You are on page 1of 9

Received: 16 July 2015 Revised: 10 February 2016 Accepted: 27 September 2016

DOI 10.1002/stc.1959

RESEARCH ARTICLE

Real‐time structural monitoring of Building 350 at Del Valle


University
Lisandro Arturo Jiménez‐Roa | Johannio Marulanda‐Casas | Alejandro Cruz‐Escobar

School of Civil Engineering and Geomatics, Del


Valle University, Santiago de Cali, Colombia Summary
Correspondence Design errors, misuse, and natural or accidental events can cause small and cumula-
Lisandro Arturo Jiménez‐Roa, School of Civil tive damages or large and sudden damages that can deteriorate structural systems or
Engineering and Geomatics, Del Valle University, cause their collapse. Structural monitoring consists in the permanent real‐time track-
Santiago de Cali, Colombia.
Email: lisandro.jimenez@correounivalle.edu.co
ing of information about the condition and performance of a structure. This paper
presents the development, validation, and implementation of the structural monitor-
ing system for building 350 at Del Valle University, located in Santiago de Cali,
Colombia. The system uses the SSI and natural excitation technique with the
eigensystem realization algorithm methodologies for modal identification and a spe-
cialized event‐recognition algorithm based on the standard deviation analysis of
vibrations. The dynamic behavior monitoring of the building has been conducted
in real‐time since January 2012, and six seismic events have been successfully
recognized. The computational tool was validated using a finite element model with
which the effect of variation in mass was evaluated.

K E Y WO R D S

identification of seismic events, modal identification, seismic engineering, structural


monitoring

1 | IN T RO D U C T IO N use these types of technologies is increased by factors such


as high level of human–structure interaction, the importance
Structural health monitoring (SHM) technologies have been of the structure (e.g., hospitals, bridges, and dams), active
widely used to improve the knowledge about the behavior seismic hazard zones, and historically significant struc-
of structures that are subjected to different load types or envi- tures.[5] Consequently, the need to develop specialized hard-
ronmental, chemical, and mechanical phenomena. Therefore, ware and software for SHM is justified.
scientists and engineers around the world are developing Non‐invasive methodologies that allow diagnosing the
increasingly specialized methodologies, technologies, and integral safety of a civil structure are being developed at
techniques for analyzing the behavior of structures.[1] The Building 350 of the School of Civil Engineering and
most accepted SHM methodologies are based on the identifi- Geomatics at Del Valle University. To do so, a pilot project
cation of dynamic parameters, such as natural frequencies, that is focused on monitoring the building's structural health
damping ratios, and modal shapes, based on ambient vibra- has been conducted. Specialized monitoring hardware was
tions, which allows to formulate a dynamic model of the installed during the project, and a computational tool capable
structure. In conjunction with model updating techniques of using ambient vibrations to identify and monitor the
and algorithms capable of recognizing abnormal changes in dynamic characteristics of the building was developed using
the dynamic properties of a structure,[2] these methodologies the stochastic subspace identification (SSI) and natural exci-
are potential tools for the identification, localization, and tation technique with the eigensystem realization algorithm
quantification of damage in civil structures.[3,4] The need to (NExT‐ERA) modal identification algorithms.

Struct Control Health Monit 2016; 1–9 wileyonlinelibrary.com/journal/stc Copyright © 2016 John Wiley & Sons, Ltd. 1
2 JIMÉNEZ‐ROA ET AL.

2 | MODA L I DE NT IF ICAT IO N decreases with the same characteristics of an impulse


A L G OR I T H M S response function from the same system, therefore has proven
that it can be used as free vibration response.
Several methodologies exist for the identification of the The ERA uses minimal realization principles to obtain a
dynamic behavior of civil structures. The monitoring sys- state–space representation of the structure. Developed by
tem of Building 350 was implemented with two algorithms: Juang and Pappa,[17] ERA has proven to be an effective
the SSI method,[6] which is currently considered the most method for the modal identification of civil structures.[18]
powerful technique for modal identification in time The free vibration response data from NExT is used to build
domain,[7] and the natural excitation technique (NExT) in two Hankel matrices, which are used to calculate the state and
conjunction with the eigensystem realization algorithm output matrices of the state–space representation of the sys-
(ERA).[8] tem using singular value decomposition. The eigenvalues of
the state matrix contain information about the natural fre-
quencies and damping ratios of the system, and the vibration
2.1 | Stochastic subspace identification
modes are the result of the eigenvectors that are used on the
Subspace identification algorithms are based on three basic output matrix.
assumptions: (a) The system behaves linearly; (b) the behav-
ior of the system does not vary over time; and (c) the input
excitation is a stochastic process with Gaussian white noise 2.3 | Modal assurance criterion
and is not correlated with the response of the system.[9] The modal assurance criterion is a statistical indicator that
The algorithms estimate the dynamic properties of the sys- expresses the degree of coherence between modal shapes. It
tem (i.e., natural frequencies, modal shapes, and damping is based on a least squares linear regression analysis; it is
ratios) from a state–space representation. Van Overschee more sensitive to large differences and is relatively sensitive
and De Moor[6] provide the theoretical framework for these to small ones. Modal assurance criterion ranges from 0 to
subspace identification algorithms. If the input excitation of 1, where 1 indicates totally consistent modal shapes, and
the system cannot be measured and is stochastic, then the 0 indicates that the modes are different.[19]
algorithm is called SSI. The SSI method has proven to be a
simple and reliable tool for modal identification[10] and has
been used successfully in industrial applications[11] and civil 3 | B U I L D I NG 3 5 0
infrastructure monitoring,[12] as well as in other applications.
The operations of SSI can be summarized in the follow- Building 350 at Del Valle University is located on the
ing steps: create a block Hankel matrix with the system's Meléndez Campus in Cali, Colombia (N: 3°22′27.527″,
vibration response data; divide the matrix into two compo- W: 76°31′55.506″). The structure contains offices, class-
nents: “past” and “future”; calculate the orthogonal projec- rooms, and laboratories for teaching, research, and consult-
tion of the future component over the past component; ing. The Building 350 is a two‐story reticular celled structure
apply singular value decomposition to the projection and, (Figure 1) with an additional roof slab (Figure 2b). The
using the matrix factorizations, calculate an observability distance between the column axes is 7.2 m. The three slabs
matrix that, in a state–space representation, is a function are 40 cm thick and are reinforced with 10‐cm‐wide joists
of the state matrix and the output matrix. The eigenvalues oriented in two directions. All the columns have a 40‐cm‐
of the state matrix contains information about the natural square cross section. The first slab is at a height of 3.2 m;
frequencies and damping ratios of the system, and the the second slab is at 6.4 m; and the third slab is at 8 m.
eigenvectors that operate on the output matrix provide the The instrumentation system is composed of a REFTEK
vibration modes. 130SMHR triaxial station, which is located at the base of
the building, and three REFTEK 131B‐01/1 seismic acceler-
ometers on the second slab (N + 6.4 m; Figure 2a). All the
2.2 | Natural excitation technique with the eigensystem
accelerometers have the capacity to measure up to ±4 g at a
realization algorithm noise level of 2 μg2/Hz, and the acquisition frequency is
The NExT was first used for modal testing of a wind turbine configurable from 1 to 1000 Hz. In addition, the system
under wind loading excitation,[8,13] and it has been used in has a GPS unit for time synchronization. The system is
numerous studies.[14,15] NExT is usually used in conjunction currently configured to continuously acquire data at a sam-
with the ERA.[16] NExT requires long response records to pling rate of 100 Hz.
ensure the assumption of relatively stationary processes and A detailed dynamic characterization to obtain a reference
to guarantee that the excitation is not correlated with the identification of the structure, as recommended by
response of the system. NExT proves that the cross‐correla- Nayeri et al.,[20] was performed on Building 350 as reported
tion function between two response records of the system's in.[21] The results of the frequencies identification are shown
vibration response corresponds to a sinusoidal function that in Table 1.
JIMÉNEZ‐ROA ET AL. 3

FIGURE 1 Distribution of sensors in Building 350.[27]

TABLE 1 Dynamic characteristics of Building 350, using stochastic sub-


space identification[21]

Natural frequency (Hz) Damping ratio (%)


Average Standard deviation Average Standard deviation

4.16 0.18 6.13 0.88


4.74 0.14 5.48 2.29
5.13 0.11 3.11 0.48

sensors, a digitalization and synchronization instrument,


and data storage in a server. The algorithms for SM run in
real‐time in the computer server. Once the data are stored,
the following protocol is used:

• The instrumentation system records the acceleration in


files with PASSCAL format, which are changed to DAT
format, to be processed in MATLAB®.
• The data analysis is performed, and it includes the follow-
FIGURE 2 Floors of Building 350. (a) Floors at N + 3.2 m and N + 6.4 m. ing: modal identification using SSI and NExT‐ERA,
(b) Floor at N + 8.0 m.[27] where operational frequencies, modal shapes, and
damping ratios are obtained by each methodology, and
also other quantities such as peak acceleration, sample
ratio, date and hour recorded, and sensor.
4 | C O M P U TATI O NA L TO O L F O R
ST RU C TU R A L M O N I T O R I NG • The previous information is stored in a database. With the
aim to make easier and faster the loading of information,
The computational tool for structural monitoring (SM) was the database is organized in a structure array. The database
developed in the MATLAB® programing language is saved in folders according with the date of the record.
(Figure 3). The computational tool uses ambient vibrations, • The main interface is updated with the last information
such as those produced by the users of the building, the oper- processed.
ation of mechanical devices, the traffic of vehicles near the
structure, micro‐tremors, and the wind. Using these excita- An important characteristic of the SM computational tool
tion sources with digital signal processing techniques and is to be an open source code, which makes possible to be
modal identification methodologies, it is possible to estimate continuously modified and improved. Figure 3 shows the
the dynamic characteristics of the structure. main interface of the computational tool, which is composed
The basic idea of the proposed protocol is to work in par- of a series of panels that display the results, including the
allel with a dynamic instrumentation system composed of following:
4 JIMÉNEZ‐ROA ET AL.

FIGURE 3 Main interface of the SM computational tool.[27]

1. General information about the most recently processed 4.1 | Seismic event recognition algorithm
recording. Variations in acceleration can be evaluated in real time
2. Visualization of the modal shapes. directly in the acceleration record at any given time and
3. Natural frequencies and damping ratios variation. can be used to identify sudden changes in the signal.
4. Natural frequencies and damping ratios histogram. Figure 4a shows a representative record of accelerations
from ambient vibrations, and Figure 4b shows a record that
5. Peak acceleration variation.
contains information from a seismic event; both records
6. Time history of accelerations registered by each sensor. correspond to sensor 5 (Figure 2). The algorithm deter-
mines the behavior of the acceleration signals using the
The dynamic characteristics of the structure are identified standard deviation. Figure 5 shows the peak values of the
with the SSI and NExT‐ERA algorithms. Once the computa- acceleration signal for both cases shown in Figure 4. Using
tional tool processes the acceleration signals, data are stored those data, maximums are determined throughout the
in a database for future analyses. The parameters of the acceleration records; the vector of maximums is divided
identification algorithms used in the monitoring tool and into segments (the criteria to define the size of the segment
the ones defined for the sensitivity and stability analysis are is that the sample may have enough information to repre-
as follows: for the SSI algorithm, blocks: 40; order: 8. The sent the variation of acceleration through time; in this par-
parameters for the NExT‐ERA algorithm are as follows: ticular case, it has defined segments second by second,
reference: channel 4; order: 8; rows: 150; columns: 30; and each of 100 data, due the sample ratio was 100 Hz), and
overlap: 75%. the standard deviation is calculated between points of each
In addition, the computational tool includes a seismic segment. The algorithm filters the standard deviation
event recognition algorithm. When a signal is cataloged results to differentiate those that fall within a range greater
as a seismic event, relevant information is sent immedi- than or equal to 37% (value obtained from a sensitivity
ately via e‐mail. The email report includes the following: analysis for a representative period) of the maximum repe-
event record files; a document with information about tition value determined in a histogram of the standard devi-
the event such as time and date, sample ratio, peak values, ation results. In this way, the information in the signal
duration; and acceleration versus time plots for each corresponding to normal vibration is eliminated, living
sensor. Data will be available in http://eicg.univalle.edu. the information that may correspond to an event. The dura-
co/G‐7. tion of the event is then determined for the data that passed
JIMÉNEZ‐ROA ET AL. 5

FIGURE 5 Peak values of the representative acceleration records. (a)


Ambient vibration. (b) Seismic event.[27]

FIGURE 4 Representative records of accelerations. (a) Ambient vibration.


(b) Seismic event.[27]

the standard deviation filter. If data extend over more than


2 s (value also obtained from the sensitivity analysis for a
representative period), the algorithm identifies the group of
values as corresponding to an event; otherwise, it will
eliminate the data. Once the computational tool identifies
an event, it establishes the initial time (ti) and the end time
(te) of the event (Figure 6). Six seismic events were auto-
matically recognized using this methodology (Table 2).

4.2 | Mass variation


FIGURE 6 Results of the event identification algorithm for a seismic
The performance of the computational tool was evaluated event.[27]
with numerical tests using a finite element model of Building
350 developed in SAP2000®.[22] The model includes the of the first three modes of vibration obtained from the
geometric characteristics of the building's structural ele- SAP2000® model are as follows: Mode 1: 4.07 Hz; Mode
ments, including the cross sections of the columns, beams, 2: 5.35 Hz; and Mode 3: 6.16 Hz.
joists, tie beams, capitals, and nonstructural elements such The evaluation protocol consisted of using the SM com-
as masonry walls. The model takes into account the proper- putational tool to determine continuously the dynamic char-
ties of the materials of each of these elements, and a detailed acteristics of the model. An algorithm was developed in
analysis of live loads from the equipment and furniture was MATLAB® using the Open Application Programming Inter-
performed. The model takes the foundation–ground interac- face of SAP2000®. This allowed experimental acceleration
tion into account by modeling the footings with translational signals (obtained with the acquisition system) to be input
and rotational spring‐type elements. The natural frequencies automatically as excitations to the base of the model and
6 JIMÉNEZ‐ROA ET AL.

TABLE 2 Seismic events identified by the Building 350 monitoring system[28]


Location Date (UTC) Time (UTC) Magnitude Distance to Cali (km)

Puerto Tejada – Cauca 12‐02‐2012 12:20:35 3.6 Ml 17.73


Sípi – Chocó 01‐03‐2012 01:41:12 4.5 Ml 94
Guacarí – Cauca Valley 27‐06‐2012 03:47:42 4.5 Ml 52.2
La Vega – Cauca 30‐09‐2012 16:31:34 7.1 Mw 150
Nuquí – Chocó 13‐08‐2013 15:43:04 6.5 Ml 266.13
Dagua – Cauca Valley 27‐08‐2013 23:22:05 4.2 Ml 32.47

UTC, Coordinated Universal Time.

the responses of the degrees of freedom associated with sen- coherence when the mass increases, the frequencies decrease
sors 4, 5, and 6 (numerical acceleration signals) to be calcu- and vice versa. Figure 8 shows the results obtained from the
lated. Finally, as with the experimental records obtained identification with the model, using SSI and NExT‐ERA,
from the instrumentation system, the SM computational tool the direct results from SAP2000®, and the variation of the
continuously estimated the dynamic parameters of the model. mass through the day, assumed as a Gaussian distribution,
The protocol used 144 experimental acceleration records simulating the variation of mass through day, reaching the
with 10 min of duration each, recorded at a sampling frequency higher value at midday, as expected. The computational tool
of 100 Hz. Because the dynamic behavior of a structure is sus- is sensitive to mass variation, which is important to differen-
ceptible to mass changes, the mass of the model was varied tiate it from variations due damage, and avoid false alarms.
from 0 to 0.3 t/m2 and was modeled as a distributed load on
the second floor of the building (simulating the minimum and
maximum load expected in the second slab according to the 5 | M O N IT OR I NG R E S U LTS
structural design). From the simulation, the acceleration
corresponding to sensors 4, 5, and 6 was obtained with the One of the most important aspects to consider in SM is the con-
same characteristics as the excitation input signal (i.e., the same tinuous observation of the variations in dynamic parameters,
sampling frequency and number of points). As noise was not which could indicate the presence of structural damage or
included in the numerical response signals, the modal identifi- changes in the structural configuration. Generally, these
cation was cleaner than when the experimental records were variations are grouped into two types: (a) variations caused
used. High precision and accuracy in the identified natural by damage and (b) variations caused by external factors, such
frequencies and modes of vibration was obtained, as shown in as temperature, mass, and humidity. The latter group of factors
Figure 7. The results show that the modal identification for leads to changes in the dynamic properties that must be differ-
the finite element model using the SM computational tool entiated to correctly estimate the effects of the damage.[20,23,24]
was satisfactory when the results are compared with the Figure 9 shows the variation of the second natural fre-
obtained directly from SAP2000® (Figure 8). quency of Building 350, caused by external factors. It shows
The test results indicate that the three frequencies are sen- the variation over a period of 2 days and 12 hr. The dark line
sitive to variations in mass; the results for the first mode are represents the mean value of the cloud of points showed
shown, as an example, in Figure 8, where correlation shown behind. The frequency varies from 4.6 to 4.8 Hz. Despite
the high dispersion, the results illustrate the tendency for
decreases in the frequency during the day due to the increase
of mass and temperature and for increases in the frequency at
night due the lower mass and temperature.
Several possible causes make difficult to identify a trend of
the natural frequencies caused by external factors. Another
factor that increases the difficulty of the identification
process is the high stiffness of the building and the low number
of sensors, which causes a high noise/signal ratio and reduces
the precision of the results. Also, the variation of the modal
frequencies due changes in temperature could be within the
range of the standard deviation characteristic of the methods.

5.1 | Analysis of seismic events


FIGURE 7 Frequency histogram obtained with structural monitoring (SM) One of the fundamental objectives of an SM system is to
for the finite element model.[27] determine the presence and magnitude of structural damage
JIMÉNEZ‐ROA ET AL. 7

FIGURE 8 Frequency variation results of


SAP2000, stochastic subspace identification
(SSI) and natural excitation technique with the
eigensystem realization algorithm (NExT‐ERA)
due the mass variation for the first mode of the
analytical model.[27]

FIGURE 9 Variation of the second natural frequency of Building 350.[27]

caused by natural or artificial events[25] and to assess percentage of the critical damping of each mode as a function
whether the structure is in a suitable condition to continue of time, is shown at the bottom. A dense gray stripe can be
in service. In the case of seismic events, the results of SM seen at approximately 5 Hz, and several black stripes are
also help to improve the design and construction processes located in the central part (during the seismic event), indicat-
and to complement the design standards.[26] Six seismic ing a greater density in the power spectrum and, thus, the
events were recorded at Building 350 during 2012 and characteristic frequencies. The dense gray stripe continues
2013 (Table 2), and they were used to perform a thorough with the same magnitude of frequency as before the occur-
study of the conditions before, during and after the events, rence of the seismic event, which confirms that the events
using the SSI and NExT‐ERA, with the goal of identifying caused no structural damage to the building or caused minor
the presence of new characteristic frequencies or variations damage that modifying its natural frequencies within the
of the current ones. The study was complemented with fre- range of standard deviation characteristic of the implemented
quency spectrograms, which help to visualize the frequency method. In addition to the evidence of no significant varia-
content of acceleration signals as a function of time. Every tion in the characteristic frequencies, a visual inspection pro-
spectrogram was calculated with different parameters set to cess confirmed that no damage occurred to the structural and
visually achieve the best contrast that allows the visualiza- nonstructural elements.
tion of the frequency content.
Figure 10 shows the frequency spectrograms correspond-
ing to sensor 5 for each seismic event. The upper side of each 6 | C O NC LU S I ON S
plot shows the acceleration signal versus time; the power
spectral density is shown on the left; the frequency spectro- The SM computational tool satisfactorily uses the SSI and
gram is shown in the middle; and the damping ratio, as NExT‐ERA modal identification methodologies. It estimates
8 JIMÉNEZ‐ROA ET AL.

FIGURE 10 Spectrograms of frequencies, accelerations, power density functions, and damping percentages for the seismic records from (a) Puerto Tejada, (b)
Sipí, (c) Vijes, (d) La Vega, (e) Nuquí, and (f) Dagua.[27]
JIMÉNEZ‐ROA ET AL. 9

the dynamic parameters of the structure, including the opera- [13] G. James, T. G. Carne, J. P. Lauffer, Modal Analysis‐the International Jour-
tional frequencies, damping ratios, and modal shapes, in real‐ nal of Analytical and Experimental Modal Analysis 1995, 10, 260.

time, using ambient vibrations and seismic events responses. [14] J. M. Caicedo, J. Marulanda, Struct. Control Health Monit. 2011, 18, 416.

The results of mass variation in the finite element model of [15] J. M. Caicedo, Exp. Tech. 2011, 35, 52.

Building 350 were used to evaluate the computational tool [16] H. Moncayo, J. Marulanda, P. Thomson, J. Aerosp. Eng. 2010, 23, 99.

by monitoring the dynamic parameters. Based on the modal [17] J.‐N. Juang, R. S. Pappa, J. guid control dynam. 1985, 8, 620.

identification before and after each of six seismic events [18] J. M. Caicedo, S. J. Dyke, E. A. Johnson, J. Eng. Mech. 2004, 130, 49.

recorded at Building 350, and the visual inspection of the [19] M. Pastor, M. Binda, T. Harčarik, Procedia Eng. 2012, 48, 543.

structure, it is stated that the events may have caused minor [20] R. D. Nayeri, S. F. Masri, R. G. Ghanem. A novel approach for the structural
identification and monitoring of a full‐scale 17‐story building based on
damage only modifying its operational frequencies within ambient vibration measurements, vol. 17, p. 025006, 2008.
the range of the standard deviation characteristic of the [21] J. M. Franco, J. Marulanda, J. M. Caicedo. Modal identification of a full‐
implemented methods. scale building under seismic excitation using the fast mode identification
technique. Experimental Techniques, 2015.
REFERENCES [22] S. Valencia, A. Cruz, P. Thomson, Actualización en línea de modelos en
[1] J. Ou, H. Li, Struc. Health Monit. 2010, 9, 219. elementos finitos: caso Edificio 350 ‐ Universidad del Valle, Facultad de
Ingeniería, Escuela de Ingeniería Civil y Geomática, Universidad del Valle,
[2] F. Magalhães, Á. Cunha, E. Caetano, Structural health monitoring based on Santiago de Cali, Colombia, 2014.
automated operational modal analysis: Application to an arch bridge, in
Earthquakes and health monitoring of civil structures, Springer, 2013, 241. [23] Y. Xia, B. Chen, S. Weng, Y.‐Q. Ni, Y.‐L. Xu, J. Civ. Struc. Health Monit.
2012, 2, 29.
[3] R. Jafarkhani, S. F. Masri, Comput.‐Aided Civ. Infrastruct. Eng. 2011, 26,
207. [24] A. J. Croxford, J. Moll, P. D. Wilcox, J. E. Michaels, Ultrasonics 2010, 50,
517.
[4] S.‐H. Chao, C.‐H. Loh, M.‐H. Tseng, J. Intell. Mater. Syst. Struct. 2014, 25,
1097. [25] W. Zhou, H. Li, C. Mao, L. Mevel, J. Ou, Adv. Struc. Eng. 2013, 16, 605.

[5] C. Rainieri, G. Fabbrocino, E. Cosenza, Struc. Health Monit. 2011, 10, 291. [26] H. S. Ulusoy, M. Q. Feng, P. J. Fanning, Earthquake Eng. Struct. Dyn. 2011,
40, 661.
[6] P. Van Overschee, B. De Moor, Subspace identification for linear systems:
Theory, implementation, applications, status: published, 1996. [27] L. Jiménez, Desarrollo e implementación de un sistema de monitoreo en
tiempo real del comportamiento dinámico del Edificio 350 de la Universidad
[7] R. Astroza, H. Ebrahimian, J. P. Conte, T. C. Hutchinson, J. I. Restrepo, del Valle, Facultad de Ingeniería, Escuela de Ingeniería Civil y Geomática,
Evolution of dynamic properties of a 5‐story RC building during con- Universidad del Valle, Santiago de Cali ‐ Colombia, 2014.
struction, in Topics in dynamics of civil structures, volume 4, Springer,
New York, 2013, 163. [28] Servicio Geológico Colombiano [online]. Available (Noviembre de 2013)
2013: http://seisan.ingeominas.gov.co/RSNC/index.php/consultas
[8] G. H. James III, T. G. Carne, J. P. Lauffer, NASA STI/Recon Technical Report
N 1993, 93, 28603.
[9] L. Hermans, H. van der Auweraer, Mech. Syst. Sig. Process. 1999, 13, 193. How to cite this article: Jiménez‐Roa, L. A.,
[10] B. Peeters, G. De Roeck, J. Dyn. Syst. Meas. Contr. 2001, 123, 659. Marulanda‐Casas, J., and Cruz‐Escobar, A. (2016),
[11] P. Andersen, Estimation of modal parameters and their uncertainties. Real‐time structural monitoring of Building 350 at
Dept. of Building Technology and Structural Engineering, Aalborg
University, 1998.
Del Valle University, Struct Control Health Monit,
[12] D. F. Giraldo, W. Song, S. J. Dyke, J. M. Caicedo, J. Eng. Mech. 2009, 135, 759.
doi: 10.1002/stc.1959

You might also like