca. 2011. Continuum damage
based four-node shell element for general non-linear ana­ model for orthotropic materials: Application to masonry.
lysis. Engineering Computations 1(1):77–88. doi:10.1108/ Computer Methods in Applied Mechanics and Engineering
eb023562. 200 (9–12):917–30. doi:10.1016/j.cma.2010.11.010.
Hughes, T. J. R., and F. Brezzi. 1989. On drilling degrees of Pelà, L., M. Cervera, and P. Roca. 2013. An orthotropic
freedom. Computer Methods in Applied Mechanics and damage model for the analysis of masonry structures.
Engineering 72 (1):105–21. doi:10.1016/0045-7825(89) Construction and Building Materials 957–67. doi:10.1016/j.
90124-2. conbuildmat.2012.07.014.
Lourenço, P. B. 1996. Computational strategies for masonry Petracca, M., L. Pelà, R. Rossi, S. Oller, G. Camata, and
structures. Ph.D. thesis, TU Delft, Delft University of E. Spacone. 2016. Regularization of first order computa­
Technology tional homogenization for multiscale analysis of masonry
Lourenço, P. B., and J. Rots. 1997. Multisurface interface structures. Computational Mechanics 57:257–76. doi:10.
model for analysis of masonry structures. Journal of 1007/s00466-015-1230-6.
Engineering Mechanics 123:75–83. 7 1997] has been Petracca, M., L. Pelà, R. Rossi, S. Oller, G. Camata, and
updated. OK. doi:10.1061/(ASCE)0733-9399(1997) E. Spacone. 2017a. Multiscale computational first order
123:7(660). homogenization of thick shells for the analysis of out-of-
Lubliner, J., J. Oliver, S. Oller, and E. Oñate. 1989. A plane loaded masonry walls. Computer Methods in Applied
plastic-damage model for concrete. International Journal Mechanics and Engineering 315:273–301. doi:10.1016/j.cma.
of Solids and Structures 25 (3):299–326. doi:10.1016/0020- 2016.10.046.
7683(89)90050-4. Petracca, M., L. Pelà, R. Rossi, S. Zaghi, G. Camata, and
Magenes, G., G. M. Calvi, and G. R. Kingsley. 1995. Seismic E. Spacone. 2017b. Micro-scale continuous and discrete
testing of a full-scale, two-story masonry building: Test numerical models for nonlinear analysis of masonry shear
procedure and measured experimental response. walls. Construction and Building Materials 149:296–314.
Experimental and numerical investigation on a brick doi:10.1016/j.conbuildmat.2017.05.130.
masonry building prototype - numerical prediction of the Quinteros, R. D., S. Oller, and L. G. Nallim. 2012. Nonlinear
experiment, Report 3.0. Gruppo Nazionale La Difesa Dai homogenization techniques to solve masonry structures
Terremoti, University of Pavia, Pavia, Italy. problems. Composite Structures 94 (2):724–30. doi:10.
Massart, T. J. 2003. Multi-scale modeling of damage in 1016/j.compstruct.2011.09.006.
masonry structures. Ph.D. thesis. Saloustros, S., M. Cervera, and L. Pelà. 2018. Tracking
Massart, T. J., R. Peerlings, and M. Geers. 2007. An enhanced multi-directional intersecting cracks in numerical model­
multi-scale approach for masonry wall computations with ling of masonry shear walls under cyclic loading. Meccanica
localization of damage. International Journal for Numerical 53 (7):1757–76. doi:10.1007/s11012-017-0712-3.
Methods in Engineering 69:1022–59. doi:10.1002/nme.1799. Wu, J. Y., J. Li, and R. Faria. 2006. An energy release rate-based
McKenna, F. 2011. OpenSees: A framework for earthquake plastic-damage model for concrete. International Journal of
engineering simulation. Computing in Science & Solids and Structures 43:583–612. doi:10.1016/j.ijsolstr.
Engineering 13:58–66. doi:10.1109/MCSE.2011.66. 2005.05.038.
 INTERNATIONAL JOURNAL OF ARCHITECTURAL HERITAGE 13

Yazdani, M., and H. Habibi. 2021. Residual capacity evalua­ Journal of Solids and Structures 39 (12):3233–55. doi:10.
tion of masonry arch bridges by extended finite element 1016/S0020-7683(02)00230-5.
method. Structural Engineering International 1–12. doi:10. Zucchini, A., and P. B. Lourenço. 2009. A micro-mechanical
1080/10168664.2021.1944454. homogenisation model for masonry: Application to shear
Zucchini, A., and P. B. Lourenço. 2002. A micro-mechanical walls. International Journal of Solids and Structures 46 (3–
model for the homogenisation of masonry. International 4):871–86. doi:10.1016/j.ijsolstr.2008.09.034.