Received: 10 December 2015 Revised: 2 November 2016 Accepted: 7 November 2016

DOI 10.1002/stc.1969

RESEARCH ARTICLE

Laboratory validation of buried piezoelectric scour sensing rods

Faezeh Azhari1 | Kenneth J. Loh2

1
Department of Mechanical and Industrial
Engineering, and Department of Civil Summary
Engineering, University of Toronto, Toronto, Scour, or the erosion of soil and sediments near bridge piers and abutments,
Ontario M5S 3G8, Canada accounts for the majority of overwater bridge failures. This study focuses on evalu-
2
Department of Structural Engineering, University ating the use of a driven piezoelectric scour sensing rod, where the real‐time dynam-
of California‐San Diego, La Jolla, California
92093, USA
ics of the voltage response of the sensing rod is used to determine scour depths using
Correspondence
the inverse relation between natural frequency and the rod’s exposed length. A poly
Kenneth J. Loh, Department of Structural (vinylidene fluoride) polymer strip forms the main sensing component of this proto-
Engineering, University of California‐San Diego, type sensor. After confirming the viability of the sensing concept through various
La Jolla, CA 92093, USA
Email: kenloh@ucsd.edu
idealized tests, the response of the sensors was studied in scour conditions simulated
in a laboratory flume. The sensors were driven into the soil surrounding a cylindrical
Funding information
pier. As the scour hole evolved, the exposed length of the sensors changed, causing
American Society of Civil Engineers (ASCE) the measured natural frequencies to also vary. Scour depth at each sensor location
Freeman Fellowship and U.S. National Science was determined using a simple cantilever beam eigenfrequency analysis where the
Foundation, Grant/Award Number: CMMI‐
soil support fixity was modeled with a rotational spring. The results were promising
1234080.
in that the sensors were capable of detecting scour depths and the scour hole topog-
raphy with reasonable accuracy. As is the case with other rod‐like scour sensors, vul-
nerability to debris and installation difficulties are some of the limitations that need
to be addressed in future real‐world implementations.

K E Y WO R D S

bridge scour, piezoelectric, polyvinylidene fluoride (PVDF), sensor, structural health

monitoring, vibration

1 | IN T RO D U C T IO N overwater bridge failures.[12–14] The 1987 collapse of the U.

S. Interstate‐90 bridge crossing Schoharie Creek in New
Many bridges and marine structures are scour critical, mean- York is a historic and prominent example of scour‐induced
ing that they are susceptible to damage or failure due to bridge failure. More recent cases include the collapse of the
scour.[1,2] Scour is the removal of bed sediments from around Houfeng bridge in Taiwan caused by a 2008 typhoon and
the structure’s supports by flowing water, especially during flooding,[12] the 2009 failure of the Malahide Viaduct in
floods. Numerous factors can affect the rate and extent of this Dublin, Ireland,[15] and the Bonnybrook Bridge failure as a
dynamic phenomenon, including support geometry, inclina- result of the June 2013 flood event in Calgary, Canada.[16]
tion, geometry, spacing, and group arrangements, as well as In the United States, the National Bridge Inventory
water depth, flow rate, angle of attack, and sediment proper- (NBI)[17] records a scour rating for all overwater bridges.
ties.[2–8] If a large enough scour hole develops around the Bridges rated as “scour‐critical” are deemed unstable due to
supports, the reduction of the axial and lateral capacity of observed scour or through scour evaluations. Figure 1 pre-
the structure can lead to undesirable deflections, structural sents a map of scour‐critical bridges in the United States, gen-
instability, or even failure. Scour is a major problem world- erated using the data recorded in the 2014 NBI.[17]
wide, including the United States[9] and many East‐Asian Various preventative bridge design guidelines and coun-
countries that regularly experience typhoons and floods.[10,11] termeasure strategies have been established to minimize the
In fact, scour has jeopardized the structural integrity of risk of scour‐induced loss of capacity and failure.[18] Realistic
many bridges around the world and is a leading cause of knowledge about the current or potential scour hole geometry

Struct Control Health Monit 2016; 1–14 wileyonlinelibrary.com/journal/stc Copyright © 2016 John Wiley & Sons, Ltd. 1
2 AZHARI AND LOH

FIGURE 1 Scour‐critical bridges in the United

States as recorded in the 2014 National Bridge
Inventory[17]

is usually a prerequisite for incorporating such measures into bridges at risk of scour‐induced failures. The uncertainty
the design and maintenance of bridges. The most important involved in identifying scour‐critical bridges brings about
geometric parameter is the maximum (or equilibrium) scour another raison d’être of permanent and real‐time scour mon-
depth in the case of live‐bed scour, and the temporal develop- itoring systems on overwater bridges.
ment of the scour depth in the case of clear‐water scour.[19] Sensing technologies used in scour monitoring systems
One can estimate maximum scour depths through various need to perform accurately in the presence of current‐induced
mathematical models and empirical equations.[18–22] While vortices, which is by no means simple. Sensing technologies
these methods are essential at the design stage, the fact that currently used by transportation agencies and bridge owners
they are only predictive and typically not inclusive of all sit- include float‐out devices,[23] magnetic collars,[24] sonar,[25]
uations diminishes their applicability at the maintenance radar,[26] tilt sensors,[27] digital switch sensors,[28] and time‐
stage where scour has already occurred, and a more accurate domain reflectometry.[29] Float‐out and sonar devices have
estimation of scour depths is required. Periodic inspections been more popular than others. Float‐out devices, as the
using portable instruments or visual assessments by trained name implies, float out from their initial buried position
divers can provide fairly accurate information regarding the when scour occurs directly above them and then wirelessly
bed levels around bridge piers and abutments. However, tur- transmits sensor status to a remote station. Sonar devices
bulent and murky waters often prevent this type of assess- are currently the highest‐rated sensing systems for monitor-
ment,[16] not to mention that these inspections are ing scour at bridge piers.[23] Sonar‐based technologies deter-
performed intermittently and may not capture all critical mine the distance to the riverbed, and hence scour depth, by
events. sending acoustic signals underwater and then monitoring the
Therefore, due to the limited predictive capabilities of time of pulse reflections.
models and equations, as well as safety and efficiency issues Despite their prevalence, the aforementioned existing scour
concerning visual inspections, autonomous and continuous sensing technologies suffer from numerous shortcomings,
scour monitoring through permanent (fixed) instrumentation including but not limited to low resolution, low reliability, sus-
have become necessary to prevent scour‐induced failures. ceptibility to interferences, costly and difficult installation and
The Federal Highway Administration requires bridge owners maintenance procedures, tedious data post‐processing require-
to devise and implement bridge‐specific scour plans of action ments, and vulnerability to debris.[2,29,30] Float‐out devices are
for scour‐critical bridges and bridges with unknown founda- battery operated, which limits their application to short‐term
tions. Monitoring, through regular inspections or fixed sens- use. Moreover, they provide discrete measurements of scour
ing devices, is a key part of scour plans of action. While depth and require reinstallation once activated, which makes
scour‐critical bridges are most vulnerable, a bridge that is them less convenient for situations where multiple rounds of
not rated scour‐critical is not necessarily immune from scour and refill occur. Sonar devices can produce topographic
scour‐induced damage and collapse. For example, severe maps of the scour holes; however, they are very susceptible to
flash floods in southern California recently (in July 2015) debris and ice interferences, and during turbulent scour events,
scoured the abutment of an Interstate‐10 bridge, causing the they can usually only be used to measure the final scour depth
bridge to partially collapse, closing the main route between as opposed to continuously monitoring the scour hole evolu-
Los Angeles and Phoenix. This bridge, according to the tion.[2,23] In the case of time‐domain reflectometry, which is
NBI,[17] was previously rated stable for the assessed scour based on a guided radar system and uses contrast in dielectric
conditions (code 113 = 8). The effect of climate change on properties at interfaces for estimating scour depths, the signal
the intensity and frequency of floods can also place more analysis for finding the reflections on the waveform can be
AZHARI AND LOH 3

quite challenging.[27,29] Prendergast and Gavin[15] have which the sensors were examined with idealized fixed‐end
reviewed a variety of scour sensing technologies in more detail. boundary conditions. Subsequently, an extensive set of tests
A growing body of research has aimed at developing new was conducted in a flume that simulated scour around a
and improved scour sensing technologies. These efforts cylindrical bridge pier. Through these small‐scale experi-
have led to various innovative research prototypes, such as ments, the scour sensing capability of the piezo‐rod sensors
micro‐electromechanical systems,[31] smart rocks,[32] fiber‐ and the effect of sensor location with respect to the pier were
optics,[33–35] dissolved oxygen sensors,[36] and erosion studied. Results from the aforementioned experiments were
function apparatus.[37] The improvements made to current compared against visual measurements.
scour sensors and advances in emerging scour monitoring
techniques are promising. Nevertheless, much work is still
2 | S E N S I NG M E C H A N I S M
required to further address the challenges involved in
monitoring bridge scour (e.g., low resolution, low reliability,
As described in the introduction, the proposed scour sensor is
inadequate ability to monitor scour depth evolution, expen-
a piezoelectric rod that functions based on the fact that the
sive and onerous installation and maintenance procedures,
fundamental (natural) frequency, fn, of a cantilevered rod is
high power consumptions, and vulnerability to debris and
inversely proportional to its length according to the well‐
vandalism).
known eigenvalue solution of a distributed‐mass cantilevered
This study intends to develop a simple piezoelectric
beam:
sensing rod (piezo‐rod) for scour depth monitoring. As aptly
categorized by Kong et al.,[30] driven scour sensing rods may
sffiffiffiffiffi
function on the basis of changes in one of three parameters: 1 3:5156 EI
fn ¼ ; (1)
vibration frequency, bending moment profile, or modal strain 2π L2 μ
profile. The proposed sensor falls in the first category and
identifies scour depths from changes in the measured natural where L is exposed length, E is elastic modulus, I is second‐
frequency of the exposed (cantilevered) portion of the rod. moment‐of‐area, and μ is linear mass density. An illustration
The piezo‐rod sensor can be driven into the streambed where of the sensing scheme is presented in Figure 2. The sensing
scour depth measurements are desired. Scour exposes the rod is driven into the bed material close to the bridge pier
sensor in a gradual manner, increasing the cantilevered where scour depth measurements are desired. In the event
length. The dependence of natural frequency on the exposed of scour, the rod’s exposed length increases as the scour hole
length is used to calculate the exposed length and, in turn, the deepens. Excited by flowing water, the piezoelectric PVDF
scour depth. A similar scour sensing rod was developed by polymer generates a time‐varying voltage signal correlated
Zarafshan et al.[38] in which the rod was instrumented with to the vibration of the cantilevered rod. The natural frequency
a fiber‐optic Bragg grating sensor. The main sensing compo- of the rod is calculated by converting the voltage time history
nent of the sensor developed in this study is a continuous response to the frequency domain through the fast Fourier
poly (vinylidene fluoride) (PVDF) film, which is embedded transform (FFT).
inside the rod. PVDF films are relatively low cost, mechani- As scour progresses, a fixed length of the time‐domain
cally strong, impact resistant, and chemically stable. When response is considered, and fn is calculated continuously.
embedded into the rod structure, the PVDF film does not dis- The exposed length is back‐calculated at each time step.
rupt the rod’s mechanical motions, because it is highly flexi- Subtracting the initial exposed length gives the scour depth.
ble and lightweight. As transducers, PVDF films are highly
sensitive to forces and perform well over a wide range of fre-
quencies. In fact, the use of PVDF piezoelectric films for
scour monitoring has been previously investigated by Wang
et al.,[39] where multiple PVDF films are installed at discrete
locations along the sensor length. The key improvement in
the present approach is that only a single PVDF film covers
the entire sensor length. Therefore, this sensor identifies
scour and refill depths in a continuous manner and not at dis-
crete and pre‐defined depths.
In the following sections, a more detailed description of
the sensing mechanism will be provided, followed by
explanations of experimental procedures for fabrication,
FIGURE 2 As the scour hole deepens, the exposed length of the
characterization, and validation of the sensor in a laboratory
polyvinylidene fluoride‐based piezoelectric sensor increases. Scour depth at
setting. Following the characterization tests to determine the that location is determined using the dynamics of the flow‐induced voltage
mechanical properties of the sensor and the soil‐support con- time history response as related to the exposed length of the piezoelectric
ditions, a number of proof‐of‐concept tests were performed in sensor
4 AZHARI AND LOH

It is, of course, important that an accepted range of exposed First, one end of a 30 cm‐long plastic tube with a diame-
length be identified prior to installation so that the initial ter of 6 mm was capped using fast‐curing epoxy and set
exposed length is reasonable (i.e., provides a measurable aside to dry. Next, a 25 cm‐long and 3 mm‐wide strip of
fn), and the shortest anticipated embedded length (at maxi- 110 μm‐thick metallized PVDF sheet (Measurement Special-
mum scour depth) is enough to support the rod. Furthermore, ties) was furnished with two electrodes using conductive
as will be described in later sections, the effect of water and soil copper tape (Ted Pella). To do this, lead wires were soldered
support should be taken into account when back‐calculating to the nonadhesive side of the copper tape, which was then
exposed lengths. affixed onto the metallized surfaces of the PVDF strip. A
small amount of colloidal silver paste (Ted Pella) was painted
over the copper tape and film surface to ensure good contact.
3 | E XP E R IM E NTA L DE TAI L S In the third step, a two‐part PDMS kit (Dow Corning) was
used to prepare the PDMS silicone elastomer. The base and
curing agent were combined at a 10:1 mass ratio, and the
3.1 | Sensor design and fabrication mixture was vigorously stirred by hand for 1 min before it
The sensor prototypes were made by encasing the PVDF pie- was placed in a fume hood for ~3 min until entrapped air bub-
zoelectric transducer in a thin‐walled plastic tube. The casing bles were allowed to escape. The PDMS was then slowly and
provides environmental protection and appropriate mechani- incrementally injected into the capped tube using a pipette.
cal stiffness suitable for subsequent laboratory testing. The Finally, the PVDF strip was carefully inserted into the
PVDF film was kept in place inside the plastic casing using tube. The sensor was placed vertically inside a fume hood
a polydimethylsiloxane (PDMS) silicone elastomer, which (for ventilation) for 3 days to allow the PDMS to completely
is hydrophobic and thus also helps with waterproofing the cure. Once cured, the PDMS elastomer provided mechanical
sensor. It should be mentioned that the sensor design is only coupling between the PVDF and the casing. To further water-
suitable for laboratory‐based tests. While the sensor design proof the sensor, a small amount of fast‐curing marine epoxy
and operation concept can remain the same for large‐scale was applied to the uncapped end of the sensor, and a thin
and real‐world implementations, alternate sensor casing layer of Parafilm sealing film was wrapped around the tip,
materials need to be considered for robustness. Four main including a short length of the wires. Figure 3a illustrates a
steps were followed in the fabrication of the proposed schematic of the sensor, and Figure 3b shows a set of pre-
PVDF‐based sensor. pared sensors.

FIGURE 3 (a) In the prototype scour sensing

rods, the piezoelectric polyvinylidene fluoride
strip, which is the main sensing component, was
kept in place in a plastic sheath using
polydimethylsiloxane elastomer. (b) Sensors were
prepared in sets of four
AZHARI AND LOH 5

3.2 | Data acquisition system 3.3.2 | Soil p‐delta tests

A low‐noise signal conditioning chassis, National Instru- In practice, the piezo‐rods are driven into soil so the base
ments (NI) SCXI 1000, capable of high‐speed multiplexing boundary conditions are not quite fixed. It is important,
was used and controlled using the NI‐DAQmx software. Up therefore, to account for the soil support when calibrating
to eight sensors were simultaneously interrogated in this the sensor based on theoretical calculations. In this study, a
study. The sensors were connected to the SCXI channels rotational spring was used for this purpose. To find an
through a terminal block. The SCXI module created a single appropriate spring coefficient, a number of load‐displacement
analog output consisting of the values sampled on all the (p‐delta) tests were performed in soil. Figure 4 shows the test
SCXI channels. An NI USB‐6259, which is a high‐speed setup. The sensor was driven into saturated sand inside a large
multifunction data acquisition (DAQ) module, was then used container. An acrylic cover with a 10 cm‐long opening was
to digitize the single analog output and sort the data to pro- placed on the container such that the sensor was located at
duce the data acquired from each of the multiple channels the center of the opening. Monotonic and cyclic p‐delta tests
that were sampled. were performed by incrementally increasing and decreasing
displacement. Three different embedment lengths, corre-
sponding to exposed lengths, Le, of 10, 13, and 15 cm were
examined, while both load and displacement values were
3.3 | Sensor characterization and proof‐of‐concept tests
recorded at 10 cm from the soil level. An EXTECH 465040
The characterization experiments involved a set of laboratory (Nashua, NH, USA) digital force gauge capable of measuring
tests to determine the sensor properties, soil‐support condi- both pushing and pulling loads was used to accurately mea-
tions, and to confirm the functionality of the sensors. The sure the applied force. The displacement increments were
ability to back‐calculate cantilevered lengths of the sensor applied manually using the gradations along the opening,
from its frequency response was examined for idealized while the actual applied displacements were accurately mea-
fixed‐end conditions. The experimental details of these pre- sured using an MTI LCT‐120‐40‐SA laser displacement
liminary tests are described in this section. transducer. The results from these experiments provided an
estimate of the rotational spring coefficient used in the cali-
bration model for the small‐scale flume validation tests.
3.3.1 | Sensor mechanical properties and density
The sensor density and elastic modulus were required for
3.3.3 | Fixed‐support free vibration tests
back‐calculating the cantilevered length from natural fre-
Idealized fixed‐end free vibration tests were performed to
quency, as demonstrated by Equation 1. The density of
demonstrate the sensing concept and functionality. The test
the piezo‐rod was measured at 1,142 kg/m3. Its elastic
setup included a flat and elevated surface to house the
modulus was evaluated using the Euler‐Bernoulli beam the-
cantilevered sensing rod in a fixed‐end manner. These exper-
ory and load‐displacement measurements from cantilevered
iments were conducted in both air and water. For the experi-
beam tests. In these tests, the sensor was clamped to a flat
ments in water, the test setup was placed inside a flume filled
and elevated surface to create a cantilevered cylindrical
with water such that the sensor was fully submerged. Tests
beam. Displacement‐controlled cantilevered tests were per-
were performed with unsupported lengths varying from 8 to
formed by applying a point load to the sensor tip using a
20 cm in 1 cm increments. Free vibration was induced by
Test Resources 150R load frame equipped with a 1.1 lbf
gently tapping the piezo‐rod tip, and the generated voltage
(4.9 N) load cell. Tip displacement was measured using
time history was recorded using the DAQ described in
an MTI Instruments LCT‐120‐40‐SA laser displacement
Section 3.2 at a sampling rate of 2,048 Hz. The voltage time
transducer (Albany, NY, USA). The applied load P [N]
histories were then converted to the frequency domain using
and displacement δ [mm] were recorded simultaneously
FFT in MATLAB. The sensor’s natural frequencies at each
using a customized LabVIEW program at a sampling rate
exposed length was identified and compared to expected the-
of 30 Hz. Having the P‐δ relation, the elastic modulus
oretical values calculated for the corresponding unsupported
was evaluated as E = PL3/3δI on the basis of the Euler–
length.
Bernoulli beam theory. Multiple tests were performed with
varying cantilever lengths. For example, at L = 110 mm,
the P‐δ relation was found to be P = 0.045 δ using linear 3.3.4 | Fixed‐support flow‐induced vibration tests
regression (R2=0.95). This relation along with To further prove the sensing concept, in the next phase of the
L = 110 mm and I = 63.62 mm4 (for the solid circular idealized tests, more realistic flow‐induced vibration condi-
cross‐section of the sensor with a diameter of 6 mm), tions were simulated. These experiments were performed on
was used to estimate E at ~313 MPa. The elastic modulus fixed‐end piezo‐rod sensors so that the effect of forced vibra-
estimated from tests with L = 100 mm to L = 160 mm tion alone could be isolated. The key difference between
ranged from 300 to 320 MPa. E = 310 MPa was used in these tests and those described in Section 3.3.3 was that the
subsequent calculations. sensor was excited by flowing water. To investigate the
6 AZHARI AND LOH

FIGURE 4 The (a) schematic and (b) actual top

view of the soil p‐delta testing setups are shown.
Pushing and pulling forces were recorded
accurately, while lateral displacements (δ) were
incrementally increased and decreased

effects of flow velocity on sensor performance, five different straighten the flow. A plan view of the test setup is illustrated
flow velocities (0.19, 0.21, 0.26, 0.30, and 0.35 m/s) were in Figure 5.
tested for each exposed length between 8 and 20 cm (in
1 cm increments). In addition, each test was repeated with 3.4.2 | Experimental procedure
the piezo‐rod placed perpendicular to its original orientation To ascertain the geometry of the potential scour hole, scour
and such that the PVDF film was oriented parallel to the flow was first simulated by running the flume test with the mock
direction. bridge pier in place but prior to installing the scour sensing
rods. For this test, the flow rate was increased to the maxi-
3.4 | Small‐scale scour validation experiments mum allowed by the flume geometry. According to this
preliminary simulation, it was decided that each scour
After confirming the viability of the piezo‐rod’s scour sens- sensing test should run for at least 20 min to capture the scour
ing concept through various idealized tests, the response of hole evolution sufficiently and adequately. While equilibrium
the sensors was studied in scour conditions simulated in a scour depth was not achieved after 20 min, a deep enough
laboratory flume. The experimental setup and testing proce- scour hole was formed that would still accommodate the
dure are described in this section. piezo‐rod length restrictions in a flume setting. The flow
velocity profile, as shown in Figure 6, was measured period-
3.4.1 | Flume test setup ically using a FlowTracker Handheld Acoustic Doppler Velo-
A clear acrylic laboratory flume with a rectangular cross‐ cimeter from SonTek/YSI (San Diego, CA, USA (Rented
section (45.7 cm wide and 61 cm deep) was used for the scour from HydroScientific West; Poway, CA)). The average
experiments. The full length of the flume was 7.3 m; to reduce current velocity upstream was measured at 0.35 m/s, which
 
the amount of soil placed inside the flume, two boxes span- was just above the mean critical velocity U cr or the
ning the entire width of the flume were placed inside the depth‐averaged threshold speed above which sediment trans-
flume such that the gap between the boxes was approximately port occurs. This indicates that live‐bed scour was initiated.
3.7 m. Another box was placed downstream of the boxes, cre- For the well‐pack sand inside the flume, U cr = 0.329 m/s
ating a sedimentation area to prevent soil from entering the was estimated using van Rijn’s formula[40]:
flume cistern. To simulate a bridge pier, a 7.6 cm‐diameter
acrylic cylinder was mounted vertically inside the flume. U cr ¼ 8:5 ðD50 Þ0:6 log10 ð4h=D90 Þ; (2)
The flume was then filled with a 30 cm‐deep layer of clean
and uniformly graded well‐pack sand (#20 WS from Red Flint where h is the water depth (~0.2 m). The units used for all the
Sand & Gravel, LLC) with D50 = 0.75 mm and D90 = 1 mm; variables are in meters and seconds.
here, D50 and D90 correspond to the particle sizes for which It was postulated that the location of the piezo‐rod sensor
50% and 90% of the soil grains are finer, respectively. A series with respect to the pier may affect the sensing performance.
of 50 cm‐long pipes were also affixed to the upstream box to Therefore, 12 sensor locations (eight unique ones considering

FIGURE 5 The plan view of the flume experiment is illustrated

AZHARI AND LOH 7

various locations around the pier. The lead wires at the bot-
tom of the rods were carefully arranged and attached to the
pier side such that no tugging occurred during the tests to
avoid interfering with sensor outputs. The area around the
piezo‐rods was then filled with soil and leveled using a
trowel. On the basis of the results from the characterization
tests and the preliminary scour simulation, the sensors were
placed such that their exposed length was 9 cm before scour
and increased to no more than 16 cm, so that enough soil sup-
port was maintained. Depending on the sensor arrangement,
some sensors had to be offset by about 1 cm to eliminate
any cross‐sensor effects. The voltage time histories for each
sensor were recorded at a sampling rate of 2,048 Hz using
the DAQ system described in Section 3.2. Actual exposed
lengths were manually recorded for all eight sensors at vari-
ous times during the scour progression using the graduations
marked along their length.

4 | R E S U LTS AN D D I S CU S S I ON
FIGURE 6 The velocity profile of the approach flow was measured using a
handheld acoustic Doppler velocimeter 4.1 | Idealized validation tests
Voltage time history signals from the free vibration tests per-
symmetry) inside the predefined scour hole geometry were formed at different exposed lengths were used to obtain cor-
examined in these experiments (Figure 7). Moreover, it responding power spectral density functions by converting
may be possible to determine the scour hole topography the time‐domain signals to the frequency‐domain using
by installing sensors at various locations around the pier. FFT. A peak‐picking algorithm was implemented in
Figure 8 shows the experimental setup and a representa- MATLAB and was utilized for determining the natural fre-
tive sensor arrangement. Over 10 rounds of scour simula- quency corresponding to each exposed length. The natural
tion tests were completed for this part of the study. The frequencies were then plotted against measured exposed
sensor locations and arrangements were varied in each test lengths, as shown in Figures 9a and 9b for experiments in
such that all the locations shown in Figure 7 were cov- air and water, respectively. As expected, the natural frequen-
ered. Sensor placements were symmetrical in all tests to cies decrease with increased exposed lengths, indicating that
verify repeatability, considering the anticipated symmetri- the piezo‐rod sensors are capable of identifying changes in
cal scour hole. their exposed lengths.
For each round of testing, the soil surrounding the pier A similar method (considering 7 s of time‐domain data)
was partially removed, and eight sensors were driven at was used to process the results from flow‐induced vibration

FIGURE 7 Sensors were arranged symmetrically

with respect to the mock bridge pier throughout
the flume experiments. Eight sensor locations
were examined
8 AZHARI AND LOH

FIGURE 8 The scour sensing experimental setup

is shown. The sensors were interrogated by the NI
DAQ system, while a custom LabVIEW program
digitized and recorded the data

FIGURE 9 For validation purposes, idealized fixed‐end sensors were subjected to (a) free vibrations in air and (b) free and flow‐induced vibrations in water.
Experimental results followed theoretical expectations, and flow rate did not affect sensor response

tests. The measured frequencies are juxtaposed with those of generated with higher flow rates. The frequency‐length rela-
free vibration in Figure 9b. As revealed by Figure 9b, the tion in Equation 1 was derived from the equation of motion
flow‐induced and free vibration frequencies were compara- of a rod of length L in vacuum. The presence of water
ble, both decreasing with increased lengths, and changes in decreases the natural frequency by increasing the effective
water flow rate had no apparent effect on sensor response. mass of the rod.[41] To derive the relation between natural fre-
The effect of PVDF orientation (with respect to the flow) quency and length in water, the following differential equa-
on the piezoelectric voltage response was also investigated tion was considered:
during these experiments. The orientation of the piezo‐rod
had no apparent effect on its frequency response. These ∂ 4 yðx; t Þ ∂2 yðx; t Þ
results further confirmed the sensing concept and theory. EI þ ð μ þ μ Þ ¼ 0; (3)
∂x4 s w
∂t 2
With regards to signal strength, larger voltage amplitudes
were generated when the sensors were placed such that the where μs is the mass per unit length of the sensing rod, and μw
PVDF film was parallel to the flow direction. This was is the commoved mass per unit length, meaning the water
because flow‐induced vibrations are predominantly stronger mass that is displaced by the rod. Therefore, the natural fre-
in the cross‐flow direction. Higher voltages were also quency in water is:
AZHARI AND LOH 9

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 3:5156 EI f vac
f nw ¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n
ffi; (4)
2π L2 ð ρs þ ρw ÞA 1 þ ρw =ρs

where ρs and ρw are the sensor and water densities, respec-

tively. Here, the effect of water was approximated using the
relation for added mass per unit length of a cylinder in quies-
cent and unrestricted domain.[41]
Damping can also affect the sensor’s dynamic response.
To take the calculations one step further, the effect of
damping was also considered:
qffiffiffiffiffiffiffiffiffiffi
f dw ¼ f nw 1−ζ2 ; (5)

where fnw and fdw are the un‐damped and damped natural fre-
quencies of a cantilever submerged in water, respectively, and
ζ is the fluid damping ratio. ζ was estimated to be 0.061 using
the logarithmic decrement method applied to time history
data from the free vibration tests in water. Experimental fre-
quencies obtained from fixed‐end tests followed the theoreti-
cal curves (Figure 9).

4.2 | Sensor calibration

To back‐calculate the exposed lengths from the frequency
measurements during the small‐scale scour experiments in FIGURE 10 For calibration purposes, the sensor was modeled using a
the flume, a calibration curve was required. This could be rotational spring to take into account the effect of soil support in a
achieved through curve‐fitting of experimental data, where simplified manner
the actual exposed lengths are known during scour, as is the
case in this study. The calibration curve could also be
determining k for 9 ≤ Le ≤ 10.5, 11 ≤ Le ≤ 13.5, and
obtained theoretically. Equation 5 approximates the damped
14 ≤ Le ≤ 16 cm, respectively. For each Le range, Equation 6
natural frequency of the sensing rod as a function of its
was used to calculate ϴ from displacement measurements, δ,
exposed length if it were fixed at the bottom. However, the
recorded during the scour tests. An average value of
sensors were embedded in soil, and the support conditions
k = 0.84 N‐m/rad was estimated for the spring coefficient.
were not fixed. The soil‐supported cantilevered sensor was
Eigenfrequency analyses were performed on the sensor
modeled as shown by the schematic of Figure 10. In this sim-
model (shown in Figure 10) using the Structural Mechanics
plified model, a rotational spring was used to account for the
module of the finite element software COMSOL
soil support boundary conditions. The spring stiffness k (in
Multiphysics. The sensing rod was modeled as a beam ele-
units of N‐m/rad) was estimated from the load‐displacement
ment with a circular cross section (6 mm in diameter) and
tests described in Section 3.3.2.
Le varying from 8 to 17 cm using the parametric sweep
Figure 11a shows typical results from the soil‐supported
option. The sensor tip was free, while the boundary condition
load‐displacement tests. M‐ϴ and normalized k‐ϴ plots were
at the supporting end was a combination of pinned and spring
produced as shown in Figures 11b and 12c. The moment
foundation with the rotational spring constant set to
values, M, were obtained by multiplying the applied load
k = 0.84 N‐m/rad. The experimentally‐measured sensor
by the lever arm: M = PL. The angles of rotation, ϴ, were
material properties were used in the model; ρs = 1,142 kg/m3
obtained from Equation 6. The measured displacement, δ, is
and E = 310 MPa. The added mass of water was
composed of the cantilevered flexural deflection
modeled using the software’s added mass option (with an
(δF = PL[3]/3EI) and the rotational deflection, δR = Lθ
added mass per unit volume of ρw = 1,000 kg/m3). The
(assuming small deformations); here, L is 10 cm. Values for
physics‐controlled mesh was set to extremely fine, generating
k were then estimated from the relation M = kϴ.
50 edge elements. Results from the eigenfrequency analyses
  were used as the calibration curve for subsequent scour tests
θ ¼ δ−PL3 =3EI =L: (6)
in the flume. In Figure 12, the calibration curve is compared
As explained in Section 3.3.2, the soil p‐delta tests were against the theoretical curves in the case of idealized fixed‐
performed with three different embedment lengths. k‐ϴ plots end experiments in air and water. As expected, the soil‐
from tests with Le equal to 10, 13, and 15 cm were used for support softens the response for a given exposed length.
10 AZHARI AND LOH

FIGURE 11 Representative soil (a) p‐δ test

results, (b) M‐θ curve, and (c) normalized k‐θ
curve (kmax = 1.2 N‐m/rad) are shown

this study. Such a model would require several assumptions

to consider the interfacial properties between PVDF and
PDMS, which would in turn introduce significant uncer-
tainties. Regardless, the results from such a model would
not be used in the “structural monitoring” aspect of this work.
Instead, the mechanics model provided a sufficiently accurate
calibration curve that relates the rod’s resonant frequency to
its length.

4.3 | Small‐scale scour validation experiments

As described in Section 3.4, small‐scale laboratory scour tests
were conducted to demonstrate the applicability of the pro-
posed piezoelectric sensors for monitoring scour depths. As
FIGURE 12 The sensor calibration curve (from eigenfrequency analyses of the sensors were continuously excited in these experiments,
the simplified model) is compared with idealized fixed‐end frequency the peak‐picking algorithm was revised to obtain the funda-
responses
mental frequencies at 1‐s intervals, each time considering a
2‐s window of voltage time history data, as illustrated in
It is important to note at this point that the COMSOL Figure 13. Representative results from three sensor locations
model was strictly a structural model and not one that con- close to the mock pier are presented in Figure 14. The decreas-
siders electromechanical effects (i.e., the piezoelectric sens- ing trend of natural frequencies with time indicated increased
ing effect). The goal of the model was to simulate the exposed lengths as a result of increased scour depths. As
cantilevered rod’s mechanical response, considering a rota- depicted by the levels of scatter in the results, if placed on
tional spring at its support to model the non‐rigid connection the side of and closer to the pier (S4), the sensor would
due to soil. An analytical solution for computing the natural respond more definitively. Results from sensor S4 were, there-
frequency of such a system can be derived and used for fore, used for direct comparisons with observed scour.
eigenfrequency calculations. In this study, a simple finite ele- Scour depths calculated from sensor S4 were strongly
ment model was chosen as an alternative. The software correlated to the observed scour depths as depicted by
COMSOL was used in lieu of other finite element modeling Figure 15a where the time history of scour depth is plotted
software because of convenience and familiarity with the for S4 in one scour experiment. The yellow diamond symbols
software and not for its multi‐physics capabilities. Although represent the observed scour depths (recorded intermittently
one, in principle, could incorporate the piezoelectric element during the scour tests), while the blue dots depict the scour
and simulate the multi‐physics effect, this was not a focus of depth obtained from the sensor. The black dashed line
AZHARI AND LOH 11

FIGURE 13 A peak‐picking algorithm was used to extract natural frequencies from the voltage time history data at 1‐s intervals (each time considering 2 s of
time history data) during scour hole evolution

FIGURE 14 The time histories of the natural frequency of the scour sensors at three locations around the pier are plotted

FIGURE 15 Scour depths calculated from the

sensor located at the side of the pier (S4) are
strongly correlated to the observed scour depths.
(a) The time history of scour depth is plotted for
S4 in one scour experiment. (b) Scour depths
calculated from S4 in five tests are directly
compared with the observed scour depths

represents the moving average of scour sensor measurements. depths calculated from sensor S4 in four additional tests are
The moving average was used to directly compare sensor also compared with the observed scour depths in Figure 15b,
measurement with observed scour in Figure 15b. Scour which show that sensor response is repeatable. The measured
12 AZHARI AND LOH

and observed values in these direct comparisons are very close Another notable outcome from the small‐scale scour
to the equivalency line. This implies that the piezo‐rod sensor experiments was determining the piezoelectric sensor’s abil-
could determine scour depths with reasonable accuracy; the ity to detect sedimentation downstream of the pier.
root‐mean‐square error was 0.33 cm. The small deviations Figure 16 displays the time history of measured and observed
may originate from model approximations (e.g., the elastic scour depths for sensor S8, located downstream and farther
modulus, spring stiffness, and so forth), as well as from obser- from the pier. As depicted by the negative values on the ver-
vational errors and estimations when recording the actual tical axis, there was no scour at this location but rather accu-
exposed lengths. In large‐scale implementations, these effects mulation of sediments. Sensor S8 was adequately capable of
may be less pronounced depending on the sensor dimensions detecting this accumulation. The main reason behind the sen-
and soil properties. sor’s apparent under‐prediction of sediment deposition is that
the thin top layer of soil around the sensor was fairly loose
and did not provide much support. The scour depth observa-
tions, on the other hand, were recorded on the basis of the
visible length of the sensor and did not account for the effect.
While the side of the pier (where sensor S4 was located)
was chosen as the most promising location for this type of
sensor, all sensor locations provided acceptable responses
(considering the scale of the experiments). Therefore, having
an array of sensors at various locations, one can plot the scour
hole topography at any given time during scour progression.
Knowing the scour hole dimensions and topography is useful
in countermeasure design and in determining the required
extent of countermeasures.[14] In Figure 17, the final sensor
frequencies (i.e., after 20 min of scour) were used to deter-
mine the scour hole topography. In this depiction, a small
amount (0.5 cm) of general scour was taken into account,
and a third‐order spline interpolation was used to realize the
scour depths between sensor locations. The detected scour
hole was found to be comparable to the observed scour hole,
also shown in Figure 17.
FIGURE 16 The time history of scour depth is plotted for the sensor located Although this simple sensing scheme and straightforward
downstream of the pier (S8) where sediment accumulation occurred data processing deem piezo‐rod sensors very effective in

FIGURE 17 The scour hole topography after 20 min of scour testing was determined using experimental data from all of the sensors instrumented in the
vicinity of the mock bridge pier
AZHARI AND LOH 13

FIGURE 18 Preliminary design for a potential

large‐scale piezoelectric scour sensor is presented
along with the corresponding calibration curves
for various values of rotational spring constant k

detecting scour hole topographies, installation difficulties systems have been developed and deployed over the years
(mainly due to wires) and vulnerability to debris may hinder to prevent scour‐induced failures through timely detection
their use for this purpose. For example, debris may restrict of scour depths. Many of these sensing schemes, however,
the motion in the lower part of the sensing rod, effectively suffer from issues including signal inaccuracy, installation
reducing the perceived exposed length. As such, it is sug- difficulty, and high costs. In this work, a simple piezoelectric
gested that one sensor be placed at a critical location close sensor was developed and examined for use in continuous,
to the side of the pier. real‐time scour detection and scour depth monitoring. The
In a real application, the sensor dimensions would be much sensor’s design was based on a piezoelectric PVDF thin film
larger, and the PVDF film would be encased in a casing made encased in a plastic sheath to create a sensing rod that could
of a stiffer and more robust material. A preliminary design of a be driven into the soil close to the pier. The relation between
large‐scale sensor (with a diameter about the size of a cone the dynamics of the voltage time history response generated
penetrometer) is considered in Figure 18. This potential by the piezoelectric sensor and the exposed length of the sen-
large‐scale sensor has a steel casing with an outside diameter sor could be used to reliably back‐calculate the sensor’s
of 3.6 cm, and is 10 m long. The calibration frequency range exposed length and, in turn, scour depth.
can be predicted using the same type of model used for the pro- Extensive laboratory experiments were performed to val-
totype sensors. Figure 18 provides calibration curves relating idate the sensing mechanism. In these experiments, the
the natural frequencies of such a scaled piezo‐rod sensor to response of the sensor with idealized support and excitation
exposed lengths of 20 to 300 cm, assuming various values of conditions were examined and confirmed through compari-
k. For higher values of k, the effect of support fixity is reduced, sons with approximate theoretical calculations. Subsequently,
meaning that the sensor can potentially be modeled as a fixed‐ the performance of the sensors was studied in simulated scour
end cantilever while still maintaining reasonable accuracy. conditions in a laboratory flume. Eight sensors were driven at
The large‐scale sensor could also be calibrated on the basis different locations around the pier. Natural frequencies were
of the Winkler model, routinely used in the analysis and design continuously extracted from the voltage time history
of laterally–loaded piles, which is fully described in American responses of the sensors using a peak‐picking algorithm and
Petroleum Institute codes.[42] In a Winkler model, a series of FFT. While all sensor locations provided acceptable
uncoupled lateral springs along the pile length represent the responses (considering the scale of the experiments), the side
pile‐soil interaction. Alternatively, an appropriate value for k of the pier (where sensor S4 was located) was chosen as the
could be obtained through the frequency response of the sen- most promising location for this type of sensor. Sensor natu-
sor at the initial known exposed length. ral frequencies decreased with increasing exposed lengths,
thereby indicating an increase in the scour depth. The
piezo‐rod sensor was also proven effective in detecting soil
5 | S UM M A RY A N D C ONCLU S IO N S deposition downstream of the pier, where its natural fre-
quency increased with decreasing exposed lengths.
Flow‐induced vortices (e.g., in the event of a flood) can A simplified cantilevered beam model with a rotational
locally scour sediments from around overwater bridge piers spring at the base was used to perform eigenvalue analyses
and abutments. If the scour hole is extended over a critical that provided an approximate theoretical calibration curve,
depth, it can cause significant damage, instability, and poten- which was used to back‐calculate scour depths from the sen-
tially catastrophic failures. A number of promising sensing sors’ frequency responses. Comparing the measured and
14 AZHARI AND LOH

observed scour depths at the side of the pier (S4) confirmed [18] E. Richardson, S. Davis. Evaluating Scour at Bridges, Hydraulic Engineering
that the piezo‐rods are capable of detecting scour depths Circular No. 18 (HEC‐18). US Department of Transportation, Federal High-
way Administration, 2001.
accurately. Results from all sensor locations were also used
[19] B. Melville, Y. Chiew, J. Hydraul. Eng. 1999, 125(1), 59, DOI: 10.1061/
to plot the scour hole topography. Future research will focus (ASCE)0733-9429(1999)125:1(59)
on large‐scale implementations of these sensors and their [20] S. Benedict, N. Deshpande, N. Aziz, Trans. Res. Record J. Trans. Res. Board
effectiveness in harsh conditions, where debris impact may 2007, 2025(1), 118.
become a challenge. [21] S.‐U. Choi, S. Cheong, JAWRA Journal of the American Water Resources
Association 2006, 42(2), 487, DOI: 10.1111/j.1752-1688.2006.tb03852.x

AC KNOWLEDGMENTS [22] B. Mutlu Sumer, J. Hydraul. Res. 2007, 45(6), 723, DOI: 10.1080/
00221686.2007.9521811
The authors gratefully acknowledge support from the U.S. [23] M. Lueker, J. Marr, C. Ellis, A. Hendrickson, V. Winsted. Bridge Scour
National Science Foundation (NSF grant no. CMMI‐ Monitoring Technologies: Development of Evaluation and Selection Proto-
1234080). Additional support was provided by the American cols for Application on River Bridges in Minnesota, in Proc. 5th Int. Conf.
on Scour and Erosion (ICSE‐5), San Francisco, ASCE, 2010.
Society of Civil Engineers (ASCE) Freeman Fellowship and
[24] J. R. Richardson, G. R. Price, E. V. Richardson, P. F. Lagasse. Modular mag-
the College of Engineering, University of California, Davis.
netic scour monitoring device and method for using the same, Google
Mr. Peter Scheel, Dr. Dan Wilson, and Dr. William Fleenor Patents: United States of America, 1996.
are also recognized for their assistance throughout this study. [25] F. De Falco, R. Mele, NDT E Int. 2002, 35(2), 117.
In addition, Prof. Fabian A. Bombardelli is acknowledged for [26] S. G. Millard, J. H. Bungey, C. Thomas, M. N. Soutsos, M. R. Shaw, A.
his collaboration. Patterson, NDT E Int. 1998, 31(4), 251.
[27] C. Yao, C. Darby, S. Hurlebaus, G. Price, H. Sharma, B. Hunt, O. Yu, K.
REFERENCES Chang, J. Briaud. Scour Monitoring Development for Two Bridges in Texas,
in Proc. 5th Int. Conf. on Scour and Erosion (ICSE‐5), ASCE, San
[1] R. Whitehouse, J. Harris, T. Mundon, J. Sutherland. Scour at offshore struc-
Francisco, 2010.
tures, in Proc. 5th Int. Conf. on Scour and Erosion (ICSE‐5). San Francisco,
ASCE, 2010. [28] J.‐L. Briaud, S. Hurlebaus, K.‐A. Chang, C. Yao, H. Sharma, O.‐Y. Yu, C.
Darby, B.E. Hunt, G. R. Price. Realtime monitoring of bridge scour using
[2] L. Deng, C. Cai, Pract. Period. Struct. Des. Construct. 2010, 15(2), 125,
remote monitoring technology, Texas Transportation Institute, Texas A&M
DOI: 10.1061/(ASCE)SC.1943-5576.0000041
University System, 2011.
[3] M. Beg Characteristics of Developing Scour Holes around Two Piers Placed
[29] X. Yu, X. Yu. Field Monitoring of Scour Critical Bridges: A Pilot Study of
in Transverse Arrangement, in Proc. 5th Int. Conf. on Scour and Erosion
Time Domain Reflectometry Real Time Automatic Bridge Scour Monitoring
(ICSE‐5). San Francisco, 2010.
System, Ohio Department of Transportation, 2010.
[4] A. Molinas, S. Jones, M. Hosny, Trans. Res. Record J. Trans. Res. Board
[30] X. Kong, C. Cai, S. Hou, Smart Mater. Struct. 2013, 22(5), 055011.
1999, 1690(1), 164.
[31] Y. B. Lin, J. S. Lai, K. C. Chang, W. Y. Chang, F. Z. Lee, Y. C. Tan, J. Chin.
[5] B. W. Melville, A. J. Raudkivi, J. Hydraul. Eng. 1996, 122(4), 203.
Inst. Eng. 2010, 33(1), 25.
[6] Z. Bozkus, O. Yildiz, J. Hydraul. Eng. 2004, 130(8), 827, DOI: 10.1061/
[32] G. Chen, D. Pommerenke, R. Zheng. Smart Rocks and Wireless Communi-
(ASCE)0733-9429(2004)130:8(827)
cation Systems for Real‐ Time Monitoring and Mitigation of Bridge Scour,
[7] B. Ataie‐Ashtiani, A. Beheshti, J. Hydraul. Eng. 2006, 132(10), 1100, DOI: Missouri University of Science and Technology, 2012.
10.1061/(ASCE)0733-9429(2006)132:10(1100)
[33] Y. B. Lin, J. C. Chen, K. C. Chang, J. C. Chern, J. S. Lai, Smart Mater.
[8] N. Apsilidis, P. Diplas, C. Dancey, P. Vlachos, S. Raben. Local Scour at Struct. 2005, 14(4), 664.
Bridge Piers: The Role of Reynolds Number on Horseshoe Vortex Dynam-
[34] W. Xiong, C. S. Cai, X. Kong, Appl. Opt. 2012, 51(5), 547.
ics, in Proc. 5th Int. Conf. on Scour and Erosion (ICSE‐5). San Francisco,
ASCE, 2010. [35] Z. Zhou, M. Huang, L. Huang, J. Ou, G. Chen, Adv. Struct. Eng. 2011, 14(1), 67.
[9] B. E. Hunt, Trans. Res. Record J. Trans. Res. Board 2005, 531. [36] F. Azhari, P. Scheel, K. J. Loh, Struct. Monit. Maint. 2015, 2(2), 145, DOI:
10.12989/smm.2015.2.2.145.
[10] J. Lu, J. Hong, C. Su, C. Wang, J. Lai, J. Hydraul. Eng. 2008, 134(6), 810,
DOI: 10.1061/(ASCE)0733-9429(2008)134:6(810) [37] J. L. Briaud, F. C. K. Ting, H. C. Chen, Y. Cao, S. W. Han, K. W. Kwak,
J. Geotech. Geoenviron. Eng. 2001, 127(2), 105.
[11] Y. B. Lin, K. C. Chang, C. C. Chen, S. C. Wong, L. S. Lee, Y. K. Wang, M.
H. Gu, Int. J. Autom. Smart Technol. 2011, 1(2), DOI: 10.5875/ausmt. [38] A. Zarafshan, A. Iranmanesh, F. Ansari, J. Bridge Eng. 2011, 17(6), 829.
v1i2.127 [39] C.‐Y. Wang, H.‐L. Wang, C.‐C. Ho, Adv. Struct. Eng. 2012, 15(6), 897.
[12] J. Hong, Y. Chiew, J. Lu, J. Lai, Y. Lin, J. Hydraul. Eng. 2012, 138(2), 186, [40] L. van Rijn, J. Hydraul. Eng. 1984, 110(12), 1733, DOI: 10.1061/(ASCE)
DOI: 10.1061/(ASCE)HY.1943-7900.0000430 0733-9429(1984)110:12(1733)
[13] R. L. Parsons, C. Bennett, J. Han, C. Lin, Case history analysis of bridge fail- [41] R. D. Blevins, Flow‐induced vibration, Van Nostrand Reinhold Co., Inc.,
ures due to scour, in Climatic Effects on Pavement and Geotechnical New York 1990, 451.
Infrastructure, ASCE Publications, Fairbanks, Alaska, USA 2014, 204.
[42] American Petroeum Institute (API), API Recommended Practice 2A‐WSD:
[14] G. K. Butch. Scour‐hole dimensions at selected bridge piers in New York, in Planning, Designing and Constructing Fixed Offshore Platforms ‐ Working
Stream Stability and Scour at Highway Bridges: Compendium of Stream Sta- Stress Design, Twenty‐Second Edition. 2014, 310.
bility and Scour Papers Presented at Conferences Sponsored by the Water
Resources Engineering (Hydraulics) Division of the American Society of
Civil Engineers. ASCE, 1996.
How to cite this article: Azhari, F., and Loh, K. J.
[15] L. Prendergast, K. Gavin, J. Rock Mech. Geotech. Eng. 2014, 6(2), 138.
(2016), Laboratory validation of buried piezoelectric
[16] J. Wingrove Train cars carrying petroleum products safely removed from
partially collapsed Calgary bridge, in The Globe and Mail, The Globe and scour sensing rods, Struct Control Health Monit, doi:
Mail, 2013. 10.1002/stc.1969
[17] Federal Highway Administration. National Bridge Inventory Database, 2014.