Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Engineering Structures 276 (2023) 115372

Contents lists available at ScienceDirect

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

Long-term continuous seismic monitoring of multi-span highway bridge


and evaluation of bearing condition by wireless sensor network
Dionysius M. Siringoringo a, *, Yozo Fujino a, Makoto Suzuki b
a
Institute of Advanced Sciences, Yokohama National University, 79-1 Tokiwadai, Hodogayaku, Yokohama, Kanagawa 240-8501, Japan
b
SONAS Co Ltd, 5-24-2 Hongo, Bunkyoku, Tokyo, Japan

A R T I C L E I N F O A B S T R A C T

Keywords: A 45-month continuous seismic monitoring of a multi-span highway bridge by wireless sensor network (WSN) is
Highway bridge described in this paper. The WSN monitoring system is developed using a simultaneous high-speed flooding
Wireless sensor network technique and deployed to monitor a continuous 12-span 381.8 m long highway bridge in Katsuta city, Ibaraki
Seismic monitoring
prefecture, Japan. The WSN monitoring system consisted of 20 triaxial accelerometers placed on bridge girder,
System identification
Isolation bearing
pier caps, abutment, and on the ground. The monitoring system has successfully recorded seismic responses of 63
earthquakes from August 2017 to December 2020. The paper describes the WSN monitoring system, analyses of
seismic responses, and structural assessment using the recorded seismic responses. The analyses include inves­
tigation on seismic response characteristics of girder, piers, and bearings, system identifications, and seismic
performance evaluation of isolation bearings. The WSN monitoring system provides high quality of seismic re­
sponses. Evaluations of ground motions recorded by WSN with the existing strong motion KNET seismograph
network have demonstrated the accuracy, reliability, and robustness of WSN monitoring system. Using three
system identification methods, dominant bridge modal parameters and their variations with respect to earth­
quake levels were investigated. High-frequency filtering effect was utilized as indicator in assessment of isolation
bearing performance. Systematic classification by clustering algorithm was conducted to determine whether the
isolation bearings have functioned normally.

1. Introduction short-term deployment on civil structures including bridge structures


[10–15]. Most of WSN full-scale deployments on bridge structures,
Seismic monitoring systems of bridge structures have been imple­ however, have been of temporary short-term monitoring campaigns for
mented in many countries and have shown great potential for immediate demonstration and research purposes. Long-term and continuous de­
structural assessment after an earthquake. Seismic monitoring as a part ployments for structural health monitoring are still relatively rare.
of structural health monitoring (SHM) systems can improve the safety, Notable studies of long-term deployments of WSN system in literature
and reliability of structures, reduce the possibility of undetected struc­ are the cable-stayed bridges Jindo bridge [16,17], Geumdang bridge
tural damage, and extend service life. Currently, the majority of the [18] and Hwamyung bridge in Korea [19], New Carquinez Suspension
permanent seismic monitoring systems deployed on large bridges use Bridge [20] and Golden Gate Bridge in the US [21], Fatih Sultan Mehmet
wired sensing system [1–9]. Compared to conventional monitoring by Bridge in Turkey [22], Gi Lu bridge in Taiwan [23], and Ferriby Road
wired sensors, monitoring by wireless sensor network (WSN) offers Bridge in UK [24].
various benefits: elimination of the need for cabling, installation, and Implementation for continuous seismic monitoring application is
protection which yields to a simpler and low-installation cost. WSN even more limited. Some studies have proposed the use of WSN in
hardware is also highly integrated and more compact, since the sensor structural monitoring for earthquake-induced damage evaluation of
nodes include supporting components like signal analyzer, battery, and structures in the laboratory-scale experiments and actual field imple­
data memory. There have been various research and developments on mentations on buildings and bridges [25–26]. These studies and
WSN in the past two decades. Accuracy, reliability, and robustness of the implementations have shown promising results, although challenges
developed WSN have been investigated in model-scale experiments and remained on how to deal with effective use of power consumption,

* Corresponding author at: IAS Assoc. Professor Institute of Advanced Sciences, Yokohama National University, Japan.
E-mail address: dion@ynu.ac.jp (D.M. Siringoringo).

https://doi.org/10.1016/j.engstruct.2022.115372
Received 11 July 2022; Received in revised form 22 November 2022; Accepted 23 November 2022
Available online 5 December 2022
0141-0296/© 2022 Elsevier Ltd. All rights reserved.
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

synchronization and automatization of the recording and data analysis. downstream girder. From the south end to the north end, the girder is
The WSN monitoring system has also been used for screening the post- supported by 12 rectangular reinforced concrete piers and 1 abutment.
earthquake structural conditions using ambient responses of the struc­ The heights of the piers are decreasing from 11.18 m at the south end
tures to detect the sign of structural damages [26,27]. (P35) to 8.9 m at the north end (abutment A2). The piers in the middle
The work presented in this paper is a part of comprehensive research (P36 – P46) consist of two wall type piers each with uniform width and
works for development and implementation of large-scale WSN moni­ thickness 5.5 m and 1.7 m, respectively. Meanwhile, the south end pier
toring system on large bridges. In this research work, the WSN system (P35) is a single wall pier with the width of 22.5 m and thickness 4.8 m,
was deployed on two large bridges, a cable-stayed bridge [28] and a which also serves as a connection to approach span of Shin-Nakagawa
multi-span highway girder bridge presented in this paper. The paper has cable-stayed bridge. At the north end, the bridge girder is supported
three main objectives: 1) to describe a development and deployment of a by abutment A2, which is of single wall type with the width of 22.9 m
stable, efficient and reliable continuous long-term seismic monitoring and thickness1.8 m. The girder is 381.1 m long and 10.9 m wide open
system on multi-span highway bridge using WSN and explain the lessons cross section made of composite steel. Upper sections of the girder are
learned from monitoring experience for improvement of WSN-based connected by shear studs to the reinforced concrete slab supporting the
monitoring system, 2) to describe procedures for structural assessment bridge deck , while the lower part is braced with steel bracings. The
of a multi-span highway bridge after an earthquake by identifying the girder rests on two types of isolation bearing pads: the south end pier
global and local structural parameters and their possible changes caused P35 and abutment A2 use the steel sliding bearing, while all piers in the
by the earthquake, and 3) to describe a methodology for quick assess­ middle span use natural rubber bearing (NRB). Five units of NRB with
ment of isolation bearing and to check possibility of locked bearing dimension: 52 × 52 × 15.5 cm were installed on each pier and abutment.
incident using recorded seismic responses in an attempt to ensure
functionality of seismic isolation on the highway bridge.
The paper is arranged in the following order. Next section describes 2.2. Wireless sensor network system
the monitored bridge and sensing system followed by a brief description
of wireless sensor network and its implementation. Then the paper de­ The deployed WSN monitoring system comprises of wireless sensing
scribes the recorded seismic events and analyses of the seismic records system that uses the Concurrent Transmission Flooding (CTF) principle
via system identifications. This is followed by investigation on seismic within a protocol called Choco [29]. The protocol was recently devel­
performance of isolation bearing using time–frequency analysis of pier oped, and it has five main features: low power consumption, robust
and girder seismic responses. Finally, the paper concludes with discus­ system, fast synchronization, reliable transmission, and high
sion on the results of monitoring. throughput. The sensor requires relatively low power consumption in an
order of several hundred microamperes for all stages including sam­
2. Description of bridge, wireless sensors network and pling, storage, and synchronization. The power for one unit of sensor
monitoring system node is provided by five D-cell batteries and has effective lifespan of 18
months. The sensor system is designed to be stable and robust such that
2.1. Bridge description it can be installed by those even without professional qualification and
experience in wireless sensors installation and communication. Syn­
Object of monitoring is Katsuta viaduct, a 12-span 381.1 m long chronization process among sensor nodes is achievable within 30–50 μs.
highway bridge located on north of Nakagawa riverbank in the Katsuta The transmission system is designed by end-to-end ACK mechanism that
city, Ibaraki Prefecture, east Japan. It connects Katsuta highway on the can minimize data losses. The throughput rate of the system at the
north and Shin-Nakagawa cable-stayed bridge on the south. Basic in­ application layer is 2 KB/s. Communication in the sensor network is
formation on bridge layout, dimensions and sensor arrangement are controlled by a sink node (or master node) via CTF by periodically
depicted in Fig. 1. The Katsuta viaduct is a continuous span bridge with transmitting time-synchronization packets to the other nodes. The CTF-
two separated girders supporting two traffic lanes: Katsuta-bound Lane based communication system is more robust compared to the routing-
on the upstream girder and the Shin-Nakagawa-bound Lane on the based communication system because it utilizes numerous paths
concurrently. The Choco protocol is effective in reducing the overhead

Fig. 1. Katsuta highway multi-span viaduct (a) Photos (b) Dimension and (c) layout of wireless sensor network.

2
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

of idle listening by letting the nodes to sleep when not collecting the cement. Piers P35, P38, P41, P44, and abutment A2 have the set of
packets, while the sink node keeps rearranging the lost packets until sensors on the pier caps and on the girder right above the isolators. As
they are all delivered. Meanwhile, a high throughput system is achieved will be shown later, the arrangement of piers and girder sensor nodes
by continuously organizing the slots for nodes until they all received the was intended to investigate isolator bearings performance during
packets. The above-mentioned features are effective to realize a high- earthquake. In the placement of all sensors the x, y, and z directions
precision sensor with low power consumption. correspond to bridge vibrations in transversal, longitudinal and vertical
In the field deployment, sensor nodes are assigned to sleep almost directions, respectively.
constantly. The power consumption increases every 0.005 s to about 1 The WSN monitoring system was started in August 2017 and
mA due to CPU’s activity of generating a square wave with 10 ms cycle. completed in December 2020. In the 45 months continuous monitoring,
Next, to support the activity of CPU and SRAM in reading and writing the monitoring system has successfully recorded 63 seismic events.
the recorded acceleration, the power consumption rises to 2 mA for Majority of the recorded seismic events are from small and moderate
every 31.25milli-second. In average, data sampling and writing opera­ level earthquakes caused by near-field earthquakes (NFE) and far-field
tion require 110μA, time-synchronization process requires 5.6μA, and in earthquakes (FFE). In this paper, classification of a small and moder­
an idle condition each sensor consumes 60μA of power. One unit of ate seismic event is made according to the JMA’s seismic intensity scale.
sensor node includes two types of accelerometers: analog ADXL362 for A seismic event with seismic intensity less than 4 is classified as a small
low-power type and EPSON M− A351AU for high precision sensor type. event, whereas an event with seismic intensity between 4 and 5- is
In this paper, however, only records from low-power sensor ADXL362 categorized as a moderate one. Table 1 lists chronically all earthquakes
are presented and analyzed. More details on specification of the sensors recorded during the 45-month monitoring period. Note that seismic
are given in Fig. 2(b). intensity classification according to peak ground acceleration (PGA)
To accommodate outdoor environment condition, the hardware recorded on the site is given as follows. Seismic Intensity (SI) 1 is for
component was designed against direct heat, water, chemical, and me­ PGA less than 2.5 cm/s2, SI = 2 is for PGA between 2.5 and 8 cm/s2, SI =
chanical impact. . All sensor nodes were encased in a transparent water- 3 is for PGA between 8 and 25 cm/s2, SI = 4 is for PGA between 25 and
tight PVC box mounted with external antenna and equipped with five D- 80 cm/s2, and SI = 5- (lower 5) is for PGA between 80 and 140 cm/s2
cell batteries each. To monitor sensor condition, switches and status The table shows that of the 63 recorded seismic events, 18 events are of
LEDs are placed on location that can be easily accessible from outside SI 1, 20 are of SI 2, 14 events are of SI 3, 7 events are of SI 4, and 5 events
the box. Fig. 2(b) illustrates the photo of a sensor node. Further infor­ are of SI 5-.
mation on the wireless sensor system and its detailed features are Table 2 shows the list of 50 earthquakes sorted from the largest to
available in [29]. smallest peak ground accelerations. Notable earthquakes in this table
are the near-field earthquakes on March 30th, 2018 (M5.1), July 17th,
2018 (M4.8), June 17th, 2019 (M5.1), June 4th, 2020 (M4.7), and
2.3. WSN seismic monitoring system and recorded seismic events
December 30th, 2020 (M5.1). All these near-field earthquakes were
within 25 km away from the bridge site. A few FFEs have been recorded
The sensor layout is illustrated in Fig. 1(c). Totally 20 sensor nodes
successfully, the farthest one was recorded on February 13th, 2020
were installed, with the following configurations: 7 nodes along the
(M7.0) with epicenter 1163 km away from the bridge.
girder, 8 nodes on piers, 2 nodes on abutment, and 3 nodes on the
The recorded raw data was subjected to data preprocessing and
ground. Note that only the upstream girder bridge was instrumented by
cleansing procedures before being used in analyses. The first procedure
WSN monitoring system. The pier sensor nodes were installed on the
is a quick data scanning to detect any possible sensor malfunction, data
pier caps next to the isolation bearing plate (Fig. 2(c)). Fig. 2(d) shows
loss, and synchronization problems. The next procedure is to correlate
the installation of sensor nodes on the lower flange of the I-girder
part of the data associated with earthquake responses, this includes
directly above the isolation bearing. All ground sensors were placed
procedure to define the beginning and the end of an earthquake
inside waterproof containers that were fixed to the ground by mortar

Fig. 2. (a). Schematic figure of the WSN-based seismic monitoring system on the Katsuta highway multi-span viaduct, (b) specification of sensor node, (c) placement
of sensor node on pier cap, (d) placement of the sensor node on the girder above the bearing, (d) placement of the sensor node on the ground.

3
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Table 1
List of all 63 recorded seismic events by WSN monitoring system at Katsuta Bridge sorted chronically with details on epicenter, magnitude and seismic intensity at
bridge site.
No Date(mm, Epicenter Magnitude (M) Distance from No Date Epicenter Magnitude (M) Distance from
dd,yy) coordinate [JMA Seismic Epicenter to bridge (mm,dd, coordinate [JMA Seismic Epicenter to bridge
Intensity] site [km] yy) Intensity] site [km]

1 3/8/2017 36.04 M4.6 [1] 98.74 33 12/19/ 36.3 N,140.7E M4.2 [3] 14
N,139.53E 2019
2 5/8/2017 36.27 M4.1 [1] 20.2 34 1/3/2020 35.8 N,141.2E M5.9 [3] 85
N,140.36E
3 8/18/ 36.35 M3.9 [1] 43.19 35 1/14/ 36.1 N,139.9E M5.0 [2] 66
2017 N,141.04E 2020
(04:46)
4 8/26/ 37.01 M4.7 [1] 74.92 36 1/14/ 36.1 N,140.9E M4.9 [3] 42
2017 N,140.35E 2020
(13:25)
5 8/27/ 36.44 N, M4.8 [1] 21.56 37 1/21/ 36.4 N, 140.7E M4.3 [3] 124
2017 140.34E 2020
6 10/19/ 36.29 N, M3.8 [1] 8.61 38 2/1/2020 36.4 N, 140.7E M5.1 [2] 74
2017 140.51E
7 11/2/ 36.48 N, M4.3 [1] 16.22 39 2/1/2020 36.0 N, 140.1E M5.3 [3] 57
2017 140.46E
8 11/3/ 36.81 N, M4.8 [2] 55 40 2/6/2020 36.4 N,141.6E M5.6 [1] 93
2017 140.53E
9 2/6/2018 36.64 N, M3.7 [1] 45 41 2/12/ 37.3 N,141.4E M5.5 [1] 129
140.90E 2020
10 3/30/ 36.44 N, M5.1 [2] 15 42 2/13/ 44.7 N,148.9E M7.0 [1] 1163
2018 140.62E 2020
11 4/3/2018 36.75 N, M3.8 [1] 50 43 4/12/ 36.2 N,140.0 E M5.1 [2] 53
140.71E 2020
12 5/17/ 36.35 M5.3 [1] 75 44 4/18/ 27.2 N,140.7E M6.9 [1] 1018
2018 N,140.60E 2020
13 7/17/ 36.43 M4.8 [3] 15 45 4/26/ 36.2 N,140.1E M4.8 [1] 45
2018 N,140.69E 2020
14 9/5/2018 36.48 M5.5 [2] 72 46 5/4/2020 36.4 N,141.0E M4.4 [1] 41
N,141.34E
15 11/27/ 36.07 M5.0 [1] 75 47 5/6/2020 35.6 N,140.1E M5.0 [1] 94
2018 N,139.86E
16 5/1/2019 36.27 M4.5 [1] 42 48 5/11/ 36.4 N,141.1E M5.5 [2] 50
N,141.01E 2020
17 5/25/ 35.3 N 140.3E M5.1 [1] 119.8 49 5/19/ 37.5 N,141.5E M5.3 [1] 152
2019 2020
18 5/27/ 36.7 N, 140.7 M4.2 [1] 40.18 50 6/1/2020 36.2 N,140.4E M5.3 [2] 22
2019 E
19 6/17/ 36.5 N,140.6E M5.2 [2] 16.32 51 6/4/2020 36.4 N, 140.7E M4.7 [3] 14
2019
20 6/18/ 38.6 N,139.5E M6.8 [1] 266.3 52 6/25/ 35.5 N,141.2E M6.1 [2] 111
2019 2020
21 7/17/ 37.1 N,141.6E M4.4 [1] 124 53 7/9/2020 36.1 N,139.9E M4.7 [1] 66
2019
22 7/20/ 36.4 N, 140.8E M3.6 [1] 22 54 7/29/ 35.9 N, 140.9E M4.6 [2] 59
2019 2020
23 7/25/ 35.1 N,140.6E M5.3 [1] 140 55 8/30/ 36.2 N, 139.8E M4.4 [2] 70
2019 2020
24 7/28/ 33.0 N,137.4E M6.5 [2] 472 56 9/7/2020 36.7 N, 141.6E M5.2 [1] 101
2019
25 7/29/ 35.8 N,141.0E M4.6 [1] 74 57 9/10/ 36.5 N,140.7E M4.1 [2] 20
2019 2020
26 7/30/ 32.9 N,140.8E M6.0 [1] 385 58 9/12/ 38.7 N,142.3E M5.0 [1] 302
2019 2020
27 8/4/2019 37.7 N,141.7 E M6.4 [2] 181 59 9/17/ 36.5 N, 140.6E M4.3 [2] 16
2020
28 8/24/ 37.4 N,142.5E M5.5 [1] 208 60 9/20/ 35.0 N, 140.2E M4.5 [1] 154
2019 2020
29 12/4/ 36.8 N,140.6E M4.8 [3] 49 61 11/22/ 36.6 N, 141.1E M5.7 [3] 55
2019 2020
(10:30)
30 12/4/ 36.9 N,139.7E M4.7 [1] 97 62 11/25/ 36.1 N,139.9E M4.3 [1] 66
2019 2020
(19:35)
31 12/5/ 36.8 N, 140.6E M4.5 [2] 49 63 12/30/ 36.4 N,140.6E M5.1 [3] 60
2019 2020
32 12/11/ 37.7 N,141.9E M5.2 [1] 191
2019

4
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Table 2
List of 40 selected recorded earthquakes at Katsuta Bridge sorted from the largest to the smallest. The earthquakes in this list are used in system identification and
analysis of bearing seismic performance. Note: items in bold are the three largest NFE, and items written in Italic are the three farthest FFE.
Earthquake Date Magnitude Epicenter Distance from Epicenter to bridge site (km) Peak Ground Acc (cm/s2) RMS of Ground Acc.(cm/s2)
Number (mm/dd/ (depth in km)
yy)

1 7/17/2018 M4.8 (52) 36.43 N,140.69E 15 153.85 6.29


2 6/17/2019 M5.1 (80) 36.50 N,140.6E 16 135.61 5.58
3 6/4/2020 M4.7 (50) 36.4 N, 140.7E 14 130.75 4.42
4 11/22/2020 M5.7 (50) 36.6 N, 141.1E 55 101.60 5.16
5 3/30/2018 M5.1 (56) 36.44 N,140.62E 11 97.57 5.48
6 12/30/2020 M5.1 (60) 36.4 N,140.6E 6 73.51 2.83
7 1/21/2020 M4.3 (50) 36.4 N, 140.7E 14 49.97 1.62
8 4/12/2020 M5.1 (50) 36.2 N,140.0 E 53 42.25 2.12
9 1/3/2020 M5.9 (30) 35.8 N,141.2E 85 36.76 3.13
10 6/25/2020 M6.1 (30) 35.5 N,141.2E 111 32.85 3.85
11 9/17/2020 M4.3 (60) 36.5 N, 140.6E 16 30.09 1.10
12 6/1/2020 M5.3(1 0 0) 36.2 N,140.4E 22 27.93 1.68
13 9/5/2018 M5.5 (60) 36.5 N,141.3E 72 27.08 1.73
14 5/11/2020 M5.5 (50) 36.4 N,141.1E 50 21.41 1.97
15 9/10/2020 M4.1 (20) 36.5 N,140.7E 20 20.13 0.67
16 8/4/2019 M6.2 (50) 37.7 N,141.7 E 181 19.04 1.50
17 12/4/2019 M4.8 (10) 36.8 N,140.6E 49 17.33 1.97
18 2/1/2020 M5.3 (70) 36.0 N, 140.1E 57 17.01 1.95
19 12/19/2019 M42 (90) 36.3 N,140.7E 14 15.51 0.91
20 2/12/2020 M5.5 (80) 37.3 N,141.4E 129 14.85 1.50
21 1/14/2020 M4.9 (50) 36.1 N,140.9E 42 14.42 1.33
22 7/28/2019 M6.5(4 2 0) 33.00 N,137.4E 472 11.34 1.96
23 11/27/2018 M5.0 (44) 36.07 N,139.86E 75 10.38 0.92
24 12/5/2019 M4.5 (10) 36.8 N, 140.6E 49 10.14 0.93
25 5/1/2019 M4.5 (42) 36.27 N,141.01E 42 9.97 0.60
26 9/7/2020 M5.2 (20) 36.7 N, 141.6E 101 8.52 1.08
27 5/17/2018 M5.3 (52) 36.35 N,140.60E 75 8.48 0.84
28 6/18/2019 M6.8 (10) 38.6 N,139.5E 266.3 8.05 1.93
29 5/4/2020 M4.4 (50) 36.4 N,141.0E 41 8.02 0.90
30 5/27/2019 M4.2 (10) 36.7 N, 140.7 E 40.18 7.08 0.44
31 2/6/2020 M5.6 (40) 36.4 N,141.6E 93 6.88 1.04
32 4/26/2020 M4.8 (70) 36.2 N,140.1E 45 6.82 0.74
33 8/30/2020 M4.4 (60) 36.2 N, 139.8E 70 6.59 0.31
34 9/20/2020 M4.5 (80) 35.0 N, 140.2E 154 6.53 0.48
35 5/19/2020 M5.3 (50) 37.5 N,141.5E 152 6.05 0.64
36 7/20/2019 M3.6 (60) 36.4 N, 140.8E 22 5.68 0.35
37 7/29/2020 M4.6(20) 35.9 N, 140.9E 59 4.76 0.54
38 7/9/2020 M4.7(50) 36.1 N,139.9E 50 4.73 0.53
39 2/13/2020 M7.0 (1 6 0) 44.7 N,148.9E 1163 2.46 0.35
40 4/18/2020 M6.9 (4 9 0) 27.2 N,140.7E 1018 2.73 0.60

Fig. 3. Flowchart of data preprocessing and cleansing procedures.

5
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

response. The arrival time of the P-wave on the ground motion was Because of Typhoon Hagibis, all ground sensors had to be replaced, and
selected as the beginning of the earthquake response. Meanwhile, the the sink node needed repairment. The measurement system went offline
time in the energy plot when 95 % of cumulative energy has been for several weeks for overall checking and repairment.
reached was defined as the saturation point or the end of the earthquake
response. Afterward, all records corresponding to the selected data 3. Characteristics of recorded seismic responses
segments were examined for possibility of erroneous responses. Typical
erroneous responses include data loss, high spikes, and linear or low- 3.1. Comparison of recorded ground accelerations by WSN and existing
frequency offset of the accelerations. seismometer
By scanning through and comparing the records using cumulative
earthquake energy plot, data loss in the acceleration of each sensor node To examine performance and accuracy of WSN monitoring system in
can be detected. This is because the saturation points that indicate 95 % capturing and recording an earthquake, ground motions recorded by
of cumulative energy typically appear on equal time stamp. Therefore, a WSN sensor node 2 and IBR007 of KNET Seismograph network are
very different time stamp of saturation point is generally caused by a compared. IBR007 is a triaxial seismometer installed by National
significant data loss. Next, by comparing the peak responses of several Research Institute for Earth Science and Disaster Resilience (NIED) [30]
sensor nodes within the same cluster, we can localize the possible of Japan about 3 km away from the bridge. Fig. 4 shows the comparisons
occurrence of large erroneous spike caused by signal noise. The data of ground accelerations and the Fourier spectra recorded in the largest
associated with erroneous spike was removed and then rectified by NFE (July 17th,2018) and in the largest FFE (July 28th, 2019). Although
interpolation procedure using the data prior to and following the achieving a precise quantitative comparison between WSN sensor node
removed data segments. Likewise, interpolation functions were utilized 2 and IBR007 sensor is somewhat difficult due to the significant distance
to rectified data loss on some small segments. Finally, to remove the between the two locations and relatively dissimilar soil conditions,
linear or low-frequency offset, the data was detrended. The pre- reasonable and comparable characteristics of ground accelerations
processing procedure above was conducted offline for all data sets. would demonstrate the capability of WSN system. Four characteristics of
The complete procedures for data pre-processing and cleansing are the ground motions were compared: 1) P-wave arrival time, 2) ampli­
schematically illustrated as a flowchart in Fig. 3. tude and the occurrence time of PGA, 3) the presence of impulse on NFE
The monitoring system has an ability to detect the malfunctioning ground motion or its absence on the FFE ground motion, and 4)
sensor before data processing. As explained previously, the Choco pro­ resemblance in frequency contents of the ground motions.
tocol works such that the sink node keeps sending signal and rearranging Fig. 4(a) and (b) clearly demonstrates that the ground accelerations
the packets until they are all delivered. This procedure can be monitored of WSN node 2 and KNET-IBR007 have the same characteristic of im­
online for each sensor on the website designed to monitor the system pulse typically observed in a near-field ground motion. The P-wave
operation. The website also provides information on level of power arrival times recorded by WSN Node 2 and KNET-IBR007 are very close,
consumption on each sensor and the time elapses since the last time separated by less than 1 s. The peak ground accelerations on both sen­
signal was sent from one sensor. With such information available, we sors also occur almost simultaneously with time lag less than 1 s. Note
can monitor condition of each sensor and the overall monitoring system. that the acceleration response spectra (Fig. 4(c) reveal a good agreement
Therefore, when one or more sensors do not function properly, the in all directions respectively. Some discrepancies appear in the re­
system can show the number of malfunctioning sensor and its possible sponses are believed to be associated with the differences in soil con­
source of problem such as communication problem, package loss or ditions at the two locations of sensors. However, they are relatively
battery exhaustion. small, with the mean absolute error (MAE) between the response spectra
It should be mentioned that during 45-month continuous seismic obtained by KNET and WSN sensors being less than 7 %. For the largest
monitoring, the measurement system experiences several challenges. FFE, comparison of ground accelerations and their response spectra
While the overall performance and stability of the system is satisfactory, recorded in the FFE (July 28th, 2019) earthquake are depicted by Fig. 4
the system suffers from two type of temporary failures in 45-month (d) and (e). Similar to the previous figure, it is evident that P-wave
continuous seismic monitoring. The first type of failure is the local arrival time recorded by WSN Node 2 and KNET-IBR007 are nearly
failure in which one or two sensor nodes fail to record or transmit the identical with the time lag less than 1 s. In contrast to NFE ground
data. This type of failure mostly relates to loss of power because the motions, both FFE ground accelerations reveal the same characteristics
battery is exhausted. Note that power consumption of each sensor is where the responses develop slowly from the time of P-wave arrival until
monitored online and reported on the monitoring website, so that the the peak excitation is reached. Note that unlike the ground motion of
remaining lifetime of a sensor can actually be predicted, and the time for NFE, there is no presence of sudden pulse in the FFE ground motions.
battery replacement can be decided before. However, there were cases These are the typical characteristics of FFE ground motions. The figures
where the power on some sensors dropped suddenly out of the predic­ demonstrate that time of occurrence and amplitude of PGAs for both
tion causing local failure on one or more sensors. This occurred mostly sensors are very close. A good agreement is shown by the spectra as
during winter period or rainy season (i.e., June to July). In this case, the indicated by the MAE values being less than 4 % (Fig. 4(f)). Comparisons
solution is to replace the battery or change the sensor case with the new of record ground motions recorded from both NFE and FFE have
one. Because of its local nature, this type of failure does not affect the confirmed the robustness and effectiveness of WSN monitoring system in
entire system. Other sensors and the data transmission system still recording both small and large seismic events reliably and accurately.
function normally. The adjustment, however, must be taken when
dealing with the data analysis because information from the failed 3.2. Seismic responses of the girder
sensors must be excluded in the analysis. The second type of failure is the
complete system failure because of failure of the sink node and/or the As explained previously, the girder is continuous over the entire span
ground nodes. During the 45-month of monitoring period, the system and made of composite steel with open cross section configuration.
experienced the complete system failure once following the passage of Under this circumstance, the girder is supposed to move as a rigid body
Typhoon Hagibis (No.16) on October 11, 2019. The typhoon passed longitudinally without relative in-plane longitudinal motion. This im­
through the bridge site, and the strong wind caused by the typhoon hit plies that all sensor nodes along the girder are supposed to have identical
the sink node causing the failure of its communication system. Once the longitudinal accelerations without significant difference in response
sink node failed, the entire system failed because communications characteristics. Fig. 5 reveals that such expectation is confirmed. The
among sensors were interrupted, and data transmission became impos­ longitudinal accelerations recorded by nodes 5,8, 9, 10,13, and 16 in the
sible. The flood that followed the typhoon inundated all ground sensors. largest NFE (July 17, 2018) are equal (Fig. 5(a) and (b)). Likewise in the

6
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Fig. 4. Ground accelerations recorded by node 2 of WSN system and by IBR007 of nearby KNET seismometer network in the largest NFE (July 17th, 2018): a) IBR007
acceleration time-histories, b) node 2 of WSN acceleration time-histories, c) acceleration response spectra. In the largest FFE (July 28th, 2019): d) IBR007 accel­
eration time-histories, e) node 2 of WSN acceleration time-histories, f) acceleration response spectra.

Fig. 5. Longitudinal girder accelerations obtained from sensors along the girder, in the largest NFE (July 17th,2018): (a) acceleration on nodes 5,8,9 and (b) ac­
celerations on nodes 10,13,16; in the largest FFE (July 28th,2019): (c) acceleration on nodes 5,8,9 and (d) Accelerations on nodes 10,13,16. Spectra of acceleration in
the largest nearfield earthquake (July 17th,2018): (e) on nodes 5,8,9. (f) on nodes 10,13,16. Spectra of acceleration in the largest FFE (July 28th,2019): (g) on nodes
5,8,9. (h) on nodes 10,13,16.

largest FFE (July 28, 2019), amid smaller amplitudes, all girder sensor low frequency peaks around 1.4–1.6 Hz and 4 Hz.
nodes recorded equal longitudinal accelerations (Fig. 5(c) and (d)). This Girder responses in lateral direction are generally larger than in the
indicates the girder indeed moved longitudinally as a rigid body, and longitudinal and vertical directions as shown by Fig. 6 and Fig. 7 for the
this vibration characteristic can be captured accurately by the WSN sensor nodes on piers P36, P37, P38, P41, and P44. For abutment A2,
monitoring system. Spectra of accelerations in Fig. 5(e-h) reveal that however, the amplitude of girder lateral vibration is not significantly
girder’s longitudinal vibrations in both NFE and FFE were dominated by larger than vibration in the longitudinal and vertical directions. Note

7
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Fig. 6. Comparisons of lateral girder accelerations and their frequency spectra obtained from sensors along the girder in the largest NFE (July 17th,2018). Ac­
celeration responses: (a) Nodes 5,8, 9,10, (b) Nodes 13,16,19. Fourier spectra of accelerations: (c) Nodes 5,8,9,10, and (d) Nodes 13,16,19.

Fig. 7. Comparisons of lateral girder accelerations and their frequency spectra obtained from sensors along the girder in the largest FFE (July 28th,2019). Accel­
eration responses: (a) Nodes 5,8, 9,10, (b) Nodes 13,16,19. Fourier spectra of accelerations: (c) Nodes 5,8,9,10, and (d) Nodes 13,16,19.

that the frequency spectra plots reveal that girder lateral vibrations are previously, the piers in the middle of bridge continuous span use bear­
dominated by single mode around 1.5 Hz for all sensor nodes except the ings made of natural rubber, while the ones at the ends (i.e., P36 and A2)
two nodes at the girder ends, namely node 5 on P36 and node 19 on A2, use combination of natural rubber bearing and steel sliding friction
which have higher frequency peaks around 2 Hz and 4 Hz, respectively. bearing, which have higher initial stiffness. In terms of spectra of girder
The difference in both amplitude and frequency characteristics of girder lateral accelerations, Fig. 7(c and d) reveal that in case of FFE, lower
lateral vibration is thought to be caused by the different type of bearing frequency peak at 0.75 Hz was also excited by the long-period ground
used in the middle span and at the ends of girder. As explained motion component. Frequency peak within the similar frequency range

8
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

was also observed in the longitudinal girder acceleration (Fig. 5(g) and later, this different characteristic is due to the different behavior of
(h)). isolations in the large and small earthquakes.
In the vertical direction, vibration of girder is recorded only by Fig. 12 and Fig. 13 illustrate the lateral accelerations of the piers for
sensors located directly above the piers. Therefore, their amplitudes are the largest NFE and FFE, respectively. In the case of NFE (July 17, 2018),
notably smaller since vertical movement of girder at that location is the acceleration peaks were almost as twice larger as the longitudinal
limited by vertical stiffness of the bearing. Fig. 8 and Fig. 9 show the ones for all piers in the middle spans. Whereas for the largest FFE (July
typical girder vertical accelerations and their spectra for the largest NFE 28, 2019), the amplitudes of lateral and longitudinal accelerations were
and FFE. One can observe similar characteristics of the spectra where relatively equal. Similar to the responses in longitudinal direction, the
two frequency peaks around 1.46 Hz and around 5.1 Hz dominate the spectra of lateral accelerations show different characteristics between
spectra. For the NFE, the 5.1 Hz peak dominate the responses, while for the largest NFE and FFE. In the case of largest NFE, the pier vibrations
the FFE the lower frequency 1.46 Hz dominate the responses. This is are dominated by many frequency peaks around 1–5 Hz, whereas for
believed to be related to the characteristics of FFE ground motions largest FFE the dominant peaks are around 1–2 Hz. The differences in
which are normally more dominated by long-period component. Note the spectra characteristics are thought to be related to different behavior
that the first peak in this spectra figure appear around the same fre­ of isolations and the effect of different frequency contents in the
quency with peak of the girder’s lateral acceleration, which is related to respective earthquakes.
three dimensionality of girder vibration.
4. System identifications of the bridge using seismic responses
3.3. Seismic responses of the piers and abutment
To obtain more comprehensive information of dynamic character­
Fig. 10. exhibits time-histories and frequency spectra of piers lon­ istics of the structure during different levels of earthquakes, system
gitudinal accelerations recorded in the largest NFE (July 17, 2018). The identifications were employed using recorded seismic responses. Iden­
figure shows that all piers have relatively the same time-histories as well tification of dynamic characteristics was conducted using two types of
as frequency spectra characteristics. The peak accelerations for the piers time-domain system identification methods. The first type is system
in the middle spans were between 44 and 55 cm/s2. Smaller accelera­ identification based on system realization technique. Two state-of-the-
tions were recorded on the end pier (P35). As for the abutment A2, art system identifications based on system realization technique were
except for the spike around 10 s, the amplitude of acceleration was far employed namely, the System Realization using Information Matrix
smaller than that of the piers. The spike in abutment acceleration was (SRIM) [31] and the Observer (Kalman Filter) Identification (OKID) [32]
thought to be caused by pounding with the non-structural or secondary combined with multireference Eigensystem Realization Algorithm
structural element near the sensor node, as it does not appear on the (ERA) [33]. The second type is based on recursive least-square (RLS)
records from other seismic events. For the largest FFE, similar charac­ method using autoregressive exogenous (ARX) model. Brief explana­
teristics can be observed (Fig. 11), where piers in the middle span tions of the methods and their implementations are given in the
generally have the same shape and amplitudes of acceleration responses. following sections.
It is interesting to note that the acceleration spectra of the largest NFE
and the largest FFE have different characteristics. In the case of the 4.1. System-Realization-based methods
largest NFE (July 17, 2018), the pier vibrations are dominated by higher
frequency peaks around 5 Hz. For the largest FFE (July 28, 2019), In both SRIM and OKID-ERA methods, the system matrices con­
however, the dominant peaks are around 1–1.5 Hz. As will be explained taining information of modal parameters are obtained from a state-space

Fig. 8. Comparisons of vertical girder accelerations and their frequency spectra obtained from sensors along the girder in the largest NFE (July 17th,2018). Ac­
celeration responses: (a) Nodes 5,8, 9,10, (b) Nodes 13,16,19. Fourier spectra of accelerations: (c) Nodes 5,8,9,10, and (d) Nodes 13,16,19.

9
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Fig. 9. Comparisons of vertical girder accelerations and their frequency spectra obtained from sensors along the girder in the largest FFE (July 28th,2019). Ac­
celeration responses: (a) Nodes 5,8, 9,10, (b) Nodes 13,16,19. Fourier spectra of accelerations: (c) Nodes 5,8,9,10, and (d) Nodes 13,16,19.

Fig. 10. Comparisons of piers accelerations in longitudinal direction (bridge axis) and their frequency spectra obtained from sensors on the pier caps in the largest
NFE (July 17th,2018). Acceleration responses: (a) Nodes 7,11,12,14, (b) Nodes 15,17,18,20. Fourier spectra of accelerations: (c) Nodes 7,11,12,14, and (d)
Nodes 15,17,18,20.

model using realization algorithm. The methods accommodate identi­ information-matrix is constructed by assigning ground motions as the
fication of multiple-input multiple-output (MIMO) system that suits the inputs and structural responses as the outputs. As explained in [31] the
condition for identification of multi-span highway bridge where the solution for observability matrix can be estimated by factorization using
sources of input excitations are multiple ground motions and the output singular value decomposition (SVD) of information matrix. Afterwards,
being the seismic responses along the bridge. the system matrix A is estimated from the observability matrix. Upon
In the SRIM algorithm, an input–output correlation matrix called an obtaining the system matrix A, natural frequency and damping ratio are

10
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Fig. 11. Comparisons of piers accelerations in longitudinal direction (bridge axis) and their frequency spectra obtained from sensor nodes on the pier caps during the
largest FFE (July 28th,2019). Acceleration responses: (a) Nodes 7,11,12,14, (b) Nodes 15,17,18,20. Fourier spectra of accelerations: (c) Nodes 7,11,12,14, and (d)
Nodes 15,17,18,20.

Fig. 12. Comparisons of piers accelerations in lateral direction (bridge axis) and their frequency spectra obtained from sensor nodes on the pier caps during the
largest NFE (July 17th,2018). Acceleration responses: (a) Nodes 7,11,12,14, (b) Nodes 15,17,18,20. Fourier spectra of accelerations: (c) Nodes 7,11,12,14, and (d)
Nodes 15,17,18,20.

identified from eigenvalues of matrix A, while mode shapes matrices are Observer (Kalman Filter) Identification (OKID) [32]. Using the impulse
estimated from eigenvectors of matrix A and the output-transformation response functions, the blocks of Hankel matrix are constructed. After­
matrix C. wards, the procedure is continued with factorization of Hankel matrix
OKID-ERA system identification starts by recovering the impulse using singular value decomposition to obtain the extended observability
response functions using input–output data by a procedure named matrix and the extended controllability matrix. Similar to the procedure

11
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Fig. 13. Comparisons of piers accelerations in lateral direction (bridge axis) and their frequency spectra obtained from sensor nodes on the pier caps during the
largest FFE (July 28th,2019). Acceleration responses: (a) Nodes 7,11,12,14, (b) Nodes 15,17,18,20. Fourier spectra of accelerations: (c) Nodes 7,11,12,14, and (d)
Nodes 15,17,18,20.20.

for SRIM method explained above, the system matrices (A, B, C) were average (MA) coefficients {ai }, {bi }:
estimated from the identified extended observability matrix and
A(q) = 1 + a1 q− 1 + ⋯ + ana q− na
(2.a)
controllability matrix. Finally, modal parameters were extracted by
solving the eigenvalue problem of observability matrix.
B(q) = b1 q− 1 + ⋯ + bnb q− nb
(2.b)
In the above explanation, the sources of input to system are the
ground accelerations. However, in the implementation, the presence of where na is the order of the system output and nb is the order of
traffic-induced vibration on the accelerations especially on the seismic input, and q− j is the backward shift operator defined as q− j y(k) =
responses of girder sensor nodes is unavoidable. Therefore, to avoid y(k − j). The AR and MA coefficients {ai }, {bi } are defined by the input
errors in system identification, the part of the bridge responses that and output sequences z(k) and y(k), where the input is the ground ac­
significantly contain traffic-induced vibration were excluded from the celeration, and the output is the bridge response. Parameters of the
analysis. Furthermore, ground accelerations recorded by sensor nodes 2, model are estimated by the following equation:
3, and 4 were selected as the inputs, while the other 17 sensor nodes on
piers and girder were used as the outputs. Since all sensor nodes are θ(N) = R(N)− 1 Ψ(N) (3)
triaxial accelerometer, in an ideal case it results in a 9-input and 51- where
output MIMO system. However, it should be mentioned that a number
of sensors failed to function in several recorded earthquakes. In this θ(N) = [ a1 , a2 , ⋯ana , b1 , b2 , ⋯bnb ]T (4)
situation, the failed sensor nodes were removed and number of input or
output in the MIMO system was adjusted. Since both methods are linear 1 ∑N N− k
R(N) = λ ϕ(k)ϕT (k) (5)
system identifications, modal parameters are assumed to be constant in N k=1
the specific time-window where data is sampled. In the analysis only 60
1 ∑N N− k
s of input and output data containing the peak excitation was utilized to Ψ(N) = λ y(k)ϕT (k) (6)
N k=1
generate modal parameters. More details on the applications of the
methods and case-studies of modal parameter identification of long-span In the above equations, λ is the forgetting factor coefficient, and ϕ(k)
bridges using the above methods are available in previous work [34]. is given by:

ϕ(k) = [ − y(k − 1), ⋯, − y(k − na), z(k − 1), ⋯, z(k − nb)]T (7)
4.2. System identification by autoregressive model with exogenous input
(ARX Model) The natural frequencies ωj and damping ratios ξj are then estimated
from the modulus rj and argument pj of the polynomial function’s poles.
The second type of system identification is recursive least-square In this study, responses from the girder sensors (node 8 and 9) and
(RLS) using autoregressive exogenous (ARX) model. Brief description pier sensors (node 11, 14 and 17) were assigned as the outputs, while
of the method is given here, while detailed information of the method is accelerations from ground sensors (node 2,3, and 4) were selected as the
available in [35,36]. Consider an ARX model written as: inputs. Consequently, a number of separate systems were utilized for
piers and girder in the longitudinal (bridge axis) and lateral directions.
A(q)y(k) = B(q)z(k) + w(k) (1)
Note that while the RLS-ARX method accommodates time-variant
In Eq.(1), y(k), z(k) and w(k) denote the output, input, and white identification of modal parameters, in this study the average values of
noise, respectively. Next, A(q) and B(q) are defined to characterize modal parameters obtained during the duration of peak excitation were
polynomial functions consisting of autoregressive (AR) and moving selected as the representative values of modal parameters. This process

12
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

is considered reasonable since the change in modal parameters as the vertical lines or appear in most of the model orders. As shown in Fig. 14
result of nonlinearity effect is expected to be insignificant within the (a), the first natural frequency is evidently identified at 1.45 Hz, and the
selected time frame for the relatively modest ground excitations. In the second one is at slightly higher frequency 1.65 Hz. Both frequencies
calculation, λ = 0.99 and 5 s delay response delay were selected for the stabilize at smaller model orders of 20. At 3.85 Hz, the third frequency
identification. Note that the smaller the lambda value is, the sharper is appears and stabilizes at larger model order of around 40. The fourth
the decay of the weighting factor. In this study we use λ = 0.99 as and fifth frequency at 4.45 Hz and 5.25 Hz also stabilize at larger model
suggested in [37,38]. Delay parameters were selected as 5 s to obtain orders, around 60 and 80, respectively. Whereas the other three modes
more stable results. Consistency of the identification results depends on (i.e., 5.6 Hz, 6.45 and 7.3 Hz) stabilize at even larger model orders over
stationarity of parameter conditions. In our observation, the delay factor 100 indicating weak contribution of the modes on the overall bridge
5 s gives relatively stable results. Moreover, since the objective is to response. Note that the normalized transfer function plots are included
identify modal parameters during the main excitation, only responses in the stabilization diagram. In the transfer functions responses from
from the largest part of the records were used. sensor node 9 on the girder and sensor node 14 on the pier in longitu­
dinal and lateral direction are used as the outputs, while the ground
motions of nodes 2 and 3 are used as input. The figure clearly shows that
4.3. Results of system identification and Comparisons the dominant peaks in the transfer function plots generally coincide with
the stabilized poles signifying the correlation between the real modes
In implementation of SRIM and OKID-ERA algorithms, stabilization identified by system identifications and the ones indicated by transfer
diagrams were used to assist selection of the modes. The diagram is an functions.
instrument to differentiate the spurious modes from the structural The system identification methods were applied to all earthquakes in
(physical) vibration modes [39]. Fig. 14(a) shows an example of the Table 2, Generally, one input dataset generated eight modes between
diagram used for SRIM identification for responses due to the largest 1.4 Hz and 8 Hz, where the mode shapes were dominated mostly by the
earthquake (EQ1). Horizontal axis of the diagram denotes the identified girder modal displacements. The lowest mode identified was the girder
frequencies, and vertical axis represents model order. The poles (modes) first lateral mode at 1.45 Hz, after that the first longitudinal mode at
associated with a specific model order are examined and assessed by 1.65 Hz. Fig. 14(b) and(c) exhibit the typical shapes of the first and the
comparing them with the poles of a lower-order model. With the in­ second mode, respectively. There were also several modes dominated by
crease of model order, the physical poles (modes) at identical frequency modal displacement of piers in longitudinal and lateral direction
are evaluated. The stable poles are selected as the ones that form stable

Fig. 14. (a) Stabilization diagram generated using SRIM system identification for the largest near-field earthquake (EQ1). Note comparison between transfer
functions and the stable poles with respect to model order. Sensors used for transfer functions are node 2, 3 on the ground and nodes 9 and 14 on the bridge. Examples
of mode-shapes identified by SRIM (b) 1st mode (c) 2nd mode.

13
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

identified at higher frequencies (i.e., between 4 and 6 Hz). This is un­ by this type of isolation bearing is naturally quite limited. It should be
derstandable because the piers are stiffer than girder in both longitu­ mentioned that unlike the natural frequencies and damping ratios, the
dinal and lateral direction. Table 3. provides the list of modes identified mode-shapes do not show significant changes with respect to the
from the largest earthquake (EQ1) on July 17, 2018, and the smallest amplitude earthquake excitation. For this reason, they are not discussed
earthquake (EQ40) on April 18, 2020. The table shows that for the same in detailed in this paper.
earthquake, natural frequencies identified by SRIM, OKID-ERA and ARX
methods are nearly identical, with the differences less than 5 %. 5. Performance assessment of isolation bearing using seismic
Damping ratios, however, show higher differences that can reach as high responses
as 150 %. In general, natural frequency of the same mode was identified
at slightly lower frequency in the EQ1 than in the EQ40, meanwhile 5.1. Locking of isolation bearing
damping ratios of the same modes were identified at higher values in the
EQ1 than in the EQ40. In the case of seismically isolated multi-span highway bridges, pre­
Fig. 15 exhibits variations of the identified natural frequencies vious studies in Japan using data from seismic monitoring systems of
plotted with respect to root mean square (RMS) of ground acceleration Matsunohama bridge, Yamaage bridge, Onneto bridge, and Yamada
for all earthquakes in Table 2. The figure demonstrates that natural bridge revealed that structural details of isolation bearing such as steel
frequencies identified from all earthquakes decrease slightly as ground bearing, side stoppers, and upper or lower bearing plates significantly
motions get stronger (i.e., higher RMS at sensor node 2). The decreases affect behavior of seismically isolated bridges during earthquakes
vary among the modes, but they are all well within 10 %. The same trend [42–45]. In the case of Onneto bridge [43,44] and Yamada bridge [45].,
is apparent for all modes identified by the three system identifications. for example, the steel plate girders became locked with the lateral side-
The reduction of identified natural frequencies can be explained as the stopper on several bearings causing the deck not isolated completely
consequence of the change of support condition, which in this case are from the pier. The locking of isolation bearings at one or more pier
the NRB and sliding bearings. At higher excitations, the isolation bearing during a large earthquake would result in an unexpected load redistri­
starts to engage and the structure becomes more flexible thus resulting bution to the substructure because significant inertia force of super­
in the reduction of natural frequencies. structure will be transmitted to the substructure. This phenomenon has
Fig. 16 shows the identified damping ratios from forty earthquake been observed in many more instances other than the ones reported in
events listed in Table 2. It can be observed that variations of damping are the literature above but not yet anticipated in design and therefore
higher than that of natural frequencies within the same mode. In general should be identified as early as possible and mitigated to avoid further
damping ratios were identified between 0.4 % and 4 %, however, in malfunction of the isolation bearings at larger earthquake [44].
most cases the damping ratios were within 1–2 %. The amount of Modal parameters estimated by system identifications explained
damping identified in this study is relatively smaller compared to other above are normally utilized for characterization of global structural
type of highway bridges. The source of damping in isolated bridges condition. Evaluation of a local structural condition such as bearing
primarily comes from dissipative mechanism provided by the isolation using only the changes of modal parameters, however, is rather
bearings. In this bridge, the isolation bearings are made of natural complicated. This is because other factors like temperature in addition
rubber bearing (NRB) which is known to have smaller damping than the to bearing condition can cause changes in modal parameters. Therefore,
high damping rubber (HDR) or lead rubber bearing (LRB). Nevertheless, it is more desirable to evaluate isolation bearing performance from vi­
the range of identified damping ratio is consistent with the experiments bration characteristics of local structural elements near the bearing,
conducted on natural rubber bearing commonly used for bridge in Japan which in this case is the seismic response of girder and pier. In a con­
[40,41]. Therefore, it can be reasonably argued that the relatively ventional non-isolated bridge, seismic responses of the girder typically
smaller damping in this highway bridge is caused by the type of isolation comprise of high-frequency components originated from the piers
bearings that does not provide damping as high as in the lead-rubber or because of rigid connection between pier and girder. In a seismically
high-damping rubber bearings. The figure exhibits slight increase in isolated bridge, when isolation system functions properly, the high-
damping ratios as the ground motion becomes stronger (i.e., higher RMS frequency components would be filtered out of the girder’s seismic
of sensor node 2). Again, a possible explanation for the increase of response by isolation bearing in a mechanism that decouples piers and
damping is the change in the bearing conditions. It can be reasonably girder at the connection. The phenomenon commonly identified as the
inferred that as the isolation bearings start to engage at higher excita­ high-frequency filtering effect thus can be utilized as an indicator of
tion, the structure becomes more flexible, and contribution of dissipa­ isolation bearing performance during an earthquake.
tive energy mechanism provided by NRB and sliding bearing on the Elastomeric bearing used in isolated bridge is normally equipped
overall structural damping increases. The increase, however, is not with side stopper made of welded or stiffened steel plates. Normally, a 5-
substantially high as the total amount of damping that can be provided mm gap exist between isolator bearing and side stoppers for relative

Table 3
List of modes identified from the largest (EQ1) July 17, 2018, and the smallest (EQ40) April 18, 2020, earthquake identified by the system identifications SRIM, OKID-
ERA and AR.
Mode Type Largest Earthquake (EQ1) July 17, 2018, Smallest Earthquake (EQ40) April 18, 2020,
Natural Freq. [Hz] Damp. Ratio [%] Natural Freq. [Hz] Damp. Ratio [%]
SRIM OKID- ARX SRIM OKID- ARX SRIM OKID- ARX SRIM OKID- ARX
ERA ERA ERA ERA

1 1st Girder Lateral direction 1.45 1.47 1.45 2.53 2.45 2.31 1.43 1.43 1.49 1.08 1.08 1.06
2 1st Girder Longitudinal direction 1.64 1.64 1.66 1.74 2.03 2.46 1.63 1.62 1.68 1.25 1.17 0.85
3 2nd Girder Longitudinal direction 3.84 3.83 3.79 4.20 3.25 3.30 3.87 3.89 3.83 0.48 0.74 0.99
4 Pier Combination #1 in Longitudinal 4.45 4.36 4.40 3.00 1.68 2.49 4.71 4.73 4.53 0.48 0.64 0.86
direction
5 2nd Girder lateral direction 5.30 5.23 5.28 1.15 1.15 2.01 5.43 5.46 5.44 0.51 0.75 0.89
6 Pier Combination #1 in Lateral direction 5.55 5.67 5.69 2.85 1.66 3.12 5.79 5.73 5.87 0.87 1.17 1.08
7 Pier Combination #2 in Longitudinal 6.45 6.44 6.44 2.22 2.22 1.78 6.76 6.72 6.53 1.01 0.70 0.92
direction
8 3rd Girder Longitudinal direction 7.28 7.26 7.31 1.13 1.61 2.78 7.36 7.36 7.44 0.81 0.45 0.93

14
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Fig. 15. The trends of natural frequencies against root mean square (RMS) of ground accelerations identified from 40 selected earthquakes by three system
identifications.

Fig. 16. The trends of damping ratios against root mean square (RMS) of ground accelerations identified from 40 selected earthquakes by three system
identifications.

lateral displacements between girder and isolator. There were cases, structure that has higher fundamental natural frequency in the longi­
however, where the gap was sealed causing the side stopper to stick with tudinal direction. The increase in the bridge fundamental frequency in
bearing’s upper steel plate. Such conditions may exist due to seismic longitudinal direction would depend on how many piers have the locked
pounding, poor maintenance, corrosion, or misalignment. When it oc­ bearings. Obtaining information on the location and the number of
curs, friction force between the contacting surfaces would obstruct the locked bearings directly from modal parameters (i.e., natural frequency,
movement of upper steel plate. The initial stiffness of girder-pier damping and mode-shape) is challenging because of its locality effect.
connection would become higher because stiffness of the side-stopper The locked bearing condition would typically result in a semi-rigid pier-
is higher than the elastomeric bearing stiffness. Consequently, the girder connection that is closer to the one in a conventional non-isolated
isolation system would not work properly as intended until seismic force bridge [44,45]. In such condition, girder and pier will move together
can overcome the friction force. Isolation bearing in this condition is and seismic response of the pier with locked bearing would be domi­
defined categorically as a malfunction bearing. nated by girder’s frequency which is lower than pier’s frequency. In
In a typical bridge with laminated rubber bearing , the initial stiff­ contrast, when a pier has a normally functioning isolation bearing,
ness of the bearing is in the range of 15000–16000 kN/m with the sec­ girder and pier would have different seismic response characteristics.
ondary stiffness about 40–45 % of the primary ones [45]. However, The pier responses would be dominated by higher frequency content
when the bearing is locked, primary stiffness and secondary stiffness can associated with its own frequency, while girder response would be
reach to about 196,000 kN/m and 49000kN/m, respectively. These dominated by its own longitudinal frequency which is lower than the
values were estimated from the stiffness of the side-stopper and friction pier’s frequency.
force between side-stopper and isolator’s upper steel flange. The in­ In this study, with the premise of condition explained above, eval­
crease in the primary stiffness of the bearing would result in a stiffer uation of isolation bearing condition is conducted using time–frequency

15
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

analysis of seismic responses recorded on the girder and piers. The an­ domain for which the ts (a,b) is equal to b. The curve that is a function of
alyses were conducted using continuous wavelet transform (CWT) via b (e.g. a = f(b)) is a ridge of the wavelet transform.
complex Morlet wavelet [46]. The complex Morlet wavelet is described Evolution of vibration characteristics throughout the excitation can
as: be traced from the ridge of wavelet transform from which the instan­
taneous frequency (IF) of signal is estimated. The IF indicates the main
1 ( ) t2
(8) concentration of vibrational energy in the time-scale plane of energy
2
ψ(t) = √̅̅̅̅̅̅̅ ei2πf c t − e− f b (πf c ) e− fb
πf b
density and thus dominates vibration characteristics of the signal at a
where fb denotes the bandwidth parameter, and fc denotes the specific time. The wavelet ridge is obtained using an amplitude-based
√̅̅̅̅̅
mother wavelet central frequency. As explained in [47], when f b f c ≥ ridge extraction method, where the local maxima along the scale axis
√̅̅̅ 2 of the magnitude of wavelet transform is identified via optimization
2 the term e− f b (πf c ) becomes negligible, and the approximate version
[48].
of the modified Morlet wavelet can be expressed as:
1 t2
5.2. Results on observation on possible locked bearing incident using
ψ(t) = √̅̅̅̅̅̅̅ei2πf c t e− fb
(9)
πf b monitoring data
The dilated version of the Fourier transform of the modified Morlet
wavelet becomes: Fig. 17 shows the acceleration time–frequency maps recorded on the
piers and the girder during the largest NFE (July 17th, 2018), (EQ1). The
(10)
2
dotted lines in the figure represent the IF line explained above. Note that
π2 f fc )
F[ψ(t) ] = e− b (af−

sensor node 6 on P35 was failed during this earthquake, thus only ac­
Equation (10) shows that the relation between scale a and frequency
celerations from the remaining four piers are depicted in the figure. It
f is given by fc/a that is the localized Fourier frequency obtained when
should be mentioned that characteristics of time–frequency of longitu­
F[ψ(t) ] reached the maximum. The time and frequency resolution of
dinal accelerations obtained from all seven sensor nodes on the girder
wavelet at frequency fi can be determined by the Heisenberg uncertainty
are identical since the girder moves as a continuous rigid span. Fig. 17
principle. Additionally, the end effects that can significantly influence
(a) and (b) illustrate two sensor nodes along the girder (i.e., node 10 and
the quality of wavelet coefficient is estimated by setting the time interval
√̅̅̅ node 13 above P38 and P41, respectively). The figures clearly exhibit
Δt = β c 2fi b where the value of β is taken as [47].
f f
similar characteristics of time–frequency responses where the IF around
The ridge of wavelet transform is related to instantaneous frequency 1.65 Hz corresponding to the girder’s first longitudinal mode dominates
of the signal. A continuous wavelet transform can be represented as the girder accelerations from beginning until the end of excitation. On
scalar product [48]: contrary, pier accelerations exhibit different time–frequency charac­
[ ( )] teristics. During peak excitation that is between t = 10 s and 25 s, the IF
1
∫∞ (
t− b
) i ϕ (t)− ϕ t− b
x ψ a
values fluctuate around 4–8 Hz on pier P38 (node 11), pier P41 (node
W(a, b) ≅ Λx (t)Λψ e dt (11) 14), pier P44 (node 17), and 4–8 Hz on abutment A2 (node 20). Two
2a − ∞ a
separate peaks can be observed from the piers’ time–frequency map,
in which Λ and ϕ are the envelope and instantaneous phase of the namely at 4.45 Hz and between 6 and 8 Hz. These two frequency ranges
asymptotic signal x(t) and wavelet ψ(t), respectively. Note that as correspond to the pier’s first and second combination mode in longitu­
pointed out by [48,49], the main contribution to Equation (11) is pro­ dinal direction (see Table 3). After t = 25 s, the IF values change to about
vided by a set of stationary points t = ts of the function: 1.65 Hz that corresponds to girder’s longitudinal mode. This finding
( ) suggests that at the peak excitation (t = 10–25 s), seismic responses of
t− b
Φ(a, b)(t) = ϕx (t) − ϕψ (12) girder and piers were mainly influenced by their individual resonant
a
frequencies indicating that they completely acted as two separate
where the stationary points can be determined from equation: (decoupled) structural systems. In such circumstance, the girder was
( ) isolated from the piers and high frequency components of piers accel­
1 ts − b
Φ̇(a, b)(ts ) = ϕ̇ψ (13) erations were not transmitted to girder acceleration. It describes a
a a
normal and anticipated condition of a fully operational isolation
The stationary points ts are the set of points in time-scale (a,b) bearing. The same condition was noted on the seismic responses of piers
and girder obtained from the other large earthquakes (PGA > 100 cm/

Fig. 17. Time-frequency maps and IF values (dotted red lines) of longitudinal accelerations due to EQ1 (July 17th,2018): (a) on girder node 10, (b) on girder node
13, (c) on P38 node 11, (d) on P41 node 14, (e) on P44 node 17, and (f) on abutment A2 node 20. (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

16
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

s2) such as EQ2, EQ3 and EQ4. Their responses are not shown in this responses. We can classify isolation bearing performance as movable or
paper because of space limitation. functioning bearing and as a malfunction or locked bearing based on the
Example of the girder and piers responses for earthquake EQ5 with patterns of IF values. In a case of a normally functioning bearing, the IF
PGA less than 100 cm/s2 are shown in detailed by Fig. 18. Throughout values of pier acceleration at the peak excitation are nearly identical
the entire excitation, the IF values of the girder acceleration are rela­ with pier’s fundamental frequency. Meanwhile, when the bearing is
tively stable value around 1.65 Hz. On the other hand, the IF values of locked or malfunction, the IF values of pier acceleration become closer
piers and abutment reveal two frequencies: the higher one during peak to the girder’s fundamental frequency.
excitation and the lower one near the end. During peak excitation (t = Classification by k-means clustering technique [50] is conducted to
10–25 s), the IF values fluctuate between 4 and 8 Hz on pier P35 (node give a more systematic evaluation of bearing condition. In this method,
6), pier P38 (node 11), pier P41 (node14), pier P44 (node 17), and be­ the IF values are grouped automatically using clustering technique, and
tween 6 and 10 Hz on abutment A2 (node 20). Again, time–frequency the distance between centroids of the IF clusters observed during peak
maps and IF results indicate that girder was decoupled from the piers excitation is normalized to the absolute difference between girder and
during peak excitation, a condition expected in a functioning isolation pier’s fundamental frequencies. The normalized distance index (NDI)
bearing. between the clusters of IF values is defined as:
Next, Fig. 19 describes typical responses of small earthquake (PGA ⃒ ⃒
⃒(Uc1 − Uc2 ) ⃒
less than 10 cm/s2) obtained from April 18, 2020 (EQ40) with PGA 2.73 NDI(Uc1 , Uc2 ) = ⃒⃒ ⃒ (14)
cm/s2. Note that similar to the case of large earthquake, single frequency ωp − ωg ⃒
around 1.7 Hz, slightly higher than that of the large earthquakes, In Equation (14), Uc1 and Uc2 are the centroids of IF values for cluster
dominates the IF values of girder longitudinal acceleration throughout 1 and cluster 2, respectively, whereas ωp and ωg are the fundamental
the excitation. Likewise, single frequency around 1.7 Hz also controls frequencies of the pier and girder, respectively. The fundamental fre­
the IF values of all piers’ longitudinal accelerations. One can observe quencies can be obtained from system identification or finite element of
that the dominating IF values of piers accelerations do not vary the structure. In this study, information from system identification is
throughout the excitation in the same way as the one observed in large used. It can be expected from Equation (14) that NDI value should be
earthquakes. The fact that both piers and girder accelerations are closer to one for a normally functioning bearing because the centroid
dominated by identical frequency suggests that they behave as a coupled distance between IF clusters should also be closer to the difference be­
structural system. One possible explanation for this result is that during tween the fundamental frequencies of girder and piers. On the contrary,
small excitation such as the one in the EQ40, the isolation system is still when the bearing is locked or malfunction, the NDI value would be
at initial state. A certain level of excitation is needed to mobilize girder closer to zero because the distance between IF clusters’ centroid would
and engage the isolation mechanisms, and an even larger excitation is become considerably smaller compared to the difference between
needed to reach the critical stage of yielding. This means that even fundamental frequencies of girder and piers.
though girder and piers behave as a coupled structural system in case of Fig. 20 summarizes the outcomes of classification using IF values
small excitation level, it should not be directly considered as a case of obtained via CWT. In this figure the results are shown in a heat map
locked or malfunction bearing. However, if the observation of IF values table, where the abscissa denotes the earthquake number sorted ac­
indicate a coupled structural system under large earthquake excitation, cording to PGA from the largest to the smallest. The ordinate shows the
then the result can be viewed as an indication of bearing locking or pier number, and the value along with color gradation in the table de­
malfunction. The same condition was noted on the seismic responses of scribes the normalized distance index (NDI) computed by Equation (14).
piers and girder obtained from the other small earthquakes (PGA less The outcomes shown in the figure are from 40 earthquakes listed in
than 5 cm/s2) such as EQ37, EQ38 and EQ39. In these earthquakes, Table 2. In general, the table demonstrates that the NDIs are closer to
single frequency around 1.7 Hz controls the IF values of girder and all one in all cases of large and moderate earthquakes suggesting that the
piers’ longitudinal accelerations and the dominating IF values remain piers have normally functioning bearings. The table shows that from
largely the same without significant variation throughout the excitation. EQ1 to EQ25 the NDIs are>0.5 in all piers. For the rather smaller
Because of the space limitation, their responses are not shown in this earthquakes with PGA ≤ 10 cm/s2 (EQ25-EQ37), some piers have NDIs
paper. less than 0.5 and some are>0.5. Moreover, for far smaller earthquakes
The foregoing explanation suggests that indication on isolation with PGA less than 5 cm/s2 (EQ38-40), majority of the piers have NDIs
bearing performance during an earthquake can be inferred from eval­ less than 0.2 indicating a coupled structural system of girder and piers.
uation of IF values derived from CWT of piers and girder seismic NDI 0.2 means that the ratio the ratio between the distance of the

Fig. 18. Time-frequency maps and IF values (dotted red lines) of longitudinal accelerations due to EQ5 (March 30th,2018): (a) on girder node 13, (b) on P35 node 6,
(c) on P38 node 11, (d) on P41 node 14, (e) on P44 node 17, and (f) on abutment A2 node 20. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.)

17
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

Fig. 19. Time-frequency maps and IF values (dotted red lines) of longitudinal accelerations due to EQ40 (April 18, 2020): (a) on girder node 13, (b) on P35 node 6,
(c) on P38 node 11, (d) on P41 node 14, (e) on P44 node 17, and (f) on abutment A2 node 20. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.)

Fig. 20. Classification of isolation bearings condition on the piers (P35-P44) and on the abutment A2 for 40 selected earthquakes based on k-means clustering
method. (a). EQ1-EQ20, (b) EQ21-EQ40. Note ND is the normalized distance index.

clusters and the difference between pier and girder’s main frequency is earthquakes, the isolation bearings on most piers have functioned well
20 % as denoted by Equation (14). Considering the girder’s frequency in as intended and there was no case where the NDIs are less than 0.2. As
a completely separated system is about 1.7 Hz and pier’s main frequency for the smaller earthquakes, while the outcomes suggest a possibility of
is about 4 Hz, then the difference between piers and girder’s main fre­ malfunction or locked bearing, it could be explained as a condition when
quency is 2.3 Hz. Therefore, the distance of the two IF clusters when the the girder-pier connection still functions as a coupled structural system
NDI equals to 0.2 is 0.46 Hz. This amount of frequency difference is because the amplitude of excitation is too small to initiate the isolation
reasonable for the increase of girder’s frequency in the case of locked mechanism. Using this table, information and quick assessment of
bearing, that is when the girder’s frequency becomes 2.16 Hz. isolation bearing condition can be conducted immediately after an
The table evidently shows that in most cases of recorded large earthquake. In the case where the NDI value in one pier is closer to zero

18
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

during a large earthquake with PGA > 100 cm/s2, then there is a pos­ cost competitiveness of a WSN-based semi-permanent continuous
sibility that the bearing does not function properly. Further detailed seismic monitoring. Total cost of structural monitoring generally varies
investigation or site visit can be performed to confirm the observation according to the monitoring strategy, period, and objective. A perma­
from the seismic records. nent long-term seismic monitoring of a large bridge normally lasts for
It should be mentioned that the data from 45-month monitoring are 20–30 years. Such a monitoring system in Japan typically utilizes high-
mostly from small and moderate earthquake, therefore there has been no accuracy servo-type wired accelerometers where the total cost including
data available showing that the locked bearing can or cannot occur installation is about 27,000–36,000USD (approximately 3–4 million
during large earthquake. However, based on the experiences from yen) per unit sensor. The cost of cabling and maintenance is estimated
similar same type and size of multi-span seismically isolated highway about 1800USD (approx. 0.2 million yen) per sensor node. A similar or
bridges like the Onneto bridge [43,44] and the Yamada bridge [45], system with lesser sensor specification is typically deployed for a semi-
there is a possibility that locked bearing on one or more piers occur permanent monitoring in a shorter period of 3–5 years.
during a large earthquake. Both bridges have experienced locked A WSN-based monitoring system is an excellent alternative for a
bearing condition during large earthquakes in the past, that is during the continuous semi-permanent monitoring of a 3-to-5-year period. The
April 29, 1995, on Onneto bridge and the 2011 Tohoku earthquake on initial costs of WSN used in this study are 2300 USD (approx. 0.25
Yamada bridge. Therefore, it is reasonable to assume that there is a million yen) and 4600 USD (approx.0.5 million yen) per sensor node and
possibility of locked bearing occurring during a large earthquake in the sink node, respectively, without any additional expense for cabling. The
future considering that Katsuta bridge, which is a typical highway only additional expense is for replacement of batteries once a year. It can
bridge in Japan, has the same type, size, and structural configuration as be concluded that the WSN-based monitoring system used in this study
the Onneto bridge and Yamada bridge. Detecting the future occurrence would require only 20 % of the initial cost typically needed for servo-
of locked bearing during large earthquake and identifying the underly­ type sensors with wired monitoring system. The significant reduction
ing cause directly from seismic monitoring data is important for quick in cost is achieved with relatively similar reliability and monitoring
seismic assessment of isolated bridges. The method presented above is performance. Therefore, the WSN-based monitoring system can be
expected to facilitate the detection and quick assessment of bearing considered as a promising option for semi-permanent monitoring system
condition. of a large bridge. The paper describes an example of the use of seismic
monitoring data for assessment of isolation bearing condition. In the
6. Conclusions future, WSN seismic monitoring platform developed in this research can
be extended and improved for other diagnostic systems using seismic
Long-term continuous seismic monitoring of multi-span highway responses of the structure.
bridge by WSN system is described in this paper. The monitoring system Data availability statement.
consists of 20 nodes of triaxial wireless accelerometers installed on The seismic monitoring data used in this paper are proprietary of the
numerous locations on the girder, piers, abutments, and on the ground. authors and may only be provided upon reasonable request to the cor­
During the 45-month continuous monitoring period, the WSN system responding author.
has successfully recorded 63 earthquakes of small and moderate levels.
The recoded seismic responses were analyzed by comparison with CRediT authorship contribution statement
existing seismic monitoring system, system identifications, and time­
–frequency analysis. The following conclusions are obtained: Dionysius M. Siringoringo: Conceptualization, Methodology,
Formal analysis, Investigation, Data curation, Validation, Writing - re­
1. The WSN monitoring system has proved to be effective in recording view & editing. Yozo Fujino: Conceptualization, Resources, Supervi­
seismic responses from the near-field as well as the far-field earth­ sion, Funding acquisition, Project administration. Makoto Suzuki:
quakes with excellent quality. Accuracy, robustness, and reliability Software, Data curation, Resources, Conceptualization, Validation.
of WSN monitoring system were confirmed by comparing the ground
motions recorded by WSN monitoring system with the existing Declaration of Competing Interest
strong-motion seismograph KNET network.
2. Modal parameters of the bridge of were successfully estimated with The authors declare that they have no known competing financial
excellent accuracy and consistency by time-domain system identifi­ interests or personal relationships that could have appeared to influence
cations using high quality seismic responses recorded by WSN. The the work reported in this paper.
outcomes of modal parameters were analyzed with respect to
amplitude of seismic excitation and some trends on dependence of Data availability
modal parameters on the amplitude of excitation were observed.
3. Seismic performance evaluation of isolation bearings was conducted Data will be made available on request.
by continuous wavelets transform analysis using information ob­
tained from instantaneous frequency of girder and the piers longi­ Acknowledgements
tudinal accelerations. High-frequency filtering effect was utilized as
indicator of isolation bearing performance. Systematic classification We gratefully acknowledge the support and cooperation given by the
by clustering algorithm was performed to determine whether the East Nippon Expressway Co., Ltd. (East NEXCO) Japan in this research.
bearings have functioned normally. The analyses revealed that The study is part of a JSPS-supported research to the first author under
isolation bearings have functioned properly during moderate earth­ the Grant-in-Aid Kakenhi C No. 18K04320 (PI: Dionysius M. Siringor­
quakes (PGA > 20 cm/s2). Possibilities of bearing malfunction or ingo). Findings and statements in the paper are of the authors and not of
locked bearings on some piers were observed during smaller earth­ the abovementioned institutions.
quakes (PGA less than 5 cm/s2). These outcomes can be explained as
a condition where the girder and pier are still functioning as a References
coupled structural system because the amplitude of earthquake is not
strong enough to initiate the isolation mechanism. [1] Ganev T, Yamazaki F, Ishizaki H, Kitazawa M. Response analysis of the Higashi-
Kobe bridge and surrounding soil in the 1995 Hyogoken-Nanbu earthquake. Earthq
Eng Struct Dyn 1998;27(6):557–76.
In addition to analysis of the recorded seismic responses, the expe­ [2] Boroschek RL, Moroni MO, Sarrazin M. Dynamic characteristics of a long span
rience gained during 45 months of monitoring period reveals the overall seismic isolated bridge. Eng Struct 2003;25(12):1479–90.

19
D.M. Siringoringo et al. Engineering Structures 276 (2023) 115372

[3] Memisoglu Apaydin N, Zulfikar A, Cetindemir O. Structural health monitoring [26] Lynch, J. P., Y. Wang, K. C. Lu, Tsung-Chin Hou, and C. H. Loh. (2006). Post-
systems of long-span bridges in Turkey and lessons learned from experienced seismic damage assessment of steel structures instrumented with self-interrogating
extreme events. J Civ Struct Heal Monit 2022:1–38. wireless sensors. In Proc. 8th Natl. Conf.on Earthq. Eng., 18. 2006.
[4] Çelebi M. Real-time seismic monitoring of the New Cape Girardeau Bridge and [27] Amditis, A., Y. Stratakos, D. Bairaktaris, M. Bimpas, S. Camarinopolos, S.
preliminary analyses of recorded data: an overview. Earthq Spectra 2006;22(3): Frondistou-Yannas, and T. Torfs. (2010). An overview of MEMSCON project: An
609–30. intelligent wireless sensor network for after-earthquake evaluation of concrete
[5] Lekidis V, Tsakiri M, Makra K, Karakostas C, Klimis N, Sous I. Evaluation of buildings. In Proc. 14th Eur. Conf. Earthquake Eng. 2010.
dynamic response and local soil effects of the Evripos cable-stayed bridge using [28] Siringoringo DM, Fujino Y, Suzuki M, Mehta V. Implementation of Wireless Sensor
multi-sensor monitoring systems. Eng Geol 2005;79(1–2):43–59. Network for Continuous Seismic Monitoring of Isolated Cable-Stayed Bridge.
[6] Limongelli MP, Çelebi M, editors. Seismic structural health monitoring: from J Bridg Eng 2022;27(4):04022011.
theory to successful applications. Springer; 2019. [29] Suzuki M, Jinno K, Tashiro Y, Katsumata Y, Liao CH, Nagayama T, et al.
[7] Siringoringo DM, Fujino Y. System identification applied to long-span cable- Development and field experiment of routing-free multi-hop wireless sensor
supported bridges using seismic records. Earthq Eng Struct Dyn 2008;37(3): network for structural monitoring. In Proc ICSIC 2016:179–84.
361–86. [30] Network Center for Earthquake, Tsunami and Volcano (NIED). Accessed July 5,
[8] Fujino Y, Siringoringo DM, Abe M. Japan’s experience on long-span bridges 2022. http://www.kyoshin.bosai.go.jp/kyoshin/quake/index_en.html.
monitoring. Structural Monitoring and Maintenance 2016;3(3):233–57. [31] Juang JN. System realization using information matrix. J Guid Control Dynam
[9] Smyth AW, Pei JS, Masri SF. System identification of the Vincent Thomas 1997;20(3):492–500.
suspension bridge using earthquake records. Earthq Eng Struct Dyn 2003;32(3): [32] Juang JN, Phan M, Horta LG, Longman RW. Identification of observer/Kalman
339–67. filter Markov Parameters-Theory and experiments. J Guid Control Dynam 1993;16
[10] Spencer BF, Jo H, Mechitov KA, Li J, Sim S-H, Kim RE, et al. Recent advances in (2):320–9.
wireless smart sensors for multi-scale monitoring and control of civil [33] Juang JN, Pappa RS. An Eigensystem Realization Algorithm for Modal Parameter
infrastructure. J Civ Struct Heal Monit 2016;6(1):17–41. Identification and Model reduction. J Guid Control Dynam 1985;8(5):620–7.
[11] Lynch JP. An overview of wireless structural health monitoring for civil structures. [34] Siringoringo DM, Fujino Y. Seismic response of a suspension bridge: Insights from
Philosophical Transactions of the Royal Society A: Mathematical, Physical and Eng long-term full-scale seismic monitoring system. Struct Control Health Monit 2018;
Sciences 2007;365(1851):345–72. 25(11):e2252.
[12] Nagayama T, Spencer Jr BF. Structural health monitoring using smart sensors. [35] Safak E. Adaptive modeling, identification, and control of dynamic structural
Newmark Structural Engineering Laboratory: University of Illinois at Urbana- systems. I: theory. Journal of Engineering Mechanics ASCE 1989;115(11):
Champaign; 2007. 2386–405.
[13] Nagayama T, Moinzadeh P, Mechitov K, Ushita M, Makihata N, Ieiri S, et al. [36] Safak E. Adaptive modeling, identification, and control of dynamic structural
Reliable multi-hop communication for structural health monitoring. Smart Struct & systems. II: Applications. Journal of Engineering Mechanics ASCE 1989;115(11):
Systems 2010;6(5–6):481–504. 2406–26.
[14] Zhou G-D, Yi T-H. Recent Developments on Wireless Sensor Networks Technology [37] Ljung L, Soderstrom T. Theory and practice of recursive identification. Cambridge,
for Bridge Health Monitoring. Math Probl Eng 2013;2013:1–33. Mass: MIT Press; 1983.
[15] Chen GW, Beskhyroun S, Omenzetter P. Experimental investigation into amplitude- [38] Young P, editor. Recursive Estimation and Time-Series Analysis. Berlin,
dependent modal properties of an eleven-span motorway bridge. Eng Struct 2016; Heidelberg: Springer Berlin Heidelberg; 1984.
107:80–100. [39] Van der Auweraer H, Peeters B. (2004). Discriminating physical poles from
[16] Jang S, Jo H, Cho S, Mechitov K, Rice JA, Sim S-H, et al. Structural health mathematical poles in high order systems: use and automation of the stabilization
monitoring of a cable-stayed bridge using smart sensor technology: deployment diagram. Proc.of the 21st IEEE, Instrumentation and Measurement Technology
and evaluation. Smart Struct Syst 2010;6(5_6):439–59. Conference, IMTC 2004; 3: 2193– 2198.
[17] Cho S, Jo H, Jang S, Park J, Jung H-J, Yun C-B, et al. Structural health monitoring [40] Unjoh S (2014). Menshin (Seismic Isolation) Bridges in Japan, Technical Note of
of a cable-stayed bridge using wireless smart sensor technology: data analyses. PWRI Japan No.4288.
Smart Struct Syst 2010;6(5_6):461–80. [41] Kawashima K, Nagashima H, Iwasaki H. Evaluation of modal damping ratio based
[18] Lynch JP, Wang Y, Loh KJ, Yi JH, Yun CB. Performance monitoring of the on strain energy proportional damping method. Journal of Structural Engineering
Geumdang Bridge using a dense network of high-resolution wireless sensors. Smart JSCE 1994;40:953–65.
Mater Struct 2006;15(6):1561–75. [42] Chaudhary MTA, Abe M, Fujino Y, Yoshida J. System identification of two base
[19] Nguyen K-D, Kim J-T, Park Y-H. Long-term vibration monitoring of cable-stayed isolated bridges using seismic records. J Struct Eng ASCE 2000;126(10):1187–95.
bridge using wireless sensor network. J of Distributed Sensor Networks 2013;9(11): [43] Chaudhary MTA, Abé M, Fujino Y. Investigation of atypical seismic response of a
804516-1-9. base-isolated bridge. Eng Struct 2002;24(7):945–53.
[20] Kurata M, Kim J, Lynch JP, van der Linden GW, Sedarat H, Thometz E, et al. [44] Chaudhary MTA, Abe M, Fujino Y. Role of structural details in altering the
Internet-enabled wireless structural monitoring systems: development and expected seismic response of base-isolated bridges. Mech Syst Sig Process 2002;16
permanent deployment at the New Carquinez Suspension Bridge. Journal of (2–3):413–28.
structural 2013;139(10):1688–702. [45] Yunsong X, Siringoringo DM, Fujino Y. Condition assessment of seismically isolated
[21] Pakzad SN, Fenves GL. Statistical analysis of vibration modes of a suspension multi-span highway bridge bearings using recorded and simulated seismic
bridge using spatially dense wireless sensor network. J Struct Eng 2009;135(7): responses. Advances in Structural Engineering 2022;25(16):1–17. https://doi.org/
863–72. 10.1177/136943322211331.
[22] Picozzi M, Milkereit C, Zulfikar C, Fleming K, Ditommaso R, Erdik M, et al. [46] Grossmann A, Morlet J. Decomposition of Hardy functions into square integrable
Wireless technologies for the monitoring of strategic civil infrastructures: an wavelets of constant shape. SIAM J Math Anal 1984;15(4):723–36. https://doi.
ambient vibration test on the Fatih Sultan Mehmet Suspension Bridge in Istanbul. org/10.1137/0515056.
Turkey Bulletin of Earthquake Engineering 2010;8(3):671–91. [47] Yan BF, Miyamoto A, Brühwiler E. Wavelet transform-based modal parameter
[23] Weng JH, Loh CH, Lynch JP, Lu KC, Lin PY, Wang Y. Output-only modal identification considering uncertainty. J Sound Vib 2006;291(1–2):285–301.
identification of a cable-stayed bridge using wireless monitoring systems. Eng https://doi.org/10.1016/j.jsv.2005.06.005.
Struct 2008;30(7):1820–30. [48] Tchamitchian P, Torrésani B. Ridge and skeleton extraction from the wavelet
[24] Hoult NA, Fidler PR, Hill PG, Middleton CR. Long-term wireless structural health transform 1990;No. M-CPT-2467.
monitoring of the Ferriby Road Bridge. J Bridg Eng 2010;15(2):153–9. [49] Staszewski WJ, Wallace DM. Wavelet-based frequency response function for time-
[25] Cheng L, Pakzad SN. Agility of wireless sensor networks for earthquake monitoring variant systems—an exploratory study. Mech Syst Sig Process 2014;47(1–2):35–49.
of bridges. In: In 2009 sixth international conference on networked sensing systems https://doi.org/10.1016/j.ymssp.2013.03.011.
(INSS). IEEE; 2009. p. 1–4. [50] Chawla S, Aristides G. In: k-means–: A unified approach to clustering and outlier
detection. Society for Industrial and Applied Mathematics; 2013. p. 189–97.

20

You might also like