Recent advances in wireless smart sensors for multi-scale monitoring and control of civil infrastructure

Abstract

While much of the technology associated with wireless smart sensors (WSS) has been available for over a decade, only a limited number of full-scale implementations have been realized for civil infrastructure, primarily due to the lack of critical hardware and software elements. Using the Imote2, a flexible WSS framework has been developed for full-scale, autonomous structural health monitoring (SHM) that integrates the necessary software and hardware elements, while addressing key implementation requirements for civil infrastructure. This paper discusses the recent advances in the development of this WSS framework and extensions to structural control. Their successful implementations at full-scale for SHM of the 2nd Jindo Bridge in South Korea and the Government Bridge at the Rock Island Arsenal in Illinois, USA, as well as for wireless control of a lab-scale structure are presented.

1 Introduction

The declining condition and performance of civil infrastructure, as well as its potential vulnerabilities to natural hazards, are critical issues in both the US and most industrialized nations. Accordingly, the efforts for maintaining and enhancing such civil systems can consume a significant portion of a nation’s capital investment. For example, the U.S. Department of Transportation [46] has reported that $46.6 billion was used for rehabilitating highways and bridges and $11.0 billion for system enhancements in 2008; this expenditure amounts to about 63.2 % of the total $91.1 billion capital outlay by all levels of government; on the other hand, only $33.6 billion (36.8 %) was used for system expansion. Improving and modernizing infrastructure is an important part of current government policy on ensuring homeland security and maintaining economic vitality.

The ability to assess the performance of such civil infrastructure in a timely manner and detect damage at an early state can potentially reduce the cost, as well as the downtime associated with repairing or rehabilitating systems, all while providing increased public safety. Visual inspection has been a current practice in monitoring the safety of civil infrastructure. However, high cost often limits the use of visual inspection to infrequent occurrences. Moreover, the Federal Highway Administration (FHWA) in the US reported significant variability in the ratings assigned by a cohort of highly trained inspectors [30]. As a result, failure of bridges in the US is not as rare of an event as the public may believe. For example, between 1989 and 2000, a total of 134 bridges are reported to have partially or totally collapsed in the US due to triggering events (e.g., earthquake or vehicle collision), design and construction error, and undetected structural deterioration (e.g., scour, fatigue) [50]. Visual inspection would benefit by supplementing it with continuous, autonomous, impartial, data driven monitoring.

Structural health monitoring that combines various sensing technologies with embedded measurements provides an essential tool for assessing the status of structures. Moreover, the principles of SHM and its application will aid not only in the inspection of existing infrastructure, but allow rapid assessment on the safety of emergency facilities and evacuation routes, including bridges and highways, after natural disasters. Continuous and autonomous structural health monitoring can facilitate these goals.

Conventional systems based on wired sensors and data acquisition systems have long been the standard for SHM; however, realizations of such wired systems are limited by high cost, as structures become larger and more complex, scalability is essential. For example, the total cost of the monitoring system on the Bill Emerson Memorial Bridge in Cape Girardeau, Missouri is approximately $1.3 million for 86 accelerometers, which makes the average installed cost per sensor a little more than $15,000; this cost is not atypical of today’s wired SHM systems [3]. Considering that a large portion of the cost results from the cabling between the sensors and the data acquisition system, wireless sensors are an attractive alternative to such wired systems, especially for large civil structures, offering the potential for low-cost, continuous, and reliable SHM.

While wireless sensors have been commercially available for over a decade, only a limited number of full-scale implementations have been realized, primarily due to the lack of critical hardware and software elements. An example of one such critical issue is network scalability. A wireless sensor network implemented on the Golden Gate Bridge in 2008 took approximately 10 h to collect 80 s of data (sampled at 1000 Hz) from 56 sensor nodes to a central location [36]. To assist in dealing with the large amount of data generated by a monitoring system, on-board processing can be performed locally on the sensor’s embedded microprocessor. This strategy is a radical departure from the conventional approach to monitoring structures.

This paper discusses recent advances in the development of a low-cost, wireless smart sensor framework for continuous and reliable structural health monitoring, as well as its successful deployment at full scale on the 2nd Jindo Bridge in South Korea and Rock Island Government Bridge in Illinois. In addition, efforts toward structural control applications of the framework are introduced, focusing on the real-time performance improvements of hardware and software developed herein. The wireless SHM system on the Jindo Bridge constitutes the largest deployment of a wireless smart sensor network (WSSN) for long-term monitoring civil infrastructure, and the wireless SHM system on the Rock Island Bridge exhibits a great compensation of existing, but incomplete, wired SHM system. These wireless SHM examples, as well as the lab-scale implementation for wireless feedback control, demonstrate the tremendous potential of this technology.

2 Wireless smart sensors for SHM

Wireless smart sensors differ from traditional wired sensors in significant ways. Each sensor has an on-board microprocessor that can be used for digital signal processing, self-diagnosis, self-calibration, self-identification, and self-adaptation functions. Furthermore, all WSS platforms are employed with wireless communication technology. Numerous WSS platforms have been developed in academia and industry [28], and they have experienced substantial progress through interdisciplinary research efforts to address issues in sensors, networks, and application-specific algorithms.

The Imote2 (see Fig. 1), developed by Intel Research, is a wireless sensor platform designed for data-intensive applications such as SHM, and thus selected for this research. The Imote2 includes a high-performance X-scale processor (PXA27x), whose clock speed can be varied according to application demands and power management, ranging from 13 to 416 MHz. It has 256 K SRAM, 32 MB FLASH, and 32 MB SDRAM, which enables the on-board computations required for data-intensive applications, as well as storage of long-term measurements. Sensing with the Imote2 is facilitated by sensor board(s) stacked on the Imote2 via two advanced board-to-board connectors.

Fig. 1

Imote2: top (left), right (middle), and ITS400 board stacked on Imote2 (right)

This section briefly reports the previous achievements using the Imote2 for SHM, in terms of both hardware and software.

2.1 Enabling hardware for SHM

Intel provided a reference Basic Sensor Board (ITS400) that can interface with the Imote2. Although the ITS400 can measure 3-axes of acceleration, light (TSL2561), temperature, and relative humidity (SHTx), its application to SHM has been limited due to low (12-bit) analog-to-digital conversion (ADC) fidelity, poor flexibility in selecting sampling rate, inconsistent sampling rate, and inadequate anti-aliasing filters [31].

To address the shortcomings of the ITS400 sensor board, a versatile accelerometer sensor board, designated SHM-A (see Fig. 2), was developed for SHM applications of civil infrastructure [38]. The key component of the SHM-A board is the 16-bit 4-channel QuickFilter ADC (QF4A512) that provides the anti-aliasing filter, programmable gain amplifier (PGA), user-selectable sampling rates and customizable digital FIR filters. The SMH-A board contains the ST Microelectronics LIS344ALH accelerometer for 3-axis acceleration sensing (ranging ±2 g), in addition to the Sensirion STH11 for temperature and humidity measurement, and the TAOS 2561 for light intensity sensing.

Fig. 2

SHM-A board: perspective (left), top (middle), and bottom (right)

2.2 Enabling software for SHM

SHM applications for WSSNs require complex programming, ranging from network functionality to embedded algorithm implementation. Software development is made very difficult by the fact that many smart sensor platforms are employed with their own special-purpose operating systems out of common programming environments. Furthermore, the extensive expertise required for developing SHM applications has severely limited the use of smart sensing technology for monitoring civil infrastructure. For example, application development on the Imote2 requires knowledge and expertise on TinyOS, an operating system of the Imote2 developed by University of California at Berkeley, and its associated programming language nesC [22].

2.2.1 Illinois SHM services toolsuite

To address the complexity associated with creating WSSN applications, a software framework has been developed based on the design principles of Service Oriented Architecture. The software framework aims to provide continuous and reliable monitoring of civil infrastructure using a dense network of the Imote2 smart sensors and was developed under the Illinois Structural Health Monitoring Project (ISHMP), a collaborative effort between researchers in civil and environmental engineering and computer science departments at the University of Illinois at Urbana-Champaign. This framework provides a suite of services, called the Illinois SHM Services Toolsuite, implementing the key middleware infrastructure necessary to provide high-quality sensor data and to reliably communicate it within the sensor network, as well as number of commonly used numerical algorithms for SHM. Intended to allow researchers and engineers to focus their attention on the advancement of SHM approaches without the headaches of low-level programming, the software is available for public use at http://shm.cs.illinois.edu/software.html. Some of the key foundation service modules include: time synchronization ( TimeSync ¹), reliable wireless communication of both short messages and long data records ( ReliableComm ), flexible network-wide synchronized sensing service ( SensingUnit ), a service that supports the reliable dissemination of network and utility commands ( RemoteCommand ). Application services provide the numerical algorithms for easy implementation of SHM applications include, but are not limited to: correlation function estimation ( CFE ), Eigensystem Realization Algorithm ( ERA ), Stochastic Subspace Identification ( SSI ), and Stochastic Damage Locating Vector method ( SDLV ). More detailed information about the software can also be found in Rice and Spencer [38], Rice et al. [39], and Sim and Spencer [41].

2.2.2 Autonomous network operation

Continuous and autonomous operation is another key issue in developing a large-scale WSSN for SHM. One of the services in the Illinois SHM Services Toolsuite, called AutoMonitor ,² was designed to facilitate autonomous operation of all of the necessary functions in a full-scale WSSN deployment. To minimize power usage in the WSSN, AutoMonitor incorporates with power-efficient management services for sensor nodes, called SnoozeAlarm and ThresholdSentry . The Imote2 allows the processor to be put into a sleep mode with minimal power consumption by keeping only the real-time clock powered on. The SnoozeAlarm service implements low-power listening (LPL) functionality: it places sensor nodes into a deep-sleep state (for 10 s by default), waking periodically for a brief time (0.6–0.75 s by default) to listen for beacon signals sent from the gateway node; should a signal be heard, the node becomes fully active [38]. To monitor activities in the network, a threshold triggering strategy is utilized. ThresholdSentry service runs on several designated sentry nodes periodically execute sentry task, i.e., collecting data for given period of time and checking if the threshold value is exceeded. If this threshold is exceeded, the sentry nodes send notification to the gateway node, which subsequently wakes up the entire network and initiates designated network-wide sensing tasks. The carefully designed scheduler embodied in AutoMonitor enables power-efficient and continuous management of the WSSN, as well as a combined operation of the other functionalities such as decentralized analysis, sensor diagnosis, and multi-hop communications [38].

2.2.3 Energy harvesting for sustainable operation

Availability of a sustainable power supply is one of the biggest concerns when large numbers of wireless sensors are distributed in the field. Though the battery life of a WSS node can be prolonged by efficient power management software, data compression, and in-network processing, the use of ordinary batteries still requires regular replacement. The Imote2 has a power management integrated circuit (PMIC) that facilitates sustainable energy harvesting. The PMIC interfaces directly with a rechargeable battery pack and handles unregulated power from energy harvesters up to 10 V. The PMIC charger manipulates voltage and current from energy sources for faster and more stable charging at the rechargeable battery.

Among various available energy sources, solar energy appears to be the best sustainable power source sufficient to reliably operate Imote2s [29]. Validation of solar energy harvesting has been successfully carried out in the Jindo Bridge deployment using solar panels [SCM-3.3W from SolarCenter (9 V–370 mA)] and lithium-polymer rechargeable batteries (the Ainsys 3.7 V–10,000 mAh; [16, 17]). In addition, a prototype wind turbine (HR-W35V, Hankukrelay) can be used to power Imote2 in a windy area such as the Jindo Bridge [35] (Fig. 3).

Fig. 3

Rechargeable battery (middle) and energy harvesters: solar panel (left) and wind turbine (right)

To enable energy harvesting during the SnoozeAlarm mode, a software service called ChargerControl , is additionally developed [29]. Enabling ChargerControl , the Imote2 determines whether it will continue in sleep mode or initiate charging mode by assessing the battery voltage and charging current. If the battery voltage is low and the charging current is sufficient, the Imote2 will start charging until the battery voltage achieves the target value of 4.2 V.

2.2.4 Multi-hop communication

Large-scale deployment of a WSSN gives rise to the need for multi-hop communication to provide adequate wireless coverage. The limited radio range of general WSS using IEEE802.11 or IEEE802.15 wireless protocols [14], combined with the impact of various environmental effects on radio transmission, makes direct communication between all nodes impractical. On the other hand, an important requirement of any communication scheme is data transfer reliability. Multi-hop communication, together with appropriate packet-loss compensation, addresses these issues by allowing sensors to cooperate to deliver data reliably between nodes outside of direct communication range.

The Ad hoc On-demand Distance Vector (AODV) protocol is a widely used routing method to discover optimal routes for multi-hop communication [37]. Figure 4 illustrates the AODV method. The route request (RREQ) message initiated from a source node is rebroadcasted by neighbor nodes until it reaches the destination node. Then, a route reply (RREP) message originating at the destination or intermediate nodes knowing a path to the destination is sent back to the source node, establishing the route in the reverse order. Among the received routes, the one with the minimum hop count is selected.

Fig. 4

Example of AODV route discovery [33]

The modified AODV protocol, termed General Purpose Multi-hop (GPMH), is developed to support diverse data flow patterns such as central data collection, dissemination, as well as decentralized communication that are possible in SHM applications [33]. The standard AODV protocol uses periodic probe messages to update routing information frequently between mobile nodes, which consumes significant power. GPMH omits the periodic probe messages, because sensor mobility is not an issue in the SHM system. To reduce the delay caused by route formation, GPMH does not regenerate route request (RREQ) messages when route discovery is unsuccessful; instead, the task is handled by the reliable data transfer service in the Illinois SHM Services Toolsuite. GPMH employs an alternative to the standard hop-count routing metric used for evaluating different paths. The hop-count routing metric may lead to construction of long-distance, unreliable links, which may result in significant radio loss and increased power consumption. The new metric uses a combination of the link quality indicator (LQI) and received signal strength indicator (RSSI), calculated by the radio upon packet reception. Additional details can be found in the next section and in Nagayama et al. [33].

3 Recent advances for improved SHM and control using WSS

Many recent hardware and software advances have been achieved toward improved SHM of civil structures and extension to structural control, including new developments of multi-metric sensor and control boards, decentralized computing strategies, multi-hop communication protocols, real-time streaming data transmission, and more fault-tolerant features in the Illinois SHM Services Toolsuite. These advances, detailed in this section, are expected to increase the robustness of WSSNs, broaden the field of application, and support comprehensive monitoring and control of civil infrastructure.

3.1 Hardware improvements

3.1.1 Multi-metric sensor boards for SHM

3.1.1.1 High-sensitivity accelerometer board (SHM-H board)

A high-sensitivity sensor board (designated SHM-H; see Fig. 5) has been developed for measuring low-level ambient vibration (i.e., less than 1.0 mg), which is a common phenomenon for large civil structures, such as high-rise buildings and long-span bridges [19]. The SHM-H board employs a single-axis low-noise accelerometer (Silicon Designs, SD1221-002L) having noise density of 5.0 μg/√Hz for normal direction (i.e., z-axis), and the ST Microelectronics LIS344ALH, same as the SHM-A board, for tangential acceleration (i.e., x- and y-axis). The typical RMS noise levels of the SHM-H board at 20 Hz bandwidth are 0.05 mg (z-axis using SD1221), which is low enough to measure the ambient vibration of most civil infrastructure, and 0.3 mg (x- and y-axes using LIS344ALH). This high-sensitivity sensor board is found to be exceptionally effective when used as a reference sensor for decentralized data aggregation strategies employed in large WSSNs [19].

Fig. 5

SHM-H board: perspective (left), top (middle), bottom (right) [19]

3.1.1.2 Data acquisition board (SHM-DAQ board)

The SHM-DAQ board (see Fig. 6), which is another Quickfilter-based sensor board, opens all 4 channels of the ADC for external analog inputs (0–5 V or −5 to 5 V range selectable). The SHM-DAQ has terminal blocks (wire to board) and advanced connectors (board to board) for direct interface between external sensors or sensor boards. Digital sensors using the I²C or SPI interface can also be accommodated in the same way. In addition, the SHM-DAQ board provides various regulated power options (1.8, 3.3, and 5.0 V) for external sensors. The SHM-DAQ board was initially developed to accommodate a 3D ultra-sonic anemometer for wind environment monitoring [18], but it has been extended subsequently to interface with generic analog and digital sensors.

Fig. 6

SHM-DAQ board: perspective (left), top (middle), and bottom (right)

3.1.1.3 High-precision strain sensor board (SHM-S board)

Strain provides another important measure of a structure’s behavior. A new strain sensor board for the Imote2, designated SHM-S board (see Fig. 7), has been developed which includes an autonomously controllable Wheatstone bridge circuit. In the SHM-S board, the Wheatstone bridge is precisely balanced prior to the amplification of the signal so that the ADC is not saturated due to the amplified offset. To this end, digital potentiometers are used, connected to the two arms of the Wheatstone bridge in parallel, and autonomously controlled by software services RemoteCommand , SHMSAutoBalance in the Illinois SHM Services Toolsuite [20]. The SHM-S board supports up to 2500 times signal amplification, resulting in 0.3 με resolution below 20 Hz bandwidth, as well as temperature compensation and on-board shunt calibration capabilities. In addition, any analog signal from other types of sensors can be amplified with this sensor board, as long as the signal is in the range of the ADC. The SHM-S board has been designed to be used with SHM-DAQ or SHM-A board in combination, because the strain board includes only signal conditioning circuits without the ADC for the sake of simplicity.

Fig. 7

SHM-S board: top (left), stacked on SHM-A (middle), and stacked on SHM-DAQ (right) [20]

3.1.1.4 Wind pressure sensor board (SHM-P board)

The measurement of wind pressure occurring at the surface of bridges or buildings provides the distribution of wind forces surrounding the structures. Such measurement has generally been carried out through wind tunnel tests simulating the wind environment of the structure [34]. To enable direct wind pressure measurement on full-scale structures along with wireless smart sensor, a sensor board to measure wind pressure (SHM-P board) has been developed under the ISHMP. The SHM-P board provides wind pressure measurements using the AMSYS 5812 high-precision analog pressure sensor and appropriate signal conditioning for quality pressure measurement. This board is a simplified sensor board without an ADC, as the SHM-S board is; it can be stacked on the SHM-DAQ board to be compatible with the Imote2 as shown in Fig. 8.

Fig. 8

SHM-P board: bottom (left), and stacked on SHM-DAQ (right)

3.1.1.5 24-bit data acquisition board (SHM-DAQ24 board)

Many issues in measuring low-level ambient vibrations using existing wireless smart sensors are caused by the limited resolution of the ADC on existing sensor boards. The use of 24-bits ADC can resolve most of these concerns. However, such 24-bits ADC was not efficient to be used in wireless sensing system due to the high cost and high power consumption. Recent noise shaping technology using delta-sigma modulation [2] has significantly enhanced the ADC. Unlike most quantizers, the delta-sigma modulator includes an integrator before the actual comparator process, which consequently acts as a high-pass filter for the quantization noise [48], pushing the noise to higher-frequency area (see Fig. 9). Therefore, a much lower rate of oversampling is required to obtain the level of quantization power for 24-bits, resulting in lower-cost and lower-power 24-bit ADCs.

Fig. 9

Principle of delta-sigma analog-to-digital conversion [48]

A new 24-bit data acquisition board for Imote2 has been developed, employing the Texas Instruments (TI) ADS1274 delta-sigma type 24-bit ADC. In addition to the ADS1274, a 27 MHz crystal oscillator (CTS CB3LV-5I-27M0000) is used as the master clock of the ADS1274. The accuracy in the sampling rate of the output signal depends on the accuracy of the external crystal, of which frequency stability is specified as ±25 ppm with this CTS crystal. A voltage regulator (MAXIM 8878_5V) is used to provide a clean 5 V power to the ADS1274. The other 1.8 and 3.0 V power required for the ADS1274 are provided by Imote2 with its own regulator. For the reference voltage to be used in the comparator process, which directly affects the ADC performance, a low-noise low-drift voltage reference (TI REF5025) is used. Additional flip-flop (TI SN74LVC2G74) and single inverter gate (TI SN74LVC1G04) components are used to re-clock the data output to accurately interface with SPI of Imote2, which can potentially reduce the possible timing errors between the data output from the ADC and SPI clock of the Imote2.

As the ADS1274 provides a fixed data rate of 10,574 Hz for the low-speed mode (lowest power consumption, 7 mW/ch), additional real-time digital filtering and decimation software have been designed to obtain the desired data rate as well as better effective number of bits (ENOB) using the Illinois SHM Services Toolsuite. Figure 10 shows the SHM-DAQ24 board and a block diagram of the on-board FIR filtering and decimation process.

Fig. 10

SHM-DAQ24 board (left) and block diagram of on-board FIR filtering and decimation process (right)

3.1.2 Structural control boards

3.1.2.1 SAR-type data acquisition board (SHM-SAR board)

While the QuickFilter QF4A512 and TI ADS1274 ADCs, described above with SHM boards, are well-suited for SHM applications where high-throughput and high-resolution data acquisition is important, the pipeline architecture of the QF4A512 and Delta-sigma modulation technique of ADS1274 introduce inherent latency issues, which is critical for structural control application. To achieve low-latency data acquisition while maintaining good resolution, the SHM-SAR board has been developed around a SAR-type ADC. A Successive Approximation Resister (SAR) type ADC is a single-shot converter that completes the conversion within one sampling interval. The SHM-SAR board, shown in Fig. 11, offers an on-board 3-axis MEMS accelerometer or four channels of analog input, which can be selected with a switch. At an Imote2 processor speed of 416 MHz, the maximum rate for sampling a single channel or multiple channels are 3700 and 2700 Hz, respectively. The RMS noise floor is 2.9 mg when unfiltered, but can be limited to 1.8 mg with a simple analog filter [27]. Overall, the latency due to the SHM-SAR is 200 μs, which is significantly lower than oversampling type architectures. Thus, the latency due to the data- acquisition hardware alone can be neglected.

Fig. 11

SHM-SAR board: top (left) and bottom (right) [27]

3.1.2.2 DAC board

A digital-to-analog converter (DAC) in combination with a wireless smart sensor can enable both sensing and actuation in the WSSN, such as structural control or active sensing [49]. The SHM-D2A board (see Fig. 12) has a DAC to support output of analog signals. The four channels of digital inputs delivered from the I²C interface of the Imote2 are converted to external analog outputs ranging 0–2.5 V using the TI-DAC8565. The TI-DAC8565 offers comparable performance to the SHM-SAR, so no performance is lost with the SHM-D2A. This DAC functionality provides an effective tool for wireless control.

Fig. 12

SHM-D2A board: perspective (left) and bottom (right) [26]

3.2 Software improvements

3.2.1 Effective network power management

As effective management of power consumption is critical in the long-term and autonomous operation of the WSSNs, substantial improvements have been made for (1) power saving, (2) energy harvesting, and (3) network power monitoring. Essential improvements include:

3.2.1.1 Data storage in non-volatile memory

Non-volatile flash memory is available to store measured data sets. This feature enables sensor nodes to sleep after data collection without waiting for subsequent transmission to the gateway node. The waiting time for each sensor node increases proportionally to the network size, which causes significant energy consumption in the large-scale WSSN. The data in the flash memory does not go away due to the sleep function, and data can then be retrieved when the each sensor node is requested to wake up and send data sequentially.

3.2.1.2 Leaf node-operated ThresholdSentry

In the previous approach, when the gateway node wants to wake up the sentry nodes in multi-hop communication, it first broadcasts wake-up messages to all sensor nodes in the network, which are subsequently transmitted to sentry nodes. Then, the sensor nodes hearing the message stay awake for multi-hop route establishment to the sentry nodes, consuming too much energy while waiting for the route to be established. In the current implementation, the sentry tasks are scheduled from sentry nodes themselves, which are autonomously checking the vibration threshold periodically based on the parameters received from the gateway and stored in flash memory at the beginning, without the need to be woken up by gateway node every time they perform the sentry task.

3.2.1.3 Long-term sleep mode

If a sensor node has relatively low battery power, the long-term sleep mode is enabled to power off during, or after, sensing. Since the Imote2 uses internal PMIC, it is impossible to rerun the energy harvesting once a sensor node is turned off by insufficient power supply. In the current implementation, sensor nodes whose battery voltage is lower than a preset threshold autonomously enter the long-term sleep mode where only charging is performed in the sensor nodes. Naturally, they are excluded from the measurement before entering long-term sleep and integrated back into the network again after the battery has recharged beyond another preset threshold.

3.2.1.4 Autonomous monitoring of network power

The utility commands Vbat and ChargeStatus of the Illinois SHM Services Toolsuite provide information about power supply for the Imote2, such as battery voltage, charging voltage, and current. AutoUtilsCommand enables autonomous tracking of battery voltage and charging status of the entire sensor network and is scheduled within AutoMonitor .

3.2.2 Reliable network operation

WSSNs can be exposed to various unexpected conditions that can possibly cause the network to malfunction. To minimize network malfunctions and to operate the network more reliably, fault-tolerance features have been implemented.

3.2.2.1 Unresponsive nodes

With AutoMonitor , each sensor node is requested to wake up when the WSSN starts measurement. However, some sensor nodes, after they notified the gateway node that they were awake at the beginning, can become unresponsive due to low battery power, casual radio communication block, or an unexpected hardware malfunction before or after sensing. In the current implementation of the software, the radio response is checked before data retrieval to avoid any unnecessary waiting time and excessive power use. The unresponsive nodes are set to be excluded in the network until it responds again.

3.2.2.2 Autonomous resumption of AutoMonitor

During the long-term operation, the AutoMonitor service running on the gateway node may experience unexpected resets due to casual malfunctioning or disconnection to the base station of the gateway node. AutoMonitor is resumed after unexpected resets by creating a checkpoint which saves into flash memory the parameters defining the current state of AutoMonitor after the completion of each scheduled task. If the gateway node experiences an unexpected reset, the parameters saved in flash memory are reloaded to restart AutoMonitor from the last saved checkpoint.

3.2.2.3 E-mail notification of structure and network anomalies

To cope with any anomalies in the WSSN, such as outlying responses, unresponsive nodes, poor wireless communication, and recharging inability, e-mail notification service has been added. This e-mail notification is based on parsing log messages from gateway node during operation. If any anomalies are found in the log messages, they are automatically forwarded to a predefined group of individuals responsible to take proper actions. This service has significantly contributed to debugging the WSSN and developing fault-tolerance features.

3.2.3 Multi-hop communication with an advanced protocol

The initial development of a multi-hop network identified multiple shortcomings and inadequacies of commonly used protocols, such as AODV [37] and CTP [12], for data-intensive applications needing reliable multi-hop communication under real-world deployment conditions including sparse, predominantly poor-quality radio links and bursty traffic patterns. In particular, the metrics used for link quality estimation, such as ETX [8], are not capable of quickly and accurately estimating end-to-end route quality during short data bursts.

The empirical measurements collected during the first deployment were used to identify critical factors affecting the performance of multi-hop routing in data-intensive WSSN applications and were employed in the design of an improved multi-hop communication and reliable data transfer protocol. As discussed in Sect. 2, the General Purpose Multi-hop (GPMH) protocol is developed to support diverse data flow patterns, including intermittent bursty traffic, such as centralized data aggregation, dissemination, as well as decentralized communication that are possible in SHM applications [33].

Under AODV, route request messages are first flooded from the source node, and then route reply messages are flooded back from the destination node. The “next hop” at each node in the route is selected using the minimum hop count metric among the received route reply messages. In WSSNs, link quality metrics provided by radio hardware, such as received signal strength (RSSI) and link quality indication (LQI), are often used in addition to minimum hop count to select the most reliable links. Unfortunately, both LQI and RSSI are extremely noisy and cannot reliably predict link quality on the basis of a single measurement, particularly when differentiating among several poor-quality links.

The GPMH protocol employs an alternative routing metric specifically designed to rapidly (on the basis of a single measurement) provide an accurate assessment of link quality, which is a desirable feature for the short, bursty communication patterns often found in data-intensive applications such as SHM. The metric combines RSSI and LQI measurements into a regression tree, a decision tree data structure automatically generated using least-squares regression, to sort links into quality buckets based on prior on-site measurement. In this way, it combines the long-term link measurements suitable for high-throughput data transfer, as in ETX, with instantaneous, single-packet link quality estimation using RSSI or LQI, which is needed for quickly building reliable routes for bursty traffic. Figure 13 demonstrates that the regression tree metric performs better than either LQI or RSSI at estimating the packet reception rate (PRR) for bursty traffic and is especially good at identifying and distinguishing poor-quality links, which were very common on the Jindo Bridge deployment.

Fig. 13

Estimation of link packet reception rate (PRR) using LQI, RSSI, and Regression Tree methods

3.2.4 Decentralized in-network processing

Centralized data acquisition and processing schemes (see Fig. 14a) that are commonly used in traditional wired sensor systems are not tractable in WSSNs, because bringing all data to a centralized location will result in severe data congestion in the WSSN [41]. Decentralized processing schemes significantly reduce the amount of data transferred through wireless communication, so that it ensures the scalability of WSSNs required to enable a dense array of sensors deployed on full-scale civil infrastructure [31]. Two decentralized processing software services have been developed; one uses independent data processing to estimate the tension in stay cables (see Fig. 14b) and the other is based on coordinated computing to aggregate cross-correlation data (see Fig. 14c).

Fig. 14

Data acquisition and processing schemes: a centralized data collection, b independent processing, and c coordinated computing strategy [31]

CableTensionEstimation [44] is an independent processing-based service that autonomously interrogates cable tensions using the vibration-based method developed by Shimada [40]. Stay cables are the main load carrying members of a cable-stayed bridge, and tension forces of cables are direct indicators of a cable’s integrity as well as overall structural health of the cable-stayed structure. Previously, Cho et al. [4] implemented a vibration-based tension estimation method proposed by Zui et al. [51] on an academic WSS platform. In CableTensionEstimation , a closed form relationship between multiple natural frequencies and the tension force [40] is implemented as a practical alternative of Zui’s method, because Zui’s method uses lower natural frequencies that are easily contaminated by cable-deck interaction on cable-stayed bridges. CableTensionEstimation performs unsynchronized sensing, estimates the power spectrum density (PSD), and picks peaks of PSD in the automated manner [44] to estimate cable tensions.

Built based on the coordinate processing, DecentralizedDataAggregation [41] implements the decentralized calculation of cross-correlations (CC) for the natural excitation technique [15] and the random decrement (RD) technique [7] (see Fig. 9c). As opposed to the independent processing used in CableTensionEstimation , DecentralizedDataAggregation allows sensor nodes to communicate with each other and share information. The sensor network is divided into local sensor communities where data communication and processing are taking place to estimate either CC or RD function, depending on the user-specified input. The CC or RD functions can be either collected at the base station or retained at the cluster-heads for further analysis such as in-network system identification and damage detection. DecentralizedDataAggregation can be used as a foundation for other application development that requires the coordinated processing such as decentralized modal analysis [42, 43], decentralized system identification [41], and decentralized damage detection [17].

3.2.5 Autonomous operation of multi-task network

Long-term autonomous SHM with WSSNs requires a high level software manager to handle the overall network activities. The major goal of the software manager is to ensure that all the tasks of the network are properly scheduled. Especially the trigger-based tasks are executed without interrupting the on-going or other scheduled tasks. An application called AutoMonitor [39] was developed for the Imote2 WSSNs as described earlier.

One issue with AutoMonitor is that the operations of the timers for scheduling tasks and the actual code for executing the tasks are all lumped in the same file, making the maintenance and expansion of AutoMonitor extremely difficult—especially when a large number of tasks are involved. For example, all execution of tasks is implemented in the single SnoozeAlarm.awake event. Therefore, many different flags need to be carefully checked before a task can be executed. A better design would be separating the task scheduling (timer) component from the task execution component.

To make the AutoMonitor easier to maintain and expand, a new service module called AutoScheduler [23] has been developed in the ISHMP Services Toolsuite which takes care of the scheduling timers and the initiation of task execution after the timers are fired. In the AutoScheduler module, a parametric interface ScheduleApp provides a command ScheduleApp.registerApp to register different network-wide tasks and a command ScheduleApp.executeApp to execute these tasks. The registration involves assigning a scheduling timer to each task, defining the name of the task, execution period, priority, and the name of the function which will be executed once the command ScheduleApp.executeApp is called. Meanwhile, another interface called AutoScheduler provides two commands, AutoScheduler.start and AutoScheduler.stop , to start and stop the scheduling timers. AutoScheduler provides a self-contained control center for scheduling all network tasks and can avoid any potential conflict between tasks in order to facilitate smooth operation of autonomous monitoring. Figure 15 shows the block diagram of the new AutoMonitor and embedded AutoScheduler .

Fig. 15

Block diagram of the new AutoMonitor and embedded AutoScheduler

3.2.6 Multi-scale time synchronization

As in many other WSSN applications, synchronization of the network is highly desirable in SHM applications. However, due to the specific features in SHM, such as high sampling frequency, extended sensing duration, and uncertainties in the software and hardware, synchronization of data is not automatically guaranteed even with accurately synchronized clocks [24, 31]. The time synchronization strategy implemented in the Illinois SHM Services Toolsuite was composed of a two-stage synchronized sensing; The Flooding Time Synchronization Protocol (FTSP) was adapted to provide clock synchrony in the first stage, then the data is resampled after sensing is finished to remove the errors due to uncertainties in hardware and software in the second stage. Particularly, before sensing, a 30-s period is used to broadcast beacon messages for estimating the clock drift rates at the sensing units through linear regression. The linear clock drift rates are then used to correct the clocks so that the samples can be accurately time-stamped during sensing.

3.2.6.1 Nonlinear clock drift compensation

The limitation of the assumption of linear clock drift is that during long-duration sensing, nonlinearity in clock drift may become significant due to temperature effects. When the temperature changes, the resonant frequency of the clock crystal oscillators will change, leading to nonlinear clock drift. In addition, the 30-s period for broadcasting beacon messages before sensing delays the start of sensing and is therefore undesirable in SHM.

To address the issue of nonlinear clock drift due to temperature change, an improved time synchronization strategy is proposed and implemented in the ISHMP Services Toolsuite. The strategy aims to capture the full picture of nonlinear clock drift during the sensing period so as to fully compensate for the nonlinear clock drift in data timestamps. As illustrated in Fig. 16, a single synchronization message/beacon is used to synchronize the clocks and then all leaf nodes in the network can start sensing at roughly the same time. During sensing, the gateway node continues to broadcast beacons periodically with its global time. Upon receiving the beacons, the leaf nodes time-stamp the beacons and compute the offsets. Once sensing is finished, the recorded local timestamps and offsets are used to depict the complete history of clock drift during the sensing period through nonlinear regression analysis. Subsequently, the data timestamps can be corrected using the fitted nonlinear curve of the clock drift. Finally, resampling is performed based on the drift-compensated timestamps to achieve data synchronization. In order to take into account potential beacon loss due to packet collision during broadcast, multiple beacons are transmitted during sensing to ensure the accuracy and robustness of the nonlinear regression analysis. Using the nonlinear clock drift compensation method, controlled lab experiments showed great improvement of synchronization accuracy, less than 50 µs maximum error over an extended measurement period [24].

Fig. 16

Implementation of the improved synchronized sensing

3.2.6.2 GPS-based synchronization of multiple sub-networks

While compensating the nonlinear clock drift, to achieve synchronized sensing among multiple sub-networks, a GPS-based synchronization feature has been added to the Illinois SHM Services Toolsuite. This new stand-alone application periodically adjusts the gateway nodes’ clocks using Pulse Per Seconds (PPS) signals from GPS receivers and the corrected clock information is shared within a sub-network. This protocol aims to enable precise synchronized sensing for multiple networks, providing scalability by removing communication overhead among sub-networks.

The clocks on multiple gateway nodes are synchronized through the hardware interrupt and the Coordinated Universal Time (UTC) information. As the application PPS_based_RemoteSensing starts, PPS signals from a GPS receiver interrupt the general purpose input output (GPIO) pin on the gateway Imote2. The interrupt handler passes the signals until time reaches the input UTC time. As shown in Fig. 17, at the desired UTC time, all gateway nodes adjust their local clocks to a common value (for example, zero as implemented in the application). Once the local clocks are adjusted to a common value, the gateway nodes broadcast the synchronized time stamps and the calculated sensing start time to the leaf nodes in their sub-network. Because the timestamps on the gateway nodes are synchronized, leaf nodes in multiple networks are expected to synchronize their clock based on the same timestamps without requiring message exchanges among sub-networks. Once all the gateway nodes of multiple sub-networks are synchronized, the two-stage approach with nonlinear clock drift compensation, described above, is followed for clock and data synchronization among leaf nodes in each sub-network. The application achieved precise synchronized sensing with less than 50 µs maximum error among multiple sub-networks requiring only a single GPS receiver in each sub-network.

Fig. 17

Clock synchronization of multiple sub-networks using PPS signals of GPS

3.2.7 Wireless structural control

3.2.7.1 Real-time data streaming

Real-time wireless data acquisition expands the applications of WSS, allowing wired data acquisition systems to be mimicked and offering real-time visualization of structural response. Also, this approach is critical when real-time state knowledge is required (i.e., feedback for structural control). Thus, high-throughput, real-time, wireless data acquisition service for the Imote2 is developed [25].

Real-time data acquisition using WSS is challenging due to operating system limitations, tight timing requirements, sharing of transmission bandwidth, and unreliable radio communication. In the case of the Imote2, TinyOS completes tasks in a first-in-first-out (FIFO) manner along with interrupts to facilitate interaction with hardware [22]. To realize real-time data acquisition under the FIFO data handling, any processing of a data sample (i.e., temperature correction, time stamping, and data transmission) should be completed before the next data sample is acquired; as a result, the maximum sampling rate is limited by the time required for each of these steps. Furthermore, the clear channel assessment used with the Chipcon CC2420 radio can increase the communication time when multiple nodes transmit at the same time [21]. For time synchronization of a WSSN, the random offset in the sensing initiation time, which is inevitable due to hardware variability, requires the timestamps to be transmitted with the data sample; this additional transmission increases the overhead for each sample [32].

To achieve real-time data acquisition in light of these constraints, a tightly coordinated scheduling (TCS) approach is proposed. The TCS approach requires limiting the time of each processing and sending step to achieve high data throughput. First, the data payload is limited to one packet to reduce the communication time. The maximum packet payload is 112 bytes, which corresponds to nine samples of four channels of 16-bit data and a 32-bit timestamp [45]. Thus, nine samples can be buffered prior to sending. Second, a staggered time division multiple access (TDMA) scheme, which only allows each node to communicate with the gateway node during a specified timeslot, is used to limit the variability in communication time due to contention and back-off delays and reduce packet loss due to collisions. The number of nodes and sending/processing times are accounted for in the TDMA approach making this approach different from other MAC-layer protocols, which cannot account for these variables [13, 47].

The complete application combines accurate time synchronization, reliably broadcasted commands to initialize sensing and compute the TDMA send time, and the scheduled communication protocol to achieve high-throughput near-real-time data acquisition. The scheduled communication approach leads to a tradeoff between network size and maximum sampling rate. The communication and processing protocols allow for near-real-time sensing of 108 channels across 27 nodes at up to 25 Hz with minimal data loss as shown in Table 1. As the network size increases, the corresponding maximum sampling rate decreases and the data throughput remains relatively unchanged due to the increased number of sensor nodes.

Table 1 Real-time wireless data acquisition performance

3.2.7.2 Centralized and decentralized control

Essential for wireless control implementation are consistent sensing and actuation times among nodes, accurate computation, and flexibility for a variety of feedback architectures. The tight scheduling and communication protocol insights developed from streaming data are applied to achieve these goals [25]. Two wireless control implementations are available: centralized and fully decentralized. In fully decentralized control, there is no sensor communication between nodes. The controller node processes the sensor data, calculates the control command based on the embedded controller, and commands the SHM-D2A. The current control implementation uses a predictor–corrector form of the discrete-time Kalman filter and a constant gain feedback. The maximum sampling rate possible depends on the number of states in the calculation. For a system with four states, the maximum sampling of the fully decentralized control system is 950 Hz.

The control node framework developed for fully decentralized control is extended with real-time data acquisition (data streaming) for centralized control applications. The TDMA protocol uses tight timing to achieve good sampling performance while limiting data loss. In this case, the timeslot length is 7.8 ms. If all samples are received, the command is calculated based on the designed feedback controller. On the other hand, if all samples are not received, the previous command is held constant. The overall sampling rate is a function of the number of nodes in the network and the states in the control calculation. For four remote nodes, the maximum sampling rate is about 31 Hz.

4 Implementation and validation

The majority of hardware and software developments and advances mentioned in the previous section were validated through full-scale deployments on a cable-stayed bridge (the 2nd Jindo Bridge) test-bed in Korea and on a historic truss bridge (the Government Bridge at Rock Island Arsenal) test-bed in the US and through lab-scale experiments on a shear building (structural control). This section describes these deployments and experiments.

4.1 Full-scale deployment for monitoring a cable-stayed bridge in Korea

Deployment of the WSSN on the Jindo Bridge was carried out as an international collaborative research effort between the US (University of Illinois at Urbana-Champaign), South Korea (KAIST), and Japan (University of Tokyo). The collaborative project resulted in the first long-term and the world’s largest deployment of WSSs for monitoring civil infrastructure.

The Jindo Bridges are twin cable-stayed bridges. The bridges connect Jindo Island and Haenam located at the southwestern tip of Korean Peninsula. The 1st Jindo Bridge (right of Fig. 18) was constructed in 1984 with the design live load of DB-18 (similar to AASHTO HS-20). The 2nd Jindo Bridge (left of Fig. 18) was constructed in 2006 with the enhanced design live load of DB-24 (about 30 % larger than AASHTO HS-20) due to the increased volume and weight of the traffic between Jindo and Haenam. The 2nd Jindo Bridge has a three-span streamlined steel-box girder with 344 m of main span and 70 m of two side spans. The girder is supported by the 60 stay cables connecting the two A-shape steel pylons on concrete piers. The 2nd Jindo Bridge was selected as the test-bed with the allowance of unfettered access by the bridge authority.

Fig. 18

The Jindo Twin Bridges

4.1.1 Initial deployment efforts

The initial deployment on the bridge test-bed was carried out in 2009. The 70 sensor nodes in the network were divided into two subnets: one on the Jindo-side and the other on the Haenam-side. The Jindo subnet consists of 33 nodes (22 deck nodes, 3 pylon nodes, 8 cable nodes). The Haenam subnet consists of 37 nodes (26 deck nodes, 3 pylon nodes, 7 cable nodes). Two base stations were installed on the piers of both pylons of the first Jindo Bridge in charge of each subnet. Each base station consists of an industrial-grade PC (AAEON AEC-6905) running Window XP Embedded OS, an uninterruptable backup power supply (APC ES550), and a gateway node. Each gateway node, consisting of an Imote2, interface board, antenna, and environmentally hardened enclosure, interfaces with the respective subnets. Each leaf node is composed of an Imote2, battery board, sensor board, antenna and environmentally hardened enclosure. Three D-cell batteries were used to power most of the leaf nodes, except 8 nodes self-powered using solar panels and rechargeable batteries. SHM-A boards were used to measure acceleration, temperature, humidity, and light for most of nodes; the SHM-W sensor board (an early version of the SHM-DAQ) was used to measure the signal from a 3D ultra-sonic anemometer. The deployment details, evaluation, and data analysis can be found out in Jang et al. [16, 17] and Cho et al. [5].

4.1.2 2010–2011 deployments

Advanced hardware and software were implemented on the test-bed bridge in the 2010 deployment. Energy harvesting strategies using solar panels and rechargeable batteries were employed for all sensor nodes based on the satisfactory performance during 2009; additionally, a mini wind turbine was installed on one node to assess the potential for wind energy harvesting [35]. The network size was also increased to a total of 659 channels on 113 sensor nodes; which is currently the world’s largest WSSN established for SHM. To better understand the wind conditions on the bridge, three 3D ultra-sonic anemometers were installed with SHM-DAQ boards. Ten SHM-H boards were implemented as cluster heads for DecentralizedDataAggregation to increase the accuracy of decentralized modal analysis of the deck and pylons [18]. Figures 19 and 20 show example sensor nodes installed on the deck and cable, enclosure assembly detail, and energy harvesters.

Fig. 19

Enclosure assembly (left), sensor module mounting (middle) and installation using magnet (right)

Fig. 20

Solar panel on cable node (left), mini wind turbine (middle) and 3D ultra-sonic anemometer (right)

To efficiently operate such a large sensor network, the WSSN was divided into four subnets, considering the functionalities, network size, communication range, and communication protocol of each network. The sensor topology of the 2010–2011 deployment is shown in Fig. 21. The four subnets share a common software configuration that includes reliable data acquisition ( RemoteSensing ), autonomous operation ( AutoMonitor ), power harvesting ( ChargerControl ), sleep-cycling ( SnoozeAlarm ), and monitoring of the network status ( AutoUtilsCommand ). In addition to the common software, each subnet was designed to have its own unique software application. For example, deck subnets employed DecentralizedDataAggregation for decentralized modal analysis, while cable subnets employed CableTensionEstimation to estimate tension forces of cables in the automated manner. The Jindo-side cable subnet featured multi-hop communication for the purpose of evaluating the newly developed multi-hop protocol.

Fig. 21

Sensor topology with node IDs (2010–2011 deployment on the 2nd Jindo Bridge)

In the 2011 deployment, several hardware and software features were updated from the 2010 deployment. ThresholdSentry in multi-hop communication employed in the Jindo-side cable subnet was updated to operate autonomously without having to receive commands from the gateway node to prevent energy consumption by frequent re-routing. Additionally, a long-term sleep mode was implemented to enhance the long-term survival of the WSSN. The e-mail notification function enabled the gateway nodes to notify the network manager when abnormalities are detected in the WSSN. An improved multi-hop protocol was deployed which includes a better optimized routing algorithm. The newly developed strain sensor boards (SHM-S) were installed and validated on the bridge (see Fig. 7).

4.1.3 Example analysis of measured data during Typhoon Kompasu

The Jindo Bridge experienced Typhoon Kompasu, having the 960 hPa of central pressure and 40 m/s of max wind speed just after the 2010 deployment (Aug. 31st–Sep. 2nd, 2010). The typhoon passed quite close to the bridge, as shown in Fig. 22a. According to the Korea Meteorological Administration (KMA) records, the wind was coming from the SSE (see the dashed arrow in Fig. 22b) with the speed of 14–20 m/s in the Jindo area at 21:00 on Sep. 1st, 2010. The measured wind data at the same time using a 3D ultra-sonic anemometer interfaced with the SHM-DAQ board is shown in Fig. 22c. It indicates 15–25 m/s of wind speed and 170°–200° (from SE direction) of wind direction (see the solid arrow in Fig. 22b).

Fig. 22

Typhoon Kompasu a path on Sep. 1st, 2010, b measured wind direction, and c measured data at 21:00 using 3D ultra-sonic anemometer with SHM-DAQ board

Each sensor node equipped with an SHM-A or SHM-H sensor board provided 3-axes accelerations at the sampling rate of 25 Hz (which is user-selectable). Figure 23 shows the example vertical (z-axis) acceleration excited by the Typhoon Kompasu and heavy truck passage on the bridge, measured at the center of main span. The acceleration level by the Typhoon is consistently around 20 mg, which is comparable with a continuous heavy truck loading.

Fig. 23

Example vertical acceleration measured from center of main span. a Excited by a typhoon. b Excited by a heavy truck

The acceleration responses collected from the two deck subnets were utilized to identify modal properties of the bridge, which are compared with those obtained before the typhoon. For the output-only modal identification technique, the Natural Excitation Technique in conjunction with Eigensystem Realization Algorithm (NExT/ERA) was implemented resulting in Table 2. The estimated modal properties are found to be consistent with those obtained from the wired sensor system before the typhoon [5].

Table 2 Identified natural frequencies and comparisons

4.2 Full-scale deployment for monitoring a historic truss bridge at Rock Island

Currently, much of the national bridge stock of the United States has reached or is reaching the end of its design life. SHM of the old and deteriorated structures is important in ensuring that they are still functional and safe for their intended uses. The continued monitoring and use of these structures represents a sustainable approach to meeting the transportation needs of today. An implementation of an integrated fiber optic and wireless SHM system on the 115-year-old steel truss Government Bridge at the Rock Island Arsenal illustrates this important use of SHM technology.

Located over the Mississippi River between Rock Island in the state of Illinois and Davenport in the state of Iowa, the Government Bridge is one of over two hundred steel bridges owned and operated by the US Army and the Army Corps of Engineers. The Government Bridge is an eight-span, double decker, steel truss bridge, as shown in Fig. 24a, where the upper deck carries rail traffic and the lower deck carries vehicular traffic and pedestrians. The second span of the bridge, shown in Fig. 24b, is a draw span that can swing 360° in either direction to allow boats to pass and has been in near continual operation since its construction. The Government Bridge is still a relevant and vital link in the national transportation network, and its importance has increased in the last few years due to a large portion of the nation’s ethanol and bio-diesel crossing the Mississippi over this bridge [9].

Fig. 24

Government Bridge. a Full spans. b Monitored draw span

Unlike conventional bridges, the rotational position of the swing span must be identified prior to the analysis of the measured data. Because the bridge is symmetric, the only distinguishing feature of the draw span position is a staircase used to access the operator’s house. When the bridge is locked to allow cars to cross and these stairs face upstream of the river (original position), the structural responses differ from those when the stairs face downstream (opposite position). The bridge response is also different when it opens to allow boats to go by and acts as two cantilevers. These three positions will be designated as ‘closed-upstream’, ‘closed-downstream’, and ‘swing’.

4.2.1 Integrated SHM system

To supplement the traditional visual inspections, the Army Corps of Engineers installed a fiber optic SHM system with 34 FBG strain sensors on the draw span of the Government Bridge [10]. To enable comprehensive structural health monitoring, a wireless SHM system, mainly containing tri-axial MEMS accelerometers, was also installed in July, 2011. A wired digital compass was also installed to help determine the bridge position.

The measurements obtained from the integrated SHM system provide a comprehensive view of the bridge condition. The strain data from the fiber optic system is processed to provide information on when train and swing events occur on the bridge along with the raw strain measurement. The change in strain due to the bridge swinging is found to be constant, within statistical limitations, when compensation for temperature and sun exposure is performed. The acceleration data from the wireless system is first segmented, based on the detected events, and then processed to determine the modal properties of the structure. Both the modal properties of the bridge and change in strain due to swing events can be fed into a model updating algorithm to determine likely locations of changes to the health of the structure. The acceleration data also can be used in damage detection algorithms to determine damage indices for the members in the system [10].

4.2.2 Installation of wireless SHM system

In July 2011, a WSSN consisting of 22 sensor nodes was deployed on the Government Bridge at the Rock Island Arsenal as shown in Fig. 25 [6]. The sensor nodes are composed of the Imote2, battery board, sensor board, antenna, and environmentally hardened enclosure, similar to those deployed on the Jindo Bridge. Only SHM-A sensor boards measuring 3-axis acceleration are employed in the WSS nodes. The sensor nodes were distributed optimally in a pattern designed to capture appropriate modal data, based on a previously conducted finite element model analysis. All the sensor nodes are equipped with solar panels and rechargeable batteries. The solar panels for the sensor nodes were placed to ensure maximum sunlight and to keep them out of reach of curious pedestrians. The recently developed AutoMonitor with its fault-tolerant features was employed for both the gateway and sensor nodes for autonomous, but reliable, operation of the monitoring system.

Fig. 25

Installed sensors on Government Bridge. a FOS stain gauges (diamond), digital compass (square), and WSS (circle). b WSS on top chord. c WSS on railway deck

4.2.3 Example analysis of measured data

Three types of measured data worked together to monitor activities on the bridge. The compass reading gave the heading of the bridge indicating when the bridge is turning. Figure 22a shows the compass heading plotted with an acceleration measurement when the bridge swings. The compass reading indicates the bridge has left its initial heading close to 270° and begun to turn counter clockwise indicated by approximately 90° reduction; the wireless system recorded the vibration of the bridge caused by its rotation with a lag of approximately 3 min due to time synchronization of the whole sensor network. The bridge stopped for a period of time half-way through its swing while the boats were passing; subsequently, the bridge began to rotate and the vibration started again. When the bridge was closed after the swing, locking of the roller jacks on both ends of the span caused a large impulse in the acceleration record. After the pulse, normal traffic crossed the bridge. The figure shows how the compass reading can help interpret the data with regards to the position and state of the bridge.

Strain data recorded by the fiber optic system can also help interpret the accelerations recorded by the wireless SHM system. Figure 26b shows a vertical acceleration from a sensor node plotted with a compass heading and the strain from the FO sensor closest to the WSS. The compass heading remains constant, indicating the bridge did not swing. Instead, a train entered the bridge at about 17:28, as indicated by the increase in the strain, and the wireless node began recording acceleration after time synchronization. Just before 17:34, the train stopped moving on the bridge. During the train’s pause, the acceleration level became very small while the strain level remained elevated with no dynamic amplification. After the train movement resumed, acceleration and strain levels returned to the levels before the pause. The example shows how strain data can be used to analyze the traffic patterns from the acceleration measurement [11].

Fig. 26

Example measured acceleration (by WSS), heading (by digital compass), and strain (by FOS strain gauge). a Acceleration and heading when bridge swings. b Acceleration, heading, and strain when a train passes with a pause

With the help of complementary data (i.e., compass heading and strain), the acceleration measurements were divided according to the traffic condition and bridge positions. Separating the data based on traffic is necessary as the weight of trains will greatly affect the modal properties of the bridge. Therefore, the acceleration data excited by automobile traffic at closed bridge positions and ambient vibration during swings was used to identify the bridge’s modal properties. An output-only modal identification method, frequency domain decomposition (FDD) [1], was employed for the modal analysis. Figure 27 shows an example of the modal properties (i.e., natural frequencies and mode shapes) identified by the FDD method. Acceleration was measured for 10 min at 100 Hz when the bridge swung. Figure 27b–e shows the first four mode shapes of the bridge in the open position. The 1st mode that appears at 0.391 Hz is a rocking/vertical mode because of the cantilever nature of the bridge in the swung position. The other experimental mode shapes are consistent with the bridge’s shape, geometry, and boundary conditions. The identified modal properties were successfully verified using a detailed FE model of the bridge reported in Cho et al. [6] and Giles [11]. Using the modal information, many vibration-based system identification techniques, such as FE model updating, modal strain energy method, and damage locating vector method, can be employed in assessing the bridge’s integrity.

Fig. 27

Example modal analysis result of Government Bridge: swing position. a Singular values from FDD. b 1st vertical mode (rocking): 0.391 Hz. c 1st lateral mode: 1.074 Hz. d 2nd lateral mode: 1.831 Hz. e 2nd vertical mode: 2.075 Hz

4.3 Lab-scale implementation of wireless feedback control

Prior to full-scale implementation, the use of WSSN for structural control needs to be tested in a laboratory environment to evaluate control performance and ensure stability in the presence of data loss, sampling rate limitations, and time delay. The previously discussed WSS developments for control are implemented on an experimental system to evaluate the performance of two wireless structural control strategies: centralized and fully decentralized. The four-story small-scale structure is fitted with two active mass driver (AMD) control systems on the 2nd and 4th stories, respectively (Fig. 28). The combination of the two AMDs allows for reduction of the response in the higher modes and different decentralized control strategies. A comparison of single AMD and multiple AMD performance can be found in Linderman [26], which considers failures in controller nodes. The overall height of the structure is about 2.2 m and has a first natural frequency of 0.68 Hz. In addition, a small, single-axis shake table is used to excite the structure.

Fig. 28

4-story experimental structure (left) and three control configurations (right): wired, centralized wireless, and fully decentralized wireless

Overall, three strategies are implemented on the system to evaluate the wireless control performance. The three strategies are illustrated in Fig. 28. The traditional wired system operates at 1000 Hz and can be approximated as continuous in the control design. The centralized wireless control system operates at 30 Hz due to the time required for feedback. A discrete-time control approach is used to account for the slow rate in the design. The fully decentralized control system uses two subsystems that operate at 725 Hz. There is no communication between the two subsystems. These are all compared to a ‘zeroed’ control case where a zero command is issued to the AMD.

The control strategies are evaluated under several earthquake excitations: El Centro, Northridge, Kobe, and Chi Chi. The peak and RMS responses normalized to the uncontrolled are presented in Figs. 29 and 30, respectively. Overall, both wireless control strategies are able to control the response of the structure and no instability results. However, the peak responses of the wireless centralized surpass the uncontrolled at the 2nd and 4th stories due to the ‘jumpier’ control performance. The fully decentralized system outperforms the wireless centralized controller despite having less system knowledge. This improvement is likely due the higher sampling rate, which improves the estimator performance. In addition, both wireless systems have more trouble with the Chi Chi and Kobe ground motions because of their cyclic, impulsive load and associated saturation of the SHM-SAR measurements. However, in general, the fully decentralized wireless controller results in good performance.

Fig. 29

Comparison of normalized peak responses under earthquake excitation

Fig. 30

Comparison of normalized RMS acceleration responses under earthquake excitation

5 Conclusions

This paper discussed recent advances and laboratory and field validation of a state-of-the-art smart sensor framework developed for SHM and control applications. Various hardware and software issues are taken into account based on the lessons learned from recent efforts to develop, stabilize, and operate full-scale wireless smart sensor networks for SHM and control of civil infrastructure. High-fidelity, multi-scale responses can now be captured using a variety of sensor boards that were recently developed. Time synchronization algorithms assure that not only the clocks on each Imote2 in the network but data collected from those sensors are accurately synchronized for extended measurement periods. Time synchronization among multiple networks is available using GPS receivers. Decentralized computing is realized for decentralized data aggregation and autonomous cable tension estimation by fully utilizing computing capabilities and wireless communication. Routing in multi-hop communication is advanced to minimize the data loss and energy consumption. Data streaming is available for mimicking wired sensors and real-time wireless control. Centralized and decentralized control schemes have been developed. The Illinois SHM Services Toolsuite is updated with more services and fault-tolerance features. Using new (the 2nd Jindo Bridge) and historic bridges (the Government Bridge), full-scale WSSNs are realized for the purpose of SHM. Wireless active control has successfully been implemented at lab scale using the WSSN developed herein. The evaluation of the WSSNs as well as the data analysis shows the practicality of wireless SHM and control systems for civil infrastructure. Indeed, wireless smart sensor technology has reached a level of maturity that makes it an important tool for management of civil infrastructure.