Sensor
network application framework for autonomous structural health monitoring of
bridges
Edward Sazonova, Kerop Janoyan b, Ratan Jhac
a Dept. of Electrical and Computer Engineering
b Dept. of Civil and Environmental Engineering
c Dept. of Mechanical and Aeronautical Engineering
ABSTRACT
Life cycle monitoring of civil infrastructure such as bridges and buildings is critical to the long-term operational cost and safety of aging structures. Nevertheless, there is no commonly accepted and recognized way to perform automated monitoring of bridges. One of the important issues is the cost of the data acquisition subsystem and its installation and maintenance costs, which are tightly connected to the choice of monitoring methodology.
The
presented application framework includes: first, Wireless Intelligent Sensor
and Actuator Network (WISAN) as an inexpensive way to perform data acquisition
for the tasks of structural health monitoring; second, a vibration-based SHM
method for bridges; and third, a fully autonomous SHM system for bridges,
ambient-energy-powered and minimally dependent of human involvement.
Design of
the sensor network reflects the particularities of the application: proactive
rather than reactive nature of the data streams; fault-tolerant architecture
ensuring protection from extreme events; and real-time data acquisition
capabilities. Other issues include operating a massive array of heterogeneous
sensors, achieving a low cost per sensor, cost and sources of energy for the
network nodes, energy-efficient distribution of the computational load,
security of communications and coexistence in the ISM radio bands.
The modal
SHM methods under consideration are the method of modal strain energy with
fuzzy uncertainty management, method of damage index and a method based on
Hilbert-Huang transform. Modal identification through ambient vibrations is
performed though auto-regressive moving average models.
The final
step in the monitoring methods is the determination of bridge deterioration
rate and prediction of its remaining useful life based on measurements provided
by the sensor network and modal methods used.
The deterioration curves are generated at both the element and bridge
levels and are compared to existing inspection-based methods.
Keywords:
wireless sensor networks, structural health monitoring, ambient vibrations,
autonomous systems, modal strain energy, deterioration rate, life-cycle
monitoring.
1. INTRODUCTION
Life cycle
monitoring of civil infrastructure such as bridges and buildings is critical to
the long-term operational cost and safety of aging structures. Knowledge of the
structure’s health, load bearing capacity, and remaining life is the primary
goal of any strategy of Structural Health Monitoring (SHM). Bridges constitute
the most vulnerable element of the transportation infrastructure. An
out-of-service bridge creates economic losses both for the bridge users (in
terms of traffic delays and detours) and for the bridge and road operators. At
the end of 2003, the Federal Highway Administration (FHWA) has listed 27
percent of the country’s 591,000 bridges as structurally deficient or
functionally obsolete, in terms of dimensions, load or other characteristics
(FHWA, 2003).
Repair and
replacement of the structurally deficient components based on the results of
objective evaluation of the bridge status can significantly reduce money spent
on maintenance. The key is to have a low cost and reliable way to evaluate
structural integrity, identify deteriorated structural components and quantify
changes in terms of the load capacity and remaining service life estimates.
This goal can be accomplished by a system that is capable of performing
continuous health monitoring and evaluation of bridges and could complement
FHWA-mandated periodic visual inspections of the bridges
At the
present time there is no commonly accepted and
recognized way to perform automated monitoring of bridges. One of the important
issues is the cost of the data acquisition subsystem and its installation and
maintenance costs, which are tightly connected to the choice of monitoring
methodology. Lynch et
al. (2003) reported the costs over $300,000 per bridge ($5000 per sensor) to
install a monitoring system with 60 accelerometers in
The high
costs of instrumentation can be alleviated by employing the concept of a highly
distributed, networked data acquisition system such as a wireless sensor
network, consisting of many inexpensive low-power nodes. However, the full
potential even of a low-cost SHM system can only be achieved if it requires
minimal human involvement in reviewing periodic reports produced by the system
and responding to alarms. Every step of data processing starting from data
acquisition up to computing remaining life estimates has to be automated and
performed by a computer without human involvement, i.e. perform a fully
automatic decision support.
2. DESIRED
FEATURES of THE SHM System
The design
of an autonomous SHM system should follow a global view on the system as a complex
and tightly integrated sequence of information processing, where each step
depends on the desired features of the SHM system and defines methods to be
used. Therefore, desired features should be defined as the first step of the
design and application process.
In the
proposed SHM system, we are targeting the following features:
1. The SHM system should utilize
low-cost equipment and sensors, have low installation and maintenance costs.
Ideally, the cost of an installed system should constitute 1-2% of the
structure’s cost.
2. Massive arrays of heterogeneous
should be employed on the structure, collecting a variety of information from
the structural components.
3. The system should be easily
configured to be used with various types of bridges with different structural
geometry and independent of used construction materials.
4. Data collected from the sensor
arrays should be processed in a streamlined, fully automatic manner, with the
final result being a periodic report delivered to the monitoring agency over
inexpensive, commonly available and reliable data link. The system should be
able to raise alarms in case of extreme events, pronounced and rapid changes in
structure’s condition and major equipment malfunctions.
5. The system should function
unattended for prolonged periods of time (years), provide for self-diagnostic
capabilities and easy repairs.
6. To be fully autonomous, the SHM
system should not utilize wired power for data acquisition and, most
importantly, damage detection and localization, i.e. it should utilize ambient
energy for these goals.
To satisfy
stated requirements, we are proposing a sensor-network-based application
framework of structural health monitoring of bridges. Utilization of a sensor network addresses
cost and power requirements. Installation costs for wireless sensors are much
lower then for wired systems. At a low price tag per sensor, repair by
replacement is a viable and inexpensive way to maintain the system. With a
low-power design (Sazonov et al 2004), a networked sensor will consume very
little energy and can utilize harvesting of ambient energy (vibration, solar,
wind, etc.) for powering of the data acquisition and network communications,
eliminating the need for periodic battery changes.
Modal-based
damage detection and localization methods allow monitoring of various types of
bridges and construction materials. Being driven by ambient excitation these
methods can perform monitoring of structure excited only by passing traffic and
wind, therefore enabling truly autonomous monitoring not dependent on external,
dedicated excitation sources. Modern low-power general purpose computers (such
as notebook computers) possess enough computing power to perform the full cycle
of information processing directly on site, while being powered by a solar cell
or a wired source. Processing the information from the damage detection and
localization method as well as from auxiliary sensor, methods of computational
intelligence such as neuro-fuzzy systems allow connecting damage and sensor
information to element-level and bridge-level health reports. These periodic
reports will be emailed to the monitoring agency via an inexpensive cellular
link or a satellite link in places with no cellular coverage.
The
general structure of the proposed system is shown in Fig. 1. The following
sections of this paper present more details on the proposed Wireless
Intelligent Actuator and Sensor Network (WISAN), damage detection and
localization from ambient vibrations and estimating remaining service life of a
bridge.
3. WIRELESS INTELLIGENT SENSOR
NETWORK
Utilization
of a sensor network for long-term SHM is performed in an attempt to minimize
cost and maximize utility of the system as a whole. Therefore, the sensor
network design should be closely tied-in with the design of the monitoring
methodology and do not necessarily follow the concepts formulated for other
applications.
The key
differences include: proactive rather than reactive nature of the data streams,
where data is collected from multiple sensors at a constant rate rather than
reflecting events or changes in values; the fault-tolerant architecture
ensuring protection from extreme events and making the system insensitive to
loss of one or multiple nodes does not assume such node losses under normal
operating conditions; and real-time data acquisition capabilities require not
only timestamping and prompt delivery of messages, but closing the feedback
loop directly on a sensor node.
The design
issues include minimization of power consumption by a sensor node, so it can be
powered by harvested energy; accounting for coexistence with other devices
functioning in the Industrial, Scientific and Medical (ISM) range of radio
frequencies; guaranteeing reliable and secure way of communication in a network
containing a large number of nodes; performing energy-efficient distribution of
computational load and providing on-sensor intelligence supporting the
monitoring methodology; enabling self-localization of the sensor nodes for easy
configuration of the network.
The
Wireless Intelligent Sensor and Actuator Network (WISAN, “the sensor network”)
is designed with the goal of continuous structural health monitoring in mind.
The design of the network conforms to the requirements stated in the previous
section and provides a sound foundation for practical implementation of health
monitoring systems for bridges.
The sensor nodes are built around an ultra-low-power microcontroller MSP430F1611 from Texas Instruments (TI 2004). A 2.4Ghz module CC2420 from Chipcon (Chipcon 2004) is used for the IEEE 802.15.4 compatible network interface.
nd 24-bit ity of 16-bit expansinsitionem.The CC2420 is a low-power,
low-cost, IEEE 802.15.4 compliant transceiver designed for RF applications in
the 2.4 GHz unlicensed ISM band. The transceiver module provides 16-channel
direct sequence spread spectrum modem with 2 Mchips/s and 250 kbps effective
raw data rate, low power consumption (RX: 60 mW, TX: 52 mW), effective range 10
to 75 meters, programmable output power, hardware MAC encryption and
authentication (AES-128), signal strength indicator and battery monitor.
The low
power consumption of sensor nodes enables truly autonomous and continuous
operation from an energy-harvesting device. In practical terms, a virtually
maintenance-free SHM system could be created by utilization of
electro-mechanical or piezo-electrical power generating devices for powering of
the WISAN nodes.
Selection of
the 802.15.4-compatible protocol for the physical and data link layers of the
network protocol also resolves the issues with electromagnetic compatibility,
reliability and security. The radio frequency (channel) allocation for 802.15.4
devices foresees the coexistence in the presence of other popular network
protocols, such as 802.11 (WiFi). The communications between sensor nodes can
be made secure through Advanced Encryption Standard (AES) encryption and
authentication.
The signal
strength meter located in the Chipcon RF interface module allows implementation
of localization capabilities in the sensor nodes by performing cross-node
measurements of signal strength.
WISANs
architecture provides flexibility and fault-tolerance through two-tier cluster
organization (Sazonov et al 2004). Each cluster head collects data from the
sensor nodes and passes it on to a supernode, where data processing takes place
(Fig.1). For many small-scale bridges the network will contain a single cluster
head / supernode.
4. THE
MONITORING METHODOLOGY
Dynamics
based SHM can provide a quantitative global damage detection method that can be
applied to complex structures. A recent paper by Chang et al. (2003) reviews
health monitoring methods for civil infrastructure.
The basic idea
in most dynamics-based SHM is that changes in physical parameters (mass,
stiffness, damping) cause changes in structural properties (natural frequency,
mode shape, modal damping) hence damage can be determined by changes in dynamic
properties. However, standard modal properties represent a form of data
compression and modal properties are independent of excitation amplitude,
frequency, and location. Lower frequency modes tend to capture global response,
but damage is typically a local phenomenon and local response is captured by
higher frequency modes. The task of damage detection by finding shifts in
resonant frequencies or changes in structural mode shapes is further compounded
by changes in these characteristics due to environmental factors (temperature,
moisture, etc.).
We are
investigating several methods to bring out their relative advantages and
disadvantages. The first two methods are similar conceptually, but differ in
treating noise and uncertainty of the real-world systems. First is the method
of modal strain energy combined with a fuzzy expert system (Sazonov et al,
2002) and second is the method of damage index (Kim and Stubbs, 2002). Both
methods utilize the modal data extracted through structural identification over
several measurements performed under the same environmental and load
conditions.
Modal
parameter identification with the aid of ARMA (auto-regressive moving average)
models leading to a state space formulation of the structural dynamic equations
of motion assumes that the dynamic behavior of the structure can be described
by a stationary, infinite order, linear, dynamical system. The input forces
from the natural excitations may be in the form of white or colored noise,
mixed with harmonics and non-stationary noise. At
Another
method under investigation is based on advanced signal processing for damage
detection and localization. Historically, structural vibration signals have
been analyzed as frequency response using Fourier transform or as
energy-frequency-time response using short-time Fourier transform. A novel
method, known as Hilbert-Huang Transform (HHT) (Huang
et al., 1998),
produces instantaneous frequencies as functions of time that give sharp
identifications of imbedded structures. The final presentation of the results
is an energy-frequency-time distribution, designated as the Hilbert spectrum.
This means we obtain the instantaneous frequency and energy defined locally,
rather than the global frequency and energy defined by traditional Fourier
Transform. Also, Fourier analyses are valid for problems involving linear
systems with periodic or stationary response. However, the Fourier analysis has
little physical sense for nonlinear systems and/or non-stationary response. We
have conducted a comparative study of vibration signal analysis using
short-time Fourier transform and Hilbert-Huang transform to gain an insight
into their benefits (Jha et al., 2004). The HHT spectrum is more sensitive to
the dynamic energy-frequency distribution and it is capable of capturing the
difference in the structural response caused by damage. The short time Fourier
transform indicates the natural frequency changes, but fails to capture the
magnitude changes.
5. Condition Assessment and Life Cycle
Monitoring
The ultimate goal of the damage detection scheme is to assess the bridge’s condition and determine its remaining useful or service life. Currently, State and metropolitan planning organizations turn to Bridge Management Systems (BMS) to provide them with a systematic approach to bridge programming. However, the deterioration rates are extrapolated from visual inspection data and are highly subjective and prone to human errors. Rational bridge deterioration rate information based on measured performance response data would provide a cogent basis for condition assessment of bridges.
The two BMS software that are most frequently used today are PONTIS (Golabi et al., 1993) and BRIDGIT (Hawk and Small, 1998). Both systems attempt to predict the remaining life of a bridge by generating life cycle curves using empirical (visual inspection) data input in the BMS routines. In PONTIS, originally funded by the FHWA, a population of bridges is represented on a network level by the individual bridge elements (deck, girder, bearings, etc.) with field inspection data providing numerical condition states for each element. The prediction model, a probabilistic second order Markovian chain, is applied at the network level, estimating the proportion of each bridge element that is expected to deteriorate in the next inspection cycle. A rank order of the bridge element condition states in any inspection cycle leads to an application at the bridge level. BRIDGIT, another popular BMS developed under the National Cooperative Highway Research Project (NCHRP) 12-28(2), is similar to PONTIS in that a Markovian prediction model is applied at the element level. The primary difference between the two systems lies in the optimization model, which is more bridge specific in BRIDGIT and that it addresses the issue of element interaction more extensively than PONTIS.
Deterioration curves could also be generated
at both the element and bridge system levels based on measured performance data
then compared to those produced by various BMS software. Element-level
parameters determined could be used to generate condition ratings for the
individual structural elements. The input parameters will include damage report
information, such as location and magnitude of change in damage, as well as
environmental factors such as temperature. These element ratings would be input
into the existing BMS models through general mapping from a physical order to a
logical order. The mapping scheme would take measured performance information
and relate them to the existing inspection-based rating description tables
(such as number of cracks, location of damage, etc.). 
The generated curves would
also act as a baseline for comparison in the case when only damage detection
measurements are available for computation of the final bridge deterioration
curve. A knowledge-based expert system that can be defined analytically must be
used to perform the mapping of input parameters supplied to the BMS.
6.
IMPLEMENTATION
Design of
the Wireless Intelligent Sensor and Actuator Network is currently being
developed under a grant from New York State Energy Research and Development
Authority (NYSERDA). Fig. 2 shows a prototype of the WISAN modules.
7. CONCLUSIONS
The presented application framework concept of a sensor network for
autonomous structural health monitoring addresses the synergetic issues of
integrating a sensor network with a vibration-based SHM method and subsequent
estimation of the remaining life.
ACKNOWLEDGEMENTS
This
research has been partially funded by a grant from New York State Energy
Research and Development Authority (NYSERDA), whose support we gratefully
acknowledge.
REFERENCES
Chang, P.C., Flatau, A., and
Chipcon Inc. (2004), www.chipcon.com
Federal Highway Administration (2003) Tables of Frequently Requested NBI Information, retrieved from http://www.fhwa.dot.gov/bridge/britab.htm.
Golabi, K., Thompson, P.D. and Hyman, W.A. (1993), Pontis Technical Manual, Tech. Rep. No. FHWASA-94-031. Optima Inc. and Cambridge Systematics, Inc.
Hawk,
H. and Small, E. (1998), The
He, C., and Jha, R., “Experimental Evaluation of Augmented UD Identification Based Vibration Control of Smart Structures,” Journal of Sound and Vibration, Vol. 274, No. 3-5, July 2004, pp. 1065-1078.
Huang, N. E., Shen, Z., Long, S. R., Wu, M. C., Shih, H. H., Zheng, Q., Yen, N., Tung, C. C., and Liu, H. H. (1998), “The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis,” Proceedings of the Royal Society of London A, Vol. 454, pp. 903-995.
Jha, R., Yan, F., and Ahmadi, G. (2004), “Energy-Frequency-Time Analysis of Structural Vibrations Using Hilbert-Huang Transform,” AIAA 2004-1975, 12th AIAA/ASME/AHS Adaptive Structures Conference, April 2004, Palm Springs, CA.
Kim, J., and Stubbs, N., “Improved Damage Identification Method Based on Modal Information,” Journal of Sound and Vibration, 2002, Vol. 252, No. 2, pp. 223-238.
Lynch, J.P. Sundararajan, A. Law, K. H. Kiremidjian, A. S. and Carryer, E. (2003) Power-Efficient Wireless Structural Monitoring with Local Data
Processing, Proceedings of the 1st International Conference on Structural
Health Monitoring and Intelligent Infrastructure (SHMII'03) ,
Sazonov E., Klinkhachorn P., GangaRao H., and Halabe U., (2002) “Fuzzy logic expert system for automated damage detection from changes in strain energy mode shapes,” Non-Destructive Testing and Evaluation, Taylor & Francis Publishing, Volume 18, Number 1/2002, Pages 1 - 17.
Sazonov, E.S. Janoyan, K. Jha, R. (2004) Wireless Intelligent Sensor Network for Autonomous Structural Health Monitoring, Smart Structures/NDE 2004, San Diego, California.
Texas Instruments (2004), www.ti.com