Engineering Structures 95 (2015) 94–111

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: [Link]/locate/engstruct

Investigations on the structural behaviour of archaeological heritage in

Peru: From survey to seismic assessment
Rafael Aguilar a,⇑, Rui Marques a, Karim Sovero a, Carol Martel a, Fernando Trujillano a, Ruben Boroschek b
a
Department of Engineering, Civil Engineering Section, Pontificia Universidad Católica del Perú PUCP, Av. Universitaria 1801, San Miguel, Lima 32, Peru
b
Department of Civil Engineering, University of Chile, Beauchef 850, Santiago de Chile, Chile

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

Article history: The conservation of archaeological heritage is of major importance for preserving the scientific, ethno-
Received 5 March 2014 graphic and artistic values of past cultures. Once archaeological sites are exposed after being buried
Revised 25 March 2015 for centuries, they are subjected to natural hazards, which should be studied with up-to-date techniques.
Accepted 26 March 2015
Moreover, conservation works are primarily focused on aesthetic aspects or on solving localized prob-
Available online 10 April 2015
lems. In earthquake-prone areas, it is of extreme importance to carry out structural analysis studies
for assessing the actual behaviour of archaeological constructions, and for proposing adequate interven-
Keywords:
tion measures. This paper presents an extensive study on structural behaviour of archaeological building
Archaeological building remains
Stone masonry
remains in Peru, based on in-situ non-destructive testing as well as on numerical approaches. The case of
Operational modal analysis the Chokepukio Archaeological Site is presented, which was built between 1000 and 1450 AD in the
Finite element modelling Pre-Columbian era, with a mixed masonry of stone units and earthen mortar. The paper begins with a
Model calibration comprehensive description of the historical, architectural and structural aspects of the archaeological
Pushover analysis site. The possibility of applying operational modal analysis tests is then explored with reference to a
Kinematic analysis representative wall of Chokepukio. The results of the experimental field campaign are used to develop
calibrated finite element models of the wall, and to indirectly estimate mechanical characteristics of
the masonry. Basing on the investigations performed, potential failure mechanisms are identified for
the wall and validated by pushover analysis. Finally, the mechanisms are evaluated through kinematic
limit analysis, to proceed with the seismic assessment.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction remains in the Andean region. On the other hand, the Peruvian
coast is located in the Pacific Ring of Fire, which makes this country
Archaeological heritage buildings are like a constructed calen- a very relevant case regarding the development of a worldwide
dar of the history of civilizations, and thus are of high importance approach for the preservation of archaeological building remains.
for the preservation of cultural, ethnographic and artistic values of Recent seismic events such as the 2010 Chilean and 2003 Iranian
past folks. Worldwide, archaeological remains represent a signifi- earthquakes evidenced, once again, the high seismic vulnerability
cant fraction of the heritage building stock. Much of these building of historical constructions, in which archaeological building
remains, which are mostly made of masonry, have been discovered remains are included. Understanding the structural behaviour of
through archaeological excavations developed without considera- this kind of constructions is particularly complex due to the diffi-
tion of structural aspects. Moreover, the exploration works intro- culty for characterizing the geometry, materials and damage state,
duce new hazards that can occur in open spaces, such as erosion for identifying the structural system, as well as for creating reliable
due to water and wind, and particularly seismic events. Thus, the numerical models [1].
conservation of archaeological building remains requires, beyond The International Council on Monuments and Sites (ICOMOS)
a static stabilization or aesthetic operations, deep structural stud- has published different strategies for studying historical construc-
ies to assess its behaviour in case of exceptional loading events. tions. These strategies evidence the need for a deep knowledge of
Peru has a great legacy of archaeological building remains, from the monument under study, which can only be obtained through
typical earthen constructions in the coast to stone masonry extensive experimental and diagnosis campaigns by means of
laboratory and on-site investigations [2]. In this context, non-
destructive testing is an important tool, since it allows the evalua-
⇑ Corresponding author. Tel.: +51 1 6262000x4610; fax: +51 1 6262000x2813.
tion of constructions without endangering their structure.
E-mail address: raguilar@[Link] (R. Aguilar).

[Link]
0141-0296/Ó 2015 Elsevier Ltd. All rights reserved.
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 95

Vibration-based testing, namely Operational Modal Analysis mechanical parameters of the local andesite stone or masonry tex-
(OMA), is a powerful non-destructive technique for the estimation tures. Effectively, this gap needs to be addressed in future studies.
of the structural dynamic properties of a construction. On the other The height of the constructions ranges from 8 m to 10 m and
hand, the structural evaluation requires adequate methods and each wall seems to be constructed in stages with growth in length
tools for modelling and analysis, particularly regarding the seismic and height, as evidenced by marked transitions amongst stone
assessment. For this purpose, several approaches can be used con- courses. Walls were built with multi-leaf arrangement and are
cerning the nature and complexity of the construction, such as con- 2 m width in average. In general, the walls present trapezoidal or
tinuum finite element models, structural component methods or rectangular niches at different heights. In some cases, the original
rigid block analysis, e.g. [3–7]. However, there are few studies earthen coatings are still visible on walls and niches.
applied to masonry remains and archaeological heritage, e.g. [8,9]. One particular feature at Chokepukio is that higher walls have
For this last kind of structures, the seismic assessment may transversal buttresses, to improve vertical stability, and probably
require a multiple-view analysis approach, using different numeri- also to provide earthquake resistance. However, most of buttresses
cal methods for validating one against the others, and in which the are partially in ruins, making the walls more vulnerable. In order to
Finite Element (FE) simulation is usually the reference for compar- know the soil characteristics and foundation conditions, pit
ison, e.g. [8,10]. However, the development of a reliable FE model excavations were carried out near one wall located in the southeast
requires a calibration of the actual condition of the structure, corner of Sector A (see Fig. 2a and d). There it was possible to find
regarding material parameters, boundary conditions and existing the foundation of the wall, which is 3.0 m in depth and includes
damage. This is usually made based on experimental in-situ vibra- footings to increase the wall stability, see Fig. 2c and f.
tion tests, trying to approximate the experimental modal proper-
ties by the numerical simulation, through a successive process of
updating the model variables. In effect, calibration through vibra- 3. Experimental diagnosis tests
tion tests is an important issue in seismic analysis, e.g. [11].
This paper aims at the structural evaluation of archaeological In-situ experimental investigation was based on OMA [13],
heritage buildings, with application to the Chokepukio which was used as a vibration based non-destructive method to
Archaeological Site in Cusco, Peru. The study includes on-site obtain the dynamic modal properties of the structure (frequencies,
inspection, experimental testing, and numerical modelling and damping and modal shapes).
analysis, performed to assess the seismic vulnerability of the Rainieri and Fabbrocino [14] summarize the available tech-
remaining traces of Chokepukio. After a brief description of the niques for OMA. These techniques consider the measured response
archaeological site, details of the OMA tests carried out on a repre- of a structure under the unmeasured ambient excitations. Even if
sentative wall of Chokepukio are given. Then, the optimization pro- the input is not measured, it is assumed that the ambient excita-
cess of the FE model for the wall is presented, and finally, a tion is banded with a bandwidth large enough to excite most of
pushover analysis is reported together with a kinematic limit the response controlling modes. In general, the technique assumes
analysis to proceed with the seismic assessment. a white noise ambient excitation to identify the modal response
parameters. OMA is especially appropriate for civil engineering
structures with high dimensions and special characteristics, where
2. The Chokepukio Archaeological Site the application of impacts or shakers is too expensive or not feasi-
ble. Vibration based evaluation of existing constructions has
The Chokepukio Archaeological Site is located 30 km from the become a deeply investigated topic, e.g. [15–18]. However, there
city of Cusco, in the Andes of Peru. A wide variety of remaining are still few applications to archaeological heritage, e.g. [9,11].
structures made of stone masonry and mud mortar was found in OMA tests were carried out near the building remains located at
the archaeological site. The seismic hazard at the region is high the southeast corner of Sector A, namely on a couple of walls which
since the archaeological site is located in an area with active faults present the particularity that are shored one against the other with
and Cusco itself is affected by the subduction of the Nazca plate. timber struts (see Fig. 2a, b, d, and e). This sector was selected as
Archaeological investigations in Chokepukio by McEwan et al. case study due to the well preservation of the remaining struc-
[12] evidenced that the original structures were built between tures, e.g. in the studied walls the original plaster is still on the
1000 and 1450 AD. Furthermore, artefacts were found in the site interior face. For this study, only the front wall in Fig. 2 was instru-
corresponding to Lucre and Killke cultures, which are considered mented, which presents variable geometry (thickness varies from
ethnicities of transition between Wari (650–1000 AD) and Inca 1.2 m to 1.8 m at the base and from 0.4 m to 0.6 m at the top),
(1425–1532 AD) cultures. and average length and height of 20 m and 9 m, respectively. The
Chokepukio presents a particular architecture of walls forming wall presents two vertical parts with different stone masonry pat-
enclosures around open spaces (known as ‘kanchas’). McEwan terns. The bottom part is built with large stones and thin mud-
et al. [12] divided the site in three principal areas according to con- mortar joints, while the top part is made of smaller stone units
struction features and occupation periods, namely Sectors A, B and and thick mud-mortar joints. As shown in Fig. 2a–c, the change
C (see Fig. 1a). Sector A, which is the one studied here, presents the on masonry patterns coincides with the change on section in the
highest density of standing structures (twelve ‘kanchas’ at least) height of the wall (at level +6.10 m).
and its walls enclose substantial areas (2600 m2) with small rooms For the experimental tests, sixteen measurement points were
connected amongst them. Sectors B and C correspond, respectively, set in the wall in order to obtain an appropriate characterization
to the beginning of the Inca occupation and post-Wari culture, and of its dynamic response, see Fig. 3a. Due to the availability of a por-
are more degraded. Isometric views of typical walls in each sector table Data Acquisition System (DAQ) with a limited number of
are presented in Fig. 1b–d. measurement channels, only four accelerometers were used for
The constructions at Chokepukio were built using andesite the tests. With this limitation, the test planning considered seven
stone, which is an extrusive igneous rock named after the Andes. setups with two reference nodes (located at the expected higher
The masonry is composed of irregular stones interposed with modal amplitude points) and two roving sensors. The sensors lay-
mud-mortar joints of thickness ranging between 2.5 cm and out was designed in such a way that the behaviour of the bottom
10 cm. The mortar is a mixture of local soil, clay, straw, and cactus and upper part of the wall could be properly measured. Sensors
resin. Unfortunately, there are no reported studies about material were criteriously installed in two rows at the bottom part of the
96 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

SECTOR C

SECTOR B

0 100 m

SECTOR A

(a)

0 7m 7m 7m
0 7m 0
0

(b)

0 7m
0 3m

(c) (d)

Fig. 1. The archaeological site of Chokepukio: (a) location in Cusco and plan of its sectors, and typical walls of (b) Sector A, (c) Sector B and (d) Sector C.

wall, namely at the level of niches and at the top (thickness transi- relative motion between the sensors and the wall was limited
tion section), which are considered as sections representative of mainly by friction, and the rocking motions were avoided in the
the global behaviour. A third row of sensors was established at frequency of interest through an appropriated levelling of the sen-
the mid- height of the upper part of the wall to capture the average sors. Frequency analysis for each sensor indicated consistent and
behaviour of that zone. good quality records for frequencies below 10 Hz.
In the case study here, highly sensitive sensors were used to Regarding the acquisition time needed to obtain enough data
capture the dynamic response in time domain. The root mean for OMA purpose, Brincker et al. [19] propose a simple rule of
square values from the acceleration time series recorded in the thumb for the time length of recorded data, which is defined as
tests (see Fig. 4) vary from 0.012 mg to 4.8 mg, while the noise inversely proportional to the product of the damping ratio by the
level of the transducers is 0.001 mg rms. The transducers used natural frequency of a given mode. However, in this study and
were piezoelectric accelerometers with a sensitivity of 10 V/g and for all setups, the sampling rate was set to 200 Hz and the acquisi-
a dynamic range of ±0.5g together with a portable USB-powered tion time to 10 min following the recommendation by Ramos [20]
24 bits DAQ system (see Fig. 3b and c) with a maximum sampling (sampling time in the order of 1000–2000 times the first natural
rate of 51.2 kHz, bandwidth of 23.04 kHz, AC coupled (0.5 Hz), period). To compute the averaged spectrum, the cross power den-
antialiasing filters, and 102 dB of dynamic range. As shown in sity function was estimated using the Welch averaged modified
Fig. 3d, an external scaffolding (not connected to the structure) periodogram method [21] and considering 1024 points as signal
was necessary to place the transducers along the wall. In this case, length, 50% as overlap criterion, and a decimation factor of 5.
the accelerometers could not be fully fixed to the wall due to the The preliminary check of the signal quality was based on the
intangibility of the monument, and thus, the transducers were analysis of the obtained averaged spectrum presented in Fig. 5a.
screwed to metallic cubes that were conveniently perched in The clear and well-spaced peaks shown in the spectrum evidence
the wall. The sensors were placed on the wall in order to minimize the high quality of the acquired signal. Based on this spectrum, a
any relative motion between the transducers and the wall. The preliminary system identification was carried out using the Peak
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 97

(a) (b)

+9.90 +9.90

+6.10 +6.10

Superstructure

+0.00 +0.00

Foundation 0 5m
-3.00
-3.00
20 m
(c)

Sector A - Chokepukio
N
0 50 m

(d) (e) (f)

Fig. 2. General views of the studied area: (a) façade of the instrumented wall, (b) elevation of the instrumented wall, (c) front and transversal sections of the studied
structure, (d) plan of the studied sector adapted from [12], (e) internal view of the two studied walls and (f) foundation detail.

Picking (PP) method [22,23]. The results of this processing denote a The modal shapes identified with basis on the dynamic tests
high flexibility of the wall, since at least the first seven natural fre- mainly denote out-of-plane deformations, as shown in Fig. 6a–g.
quencies are located below 10 Hz. These results were later con- The first mode shape corresponds to a global translation motion
firmed using a more refined data processing technique, namely of the wall with prevalence of displacements at its top part, where
the Data-Driven Stochastic Subspace Identification (SSI-data) a slight oscillatory movement is observed. The second mode is
method [24] performed in ARTEMIS [25]. In this case, the perfectly through a global torsion of the wall and with the top part of the
aligned poles on the stabilization diagram in Fig. 5b confirm the wall oscillating in its full length. The third mode shape is like a
accurate identification of the first seven modes. The results also wing stroke of the top part of the wall that combines translational
show that the small or sharp peaks obtained in the averaged spec- and rotational movements. The fourth mode denotes a global oscil-
trum of Fig. 5a (peaks between the second and the third mode, the latory motion of the wall, mainly around the vertical axis, and
fourth and fifth ones, and the sixth and seventh modes) correspond which presents large amplitude at the top part. The fifth, sixth
to spurious modes, and should be discarded. and seventh modes are similar and correspond to local modes at
Table 1 reports a summary of the results of natural frequencies the corners of the wall, where the displacements are concentrated
and damping ratios. As shown, frequency values present very low through mixed motions.
error margins (of less than 2%) comparing the results of the PP and In order to evaluate the quality of the estimations, the AutoMAC
SSI methods. Concerning the damping, it is know that the com- matrix is presented in Fig. 6h, which correlates the set of estimated
puted ratios from OMA tests are not precise, and furthermore the mode shapes, in terms of the MAC ratio (see definition in
damping estimation is complex and very sensitive to the masonry Section 4.2), amongst themselves. The AutoMAC matrix confirms
type, but the predicted values around 3% are acceptable, e.g. [26]. the quality of the dynamic identification of the first four modes,
98 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

12 13 14 15 16 11
Ref. channels

7 8 9 10 11

1 2 3 4 5 6

(a)

(b) (c) (d)

Fig. 3. OMA tests at Chokepukio: (a) test setup, (b) close up of a measurement node, (c) central acquisition station and (d) general view of the instrumented wall and the
process of fixing sensors.

which are fully uncoupled, and denotes contamination between for the mortar density, the specific weight of the masonry is
the measured degrees of freedom in the last three higher modes. expected to be around 25–26 kN/m3.
This last aspect can also be due to the presence of a weak plane A first model (Model 1) which includes the instrumented wall
at the interface of the upper and bottom parts of the wall, and and the other one located behind it, was built taking into account
therefore the response is mainly characterized by lower modes. the interaction between the two walls through the three existing
Thus, because of this predominant behaviour and of the location timber struts, as presented in Fig. 7a. The models were assembled
and limited number of sensors used in the tests, the higher modes using eight-node isoparametric brick elements of type HX24L. The
were poorly captured. An aspect that could also have influenced timber struts are eucalyptus pieces with an elastic modulus of
the experimental results is the dynamic interaction between the 15,000 MPa [30], which were modelled as beam elements of type
instrumented wall and the other one located behind it, since the L12BE. In Model 1, the timber struts were modelled as hinged to
two walls share a common foundation and are somewhat con- allow rotations around the vertical and orthogonal planes to the
nected through the timber struts that may behave in a nonlinear wall, and no possibility of sliding was considered. The boundary
fashion. The study of this phenomenon might be of interest and conditions of the timber struts were adjusted in order to approxi-
should be considered in further experimental campaigns, particu- mate better the experimental modal properties. Even if the founda-
larly referring to the work of Rainieri et al. [27]. For further analysis tion of the structure and the restraining conditions below the
stages, only the first four mode shapes were taken into account. ground may influence the mode shapes, particularly by the fact
that the two parallel walls share a common infrastructure, it was
considered that this effect is somewhat dissipated since the foun-
4. Numerical modelling
dation is significantly deep and fully buried. Thus, for the sake of
simplicity, the structure was assumed as fixed at the base. The
Today, advanced numerical tools are available with large appli-
walls were modelled with variable thickness in height correspond-
cation in the field of structural engineering, particularly Finite
ing to section changes. The thickness was also considered variable
Element (FE) software. Here, a computationally sustainable
in the length of the instrumented wall, from 1.20 m on the left side,
approach for detailed modelling of complex structures is presented
to 1.50 m from the door to the right side.
aiming at the development of an accurate model to support the
The second considered model (Model 2) was created assuming
seismic assessment of the studied wall in Chokepukio.
the instrumented wall as decoupled, and simulating the interac-
tion from the timber struts through elastic springs with an equiva-
4.1. Model development lent axial stiffness, once the wooden struts are purely wedged
between the walls (see Fig. 7b). Considering that the struts work
Three finite element models, which are presented in Fig. 7, were in the range of small axial deformations with a low gradient com-
developed using DIANA [28] aiming at finding the numerical model pression-decompression behaviour, the adopted stiffness value is
that better represents the structure in study. Initially, the models the complete axial component EA/L, where E is the elastic modulus
were built considering the masonry as a homogeneous material of the wood, A is the cross-section area corresponding to a timber
with an elastic modulus of 800 MPa, according to reference values diameter of 0.1 m, and L is the strut length with a value of 1.0 m.
from Brignola et al. [29]. Considering that the density of the ande- The strut axial stiffness results with a value of 118.0 MN/m.
site stone in Cusco is about 2700 kg/m3, and a slightly lower value Finally, a third model (Model 3) was built assuming the two walls
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 99

Setup 1 Setup 2 Setup 3

12 12 12

8 8 8

Acceleraon (mg)
4 4 4

0 0 0

-4 -4 -4

-8 -8 -8

-12 -12 -12

0 10 0 10 0 10
Time (min)

Setup 4 Setup 5 Setup 6

12 12 12

8 8 8
Acceleraon (mg)

4 4 4

0 0 0

-4 -4 -4

-8 -8 -8

-12 -12 -12

0 10 0 10 0 10
Time (min)

Setup 7
12

8
Acceleraon (mg)

0 -------- Accelerometer 1 (ref 1)

-------- Accelerometer 2 (ref 2)
-4 -------- Accelerometer 3
-------- Accelerometer 4
-8

-12
0 10
Time (min)

Fig. 4. Plots of acceleration time series obtained from the OMA tests.

P
as completely disengaged (only the instrumented wall is consid- j n ui;EXP ui;FE j2
ered, see Fig. 7c). MAC ¼ Pn i¼12 Pn ð1Þ
i¼1 ui;EXP2
i¼1 ui;FE
As explained before, the numerical modal analysis and the sub-
sequent model updating process were performed considering only The results of frequencies, obtained for each model before the
the first four natural frequencies and corresponding modal shapes. optimization process, are summarized in Table 2. Fig. 8 shows the
In order to compare experimental and numerical modes, the Modal FMAC graphs, where the MAC values and frequency scales are
Assurance Criterion, MAC [31] and the Frequency scales with MAC globally compared for the three numerical models. All models
combination, FMAC [32] were used. The MAC ratio is to provide a provide an accurate estimation of the first mode shape, with high
measure of consistency (degree of linearity) between estimates MAC values (about 0.95) and a tolerable frequency difference
of a modal vector. In this case, it is computed through Eq. (1) to (from 20% to 36%). However, Models 1 and 3 provide in general
correlate the experimental and numerical modal vectors, respec- a better approximation concerning the second and third mode
tively uEXP and uFE, and considering the set of n estimated degrees shapes, with MAC values from 0.86 to 0.92. These two models
of freedom. In Eq. (1), the index FE corresponds to the results of the provide also better approximation between experimental and
numerical simulation, while the index EXP refers to the experimen- numerical frequencies, with a frequency difference from 20% to
tal results. MAC values close to one indicate high correspondence, 37% when excluding the worst prediction by both models. The
whereas values close to zero indicate poor resemblance. The FMAC fourth mode was in general more difficult to capture with a good
is a graphical representation that provides a general comparison of approximation in all models, i.e. MAC values lower than 0.57 and
experimental versus numerical modal properties from several frequency differences from 31% to 49%. It can be noted that
mode shapes, by considering simultaneously the mode shape Models 1 and 2 seem to suffer some bias error [22], i.e. a devia-
correlation (MAC ratio), the degree of spatial aliasing and the fre- tion between the measured and predicted frequency while its
quency comparison. range value increases.
100 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

(a) 1.99 Hz
3.16 Hz
4.30 Hz

Power/Frequency (dB/Hz)
5.12 Hz 6.33 Hz
6.72 Hz
9.18 Hz

Frequency (Hz)

Test setup
(b)

Frequency (Hz)

Fig. 5. Data processing results: (a) Welch spectrum for the PP method and (b) stabilization plot of the SSI method.

numeration in Fig. 3a. Model 1 presents lower COMAC values for

Table 1
System identification results from the experimental field campaign. points located in the higher part of the wall, in particular the cen-
tral node (in position 14), which can be because this wall part is
Mode Peak Picking Stochastic Subspace Frequency relative
significantly more flexible (presents larger displacements) than
(PP) Identification (SSI) error (%)
the remaining structure, and thus is more prone to errors.
Frequency Frequency Damping
Models 2 and 3 present a similar approximation, but Model 3 is
(Hz) (Hz) (%)
slightly better with reference to DOFs in the lower part of the wall.
1 1.99 1.98 2.6 0.51
P
2 3.16 3.19 3.2 0.94 j m
j¼1 uij;EXP uij;FE j
2
3 4.30 4.39 3.2 2.05 COMACi ¼ Pm 2 Pm 2 ð2Þ
4 5.12 5.08 2.2 0.79 j¼1 uij;EXP j¼1 uij;FE
5 6.33 6.29 2.4 0.64
6 6.72 6.70 2.6 0.30 All considered, Models 1 and 3 present similar approximations.
7 9.18 9.19 3.0 0.11 However, due to the significant bias error at Model 1 and the sim-
plicity of Model 3, this last one was selected for calibration through
an optimization process.
Considering that the MAC ratio is a global indicator, the COMAC
(Co-ordinate Modal Assurance Criterion) [20] was also evaluated, 4.2. Optimization procedure and results
which allows to obtain local information from the measurement
points. The COMAC ration is to identify which measurement The purpose of the optimization process is to find the most
degrees-of-freedom contribute negatively to a low value of MAC. appropriate values for unknown variables (which are set at the
In this case, it is computed through Eq. (2) for each degree of free- beginning of the process) in order to approximate the numerical
dom (DOF) associated to a measuring point, and its value indicates frequencies and mode shapes to those experimental. The process
the general approximation between the experimental and numeri- is monitored through computation of an error or objective func-
cal modal displacements of the DOF, for the set of considered m tion, which is minimized using a nonlinear least square method.
mode shapes. The closer the COMAC value is to one, the more simi- The objective function is defined in Eq. (3), according to Ramos
lar are the experimental and numerical modal displacements of a [20]. In this equation, the index FE corresponds to the results of
given DOF. The obtained COMAC values are presented in Fig. 9a, the numerical simulation, while the index EXP refers to the experi-
relating to the three considered models and according to the node mental results.
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 101

(a) (b) (c) (d)

f1 = 1.98Hz f2 = 3.19Hz f3 = 4.39Hz f4 = 5.08Hz
ξ1 = 2.6% ξ2 = 3.2% ξ3 = 3.2% ξ4 = 2.2%

(e) (f) (g)

f5 = 6.29Hz f6 = 6.70Hz f7 = 9.19Hz (h)
ξ5 = 2.4% ξ6 = 2.6% ξ7 = 3.0%

Fig. 6. Experimental mode shapes: (a–g) views of the first seven modes and (h) AutoMAC matrix.

Instrumented wall Instrumented wall Instrumented wall

Timber struss
Springs

Backward wall 0 5m

(a) (b) (c)

Fig. 7. Developed FE models: (a) Model 1 considering interaction between walls, (b) Model 2 assuming the instrumented wall as decoupled and (c) Model 3 considering only
the instrumented wall.

natural frequencies and mode shapes, respectively. The frequencies

Table 2
and MAC parameters, fj,FE and MACj,FE/EXP, are made explicit accord-
Experimental and predicted frequency values obtained for each model.
ing to Eqs. (4) and (5), which are based on the Douglas and Reid [33]
Mode Experimental (Hz) Model 1 (Hz) Model 2 (Hz) Model 3 (Hz) approach. In these equations, V is the number of the unknown vari-
1 1.98 2.67 [34%] 2.72 [36%] 2.40 [20%] ables; X is the vector of the unknown variables; and A, B and C are
2 3.19 4.81 [52%] 5.06 [60%] 4.32 [37%] constants which are computed through solving a system of equa-
3 4.39 5.94 [36%] 6.36 [45%] 5.95 [36%] tions considering base, lower and upper boundaries for the
4 5.08 6.67 [31%] 7.58 [49%] 7.41 [45%]
unknown variables (2V + 1 equations are considered). In the original
The difference in frequency is indicated inside brackets. formulation, the number of updating natural frequencies must be
equal or larger than the number of unknown variables.
2 !2 m 3
2 2 X
2Vþ1
1 4Xmw
f j;FE  f j;EXP X u
25 f j;FE ¼ C j þ Aj;k X k þ Bj;k X 2k ð4Þ
e¼ W f ;j 2
þ W MAC;j ðMACj;FE=EXP  1Þ ð3Þ
2 j¼1 f j;EXP j¼1
k¼1

X
2Vþ1

where W refers to weighting matrices for frequencies (Wf) and MAC MACj;FE=EXP ¼ C j þ Aj;k X k þ Bj;k X 2k ð5Þ
k¼1
values (WMAC), while mw and mu are the number of considered
102 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

8 8 8

Experimental frequencies
Experimental frequencies

Experimental frequencies
7 7 7
6 6 6
5 5 5
0.57 0.51 0.49
4 4 4
0.92 0.85 0.92
3 3 3
0.88 0.83 0.86
2 2 2
0.96 0.95 0.97
1 1 1
0 0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
FE model numerical frequencies FE model numerical frequencies FE model numerical frequencies
(a) (b) (c)
Fig. 8. FMAC graphics for: (a) Model 1, (b) Model 2 and (c) Model 3 (frequency values in Hz).

(a) 1

0.8

0.6

0.4

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Model 1 Model 2 Model 3 Position

(b) 1

0.8

0.6

0.4

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Initial Model 3 Updated Model 3 Position

Fig. 9. Comparison of COMAC values for all the measured DOFs, between: (a) initial Models 1–3 and (b) initial and updated Model 3.

In this study, the unknown variables were selected through process, were carried out considering the possibility of different
sensitivity analysis, from an initial set that included geometrical masonry qualities for the bottom and top parts of the studied wall.
aspects, boundary conditions and material properties. However, In this case, an interface exists between the two parts of the
it is known that in general the numerical mode shapes present wall. However, from the on-site inspection it was not possible to
low sensitivity to global parameter variation, e.g. to the elastic conclude about the texture of the interface, but the transition
modulus. Effectively, the conclusion was that the variables with between the two parts is probably made through interlocking of
more influence for changing frequencies were material properties stone units in the two bodies. Under this hypothesis, a third zone
(E-modulus and specific weight), while the modal shapes were in the wall is to be considered with an intermediate E-modulus
mostly influenced by the geometry (different sections in length value. However, due to uncertainty and for the sake of simplicity
and height of the wall). There are studies in the literature specifi- only two masonry qualities were considered with a perfect inter-
cally focused to FE model updating, namely using sophisticated face connection. In effect, the critical value for capturing the mode
formulations to perform model optimization at both global and shapes is most likely the ratio of E-moduli rather than the individ-
local levels, e.g. [34,35]. In this work, local optimization has been ual values per se.
only applied in a simplified manner, e.g. the geometry was locally The application of the Douglas–Reid approach requires the def-
changed based on personal sensitivity. inition of base, lower and upper bound values for the input vari-
In effect, a first model was built considering the topographic ables (E-modulus and specific weight), which were set as
survey and a preliminary manual modal shape tuning, but assum- presented in Table 3. While the base values are the assumed start-
ing only a thickness change between the bottom and top parts of ing values for variables, the lower and upper bound values circum-
the wall. Following, a third thickness change was additionally con- scribe the space of searching. The range of values in Table 3 was
sidered at the top part of the wall, see Fig. 7c, searching for a better defined by identifying the bound values of the considered vari-
approximation. This change of thickness is apparent in the real ables, for stone masonry. This range is not referred to a single
wall. Once the model geometry was set, a model calibration using optimization run, but to a sequence of runs for which the searching
the Douglas–Reid approach and an automatic optimization interval of parameters was successively changed in order to obtain
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 103

Table 3 Table 4
Range of values for the updating variables. Values for the updating variables obtained after optimization.

E-modulus (MPa) Specific weight (kN/m3) Updating variables Ebottom part (MPa) Etop part cbottom part ctop part
(MPa) (kN/m3) (kN/m3)
Bottom part (E1) Top part (E2) Bottom Top part (c2)
part (c1) Initial value 650.0 650.0 30.00 30.00
Final value 579.7 210.1 25.32 25.27
Base value 650.0 650.0 30.00 30.00
Lower value 300.0 210.0 20.00 20.00
Upper value 1300.0 1300.0 35.00 35.00

580 MPa and 210 MPa. In the case of the specific weight, a value
close to 25 kN/m3 is obtained for both parts.
a coherent (non-local) solution with a minimal error value. In some The weighting matrices resulting from applying the Douglas–
cases, the searching interval of parameters was purposively Reid approach are square diagonal matrices. In this case, the
enlarged to extend the searching space, e.g. by considering an dimension of the weighting matrices is 9  9 as result of the
upper bound of 35 kN/m3 for the specific weight. combination of base, lower and upper values for a set of four mode
The optimization process was implemented by using the shapes. In a first stage, these matrices were automatically calcu-
‘lsqnonlin’ trust-region-reflective least squares algorithm available lated with the inverse of the normal variance of each modal quan-
in Matlab [36], which is a search method based on minimizing the tity [20], and afterward, the matrices were adjusted to find a better
sum of the squares of the differences between the predicted and solution for the optimization problem. The weight values are pre-
measured values. Fig. 10 shows the evolution of the objective func- sented in the matrices in Eqs. (6) and (7), respectively for frequen-
tion, and also of the values of the input variables against the num- cies and MAC values, and which contain in each row a vector with
ber of iterations. As shown, the values became stable at the the values of the main diagonal of the weighting matrices, in
iteration 100, and convergence was attained after 300 iterations. correspondence with the four considered mode shapes. Based on
Even if the objective function curve is practically flat after 150 the obtained weights for MAC values, it can be noted the high
iterations, the elastic modulus of the top part of the wall (E2) fol- sensitivity of the objective function to the first mode shape, while
lows with a slight variation (9 MPa) until 300 iterations. It can the fourth mode shape presents a small influence.
be observed that E2 is the last converged variable, and presents
2 3 2 3
the larger value variation in the searching process. diagðW f ;1 Þ 6:16 3:96 7:51 5:48 6:34 6:36 5:98 6:93 5:60
6 diagðW Þ 7 6 7:82 5:13 10:77 6:92 8:11 8:36 7:34 8:41 7:31 7
The large decreasing of E2 was in part due to large overestima- 6
6
f ;2 7 6
7¼6
7
7
tion of its initial value. In effect, the masonry in the top part of the 4 diagðW f ;3 Þ 5 4 5:25 3:91 5:88 4:04 5:86 5:63 4:96 5:71 3:14 5
wall is most an earth-mortar/rubble stone mix, thus presenting a diagðW f ;4 Þ 7:02 5:06 8:10 5:69 7:64 7:67 6:55 7:51 6:65
significant lower elastic modulus. Furthermore, the modal ð6Þ
response is strongly determined by the deformability of the upper
2 3
part of the wall. The results of the optimization process are sum- diagðW MAC;1 Þ
marized in Table 4. The final values for the varying parameters evi- 6 diagðW 7
6 MAC;2 Þ 7
6 7
dence a clear difference of the E-modulus for the masonry at the 4 diagðW MAC;3 Þ 5

bottom and top parts of the wall, respectively with values of diagðW MAC;4 Þ
2 3
215:43 215:32 214:70 212:42 215:49 215:27 215:52 215:54 215:32
6 18:00 16:43 18:95 19:68 17:16 18:40 17:47 17:25 18:33 7
6 7
¼6 7
4 12:42 10:34 12:89 13:03 10:89 12:78 11:59 11:18 12:72 5
(a) 3:78 3:57 4:99 6:53 3:73 5:17 3:75 3:76 4:84
10000 ð7Þ
Error function value (e)

The reduced values of the elastic modulus for the studied wall
1000
are determined by the relatively low value of the E-modulus of
the earthen mortar, since the local soils present a great deformabil-
100
ity [37], and even if the mortar is improved with straw and cactus
resin. On the other hand, Pinho et al. [38] report, for rubble stone
10 masonry, an elastic modulus in the range 200–400 MPa and a com-
pressive strength around 0.4 MPa, even if the compressive strength
1 of the stone is about 50 MPa.
0 50 100 150 200 250 300
The FE model was then updated by considering the optimized
Iterations
values for variables. The final FMAC relationship, which is pre-
7.0
sented in Fig. 11a, denotes high correspondence between numeri-
(b) cal and experimental frequencies and mode shapes: the maximum
6.0
difference in frequencies is less than 8%, while the minimum MAC
Variable values

5.0 value is 0.86. For a better comparison, Fig. 11b and c presents the
4.0 first four mode shapes regarding the experimental and numerical
3.0
approaches. It is also to note the improvement, in general, of the
COMAC values of the updated model relatively to those of the
2.0
model before optimization, see Fig. 9b. Furthermore, it is observed
1.0 in Fig. 9b that, for Model 3, while the COMAC values at the top part
0.0 of the wall are improved after updating, the ones in the bottom
0 50 100 150 200 250 300 part of the wall are worsened. In fact, this tendency of the
Iterations COMAC values also reflects the accuracy of the numerical model.
E1 (x10² MPa) E2 (x10² MPa) γ1 (x10 kN/m³) γ2 (x10 kN/m³)
Considering that the updated Model 3 is a better approximation
Fig. 10. Optimization process: (a) evolution of the objective function over iterations to the reality, the lowest COMAC values are verified in the bottom-
and (b) evolution of each variable over iterations. right part of the wall, where the points most distant to the
104 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

Experimental frequencies (Hz)

6

5
0.86
4
0.96
3
0.94
2
0.96
1

0
0 1 2 3 4 5 6
FE model numerical frequencies (Hz)
(a)

1st Mode 2nd Mode 3rd Mode 4th Mode

1.98 Hz 3.19 Hz 4.39 Hz 5.08 Hz

(b)

1.88 Hz 3.37 Hz 4.03 Hz 5.29 Hz

(c)
Fig. 11. Experimental versus FEM mode shapes: (a) FMAC after optimization process, (b) experimental mode shapes and (c) FEM numerical mode shapes.

reference nodes are placed (in positions 4, 5 and 6). In effect, the push-over loading and kinematic limit analysis are presented and
COMAC values seem to suffer a kind of bias error, which increases applied to the case of Chokepukio.
with the distance to the reference nodes. The obtained results evi-
dence the utility of the optimization process and the reliability of 5.1. Pushover analysis
the updated model regarding its use in structural assessment.
Nonlinear static (pushover) analysis is an approach to evaluate
the seismic response of buildings through simulating an incremen-
5. Seismic assessment tal static lateral loading of the structure, as an alternative to non-
linear time-history analysis [42]. However, pushover analysis
The preoccupation with earthquake effects on buildings has presents limitations, particularly the consideration of a negligible
been considered from ancient times, see [39,40]. The preservation influence from the higher vibration modes, since the analysis fails
of ancient constructions is also a great challenge for the future. in predicting local damages occurring previously to the considered
Performance-based approaches have been proposed for the seismic mechanism. Certainly that a nonlinear dynamic analysis provides a
safety assessment of structures, e.g. the displacement-based N2 more realistic seismic response, which is however highly sensitive
method by Fajfar [41], which are however mostly applicable to to the seismic input and its uncertainties, and thus the pushover
the case of buildings with box behaviour [42]. The seismic assess- seems a more robust approach for practical purposes, see [44].
ment of structures without box action is generally more difficult to Once a nonlinear material law is considered, the pushover
establish, since the walls present an almost independent behaviour analysis allows capturing plastic and cracking events, and thus
and significant out-of-plane deformation and damage components. the inelastic source of the structure. Pushover has become a very
In this case, and particularly when dealing with complex geome- popular approach for seismic design of real or idealized framed
tries, the concept of structural performance needs most likely to structures, for which the nonlinear behaviour is lumped in plastic
be applied at local level, namely through reduction of accelerations hinges that are activated after a limit rotation is reached. For
and control of stresses. On the other hand, based on real and masonry structures, pushover has been mostly applied to planar
experimental evidence, it can be observed that walls tend to models assembled through shell elements, which can normally
behave as rigid bodies subject to rocking, e.g. [43]. In the following, include shear deformations through the wall length, but no bend-
approaches for seismic assessment of wall structures based on ing, e.g. [8]. In cases of structures eminently three-dimensional and
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 105

strongly influenced by bending, modelling with brick solid ele- bond of the masonry, and the tensile fracture energy equal to
ments is more adequate, since the formation of hinges is mainly 50 N/m [47]. The tensile strength adopted for the Chokepukio wall
due to flexural cracking. represents 10% of the assumed compressive strength for the
However, it is known that the available material models in FE masonry in the bottom part of the wall, which is a common
software are mostly validated with reference to bi-dimensional assumption for masonry, e.g. [46]. This value was considered con-
structures subjected to in-plane loading. Thus, further investiga- stant in the wall even though the bonding of the stones and mortar
tion is required to extend such models to a three-dimensional seems to be better in the upper part of the wall.
material domain, and to validate its accuracy in cases of out-of- In this work, the pushover analysis was mainly performed to
plane mechanisms. Material models based on the Drucker–Prager identify the critical sections and potential collapse mechanisms
criteria are commonly used for solid models of masonry construc- of the studied wall, and thus only the overturning of the wall to
tions, which present advantages from the analytical and com- outside the principal façade was considered. The pushover was car-
putational point of view, since the considered failure criterion ried out under conditions of constant gravity load, and the lateral
presents a smooth failure surface, the masonry is considered as a load pattern was assumed to be proportional to the mass regard-
continuum media and the model requires the definition of few less of the elevation. This load distribution was considered to allow
parameters [45]. On the other hand, modelling the out-of-plane a more direct comparison with the kinematic analysis, which con-
behaviour of masonry as a discontinuous material is difficult to siders mainly global rocking mechanisms. Given the large com-
handle in a FE context, and discrete element approaches have been putational effort and non-convergence associated with the
developed with this purpose, e.g. [44]. regular method when dealing with complex solid-based models,
For the Chokepukio case study, the nonlinear behaviour of the the modified Newton–Raphson method, combined with arc-length
stone masonry was considered by the adoption of a constitutive control and the line-search technique, was adopted to obtain the
law based on a total strain crack model, which considers an isotro- solution of the nonlinear problem.
pic behaviour with a compressive cap and fixed smeared cracking Fig. 12 presents the evolution of the mechanism for the wall
[28]. This model is based on direct implementation of experimen- when subjected to the push-over loading proportional to the mass.
tal observations, and furthermore, it provides stability in the crack- The damage distribution was assumed as given by the vertical
ing control and moderate computational cost [8]. Stress–strain strains as an indication of cracking. As shown in Fig. 12a, after
relations were assumed considering exponential softening for ten- the elastic stage, cracking associated to bending tensile strains
sion and a parabolic law in compression. The post-cracked shear develops contemporaneously at the wall base and thickness transi-
behaviour was considered assuming a shear retention factor of tion sections, and a combined motion of the two parts of the wall is
0.1, to allow an important shear transference after crack occur- observed. These strains mainly generate sub-superficial cracks and
rence. Based on the updated values of the E-modulus, the compres- produce a small degradation of the wall stiffness, which reflects in
sive strength was considered different for the two masonry the capacity curve (segment a–b in Fig. 12e). The cracks spread in
patterns of the studied wall at Chokepukio, and its value is the wall face areas adjacent to the wall base and thickness transi-
assumed as E/500, which is the mid-range relation of the interval tion sections (Fig. 12b). In the following stage, cracking concen-
proposed by Tomaževič [46]. The compressive fracture energy trates in the wall base and propagates to inside the wall base,
was defined by multiplying the compressive strength by a ductility while the tensile strains in the top part are released. In this phase,
factor equal to 1.6 mm [47]. The masonry tensile strength was the wall forms a plastic mechanism at the base, which is denoted in
assumed with a value of 0.1 MPa considering the relatively good the plateau b–c of the capacity curve, and starts developing a

(a) (b) (c) (d)

0.30

0.25
b
c
Load factor (g)

0.20
a d
0.15

0.10

0.05

0.00
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
(e) Displacement (mm)

Fig. 12. Pushover analysis: (a–d) evolution of the wall mechanism with representation of vertical strains (as an indicator of cracking) and (e) capacity curve.
106 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

global rocking mechanism (Fig. 12c). Finally, the mechanism which are instructed for practical purposes in the CNTC [51] – Part
evolves with the damage propagating to within the wall base C7: Design for seismic actions.
(Fig. 12d), and with the load factor decreasing in the capacity curve This procedure has been applied in seismic assessment of
against increasing deformation (segment c–d in Fig. 12e). The iden- masonry structures complementarily to FEM-based approaches,
tified sequence is fully evidenced in the capacity curve presented e.g. [4,5,8]. Limit analysis is in general not sufficient for a full struc-
in Fig. 12e, which plots the load factor, defined as the ratio between tural analysis under seismic loads, but it can be used to obtain a
the horizontal load and the wall weight, versus the displacement of simple and quick estimation of collapse loads and failure mecha-
a control point at the wall top. nisms. This is probably the most realistic approach for practical
The estimated push-over response according to the considered seismic assessment of archaeological sites, which are generally
rocking mechanism will be used to validate the kinematic constituted by a large set of substructures that are weakly con-
approach for seismic assessment presented in Section 5.2. nected and without forming a closed contour, e.g. Fig. 13.
However, it needs to be noted that the obtained results are in The key of limit analysis is the definition of potential collapse
accordance with the typical behaviour verified for slender masonry mechanisms, which can be a relatively difficult task depending
elements failing by rocking, i.e. developing a hinge at the base and on the particularities of the structure. For a cantilever wall, a
presenting a significantly ductile response. mechanism requires the formation of only one hinge, whose loca-
tion is expected to be defined, under the lower bound theorem and
corresponding to the inferior limit condition [48], with the thrust
5.2. Kinematic limit analysis line touching the section edge, as illustrated in Fig. 14a. In this case,
formation of hinges is also to be expected in sections with marked
The plastic or limit analysis has been historically developed as a geometry change, as also confirmed in Section 5.1. After definition
simplified approach for evaluation of structures that present a of the collapse mechanism, a kinematic approach is used to evalu-
behaviour determined by the formation of plastic hinges and ate the load multiplier that activates the mechanism, a0, which is
development of collapse mechanisms. In this case, a simplification the relationship between the horizontal load and self-weight
can be assumed by considering a static equilibrium approach, such applied to each body involved in the mechanism. Then, the solu-
as that developed for the theory of arches [48]. For the case of tion for the equilibrium can be obtained through application of
masonry buildings, the common approach for limit analysis is the principle of virtual work (PVW), which can be formulated
based on macro-block discretization, by assuming collapse mecha- (for a mechanism involving n bodies, m weight loads from dead
nisms for large structural assemblages. Even if limit analysis is very bodies, and o external forces) according to:
simple, it was included in seismic codes as a possible method for " #
local failure assessment. The seismic verification of local mecha- X
n X
n þm X
n X
o

nisms through limit analysis is specified in the Italian directive

a0 W i dx;i þ W j dx;j  W i dy;i  F h dh ¼ Int: Work ð8Þ
i¼1 j¼nþ1 i¼1 h¼1
DPCM [49], which is addressed to heritage constructions. This reg-
ulation prescribes the application of the procedures specified in the where Wk is the weight of the body k; dx,k and dy,k are virtual dis-
Italian building code NTC [50] – Part 7: Design for seismic actions, placements of the body k relatively to its mass centroid, in x

(a) (b)

Fig. 13. Example of building remain: (a) view and (b) idealised potential collapse mechanisms.

δ x,j
α0Wj δ y,j

Wj
δh
Fh
δ x,i
α+ W
α- W αW δy,i α0Wi
W
W W Wi
ϕ O (b)
(a)
Fig. 14. Collapse mechanism for cantilever wall: (a) activation and (b) equilibrium variables.
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 107

(horizontal) and y (vertical) directions, respectively; Fh is an exter- strength is generally high and the crushed length of the edge sec-
nal force applied to a body and dh is its corresponding virtual dis- tion is relatively small comparatively to the wall thickness.
placement on the body; see Fig. 14b. The appropriateness of applying the kinematic approach in the
After activation of the mechanism, a progressive motion of the case of Chokepukio was evaluated through comparison of the
body occurs until reaching a maximum displacement state, mechanisms and capacity curves obtained from applying the push-
corresponding to a zero value for the load multiplier (weight load over and kinematic analyses to the studied wall. Referring to the
vector is aligned with the hinge point), see Fig. 15a. The kinematic overturning of the wall to outside the principal façade, a rocking
response considers the horizontal action that the structure is pro- of the wall is identified from the pushover analysis similar to the
gressively able to stand with the evolution of the mechanism, until global rigid-body mechanism assumed in the kinematic analysis.
the complete dissipation of the horizontal force itself, i.e. as long as This is denoted in Fig. 16, where the kinematic curve present some
the structure is not able anymore of stand horizontal actions. The features similar to those of the force–displacement pushover
relationship between the load multiplier and the displacement dk response that give sense to a comparison. It can be noted the
of a control point k can be assumed as linear according to Eq. (9) matching of the two responses at the yield displacement of the
and Fig. 15b. pushover curve (40 mm), after that this curve slightly increases
up to a displacement of 90 mm, and finally it decreases quicker
a ¼ a0 ½1  dk =dk;0  ð9Þ than the kinematic curve due to internal damage to the wall.
Furthermore, until the range of displacement verified for the push-
where dk,0 is the displacement of the control point corresponding to over analysis (140 mm) the two curves denote energy equivalence,
a zero value for the load multiplier. This relation can be interpreted and they are matching for a decay of the maximum load factor in
as a linear capacity curve, which is an approximation to the load– the pushover curve of 20% (to 0.2g), which is normally considered
displacement response of the macro-block, and that is related with as the ultimate limit state for safety verification purposes through
the assumptions of infinite compressive strength of the masonry pushover methodologies.
and impossibility of sliding in the mechanism.
However, when assuming a limited compressive strength for 5.3. Seismic safety verification in Chokepukio
the masonry, the difference between the actual relationship and
the linear relationship is significant, because the actual relation- In the following, aspects of the seismic assessment through
ship allows the displacement to increase comparatively quickly kinematic limit analysis are presented considering the particulari-
as the critical point (load factor becomes zero) is approached. In ties of archaeological building remains in Peru. Finally, kinematic
this case, the compressed edge at the wall base is crushed through- limit analysis is applied to the studied wall in the Sector A of
out the mechanism and the hinge moves to the interior of the edge Chokepukio.
section, which generates an asymptotic decreasing of the load fac- For the seismic safety verification it is possible to proceed with
tor against increasing displacement. However, this is not an issue an acceleration-based approach, which is the so-called linear kine-
for the case of thick stone masonry walls, since the compressive matic analysis, or a displacement-based approach (nonlinear

d k1 d k2 d k0
k k k k α
α=0
α0

α1 α=α0 (1-dk /d k0 )
δx δx δx
δy δy δy α2
α0Wi α1Wi α2Wi
δy
dk
Wi Wi Wi ϕ Wi
d k1 d k2 d k0
ϕ O ϕ O ϕ O O
(a) (b)
Fig. 15. Evolution of mechanism: (a) motion sequence and (b) linear capacity curve.

0.30

Pushover
0.25
Load factor (g)

0.20
Kinematic
0.15

0.10

0.05

0.00
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Displacement (mm)

Fig. 16. Comparison of capacity curves obtained from the pushover and kinematic analyses.
108 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

kinematic analysis), as specified in the Italian building code NTC corresponding to the end of the plateau in the elastic acceleration
[50] – Part 7. In the first case, the computed value of the collapse response spectrum; and T1 is the fundamental vibration period of
load multiplier, see Section 5.2, is used for a verification in terms the structure.
of acceleration. On the other hand, the nonlinear kinematic analy-  
Se ðT 1 ÞwðzÞc Z Tp
sis considers the load–displacement response to proceed with a a0 P   min 2:5 ; 2:5  S  g  wðzÞ  c ð13Þ
q q T1
safety verification in terms of displacement.
The capacity parameters need to be considered relatively to a where in the Italian building code, Se(T1) is the elastic spectral
single degree of freedom (SDOF) system, i.e. the capacity curve is acceleration evaluated for the fundamental vibration period of the
to be transformed into a capacity spectrum through consideration structure, T1; w(z) is the normalized first vibration mode of the
of the dynamic properties of the structure, see Fajfar [41]. The for- structure, which can be approximated as the relationship between
mulation to obtain the capacity spectrum is not presented here in the elevation of the hinge and the total height of the structure, z/
full, since it is detailed in the Italian building code NTC [50] – Part H; and c is the modal participation factor, which can be approxi-
7, from analogy with the capacity spectrum method by Fajfar [41]. mated in function of the number of levels of the structure, N, as
For the particular case of a one-body mechanism, the spectral 3N/(2N + 1).
acceleration a⁄ can be computed as the product of the collapse load Beyond the acceleration-based, the Italian building code NTC
multiplier, a0, by the gravity acceleration, g. Then, the spectral dis- [50] – Part 7 also specifies a safety verification in terms of displace-
placement d⁄ is computed according to Eq. (10), multiplying the ment. In this case, an ultimate spectral displacement du⁄ is defined
real displacement of the control point k, dk, by a modal partic- correspondently to the life safety limit state as 0.4d⁄0, mostly based
ipation factor. Thus, a linear relation is established for the capacity on research by Doherty et al. [43]. Then, du⁄ is compared with the
spectrum according to Eq. (11), where parameters a⁄0 and d⁄0 are spectral displacement demand dd⁄, which is computed in function
respectively the collapse activation acceleration and the maximum of a secant period Ts defined for the SDOF system as illustrated in
spectral displacement in the capacity spectrum. Fig. 19a. Also here, cases of global and local mechanisms are con-
Pnþm 2
sidered, respectively requiring the verification of Expressions (14)
 i¼1 W i dx;i and (15). In case of global mechanisms, dd⁄ is obtained by inter-
d ¼ dk Pnþm ð10Þ
dx;k i¼1 W i dx;i cepting the demand spectrum in correspondence with Ts.

   T 2s
a ¼ a0 ½1  d =d0  ð11Þ du P SDe ðT s Þ ¼ Se ðT s Þ ð14Þ
4p 2
According to the Italian building code NTC [50] – Part 7, the
h i2
seismic safety of buildings needs to be verified for both a damage Ts
 T1
limit state and a life safety limit state. In this work, only the life du P SDe ðT 1 ÞwðzÞc rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð15Þ
h i 2
safety limit state, which considers a reserve of strength against col- 1  TT1s þ 0:02 TT1s
lapse due to horizontal loads, is assumed as applicable to archaeo-
logical building remains. Note that the procedure for local seismic where SDe(Ts) is the elastic spectral displacement evaluated for the
assessment as specified in the Italian building code has been secant period Ts, which can be related with the elastic spectral
mostly applied to existing structures, as referenced before. The acceleration, Se(Ts), as presented in Expression (14).
safety verification in terms of acceleration is defined for two differ- Concerning the studied wall in Chokepukio, three collapse
ent cases of mechanism concerning the elevation level of the struc- mechanisms have been considered as presented in Fig. 17, namely
ture where the hinge is formed: at ground level which is denoted two global rocking motions around the wall base, and a partial
as ‘global mechanism’, and at an upper level of the structure which rocking mechanism of the top part of the wall. The first global
is denoted as ‘local mechanism’. The two cases require, respec- mechanism is purely through a rigid-body motion, while the sec-
tively, the verification of Expressions (12) and (13), where the right ond mechanism is constrained by the timber struts that are shoring
part is an equivalent formula to compute the spectral acceleration the wall. The activation of this last mechanism requires the reach-
demand ad⁄ according to the Peruvian seismic design code [52]: ing of the axial load strength of the three wooden struts, which is
ag ðP VR ÞS Z considered as an external force applied to the wall. This force was
a0 P  Sg ð12Þ evaluated by initially assuming a composite flexural behaviour of
q q
the timber rods. However, as the effective slenderness of each
where in the Italian building code, ag(PVR) is the reference peak strut, i.e. the ratio between its effective length (0.9 m) and its
ground acceleration at the site, which is defined as a function of diameter (0.1 m), is 9.0, according to the Peruvian wood design
its probability of exceedance in a given reference period, PVR (usu- code [53] the strut is considered as a short column and thus
ally 50 years); S is the soil amplification factor; and q is the beha- behaves in pure compression. In this condition, the admissible
viour factor (assumed with a value of 2). In the Peruvian seismic compressive stress is taken as the compressive strength of the
design code, Z is the zoning coefficient; Tp is the period eucalypt wood, with a value of 7.8 MPa. Then, considering the full

hinge

timber timber
struts struts

hinge hinge
(a) (b) (c)

Fig. 17. Considered mechanisms: (a) 1st global rocking, (b) 2nd global rocking and (c) partial rocking.
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 109

diameter of the timber struts (cross-section area of 7855 mm2), the microtremor (ambient vibration) measurements at the ground
total compressive axial strength has a value of 3  61.3 kN, i.e. level. After applying the procedure by Nakamura [54] to estimate
around 184 kN. The third mechanism is through a rotation of the the ratio between the Fourier amplitude spectra of the horizontal
top part of the wall around a hinge located in the section with (H) to vertical (V) components of ambient noise vibrations, i.e.
thickness change. the H/V spectrum, the predominant frequency was found in the
The limit analysis presents the advantage of requiring mostly a range 2.0–2.5 Hz (0.4–0.5 s). Thus, an average value of 0.45 s was
geometry input, beyond the specific weight of the material. In this considered for the period Tp. The fundamental vibration period of
case, a CAD-based tridimensional model of the wall corresponding the wall, T1, was adopted as 0.53 s according to the modal analysis
to the calibrated geometry of the FE model was used for the com- of the calibrated FE model.
putations, jointly with the optimized value of the masonry specific The obtained capacity spectra for the three considered mecha-
weight. The kinematics of the three considered mechanisms is nisms are presented in Fig. 19b, where it can be observed that
illustrated in Fig. 18, from the initial static position to the free rota- the second global rocking mechanism is the one that requires the
tion stage. In this study, with only a body considered as involved in highest value of spectral acceleration for its activation (0.27g).
the kinematics, a virtual rotation is assumed around the edge hinge This is mainly due to the positive constraint effect by the timber
to apply the PVW as moment equilibrium (balance of the vertical struts, and even if in this case the ‘pseudo-ductility’ (displacement
and horizontal forces acting around the hinging point). to acceleration ratio) results to be lower than that verified for the
The seismic demand was considered according to the Peruvian unconstrained global rocking mechanism. It can be noted that
seismic design code [52], which establishes a reference peak the timber struts, even if reacting with a small force, have a signifi-
ground acceleration at the site (Zg) of 0.3g, corresponding to 10% cant influence in the safety check. On the other hand, the partial
probability of exceedance in 50 years (return period of 475 years). rocking mechanism presents the lowest activation acceleration
Considering the local ground, composed mainly of graved deposits, and is the most brittle. As shown in Table 5, the global rocking
the soil factor S was considered with a value of 1.2 (medium soil in mechanisms are safe, both in terms of acceleration and displace-
the Peruvian seismic design code). The period Tp, which is related ment checks, while the partial rocking mechanism is unsafe and
to the soil type, was in this case computed from in-situ predictable to occur for a spectral acceleration of 0.17g.

d k0 dk0 dk0
α=0 α=0
α=0
δx1
δy
α0W
δy1 ϕ
δ x2 ϕ W W
δx1 O O
δy Fh δy1 δy
δy1
α 0W δ x1 α 0W
W W W W
ϕ ϕ
ϕ O O O ϕ O
(a) (b) (c)
Fig. 18. Evolution of mechanisms: (a) 1st global rocking, (b) 2nd global rocking and (c) partial rocking.

Sa (g)
a∗ demand spectrum
0.5
a 0∗ Ts =2π d*s/a *s with d*s=0.4d*u Ts3 mechanism 1
a ∗s
Ts2 mechanism 2
a* =a*0 (1-d*/d*0) Ts1
mechanism 3

0.1
d∗
d ∗s d ∗u d ∗0 Sd (m)
0.0 d∗u3 d∗u2 d∗u1 0.5 0.8

(a) (b)
Fig. 19. Capacity spectrum: (a) definition and (b) computation for the considered collapse mechanisms.

Table 5
Results of limit analysis and seismic safety verification.

Mechanism Capacity Ts (s) Demand SFa (g/g) SFd (mm/mm) Check

a0 a0⁄ (g) d⁄u (mm) ad⁄ (g) d⁄d(mm)
1st global 0.24 0.24 300 1.53 0.18 155 1.33 1.93 Safe
2nd global 0.27 0.27 250 1.33 0.18 132 1.50 1.89 Safe
Partial 0.17 0.17 113 1.13 0.20 123 0.85 0.92 Unsafe
110 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111

The partial mechanism results unsafe due to the considered – the geometry, i.e. boundary conditions and section changes in
amplification of the ground acceleration over the elevation of the height and length, plays a fundamental role when calibrating
wall, which implies a higher seismic demand. It is also noted as FE models, particularly to capture the modal shapes, and thus
the check in terms of displacement provides in general higher the need of tools capable of reproducing the geometric details
safety factors comparatively to the force verification, which is is recognized;
mostly associated to the consideration of the dissipation of inertial – the most significant variables to capture the modal response in
horizontal force throughout the rocking motion, like a capacity frequencies were the E-modulus, which ranges from 210 MPa to
reserve. 580 MPa on the wall, and the specific weight (with a calibrated
value close to 25 kN/m3);
6. Conclusions – the modal response is strongly determined by the deformability
of the upper part of the wall, since the E-modulus of the this
This paper presents investigations concerning the layout and part of the wall varies largely during the optimization process.
structural behaviour of archaeological heritage in earthquake- It can also be noted the high sensitivity of the objective function
prone areas, with an application to the Chokepukio to the first mode shape of the wall.
Archaeological Site in Cusco, Peru. The study aims at providing
general guidelines for structural evaluation of archaeological Regarding the analytical investigation, a pushover analysis was
building remains, particularly regarding the seismic assessment. carried out aiming at simulating an overturning mechanism of the
A multi-approach method was used that includes historical and studied wall. The obtained results allowed the design of simplified
typological surveys, and experimental, numerical and analytical kinematic limit analysis taking into consideration the critical sec-
investigations. tions for the activation of collapse mechanisms. Kinematic analysis
The survey of the site layout has large importance since it pro- was then applied as a first approach for the seismic safety veri-
vides a general view of the problem and helps on selecting an ade- fication, after adapting the parameters of the Peruvian seismic
quate study strategy. In Chokepukio, a very particular layout was design code in the safety conditions and considering different
identified with contoured constructions forming an urban environ- types of collapse mechanisms. Based on the analytical com-
ment. Concerning the structural engineering point of view, it was putations, it is concluded that:
possible to observe that the builders of Chokepukio used tech-
niques to ensure the structural stability, namely the use of deep – from the pushover analysis, the out-of-plane mechanism of the
footings in foundations, and buttressing systems. On the other studied wall starts with flexural cracking at the wall base and
hand, the masonry pattern, the geometry and materials are com- thickness transition sections, after that, the cracks propagate
plex and present large variability. in the thickness of the wall base and activate a rocking
In-situ dynamic testing, namely OMA tests, was explored as a mechanism;
tool to support structural identification and assessment of archaeo- – the pushover and kinematic curves are matching at the yield
logical heritage. For a proper test design of archaeological building displacement of the pushover response, the two curves denote
remains, namely regarding the sensors layout, measurement chain energy equivalence, and they are coinciding for a decay of 20%
and data processing techniques to use, it is important to consider of the maximum load factor in the pushover curve;
the peculiarities of these sites, e.g. geometrics and boundary condi- – from the kinematic analysis, the studied wall is safe regarding
tions. From the OMA tests performed at Chokepukio it was possible the global rocking mechanisms, which are activated for a spec-
to accurately identify the relevant mode shapes for the studied sec- tral acceleration around 0.25g. However, the wall results unsafe
tor, and the conclusion is that: respecting to a partial rocking mechanism of the top part of the
wall, which is activated for a spectral acceleration of 0.17g.
– the studied wall presents high flexibility, since at least the first
seven natural frequencies are below 10 Hz. Furthermore, the The methodology applied in this study can be integrated in a
first four frequencies of the wall are in the range 2–5 Hz, respec- broad tool for assessing the vulnerability of the entire archaeologi-
tively varying from translational movements to oscillatory cal site of Chokepukio and for studying other structures with simi-
motions; lar characteristics. From this study, it was also possible to identify
– the higher modes denote some contamination between mea- the need of further investigations aiming at experimentally charac-
sured degrees of freedom, which can be due to the presence terizing the material mechanical parameters of the local masonry.
of a weak plane at the interface of the upper and bottom parts
of the wall, to limitations in the test setup or to effects of
Acknowledgements
dynamic interaction with adjacent structures.

This work was partially funded by the Dirección de Gestión de la

Numerical modelling was next developed which allowed, from
Investigación at PUCP through Grant DGI-70242.3113. An acknowl-
a set of three Finite Element (FE) models initially considered, and
edgement is also given to the Ulises program at PUCP for providing
based on measures of correlation between the experimental and
the scholarships to the masters students involved in the project,
numerical modal results, establishing an efficient model for further
and to the Ministry of Culture of Peru for facilitating the access
analytical purposes. This FE model was calibrated through sensitiv-
to the archaeological site. Finally, the authors gratefully acknowl-
ity analysis and optimization routines, allowing the definition of
edge to archaeologist Carmen Farfan for the invaluable assistance
optimal values for the elastic mechanical properties of the masonry
while carrying out the field investigations at Chokepukio.
(E-modulus and specific weight). The final obtained model pre-
sents high representativeness of the actual structural condition of
the studied wall, concerning the approximation of the experimen- References
tal modal response, i.e. maximum difference in modal frequencies
is less than 8% while the lowest MAC value is 0.86. Furthermore, [1] Lourenço PB. Recommendations for restoration of ancient buildings and the
from the numerical modelling and calibration process of the FE survival of a masonry chimney. Constr Build Mater 2006;20(4):239–51.
[2] Roca P, Cervera M, Gariup G, Pelà L. Structural analysis of masonry historical
model of the wall in Chokepukio, the following conclusions are constructions: classical and advanced approaches. Arch Comput Methods Eng
pointed: 2010;17(3):299–325.
 R. Aguilar et al. / Engineering Structures 95 (2015) 94–111 111

[3] Lourenço PB. Computations on historic masonry structures. Prog Struct Mat [28] TNO. DIANA: DIsplacement method ANAlyser, version 9.4.4. Delft: TNO DIANA
Eng 2002;4(3):301–19. BV; 2013.
[4] Chellini G, Nardini L, Pucci B, Salvatore W, Tognaccini R. Evaluation of seismic [29] Brignola A, Frumento S, Lagomarsino S, Podesta E. Identification of shear
vulnerability of Santa Maria del Mar in Barcelona by an integrated approach parameters of masonry panels through the in-situ diagonal compression test.
based on terrestrial laser scanner and finite element modelling. Int J Archit Int J Archit Heritage 2008;3(1):52–73.
Heritage 2014;8(6):795–819. [30] Sánchez-Acosta M. Characterization and use of E. grandis wood. In:
[5] de Matteis G, Mazzolani FM. The Fossanova Church: seismic vulnerability Proceedings of the ‘V Jornadas Forestales de Entre Ríos’, Concordia; 1990. p.
assessment by numeric and physical testing. Int J Archit Heritage 83–93 [in Spanish].
2010;4(3):222–45. [31] Allemang JR. The modal assurance criterion – twenty years of use and abuse.
[6] Ceroni F, Pecce M, Voto S, Manfredi G. Historical, architectural, and structural Sound Vib 2003;37(8):14–23.
assessment of the bell tower of Santa Maria del Carmine. Int J Archit Heritage [32] Ramos LF, Aguilar R, Lourenço PB, Moreira S. Dynamic structural health
2009;3(3):169–94. monitoring of Saint Torcato church. Mech Syst Signal Process 2013;35(1–2):
[7] Casarin F, Modena C. Seismic assessment of complex historical buildings: 1–15.
application to Reggio Emilia Cathedral, Italy. Int J Archit Heritage 2008;2(3): [33] Douglas BM, Reid WH. Dynamic tests and system identification of bridges. J
304–27. Struct Div ASCE 1982;108(ST10):2295–312.
[8] Lourenço PB, Trujillo A, Mendes N, Ramos LF. Seismic performance of the St. [34] Brownjohn JMW, Xia PQ, Hao H, Xia Y. Civil structure condition assessment by
George of the Latins church: lessons learned from studying masonry ruins. Eng FE model updating: methodology and case studies. Finite Elem Anal Des
Struct 2012;40:501–18. 2001;37(10):761–75.
[9] Chácara C, Zvietcovich F, Briceño C, Marques R, Perucchio R, Castañeda B et al. [35] Chellini G, De Roeck G, Nardini L, Salvatore W. Damage analysis of a steel–
On-site investigation and numerical analysis for structural assessment of the concrete composite frame by finite element model updating. J Constr Steel Res
archaeological complex of Huaca de la Luna. In: Proceedings of the SAHC2014: 2010;66(3):398–411.
9th international conference on structural analysis of historical constructions. [36] MathWorks. MatLab user manual, release 7.2. Natick MA: The MathWorks;
Mexico City; 2014 [CD-ROM]. 2011.
[10] Marques R, Pereira JM, Lourenço PB, Parker W, Uno M. Study of the seismic [37] García JM. Microzonification of the Cusco city. In: Proceedings of the ‘1er
behavior of the ‘‘Old Municipal Chambers’’ building in Christchurch, New Simposium Nacional de Prevención y Mitigación de Desastres Naturales, Lima;
Zealand. J Earthquake Eng 2013;17(3):350–77. 1987. p. 275–82 [in Spanish].
[11] Oliveira DV, Grecchi G, McCall A, Noh J, Speer E, Tohidi M. Seismic analysis of [38] Pinho FFS, Baião MFC, Lúcio VJG, Faria P. Experimental research on rubble
the Roman Temple of Évora, Portugal. In: Proceedings of the 14th world stone masonry walls. In: Proceedings of the 1st historical mortars conference:
conference on earthquake engineering. Beijing; 2008 [CD-ROM]. characterization, diagnosis, conservation, repair and compatibility. Lisbon;
[12] McEwan G, Gibaja A, Chatfield M. Monumental architecture of late 2008 [CD-ROM].
intermediate period of Cuzco: continuities of ritual reciprocity and statecraft [39] Kirikov BA. History of earthquake-resistant construction: from antiquity to our
between the middle and late horizons. Boletín de Arqueología PUCP times. Madrid: Instituto de Ciencias de la Construcción Eduardo Torroja; 1992.
2005;9:257–80 [in Spanish]. [40] Stiros SC. Archaeological evidence of antiseismic constructions in antiquity.
[13] Magalhães F, Cunha A. Explaining operational modal analysis with data from Ann Geophys 1995;38(5–6):725–36.
an arch bridge. Mech Syst Signal Process 2011;25(5):1431–50. [41] Fajfar P. Capacity spectrum method based on inelastic demand spectra.
[14] Rainieri C, Fabbrocino G. Operational modal analysis of civil engineering Earthquake Eng Struct Dynam 1999;28(9):979–93.
structures: an introduction and guide for applications. Berlin: Springer; 2014. [42] Lourenço PB, Mendes N, Marques R. Earthquake design and assessment of
[15] Rainieri C, Fabbrocino G, Verderame GM. Non-destructive characterization and masonry structures: review and applications. In: Topping BHV, Costa Neves LF,
dynamic identification of a modern heritage building for serviceability seismic Barros RC, editors. Trends in civil and structural engineering
analyses. NDT&E Int 2013;60:17–31. computing. Stirlingshire: Saxe-Coburg; 2009. p. 77–101.
[16] Caprili S, Nardini L, Salvatore W. Evaluation of seismic vulnerability of a [43] Doherty K, Griffith MC, Lam N, Wilson J. Displacement-based seismic analysis
complex RC existing building by linear and nonlinear modeling approaches. for out-of-plane bending of unreinforced masonry walls. Earthquake Eng
Bull Earthq Eng 2012;10(3):913–54. Struct Dynam 2002;36(6):801–21.
[17] Alaboz M, Ramos LF, Aguilar R. Dynamic identification and modal updating of S. [44] Casolo S, Uva G. Nonlinear analysis of out-of-plane masonry façades: full
Torcato church. In: Proceedings of the SAHC2010: 7th international conference dynamic versus pushover methods by rigid body and spring model.
on structural analysis of historical constructions. Shanghai; 2010. p. 110–16. Earthquake Eng Struct Dynam 2013;42:499–521.
[18] Gentile C, Saisi A. Ambient vibration testing of historic masonry towers for [45] Pelà L, Aprile A, Benedetti A. Seismic assessment of masonry arch bridges. Eng
structural identification and damage assessment. Constr Build Mater Struct 2009;31:1777–88.
2007;21(6):1311–21. [46] Tomaževič M. Earthquake-resistant design of masonry buildings. London:
[19] Brincker R, Ventura C, Andersen P. Why output-only modal testing is a Imperial College Press; 1999.
desirable tool for a wide range of practical applications. In: Proceedings of [47] Lourenço PB. Recent advances in masonry modelling: micromodelling and
the international modal analysis conference IMAC XXI. Kissimmee; 2003 homogenisation. In: Galvanetto U, Ferri Aliabadi MH, editors. Multiscale
[CD-ROM]. modeling in solid mechanics: computational approaches. London: Imperial
[20] Ramos LF. Damage identification on masonry structures based on vibration College Press; 2010. p. 251–94.
signatures. PhD thesis. Guimarães: University of Minho; 2007. [48] Heyman J. The stone skeleton: structural engineering of masonry
[21] Welch PD. The use of the fast Fourier transform for the estimation of power architecture. Cambridge: Cambridge University Press; 1997.
spectra: a method based on time averaging over short modified periodograms. [49] DPCM. Directive for assessment and mitigation of seismic risk of cultural
IEEE Trans Audio Electroacoust 1967;15(2):70–3. heritage, dated of 09-02-2011. Rome: Presidente del Consiglio dei Ministri;
[22] Ewins DJ. Modal testing: theory, practice, and application. 2nd 2011 [in Italian].
ed. Baldock: Research Studies Press; 2000. [50] NTC. Norme tecniche per le costruzioni (‘Italian building code’), Ministerial
[23] Felber A. Development of a hybrid bridge evaluation system. PhD thesis. Decree dated of 14-01-2008. Rome: Ministero delle Infrastrutture e dei
Vancouver: University of British Columbia; 1993. Trasporti; 2008 [in Italian].
[24] Peeters B, De Roeck G. Reference-based stochastic subspace identification for [51] CNTC. Circular of instructions for application of the building code of 14-01-
output-only modal analysis. Mech Syst Signal Process 1999;13(6):855–78. 2008. Rome: Consiglio Superiore dei Lavori Pubblici; 2008 [in Italian].
[25] SVS. ARTeMIS structural vibration solutions extractor pro user manual, release [52] NTE.030. Reglamento de edificaciones del Perú, Norma técnica E.030: Diseño
4.5. Aalborg: Structural Vibration Solutions; 2013. sismorresistente (‘Peruvian seismic design code’). Lima: Ministerio de
[26] Elmenshawia A, Sorour M, Mufti A, Jaeger LG, Shrive N. Damping mechanisms Vivienda, Construcción y Saneamiento; 2006 [in Spanish].
and damping ratios in vibrating unreinforced stone masonry. Eng Struct [53] NTE.010: Reglamento de edificaciones del Perú, Norma técnica E.010: Madera
2010;32:3269–78. (‘Peruvian wood design code’). Lima: Ministerio de Vivienda, Construcción y
[27] Rainieri C, Fabbrocino G, Manfredi G, Dolce M. Robust output-only modal Saneamiento; 2006 [in Spanish].
identification and monitoring of buildings in the presence of dynamic [54] Nakamura Y. A method for dynamic characteristics estimation of subsurface
interactions for rapid post-earthquake emergency management. Eng Struct using microtremor on the ground surface. Quart Report RTRI 1989;30(1):
2012;34:436–46. 25–33.