Engineering Structures 60 (2014) 100–110

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Identification of material parameters for low bond strength masonry

V. Sarhosis ⇑, Y. Sheng
School of Civil Engineering, University of Leeds, LS2 9JT, Leeds, UK

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

Article history: This paper investigates the parameter estimation problem for masonry constitutive models and describes
Received 29 May 2013 the development of a computational model for low bond strength masonry. The conventional method of
Revised 3 December 2013 obtaining material parameters from the results of small sample tests was thought to be problematic.
Accepted 11 December 2013
Instead, material parameters were obtained from the results from the load testing of large clay brick
Available online 16 January 2014
low bond strength masonry wall panels in the laboratory. Initially, the panels were modeled computa-
tionally using UDEC software and an assumed set of material parameters. The differences between the
Keywords:
results obtained experimentally and computationally were minimized using an optimization technique.
Material parameter identification
Brickwork masonry
The model was then used with the optimized parameters to predict the structural response of another
Discrete Element modeling wall panel that had been tested in the laboratory. Good correlation was obtained with the experimental
Optimization results. The general procedure of the material parameter identification method and the numerical results
are presented.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction identified in order to characterize the behavior of the masonry.

Many forms of masonry construction exhibit low bond strength
characteristics. With the term low bond strength masonry, we de-
fine masonry where the bond at the masonry unit/mortar joint
interface is sufficiently low to have a dominant effect on the
mechanical behavior such as the formation of cracks, re-distribu-
tion of stresses after cracking and the formation of collapse mech-
anisms. Modern mortars used for new construction tend to be
cementitious with a relatively high cement content. As a result,
the strength of the bond at the interface between the masonry
units and the mortar joints tends to be much higher than is the
case with other mortars. Low bond strength masonry may encoun-
tered: (a) in historic constructions where lime mortar were mainly
used; (b) masonry arch bridges, tunnels linings and earth retaining
walls where unit/mortar joint bond has been disrupted by the ac-
tion of water leeching through the masonry; and (c) in more recent
examples of masonry constructed with low cement content mortar
due to lack of quality control on site. Engineers often need the use
of relatively sophisticated numerical or analytical methods in order
to predict the in-service behavior and load carrying capacity of
such structures and inform repair or strengthening decisions.
However, any numerical or analytical model of analysis requires
some forms of constitutive model to simulate the mechanical re-
sponse of structures under various loading conditions. Constitutive
models also require a number of input material parameters to be
⇑ Corresponding author. Address: School of Engineering, Cardiff University, CF24
3AA Cardiff, Wales, UK. Tel.: +44 7725071212.
E-mail address: SarhosisV@cardiff.ac.uk (V. Sarhosis).
In the last few years, with the development of the sophisticated
numerical models, the number of parameters for material models
has increased significantly [1]. It is common practise to determine
such material parameters from the results of various, relatively
simple, small scale laboratory experiments. However, masonry is
a highly variable, stress-state type dependant material which
experiences non-uniform distributions of stress in real structures.
Furthermore, the boundary conditions in small scale tests are un-
likely to be representative of those exist in a larger structure
[2,3]. These have cast doubts on the effectiveness of determining
material parameters that are representative of masonry from small
scale experiments.
The aim of this study is to investigate the material parameter
identification problem and develop a computational model for
low bond strength masonry. Given the importance of the masonry
unit-mortar interface on the structural behavior of low bond
strength masonry [4–6], the micro-modeling approach based on
the Discrete Element method of analysis have been used. The soft-
ware used was the Universal Distinct Element Code (UDEC). The
material parameters for the masonry constitutive model were ob-
tained from the results of tests carried out in the laboratory on sin-
gle leaf wall panels, each containing a large opening. Each panel
was subjected to a gradually increasing vertical in-plane static
point load until it collapsed. The wall panel tests were also mod-
eled using UDEC. The load to cause first visible cracking, the ulti-
mate load and the load versus mid-span displacement
relationship obtained from the laboratory tests are compared with
the computational predictions obtained from DEM modeling. An
optimization procedure was then used to tune the parameters used

0141-0296/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.engstruct.2013.12.013
 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110
initially in UDEC in order to better simulate the pre- and post-
cracking behavior and the behavior close to collapse of the panels
tested in the laboratory. Having obtained the material parameters
for UDEC, the model was then validated by comparing the UDEC-
predicted behavior of wall panels different to those used to deter-
mine the material parameters, with those tested in the laboratory.
2. Overview of modeling low bond strength masonry with UDEC
UDEC is a two dimensional Discrete Element software and was
developed initially to model sliding rock masses in which failure
occurs along the joints [7]. This has similarities with the behavior
of low bond strength masonry. Typical examples of masonry struc-
tures that have been modeled with UDEC include masonry arches
and aqueducts [8–10]; wall panels [11,12]; and retaining walls
[13].
In UDEC, bricks can be represented as an assemblage of rigid or
deformable distinct blocks which may take any arbitrary geometry.
Rigid blocks do not change their geometry as a result of any ap-
plied loading and are mainly used when the behavior of the system
is dominated by the mortar joints. Deformable blocks are internally
discretised into finite difference triangular zones and each element
responds according to a prescribed linear or non-linear stress–
strain law. Mortar joints are represented as zero thickness inter-
faces between the blocks. The interactions between the blocks at
interfaces are governed by point contacts (Fig. 1) located at the
edges or corners of the bricks and the zones [9].
For each contact point, there are two spring connections. These
can transfer a force (or stress) either into a normal (Fn or rn) or a
shear (Fs or s) component from one brick to the other, see Fig. 2.
The response to normal loading is expressed using the normal stiff-
ness (JKn) and normal displacement (DUn):
Drn ¼ JKnDU n
The response to shear is calculated using the shear stiffness (JKs)
and shear displacement (DUs)
Ds ¼ JKsDU s
According to Souley [14], the value of normal stiffness of the inter-
face depends on the contact area between the joint surfaces and the
relevant properties of the mortar joints. Also, the value of the shear
stiffness depends on the roughness of the joint surfaces. The
‘‘Coulomb-slip model’’ is the preferred option to simulate
the mechanical behavior of the mortar joints [15]. This includes
the following parameters: the normal stiffness (JKn), the shear stiff-
ness (JKs), the friction angle (Jfric), the cohesion (Jcoh), the tensile
strength (Jten) and the dilation angle (Jdil). For this model, there
is a limiting tensile strength (Jten) for the mortar. If the tensile
strength exceeds the normal stress, then the normal stress becomes
Fig. 1. Deformable blocks with contact points [9].
101
zero. A displacement weakening response with the use of the ‘‘Cou-
lomb slip model with residual strength’’ is also available in UDEC.
However, the tensile strength of low bond strength masonry is low-
er than the conventional masonry. Thus, any tension softening (or
residual strength) values are considered to be insignificant when
modeling low bond strength masonry. Thus, tension softening has
been assumed to be negligible at this study.
Also, in UDEC the unknowns are obtained by solving explicitly
the differential equations of Newton’s Second law of motion at
all blocks and the force–displacement law at all contacts. The
force–displacement law is used to find the contact forces from
known displacements while Newton’s second law gives the motion
of the blocks resulting from the known forces acting on them. Con-
vergence to static solutions is obtained by means of adaptive
damping, as in the classical dynamic relaxation methods [16]. In
this way, large displacements along the mortar joints and the rota-
tions of the bricks are allowed with the sequential contact detec-
tion and update [15].
3. Conventional methods for masonry material parameter
identification
Conventionally, material parameters for masonry constitutive
models are determined directly from the results of compressive,
tensile and shear strength tests on small masonry prisms (Fig. 3).
These usually consist of assemblages of masonry made up with a
small number of bricks and mortar joints. It is usually assumed
that the stress and strain fields in the specimen are uniform. In
other cases, separate tests are carried out on material samples,
such as masonry units and/or mortar specimens [17,18]. The
testing of small specimens is simple, relatively inexpensive and
involves little specialist equipment.
ð1Þ However, the conventional approach of material parameter
identification is considered to be problematic and may not produce
material parameters that are representative for masonry. As
identified by Hendry [3], brick and mortar properties are highly
ð2Þ variable and depend primarily on the local supply of raw materials
and manufacturing methods. Also, the assumption that the stress
and strain in the specimen are uniform is not applicable for
masonry which is an intrinsically inhomogeneous material. More-
over, the simple conditions under which the small specimens are
tested in the laboratory do not usually reflect the more complex
boundary conditions, the combinations of stress-state types and
load spreading effects that exist in a large scale masonry structure.
In addition, some of the parameters obtained from small scale
tests are variable and sensitive to the method of testing. This is
likely to be due to the combined effects of eccentric loading, stress
concentrations and variations in the resistance to applied stress
that are likely to exist in the test specimens [3]. According to
Vermeltfoort [20], the effects of boundary conditions such as
platen restraint and the shape and size of the test specimen can
have a significant influence on the magnitude of the measured
parameter. Also, the restraint conditions on the mortar in the cube
test will be different to those existing in the mortar joint between
masonry units.
The situation is made more complex when workmanship is
considered. Usually, a much higher standard and consistency of
workmanship will be achieved by constructing small scale test
specimens in the laboratory compared with the construction of
larger masonry structures. Such variations in workmanship will
not be captured if the material parameters are based on the results
from the small scale experiments [21]. Due to these difficulties, it is
often necessary to adjust the material parameter values obtained
from small scale experiments before they can be used in the
numerical model.
102 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110
Fig. 2. Mechanical representation of the contact between blocks [15].
Fig. 3. Different types of shear tests: (a) couplet test; (b) Van der Pluijim test [18] and (c) triplet test [19].
The authors have found a further complication when using the
Discrete Element Method (DEM) to model masonry. As the mate-
rial parameters define the characteristics of the zero thickness
interfaces between the mortar joints and the blocks, they can be
difficult to measure directly from physical tests. The UDEC user
manual [15] states: ‘‘it is important to recognize that joint properties
measured in the lab typically are not representative of those for real
joints in the field’’. It further states that ‘‘scale dependence of joint
properties is a major question and the only way to guide the choice
of appropriate parameters is by comparison to similar joint properties
derived from field tests’’. The use of field test results presents an-
other set of difficulties. The stress and strain levels that are found
in structures in the field are likely to be very low and affected by
effects such as moisture movements, shrinkage and creep. Any
material parameters determined from field measurements are un-
likely to represent the behavior of masonry in the post-cracking
and near-collapse conditions. Other factors such as load spreading
effects, residual thermal stresses in bricks and large inclusions
sometimes found in bricks are all contribute to the uncertainty of
material parameters obtained from small scale experiments. Final-
ly, another problem encountered when using UDEC to model ma-
sonry is that, because the use of UDEC is relatively new for
masonry, there is very little reliable guidance available on suitable
material parameters.
4. Proposed method for material parameter identification
An alternative method of identifying material parameters for
masonry that better reflects the complex nature of masonry and
the range of stress state types that exist in practice is adopted here.
The method was initially proposed by Toropov [22] and is based on
an advanced optimization of the responses of relatively complex or
‘‘non-trivial’’ large scale masonry elements. According to the meth-
od, numerical analysis for each large scale experiment is carried
out and values of material parameters are tuned so that the differ-
ence between responses measured from a series of large scale lab-
oratory experiments and those obtained from the numerical
simulation can be minimized. Initially, a range of material
parameters based on results of conventional small-scale experi-
ments or on the codes of practice or on engineering judgment
are used in the model for the simulation of the large experiments.
These material parameters can then be modified and tuned
through an optimization process in which the function to be min-
imized is an error function that expresses the difference between
the responses measured from the large scale experiments and
those obtained from the numerical analysis. Such responses can in-
clude: failure load, load at initial visual cracks, load–deflection
characteristics etc. The use of optimization software, such as Altair
HyperStudy [23], is essential for the implementation of the optimi-
zation process. The proposed method of material parameter iden-
tification is illustrated in Fig. 4.
The aim of the identification problem is to obtain the optimum
estimate of the unknown model parameters taking into account
uncertainties which may exist in the problem, such as the inherent
variation of material properties, experimental errors and errors in
the model estimation method. The estimates of the material
parameters obtained from this approach could be referred to as
the ‘‘maximum likelihood estimates’’ and can be used to ‘‘inform’’
the computational model. Examples validating the use of the pro-
posed material parameter identification technique for large defor-
mation plasticity models include: (a) test data of a solid bar in
torsion [24] and (b) test data for the cyclic bending of thin sheets
[25]. Later, Morbiducci [26] applied the method to two different
masonry problems in order to: (a) identify the parameters of a
non-linear interface model [27] to describe the shear behavior of
masonry joints under monotonic loading, where shear tests were
chosen as the experimental tests; and (b) to evaluate the parame-
ters of a continuum model for brick masonry walls under cyclic
loading [28]. Based on the above mentioned studies, the following
points should be considered:
(a) when modeling masonry, different material parameters
influence different stages of mechanical behavior;
(b) large number of full scale experiments may be required; and
(c) a significant amount of computational time is required to
carry out parameter sensitivity studies.
 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110 103

Fig. 4. Schematic chart for the proposed material parameters identification method.

2 2
F 1ðxÞ ¼ R½ðRexp comp exp comp
1;1  R1;1 Þ þ ðR1;2  R1;2 Þ   
4.1. Formulation of the alternative material parameter identification
2
problem as an optimization problem þ ðRexp comp
1;n  R1;n Þ  ð3Þ
2 2
Consider an experimental test performed on M ¼ 1; 2; . . . ; m F 2ðxÞ ¼ R½ðRexp comp exp comp
2;1  R2;1 Þ þ ðR2;2  R2;2 Þ   
specimens in order to estimate the design variables or unknown 2
þ ðRexp comp
2;n  R2;n Þ  ð4Þ
parameters P ¼ 1; 2; . . . ; p of the constitutive model. Let
N ¼ 1; 2; . . . n be the number of responses of interest recorded from ..
the experimental data. Also, consider Rexp as the value of the nth .
n
2 2 2
measured response quantity corresponding to the large scale Fm exp comp exp comp exp comp
ðxÞ ¼ R½ðRm;1  Rm;1 Þ þ ðRm;2  Rm;2 Þ   þ ðRm;n  Rm;n Þ  ð5Þ
experiment carried out in the laboratory. Consider Rcomp n as the va-
lue of the nth measured response quantity corresponding to the FM
ðxÞ ¼ F 1ðxÞ þ F 2ðxÞ þ  þ Fm
ðxÞ
computational simulation. Then, the responses R are functions of
the design variables of the model P. The model takes the general is a dimensionless function. The problem is then to find the vector
function form x ¼ RðPÞ. To calculate values of this specific function x = [x1, x2, x3 . . . xp] that minimizes the objective function:
for the specific set of parameters, x, requires the use of a non-linear F total M M
Ai  X i  Bi ði ¼ 1 . . . . . . NÞ
ðxÞ ¼ Rh ðF ðxÞÞ; ð6Þ
numerical simulation (e.g. discrete/finite element) of the experi-
mental test under consideration. The difference between the where F total
ðxÞ is a function of the unknown parameters
experimental and the numerical response is an error function that ðx1 ; x2 ; x3 . . . xp Þ; hM is the weight coefficient which determines the
can be expressed by the difference D ¼ Rexp comp
M;N  RM;N . The optimi- relative contribution of information yielded by the M-th set of
zation problem can then be formulated as follows: experimental data, and Ai,Bi are the lower and upper limits on the
104 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110
values of material parameters. The optimization problem in Eq. (6)
has the following characteristic features:
 The objective function is an implicit function of parameters x,
wher x 2 R;
 to calculate the values of this function for the specific set of
parameters x requires the use of a non-linear numerical (e.g.
Discrete Element) simulation of the process under consider-
ation, which is usually involves a considerable amount of com-
putational time;
 function values may present some level of numerical noise.
The computational simulations of masonry wall panels with
UDEC would require a large amount of computational time. Also,
convergence of the above method cannot be guaranteed due to
the presence of noise in the objective function values. Thus, routine
task analysis such as design optimization, design space exploration
and sensitivity analysis becomes impossible since a large amount
of simulation evaluations is required. One way to mitigate against
such a burden is by constructing surrogate models (also referred to
as response surface models or metamodels). These mimic the
behavior of the model as closely as possible while at the same time
they are time effective to evaluate [29]. Surrogate models are con-
structed based on modeling the response predicted from the com-
putational model to a limited number of intelligently chosen data
points. New combinations of parameter settings, not used in the
original design, can be plugged into the approximate model to
quickly estimate the response of that model without actually run-
ning it through the entire analysis. Different methods of regression
analysis (i.e. Least Squares Regression and Moving Least Squares)
can be used to construct the expression for the function F total
ðxÞ . This
approach can result in less computational iterations and lead to
substantial saving of computational resources and time. Using this
approach, the initial optimization problem, Eq. (6) is replaced with
the succession of simpler mathematical programming sub-prob-
lems as: finding the vector xk that minimizes the objective
function:
k k
F~k ðxÞ ¼ RhM F~M k k
k ðxÞ; Ai  X i  Bi ; Ai  Ai ; Bi  Bi ; ði ¼ 1. . .. .. NÞ
ð7Þ
where k is the iteration number. The limits Aki and Bki define a sub-
region of the optimization parameter space where the simplified
functions F~M
k ðxÞ are considered as current approximations of the ori-
ginal implicit functions FM(x). To estimate their accuracy, the error
parameter rk ¼ j½Fðxk Þ  F~k ðxk Þ=Fðxk Þj is evaluated. The value of the
error parameter gives a measure of discrepancy between the values
of the initial functions and the simplified ones. Any conventional
optimization technique can be used to solve a sub-problem in Eq.
(7) because the functions involved in its formulation are simple
and noiseless [30].
5. Identification of material parameters for low bond strength
masonry
5.1. Experimental stage
Seven single leaf brickwork masonry wall panels were tested in
the laboratory. These are part of a large test program carried out by
the research sponsor Bersche-Rolt Ltd. The wall panels were devel-
oped to represent the clay brickwork outer leaf of an external cav-
ity wall containing openings for windows. Panels were built with a
soldier course immediately above the opening with the remainder
of the brickwork being constructed in stretcher bond. Four of the
wall panels (S1–S4) had an opening of 2025 mm and six courses
of stretcher bonded brickwork above the opening. Two of the wall
panels (L1 and L2) had an opening of 2925 mm and 6 courses of
stretcher bonded brickwork above the opening. Also, one wall pa-
nel (DS1) had an opening of 2025 mm and nine courses of stretcher
bonded brickwork above the opening. Typical details of the panels
are shown in Fig. 5.
All the test panels were constructed with UK standard size
215 mm 102.5 mm 65 mm Ibstock Artbury Red Multi Stock
sand-faced bricks. The bricks had average water absorption of
14% and a compressive strength of 35 N/mm2. The joints were all
nominally 10 mm thick with 1:12 (OPC:sand) weigh-batched mor-
tar. The bricks and mortar were selected deliberately to produce
brickwork with low bond strength [31]. The aim of producing
brickwork with low bond strength in the laboratory was to repre-
sent low quality, high volume wall construction which according
to Bersche-Rolt Ltd. is fairly typical of low rise domestic construc-
tion in the UK. Each wall panel was tested by applying a central
point load to the top of the wall at mid-span.
The load was applied to each panel using a hydraulic ram and
was distributed through a thick steel spreader plate which was
embedded in mortar on the top of the brickwork. A structural steel
frame bolted to the laboratory floor provided the support. The load
was applied to each wall incrementally until the panel could no
longer carry the applied load. At each load increment, vertical
deflections were measured at mid-span using a dial gauge sup-
ported on a magnetic stand and a steel base plate. The surfaces
of the panels were inspected visually for signs of cracking at each
load increment. In addition, cracking in each panel under test
was identified from the dial gauges readings, i.e. sudden increases
in the deflection measurements during testing indicated crack for-
mation and other effects such as stress redistribution following
cracking and very short-term creep effects. Deflections at ultimate
load were not taken for safety reasons and to avoid damage to the
dial gauge. The experimental test results are summarized in
Table 1. Full descriptions of the experimental tests are described
at Sarhosis [32].
5.2. Computational model for masonry wall panels
Geometrical models representing the brickwork wall panels
tested in the laboratory were created in UDEC (Fig. 6). Each brick
of the panel was represented by a deformable block separated by
zero thickness interfaces at each mortar bed and perpend joint.
To represent for the 10 mm thick mortar joints in the real wall pan-
els, each deformable block was based on the nominal brick size in-
creased by 5 mm in each face direction to give a UDEC block size of
225 mm 102.5 mm 75 mm. The fact that the dimensions of the
brick have to be expanded slightly has no significant influence on
the accuracy of the model’s mechanical behavior predictions [32].
Bricks were modeled as deformable blocks. This option was also
chosen as it allows stresses in the zones to be calculated; these
zones can be seen in Fig. 5. It was assumed that each brick would
behave as a homogeneous, isotropic continuum which will exhibit
linear stress–strain behavior as they would be loaded well below
their strength limit. Material properties have been selected such
that the bricks would remain intact at all stages of applied loading
and the predominant failure mode would be slip along the brick–
mortar interface. Such failure modes have been observed in the full
scale experiments of the low bond strength masonry wall panels.
The block parameters required by UDEC to represent the behavior
of the bricks are: density (d), elastic modulus (E), shear modulus
(G) and Poisson’s ratio (m). Mortar joints were represented by inter-
faces modeled using UDEC’s elastic-perfectly plastic coulomb slip-
joint area contact option. According to ITASCA [15], this option is
intended for closely packed blocks of any shape with area contact,
such as masonry, and provides a linear representation of the mor-
tar joint stiffness and yield limit. It is based on the elastic normal
 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110 105

Fig. 5. Typical details of masonry wall panel tested in the laboratory.

Table 1 at each time step were recorded and used to create the load versus
Masonry wall panel experimental test results. displacement relationships. This was later compared in the optimi-
Panel Clear Courses of Type of Mortar Load at Failure zation process with the experimental results obtained in the
opening stretcher masonry compressive first load laboratory.
(mm) bond above strength 106 visible (kN) The selection of the appropriate velocity applied on the top of
opening (N/m2) crack the wall panel is specified by the user and thus its value is critical.
(kN)
From a sensitivity analysis on different values of velocity applied at
S1 2025 Six Unreinforced 0.72 0.72 3.69 the steel spreader plate located on the top mid-span part of the pa-
S2 2025 Six Unreinforced 0.79 1.60 4.60
nel it was found that the higher the applied velocity, the larger the
S3 2025 Six Unreinforced 0.86 1.60 5.10
S4 2025 Six Unreinforced 1.18 1.71 5.67 impact on the steel spreader plate. Conversely, the lower the value
of velocity, the lesser the impact on the steel spreader plate and the
L1 2925 Six Unreinforced 0.64 0.10 1.60
L2 2925 Six Unreinforced 0.71 0.60 2.60 larger the computational time required to complete the analysis.
Also, the selection of the appropriate value of velocity is related
DS1 2925 Nine Unreinforced 0.96 0.72 10.60
to the value of normal and shear stiffness as these will determine
Notations: (S) refers to short span panels, (L) refers to long span panels and (D) the stiffness of the panel and thus the impact caused from the
refers to deep panels.
velocity. Convergence tests were also carried out on the magnitude
of the velocity to be applied to the spreader plate to make sure that
a quasi-static loading condition was achieved. An indicative value
of velocity that used for the analysis for the wall panels was
(JKn) and shear (JKs) stiffnesses, frictional (Jfric), cohesive (Jcoh) 0.8 mm/s.
and tensile (Jten) strengths, as well as the dilation (Jdil) character-
istics of the mortar joints. 5.3. Identification of masonry material parameters for UDEC models
The bottom edges of the UDEC wall panels were modeled as
rigid supports in the vertical and horizontal direction whilst the 5.3.1. Optimization aims
vertical edges of the wall panel were left free. Self-weight effects The material parameters used in UDEC were optimized to pro-
have also been included in the model as a gravitational load. At duce similar responses to the following aspects of behavior
first, the model was brought into a state of equilibrium under its observed from the large-scale tests in the laboratory:
own weight. Then a vertical constant velocity applied to the load
spreader plate on the top of the panel until collapse. A FISH sub- (a) The applied load and deflection of the panel at the occur-
routine was written in UDEC that was able to record the reaction rence of first cracking (point 1 in Fig. 7).
forces from the fixed velocity grid points acting on the spreader (b) The maximum load supported by the wall/beam panels
plate at each time step. Also, histories of mid-span displacement (point 4 in Fig. 7).

Fig. 6. Typical UDEC geometric model of a masonry wall panel with a 2025 mm opening.
106 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110
Fig. 7. Response evaluation points for Level 1 and Level 2 optimization (R, L and D
denotes Response, Load and Displacement, respectively).
(c) The intermedian load–displacement relationships (points 2
and 3 in Fig. 7).
(d) The propagation of cracks in the wall/beam panels with
increasing applied load.
(e) The mode of failure.
5.3.2. Sensitivity analysis
Initially a sensitivity analysis was carried out to identify the ef-
fect of the brick and brick/mortar joint interface parameters used
in UDEC on the pre- and post-cracking behavior [32]. As expected,
behavior was found to be independent of the brick properties and
was dominated by the brick/mortar joint interface properties. From
the sensitivity analysis it was also found that:
(a) The load versus displacement relationship for the masonry
wall/beam panels was linear up to the occurrence of first
cracking;
(b) Of the six parameters used by UDEC to define the character-
istics of the brick/mortar joint interface:
(i) The normal stiffness(JKn); the shear stiffness (JKs) and
the tensile strength (Jten) of the interface, have domi-
nate influences on the behavior of panels up to and
including the occurrence of the first crack;
(ii) The cohesive strength (Jcoh); the angle of friction (Jfric)
and angle of dilation (Jdil) influence more on the behav-
ior of the panels after first cracking up to collapse.
To reflect results from the sensitivity study and to minimize the
computational time, the optimization of the material parameters
was carried out in two different Levels as indicated in Fig. 6.
 Level 1 Optimization: Optimization of the joint interface material
parameters JKn, JKs and Jten up to the occurrence of the first
crack; and
 Level 2 Optimization: Optimization of the joint interface material
parameters Jfric, Jcoh and Jdil just after the occurrence of first
crack and up to the ultimate load in the panel.
Table 2
Variation of brick and mortar properties as identified from the literature.
Interface parameter Young’s modulus of brick Young’s modulus of mortar
(N/m2) (N/m2)
Symbol Eb Em
Range (4–10) 109 (1–11) 108
The two levels of optimization are described in more details
below.
5.3.3. Level 1 optimization
For the Level 1 optimization, the material parameters to be opti-
mized up to the occurrence of first crack in the panel are: JKn, JKs
and Jten. After Lourenco [19], the normal and shear stiffness of
mortar joint can be estimated according to the brick and mortar
properties as follows:
Eb Em
JKn ¼ ð8Þ
hm ðEb  Em Þ
Gb Gm
JKs ¼ ð9Þ
hm ðGb  Gm Þ
E
G¼ ð10Þ
2ð1 þ v Þ
where Eb and Em are the Young’s moduli and Gb and Gm are the
shear moduli, respectively, for the blocks and mortar and hm is
the actual thickness of the mortar joint. These equations give the ra-
tion of normal to shear stiffness as:
 
Eb v m  Em v b
JKn=JKs ¼ 2 1 þ ð11Þ
Eb  Em
Table 2 demonstrates a range of material properties determined by
using the data from the literature [3,17,18,33] for brick and mortar
combinations similar to these used for the construction of the large
scale experiments. These properties have been obtained from the
testing of small samples of material or small assemblages.
Using the range of material parameters from Table 2, a variation
analysis of the minimum and maximum values of the normal and
shear stiffness has been carried out. The analysis showed that:
 Normal stiffness (JKn) ranged from 10 to 150 109 N/m3;
 Shear stiffness (JKs) ranged from 4.3 to 65 109 N/m3 and
 The ratio of normal to shear stiffness (JKn/JKs) ranged from 2.18
to 2.5. For the purpose of this study is appropriate to assume
that the ratio of normal to shear stiffness is equal to 2.3 [32].
Since normal stiffness is directly related to shear stiffness, it
was decided to optimize the normal joint stiffness and the joint
tensile strength parameters of the mortar. A factorial design of
28 experiments has been proposed for each of the short (S1–S4)
and long panels (L1 and L2) referred to in Table 1. The material
parameters used in the computational experiments for the Level
1 optimization are shown in Table 3. The joint normal stiffness
and joint tensile strength values were varied as shown, but the
other parameters were assumed to be constant and equal to the
values reported by Lourenço [19].
The least squares differences between the experimental and
computational test results, up to the occurrence of first cracking,
were then estimated for each of the short and long panels. All the
response quantities were considered to be equally weighted for
the formulation of the objective function. A surrogate model
was created with the use of the Moving Least Squares (MLS)
approximation method. The predicted response surface created
Poisson’s ratio of brick Poisson’s ratio of mortar Height of mortar joint
(m)
vb vm hm
0.1–0.2 0.1–0.2 0.01
 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110 107

Table 3 5.3.4. Level 2 optimization

Range of brick and mortar joint properties used in UDEC models. For the Level 2 optimization, a factorial design of 175 experi-
Brick properties Symbol Value Units ments was carried out for each of the short and long panels used
Elastic parameters Density d 2000 kg/m3 in the level 1 optimization. The values of the interface parameters
Elastic modulus E 6050 106 N/m2 used when planning the computational experiments are shown in
Poisson’s ratio m 0.14 – Table 4. The range of materials properties obtained from the liter-
Mortar joint ature [12,17]. The procedure adopted was similar as per Level 1
properties
Joint normal JKn (10, 25, 50,75, 100, N/m3
Optimization. Response surfaces were created as shown in Fig. 9
stiffness 150) 109 and an optimization study was carried out using a genetic algo-
Joint shear JKs JKn/2.3 N/m3 rithm in Hyperstudy 10. From the analysis it was found that:
stiffness Jfric = 40°; Jcoh = 0.062 106 N/m2 and Jdil = 40°.
Inelastic parameters Joint friction Jfric 36.8 Degrees
angle
Joint cohesion Jcoh 0.375 10 6
N/m2 5.3.5. Verification study
Joint tensile Jten (0.09, 0.10, 0.11, N/m2 The experimental results for the short and long panels were
strength 0.12) 106 compared with the results obtained from the UDEC modeling using
Joint dilation Jdil 0 Degrees
the optimized material parameters (Figs. 10 and 11). The
angle

Table 4
Range of brick and interface material properties.

using Altair HyperStudy 10 [23] is shown in Fig. 8. Then, an opti- Brick properties Symbol Value Units
mization study has been carried out using Altair HyperStudy 10. Elastic Density d 2000 kg/m3
This optimization study is based on the response surface repre- parameters Elastic modulus E 6050 106 N/m2
sented in Fig. 8. An evolutionary algorithm method i.e. the Genet- Poisson’s ratio v 0.14 –
Mortar joint
ic Algorithm (GA) in HyperStudy 10 has been adopted. According
properties
to Toropov and Yoshida [30]), GA is more likely to find a non-lo- Joint normal JKn 13.5 109 N/m3
cal solution (i.e. the global minimum) when compared to other stiffness
gradient based methods such as Sequential Quadratic Method or Joint shear stiffness JKs 5.87 109 N/m3
the Adaptive Response Surface method. Further details of the Inelastic Joint friction angle U 20–40 Degrees
optimization process are described elswhere [32]. From the Level parameters Joint cohesion Jcoh 0.04– N/m2
1 optimization exercise it was found that: JKn = 13.5 109 N/m3; 0.16 106
Joint tensile Jten 0.101 106 N/m2
JKs = 5.87 109 N/m3 and Jten = 0.101 106 N/m2. These values strength
and the brick properties shown in Table 3 were then used in Joint dilation angle W 0–40 Degrees
the Level 2 optimization.

Fig. 8. Response surface relating the objective function with the joint normal stiffness and the joint tensile strength.
108 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110
Fig. 9. Response surface relating the objective function with the joint cohesive strength and joint friction angle when joint dilation angle is 35°.
Fig. 10. Comparison of experimental and computational results for the short
panels.
experimental curves at Fig. 10 and Fig. 11 stop at a point before the
ultimate load is reached. Records of mid-span displacements at
ultimate load were not taken for safety reasons and to avoid dam-
aging the dial gauge. Bearing in mind the inherent variations that
occur in masonry, a good level of correlation was achieved.
5.3.6. Validation study
To check the validity of the material parameter identification
procedure, the UDEC model with the optimized parameters was
used to predict the behavior of a deeper wall/beam panel (DS1).
Fig. 11. Comparison of experimental and computational results for the long panels.
Fig. 12 shows the distribution of cracks at collapse predicted with
UDEC. A similar distribution of cracks was observed in the labora-
tory testing as shown in Figs. 13–15. Also, the development of
cracks under increasing applied load predicted using UDEC was
found to be very similar to the observed propagation of cracking
in the laboratory tests. Both experimental and computational re-
sults contain four notable aspects of behavior namely: (a) initial
flexural cracking in the soffit of the panel; followed by (b) the
development of flexural cracks in the bed joint of each support;
with increasing load leading to (c) propagation of diagonal stepped
cracks at mid depth both up (towards the loading point) and down
 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110
Fig. 12. Failure mode for panel DS1 predicted using UDEC.
Fig. 13. Shear development propagated from the corners of the opening to the top
of the panel.
Fig. 14. Flexural crack at the right hand support.
(towards the corner of the opening); and (d) collapse as a result of
what is usually referred as a shear failure (or excessive diagonal
tension).
Fig. 16 shows the observed and the UDEC predicted load versus
displacement behavior of the panel DS1. The UDEC predicted value
of the load at first cracking (2 kN) is close to that observed from the
laboratory testing (1.72 kN). The ultimate load recorded in the lab-
oratory (10.6 kN) compares well with the load of 10.4 kN predicted
109
Fig. 15. View of the right hand side of the panel. Development of shear cracks.
using UDEC. However, the stiffness of the panel as observed from
the experiment is dissimilar to that predicted from UDEC, as the
load applied to the panel is more than 6 kN. A likely reason for this
is the effect of short-term creep (or load re-distribution) that oc-
curred in the panel. During the experiment, readings of the mid-
span displacement were taken when the dial gauge value had sta-
bilized. As the panel neared a state of impending collapse, the dial
gauge reading varied a great deal under constant applied load as
cracks developed and propagated throughout the panel. This could
explain the increasing discrepancies between the UDEC output and
the experimental results at the higher levels of applied load.
Fig. 16. Observed and UDEC-predicted load versus displacement relationships for
panel DS1.
110 V. Sarhosis, Y. Sheng / Engineering Structures 60 (2014) 100–110
6. Conclusions
A study of the identification of material parameters for model-
ing masonry structures with UDEC has been carried out and re-
ported in this paper. Conventionally, the material parameters
used for modeling masonry in computational models are based
on the results of small scale tests that do not reflect the more com-
plex boundary conditions and stress-state types that exist in a real
masonry structure. A method which is considered likely to provide
more representative material parameters for masonry constitutive
models has been proposed here. The method involves the compu-
tational analysis of large scale experimental tests on masonry
structures, and an optimization process to tune the masonry
material parameters by minimizing the difference between the
responses measured from the large scale lab tests and those
obtained from the computational simulation.
The results from a series of low bond strength masonry wall
panels with opening tested in the laboratory have been used in this
study. Such panels have also been modeled with UDEC software.
Material parameters for UDEC models were ‘‘tuned’’ using an opti-
mization process in order to achieve similar responses to those ob-
tained in the laboratory. The tuning was based on the need to
achieve good correlation between the pre-cracking, post-cracking
and near-collapse behavior of the masonry wall panels. Surrogate
models that relate the output responses of the UDEC model to
the input material parameters have been constructed to minimize
the computational costs. Optimization of the material parameters
was then performed using the surrogate model and a single set
of optimized material parameters were obtained by this process.
The use of surrogate modeling for creating approximations has
proved to be a useful approach as it resulted in less computational
iterations, and led to substantial saving of computational resources
and time. To validate the optimization results, the material param-
eters obtained from the optimization process were then used in a
UDEC model to predict the structural response of different wall
panels. In spite of the inherent variability of masonry, good corre-
lation was achieved between the predicted behavior from UDEC
model and that observed in the laboratory. The developed method
provides a more rigorous tool for the identification of the material
parameters for modeling masonry structures, and has reduced the
uncertainties associated with the traditional methods. As for the
future study, the accuracy of the optimization method can be im-
proved by using larger number of test data and the developed com-
putational model for masonry can be used to study the
strengthening of masonry structures. The material parameter iden-
tification approach presented in this paper is applicable to any
non-linear numerical formulations (finite element, Discrete Ele-
ment) and almost unlimited variety of experiments can be used
to account for inhomogeneous material response.
Acknowledgements
The work presented in this paper is supported by an EPSRC
Award in Science and Engineering (CASE/CAN/07/22) and Ber-
sche-Rolt Ltd. Particular thanks are due to Prof. Garrity. Also to
George Edscer and Chris Smith of Bersche-Rolt Limited for their
continued support and encouragement. Thanks are also due to
Prof. Toropov and Dr. Eves for their assistance with the optimiza-
tion elements of this research.
References
[1] Roca P, Cervera G, Gariup G, Pela L. Structural analysis of masonry historical
constructions. Classical and advanced approaches. Arch Comput Methods Eng
2010;17(3):299–325.
[2] Pluijm R. Shear behavior of bed joints. In: Hamid AA, Harris HG, editors.
Proceedings of the 6th North American masonry conference. Philadelphia,
Pennsylvania, USA: Drexel University; 1993. p. 125–36.
[3] Hendry AW. Structural masonry. 2nd Edition. London, UK: Palgrave
Macmillan; 1998.
[4] Abdou L, Ami SR, Meftah F, Mebarki A. Experimental investigations of the joint-
mortar behavior. Mech Res Commun 2006;33(3):370–84.
[5] Adami CE, Vintzileou E. Investigations of the bond mechanism between stones
or bricks and grouts. Mater Struct 2008;41(2):255–67.
[6] Garrity SW, Ashour AF, Chen Y. An experimental investigation of retro-
reinforced clay brick arches. In: Proceedings of the 8th International Masonry
Conference. Dresden; July 2010, Germany, pp. 733–742. [CD-ROM
proceedings].
[7] Cundall PA. A computer model for simulating progressive large scale movements
in blocky rock systems. In: Proceedings of the Symposium of the International
Society of Rock Mechanics, Nancy, France, vol. 1, paper No II-8; 1971.
[8] Schlegel RR. Failure analyses of masonry shear walls. In: Konietzky H, editor.
Numerical modelling of discrete materials in geotechnical engineering, civil
engineering and earth sciences. UK: Taylor and Francis Group London; 2004. p.
15–20.
[9] Lemos JV. Discrete element modelling of masonry structures. Int J Archit
Heritage 2007;1:190–213.
[10] Toth AR, Orban Z, Bagi K. Discrete element analysis of a masonry arch. Mech
Res Commun 2009;36(4):469–80.
[11] Zhuge Y. Micro-modelling of masonry shear panels with distinct element
approach. In: Chowdhury L, Fragomeni, editors. Advances in mechanics of
structures and materials. Lisse: Swets & Zeitinger; 2002. p. 131–6.
[12] Dialer C. A distinct element approach for the deformation behavior of shear
stressed masonry panels. In: Proceedings of the 6th Canadian masonry
symposium. Saskatoon, Canada: University of Saskatchewan; 1992. p. 765–76.
[13] Walker P, McCombie P, Claxton M. Plane strain numerical model for drystone
retaining walls. Proc ICE – Geotec Eng 2006;160(2):97–103.
[14] Souley M. 1993. Modelling of jointed rock masses by distinct element method,
influence of the discontinuities constitutive laws upon the stability excavation.
PhD thesis, Institute of National Polytechnique de Lorraine, France.
[15] ITASCA 2004. UDEC – Universal distinct element code manual. Theory and
background, Itasca consulting group, Minneapolis, USA.
[16] Otter JRH, Cassell AC, Hobbs RE. Dynamic relaxation. Inst Civil Eng
1996;35(4):633–56.
[17] Rots JG. Structural masonry: an experimental/numerical basis for practical
design rules. Netherlands: Balkema Publishers; 1997.
[18] Pluijm R. Out-of-plane bending of masonry behavior and strength, Ph.D thesis,
Eindhoven University of Technology, The Netherlands; 1999.
[19] Lourenço PB. Computational strategies for masonry structures, Ph.D thesis,
Delft University of Technology, Delft, Netherlands; 1996.
[20] Vermeltfoort AT. Effects of the width and boundary conditions on the
mechanical properties of masonry prisms under compression. In:
Proceedings of the 11th International Brick/Block Masonry Conference.
Shanghai, 27–29 October 1997, pp. 181–190.
[21] Lignola G, Prota A, Manfredi G. Nonlinear analyses of tuff masonry walls
strengthened with cementitious matrix-grid composites. J Compos Constr
2009;13(4):243–51.
[22] Toropov VV, Garrity SW. Material parameter identification for masonry
constitutive models. In: Proceedings of the 8th Canadian masonry
symposium. Alberta, Canada: Jasper; 1998. p. 551–62.
[23] Altair Engineering, 2010. Altair Hyperstudy 10, User’s Manual, Design of
experiments, Optimization and Stochastic studies. Alter Engineering Inc.
[24] Toropov VV, Giessen E. Parameter identification for nonlinear constitutive
models: finite element simulation – optimization – nontrivial experiments. In:
Pedersen P, editor. Proceedings of IUTAM symposium, optimal design with
advanced materials. Lyngby, Denmark: The Frithiof Niordson volume; 1993. p.
113–30.
[25] Yoshida F, Urabe M, Toropov VV. Identification of material parameters in
constitutive model for sheet metals from cyclic bending tests. Int J Mech Sci
1998;40(2):237–49.
[26] Morbiducci R. Non-linear parameter identification of models for masonry. Int J
Solids Struct 2003;40(15):4071–90.
[27] Gambarotta L, Lagomarsino S. Damage models for the seismic response of
brick masonry shear walls. Part I: the mortar joint model and its applications. J
Earthquake Eng Struct Dynam 1997;26(4):423–39.
[28] Gambarotta L, Lagomarsino S. Damage models for seismic response of brick
masonry shear walls. Part II: the continuum model and its applications. J
Earthquake Eng Struct Dynam 1997;26(4):441–62.
[29] Queipo NV, Haftka RT, Shyy W, Goel T, Vaidyanathan R, Tucker PK. Surrogate-
based analysis and optimization. Prog Aerosp Sci 2005;41(1):1–28.
[30] Toropov VV, Yoshida F. Application of advanced optimization techniques to
parameter and damage identification problems. In: Mroz Z, Stavroulakis GE
editors. Parameter Identification of Materials and Structures CISM Courses and
Lectures, vol. 469, International Centre for Mechanical Sciences; 2005. pp.
177–263.
[31] Vermeltfoort AT, Martens DEW, Zijl V. Brick mortar interface effects on
masonry under compression. Can J Civil Eng 2007;34(1):1475–85.
[32] Sarhosis V. Computational modelling of low bond strength masonry. PhD
thesis, University of Leeds, UK; 2012.
[33] Sarangapani G, Venkatarama RBV, Jagadish KS. Brick-mortar bond and
masonry compressive strength. J Mater Civil Eng 2005;17(2):229–37.