DISCRETE ELEMENT MODELING OF A TWO STOREY
UNREINFORCED MASONRY SCALED MODEL
Francisco GALVEZ 1, Marta GIARETTON 2, Shannon ABELING3,
Jason M. INGHAM4, Dmytro DIZHUR5
ABSTRACT
Unreinforced masonry (URM) buildings are known to perform poorly when subjected to earthquake induced
ground shaking. In order to enhance understanding of URM building global behaviour, a two-storey URM scaled
model building was constructed and tested on a uniaxial shake-table. The building model was designed to replicate
the global behaviour of a prototypical URM building commonly encountered in New Zealand. During the shake-
table testing, in-plane shear and two-way out-of-plane bending mechanisms were observed similar to those
documented in post-earthquake observations.
The scaled building used in the experimental shake-table test was numerically modelled using discrete element
approach and subjected to the same ground motion as that used in the experimental campaign. The arrangement
of accelerometers used during the test was maintained to model the data acquisition in the numerical simulation.
Significant correlation was found between the damage exhibited by the scaled model building test and the
numerical simulation. The numerical model helped to understand the details of the collapse mechanism formation
and was calibrated in order to be capable of reproducing the dynamic behaviour of more complex buildings and
testing the effectiveness of retrofit solutions.
Keywords: Macro-elements, Numerical Modelling, Unreinforced Masonry, Failure Mechanisms, Shake-table
1. INTRODUCTION
The poor performance of unreinforced masonry (URM) buildings in past earthquakes throughout the
world in locations such as New Zealand (Dizhur et al., 2011; Dizhur et al., 2010; Ingham & Griffith,
2011a, 2011b; Moon et al., 2014), Chile (Astroza et al., 2012), Italy (Indirli et al., 2013), California
(Galloway & Ingham, 2015), and Nepal (Dizhur et al., 2016) is well documented. Based on the existing
New Zealand URM building stock (Russell & Ingham, 2010), a scaled model of a commonly
encountered URM building was built and tested in the laboratory by means of shake-table testing. The
aim of the experimental campaign was to enhance understanding of the sequence of damage that leads
to the formation of different collapse mechanisms and to investigate the effectiveness of various retrofit
systems. To further investigate the URM behaviour, the shake-table laboratory tests were complemented
with numerical simulation using the Discrete Element Method (DEM). DEM was used before by several
authors to perform URM simulations and in some cases to compare the results with shake-table testing
(Alexandris et al., 2004; Çaktı et al., 2016; Cannizzaro & Lourenço, 2016; DeJong & Vibert, 2012; Lou
et al., 2014). While DEM have experienced large improvement during the last years, validation against
experimentally tested models is still needed. Only modelling blocks and joint interfaces that simulates
the mortar contact, the heterogeneous nature of URM can be well described allowing complex
geometries to be modelled. Furthermore, the DEM is ideal for problems that need to solve large
displacements and collapse of buildings, since the position and interactions between blocks are updated
1
PhD student, University of Auckland, Auckland, New Zealand, [email protected]
2
Research Fellow, University of Auckland, Auckland, New Zealand, [email protected]
3
PhD student, University of Auckland, Auckland, New Zealand, [email protected]
4
Professor, University of Auckland, Auckland, New Zealand, [email protected]
5
Lecturer, University of Auckland, Auckland, New Zealand, [email protected]
in every timestep (Itasca, 2013). In addition, stresses and crack opening forming collapse mechanisms
can be monitored during the simulation.
2. SHAKE-TABLE EXPERIMENTAL SETUP
A 1/4 scaled model URM building was constructed using recycled solid-clay bricks from the 1950-60s
with approximate dimensions of 110 × 60 × 45 mm (length × thickness × height). The full-sized bricks
were hand cut to scale and assembled in a running-bond with approximately 5 mm thick lime-based
mortar layers. The density of the masonry model was 1800 kg/m3. Perimetral timber elements were
placed at the first floor and roof levels and were anchored to the bricks using bolts (see Figure 1a).
Timber beams were also used as lintels above door and window openings. Timber strips were positioned
at the base and at each side of the perimetral wall to prevent sliding. Several accelerometers were placed
around the building, but only five were considered and analysed for the study reported herein (see Figure
1a).
(b) Interior view (first level)
(c) Construction process
(a) ¼ scaled model dimensions and accelerometer locations (d) Geometry of the numerical model
Figure 1: Experimental test and numerical simulation configuration
Experimental testing of the model building was undertaken after allowing 28 days for the mortar to cure.
The experiment was divided in two tests, during which harmonic motion was applied to the shake-table
and the frequency progressively increased. During test one (T1), the model was subjected to a sequence
of harmonic excitation that reads as follows (cycles x period): 11x2.99s; 10x2.37s; 7x1.96s; 8x1.80s;
9x1.70s; 13x1.57s; 9x1.46s; 9x1.38s; 21x1.30s; 12x1.23s; and 13x1.17s. When the model performed
2
near-collapse instability, the motion was stopped, and cracks were registered (Figure 2a-d). Test two
(T2) consisted of applying a similar sequence of harmonic excitation to the cracked model. T2 continued
until the model collapsed at a period of 1.17s.
(b) Right wall
(a) Front wall
(c) Left wall (d) Rear wall
(e) Accelerogram recorded during test T1 showing the points where the visible cracking occurred.
Figure 2: Crack-pattern occurred during test T1.
Figure 2e shows the times when cracks began to form during T1, with the first visible cracking appearing
at second 201. The first cracking consisted of in-plane shear diagonal cracks (1, 2, 3, 4) which appeared
3
around the openings and particularly gathered around the spandrel of the front and rear building
elevations. After the first shear cracks, horizontal cracks occurred around the floors in the left wall (2,
5). These horizontal cracks allowed for certain movements that later formed an out-of-plane mechanism
(7). The right wall exhibited low damage in the top corners (6, 7) and the beginning of an out-of-plane
crack pattern (7).
3 EXPERIMENTAL DATA PROCESSING
One input parameter for the numerical simulation was the elastic natural frequency of the scaled building
model. Signal recordings at an early stage of test T1 showed the un-cracked behaviour of the model
(Figure 3a), while the signal recorded after 154 seconds (Figure 3b) showed the behaviour of the model
with micro-cracking damage (note that first visible cracking appeared at second 201). The recorded
acceleration time-history captured the oscillation to which the model was subjected as well as the free-
vibration of the building at the peak of the oscillations when velocities are close to zero. The process to
obtain the frequencies of the first mode of vibration required to filter the input frequencies of the
harmonic motion as observed in Figure 3. The Fourier transform was applied to the resultant signal, and
the natural frequencies were observed via the Fourier amplitude. All signals in the study reported herein
were filtered using the Butterworth filter of order n=6. These two sets of accelerations were filtered with
a high-pass configuration and a cut-off frequency of 2Hz.
(a) Experimental acceleration (10-80s)
(b) Experimental acceleration (154-176s)
Figure 3: Top acceleration at different stages of test T1 before and after frequencies filtering
Figure 4 shows the Fourier transforms for each accelerometer considered in the study and for both the un-cracked
(10-80s) and micro-cracked (154-176s) signal recordings. The graphs show the peaks corresponding to the
fundamental frequencies depicted with a circle in the un-cracked signal and with a star in the cracked signal. The
difference in frequencies peaks was attributed to the fact that some accelerometers were able to better capture some
modes of vibration than other. A set of frequencies are excited around each peak giving a range of uncertainty of
±1 Hz. The first Fourier amplitude is found around 4 Hz, however this frequency may be part of the disturbances
from the shake-table input (see discussion in section 4). If lower frequencies are ignored, the first frequency was
4
found at 15 Hz (see red markers in Figure 4), the second frequency (blue markers) was encountered at 23-24 Hz
and finally, a group of frequencies corresponding to the third mode of vibration was observed around 33 Hz (see
green markers). Generally, after 154 seconds of shaking and hence during the micro-cracking zone, the first
fundamental frequency (red markers) dropped between 2 Hz and 3 Hz when compared to the un-cracked model,
thereby showing a reduction in stiffness due to the formation of micro-cracking. Higher frequencies were more
difficult to allocate after cracking initiated and this is the reason why they are not marked in Figure 4.
(a) Bottom: f1=15-12.5Hz, f2=23.5Hz (b) 1/4: f1=15.5-12.5Hz, f2=23.5Hz
(c) Middle: f2=23.5-18Hz, f3=32-27.5Hz (d) 3/4: f1=15-12Hz, f2=/-19Hz, f3=33-30Hz
Circle and star markers refer to fundamental
frequencies in the un-cracked and cracked
signal, respectively. Different colours refer
to different modes of vibration, see text for
details.
(e) Top: f1=15-12.5Hz, f2=23.5-20Hz, f3=35-32.5Hz
Figure 4: Fourier Amplitude Spectra at different times of the tests and at different accelerometer locations
5
4. MODELLING APPROACH
Numerical simulations of the reported experimental campaign were performed using the DEM by
applying the code 3DEC code formulated by Itasca (2013). The simplified micro-modelling approach
was chosen as preferred between the several strategies available to model masonry structures (Lourenço,
2002). Crushing of bricks was not observed during the experimental campaign. Therefore, assuming
rigid bricks and a linear compression constitutive relation was deemed a reasonable approximation. In
addition, the use of rigid bricks to model masonry structures for simplicity and efficiency is widely
found in the literature (Bui & Limam, 2013; Çaktı et al., 2016; Oliveira et al., 2002). The lintel above
the opening was also modelled using rigid polyhedral elements. Timber floors and boundary beams were
modelled using rigid elements connected to each other and to the masonry blocks through flexible joints
to allow deformation during the oscillations. Normal displacements of the joints were plotted to visualize
the crack opening and thus study the collapse mechanisms. The Mohr-Coulomb failure criterion was
selected to simulate the elastic properties of the mortar and the brick and the non-linear behaviour of the
mortar. The three parameters needed to define the material properties of the bricks were density (dens);
bulk modulus (K); and shear modulus (G). The five normal and shear parameters needed for the joints
were: normal stiffness (jkn); shear stiffness (jks); friction angle ( ̊ ); cohesion (c); and tensile strength (ft).
The following formula was used in order to obtain the normal joint stiffness using rigid blocks:
𝐸𝐸𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤
𝑗𝑗𝑘𝑘𝑘𝑘 = 𝑠𝑠
(1)
where s is the joint spacing and Ewall is the Young Modulus of the masonry assemblage. Using elastic
relationships from the jkn, shear stiffness can be calculated as follows:
𝑗𝑗𝑘𝑘𝑘𝑘
𝑗𝑗𝑘𝑘𝑘𝑘 = (2)
2(1+𝑣𝑣)
where ν is the Poisson’s ratio. Weak materials were used to build the masonry walls in the experimental
building model and hence it was selected a friction angle of 17 ̊, a mortar cohesion of 0.1 MPa and a
Poisson’s ratio of 0.25. Previous researches (Petry & Beyer, 2014; Vaculik, 2012) have demonstrated
that scaling down masonry structures, even using the same material, induces an increase in the elastic
properties. Therefore the estimated values adopted in the simulation were:: jkn=29,000 MPa/m,
jks=11,000 MPa/m, friction angle=17 ̊, c=0.1 MPa, and ft=0.09 MPa for the horizontal joints, while for
the vertical joints: jkn=12,000 MPa/m, jks=5,000 MPa/m, friction angle=17 ̊, c=0.1 MPa and
ft=0.09 MPa.
Eigenvalue analysis was performed prior to the harmonic motion simulation. Figure 5 shows the modes
of vibration and frequencies obtained in the analysis. The first frequency of approximately 4Hz observed
in the experimental model (see Figure 4) was not registered in the eigenvalue analysis. However, the
three following frequencies, 15 Hz, 24 Hz and 33 Hz were well captured in the analysis. Nonetheless,
as will be further examined in the reported study, frequencies of approximately 4Hz were also observed
in the harmonic simulation. This frequency issue could be attributed to the input harmonic signal that
used high periods.
Mode 1: 15Hz Mode 2: 26 Hz Mode 3: 35 Hz Mode 4: 36 Hz
Figure 5: Modal shapes contoured with displacement magnitude and frequencies of first four modes of vibration
6
Rayleigh damping was only applied proportional to the mass as the stiffness-proportional type of
damping would otherwise further decrease the time-step increasing the runtime of the simulation to a
non-practical point. The same approach was used by Çaktı et al. (2016) for the dynamic simulation of a
scaled mosque shake-table model. Similarly, (Psycharis et al., 2013) simulated marble columns using
no damping and validated such choice with a reasonably low difference from the use of only mass
proportional damping. The inputs for the damping parameters were the natural frequency and the
damping ratio. A frequency of 15 Hz was used in accordance with the results obtained from the
experimental campaign (Figure 4) and the eigenvalue analysis (Figure 5). It is to be noted that the
eigenvalue analysis is independent of these two input values. A damping ratio of 5% was estimated
based on literature (Çaktı et al., 2016) and on a parametric study performed using damping ratios of 3%
and 5% for different frequencies. Simulations with a 3% damping ratio tended to collapse at a premature
simulation stage due to a numerical instability that caused an abrupt non-realistic response of the model
as shown in Figure 6. Similar non-realistic response occurred when lower input natural frequencies were
used as damping parameter.
Figure 6: Non-realistic explosive collapse of the numerical model when using low values of damping and natural
frequency.
The input signal used for the numerical simulation follows a sequence in cycles and period that reads:
1x1.57s; 1x1.39s; 1x1.31s; 2x1.23s; 3x1.15s; 3x1.07s; 4x0.99s. Each section of the sequence will input
accelerations of 0.08g, 0.1g, 0.11g, 0.13g, 0.15g, 0.17g and 0.2g respectively.
5. RESULTS
In the numerical simulations (similar to the observations made in the experimental shake-table testing),
first shear cracks appeared in the front and rear walls, followed by horizontal cracks on the side walls,
as shown in Steps 1, 2 and 3 (see Figure 7a-c). Further out-of-plane cracks developed in the side wall
and became a clear two-way bending failure during Step 5 which showed cracking at the top and ground
floor of the left wall similarly to what observed during the laboratory testing. Crack developing initiated
at the beginning of the 1.07s period, just after the 1.15s period, which was reasonably close to what
obtained during the experiment where the triggering period for visual crack propagation was 1.17s. The
final collapse stage of the simulation well replicated the one occurred during laboratory testing as shown
in Figure 7 f, g).
To allow for the total collapse of the numerical model to be simulated within a reasonable computational
time, the period of the input motion was increased in the simulation to higher level than the motion
actually encountered in the experiment. Eighteen seconds of test was simulated in 17 hours, with the
simulation time increasing when cracking initiated. Frequencies for the numerical simulation were
analysed in the same manner as for the test, showing three main peaks. A set of frequencies peaked
around 3Hz. As previously mentioned, the lower frequencies were attributed to noise coming from the
7
shake-table input even after the high frequency pass filter was adopted with cut-off frequency of 2Hz.
Therefore, these low frequencies were ignored taking as the first peak the frequency at 15Hz that shifted
to 13Hz after the model experiences some damage. The second and third modes of vibration were found
at 20Hz and at 35Hz, respectively. Finally, the velocities were plotted against and the last seconds of
test T1 were compared to the simulation in Figure 9. The steps pictured in Figure 7 were also identified
in Figure 9.
(a) Step 1 – Period 1.07s
(b) Step 2 – Period 0.99s
(c) Step 3 – Period 0.99s
8
(d) Step 4 – Period 0.99s
(e) Step 5 – Period 0.99s
(f) Step 6 – End of the simulation (g) End of the experimental test
Figure 7: Steps of the crack opening sequence in the simulation (crack width in mm units)
A common feature of the DEM is that the crack-pattern can change every time the simulation is ran. The
phenomena is sometimes attributed to the high sensitivity of trivial perturbations of the geometry or the
input motion characteristics (Mouzakis et al., 2002). For the reported simulations, the same test was ran
several times to check the repeatability of the simulation but no differences were found in the crack
pattern nor the velocities registered in the blocks.
9
Figure 8: Fourier Amplitude Spectra at different times of the numerical simulation
Figure 9: Comparison of velocities histories. Input signal including the steps where the cracking occurred
related to Figure 7.
6. CONCLUDING REMARKS
The DEM numerical simulation model of a 1/4 scaled model URM building captured the dynamic
behaviour and damage progression observed during the experimental campaign. The crack-pattern
predicted by the numerical model were in good agreement with the actual crack pattern exhibited by the
experimental model, including the final collapse stage. In-plane shear cracks initiated at an earlier stage
than the out-of-plane cracks in both the simulation and the experiment. Shear cracking allowed
additional movement of the structure, which trigger horizontal cracking around the first floor, and finally
lead to two-way bending out-of-plane failure.
At the end of the test T1, the top floor of the left wall showed an out-of-plane failure that was between
the U-shaped supports configuration and all sides supported configuration. The simulations showed that
the main out-of-plane cracking was anticipated by a horizontal crack at the top floor, suggesting that the
U-shaped cracking often observed in URM buildings after an earthquake are induced by the formation
of horizontal cracking at the top of the wall.
The eigenvalue analysis helped to understand the dynamic behaviour of the experimental model and
input the correct frequencies for the harmonic simulation.
Runtime constraints made modelling the whole extent of the experimental testing impractical.
Therefore, differences were found between the experiment and simulation. For instance, velocities at
the top of the model were found to be higher in the experiment than in the simulation. In future studies,
longer simulations will be performed in order to better approximate assumptions for simulating
10
experimental conditions.
Despite the runtime constraints, the model offered great incite to the experiment and demonstrated its
potential for use in future applications. Similar numerical models can be utilised to study volume loss
in masonry buildings as well as debris fall area, both of which can inform casualty estimation techniques
and can help decision makers to make informed cordoning decisions after an earthquake.
The DEM numerical simulation model have shown the capacity to simulate a response of an URM
building when subject to lateral loads. Further developing of the study presented herein will be to
reproduce more complex buildings and also strengthening solutions.
7. ACKNOWLEDGMENTS
This research was partially supported by NZSEE and EQC who provided partial funding for the
experimental campaign, and by QuakeCoRE, a New Zealand Tertiary Education Commission-funded
Centre.This is QuakeCoRE publication number 0278. The authors also gratefully acknowledge the
software 3DEC provided by Itasca consulting group under the Itasca Educational Partnership program
and BBR Contech for providing testing facilities. Caitlin Cairncross and Michael Kennerley are also
acknowledge for their valuable efforts during the experimental testing.
8. REFERENCES
Alexandris, A., Protopapa, E., & Psycharis, I. N. (2004, August 1-6, 2004). Collapse Mechanisms of Masonry
Buildings Derived by the Distinct Element Method. Paper presented at the 13th World Conference on Earthquake
Engineering, Vancouver, B.C., Canada
Astroza, M., Ruiz, S., & Astroza, R. (2012). Damage Assessment and Seismic Intensity Analysis of the 2010 (Mw
8.8) Maule Earthquake. Earthquake Spectra, 28(S1), S145-S164. 10.1193/1.4000027
Bui, T. T., & Limam, A. (2013, Sep 2012). Masonry Walls under Membrane or Bending Loading Cases:
Experiments and Discrete Element Analysis. Paper presented at the The Eleventh International Conference on
Computational Structures Technology, Dubrovnik, Croatia
Çaktı, E., Saygılı, Ö., Lemos, J. V., & Oliveira, C. S. (2016). Discrete element modeling of a scaled masonry
structure and its validation. Engineering Structures, 126, 224-236. 10.1016/j.engstruct.2016.07.044
Cannizzaro, F., & Lourenço, P. B. (2016). Simulation of Shake Table Tests on Out-of-Plane Masonry Buildings.
Part (VI): Discrete Element Approach. International Journal of Architectural Heritage
10.1080/15583058.2016.1238973
DeJong, M. J., & Vibert, C. (2012). Seismic response of stone masonry spires: Computational and experimental
modeling. Engineering Structures, 40, 566-574. 10.1016/j.engstruct.2012.03.001
Dizhur, D., Dhakal, R., Bothara, J., & Ingham, J. (2016). Building Typologies and Failure Modes Observed in the
2015 Gorkha (Nepal) Earthquake. Bulletin of the New Zealand Society for Earthquake Engineering, 49(2), 211-
232.
Dizhur, D., Ingham, J., Moon, L., Griffith, M., Schultz, A., Senaldi, I., Magenes, G., Dickie, J., Lissel, S., Centeno,
J., Ventura, C., Leite, J., & Lourenco, P. (2011). Performance of masonry buildings and churches in the 22
February 2011 Christchurch earthquake. Bulletin of the New Zealand Society for Earthquake Engineering, 44(4),
279-296.
Dizhur, D., Ismail, N., Knoz, C., Lumantarna, R., & Ingham, J. M. (2010). Performance of unreinforced and
retrofitted masonry buildings during the 2010 Darfield earthquake. Bulletin of the New Zealand Society for
Earthquake Engineering, 43(4), 321-339.
Galloway, B., & Ingham, J. M. (2015). The 2014 South Napa Earthquake and its Relevance for New Zealand
SESOC Journal, 28(1), 69-94.
Indirli, M., Kouris, L. A. S., Formisano, A., Borg, R. P., & Mazzolani, F. M. (2013). Seismic Damage Assessment
11
of Unreinforced Masonry Structures After The Abruzzo 2009 Earthquake: The Case Study of the Historical
Centers of L'Aquila and Castelvecchio Subequo. International Journal of Architectural Heritage, 7(5), 536-578.
10.1080/15583058.2011.654050
Ingham, J. M., & Griffith, M. C. (2011a). The Performance of Earthquake Strengthened URM Buildings in the
Christchurch CBD in the 22 February 2011 Earthquake. Royal Commission of Inquiry into Building Failure
Caused by the Canterbury Earthquake.
Ingham, J. M., & Griffith, M. C. (2011b). The Performance of Unreinforced Masonry Buildings in the 2010/2011
Canterbury Earthquake Swarm Royal Commission of Inquiry into Building Failure Caused by the Canterbury
Earthquake.
Itasca. (2013). 3DEC (3-Dimensional distinct element code) (Version 5.2). Minneapolis (MN)
Lou, Y. F., Sun, B. T., & Zhang, X. (2014). The Masonry Crack Propagation Simulation Based on LS-DYNA.
Advanced Materials Research, 1021, 105-110. 10.4028/www.scientific.net/AMR.1021.105
Lourenço, P. B. (2002). Computations on historic masonry structures. Progress in Structural Engineering and
Materials, 4(3), 301-319. 10.1002/pse.120
Moon, L., Dizhur, D., Senaldi, I., Derakhshan, H., Griffith, M. C., Magenes, G., & Ingham, J. (2014). The Demise
of the URM Building Stock in Christchurch during the 2010-2011 Canterbury Earthquake Sequence. Earthquake
Spectra, 30(1), 253-2769. 10.1193/ 022113EQS044M
Mouzakis, H. P., Psycharis, I. N., Papastamatiou, D. Y., Carydis, P. G., Papantonopoulos, C., & Zambas, C. (2002).
Experimental investigation of the earthquake response of a model of a marble classical column. Earthquake
Engineering & Structural Dynamics, 31(9), 1681-1698. 10.1002/eqe.184
Oliveira, C. S., Lemos, J. V., & Sincraian, G. E. (2002). Modelling Large Displacements of Structures Damaged
by Earthquake Motions. European Earthquake Engineering, 56-71.
Petry, S., & Beyer, K. (2014). Scaling unreinforced masonry for reduced-scale seismic testing. Bulletin of
Earthquake Engineering, 12(6), 2557-2581. 10.1007/s10518-014-9605-1
Psycharis, I. N., Fragiadakis, M., & Stefanou, I. (2013). Seismic reliability assessment of classical columns
subjected to near-fault ground motions. Earthquake Engineering & Structural Dynamics, 42(14), 2061-2079.
Russell, A. P., & Ingham, J. M. (2010). Prevalence of New Zealand’s Unreinforced Masonry Buildings. Bulletin
of the New Zealand Society for Earthquake Engineering, 43(3)
Vaculik, J. (2012). Unreinforced Masonry Walls Subjected to Out-Of-Plane Seismic Actions. (Doctor of
Philosophy PhD), The University of Adelaide.
12
