High performance wireless sensors system for structural health monitoring
G. Dib, J. Padiya, J. Xin and , L. Udpa, and K. Balasubramaniam
Citation: AIP Conference Proceedings 1430, 1583 (2012); doi: 10.1063/1.4716403
View online: http://dx.doi.org/10.1063/1.4716403
View Table of Contents: http://aip.scitation.org/toc/apc/1430/1
Published by the American Institute of Physics
HIGH PERFORMANCE WIRELESS SENSORS SYSTEM
FOR STRUCTURAL HEALTH MONITORING
G. Dib1 , J. Padiyar2 , J. Xin1 , L. Udpa1 , and K. Balasubramaniam2
1
NDE Lab, Michigan State University, East Lansing, MI, 48824
2
CNDE, Indian Institute of Technology Madras, Chennai, India 600036
ABSTRACT. Continuous structural health monitoring (SHM) uses permanently mounted
sensor networks on critical locations of a structural component. In-situ wired sensors require
a large amount of cabling for power and data transfer, which can drive up costs of installation
and maintenance. Hence the need for developing wireless sensors for SHM. The major obstacles
preventing the widespread use of wireless sensor networks (WSN) for SHM is the availability
of portable, low cost, low powered, low footprint, and high SNR based instrumentation. This
paper presents a wireless sensor system that could be interfaced with piezoelectric transducers
for the identification of anomalous events using ultrasonic techniques. Power aware algorithms
are used to coordinate the actuator-sensor network interaction with a central processing server,
where appropriate signal processing techniques are used to quantify the damage in terms of
severity.
Keywords: Wireless Sensor Networks, Structural Health Monitoring, Ultrasonics, Lamb
Waves
PACS: 84
INTRODUCTION
Current industry practice uses conventional NDE methods such as eddy current,
ultrasound, wet fluorescent magnetic particle testing, and magnetic Barkhausen noise
at scheduled intervals to inspect a system’s components during its downtime. Those
methods are typically manually conducted by utilizing hand-held devices that capture
data from remote-installed or hand-held sensors to be transported back to a central-
repository for analysis. Although such methods are well established and provide detailed
and precise information about damage location and shape, they are prone to user errors,
and require high cost to train personnel for conducting the inspection. Moreover, the
maintenance schedules might not be adequate to detect an impending hazard. To over-
come these shortcomings, cost-efficient techniques capable of accurate and continuous
structural health monitoring are desirable for meeting the systems’ uptime, reliability,
and safety.
WSNs offer a promising solution for continuous SHM of various industrial and civil
structures. Wireless sensor networks are inherently highly scalable and configurable and
do not require high installation and maintenance cost. They allow transitioning from
calendar based industrial maintenance to condition-based maintenance. In a WSN,
sensors would be fixed in predetermined locations on critical components, and each
Review of Progress in Quantitative Nondestructive Evaluation
AIP Conf. Proc. 1430, 1583-1590 (2012); doi: 10.1063/1.4716403
© 2012 American Institute of Physics 978-0-7354-1013-8/$30.00
1583
FIGURE 1. The architecture of a wireless sensor network implemented with a star network
topology. All the sensor nodes are interfaced with a PZT transducer and have a two way wireless
communication with the base station. The sensor nodes include a sensing and actuating interface
connected to the transducer, a computational core for processing and controlling data, and a wireless
interface for the communication with the base station. The base station gives damage detection
warnings using automated analysis algorithms based on data received from the sensor nodes.
sensor, or a group of neighboring sensors, would be interfaced to a sensor node (SN)
which has computation, storage, and communication capabilities using a limited energy
source (batteries). All the SNs would be able to communicate with each other, forming
a wireless sensor network. An overview of such a scheme is shown in Fig. 1.
A review of wireless sensor networks and their current technologies can be found
in [1]. WSNs have a wide range of application, and they are particularly useful in NDE.
The feasibility of deployment of a WSN has been shown in a petroleum environment [2],
in a semiconductor plant and oil tanker [3], and on a bridge [4]. However, those WSNs
do not support active sensing techniques, and they do not have an appropriate sensing
interface for implementing NDE techniques that would be able to give information
about local damage on a specific structure with high resolution and accuracy. SHM
with active sensing utilizes active transducers, such as piezoelectric Lead Zirconate
Titanate (PZT) sensors. A PZT vibrates when an electric field is applied to it or vice
versa, which allows it to be used as a transmitter or receiver. PZTs have been largely
used for guided waves inspection of thin plate structures and tubes [5] [6] [7]. A detailed
description of guided waves, also called Lamb waves can be found in [8].
This paper extends the previous work reported in [9] for implementing an interface
for guided wave inspection with off the shelf sensor nodes. A protocol for continuous
SHM is proposed by integrating sensor technology, algorithm development and wireless
networking. A SN can sample and digitize sensed signals, store them, and communicate
data back to a central base station for further processing. Signal processing algorithms
could then be used to extract damage related information from the measurements, and
analyze them to determine the current state of the structure. The base station would
also be responsible for network control and configuration of the SNs.
1584
FIGURE 2. A picture of the IRIS mote is shown on the left. The properties of the sensing
interface, computational core, and the wireless interface are shown on the right.
The rest of this paper is organized as follows. The hardware properties of the
sensor nodes are first discussed, followed by the design of the extension board for the
SN. The base station application and the communication protocol are described next.
An experimental setup for system validation is shown in the following section, and
finally the system validation results are shown. The last section gives the concluding
remarks.
THE SENSOR NODES
To keep the cost of the system as low as possible, an off-the-shelf sensor node,
the IRIS wireless module, known as IRIS mote, sold by Memsic Corporation [10], was
chosen. The IRIS mote and the properties of its different components is shown in Fig.
2. This mote is powered using two AA batteries, and has a 51-pin extension for the
connection of analog inputs and/or digital I/O. At the wireless interface, IRIS uses
the Atmel RF230 radio, which is IEEE 802.15.4 (Zigbee) compatible. The measured
communication ranges were up to 50m in indoor environment and 300m in an outdoor
environment according to the datasheet [10].
At the computational core, IRIS has the ATmega 1281 8-bit microprocessor. The
computational core provides the sensor node with data processing and control capa-
bility, which is a key advantage of wireless sensor networks. TinyOS operating system
environment was used to write software for programming the sensor nodes. TinyOS is
an event based operating system specifically designed for use with embedded networked
sensors [11]. At the sensing interfance, IRIS provides an 8 channel analog input to a
10-bit analog to digital converter (ADC), with a maximum sampling frequency around
273 ksps for all channels. There are several sensor boards available for providing sensing
capabilities to IRIS such as providing temperature sensors, light sensors, accelerome-
ters, etc. However there are no available sensor boards that give IRIS the capability for
ultrasonic NDE using PZT sensors. For this reason, a sensor board was built especially
to address this issue and to provide IRIS and similar motes with the capability of ac-
tuation so that it can excite guided waves, and the capability of sensing high frequency
guided waves.
EXTENSION BOARD CIRCUIT ARCHITECTURE
An extension board that could be connected with the IRIS mote using the 51-pin
expansion connector was designed to address its limitations in guided wave inspection.
The architecture of this board is shown in Fig. 3a. This board includes an actuator
subcircuit and a sensor subcircuit. A switch is used to control the enabling of one of the
1585
(a) (b)
FIGURE 3. Part 3a shows the sensorboard architecture. Part 3b shows a picture of the developed
extension board connected to the IRIS mote.
FIGURE 4. The actuator subcircuit, showing the digital input signal and the output of subcircuit.
The subcircuit transfer function is shown at the bottom.
two circuits. The switch is controlled by a General Purpose I/O (GPIO) pin from the
mote. This allows switching the function of the mote during run time by requests from
the administrator at the base station. Also, the extension board provides 8 channels
I/O using a multiplexer, where the active channel can also be controlled during run
time from the GPIO pins of the mote. A voltage regulator is used to provided the
needed voltages for the different circuit parts using the AA battery supply of the mote.
The Actuator Subcircuit
The actuator subcircuit gives the mote capability to actuate a sine wave tone
burst with programmable number of cycles and frequencies up to 600 KHz. The block
diagram of the actuator subcircuit is shown in Fig. 4. The input of this subcircuit is
provided by one GPIO pin from the mote, which was configured as an output pin to
provided a digital square wave with voltage range between 0V and 3V. The square wave
is then passed through a bandpass filter whose transfer function is shown in Fig. 4 to
convert it into a sine wave tone burst with a zero DC offset. The gains of the amplifiers
are fixed to provide a signal up 16V peak-to-peak.
The Sensor Subcircuit
The sensor subcircuit detects the envelop of the input signal for decreasing the
signal bandwidth, so that the acquired analog signals would have a fequency low enough
to be sampled by the ADC of the mote. The block diagram of the sensor subcircuit
1586
(a)
(b)
FIGURE 5. Part 5a shows the block diagram of the sensor subcircuit for envelop detection. Part
5b shows the subcircuit transfer function, its input and output in the frequency domain.
is shown in Fig. 5a. This subcircuit acts as a frequency demodulator, where a full
wave rectifier is used to square the input signal, and then a second order low pass filter
removes all the non-baseband components of the signal. This procedure will effectively
get the envelop of the input signal. The analysis of this subcircuit in the frequency
domain is shown in Fig. 5b. The input signal bandwidth is 250 KHz, which then passes
through the transfer function of the sensor subcircuit for envelop detection, and the
output signal has a bandwidth around 60 KHz. With such a low signal bandwidth, it
is now possible to sample it with the mote ADC sampling rate of 273 KHz.
THE BASE STATION APPLICATION
The base station consists of a personal computer connected to a mote through the
serial port using a gateway board. An application on the PC was written using Matlab
to control, configure, and read data from nodes available in the wireless sensor network.
The application communicates through the serial port with a mote connected to the
PC. This mote forwards data received from the sensor nodes through the wireless radio
to the serial port, and vice versa. An administrator can use the base station application
to discover the available nodes in the network, and configure each one of those nodes
by changing their state.
Base Station/Sensor Nodes Communication Protocol
Each sensor node in the network has a state machine which is controlled remotely
by the basestation. Fig. 6 shows the state machine of the sensor nodes on the left, and
the right side shows the steps that an administrator using the basestation application
can go through to change the state of the sensor nodes. A sensor node would usually
start in Idle mode, where the power to the extension board is shutdown, and the
mote is just listening to the wireless interface and ready to receive commands from the
basestation. When the basestation application is started up, the administrator should
1587
FIGURE 6. The sensor node state machine is shown on the left, and the possible commands and
actions that could be taken at the base station are shown in the right part.
broadcast a network request to find the current active nodes. All active nodes will
reply back with their information (location, channel, and ID). Then the administrator
can set the mode of each discovered node to be either idle, actuator, or sensor. When
the administrator broadcasts the start command to the network, each sensor node will
behave according to its current state. The idle nodes will just discard the command
and stay in the idle mode. The actuator nodes will send a sine wave tone burst with
the frequency and number of cycles previously configured. The sensor nodes will enable
the ADC and start sampling. In order to ensure that the sensor nodes will not just
sample noise, they compare each obtained sample to a threshold. When this threshold
is exceeded, the succeeding sample points within a window of 1.5ms are wirelessly
transmitted to the basestation. The basestation application can display and analyze
data received from all the sensor nodes.
SYSTEM VALIDATION EXPERIMENTAL SETUP
The designed extention board, system software and network protocol were tested
to find if they could be used to properly find damage using ultrasonic guided wave
techniques. Two 60x60 cm, and 4mm thick aluminum plates were used as samples for
damage detection. One of the plates had a 20x2 mm and 3.2 mm deep notch, and the
other plate was a healthy plate. Two PZT patches were bonded to the plate at the
locations shown in Fig. 7a. Two sensor nodes were connected to the PZT patches.
A basestation was used to configure each of the two sensor nodes. One was set as an
actuator and the other as a sensor. An oscilloscope was also connected to the PZT
patch configured as sensor to verify the experimental results. Figure 7b shows a picture
of the experimental setup.
SYSTEM VALIDATION RESULTS
The actuation subcircuit was configured to send a sine wave tone burst with 5
cycles and a frequency of 430 KHz. The peak to peak amplitude of the actuation signal
was 14V. The actuation signal is shown in Fig. 8a. The initial spike is due to the
initial condition of the filter used in the actuation subcircuit and it caused the circuit
1588
(a) (b)
FIGURE 7. Part 7a shows a schematic of the experimental setup of the plate and the PZT sensors.
Part 7b shows the photograph of the experiment.
(a) (b)
(c)
FIGURE 8. Part 8a shows the actuation signal. Part 8b shows the comparision between the digital
signal envelop and the raw signals from the oscilloscope. Part 8c shows the signals from defective and
health plate overlapped to indicate the time of flights of different parts.
to saturate. However, it did not affect the structural response results for properly
detecting damage, and hence no extra circuitry was needed to fix this issue.
The response signals from the plate response at the sensor were obtained for the
two plates. Figure 8b shows the responses of the healthy and defective plates. In both
graphs, the detected envelop that was received at the basestation was overlapped with
the actual raw signal obtained from the oscilloscope. It can be seen that the sensing
subcircuit correctly detects the envelop of the received signals.
At 430 KHz, and 4mm plate thickness, there are three different modes in the
signal, the A0 , S0 , and A1 modes, according to the dispersion equations of Lamb waves
1589
in aluminum plates [8]. However, the group velocities of the three different modes are
very close and hence we can easily distinguish the different indications in the signal
corresponding to the incident wave, reflections from defects, and reflections from the
edges. Figure 8c shows the time of arrival of the incident wave, tS0 incident , reflection
from damage, tS0 def ect , and reflection from edge, tS0 edgeref lection . It can be clearly seen
that in the the defective plate, the signal contains an additional refelction from the
defect indicating the presence of damage.
CONCLUSION
An extension board for the IRIS motes was designed for actuating and sensing
ultrasonic guided waves. The extension board detects the envelop of a received signal
to enable the use of the ADC integrated in the IRIS mote, which reduces the amount
of power consumption. Also it enables actuation using the GPIO pin of the IRIS. A
network communication protocol was implemented for managing a large network of
sensor nodes to be used for Lamb wave inspection. The system was tested for damage
detection in Aluminum plates.
REFERENCES
1. Lynch, J. P. and J., L. K., The Shock and Vibration Digest 38 (2006) 91.
2. Johnstone, I., Nicholson, J., Shehzad, B., and Slipp, J., Experiences from a wireless
sensor network deployment in a petroleum environment, in IWCMC, Honolulu,
Hawaii, 2007.
3. Krishnamurthy, L. et al., Design and deployment of industrial sensor networks:
Experiences from a semiconductor plant and the north sea, in SenSys, San Diego,
CA, 2005.
4. Pakzad, S. N., Fenves, G. L., Kim, S., and Culler, D. E., Journal of Infrastructure
Systems 14 (2008) 89.
5. Kessler, S. S., Spearing, S. M., and Soutis, C., Smart Materials and Structures 11
(2002) 269.
6. Ihn, J. B. and Chang, F. K., Structural Health Monitoring 7 (2008) 5.
7. Giurgiutiu, V., Intelligent Material Systems and Structures 16 (2005) 291.
8. Viktorov, I. A., Rayleigh and Lamb Waves: Physical Theory and Applications,
Plenum, 1967.
9. Dib, G. et al., Wireless nde sensor system for continuous monitoring, in QNDE,
San Diego, CA, 2010.
10. Memsic, http://www.memsic.com, 2011.
11. TinyOS, http://www.tinyos.net, 2011.
1590
