wireless intelligent sensor and actuator network – a scalable
TRANSCRIPT
Wireless Intelligent Sensor and Actuator
Network – A Scalable Platform for
Time-synchronous Applications of
Structural Health Monitoring
Edward Sazonov,* Vidya Krishnamurthy and Robert Schilling
Department of Electrical and Computer Engineering, Clarkson University
Potsdam, NY, USA
Wireless sensor networks have attracted attention as a possible solution for applications
of periodic and continuous structural health monitoring. Ensuring synchronous data acqui-
sition across wireless nodes in large networks of sensors spatially distributed on a structure
is of critical importance for many methods of structural health monitoring, especially those
based on analysis of vibration. In this article we present a novel Wireless Intelligent Sensor
and Actuator Network (WISAN) addressing the issue of scalability for applications of struc-
tural health monitoring. We also present a novel time synchronization algorithm that can
keep the synchronization error between any number of globally distributed sensors nodes
less than �23 ms. We show proof of stability for the time synchronization algorithm.
We validate WISAN in laboratory experiments, testing the actual time synchronization
between randomly selected sensors in a complex network. Finally, we validate WISAN in
a field experiment by reconstructing mode shapes of a highway bridge.
Keywords wireless sensor networks � modal analysis � time synchronization.
1 Introduction
With the need for reduction in size and cost of
structural health monitoring (SHM) systems, wire-
less sensor networks (WSNs) are becoming increas-
ingly popular in these application areas [1–13]. A
monitoring system using sensors placed at various
locations on the structure is needed to monitor the
health of civil structures such as bridges. The eco-
nomic losses associated with disasters are much
higher as compared to the cost for monitoring the
structure’s health and performing early repairs
avoiding major calamities [14]. Among the different
available damage detection techniques, the class of
vibration-based damage detection techniques is one
of the most popularly used methods to monitor the
condition of structures [15–18]. As a specific case of
vibration techniques, mode shape-based damage
detection usually involves comparing specific
properties of both damaged and undamaged
mode shapes and determining locations with the
pronounced difference in properties [19–21].
*Author to whom correspondence should be addressed.
E-mail: [email protected]
Figures 1 and 6–9 appear in color online: http://shm.sagepub.com
� The Author(s), 2010. Reprints and permissions:
http://www.sagepub.co.uk/journalsPermissions.nav
Vol 9(5): 465–12
[1475-9217 (201009) 9:5;465–12 10.1177/1475921710370003]
465
Thus, the first step for damage detection is modal
identification and reconstruction of mode shapes.
A monitoring system using time-synchronized sen-
sors placed at various locations on the structure is
needed to acquire vibration response. The task of
vibration acquisition poses additional requirements
for the WSN, specifically, need for time-synchro-
nized acquisition among sensors; lossless and pref-
erably real-time data transmission; scalability to
various structure sizes and number of deployed sen-
sors. Wireless Intelligent Sensor and Actuator
Network (WISAN) has been developed to address
these issues and support arrays with large number
of heterogeneous sensors.
Time synchronization in WSNs has been an
important concern that restricted the application
of these networks in SHM applications. In our pre-
vious research [22] we have shown that for the
range of natural frequencies typical of a highway
bridge, synchronicity on the order of microseconds
is required between the wireless sensors to main-
tain signal acquisition errors on the level compara-
ble to sensor noise. Consider a network with each
wireless sensor in the network has its own hard-
ware oscillator-based clock providing timing sig-
nals. Due to varying oscillator frequencies, the
clocks on the sensors drift with respect to each
other and have to be continuously synchronized
with each other in order to maintain a common
time. While a number of methods have been devel-
oped and tested in this direction [23–27] most of
them were limited to simulation studies. As one of
the notable practical results, time synchronization
within 0.1ms [25] has been achieved using global
beacon synchronization where each wireless sensor
receives the same beacon ping and sets its internal
clock to 0. Another practical implementation of a
sensor network [28], which used a beacon-based
synchronization method, was tested on a highway
bridge in South Korea. The sensors were initially
synchronized to a 20 ms inter-sensor precision, but
accumulated up to a 5ms delay in a 6-s period.
While the complete review of available synchroni-
zation mechanisms goes beyond the scope of this
article, existing wireless networks presented in
[2–13,29–32] do not completely satisfy all the
requirements (scalability with structure size, data-
stream reliability, and time synchronization) posed
by SHM applications. Hence, there is a need for a
WSN platform that can scale up or down depend-
ing on structure size while keeping the sensors
tightly synchronous.
Satisfying both the scalability and the time syn-
chronization requirements is a challenging prob-
lem. For example, global beacon synchronization
can keep the time synchronized between sensors
within a desired upper bound of a few microsec-
onds, but this method does not scale well with the
size of the network. Indeed, presence of a common
beacon assumes that all sensors share the band-
width of a frequency channel. For applications of
vibration acquisition the channel capacity can be
quickly exhausted by a few sensors even if use of
the bandwidth is optimized by a reservation sched-
uler [33]. To resolve this dilemma we propose a
hierarchical architecture in which beacon synchro-
nization is used for local clusters of wireless sensors
and spatially distributed clusters are synchronized
by using GPS time reference.
In this article we present a detailed description
of the WISAN network architecture; the time syn-
chronization algorithm developed and imple-
mented along with proof of stability; and
laboratory and field validation tests performed
using the network. WISAN hardware and software
is discussed in Section 2. Section 3 highlights the
time synchronization algorithm implemented in
WISAN and establishes theoretical synchroniza-
tion limits. Section 4 describes the laboratory and
field tests conducted on WISAN. Results are pre-
sented in Section 5, followed by the Discussion and
Conclusions.
2 The WISAN Platform
WISAN hardware platform was built around
the ultra-low power MSP430 microcontroller
MSP430F1611 from Texas Instruments [34] and
CC2420 radio transceiver from Chipcon (now
also Texas Instruments) [35]. WISAN is fully com-
patible with IEEE 802.15.4 and can be utilized
worldwide in the 2.4GHz ISM frequency band
and coexist with Wi-Fi and other devices.
Detailed description of WISAN hardware can be
found in [36]. The network hardware is built using
two types of devices: The Coordinator and The
Wireless Sensor.
466 Structural HealthMonitoring 9(5)
The Coordinator (Figure 1(a)) maintains a
cluster of sensor nodes: sends commands, receives
and stores data, processes and/or forwards data to
a computer via a USB link. An input for the pulse-
per-second (PPS) signal from a GPS receiver
enables hardware clock synchronization with a
GPS time reference. The wireless sensors
(Figure 1(b)) are used for acquiring data and deliv-
ering them to a Coordinator. Each wireless sensor
is equipped with the peripheral circuitry for inter-
facing of sensors and actuators. The peripheral cir-
cuitry consists of a variety of components that may
be selected on application needs. A low power 3D
MEMS accelerometer (LIS Accelerometer [37]) is
used as internal on-board sensor. A gain and offset
compensation stage is provided to improve the res-
olution. Signal conditioning is done using a fifth-
order low-pass filter with cutoff frequency config-
urable over the wireless link. The noise resolution
of the accelerometer in the frequency band of
100Hz is 0.5mg. An optional buffered output
terminal allows sourcing of control and excitation
signals that can be used as actuation signals. A
connector is provided for interfacing an external
acceleration sensor. Differential signal condition-
ing circuitry is also provided with a gain of 1000
to interface high precision strain sensors. The wire-
less sensors are enclosed using commercially avail-
able NEMA4 enclosures. All circuits feature
shutdown inputs, so a part or the whole wireless
device can be put into low-power sleep mode.
WISAN software has been developed to effi-
ciently work with low power hardware and satisfy
the requirements of a SHM system. The network
protocol is the key ingredient of any networked
system, since it is responsible for reliable, error-
free, and on time data delivery. The medium
access protocol network software is based on the
low-latency, low-power IEEE 802.15.4 protocol,
which has emerged as a new standard for low
rate wireless personal area networks (LR-PANs)
[38] and hence chosen for WISAN. Since the
IEEE 802.15.4 protocol only provides the lower
layers, that is, the physical layer and the medium
access Layer, the software development for the net-
work also focused on developing application and
transport layers of the network. Reliable data
delivery and substantially reduced power con-
sumption were achieved by a reservation-based
scheduler implanted at the Coordinator level [33].
The network architecture of WISAN is a clus-
ter tree. This choice is motivated by the fact that
wireless sensors remain static on the structure and
thus allow utilization of energy efficient hierarchi-
cal cluster architecture where most of the commu-
nications between sensor nodes and dedicated
cluster heads (Coordinators) is performed in
single hop. A diagram of a WISAN network is
shown in Figure 2. A certain number of sensors
(within the maximal available bandwidth) are asso-
ciated with a particular Coordinator on one of the
16 frequency channels. The spread of a cluster is
limited by the transmission range of the sensors
(�30m ) and a maximum of 16 clusters can be
co-located in a given area enabling dense place-
ment of sensors on a structure. Each Coordinator
forwards the data collected within the cluster to a
TCP server which can be accessed by a LabView
client over a high-capacity network link (e.g., wire-
less or wired Ethernet, cellular link). The proposed
network architecture is highly scalable and allows
building networks of large size (hundreds or even
thousands of nodes) distributed over large
structures.
3 Time Synchronization Algorithms
Multiple clusters of wireless sensors covering
the monitored structure need to be closely
(a) (b)
Figure 1 (a) WISAN Coordinator, (b) WISAN sensor node.
Sazonov et al. Wireless Intelligent Sensor and Actuator Network 467
synchronous to be used with vibration-based
damage detection. The synchronization should
not depend on the network size (e.g., number of
nodes in the network or physical area spanned by
the network). In the proposed hierarchical network
shown in Figure 2, each wireless sensor connects to
a Coordinator through a low-latency single-hop
connection. The Coordinators are interfaced to
the Application client via a high-bandwidth TCP
link. Thus, time synchronization in the network
has to be maintained at two levels:
(1) Within a cluster (all sensors that belong to a
Coordinator), and
(2) Between clusters (among all Coordinators).
3.1 Time Synchronization Between Sensors
Within a Cluster
At the lower hierarchical level, sensors in the
same cluster are periodically synchronized to the
beacons emitted by the Coordinator. Each time a
beacon packet arrives, the radio chip generates a
pulse, which in conjunction with a capture-and-
compare module of the microcontroller, corrects
the internal timer by tSENSORd ¼ tTMR � tBCN
(Figure 3) to compensate for accumulated time dif-
ference. The 8MHz crystal oscillator used for the
sensors and the Coordinators has a frequency tol-
erance of �30 ppm. Hence, the maximum time dif-
ference between a sensor node and its Coordinator
operating at 8MHz would be 60 ms/s. Since in
WISAN the beacons are emitted every 245.76ms
(beacon order¼ 4) the maximal drift between a
Coordinator and a sensor synchronizing on beacons
is less than �15ms (Tsensormax ). This is the maximum
amount of accumulated time error that exists
between any two wireless sensors in the cluster.
3.2 Time Synchronization Between
Coordinators
Each cluster will be synchronous with the
global time if all Coordinators emit beacons at
the same time. A GPS receiver can be used to pro-
vide reference time for one or more Coordinators.
Most GPS receivers are capable of generating a
hardware PPS signal, which is synchronized to
Coordinated Universal Time (UTC) within
�100 ns [39]. It is not feasible to use GPS for indi-
vidual synchronization of every node in the net-
work due to power and visibility considerations.
A GPS receiver takes a significant amount of
power to operate and would reduce operational
time of the sensor nodes before requiring battery
replacement. Most importantly, GPS needs a clear
view of the sky, which is impossible for most
TCP server
Application client: Control and data collection
WISANsensors
Long-range communications
A cluster
TCP link
Coordinator 1
Coordinator 2
Coordinator 3
GPS
TCP link
TCP server
Serial connection
Coordinator 4
Coordinator 5
Coordinator 6
GPS
Short-range communications
Figure 2 WISAN network topology.
468 Structural HealthMonitoring 9(5)
sensors. However, the Coordinators can be located
anywhere on the structure within the transmission
range of the sensors and thus could easily utilize
GPS time reference. The PPS signal can be used to
synchronize spatially distributed Coordinators by
using it in the same manner a beacon is used within
a cluster (Figure 4). The Coordinators run on an
8MHz clock or 0.125 ms clock interval. Emitted
beacons are timed by a sequence of two timers:
Timer B and MAC timer. Under conditions of per-
fectly stable crystal frequency the MAC timer
counts 3125 times in 1 s (or 768 times between con-
secutive beacons). Given the maximum clock drift
of �30ms in one second [40], if the Coordinator is
slow relative to UTC then upon arrival of a PPS
signal the MAC timer count will be þ3124 (relative
to the value at arrival of the previous PPS signal)
and Timer B count will be less than the maximum
8 MHzOscillator
Timer B(0–2559)
0.125µs MAC timer(0–767)
Capture module
320 ms
GPSTPPS=1stPPS
tTMR
Dt = tTMR − tPPS
Synchronize with PPS:
•ticks of MAC timerAdjust Timer B by a every Bi
• Reset Timer B every TPPS (correct by τ ′ i =τ i − (Di −Di−1 + ε i ) ).
• Avoid double MAC timer update in case of fast Timer B.
RadioEmit beacon
245.76ms
Figure 4 Time Adjustments in the Coordinator time synchronization algorithm.
8 MHzoscillator
Timer B(0–2559)
0.125µs
ADC
Programmabledivider
Capturemodule
320 ms
Samplingtrigger
Sensor
RadioA beacon from
Coordinator every245.76ms
Correct timer B by
tdSENSOR= tTMR − tBCN
tTMR
tBCN
Figure 3 Time adjustments in the sensor node.
Sazonov et al. Wireless Intelligent Sensor and Actuator Network 469
(2560) but above the half count of 1280. If the
Coordinator is fast, the Timer B counter will be
in the range of lower half (0–1280) with a relative
MAC timer count of þ3125. If Timer B is reset to
zero and the MAC timer set to a relative count of
þ3125 upon capturing a PPS time reference, the
Coordinators will be almost perfectly time syn-
chronous, however, only at that moment in time.
The problem with this approach is that adjusting
the time difference in one step can create up to
120ms time jumps due to time differences accumu-
lated between consecutive PPS pulses. Such a large
degree of asynchronicity is unacceptable for appli-
cations of modal identification. To avoid accumu-
lation of the time errors the clock drift can be
compensated by making a number of small adjust-
ments distributed across the period TPPS. The
upper bound error between consecutive MAC
timer ticks can be minimized and kept within rea-
sonable bounds.
A more formal presentation of the procedure is
as follows. At any given time instant t, let the
actual clock drift given as �(t) such that
�(t)¼ �(t0)þR � (t� t0)¼ �(t0)þDt where t is the
current time, �(t0)� is the time offset at the last
measurement update t0, R is the frequency toler-
ance of the crystal and Dt is the drift at time t with
respect to t0. In WISAN, R can have a maximum
difference between two Coordinators of 60 parts-
per-million (PPM) due to the tolerances of the
crystal oscillator. If not corrected, the resulting
errors can accumulate to the order of a few seconds
per day. The time synchronization algorithm esti-
mates �(t0) and R at regular intervals Tpps (1 s) and
adjusts the clock to minimize �(t) in the future. For
any PPS interrupt i�1 from the GPS receiver, all
Coordinators set Timer B to zero (�i�1¼ 0). The
accumulated time difference on the next, i-th inter-
rupt is given by �i� �i�1¼R � (Tpps)¼Di.
Since Di is proportional to the crystal tolerance
R it is safe to assume that under most realistic con-
ditions value Di remains approximately constant
over the period of Tpps. Then Di can be compen-
sated over the next Tpps period in a fixed amount
A of time adjustments (increments/decrements)
of fixed size �, thus keeping the maximum synchro-
nization error significantly less than Di. The
number of adjustments Ai and adjustment period
Bi are calculated as: Ai ¼ Int Di=�� �
, where
Bi ¼ Int 3125=Ai
� �and Int . . .h i is an operator that
rounds-off the division to the lower integer (all
timers are integer devices). The adjustment period
B1 is further used to calculate the actual number of
adjustments made for Tpps¼ 1 s, A0i ¼ Int 3125=Bi
� �.
Hence, the errors induced due to the fact that
the Timer B can be adjusted only in integer incre-
ments/decrements would be "i ¼ Di � A0i� where
A0i is the number of adjustments and � is the
size of the adjustments. At the next PPS inter-
rupt the difference between the UTC refer-
ence and internal clock would be, �iþ1 ¼
Diþ1 �Di þ "i ¼ Diþ1 � A0i�. The value of A
needs to be adjusted if "i4�, where � is the error
correction threshold which is the maximal error
obtained by incrementing or decrementing Biþ1
one step above or below Bi.
The practical implementation of the Coordina-
tor time synchronization algorithm consists of two
independent parts: parameter estimation and timer
adjustment. The parameter estimation part is
shown in Figure 5. Every Tpps the time correction
parameters Ai and Bi are recalculated according to
the observed error. The time adjustment part is
shown in Figure 4. Timer B is incremented of dec-
remented by � (4 timer ticks or 0.5 ms) every Bi ticks
of MAC timer. At every TPPS the Timer B
1. Initialize variables and parameters
2. At PPS interrupt i do
a. Compute
b. if ti > Λ
else
i = i +1
Ai = Ai –1Bi = Bi –1 Ai = Ai –1
t0 = 0, A0 = 0, B0 = 0,D0 = 0, 0 = 0, Λ = 0, i = 0,a = 0.5 ms, N = 3125
§
Di = tTMR – TPPSti = Di – Di –1 + i –l
§
Ai = Int ⟨Di /a⟩
Ai′ = Int ⟨N /Bi⟩Bi = Int ⟨N /Ai⟩
i = Di –Ai′a
§
i = Di –Ai a,§= =i = Di –Ai a
§− −
Λ = max { , }§=i
§
i
=Ai = Int ⟨N /(Bi + 1)⟩, Ai = Int ⟨N /(Bi – 1)⟩
−
−
Figure 5 Parameter estimation in the Coordinator timesynchronization algorithm.
470 Structural HealthMonitoring 9(5)
is corrected by a reset to zero to avoid accumula-
tion of the integer arithmetic error �0i ¼
�i � ðDi �Di�1 þ "iÞ. Since such a reset may cause
an automatic increment of the MAC timer, a cor-
rection is made to avoid an update of the MAC
timer in case Timer B is faster than TPPS.
The stability of the proposed Coordinator time
synchronization algorithm can be proven under
conditions of slowly changing synchronization
error. When Diþ1 � Di and "ij j � ", then
�iþ1�� �� � " and the algorithm keeps the error clock
offset error within the minimal computation bound
and hence stable.
Given the crystal frequency stability of
�30 ppm, a maximum drift on each Coordinator
is �30 ms, the maximum time difference accumu-
lated between two Coordinators is �60ms/s (480
timer B ticks). To find the bound " on this error,
numerical computation of the error "i was per-
formed for a range of accumulated differences
between 0 and 120ms. Figure 6 shows that the
error due to integer round-offs lies between 0 and
8 ms suggesting an upper bound " of �8 ms. Thus,
using the time synchronization algorithm, all
Coordinators in the network run on virtually iden-
tical time synchronized within an accuracy of �8 ms
(TPANmax ).
The time synchronization mechanism in
WISAN relies on a simple hierarchical organiza-
tion where synchronous operation of the global
network is achieved by keeping wireless sensors
tightly synchronized to a Coordinator and by
keeping all Coordinators tightly synchronized to
GPS time references. Overall, the maximal syn-
chronization error among sensors in WISAN
network (TWISANmax ¼ TPAN
max þ Tsensormax ) is less than
�23ms. If, however, a crystal of �10 ppm tolerance
is used in WISAN wireless nodes, the maximal syn-
chronization error using the above algorithm
would be less than �6 ms.
4 Experimental Analysis
4.1 Laboratory Tests
Experiments were conducted to verify the per-
formance of the proposed time synchronization
algorithm. The first experiment tested synchroniza-
tion of sensors within a cluster. A sinusoidal signal
from a signal generator was physically wired to
two wireless sensors associated with the same
Coordinator. Time differences in sampling directly
correspond to a shift in the phase of the acquired
sinusoidal signal. An FFT routine is thus applied
to determine the amplitude and phase of the
received data. Let f1(t) and f2(t) be the two received
signals such that f1ðtÞ ¼ sinð!0tþ �1Þ and
f2ðtÞ ¼ sinð!0tþ �2Þ. The respective Fourier trans-
forms given as F1(x) and F2(x) have peak values at
frequency !0, with corresponding phase difference
�1 � �2. The time difference of the wireless sensors
is then calculated from the phase difference
�12 ¼ �1 � �2ð Þ=!0
�� ��.The second experiment tested accuracy of time
synchronization of WISAN network on the
Coordinator level of hierarchy. A hierarchical net-
work shown in Figure 2 was built with two sepa-
rate computers serving 3 Coordinators each. The
computers were connected by a TCP link over
wired Ethernet. Application client software was
run on one of the computers. Each Coordinator
serviced a cluster of 8 sensors for the total of 48
wireless sensors in the network. Two GPS receivers
were used to provide synchronization pulses to
three Coordinators each. A sinusoidal signal was
wired to the external sensor port on six of ran-
domly selected wireless sensors on different clusters
while the rest of the sensors were transmitting
acceleration data. Asynchronization between the
wireless devices was measured using the phase dif-
ference between the captured sinusoidal signals.
Delay value Di (ms)
0–1
0
1
2
3
Tot
al in
tege
r ar
ithm
etic
err
or e
A (
ms)
4
5
6
7
8
20 40 60 80 100 120
Figure 6 Error due to integer arithmetic in incrementing/decrementing the timers.
Sazonov et al. Wireless Intelligent Sensor and Actuator Network 471
In the third laboratory experiment we tested
the validity of the data acquired by distributed
wireless sensors. WISAN has been used to perform
mode shape reconstruction tests on a
100’’� 1’’�½’’ standard aluminum (6061) beam
using 21 wireless sensors associated with 3
Coordinators. The mode shapes were recon-
structed by output–output modal analysis using
ARTeMIS [41] and compared to the analytical
model as well as a finite element model processed
in Visual NASTRAN. This experiment has been
previously reported with the detailed description
found in [22].
4.2 Field Testing Using WISAN
To show practical applicability of the time-syn-
chronous and scalable architecture of WISAN to
applications of SHM, WISAN was tested on mode
shape extraction from ambient vibrations of a
bridge. The test bridge is an integral abutment,
single span (19.8m or 65 feet), four steel girder
bridge located in the Town of Lisbon, NY
(Figure 7(a)). The bridge is located on a rural
county road with very low traffic volume. The
underside of the bridge is easily accessible from
the ground.
The bridge was instrumented by 44 WISAN
sensors installed as 11 sensors per each of the
four girders (Figure 8). The sensors were clamped
on the bottom of the girders (Figure 7(b)) and
communicated with the Coordinator station
placed under the bridge. The Coordinator station
consisted of six Coordinator nodes connected to a
laptop computer. GPS was used to provide PPS
PanCoordinator station
Wireless sensors
Girder 1
Girder 2
Girder 3
Girder 4
Figure 8 Layout of the bridge test setup.
(a) (b)
Figure 7 (a) View of the test bridge, (b) sensor attachment.
472 Structural HealthMonitoring 9(5)
synchronization signal to all Coordinators.
Control and data retrieval from Coordinators
was performed through a TCP server by a
LabView-based client application. The sensors
formed six network clusters on six independent
frequency channels. Acceleration data from each
wireless sensor were sampled at the rate of
240.385Hz and resolution of 14 bits. Traffic pass-
ing over the bridge was the source of bridge exci-
tation. A number of experiments were conducted
for data recordings of 30 s, 2min, 5min, and
10min in duration. The cutoff of frequency of
the programmable low-pass filter was varied
between 40Hz and 100Hz to filter out high fre-
quency noise components.
5 Results
The laboratory time synchronization tests
showed that the measured time synchronization
error between the wireless nodes lies well within
the predicted limits. In the experiment that tested
sensor synchronization within a cluster the mea-
sured error for a set of five experiments varied
between 2.6 and 10.4ms with the mean of 5.1 ms
and standard deviation of 3.1 ms. These results con-
firm the upper bound on time error within a cluster
as �15ms calculated in Section 3.
The test of time synchronization between wire-
less sensors of a spatially distributed network syn-
chronized by GPS time reference established an
error range from a minimum of 0.1 ms to a maxi-
mum of 12ms with the mean of 6.73ms and stan-
dard deviation of 3.4 ms for a set of 5 experiments.
These tests confirm the capability of WISAN sen-
sors to continuously sample time synchronous data
(with a precision of �23 ms) in network configura-
tions that can be scaled to cover large structures.
Results from the reconstruction of the alumi-
num beam mode shapes have been reported in [22]
and do not fit within the space allotted for this
article. Overall, measured natural frequencies and
reconstructed mode shapes for the beam fit well
with analytical equations. The average coefficient
of correlation for the first three modes (computed
over 12 experiments) was 0.944, indicating correct
extraction of experimental mode shapes by
WISAN system. The beam experiments confirmed
the validity of the data acquired by WISAN and
correctness of the methods used for modal identi-
fication from vibration data.
Results of the field experiments demonstrate
the ability of WISAN to reconstruct mode shape
from complex structures using the proposed scal-
able and time-synchronized architecture. We have
identified several mode shapes at 9.155, 10.97,
26.23, 32.75, and 53.65Hz (Figure 9). Vibration
response of the bridge had a peak amplitude of
25mg in the range of 8–12Hz, which is rather
low for the MEMS accelerometers used as sensors.
Nevertheless, reconstructed mode shapes are clear
and easily identifiable.
(a) (b)
(c) (d)
Figure 9 Mode shapes identified in the test bridge: (a) 9.155 Hz, (b) 10.97 Hz, (c) 32.75 Hz, (d) 53.65 Hz.
Sazonov et al. Wireless Intelligent Sensor and Actuator Network 473
6 Discussion
Our goal is creation of a scalable WSN in
which every node remains time-synchronous to
all other nodes in the network regardless of how
many nodes form the network or how large of an
area the network covers. These two properties are
critical for any method of SHM that involves
modal identification of structures. The proposed
scalable and time-synchronous WISAN network
is based on hierarchical architecture in which wire-
less sensors with limited transmission range form a
cluster and synchronize with a Coordinator by
using the beaconing mechanism. Several clusters
can coexist within the same area by occupying dif-
ferent frequency channels. This allows forming
very dense sensor installations. All Coordinators
in the network are synchronized to a reference
GPS signal, thus ensuring almost perfect time syn-
chronicity of every wireless node.
The results of laboratory tests show that the
actual synchronization error is well within the
bounds (�23ms) predicted by the analysis. In
theory, because of the GPS synchronization of
the Coordinators, even very large networks can
operate with maximum synchronization errors
less than �23 ms. This capability could be useful
for applications on bridges with lengths of kilome-
ters. Even tighter synchronization can be achieved
by using better crystals (�6 ms using a �10 ppm
crystal) and newer microcontrollers from TI that
have lower interrupt latency.
Results of reconstructing beam mode shapes
[22] referenced here show that WISAN provides
valid data and corresponds well with the com-
monly used methods of modal identification.
Results of bridge testing confirm conclusions
of laboratory tests and show that WISAN can be
deployed for practical applications of monitoring
highway infrastructure. The reconstructed mode
shapes could only be obtained under conditions
of tight sensor synchronization [22]. It should be
noted that we did not attempt finite element
modeling of the bridge used in field experiments
due to labor intensity of such modeling.
However, the reconstructed mode shapes behave
as expected from a structure of the given size and
geometry. Thus, the field experiment successfully
demonstrated the applicability of WISAN to
testing of the real structures, which is a step
toward practical applications of wireless SHM.
7 Conclusions
In this article we presented a scalable, hierarchi-
cal, time-synchronous wireless sensor network,
WISAN. The scalability and time synchronization
aspects of the network architecture have been
described in detail. We presented a novel time syn-
chronization algorithm that was implemented in
WISAN and demonstrated the algorithm’s stability.
Various laboratory and field validation tests have
been presented and the results show that the pro-
posed hierarchical network architecture can satisfy
both scalability and time synchronization require-
ments. This scalable network architecture can be used
to create time-synchronous sensor networks for appli-
cations of SHM-based on analysis of vibration.
Acknowledgment
The authors would like to acknowledge support provided
by New York State Energy Research and Development
Authority.
References
1. Lynch, J.P. and Loh, K.J. (2006). A summary review of
wireless sensors and sensor. Networks for structural health
monitoring. The Shock and Vibration Digest, 38, 91–128.
2. Straser, E.G. and Kiremidjian, A.S. (1998). A Modular,
Wireless Damage Monitoring System for Structures.
Technical Report 128, John A. Blume Earthquake
Engineering Center, Stanford University, Stanford, CA.
3. Lynch, J.P., Sundararajan, A., Law, K.H., Carryer, E.,
Farrar, C.R., Sohn, H., Allen, D.W., Nadler, B. and
Wait, J.R. (2004). Design and performance validation
of a wireless sensing unit for structural health monitoring
applications. Structural Engineering and Mechanics, 17,
393–408.
4. Kottapalli, V.A., Kiremidjian, A.S., Lynch, J.P.,
Carryer, E., Kenny, T.W., Law, K.H. and Lei, Y.
(2003). Two-tiered wireless sensor network architecture
for structural health monitoring. In: Proceedings of the
SPIE: Smart Structures and Materials, San Diego, CA,
3–6 March, Vol. 5057, pp. 8–19.
5. Shinozuka, M. (2003). Homeland security and safety. In:
Proceedings of the International Conference on Structural
Health Monitoring and Intelligent Infrastructure, Tokyo,
Japan, 13–15 November, Vol. 2, pp. 1139–1145.
474 Structural HealthMonitoring 9(5)
6. Park, C., Chou, P.H. and Shinozuka, M. (2005).
DuraNode: Wireless networked sensor for structural
health monitoring. In: Proceedings of the 4th IEEE
International Conference on Sensors, Irvine, CA.
7. Zhao, F. and Guibas, L. (2004). Wireless Sensor
Networks: An Information Processing Approach, San
Francisco, CA: Morgan Kaufman.
8. Kling, R.M. (2003). Intel Mote: An enhanced sensor
network node. In:Proceedings of the InternationalWork-
shop on Advanced Sensors, Structural Health Monitoring
and Smart Structures, Tokyo, Japan, pp. 12–17.
9. Kumagai, J. (2004). Life of birds – wireless sensor net-
work for bird study. IEEE Spectrum, 41, 30–37.
10. Hill, J.L. (2003). System architecture for wireless sensor
networks. PhD Thesis, University of California, Berkeley.
11. Maurer, W. (2003). The Scientist and Engineer’s Guide
to Smart Dust and TinyOS Programming, Livermore,
CA: DSPLab Press.
12. Glaser, S.D. (2004). Some real-world applications of
wireless sensor nodes. In: Proceedings of the SPIE:
Smart Structures and Materials: Sensors and Smart
Structures Technologies for Civil, Mechanical, and
Aerospace Systems, San Diego, CA, 15–18 March,
Vol. 5391, pp. 344–355.
13. Tanner, N.A., Wait, J.R., Farrar, C.R. and Sohn, H.
(2003). Structural health monitoring using modular
wireless sensors. Journal of Intelligent Material
Systems and Structures, 14, 43–56.
14. Grosso, A.D., Inaudi, D. and Pardi, L. 2002. Overview
of European activities in the health monitoring of
bridges. In: First International Conference on Bridge
Maintenance, Safety and Management, IABMAS0020,
CIMNE, Barcelona.
15. Doebling, S.W., Farrar, C.R. and Prime, M.B. (1998).
A summary review of vibration-based damage identifi-
cation methods. Shock and Vibration Digest, 30,
91–105.
16. Cunha, A. and Caetano, E. (2005). From input-output to
output-onlymodal Identification of civil engineering struc-
tures. In: Proceedings of the first International Operational
Modal Analysis Conference, Copenhagen, Denmark.
17. Farrar, C.R. and Jauregui, D.A. (1998). Comparative
study of damage identification algorithms applied to a
bridge. Smart Materials and Structures, 7, 704–719.
18. Brincker, R., Zhang, L. and Andersen, P. (2001).
Modal identification from ambient responses using fre-
quency domain decomposition. In: Proceedings of 18th
International Modal Analysis Conference, Kissimmee,
FL, USA.
19. Sazonov, E. and Klinkhachorn, P. (2005). Optimal spa-
tial sampling interval for damage detection by curva-
ture or strain energy mode shapes. Journal of Sound and
Vibration, 285, 783–801.
20. Pandey, A.K., Biswas, M. and Samman, M.M. (1991).
Damage detection from changes in curvature mode
shapes. Journal of Sound and Vibration, 145, 321–332.
21. Shi, Z.Y. and Law, S.S. (1998). Structural damage
localization from modal strain energy change. Journal
of Sound and Vibration, 218, 825–844.
22. Krishnamurthy, V., Fowler, K. and Sazonov, E. (2008).
The effect of time synchronization of wireless sensors
on the modal analysis of structures. Smart Materials
and Structures, 17, 055018.
23. Sivrikaya, F. and Yener, B. (2004). Time synchronization
in sensor networks: A survey. IEEE Network, 18, 45–50.
24. Elson, J., Girod, L. and Estrin, D. (2002). Fine-grained
network time synchronization using reference broadcasts.
In: Proceedings of 5th Symposium on Operating Systems
Design and Implementation, Boston, MA, USA.
25. Ganeriwal, S., Kumar, R. and Srivastava, M.B. (2003).
Timing-sync protocol for sensor networks.
In: Proceedings of the First International Conference
on Embedded Networked Sensor Systems, Los Angeles,
California, USA, pp. 138–149.
26. Sichitiu, M.L. and Veerarittiphan, C. 2003, Simple,
accurate time synchronization for wireless sensor
networks. In: Proceedings of the IEEE Wireless
Communications and Networking Conference (WCNC
2003), New Orleans, Vol. 2, pp. 1266–1273.
27. Greunen, J.V. and Rabaey, J. 2003. Lightweight time
synchronization for sensor networks. In: Proceedings
of 2nd ACM lnternational Conference on Wireless
Sensor Networks and Applications, San Diego, CA,
USA, pp. 11–19.
28. Wang, Y., Lynch, J.P. and Law, K.H. (2007). A wire-
less structural health monitoring system with multi-
threaded sensing devices: design and validation.
Structure and Infrastructure Engineering, 3, 103–120.
29. Lynch, J.P., Law, K.H., Kiremidjian, A.S., Kenny,
T.W., Carryer, E. and Partridge, A. (2001). The
design of a wireless sensing unit for structural health
monitoring. In: Proceedings of the 3rd International
Workshop on Structural Health Monitoring, Stanford,
CA, 12–14 September.
30. Mitchell, K., Rao, V.S. and Pottinger, H.J. (2002).
Lessons learned about wireless technologies for data
acquisition. In: Proceedings of the SPIE: Smart
Structures and Materials 2002: Smart Electronics,
MEMS, and Nanotechnology, San Diego, CA, 18–21
March, Vol. 4700, pp. 331–341.
31. Chung, H.-C., Enomoto, T., Shinozuka, M., Chou, P.,
Park, C., Yokoi, I. and Morishita, S. (2004). Real-time
visualization of structural response with wireless
MEMS sensors. In: Proceedings of the 13th World
Conference on Earthquake Engineering, Vancouver,
BC, Canada, 2–6 August.
Sazonov et al. Wireless Intelligent Sensor and Actuator Network 475
32. Farrar, C.R., Allen, D.W., Ball, S., Masquelier, M.P. and
Park, G. (2005). Coupling sensing hardware with data
interrogation software for structural health monitoring.
In: Proceedings of the 6th International Symposium on
Dynamic Problems of Mechanics (DINAME), Ouro
Preto, Brazil, 29 February – 4 March.
33. Krishnamurthy, V. and Sazonov, E.S. (2008).
Reservation-based protocol for monitoring applications
using IEEE802.15.4 sensor networks. International
Journal of Sensor Networks, 4(3), 155–171.
34. Texas Instruments. Available at: www.ti.com (accessed
October 2008).
35. Chipcon Inc. Available at: www.chipcon.com (accessed
October 2008).
36. Sazonov, E., Jha, R., Janoyan, K., Krishnamurthy, V.,
Fuchs, M. and Cross, K. (2006). Wireless intelligent
sensor and actuator network (WISAN): A scalable
ultra-low-power platform for structural health monitor-
ing. In: Proceedings of SPIE’s Annual International
Symposium on Smart Structures and Materials, San
Diego, CA, USA, Vol. 6177, pp. 234–244.
37. LIS3L02AQ 3 axis Linear Accelerometer, ST
Microelectronics Inc. Available at: www.st.com
(accessed April 2006).
38. IEEE Standard for Information technology (2006).
Telecommunications and information exchange between
systems–Local and metropolitan area networks–Specific
requirements, Part 15.4: Wireless Medium Access
Control (MAC) and Physical Layer (PHY) specifica-
tions for Low Rate Wireless Personal Area Networks
(LR-WPANs), IEEE 802.15.4-2006.
39. Allan, D.W., Ashby, N. and Hodge, C.C. (1997). The
science of timekeeping. Hewlett Packard Application
Note 1289.
40. CS20 8 MHz Crystal, Citizen. Available at:
http://www.citizencrystal.com (accessed April 2006).
41. ARTeMIS – Operational Modal Analysis Software
Provided by Structural Vibration Solutions A/S.
Available at: http://www.svibs.com/ (accessed October
2008).
476 Structural HealthMonitoring 9(5)