sensors

Article
Development of a Reference-Free Indirect Bridge Displacement
Sensing System
Jongbin Won 1 , Jong-Woong Park 1, *, Junyoung Park 1 , Junsik Shin 1 and Minyong Park 2

1 Department of Civil and Environmental Engineering, Chung-Ang University, Dongjak, Seoul 06974, Korea;
[email protected] (J.W.); [email protected] (J.P.); [email protected] (J.S.)
2 Banseok Safety Cooperation, Namyangju-si 12014, Korea; [email protected]
* Correspondence: [email protected]; Tel.: +82-2-820-5278

Abstract: Bridge displacement measurements are important data for assessing the condition of a
bridge. Measuring bridge displacement under moving vehicle loads is helpful for rating the load-
carrying capacity and evaluating the structural health of a bridge. Displacements are conventionally
measured using a linear variable differential transformer (LVDT), which needs stable reference
points and thus prohibits the use of this method for measuring displacements for bridges crossing
sea channels, large rivers, and highways. This paper proposes a reference-free indirect bridge
displacement sensing system using a multichannel sensor board strain and accelerometer with a
commercial wireless sensor platform (Xnode). The indirect displacement estimation method is then
optimized for measuring the structural displacement. The performance of the developed system was
experimentally evaluated on concrete- and steelbox girder bridges. In comparison with the reference
LVDT data, the maximum displacement error for the proposed method was 2.17%. The proposed
method was successfully applied to the displacement monitoring of a tall bridge (height = 20 m),



which was very difficult to monitor using existing systems.



Citation: Won, J.; Park, J.-W.; Park, J.; Keywords: displacement measurement; reference-free displacement; wireless sensor; strain; acceleration
Shin, J.; Park, M. Development of a
Reference-Free Indirect Bridge
Displacement Sensing System.
Sensors 2021, 21, 5647. https:// 1. Introduction
doi.org/10.3390/s21165647
Structural health monitoring (SHM) provides information about the current condition
of a structure, thereby assisting decision-making for operation and maintenance. Con-
Academic Editor: Victor Giurgiutiu
tinuous SHM allows early-stage damage detection, reduces the downtime, and prevents
potential failure during operation. Vertical displacement induced due to traffic loading is
Received: 1 August 2021
Accepted: 18 August 2021
regarded as a crucial measurement in bridge health monitoring because the displacement
Published: 21 August 2021
change is negligible under ambient traffic loads. Ambient displacement monitoring detects
any structural damages or stiffness degradation of a bridge. Traditionally, the displacement
Publisher’s Note: MDPI stays neutral
is measured using a linear variable differential transformer (LVDT), a contact-type sensor
with regard to jurisdictional claims in
that provides accurate displacements. However, the installation of LVDTs to bridges is
published maps and institutional affil- limited, as they must be affixed on a stationary reference [1–3]. Therefore, LVDT measure-
iations. ments are limited to only some points on a structure and are unavailable to bridges built
over waterways or highways. Recently, the computer vision-based method has emerged
as an alternative to the traditional method as it accurately measures the structure using a
camera installed on a remote stationary reference instead of that on a bridge [4–7]. How-
Copyright: © 2021 by the authors.
ever, as in the case for optical sensors, the accuracy of the computer vision-based method
Licensee MDPI, Basel, Switzerland.
is highly affected by lighting conditions and wind or ground vibrations that move the
This article is an open access article
camera. Alternatively, the light detection and ranging (LiDAR)-based system and radar-
distributed under the terms and based displacement measurement systems have been introduced because of robustness to
conditions of the Creative Commons weather conditions [8–12]. However, these systems are vulnerable to self-motion; therefore,
Attribution (CC BY) license (https:// structural displacement measurements of a full-scale bridge are still challenging.
creativecommons.org/licenses/by/ To address the limitations of reference-based methods requiring stationary reference
4.0/). for sensor installation, wireless smart sensors have been developed to measure the struc-

Sensors 2021, 21, 5647. https://doi.org/10.3390/s21165647 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 5647 2 of 15

tural responses and to estimate displacement indirectly. Acceleration-based displacement

estimation for reference-free measurement can be performed through direct integration in
the frequency domain using finite impulse response (FIR) filters. Although FIR filters can
eliminate the error from double integration of acceleration, only dynamic displacement
can be reconstructed [13–15].
In addition to acceleration, strain measurements can be used to estimate reference-free
displacement based on strain-displacement relationships [16–18]. Strain-based displacement
methods can reconstruct pseudo-static displacement, but there is a lack of high-frequency com-
ponents of displacement. Moreover, since the location of a neutral axis cannot be accurately
determined, it is difficult to determine the magnitude of the displacement.
To overcome the issues of strain and acceleration, data-fusion-based, reference-free
indirect displacement estimation was developed to estimate bridge displacements [19].
Reference-free indirect measurements can be useful for bridge ratings because they can easily
estimate the flexural displacement of a bridge at any location. For instance, Park et al. [19]
proposed the use of a wireless sensing system for the estimation of indirect displacement. A
total of five wireless sensor nodes were distributed, each equipped with one strain channel
and three acceleration channels [20,21]. The wireless sensors successfully acquired the
distributed strain and acceleration measurements for the indirect displacement estimates,
but acquiring multichannel data can be challenging in practical implementations because
the bridge superstructure may interfere with the displacement signals [22,23].
This study develops a practical indirect displacement sensing system with multimet-
ric sensing that simultaneously measures three-channel accelerations and three-channel
strains. First, the multimetric board capable of simultaneously observing the three-axis
acceleration and three-channel strains was proposed. Second, to enhance the accuracy of
the displacement estimation, the placements of the three-channel strains were optimized in
a numerical study. The indirect bridge displacement estimation method was optimized for
three distributed strain measurements and one acceleration measurement. The proposed
system was realized using a commercial wireless sensor platform called Xnode [24–26], and
the proposed system’s performance was experimentally validated through full-scale bridge
applications. The estimated displacements were compared with those of a reference LVDT.
The remainder of this paper is organized as follows. Section 2 describes the develop-
ment of a multimetric sensor board and its integration onto the Xnode platform. Section 3
explains the indirect displacements derived from the strain and acceleration measurements
and determines the optimal locations of the three strain sensors through a numerical study.
In Section 4, the proposed method is experimentally validated on concrete- and steel-box
girder bridges and the experimental results are compared with the reference measurements.
The paper concludes with Section 5.

2. Development of a Multimetric Sensing Board

2.1. Design Challenges
A multimetric sensing of acceleration and strain is required for multiple and high-
sensitivity vibration and strain sensing. The accelerometer should be carefully designed to
capture the vibrations not only of lightly damped steel structures but also of highly damped
concrete structures. Multiple small strain responses (<1 us) must be captured on the bridge
surface. The wireless sensor platform on which the developed multimetric sensor board
is integrated should be carefully selected. Therefore, when developing a multimetric
sensor board for wireless sensors, the following three typical design challenges must be
considered: (1) identifying an appropriate wireless sensor platform with high processing
capability and low power consumption; (2) designing a board with a high-resolution
accelerometer and strain sensor with power management; and (3) designing a stable and
low-noise printed circuit board (PCB) for physically integrating the accelerometer with the
wireless sensor platform.
 resolution accelerometer and strain sensor with power management; and (3) designi
stable and low-noise printed circuit board (PCB) for physically integrating the acceler
eter with the wireless sensor platform.

2.2. Xnode
Sensors 2021, 21, 5647 3 of 15
To measure multiple accelerations and strains at high sampling rates (e.g., 100
the base wireless sensor platform requires sufficient processing capability for implem
ing real-time digital filtering and recording. For the wireless sensor platform for m
2.2. Xnode metric sensing, we selected Xnode for its high processing speed and sufficient sync
nous
To measure dynamic
multiple random access
accelerations memory
and strains (SDRAM)
at high sizerates
sampling for real-time
(e.g., 100 data processing.
Hz), the
base wirelessstandard Xnode consists
sensor platform requiresof three main
sufficient boards:capability
processing the processor, radio/power, and se
for implementing
boards
real-time digital (Figure
filtering 1a).recording.
and The processor board
For the was a customized
wireless sensor platformMini4357 developed by Em
for multimet-
Technology
ric sensing, we (Shenzhen,
selected Xnode for itsChina). It has an speed
high processing LPC4357 andmicroprocessor [27] with a 204-M
sufficient synchronous
dynamic randomclock access
speed memory
and 32 MB (SDRAM)
of SDRAM sizefor
fortemporary
real-time data
data processing.
storage andThe stan- Mini
processing.
dard Xnode consists of three main boards: the processor, radio/power, and sensor
has numerous interfaces, including general purpose input/output (GPIO) pins and boards
(Figure 1a). The processor
ripheral boardsuch
interfaces, was as
a customized Mini4357
serial peripheral developed
interfaces and an byinter-integrated
Embest Tech- circuit
nology (Shenzhen, China).
The It has an LPC4357
radio/processor board hasmicroprocessor
a 2.4-GHz Zigbee [27] with
radioafor
204-MHz
low-powerclockwireless c
speed and 32munication
MB of SDRAM (AtmelforAT86RF233)
temporary data [28].storage and processing.
After adding Mini4357
a radio booster, the has
communica
numerous interfaces, including
range reached general
beyond purpose
1 km. input/output
An integrated (GPIO)
circuit pins andand
for charging peripheral
regulation of po
interfaces, such
wasasalso
serial peripheral interfaces and an inter-integrated circuit.
added.

(a) (b)

Figure 1. Xnode smart

Figure sensor:smart
1. Xnode (a) three-board
sensor: (a) stack and (b)stack
three-board enclosure.
and (b) enclosure.

The radio/processor
2.3. Multimetricboard hasBoard
Sensor a 2.4-GHz Zigbee radio for low-power wireless com-
munication (Atmel AT86RF233) [28]. After adding a radio booster, the communication
The configuration of the developed multimetric board is shown in Figure 2.
range reached beyond 1 km. An integrated circuit for charging and regulation of power
board employs a 24-bit analog to digital converter (ADC) (ADS131E08, Texas Instrum
was also added.
[29]) for data acquisition. ADS131E08 is a delta–sigma ADC supporting eight differe
inputs
2.3. Multimetric with
Sensor a high sampling rate (up to 64 ksps). Three out of a total of eight chan
Board
were allocated to high-sensitivity three-axis acceleration measurements by ADXL354
The configuration of the developed multimetric board is shown in Figure 2. The board
and three channels were integrated with a Wheatstone bridge for strain sensing. The
employs a 24-bit analog to digital converter (ADC) (ADS131E08, Texas Instruments [29])
remaining channels were open for external analog voltage sensing. The three-axis a
for data acquisition. ADS131E08 is a delta–sigma ADC supporting eight differential inputs
erometerrate
with a high sampling ADXL354
(up to was selected
64 ksps). forout
Three its low-noise
of a total ofpower
eight density
channelsofwere
20 μg√Hz.
allo- ADX
provides a low noise of 0.16 mg with a bandwidth of 50 Hz under the
cated to high-sensitivity three-axis acceleration measurements by ADXL354 [30], and three ±2 g sensing ra
The ADXL354 was deployed apart from the ADC on the board
channels were integrated with a Wheatstone bridge for strain sensing. The two remaining design to minimize
channels weretemperature effect, otherwise,
open for external the ADC
analog voltage can experience
sensing. an increase
The three-axis in heat due to the
accelerometer
transferred from the ADXL354 at start-up. √
ADXL354 was selected for its low-noise power density of 20 µg Hz . ADXL354 provides
a low noise of 0.16 mg with a bandwidth of 50 Hz under the ±2 g sensing range. The
ADXL354 was deployed apart from the ADC on the board design to minimize the tem-
perature effect, otherwise, the ADC can experience an increase in heat due to the heat
transferred from the ADXL354 at start-up.
A three-channel-strain sensing circuit was designed using a quarter-bridge, which is
a type of Wheatstone bridge (Figure 3). VEXT is the input voltage, RG is the resistance of
the strain gauge, and VA and VB are the voltages measured at points A and B, respectively.
The calibration resistance RC was fixed at 100 kΩ. Depending on RG , R could be switched
between 120 and 350 Ω using a switch.
 STRAIN SENSING CHANNELS
Sensors 2021, 21, 5647 4 of 15
Sensors 2021, 21, 5647 4 of 16

Low-noise
Op.Amp
OPA4344

Power Switch 24-bit ADC 3-AXIS Accelerometer

TPS22860 ADS131E08 ADXL354
STRAIN SENSING CHANNELS
XTAL
SPI
3.3V
Low-noise 3.3V
Op.Amp
Wireless Smart Sensor Platform
OPA4344

Power Switch 24-bit ADC 3-AXIS Accelerometer

Figure 2. Configuration
TPS22860
of the developed multimetric board.
ADS131E08 ADXL354
XTAL
A three-channel-strain sensing SPIcircuit was designed using a quarter-bridge, which is
3.3V of Wheatstone bridge (Figure 3). 𝑉𝐸𝑋𝑇 is the input voltage,
a type 3.3V 𝑅𝐺 is the resistance of
the strain gauge, and 𝑉𝐴 and 𝑉𝐵 are the voltages measured at points A and B, respec-
Wireless Smart Sensor Platform
tively. The calibration resistance 𝑅𝐶 was fixed at 100 kΩ. Depending on 𝑅𝐺 , 𝑅 could be
switched
Figure 2.
Figure between 120 of
2. Configuration
Configuration andthe350
of the Ω usingmultimetric
developed
developed a switch. board.
multimetric board.

A three-channel-strain sensing circuit was designed using a quarter-bridge, which is

a type of Wheatstone bridge (Figure 3). 𝑉𝐸𝑋𝑇 is the input voltage, 𝑅𝐺 is the resistance of
the strain gauge, and 𝑉𝐴 and 𝑉𝐵 are the voltages measured at points A and B, respec-
tively. The calibration resistance 𝑅𝐶 was fixed at 100 kΩ. Depending on 𝑅𝐺 , 𝑅 could be
switched between 120 and 350 Ω using a switch.

Figure
Figure3.3.Wheatstone
Wheatstonebridge.
bridge.

Thevoltage
The voltagedifference
differencebetween
between points
points A
A and
and BB was
was calculated
calculated as
as
𝐺𝐹𝜀 ∆𝑅
∆𝑉 = GFε 𝑉𝐸𝑋𝑇 = 𝐺 𝑉∆R 𝐸𝑋𝑇G, (1)
∆V = 4 VEXT 4𝑅 = VEXT , (1)
4 4R
where 𝐺𝐹 is the gauge factor and ε is the strain. The resistance change in the strain gauge
iswhere
linearly related to thefactor
voltage difference. Using Equation (1) with GF = 2, the voltage
Figure GF is the gauge
3. Wheatstone bridge. and ε is the strain. The resistance change in the strain gauge
corresponding to a strain
is linearly related to the voltage of 1 μs wasUsing
change difference. obtained as 1.65(1)
Equation μV. In the
with GF 24-bit
= 2, the ADC, the
voltage
effective numbertoofa bits
corresponding strainwas 18 and
change ofthe µsprogrammable
1points
wasAobtained gain wasµV.
12;In
as calculated
1.65 therefore, the resolu-
The voltage difference between and B was asthe 24-bit ADC, the
tion of thenumber
effective ADC was calculated
of bits was 18 and as 1.02
the
𝐺𝐹𝜀
μV, affirming
programmable
∆𝑅𝐺
that
gain the
was ADC
12; can measure
therefore, the within 1
resolution
μs. Besides
of the ADCthe was default
calculated ∆𝑉 = µV,strain
strainassensing,
1.02 𝑉affirming
𝐸𝑋𝑇 =calibration
𝑉 𝐸𝑋𝑇
that ,thewasADCenabled
can by
measurea shunt-sensing
within 1(1)
µs.
4 4𝑅
circuit.
Besides
where 𝐺𝐹 Shunt calibration
the default finds
strainfactor
is the gauge the
sensing, conversion
andstrain
ε is the factor
calibration
strain. Thewasbetween the measured
enabled change
resistance by a shunt-sensingvoltages
in the straincircuit.and
gauge
their
Shunt corresponding
calibration
is linearly tostrain
related finds thethe values. The shunt
conversion
voltage can
factorUsing
difference. be
between calibrated (1) by
the measured
Equation with connecting
voltages
GF = 2, the aand
known
their
voltage
resistance
corresponding
corresponding to the tostrain
strain
a straingauge
values. in shunt
The
change parallel,
of 1 μs calculating
canwas
be calibrated
obtained theastheoretical
by μV. In strain
connecting
1.65 a known
the change,
24-bit ADC,and
resistance the
measuring
to the strain
effective the corresponding
gauge
number ofinbits
parallel,
was 18 voltage difference.
calculating
and the Figure 3gain
the theoretical
programmable is a was
straincircuit
12;diagram
change, of the
and measuring
therefore, the resolu-
shuntthe
corresponding
tion of the ADCvoltage difference.
was calculated as Figure
1.02 μV, 3 isaffirming
a circuit diagram of thecan
that the ADC shunt calibration.
measure withinIn1
this figure,
μs. Besides the R denotes the known resistance for calibration. When
C default strain sensing, strain calibration was enabled by a shunt-sensing the calibration resistor
R is
circuit.
C connected with
Shunt calibration R , the resistance changes as follows:
G finds the conversion factor between the measured voltages and
their corresponding strain values. The shunt can be calibrated by connecting a known
R G RC
resistance to the strain gauge in∆R G =
parallel, − RGthe
calculating . theoretical strain change, and (2)
R G + RC
measuring the corresponding voltage difference. Figure 3 is a circuit diagram of the shunt
 Sensors 2021, 21, 5647

Sensors 2021, 21, 5647 5 of 15 When t

calibration. In this figure, 𝑅𝐶 denotes the known resistance for calibration.
ibration resistor 𝑅𝐶 is connected with 𝑅𝐺 , the resistance changes as follows:
𝑅 𝑅
∆𝑅𝐺 = 𝐺 𝐶 − 𝑅𝐺 .
𝑅𝐺 + 𝑅𝐶
Based on the relationship between the GF and the resistance change given by Equation (3),
Based on the relationship between the GF and the resistance change given by
we can compute the corresponding strain change:
tion (3), we can compute the corresponding strain change:
∆RG 𝐺𝐹 = ∆𝑅𝐺.
GF = . 𝜖𝑅𝐺 (3)
eRG
Comparing the actual voltage difference before and after connecting the k
Comparingshunt the actual voltage
resistance, wedifference
obtain thebefore and after
conversion connecting
factor betweenthe theknown
voltageshunt
and strain un
resistance, we obtainNotethe conversion factor between the voltage and strain units.
that the half-bridge can be configured by replacement resistance between
Note that the
D in half-bridge can abedummy
Figure 3 with configured
gauge byfor
replacement
temperature resistance between
compensation to C and self-hea
avoid
D in Figure 3 with a dummy gauge for temperature compensation to avoid self-heating
a strain gauge. The dummy gauge can be installed along an unstrained direction p
of a strain gauge.
dicularThetodummy gauge
the active straincan be installed
gauge along
to cancel out an unstrained
temperature direction
effects.
perpendicular to theTo active strain
handle thegauge
limitedtopower
cancelresources
out temperature
availableeffects.
on the wireless sensor, a powe
To handle agement
the limited power resources available
circuit was developed to control the on thepower
wireless sensor, a The
on demand. powermost powe
management circuit was developed to control the power on demand. The
suming part is strain sensing, where one 120-Ω strain gauge draws 𝑉𝐸𝑋𝑇 / most power-
consuming part 13.8
is strain
mA sensing, where
of current; thus,one 120-Ω
three strain
strain draws VEXT
gaugeconsume
gauges mA.=The
41.3/2R 13.8 mA on the
power
of current; thus, three strain gauges consume 41.3 mA. The power on the multimetric sensor
metric sensor board was controlled by a digital power switch (TPS22860) using GP
board was controlled by a digital power switch (TPS22860) using GPIO on the wireless
the wireless sensor platform. To minimize the power operation, the multimetric
sensor platform. To minimize the power operation, the multimetric sensor board was
board was maintained in deep sleep mode until triggered.
maintained in deep sleep mode until triggered.
Figure 4 shows the developed multimetric sensor board which has an onboard
Figure 4 shows the developed multimetric sensor board which has an onboard strain
gauge connector and three-axis accelerometers on the top side, and a Wheatstone
gauge connector and three-axis accelerometers on the top side, and a Wheatstone bridge
on the bottom side.
on the bottom side.

Strain gauge
(a) connector three-axis
accelerometer

Strain resistance
Wheatstone
switch
bridge
(120 Ω, 350 Ω)

Multimetric
Sensor board

(b)
Radio board

Micro Controller
Unit board

Figure 4.Figure 4. (a) Developed

(a) Developed multimetric
multimetric sensorand
sensor board board
(b) and (b) its integration
its integration with Xnode.
with Xnode.

The noise level The noise

of the level of multimetric
developed the developed multimetric
board board was
was validated via avalidated
laboratory-via a labor
scale
scale evaluation. evaluation.
This This test
test confirmed confirmed
whether the whether
PCB metthe thePCB met theaccelerometer
specified specified acceleromet
strain performances.
and strain performances. The resistance
The resistance and GF of and GFstrain
the of thegauge
strainwere
gauge Ω and
were
120 120 Ω and 2
spectively.
2.1, respectively. A strain A strainwas
gauge gauge was attached
attached at the
at the top of top of a 1.5-high
a 1.5-high cantilever
cantilever beam.beam. Th
The integrated multimetric sensor sensor
grated multimetric board board
with Xnode was deployed
with Xnode at theatsupport
was deployed of theof the bea
the support
beam and measured at a sampling rate of 100 Hz. The conversion factor obtained in the
shunt calibration was 48,287 µs/V. The acceleration and strain noise levels were measured
over 20 s in a vibration-free environment, and the results are plotted in Figure 5. The
standard deviations of the noises related to acceleration and strain were 0.15 mg and 0.29 µs,
Sensors 2021, 21, 5647 6 of 16

measured at a sampling rate of 100 Hz. The conversion factor obtained in the shunt cali-
Sensors 2021, 21, 5647 bration was 48,287 us/V. The acceleration and strain noise levels were measured over 6 of 20
15
s in a vibration-free environment, and the results are plotted in Figure 5. The standard
deviations of the noises related to acceleration and strain were 0.15 mg and 0.29 us, re-
spectively. The acceleration noise was very similar to the datasheet value, indicating that
respectively. The acceleration noise was very similar to the datasheet value, indicating that
the PCB was well designed and introduced no further signal noise. Meanwhile, the noise
the PCB was well designed and introduced no further signal noise. Meanwhile, the noise
in the strain measurement was <1 us with a programmable gain of 12.
in the strain measurement was <1 µs with a programmable gain of 12.

(a) (b)

Figure 5.
Figure 5. Noise level of the developed
developed multimetric
multimetric board:
board: (a) acceleration
acceleration and
and (b)
(b) strain.
strain.

3. Optimal Sensor
3. Optimal Sensor Location
Location for
for Reference-Free
Reference-Free Indirect
Indirect Displacement
Displacement Estimation
Estimation
This section estimates the performance of the reference-free indirect displacement
This section estimates the performance of the reference-free indirect displacement from
the strain and acceleration measurements, and investigates the optimal sensor
from the strain and acceleration measurements, and investigates the optimal sensor loca-location
for
tionthe
for three strain
the three measurements
strain measurementsavailable
availableonon the
themultimetric
multimetricsensor
sensorboard
board through
through
numerical analysis.
numerical analysis.
3.1. Reference-Free Indirect Displacement Estimation
3.1. Reference-Free Indirect Displacement Estimation
The displacement can be indirectly estimated from the vertical acceleration and three-
The displacement can be indirectly estimated from the vertical acceleration and
channel strains measured using the developed multimetric board. Park et al. [19] proposed
three-channel strains measured using the developed multimetric board. Park et al. [19]
a multimetric data fusion method for displacement estimation. The indirect displacement
proposed a multimetric data fusion method for displacement estimation. The indirect dis-
estimation method is based on the following relationship between strain and displacement:
placement estimation method is based on the following relationship between strain and
displacement: ustrain = Φq = ΦΨε = αΦΦ00 ε. (4)
𝑢𝑠𝑡𝑟𝑎𝑖𝑛 = Φ𝑞 = ΦΨ𝜀 = 𝛼ΦΦ′′ 𝜀. (4)
where ustrain is the strain-based displacement, Φ and Ψ are the displacement- and strain-
mode 𝑢𝑠𝑡𝑟𝑎𝑖𝑛
whereshape is the strain-based
matrices, respectively, displacement,
and α is the scaling Φ and Ψ which
factor, are thecritically
displacement- and
determines
strain-mode shape matrices, respectively, and 𝛼 is the scaling factor, which
the strain–displacement relationship. α can be obtained using both acceleration- and strain- critically de-
termines
based the strain–displacement
displacements from Equation relationship.
(4) by matching𝛼 can the be obtained
magnitude using both accelera-
of power spectral
tion- and
density of strain-based displacements
the strain-based displacement from Equation
to the (4) by matching
acceleration-based the magnitude
displacement (obtained of
power spectral density of the strain-based
by double integration of the acceleration): displacement to the acceleration-based dis-
placement (obtained by double integration v of the acceleration):
u disp
u S𝑑𝑖𝑠𝑝 ( f )
u strain,xi(𝑓𝑛n)
α 𝛼==t√𝑆𝑠𝑡𝑟𝑎𝑖𝑛,𝑥𝑖
disp . . (5)
(5)
𝑑𝑖𝑠𝑝
S𝑆𝑎𝑐𝑐,𝑥𝑖
acc,xi (
(𝑓 f
𝑛 )n )

𝑑𝑖𝑠𝑝 𝑑𝑖𝑠𝑝
Here, 𝑆𝑠𝑡𝑟𝑎𝑖𝑛,𝑥𝑖
disp and 𝑆 disp are the power spectral densities of the strain- and acceleration-
Here, Sstrain,xi and𝑎𝑐𝑐,𝑥𝑖
Sacc,xi are the power spectral densities of the strain- and acceleration-
based displacements atatxi.x, i.respectively.
based displacements , respectively. 𝑓𝑛the
f n is is most
the most dominant
dominant natural
natural frequency,
frequency, which
which is generally the first natural frequency of
is generally the first natural frequency of the structure.the structure.

3.2. Setup for Numerical Analysis

measurements
A numerical study was conducted on the locations of the three strain measurements
available for the developed multimetric sensor board. The sensor deployment was sym-
suggested by
metric about the center, as suggested by Park
Park et
et al.
al. [19].
[19]. Specifically, one of
of the
the strain
strain
the bridge
gauges was fixed at the center of the bridge and
and the
the other
other strain
strain gauges
gauges were
were deployed
deployed
equidistantly from
equidistantly fromthe
thecenter.
center.However,
However, thethe length
length of the
of the bridge
bridge is very
is very diverse,
diverse, and
and it wasit
necessary to determine the optimal installation position according to the span length of
the bridge. In addition, the greater the distance from the center, the more difficult it is to
install the gauge, so the optimal location with a minimum distance had to be determined.
Sensors 2021, 21, 5647 7 of 15

In this section, the effect of strain-gauge distance from the center was evaluated on bridges
of different lengths.
Simulations were performed on simply supported beams of various lengths. The
cross section of the model was 2 m × 1 m (width × height), the modulus of elasticity
was 210 GPa, the density was 7850 kg/m3 , and the damping ratio was 5%. The structure
was excited by a traffic load comprising a two-wheel car and a three-wheel truck. For
the two-wheel car, the force on both the wheels was 20 kN and a 10% random force was
applied to the front and rear axles, which were separated by 2.9 m. For the three-wheel
truck, the force values were 35 kN with a 10% random force on the front axle and 145 kN
with a 10% random force on both of the rear axles. The front and rear axles were separated
by 4.3 m, and the separation of the two rear axles was arbitrarily adjusted in the range
of 1–9 m, imposing randomness in the moving load. The vehicle configurations in the
numerical simulation are given in Table 1.

Table 1. Configurations of the moving load in the simulation.

Configurations Front Rear

Two-wheel car 20 kN 20 kN
Three-wheel truck 35 kN 145 kN 145 kN

The simply supported beam was modeled using an n-element Bernoulli beam, where
n equals the length of the beam. For instance, a 30–m long beam comprised 30 elements
and 29 nodes for simulating the acceleration, strain, and displacement. The estimated
displacement in the proposed method was derived by simulating the vertical acceleration
of the beam center and the three strain channels. The estimated displacement was then
compared with the exact displacement obtained in the simulation. Three cases with
different beam lengths (30, 40, and 50 m) were simulated using symmetric strain-gauge
deployments. One gauge was located at the center of the bridge while the other gauges
were deployed equidistant from the center. In the simulation, the distance between the
center and the two side gauges was manipulated, and the estimated displacement was
compared with the exact displacement. For instance, along the 30-m long beam, the off-
center gauges were located at 1–14 m from the center. At each strain measurement location,
the acceleration and strains were simulated 100 times under various moving loads and
the displacement was estimated. The error was defined as the root mean square error
(RMSE) of the difference between the estimated and exact displacement divided by the
maximum displacement (Equation (6)). The errors in the 100 simulations were averaged
for comparison.
σ (uestimated − uexact )
Error = . (6)
max(|uexact |)
where uestimated and uexact are estimated and exact displacement, respectively.

3.3. Results and Discussion

Figure 6 shows the average error in the strain measurements as the location changed
from 0 to 0.5 × L from the center, where L is the bridge length. The optimal location of the
strain measurements (normalized by the bridge length) was 0.2–0.25 × L from the center.
For example, on the 30-m long bridge, the optimal position was 0.2 × 30 m (i.e., 6 m) from
the center. Interestingly, the measurements of the location of the strain from the center did
not critically affect the error (the average error increment was 1–2%). At the sites where
sensor installation was difficult, the strain sensors were located close to the center for the
estimation of displacement.
 Sensors 2021, 21, 5647 8 of 16

Sensors 2021, 21, 5647

Sensors 2021, 21, 5647 8 of 15 8

Figure 6. RMSE error dependence on the normalized distance from the center of the bridge.
Figure 6. RMSE error dependence on the normalized distance from the center of the bridge.
4. Experimental Validation
4.Figure 6. RMSE error
Experimental dependence on the normalized distance from the center of the bridge.
Validation
4.1. Validation on the Keukrak Bridge
4.1. Validation on the Keukrak Bridge
4. Experimental
The proposed Validation
reference-free indirect displacement sensing system was experimen-
The proposed reference-free indirect displacement sensing system was experimen-
tally validated4.1. on a prestressed
Validation on the concrete
Keukrak bridge located at Gwangju-si, South Korea. The
Bridge
tally validated on a prestressed concrete bridge located at Gwangju-si, South Korea. The
experimental setup is presented in Figure 7. The integrated wireless sensor was installed
experimental setupThe proposedinreference-free
is presented Figure 7. The indirect
integrateddisplacement sensing
wireless sensor system
was installed was
at experi
at the mid span of the beam, which was 21-m long, and two strain gauges were attached
the mid span of the beam, which was 21-m long, and two strain gauges were attached 4 m Korea
tally validated on a prestressed concrete bridge located at Gwangju-si, South
4 m from theexperimental
either side ofsetup
the center, which isin0.2 of the7.span length (Table 2). Thesensor
resultswas inst
from the either side of the center,iswhichpresented
is 0.2 ofFigure
the span The integrated
length (Table 2).wireless
The results of
of the
the strain
strain atsensing
sensing midsystem
thesystem span installed
of
installed theatmid-span
theatbeam, the
whichmid-span
was
were were
21-m compared
long,
compared and
withtwo withofthose
strain
those of an
gauges
an LVDT were atta
LVDT installed
installed 4m
at the at
same the
from same
the location.
either
location. side of the center, which is 0.2 of the span length (Table 2). The re
of the strain sensing system installed at the mid-span were compared with those o
(a) LVDT installed at the same location. (b)

(a) Wireless
(b) Sensor

Wireless Sensor
Strain Gauge

LVDT Strain Gauge

LVDT

Figure7.7.Experimental
Figure Experimentalsetup:
setup:(a)
(a)overview
overviewand
and(b)
(b)sensor
sensordeployment.
deployment.

Table 2. Configurations of the validation experiment on the Keukrak Bridge.

Figure 7. Experimental setup: (a) overview and (b) sensor deployment.
Bridge Type Span Strain Measurement Distance from Center
Prestressed concrete 21 m 4m
 Table 2. Configurations of the validation experiment on the Keukrak Bridge.

Bridge Type Span Strain Measurement Distance from Center

Sensors 2021, 21, 5647 9 of 15
Prestressed concrete 21 m 4m

Figure 8 shows the measured acceleration and a comparison of the measured strain
and reference
Figure 8 showssignals. The strainacceleration
the measured of the mid-span
and apoint measured
comparison of by
thethe developed
measured sensor
strain
aligned very well with the reference strain measurements (Figure 8b). Figure
and reference signals. The strain of the mid-span point measured by the developed sensor9 compares
the displacements
aligned very well withestimated using
the reference the proposed
strain method
measurements and the
(Figure 8b).LVDT.
FigureTable 3 presents
9 compares
the the maximum peak
displacements displacements
estimated obtained by
using the proposed the proposed
method methodTable
and the LVDT. and 3the reference.
presents
the The error in
maximum the displacements
peak displacements obtained
of the proposed method relative
by the proposed methodtoandthethe
LVDT displace-
reference.
ments was no greater than 2.17%.
The error in the displacements of the proposed method relative to the LVDT displacements
was no greater than 2.17%.

(a)

(b)

Sensors 2021, 21, 5647 10 of 16

Figure 8. Measured structural responses of the Keukrak Bridge: (a) vertical acceleration and (b) strains at mid-span point.
Figure 8. Measured structural responses of the Keukrak Bridge: (a) vertical acceleration and (b) strains at mid-span point.

FigureFigure
9. Comparison of structural
9. Comparison displacements
of structural at the at
displacements mid-span point point
the mid-span of theof
Keukrak Bridge.
the Keukrak Bridge.

Table 3. Comparison of displacement measurements on the Keukrak Bridge.

Method Maximum Displacement (mm) Error (mm) Error (%)

Proposed method 3.18 0.07 2.17
Reference displacement 3.11 - -

4.2. Validation on the Jojungchun Bridge

A second experiment was performed on a 40-m long steel-box girder bridge located
Sensors 2021, 21, 5647 10 of 15

Figure 9. Comparison of structural displacements at the mid-span point of the Keukrak Bridge.

Table3.3.Comparison
Table Comparisonofofdisplacement
displacementmeasurements
measurementson
onthe
theKeukrak
KeukrakBridge.
Bridge.

Method
Method Maximum
MaximumDisplacement (mm)
Displacement (mm) Error(mm)
Error (mm) Error(%)
Error (%)
Proposed
Proposedmethod
method 3.18
3.18 0.07
0.07 2.17
2.17
Reference displacement
Reference displacement 3.11
3.11 -- --

4.2.Validation
4.2. Validationon onthe
theJojungchun
JojungchunBridge
Bridge
AAsecond
secondexperiment
experimentwas wasperformed
performedon onaa40-m
40-mlong
longsteel-box
steel-boxgirder
girderbridge
bridgelocated
located
at Gapyeong-gun, South Korea. An overview of the experimental setup is presented inin
at Gapyeong-gun, South Korea. An overview of the experimental setup is presented
Figure10.
Figure 10.Two
Tworeference
reference LVDTs
LVDTs werewere installed
installed at endpoint
at the the endpoint of each
of each side(rather
side cell cell (rather
than
than the middle cell) to monitor the bidirectional passage of trains (Figure
the middle cell) to monitor the bidirectional passage of trains (Figure 10b). An integrated 10b). An inte-
grated wireless
wireless sensor was sensor was installed
installed at the at the center
center of theofbox.
the box.
As Asthisthis bridge
bridge is isa acontinuous
continuous
bridgewith
bridge witha asimple
simplesupport
support at at
oneone
end,end,
its its displacement
displacement should
should be atbeitsatmaximum
its maximum at theat
the point
3/8 3/8 point
(i.e.,(i.e.,
at 15at m).
15 m). Hence,
Hence, thethe LVDT
LVDT was was installed
installed 1515
mm fromthe
from thesimply
simplysupported
supported
end. The displacement
end. displacementatatthe thesame
samepoint
pointwaswas indirectly estimated,
indirectly estimated, andand thethe reference dis-
reference
placement at the sensor location was interpolated
displacement at the sensor location was interpolated between between the two LVDT displacements
LVDT displacements
forcomparison.
for comparison.Table Table44gives
givesthetheconfigurations
configurationsofofthe theexperimental
experimentalvalidation.
validation.

(a) (b)

Sensor location
(c) LVDT location

Figure10.
Figure 10.Experimental
Experimentalsetup
setupshowing
showing(a)
(a)the
thesteel-box
steel-boxgirder
girderbridge
bridgeand
and(b)
(b)its
itssectional
sectionaland
and(c)
(c)lateral
lateralviews.
views.

Table4.4.Configurations
Table Configurationsfor
forexperimental
experimentalvalidation
validationon
onthe
theJojungchun
JojungchunBridge.
Bridge.

BridgeType
Bridge Type Span
Span Strain
StrainMeasurement Distancefrom
Measurement Distance from Center
Center
Three-cell steel-box girder 40 m 10 m
Three-cell steel-box girder 40 m 10 m

Figure 11 presents the vertical acceleration at the mid-span point and the strains
Figureat11three
measured presents thewhere
points vertical
the acceleration at the mid-span
strain measurements point and
were collected at athe strainsof
distance
measured at three points where the strain measurements were collected at a distance
10 m from the simply supported end (at 0.25 of the span length). The strain responses of
10 m from
were the simply
asymmetric supported
because end (at conditions
the boundary 0.25 of the differed
span length).
at eachThe strain
end. Oneresponses
end was a
were asymmetric because the boundary conditions differed at each end. One end was
a simple support while the other was continuous. Figure 12 compares the reference
and indirect displacements estimated by the proposed sensing system. The maximum
displacement estimated by the proposed method was 0.8795 mm. Compared with the
0.8780 mm estimated by the reference data, this gave a maximum displacement error of
0.17% (Table 5).

Table 5. Comparison of displacement measurements on the Jojungchun Bridge.

Method Maximum Displacement (mm) Error (mm) Error (%)

Proposed method 0.8795 0.0015 0.17
Reference displacement 0.8780 - -
 indirect displacements estimated by the proposed sensing system. The maximum dis-
simple
placement support
estimated by while the other
the proposed was continuous.
method Figure
was 0.8795 mm. 12 compares
Compared with the 0.8780
reference and
indirectby
mm estimated displacements
the reference estimated by the
data, this gave proposed sensing
a maximum system.
displacement Theofmaximum
error 0.17% dis-
placement estimated by the proposed method was 0.8795 mm. Compared with the 0.8780
(Table 5).
Sensors 2021, 21, 5647 mm estimated by the reference data, this gave a maximum displacement error11of 0.17%
of 15
(Table 5).

(a)

(a)

(b)

(b)

Figure 11. Measured structural responses from the Jojungchun Bridge: (a) vertical acceleration and (b) strains at three
points.
Figure 11. Measured
Figure structural
11. Measured responses
structural from the
responses Jojungchun
from Bridge: (a)
the Jojungchun vertical
Bridge: (a)acceleration and (b) strains
vertical acceleration and at
(b)three points.
strains at three
points.

Figure 12. Comparison

Figure of displacements
12. Comparison at the
of displacements at3/8
thespan pointpoint
3/8 span of theofJojungchun Bridge.
the Jojungchun Bridge.

4.3. Figure
Application of Developed
12. Comparison System to the Geumgok
of displacements at the 3/8 Bridge
span point of the Jojungchun Bridge.
To demonstrate the efficiency of the proposed cloud-based monitoring system, the
developed sensor was instrumented on a tall (20-m high) concrete bridge located at
Yeongdong-gun, South Korea. As the proposed indirect displacement sensing system
is applicable to any type of bridge, the Geumkok Bridge was selected as the testbed be-
cause, to the best of our knowledge, its displacement has not been previously monitored.
The developed sensor was installed beneath the bridge deck, where it could detect ambient
vibrations. The sensor and strain gauges were installed and instrumented by a bridge
inspection truck (Figure 13). The span of the bridge was 35 m and the strain measurements
were collected at a distance of 5 m from the center (0.15 of the span length).
 Yeongdong-gun, South Korea. As the proposed indirect displacement sensing system is
applicable to any type of bridge, the Geumkok Bridge was selected as the testbed because,
to the best of our knowledge, its displacement has not been previously monitored. The
developed sensor was installed beneath the bridge deck, where it could detect ambient
Sensors 2021, 21, 5647
vibrations. The sensor and strain gauges were installed and instrumented by a bridge in-
12 of 15
spection truck (Figure 13). The span of the bridge was 35 m and the strain measurements
were collected at a distance of 5 m from the center (0.15 of the span length).

(a)

(b)

Figure 13. Experimental

Figure 13. Experimental setup:
setup: (a)
(a) the
the Geumgok
Geumgok Bridge;
Bridge; (b)
(b) strain-gauge
strain-gauge locations.
locations.

The scaling factor was obtained from the strain and acceleration measurements using
The scaling factor was obtained from the strain and acceleration measurements using
Equation (5). Figure 14 plots the cross power spectral densities (CPSDs) of the acceleration-
Equation (5). Figure 14 plots the cross power spectral densities (CPSDs) of the accelera-
and strain-based displacements. The scaling factor was 0.044 at the first natural frequency of
Sensors 2021, 21, 5647 13 of 16
tion- and strain-based displacements. The scaling factor was 0.044 at the first natural fre-
4.49 Hz. At this frequency, the CPSD of the calibrated strain-based displacement exhibited
quency of 4.49 Hz. At this frequency, the CPSD of the calibrated strain-based displacement
the same power as the acceleration-based displacement.
exhibited the same power as the acceleration-based displacement.

Figure 14.14.
Figure Scaled cross
Scaled power
cross spectral
power densities
spectral (CPSDs)
densities of acceleration
(CPSDs) and strain
of acceleration on theon
and strain Geumgok
the Geum-
Bridge.
gok Bridge.

The
The structural
structural responses
responses with
with unknown
unknown input
input were
were monitored
monitored viavia video
video recordings
recordings
during
during thethe measurements.Figure
measurements. Figure1515shows
showsan aninstance
instance of
of the measured
measured vertical
verticalacceleration
accelera-
and
tion three
and threechannel
channelstrains
strainsas well as the
as well estimated
as the displacement
estimated caused
displacement by aby
caused five-wheel
a five-
trucktruck
wheel carrying steel steel
carrying platesplates
(Figure 16). The
(Figure 16).developed system
The developed captured
system displacements
captured displace-as
smallasassmall
ments 0.1 mm from
as 0.1 mm thefrom
set of strain
the set ofand acceleration
strain measurements,
and acceleration and the maximum
measurements, and the
displacement was 0.42 mm.
maximum displacement was 0.42 mm.

(a)
 The structural responses with unknown input were monitored via video recordings
during the measurements. Figure 15 shows an instance of the measured vertical accelera-
tion and three channel strains as well as the estimated displacement caused by a five-
wheel truck carrying steel plates (Figure 16). The developed system captured displace-
Sensors 2021, 21, 5647 13 of
ments as small as 0.1 mm from the set of strain and acceleration measurements, and the15
maximum displacement was 0.42 mm.

(a)

(b)

(c)

Sensors 2021, 21, 5647 Figure

Figure15.
15.Measured
Measuredstructural responses
structural responsesand estimated
and estimateddisplacements
displacements of the Geumgok
of the 14 of
Bridge:
Geumgok (a)16
Bridge:
measured vertical
(a) measured acceleration;
vertical (b) three-channel
acceleration; strains;
(b) three-channel andand
strains; (c) estimated displacements.
(c) estimated displacements.

Figure16.
Figure 16.Five-wheel
Five-wheelheavy
heavytruck
truckcrossing
crossingthe
theGeumgok
GeumgokBridge.
Bridge.

5.5.Conclusions
Conclusions
In
In this
this paper,
paper,we
wedeveloped
developedandandevaluated
evaluatedaareference-free
reference-freeindirect
indirectdisplacement
displacement
sensing
sensing system. Thesystem
system. The systemintegrates
integratesa amultimetric
multimetric board
board that
that measures
measures three
three strains
strains and
and the three-axis acceleration using a commercial wireless smart sensor. The
the three-axis acceleration using a commercial wireless smart sensor. The indirect dis- indirect
displacement estimation
placement estimation method
method waswas optimized
optimized forfor three
three strain
strain measurements
measurements andandoneone
ac-
acceleration measurement. In actual bridge monitoring, the number of wireless
celeration measurement. In actual bridge monitoring, the number of wireless signals are signals
inevitably reduced and the communication among multiple wireless sensors becomes dif-
ficult. For this reason, the proposed sensing system uses a single sensor unit for reference-
free bridge displacement estimation.
The main contributions of this study are as follows:
Sensors 2021, 21, 5647 14 of 15

are inevitably reduced and the communication among multiple wireless sensors becomes
difficult. For this reason, the proposed sensing system uses a single sensor unit for reference-
free bridge displacement estimation.
The main contributions of this study are as follows:
• The developed multimetric board simultaneously observed the three-axis acceleration
and three-channel strains and estimated the reference-free displacement.
• To enhance the accuracy of the displacement estimation, the placements of the three-
channel strains were optimized in a numerical study.
When developing the multimetric sensor board for wireless sensors, the following de-
sign challenges were addressed: (1) the board required a high-resolution accelerometer and
strain sensor with power management; (2) the board was integrated with three Wheatstone
bridges for three-channel strain sensing; and (3) an automated shunt calibration was devel-
oped and implemented in the software. Furthermore, the optimal location of strain sensing
for the estimation of indirect displacement (0.2–0.25 of the span length) was numerically
validated. For validation purposes, the developed multimetric board was integrated with
the commercial wireless sensor platform Xnode. The integrated system was experimentally
validated on two differently constructed railroad bridges and the estimated displacements
were compared with those of LVDTs. On the Keukrak and Jojungchun Bridges, the max-
imum displacement errors were 2.17% and 0.17%, respectively. Finally, the developed
system was applied to a tall (20-m high) bridgewhich was very difficult to monitor by
using existing systems. The indirect displacements were successfully obtained, confirming
that the proposed system can capture very small (<0.1 mm) reference-free displacements.
Future work is underway to develop ambient bridge displacement monitoring by
expanding the proposed sensing system to cloud-based bridge management, where sensor
data is uploaded in a cloud database and data processing for displacement estimation can
be implemented in an autonomous manner.

Author Contributions: All authors contributed to the main idea of this paper. Methodology, Data
curation, J.P., J.S. and M.P.; Supervision, J.-W.P.; Validation, J.W., J.P. and J.S.; Writing—original draft,
J.W.; Writing—review & editing, J.-W.P. All authors have read and agreed to the published version of
the manuscript.
Funding: This research was funded by National R&D Project, grant number 21CTAP-C164014-01
and Chung-Ang University Research Scholarship Grants in 2020.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Acknowledgments: This work was supported by the Korea Agency for Infrastructure Technology
Advancement (KAIA) grantfunded by the Ministry of Land, Infrastructure and Transport (Grant
21CTAP-C164014-01) and Chung-Ang University Research Scholarship Grants in 2020.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Gomez, F.; Park, J.-W.; Spencer, B.F., Jr. Reference-free structural dynamic displacement estimation method. Struct. Control Health Monit.
2018, 25, e2209. [CrossRef]
2. Sarwar, M.Z.; Park, J.-W. Bridge displacement estimation using a co-located acceleration and strain. Sensors 2020, 20, 1109.
[CrossRef]
3. Lee, J.; Lee, K.-C.; Cho, S.; Sim, S.-H. Computer vision-based structural displacement measurement robust to light-induced image
degradation for in-service bridges. Sensors 2017, 17, 2317. [CrossRef]
4. Hoskere, V.; Park, J.-W.; Yoon, H.; Spencer, B.F., Jr. Vision-based modal survey of civil infrastructure using unmanned aerial
vehicles. J. Struct. Eng. 2019, 145, 04019062. [CrossRef]
5. Yoon, H.; Elanwar, H.; Choi, H.; Golparvar-Fard, M.; Spencer, B.F., Jr. Target-free approach for vision-based structural system
identification using consumer-grade cameras. Struct. Control Health Monit. 2016, 23, 1405–1416. [CrossRef]
Sensors 2021, 21, 5647 15 of 15

6. Won, J.; Park, J.-W.; Park, K.; Yoon, H.; Moon, D.-S. Non-target structural displacement measurement using reference frame-based
deepflow. Sensors 2019, 19, 2992. [CrossRef]
7. Feng, M.Q.; Fukuda, Y.; Feng, D.; Mizuta, M.J. Nontarget vision sensor for remote measurement of bridge dynamic response.
J. Bridge Eng. 2015, 20, 04015023. [CrossRef]
8. Lee, J.; Lee, K.C.; Lee, S.; Lee, Y.J.; Sim, S.H. Long-term displacement measurement of bridges using a LiDAR system.
Struct. Control Health Monit. 2019, 26, e2428. [CrossRef]
9. Kim, K.; Sohn, H. Dynamic displacement estimation by fusing LDV and LiDAR measurements via smoothing based Kalman
filtering. Mech. Syst. Signal Process. 2017, 82, 339–355. [CrossRef]
10. Pieraccini, M. Monitoring of civil infrastructures by interferometric radar: A review. Sci. World J. 2013, 2013, 786961. [CrossRef]
11. Zhang, G.; Wu, Y.; Zhao, W.; Zhang, J. Radar-based multipoint displacement measurements of a 1200-m-long suspension bridge.
ISPRS J. Photogramm. Remote Sens. 2020, 167, 71–84. [CrossRef]
12. Miccinesi, L.; Beni, A.; Pieraccini, M. Multi-Monostatic Interferometric Radar for Bridge Monitoring. Electronics 2021, 10, 247.
[CrossRef]
13. Lee, H.S.; Hong, Y.H.; Park, H.W. Design of an FIR filter for the displacement reconstruction using measured acceleration in
low-frequency dominant structures. Int. J. Numer. Methods Eng. 2010, 82, 403–434. [CrossRef]
14. Hong, Y.H.; Kim, H.-K.; Lee, H.S. Reconstruction of dynamic displacement and velocity from measured accelerations using the
variational statement of an inverse problem. J. Sound Vib. 2010, 329, 4980–5003. [CrossRef]
15. Liu, B.; Ozdagli, A.I.; Moreu, F. Direct reference-free measurement of displacements for railroad bridge management.
Struct. Control Health Monit. 2018, 25, e2241. [CrossRef]
16. Wang, Z.-C.; Geng, D.; Ren, W.-X.; Liu, H.-T. Strain modes based dynamic displacement estimation of beam structures with strain
sensors. Smart Mater. Struct. 2014, 23, 125045. [CrossRef]
17. Jeon, J.-C.; Lee, H.-H. Development of displacement estimation method of girder bridges using measured strain signal induced
by vehicular loads. Eng. Struct. 2019, 186, 203–215. [CrossRef]
18. Li, C.-J.; Ulsoy, A.G. High-precision measurement of tool-tip displacement using strain gauges in precision flexible line boring.
Mech. Syst. Signal Process. 1999, 13, 531–546. [CrossRef]
19. Park, J.W.; Sim, S.H.; Jung, H.J. Displacement estimation using multimetric data fusion. IEEE/ASME Trans. Mechatron. 2013, 18,
1675–1682. [CrossRef]
20. Park, J.W.; Lee, K.C.; Sim, S.H.; Jung, H.J.; Spencer, B.F., Jr. Traffic safety evaluation for railway bridges using expanded
multisensor data fusion. Comput.-Aided Civ. Infrastruct. Eng. 2016, 31, 749–760. [CrossRef]
21. Park, J.W.; Sim, S.H.; Jung, H.J. Wireless displacement sensing system for bridges using multi-sensor fusion. Smart Mater. Struct.
2014, 23, 045022. [CrossRef]
22. Linderman, L.E.; Rice, J.A.; Barot, S.; Spencer, B.F., Jr.; Bernhard, J.T. Characterization of wireless smart sensor performance.
J. Eng. Mech. 2010, 136, 1435–1443. [CrossRef]
23. Fasl, J.; Helwig, T.; Wood, S.L.; Pasupathy, P.; Neikirk, D.P.; Raffaeli, F. Factors affecting wireless network communication in
monitoring systems for steel bridges. J. Civ. Struct. Health Monit. 2012, 2, 87–95. [CrossRef]
24. Spencer, B.F., Jr.; Park, J.W.; Mechitov, K.A.; Jo, H.; Agha, G. Next, generation wireless smart sensors toward sustainable civil
infrastructure. Procedia Eng. 2017, 171, 5–13. [CrossRef]
25. Zhu, L.; Fu, Y.; Chow, R.; Spencer, B.F.; Park, J.W.; Mechitov, K.J.S. Development of a high-sensitivity wireless accelerometer for
structural health monitoring. Sensors 2018, 18, 262. [CrossRef] [PubMed]
26. Fu, Y.; Mechitov, K.; Hoang, T.; Kim, J.R.; Lee, D.H.; Spencer, B.F., Jr. Development and full-scale validation of high-fidelity data
acquisition on a next-generation wireless smart sensor platform. Adv. Struct. Eng. 2019, 22, 3512–3533. [CrossRef]
27. NXP. LPC4357. Available online: https://www.nxp.com/products/processors-and-micro-controllers/arm-microcontrollers/
general-purpose-mcus/lpc4300-cortex-m4-m0/high-performance-32-bit-microcontroller-mcu-based-on-arm-cortex-m4-m0
-cores:LPC4357FET256 (accessed on 15 July 2021).
28. Microchip. AT86RF233 Datasheet and Product Info. Available online: https://www.micro-chip.com/wwwproducts/en/AT8
6RF233 (accessed on 15 July 2021).
29. Texas Instruments. ADS131E08 Datasheet and Product Info. Available online: https://www.ti.com/document-viewer/ADS131
E08/datasheet (accessed on 15 July 2021).
30. Analog Devices. ADXL354 Datasheet and Product Info. Available online: https://www.ana-log.com/en/products/adxl354.html#
(accessed on 15 July 2021).