PROCEEDINGS of the 22 nd International Congress on Acoustics Structural Health Monitoring and Sensor Networks: Paper ICA2016-857 Identification of dynamic response parameters of a concrete building during recent earthquakes by using structural vibration monitoring Pablo Alcaíno (a), Evelyn Álvarez (b), Tomás Orrego (c) (a) Pontificia Universidad Católica de Valparaíso, Chile, pablo.alcaino@pucv.cl (b) Pontificia Universidad Católica de Valparaíso, Chile, evelyn.alvarez.sotomayor@gmail.com (c) Pontificia Universidad Católica de Valparaíso, Chile, tomas.orrego.ferrada@gmail.com Abstract The main building of Faculty of Engineering at Pontificia Universidad Católica de Valparaíso (Chile) has been instrumented with a vibration sensor network that consists of three synchronized tri-axial accelerometers installed on three different building s floors: base (-1), 2 nd and 4 th floor. Since September of 2015 to May 2016, this concrete building has been subjected to more than 400 sensible earthquakes with magnitude (M I or M W) between 4.0 and 8.4, and peak ground acceleration (PGA) between 0.006% and 5.41% of g. Previous to begin to operate the vibration sensor network, the soil and structure have been studied using microtremor data to identify the soil's resonant period and building's vibration periods. This work reports the main considerations for the configuration network; the pre-processing of each seismic vibration record; the analysis with technics of dynamic systems identification based on the time domain and on the frequency domain in order to obtain the vibration periods. With these results the time variation and the correlation of the vibration periods with seismic intensity index were studied. The vibration sensor network s configuration identified adequately the most sensible earthquakes but produces a lot of records of false earthquakes. In other hand, the techniques are not equally effective to their purpose, but all are applicable, in fact is recommendable to use two or more of this techniques simultaneously. No structural damages were observed due to the analysed earthquakes, but significant variations of the vibration periods of building were observed. Then, the time variation of periods is not necessary a good predictor to the damage. Finally, the use of the post-event period pre-event period ratio was proposed as a better damage predictor. Keywords: vibration monitoring, earthquake response, dynamic response parameters.
Identification of dynamic response parameters of a concrete building during recent earthquakes by using structural vibration monitoring 1 Introduction In the practice of structural engineering is well known that the building s standard design and the standard calculus of dynamic response of buildings are developed considering a mathematical model with constant properties of the materials, geometry and soil foundation. Also, is well known that those suppositions are not necessarily the best, and the verification of the consistency between the mathematical model and the as-built building is not a common action. Then, questions as: How well the used mathematical model represent the response of the building? and How accurate are the results of calculus obtained? are very pertinent. Using analysis of building s vibration records is possible to identify some dynamic response parameters of the system, such as vibration frequencies, modal shapes, damping ratios, among others, in order to validate (total or partially) the mathematical model used for the structural calculus. However, several studies show that the dynamic response parameters are not constant and they are depending of environmental conditions, [1, 2, 3, 4]. In this work, a short discussion about the characterization of the dynamic response of a building using vibration monitoring is presented: a concrete building that belongs to the Faculty of Engineering of the Pontificia Universidad de Valparaíso has been instrumented with a tri-axial synchronized accelerometer network. This document reports the main considerations for the configuration network; the pre-processing of each seismic vibration record; the analysis using technics of dynamic systems identification based on the time domain and on the frequency domain, in order to obtain the vibration periods. With these results the time variation and the correlation of the vibration periods with seismic intensity index were studied. 2 Methodology and results 2.1 Building and vibration network information The main building of the Engineering Faculty (FIN-building) at Pontificia Universidad Católica de Valparaíso (PUCV) has been instrumented with the aim of studying its dynamic behaviour. The FIN-building has five stories and one basement floor. Its structure consists of reinforced concrete shear walls and reinforced concrete frames, with a lightweight shed of steel in the roof (fifth floor). This building, whose purpose are the academic work and engineering research, is located in the port-city of Valparaíso (Chile), a few blocks of the coastline. Figure 1 shows the location of Region of Valparaíso, the building s general emplacement in the city and a frontal view of the building. More information is available in [3]. 2
Source: Google earth, 2016 and self-prepared Figure 1: FIN-building: General location and frontal view. The vibration sensor network consists of three tri-axial force balance accelerometers, model SARA-SL06 with bandwidth g and a 24bit A/D converter, installed in three different levels of the building: the basement (underground level), second floor and fourth floor respectively. The location of each sensor at each floor and their identification code are indicated in the Figure 2. Also, Figure 2 shows a picture of one of the sensors. Is important to note that the longitudinal FIN-building s direction and the sensors direction is aligned with the North-South direction, while the transversal FIN-building s and sensors direction is aligned with the East-West direction. Source: Alvarez E, 2015 [4] Figure 2: FIN-building: Accelerometers network location. All the sensors (accelerometers) are networked so as to obtain vibration's data-record in way synchronized and simultaneous with a sampling ratio of 200 Hz. Also, the accelerometers network is connected to an external GPS for UTC time synchronization and, every minute it corrects the inherent micro-delay between the timers of the different sensors. In the case of blackouts, each one has its own support energy that consists in a battery system for autonomous operation during about 24h. Five days of continuous data is recorded before its permanently erased. For the strong motion recording, an STA/LTA trigger was initially configured, using the recommendations of Trncoczy [5] and considering additional time records of pre-event and postevent with long of 60 s each. The objective of the first trigger configuration implemented is to 3
acquire the maximum quantity of earthquake data for future reconfigurations considering the intensity as the parameter of interest. This configuration has allowed to obtain seismic data with small amplitudes, which was confirmed by the examination of the minimum intensity values obtained within the acceleration records of the database: the minimum peak ground acceleration (PGA) observed was approximately 0.0060% of g, while the minimum value with the Riddell y García intensity index (I a) [6] was 0.2276 cm/s 5/3. 2.2 Previous identification Previous analysis were developed in 2014 [3] before the operation of the seismic sensor network started, using a Tromino Engy 3G sensor (by Micromed/Moho). The Nakamura s method [7] was applied to analyze the microtremors records with duration of 16 minutes so as to identify the soil resonant period. The Figure 3 shows the soil resonant period identification through the average Fourier s spectrum H/V ratio. Source: Orrego T. 2015 [3] Figure 3: Soil resonant period identification. Average Fourier s spectrum ratio. On the other hand, the building s vibration modal periods in both orthogonal-horizontal direction (N-S and E-W) were estimated by applying the Enhanced Frequency Domain Decomposition method (EFDD), proposed by Brincker, Zhang and Andersen [8], to tri-axial microtremor records with 30 minutes long in each floor (2 nd and 4 th ), obtained by the accelerometer network. Table 1 contains the modal periods of the FIN-building. Table 1: Previous analysis: Building s modal period identification with EFDD method. Mode Modal period left limit (s) Modal period (s) Modal period right limit (s) 1 0.2962 0.3005 0.3055 2 0.1886 0.2072 0.2308 3 0.1254 0.1297 0.1339 4 0.0962 0.0989 0.1017 2.3 Strong motion database and pre-process of records The earthquakes database used in the analyses comprehends strong motion records from more than 400 earthquakes occurred between September, 16 th 2015 and May, 31 th 2016 with M I or 4
M w reported between 4.0 and 8.4, and approximate epicenter distances reported between 537.6 km and 23.5 km. The first earthquake considered in the database corresponds to September, 16 th 2015 Illapel earthquake (M w8.4) that occurs near the coastal cities of Coquimbo and La Serena, Chile. The identification number for each event is assigned as they occur. Continuous data of that day was saved to identify the initial conditions of the records. The date, magnitude (M I or M w) and epicenter coordinates were obtained from Centro Sismológico Nacional (information available on www.sismologia.cl). Some seismic intensity parameters studied by Riddell [9] were obtained for each record at each direction (N-S and E-W): PGA and I a [6] were calculated for each earthquake record. These indices were selected due to their good correlation with the response of structures with short vibration periods, as shown in Riddell [9]. Before characterizing the seismic intensity of each seismic record, a pre-process was applied: first a baseline correction was applied using a quadratic least square correction, after that, an 8 th Butterworth bandpass filter with cutfrequencies between 0.1 Hz and 30 Hz were applied to each record. Figure 4 contains the mentioned intensity values. Values of PGA between 0.006% and 5.41% of g were obtained. In other hand, the achieved I a values oscillate between the 0.2276 cm/s 5/3 and 217.73 cm/s 5/3. Figure 4: Seismic intensity indices (PGA and Ia) of earthquake database. 2.4 Vibration period s estimation Considering results of previous research [3, 4], for this work it was decided to use three well known techniques for structural vibration period estimation: the Space-State method (N4SID), used in system identifications (as can be seen in Ljung [10]), the Peak Picking Method (as described in Bendat and Piersol [11]), and the EFDD method [8]. 5
2.4.1 Space-state method (SSM) This technique [10], which works in the time domain, is a Single Input-Single Output (SISO) stochastic method. The parametric identification works iteratively with different model orders to discriminate the physical modes of the structure, based on stability analysis (stability diagrams). The model used in this work start with the 4th order and it grows in pairs until the 24th order. Due to the existence of two outputs (2nd and 4th floor records), two different models has been identified with theoretically the same information. The automatization of this analysis was made with a N4SID algorithm and implemented in Matlab. The criteria for the stability diagrams interpretation was as follow: first, the validation of the dynamic response parameters was tested and then the dynamic response parameters were grouped. In Table 2 are described the different criteria applied to the period and damping ratio values, while Figure 5 contains the modal periods identified using this State-space method. Table 2: Validation and grouping criteria. Validation Criteria Damping ratio less than 20% Period analysis range between 0.07 and 0.5 s Period and damping ratio obtained from the state matrix eigenvalues in complex and conjugate pairs Grouping Criteria Frequencies with differences less than 15% of the lower value are considered the same mode Minimum 5 frequency values per group. Frequency estimation bandwidth: 40% of first frequency The representative modal frequency is the average value of the group 2.4.2 Peak picking method (PPM) This non-parametric method, which works in the frequency domain with only the output signal [11], applies the Fast Fourier Transform (FFT in Matlab ) on the seismic recorded signal to compute and obtain the power spectral density function. Such information is used to identify a structure's vibration period, which is the period associated with each local peak. This only works properly if the identified periods are well separated of each other. The period range used in this analysis case (0.07 to 0.50 s) was the only consideration taken into account with this technique. 2.4.3 Enhanced frequency domain decomposition (EFDD) This technique, which works in the frequency domain using only output signals [8], is a variation of the classic Peak Picking technique described in 2.4.2. By using the singular value decomposition of the spectral matrix, a set of auto power spectral density functions is obtained. Each one of this function corresponds to a single degree of freedom system. The first singular function for every earthquake of the database was stored and analysed. The criterion for period identification considers two simultaneous conditions: (1) that the period analysis is in the range 0.07-0.50 s and (2) that the analyzed vibration period has 95% for coherence using the Modal Assurance Criterion (MAC). 6
3 Analysis of results and discussion 3.1 Sensor network configuration As can be seen in the Figure 4, the configuration of vibration network has been able to record very small (low intensity) earthquakes. Figure 5 shows that from all earthquakes detected (not only the earthquakes included in the seismic database, over 75% are false records. This configuration condition produces an inefficient memory use and an unnecessary permanent database check and then is recommendable to improve the trigger limit. Figure 5: Total, seismic and false records during the time. 3.2 Period-time and period-seismic intensity relationships According to other researches [3, 4], using a part of the same database the vibration periods of four modal shapes (M1 to M4) have shown a non-negligible variation during the time. A observed coefficient of variation (COV) varied between 1% and 21% depending the used method. Considering that the building does not have damages, the time variation of period s values is not attributable to this effects and it is due to the intrinsic dispersion of results of applied methods. Then, other parameter has been analyzed: the vibration period ratio (T f/t i), considering the vibration periods during the pre-event time (T i) and post-event time (T f). See Table 3. Table 3: Mean and Standard deviation of T f/t i ratio. Direction longitudinal (N-S) Direction transversal (E-W) Method Parameter M1 M2 M3 M4 M1 M2 M3 M4 SSM rav 1.06 1.01 0.97 0.99 1.01 0.98-1.03 Sr 0.10 0.11 0.08 0.07 0.07 0.10-0.12 PPM rav 1.04 1.00 0.99 1.01 1.00 0.97 1.00 1.00 Sr 0.08 0.08 0.09 0.04 0.09 0.11 0.05 0.04 EFDD rav 0.92 0.92 0.96 0.96 0.94 0.93 0.86 0.96 Sr 0.22 0.24 0.22 0.20 0.23 0.21 0.25 0.20 7
SSM PPM EFDD 22 nd International Congress on Acoustics, ICA 2016 This Table contends the average T f/t i ratio (r av) and its standard deviation (S r) for each modal shape and each used method (SSM and PPM only for 4 th floor records). In all cases the r av is close to 1.0 confirming no significant differences between before and after each earthquake. Figure 6: T f/t i ratio versus PGA. 8
SSM PPM EFDD 22 nd International Congress on Acoustics, ICA 2016 Figure 7: T f/t i ratio versus PGA. In the same way, Figures 6 and 7 show the relationships between the T f/t i ratio and the seismic intensity index (Figure 4). It is clear the T f/t i ratio for each modal shape (M1 up and M4 bottom) oscillates about 1.0 practically independent of the intensity index used, confirming that damages 9
did not occur due to the earthquakes analyzed. Then this parameter is a better no-damage predictor than only vibration period. Also, due to T f/t i ratio is a practically instantaneous parameter, long time effects such as temperature, weather and humidity are automatically corrected in the analysis. 4 Conclusions This study presented the installation, configuration and operation of a vibration sensor network in a multi-story concrete building with the purpose of monitoring its structural health. The configuration of the vibration sensor network identified adequately the most sensible earthquakes but produces a lot of records of false earthquakes thus, this configuration should be improved. No structural damages were observed due to the analysed earthquakes, but significant variations of the vibration periods of building have been observed with all methods. Then, the time variation of vibration periods is not necessary a good predictor to the damage. Finally, the use of the T f/t i ratio was proposed as a better damage predictor due to its consistency with the observations and because it corrects automatically long time effects. References [1] Boroschek R, Tamayo F, Aguilar R. Evaluación de los efectos ambientales en un edificio de media altura. XI Congreso Chileno de Sismología e Ingeniería Antisísmica. Santiago, Chile, 2015. In CD- ROM. [2] Moroni O, Sarrazín M, Venegas B, Villarroel J. Seismic behavior of Chilean bridges with seismic protection devices. Revista der la Construcción, Vol.14 (1), 2015 pp 53-59. [3] Orrego T. Seismic Instrumentation of Facultad de Ingeniería Building, PUCV, Pontificia Universidad Católica de Valparaíso, Valparaíso, Chile. 2015. [4] Álvarez E. Determination of Dynamic Properties of Building of Facultad de Ingeniería, PUCV. Pontificia Universidad Católica de Valparaíso, Valparaíso, Chile. 2015. [5] Trnkoczy A. Understanding and parameter setting of STA/LTA trigger algorithm, Deutsches GeoForschungsZentrum, Deutschlad. 2012. In CD-ROOM. [6] Riddell R, García JE. Hysteretic energy spectrum and damage control. Earthquake Engineering & Structural Dynamics. Vol 30 (12), 2001, pp 1791-1816. [7] Nakamura Y. A Method for Dynamic Characteristics Estimation of Subsurface using Microtremor on the Ground Surface. Quarterly Reporté of Railway Technical Research Institute RTRI. Vol 30 (1), 1989, pp 25-33. [8] Brincker R, Zhang L, Andersen P. Output-Only Modal Analysis by Frequency Domain Decomposition. International Conference on Noise and Vibration Engineering. Leuven, Belgium. 2000. In CD-ROOM. [9] Riddell, R. On Ground Motion Intensity Index. Earthquake Spectra. Vol 23 (1), 2007 pp 147-173. [10] Ljung, L. System Identification: Theory for the User. Prentice Hall, NJ (USA), 1st edition, 1987. [11] Bendat J, Piersol A. Engineering Applications of Correlation and Spectral Analysis. John Wiley & Sons, NY (USA). 3rd edition. 2000. 10